Gernot Kostorz-Phase Transformations in Materials-Wiley-VCH (2001) PDF

Gernot Kostorz (Editor)
Phase Transformations in Materials
Phase Transformations in Materials. Edited by Gernot Kostorz

Copyright © 2001 WILEY-VCH Verlag GmbH, Weinheim
ISBN: 3-527-30256-5
Related titles from WILEY-VCH
Eric Lifshin (Ed.)

X-ray Characterization of Materials
ISBN 3-527-29657-3
Dierk Raabe
Computational Materials Science
ISBN 3-527-29541-0
Gernot Kostorz (Ed.)
Phase Transformations
in Materials
Weinheim · New York · Chichester · Brisbane · Singapore · Toronto

Editor:
Prof. Gernot Kostorz
ETH Zürich
Institut für Angewandte Physik
CH-8093 Zürich
Switzerland
This book was carefully produced. Nevertheless, authors, editor and publisher do not warrant the informa-
tion contained therein to be free of errors. Readers are advised to keep in mind that statements, data, illus-
trations, procedural details or other items may inadvertently be inaccurate.
Library of Congress Card No. applied for
British Library Cataloguing-in-Publication Data:

A catalogue record for this book is available from the British Library.
Die Deutsche Bibliothek – Cataloguing-in-Publication Data:

A catalogue record for this book is available from Die Deutsche Bibliothek
ISBN 3-527-30256-5
© WILEY-VCH Verlag GmbH, D-69469 Weinheim (Federal Republic of Germany), 2001
All rights reserved (including those of translation in other languages). No part of this book may be reproduced
in any form – by photoprinting, microfilm, or any other means – nor transmitted or translated into machine lan-
guage without written permission from the publishers. Registered names, trademarks, etc. used in this book, even
when not specifically marked as such, are not to be considered unprotected by law.
Printed in the Federal Republic of Germany

Printed on acid-free paper.
Indexing: Borkowski & Borkowski, Schauernheim
Composition, Printing and Bookbinding: Konrad Triltsch, Print und digitale Medien GmbH, 97199 Ochsenfurt-
Hohestadt
To the memory of Peter Haasen
(1927 – 1993)

ISBN: 3-527-30256-5
Preface
In 1991, the late Peter Haasen, who had set out to edit a comprehensive treatment of “Ma-
terials Science and Technology” together with Robert W. Cahn and Edward J. Kramer, a
very successful series of up-to-date coverages of a broad range of materials topics, wrote
the following in a Preface to Volume 5.
“Herewith we proudly present the first volume of this Series, the aim of which is to pro-
vide a comprehensive treatment of materials science and technology. The term ‘materials’
encompasses metals, ceramics, electronic and magnetic materials, polymers and compos-
ites. In many cases these materials have been developed independently within different dis-
ciplines but are now finding uses in similar technologies. Moreover, similarities found among
the principles underlying these various disciplines have led to the discovery of common
phenomena and mechanisms. One of these common features, phase transformations, con-
stitutes the topic of this volume and rates among the fundamental phenomena central to the
Series. A phase transformation often delivers a material into its technologically useful form
and microstructure. For example, the major application of metals and alloys as mechanical-
ly strong materials relies on their multi-phase microstructure, most commonly generated by
one or more phase transformations.”
Peter Haasen, who passed away in 1993, did not live to see the overwhelming success of
this volume. A revised edition was planned as early as 1996 and finally, work on the indi-
vidual chapters began in 1998. Almost all of the original authors agreed to update their ear-
lier work, in many cases they arranged for the participation of younger colleagues who had
made major contributions to the field. It is thus with similar pride the present editor sub-
mits the second edition of “Phase Transformations” to the scholarly public. It was with some
hesitation that he assumed the task of editor, as he did not like giving the impression of pla-
giarizing a successful work. All the credit for the idea, the original chapter definitions and
the selection of the initial authors remains with Peter Haasen. He would certainly have liked
to work on a new edition himself and, assuming that he would have judged it timely to ac-
complish it about ten years after the first edition, this editor tried to help reaching this goal
in the original spirit.
Thus, all the chapters kept their original titles. The contents have been thoroughly re-
edited and updated and reflect the progress in the field up to about the middle of the year
2000. As before, the book starts with the foundations of phase transformations (“Thermo-
dynamics and Phase Diagrams” by A. D. Pelton). The sequence of the following Chapters
has been slightly modified. As most of the volume concerns the solid state, Chapter 2 (by
H. Müller-Krumbhaar, W. Kurz and E. Brener) is devoted to solidification, a subject of great
basic and technological relevance. Chapter 3 by G. E. Murch covers the most important
ideas and methods of diffusion kinetics in solids, an indispensible ingredient to many phase
transformations. Statistical theories of phase transformations are presented by K. Binder in
Chapter 4, featuring phenomenological concepts and computational methods. Diffusion con-
trolled homogeneous phase transformations are treated in Chapter 5 (“Homogeneous Sec-

ISBN: 3-527-30256-5
Preface VII
ond Phase Precipitation” by R. Wagner, R. Kampmann and P. W. Voorhees) and Chapter 6

(“Spinodal Decomposition” by K. Binder and P. Fratzl), looking at the very complex kinet-
ical aspects of the formation of new phases from the points of view of metastability and in-
stability in an initially homogeneous system. Heterogeneous phase transformations are treat-
ed in Chapter 7 (“Transformations Involving Interfacial Diffusion” by G. R. Purdy and Y.
Bréchet) while Chapter 8 by G. Inden deals with atomic ordering, mostly involving substi-
tutional alloys and intermetallic phases. Though much progress has been made in elucidat-
ing equilibrium ordered states, kinetical aspects are still widely unexplored in this field. Fi-
nally, the numerous aspects of diffusionless transformations in the solid state are taken up
by L. Delaey (Chapter 9), and a completely new Chapter on the effects of pressure on phase
transformations has been provided by M. Kunz (Chapter 10).
In working on this new edition, the editor had great pleasure interacting with the authors,
those of the first edition as well as those who joined for the new edition. He is grateful to
all of them for their friendly and competent co-operation. Thanks are due to the publisher
for expedient support and for the preparation of the subject index.
It is hoped that this book will be useful as a source of reference to active researchers and
advanced students; more up-to-date and more detailed than encyclopedic articles, but not
as complete and extensive as any monographs. Phase transformations are among the most
complex and most versatile phenomena in solid state physics and materials science – and
have considerable impact on production and processing technology. The present book should
encourage the reader to enter and more deeply appreciate this challenging field.
Gernot Kostorz, Zürich

April 2001
List of Contributors
Prof. Kurt Binder Prof. Gerhard Inden

Universität Mainz Max-Planck-Institut für Eisenhütten-
Institut für Physik forschung GmbH
Staudinger Weg 7 Max-Planck-Str. 1
D-55099 Mainz D-40237 Düsseldorf
Germany Germany
Chapters 4 and 6 Chapter 8
Prof. Yves J. M. Bréchet Dr. Reinhard Kampmann

Laboratoires de Thermodynamiques GKSS-Forschungszentrum Geesthacht
at Physico-Chimique Métallurgiques BP75 GmbH
Domaine Universitaire de Grenoble Institut für Werkstoffforschung
F-38402 Saint Martin d’Heres Postfach 1160
France D-21494 Geesthacht
Chapter 7 Germany
Chapter 5
Dr. Efim Brener
Institut für Festkörperforschung Prof. Martin Kunz
Forschungszentrum Jülich ETH Zürich
D-52425 Jülich Labor für Kristallographie
Germany Sonneggstr. 5
Chapter 2 CH-8092 Zürich
Switzerland
Prof. Luc Delaey now at:
Katholieke Universiteit Leuven Naturhistorisches Museum Basel
Dept. Metaalkunde en Toegepaste Augustinerstr. 2
Materiaalkunde CH-4053 Basel
Decroylaan 2 Switzerland
B-3030 Heverlee-Leuven Chapter 10
Belgium
Chapter 9 Prof. Wilfried Kurz
École Polytechnique de Lausanne
Prof. Peter Fratzl DMX-G, Ecublens
Institut für Metallphysik CH-1015 Lausanne
Montan-Universität Leoben Switzerland
Jahnstraße 12 Chapter 2
A-8700 Leoben
Austria
Chapter 6

ISBN: 3-527-30256-5
List of Contributors IX
Prof. Heiner Müller-Krumbhaar Prof. Gary R. Purdy

Institut für Festkörperforschung Dept. of Materials Science and Engineering
Forschungszentrum Jülich McMaster University
D-52425 Jülich 1280 Main Street
Germany Hamilton, Ontario L8S 4L7
Chapter 2 Canada
Chapter 7
Prof. Graeme E. Murch
University of Newcastle Prof. Peter W. Voorhees
Dept. of Chemical and Materials Dept. of Materials Science and Engineering
Engineering Northwestern University
Romkin Drive 2225 N. Campus Drive
Newcastle, NSW 2308 Evanston, IL 60208-3108
Australia USA
Chapter 3 Chapter 5
Prof. Arthur D. Pelton Prof. Richard Wagner

École Polytechnique de Montréal Forschungszentrum Jülich
Centre de Recherche en Calcul D-52425 Jülich
Thermochimique Germany
CP 6079 Chapter 5
Succursale A
Montréal, Québec H3C 3A7
Canada
Chapter 1
Contents
1 Thermodynamics and Phase Diagrams of Materials . . . . . . . . . . . . 1

A. D. Pelton
2 Solidification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
H. Müller-Krumbhaar, W. Kurz, E. Brener
3 Diffusion in Crystalline Solids . . . . . . . . . . . . . . . . . . . . . . . 171

G. E. Murch
4 Statistical Theories of Phase Transitions . . . . . . . . . . . . . . . . . . 239

K. Binder
5 Homogeneous Second Phase Precipitation . . . . . . . . . . . . . . . . . 309

R. Wagner, R. Kampmann, P. W. Voorhees
6 Spinodal Decomposition . . . . . . . . . . . . . . . . . . . . . . . . . . . 409

K. Binder, P. Fratzl
7 Transformations Involving Interfacial Diffusion . . . . . . . . . . . . . . . 481

G. R. Purdy, Y. J. M. Bréchet
8 Atomic Ordering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 519

G. Inden
9 Diffusionless Transformations . . . . . . . . . . . . . . . . . . . . . . . . 583

L. Delaey
10 High Pressure Phase Transformations . . . . . . . . . . . . . . . . . . . . 655

M. Kunz
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 697

ISBN: 3-527-30256-5
1 Thermodynamics and Phase Diagrams of Materials
Arthur D. Pelton
Centre de Recherche en Calcul Thermochimique, École Polytechnique,

Montréal, Québec, Canada
List of Symbols and Abbreviations . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.1.1 Notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.2 Gibbs Energy and Equilibrium . . . . . . . . . . . . . . . . . . . . . . . 5
1.2.1 Gibbs Energy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.2.2 Chemical Equilibrium . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.3 Predominance Diagrams . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.3.1 Calculation of Predominance Diagrams . . . . . . . . . . . . . . . . . . . 7
1.3.2 Ellingham Diagrams as Predominance Diagrams . . . . . . . . . . . . . . 8
1.3.3 Discussion of Predominance Diagrams . . . . . . . . . . . . . . . . . . . 9
1.4 Thermodynamics of Solutions . . . . . . . . . . . . . . . . . . . . . . . 9
1.4.1 Gibbs Energy of Mixing . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.4.2 Chemical Potential . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.4.3 Tangent Construction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.4.4 Gibbs-Duhem Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.4.5 Relative Partial Properties . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.4.6 Activity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.4.7 Ideal Raoultian Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.4.8 Excess Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.4.9 Activity Coefficient . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
1.4.10 Multicomponent Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . 14
1.5 Binary Phase Diagrams . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
1.5.1 Systems with Complete Solid and Liquid Miscibility . . . . . . . . . . . . 14
1.5.2 Thermodynamic Origin of Phase Diagrams . . . . . . . . . . . . . . . . . 16
1.5.3 Pressure-Composition Phase Diagrams . . . . . . . . . . . . . . . . . . . 19
1.5.4 Minima and Maxima in Two-Phase Regions . . . . . . . . . . . . . . . . . 20
1.5.5 Miscibility Gaps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
1.5.6 Simple Eutectic Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
1.5.7 Regular Solution Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
1.5.8 Thermodynamic Origin of Simple Phase Diagrams Illustrated by Regular
Solution Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
1.5.9 Immiscibility – Monotectics . . . . . . . . . . . . . . . . . . . . . . . . . 26
1.5.10 Intermediate Phases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
1.5.11 Limited Mutual Solubility – Ideal Henrian Solutions . . . . . . . . . . . . 29
1.5.12 Geometry of Binary Phase Diagrams . . . . . . . . . . . . . . . . . . . . . 31

ISBN: 3-527-30256-5
2 1 Thermodynamics and Phase Diagrams of Materials
1.6 Application of Thermodynamics to Phase Diagram Analysis . . . . . . 34

1.6.1 Thermodynamic/Phase Diagram Optimization . . . . . . . . . . . . . . . . 34
1.6.2 Polynomial Representation of Excess Properties . . . . . . . . . . . . . . . 34
1.6.3 Least-Squares Optimization . . . . . . . . . . . . . . . . . . . . . . . . . 35
1.6.4 Calculation of Metastable Phase Boundaries . . . . . . . . . . . . . . . . . 39
1.7 Ternary and Multicomponent Phase Diagrams . . . . . . . . . . . . . . 39
1.7.1 The Ternary Composition Triangle . . . . . . . . . . . . . . . . . . . . . . 39
1.7.2 Ternary Space Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
1.7.3 Polythermal Projections of Liquidus Surfaces . . . . . . . . . . . . . . . . 41
1.7.4 Ternary Isothermal Sections . . . . . . . . . . . . . . . . . . . . . . . . . 43
1.7.4.1 Topology of Ternary Isothermal Sections . . . . . . . . . . . . . . . . . . 45
1.7.5 Ternary Isopleths (Constant Composition Sections) . . . . . . . . . . . . . 46
1.7.5.1 Quasi-Binary Phase Diagrams . . . . . . . . . . . . . . . . . . . . . . . . 47
1.7.6 Multicomponent Phase Diagrams . . . . . . . . . . . . . . . . . . . . . . 47
1.7.7 Nomenclature for Invariant Reactions . . . . . . . . . . . . . . . . . . . . 49
1.7.8 Reciprocal Ternary Phase Diagrams . . . . . . . . . . . . . . . . . . . . . 49
1.8 Phase Diagrams with Potentials as Axes . . . . . . . . . . . . . . . . . . 51
1.9 General Phase Diagram Geometry . . . . . . . . . . . . . . . . . . . . . 56
1.9.1 General Geometrical Rules for All True Phase Diagram Sections . . . . . . 56
1.9.1.1 Zero Phase Fraction Lines . . . . . . . . . . . . . . . . . . . . . . . . . . 58
1.9.2 Choice of Axes and Constants of True Phase Diagrams . . . . . . . . . . . 58
1.9.2.1 Tie-lines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
1.9.2.2 Corresponding Phase Diagrams . . . . . . . . . . . . . . . . . . . . . . . 60
1.9.2.3 Theoretical Considerations . . . . . . . . . . . . . . . . . . . . . . . . . . 60
1.9.2.4 Other Sets of Conjugate Pairs . . . . . . . . . . . . . . . . . . . . . . . . 61
1.10 Solution Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
1.10.1 Sublattice Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
1.10.1.1 All Sublattices Except One Occupied by Only One Species . . . . . . . . . 62
1.10.1.2 Ionic Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
1.10.1.3 Interstitial Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
1.10.1.4 Ceramic Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
1.10.1.5 The Compound Energy Formalism . . . . . . . . . . . . . . . . . . . . . . 65
1.10.1.6 Non-Stoichiometric Compounds . . . . . . . . . . . . . . . . . . . . . . . 65
1.10.2 Polymer Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
1.10.3 Calculation of Limiting Slopes of Phase Boundaries . . . . . . . . . . . . 66
1.10.4 Short-Range Ordering . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
1.10.5 Long-Range Ordering . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
1.11 Calculation of Ternary Phase Diagrams From Binary Data . . . . . . . 72
1.12 Minimization of Gibbs Energy . . . . . . . . . . . . . . . . . . . . . . . 74
1.12.1 Phase Diagram Calculation . . . . . . . . . . . . . . . . . . . . . . . . . . 76
1.13 Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
1.13.1 Phase Diagram Compilations . . . . . . . . . . . . . . . . . . . . . . . . . 76
1.13.2 Thermodynamic Compilations . . . . . . . . . . . . . . . . . . . . . . . . 77
1.13.3 General Reading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
1.14 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
List of Symbols and Abbreviations 3
List of Symbols and Abbreviations
Symbol Designation
ai activity of component i
C number of components
cp molar heat capacity
E electrical potential of a galvanic cell
F degrees of freedom/variance
G Gibbs energy in J
g molar Gibbs energy in J/mol
gi partial molar Gibbs energy of i
Gi0 standard Gibbs energy of i
gi0 standard molar Gibbs energy of i
Dgi relative partial Gibbs energy i
gE excess molar Gibbs energy
giE excess partial Gibbs energy of i
DG Gibbs energy change
DG 0 standard Gibbs energy change
Dgm molar Gibbs energy of mixing
Dgf0 standard molar Gibbs energy of fusion
Dgv0 standard molar Gibbs energy of vaporization
H enthalpy in J
h molar enthalpy in J/mol
hi partial enthalpy of i
Hi0 standard enthalpy of i
hi0 standard molar enthalpy of i
Dhi relative partial enthalpy of i
hE excess molar enthalpy
hEi excess partial enthalpy of i
DH enthalpy change
DH 0 standard enthalpy change
Dhm molar enthalpy of mixing
Dh0f standard molar enthalpy of fusion
Dhv0 standard molar enthalpy of vaporization
K equilibrium constant
kB Boltzmann constant
n number of moles
ni number of moles of constituent i
Ni number of particles of i
N0 Avogadro’s number
pi partial pressure of i
P total pressure
P number of phases
qi general extensive variable
R gas constant
S entropy in J/K
s molar entropy in J/mol K
si partial entropy of i
Si0 standard entropy of i
si0 standard molar entropy of i
Dsi0 relative partial entropy of i
sE excess molar entropy
siE excess partial entropy of i
DS entropy change
DS 0 standard entropy change
Dsm molar entropy of mixing
Ds0f standard molar entropy of fusion
Dsv0 standard molar entropy of vaporization
T temperature
Tf temperature of fusion
Tc critical temperature
TE eutectic temperature
U internal energy
vi molar volume of i
vi0 standard molar volume of i
Xi mole fraction of i
Z coordination number
gi activity coefficient of i
e bond energy
h empirical entropy parameter
mi chemical potential of i
n number of moles of “foreign” particles contributed by a mole of solute
x molar metal ratio
s vibrational bond entropy
fi generalized thermodynamic potential
w empirical enthalpy parameter
b.c.c. body-centered cubic

f.c.c. face-centered cubic
h.c.p. hexagonal close-packed
1.2 Gibbs Energy and Equilibrium 5
1.1 Introduction ergy. General computer programs are avail-

able for the minimization of the Gibbs en-
An understanding of thermodynamics ergy in systems of any number of phases,
and phase diagrams is fundamental and es- components and species as outlined in Sec.
sential to the study of materials science. A 1.12. When coupled to extensive databases
knowledge of the equilibrium state under a of the thermodynamic properties of com-
given set of conditions is the starting point pounds and multicomponent solutions,
in the description of any phenomenon or these provide a powerful tool in the study
process. of materials science.
The theme of this chapter is the relation-
ship between phase diagrams and thermo- 1.1.1 Notation
dynamics. A phase diagram is a graphical
Extensive thermodynamic properties are
representation of the values of thermody-
represented by upper case symbols. For ex-
namic variables when equilibrium is estab-
ample, G = Gibbs energy in J. Molar prop-
lished among the phases of a system. Mate-
erties are represented by lower case sym-
rials scientists are used to thinking of phase
bols. For example, g = G/n = molar Gibbs
diagrams as plots of temperature versus com-
energy in J/mol where n is the total number
position. However, many other variables
of moles in the system.
such as total pressure and partial pressures
may be plotted on phase diagrams. In Sec.
1.3, for example, predominance diagrams 1.2 Gibbs Energy and Equilibrium
will be discussed, and in Sec. 1.8 chemical
potential – composition phase diagrams will 1.2.1 Gibbs Energy
be presented. General rules regarding phase
diagram geometry are given in Sec. 1.9. The Gibbs energy of a system is defined
In recent years, a quantitative coupling in terms of its enthalpy, H, entropy, S, and
of thermodynamics and phase diagrams temperature, T:
has become possible. With the use of com- G = H –TS (1-1)
puters, simultaneous optimizations of ther-
modynamic and phase equilibrium data can A system at constant temperature and pres-
be applied to the critical evaluation of bi- sure will approach an equilibrium state that
nary and ternary systems as shown in Sec. minimizes G.
1.6. This approach often enables good esti- As an example, consider the question of
mations to be made of the thermodynamic whether silica fibers in an aluminum ma-
properties and phase diagrams of multi- trix at 500 °C will react to form mullite,
component systems as discussed in Sec. Al6Si2O13
1.11. These estimates are based on structu- 13 9
ral models of solutions. Various models SiO 2 + 6 Al = Si + Al 6 Si 2 O13 (1-2)
2 2
such as the regular solution model, the sub-
lattice model, and models for interstitial If the reaction proceeds with the formation
solutions, polymeric solutions, solutions of of dn moles of mullite then, from the stoi-
defects, ordered solutions, etc. are dis- chiometry of the reaction, dnSi = (9/2) dn,
cussed in Secs. 1.5 and 1.10. dnAl = – 6 dn, and dnSiO2 = – 13/2 dn. Since
The equilibrium diagram is always cal- the four substances are essentially immis-
culated by minimization of the Gibbs en- cible at 500 °C, we need consider only the
standard molar Gibbs energies, Gi0. The action continues with the production of
Gibbs energy of the system then varies as: H2S, pH2S will increase while pH2 and pS2
will decrease. As a result, DG will become
9 0 13 0
0
dG / dn = g Al 6 Si 2 O13
+g − g progressively less negative. Eventually an
2 Si 2 SiO2 equilibrium state will be reached when
− 6 g Al
0
= DG 0 = − 830 kJ (1-3) dG/dn = DG = 0.
where DG 0 is the standard Gibbs energy For the equilibrium state, therefore:
change of reaction, Eq. (1-2), at 500 °C. DG 0 = – R T ln K (1-7)
Since DG 0 < 0, the formation of mullite 2 – 2 –1
= – R T ln ( pH2S pH2 pS2 )equilibrium
entails a decrease in G. Hence, the reaction
will proceed spontaneously so as to mini- where K, the “equilibrium constant” of the
mize G. reaction, is the one unique value of the ra-
tio ( p2H2S pH– 22 pS–12 ) for which the system will
1.2.2 Chemical Equilibrium be in equilibrium at the temperature T.
If the initial partial pressures are such
The partial molar Gibbs energy of an that DG > 0, then the reaction, Eq. (1-5),
ideal gas is given by: will proceed to the left in order to minimize
gi = gi0 + R T ln pi (1-4) G until the equilibrium condition of Eq.
(1-7) is attained.
where gi0 is the standard molar Gibbs en- As a further example, we consider the
ergy (at 1 bar), pi is the partial pressure in possible precipitation of graphite from a
bar, and R is the gas constant. The second gaseous mixture of CO and CO2 . The reac-
term in Eq. (1-4) is entropic. As a gas ex- tion is:
pands at constant T, its entropy increases.
Consider a gaseous mixture of H2 , S2 2 CO = C + CO2 (1-8)
and H2S with partial pressures pH2 , pS2 and Proceeding as above, we can write:
pH2S . The gases can react according to
dG/dn = gC + gCO2 – 2 gCO
2 H 2 + S 2 = 2 H 2S (1-5)
= ( gC0 + gCO
0
2
0
– 2 gCO 2
) + R T ln ( pCO2 /pCO )
If the reaction, Eq. (1-5), proceeds to the 0
= DG + R T ln ( pCO2 /pCO ) 2
(1-9)
right with the formation of 2 dn moles of = DG = – R T ln K + R T ln ( pCO2 /pCO 2
)
H2S, then the Gibbs energy of the system
2
varies as: If ( pCO2 /pCO ) is less than the equilibrium
constant K, then precipitation of graphite
dG/dn = 2 gH2S – 2 gH2 – gS2 will occur in order to decrease G.
= (2 gH0 2S – 2 gH0 2 – gS02 ) Real situations are, of course, generally
+ R T (2 ln pH2S – 2 ln pH2 – ln pS2 ) more complex. To treat the deposition of
= DG 0 + R T ln ( p2H2S pH– 22 pS–12 ) solid Si from a vapour of SiI4 , for example,
we must consider the formation of gaseous
= DG (1-6)
I2 , I and SiI2 so that three reaction equa-
DG, which is the Gibbs energy change of tions must be written:
the reaction, Eq. (1-5), is thus a function of SiI4 (g) = Si (sol) + 2 I2 (g) (1-10)
the partial pressures. If DG < 0, then the re-
action will proceed to the right so as to SiI4 (g) = SiI2 (g) + I2 (g) (1-11)
minimize G. In a closed system, as the re- I2 (g) = 2 I (g) (1-12)
1.3 Predominance Diagrams 7
The equilibrium state, however, is still that solid compounds of Cu, S and O. For ex-
which minimizes the total Gibbs energy of ample, at point Z, where pSO2 = 10– 2 and
the system. This is equivalent to satisfying pO2 = 10–7 bar, the stable phase is Cu2O.
simultaneously the equilibrium constants The conditions for coexistence of two and
of the reactions, Eqs. (1-10) to (1-12), as three solid phases are indicated respectively
will be shown in Section 1.12 where this by the lines and triple points on the diagram.
example is discussed further. For example, along the univariant line
(phase boundary) separating the Cu2O and
CuSO4 domains the equilibrium constant
1.3 Predominance Diagrams –2
K = pSO 2
pO– 23/2 of the following reaction is
satisfied:
1.3.1 Calculation of Predominance
Diagrams Cu2O + 2 SO2 + –23 O2 = 2 CuSO4 (1-13)
Predominance diagrams are a particu- Hence, along this line:

(1-14)
larly simple type of phase diagram which 3
log K = – 2 log pSO2 – – log pO2 = constant
2
have many applications in the fields of hot
corrosion, chemical vapor deposition, etc. This boundary is thus a straight line with a
Furthermore, their construction clearly il- slope of (– 3/2)/2 = – 3/4.
lustrates the principles of Gibbs energy In constructing predominance diagrams,
minimization and the Gibbs Phase Rule. we define a “base element”, in this case Cu,
A predominance diagram for the Cu – which must be present in all the condensed
S – O system at 1000 K is shown in Fig. phases. Let us further assume that there is
1-1. The axes are the logarithms of the no mutual solubility among the condensed
partial pressures of SO2 and O2 in the phases.
gas phase. The diagram is divided into Following the procedure of Bale et al.
areas or domains of stability of the various (1986), we formulate a reaction for the for-
Figure 1-1. Predomi-

nance diagram. log pSO2
versus log pO2 (bar) at
1000 K for the Cu – S – O
system (Bale et al., 1986).
mation of each solid phase, always from Fig. 1-3. For the formation reaction:
one mole of the base element Cu, and in-
4 Cu + O2 = 2 Cu2O (1-19)
volving the gaseous species whose pres-
sures are used as the axes (SO2 and O2 in we can write:
this example):
DG 0 = – R T ln K = R T ln ( pO2 )equilibrium
1
Cu + – O2 = CuO
2 = DH 0 – T DS 0 (1-20)
DG = DG 0 + R T ln pO–1/2 (1-15)
2 The diagonal line in Fig. 1-3 is thus a plot
1 1 of the standard Gibbs energy of formation
Cu + – O2 = – Cu2O
4 2
DG = DG 0 + R T ln pO–1/4 (1-16) of Cu2O versus T. The temperatures indi-
cated by the symbol M and 씲
2
M are the melt-
Cu + SO2 = CuS + O2 ing points of Cu and Cu2O respectively.
DG = DG 0 + R T ln ( pO2 pSO
–1
2
) (1-17) This line is thus simply a line taken from
the well-known Ellingham Diagram or
Cu + SO2 + O2 = CuSO4
DG 0 vs. T diagram for the formation of
DG = DG 0 + R T ln ( pSO
–1
2
pO–12 ) (1-18) oxides. However, by drawing vertical lines
and similarly for the formation of Cu2S, at the melting points of Cu and Cu2O as
Cu2SO4 and Cu2SO5 . shown in Fig. 1-3, we convert the plot to a
The values of DG 0 are obtained from ta- true phase diagram. Stability domains for
bles of thermodynamic properties. For any Cu (sol), Cu (l), Cu2O (sol), and Cu2O (l)
given values of pSO2 and pO2 , DG for each are shown as functions of T and of imposed
formation reaction can then be calculated. pO2 . The lines and triple points indicate
The stable compound is simply the one
with the most negative DG. If all the DG
values are positive, then pure Cu is the
stable compound.
By reformulating Eqs. (1-15) to (1-18) in
terms of, for example, S2 and O2 rather
than SO2 and O2 , a predominance diagram
with ln pS2 and ln pO2 as axes can be con-
structed. Logarithms of ratios or products
of partial pressures can also be used as
axes.
1.3.2 Ellingham Diagrams

as Predominance Diagrams
Rather than keeping the temperature
constant, we can use it as an axis. Figure
1-2 shows a diagram for the Fe – S – O sys-
tem in which R T ln pO2 is plotted versus T
at constant pSO2 = 1 bar. The diagram is of
the same topological type as Fig. 1-1.
A similar phase diagram of R T ln pO2 Figure 1-2. Predominance diagram. RT ln pO2 ver-
versus T for the Cu – O system is shown in sus T at pSO2 = 1.0 bar for the Fe – S – O system.
1.4 Thermodynamics of Solutions 9
Figure 1-3. Predominance diagram

(also known as a Gibbs energy-tem-
perature diagram or Ellingham dia-
gram) for the Cu – O system. Points
M and 앮M represent the melting
points of the metal and oxide re-
spectively.
conditions of two- and three-phase equilib- Predominance diagrams have found

rium. many applications in the fields of hot cor-
rosion, roasting of ores, chemical vapor
1.3.3 Discussion of Predominance deposition, etc. A partial bibliography on
Diagrams their construction and applications includes
In this section discussion is limited to Yokokowa (1999), Bale (1990), Bale et al.
the assumption that there is no mutual sol- (1986), Kellogg and Basu (1960), Ingra-
ubility among the condensed phases. The ham and Kellogg (1963), Pehlke (1973),
calculation of predominance phase dia- Garrels and Christ (1965), Ingraham and
grams in which mutual solubility is taken Kerby (1967), Pilgrim and Ingraham
into account is treated in Sec. 1.9, where (1967), Gulbransen and Jansson (1970),
the general geometrical rules governing Pelton and Thompson (1975), Shatynski
predominance diagrams and their relation- (1977), Stringer and Whittle (1975), Spen-
ship to other types of phase diagrams are cer and Barin (1979), Chu and Rahmel
discussed. (1979), and Harshe and Venkatachalam
We frequently encounter predominance (1984).
diagrams with domains for solid, liquid,
and even gaseous compounds which have
been calculated as if the compounds were
1.4 Thermodynamics of Solutions
immiscible, even though they may actually
1.4.1 Gibbs Energy of Mixing
be partially or even totally miscible. The
boundary lines are then no longer phase Liquid gold and copper are completely
boundaries, but are lines separating regions miscible at all compositions. The Gibbs en-
in which one species “predominates”. The ergy of one mole of liquid solution, g l, at
well known E – pH or Pourbaix diagrams of 1400 K is drawn in Fig. 1-4 as a function of
aqueous chemistry are examples of such composition expressed as mole fraction,
predominance diagrams. XCu , of copper. Note that XAu = 1 – XCu . The
Predominance diagrams may also be con- curve of g l varies between the standard mo-
structed when there are two or more base lar Gibbs energies of pure liquid Au and
0 0
elements, as discussed by Bale (1990). Cu, gAu and gCu .
nent i, and the derivative is taken with all

nj ( j9i) constant.
In the example of the Au – Cu binary liq-
uid solution, g Cu = (∂G l /∂nCu )T, P, nAu , where
G l = (nCu + nAu ) g l. That is, g Cu , which has
units of J/mol, is the rate of change of the
Gibbs energy of a solution as Cu is added.
It can be seen that g Cu is an intensive prop-
erty of the solution which depends upon
the composition and temperature but not
upon the total amount of solution. That is,
adding dnCu moles of copper to a solution
of given composition will (in the limit as
dnCu Æ 0) result in a change in Gibbs en-
ergy, dG, which is independent of the total
Figure 1-4. Molar Gibbs energy, g l, of liquid Au – mass of the solution.
Cu alloys at constant temperature (1400 K) illustrat- The reason that this property is called a
ing the tangent construction. chemical potential is illustrated by the fol-
lowing thought experiment. Imagine two
systems, I and II, at the same temperature
l
The function Dgm shown on Fig. 1-4 is and separated by a membrane that permits
called the molar Gibbs energy of mixing of only the passage of copper. The chemical
the liquid solution. It is defined as: potentials of copper in systems I and II
l
Dgm = g l – (XAu gAu
0 0
+ XCu gCu ) (1-21) are gCu I
= ∂ G I /∂nCuI
and gCu II
= ∂ G II /∂nCu
II
.
Copper is transferred across the membrane,
l
It can be seen that Dgm is the Gibbs energy with dn I = – dn II. The change in the total
change associated with the isothermal mix- Gibbs energy accompanying this transfer
ing of XAu moles of pure liquid Au and XCu is then: (1-24)
moles of pure liquid Cu to form one mole
of solution: dG = d (G I + G II ) = – (gCu
I II
– gCu II
) dnCu
I II
XAu Au (l) + XCu Cu (l) If gCu > gCu , then d (G I + G II ) is negative
II
when dnCu is positive. That is, the total
= 1 mole liquid solution (1-22)
Gibbs energy will be decreased by a trans-
Note that for the solution to be stable it is fer of Cu from system I to system II.
l
necessary that Dgm be negative. Hence, Cu will be transferred spontane-
ously from a system of higher g Cu to a sys-
1.4.2 Chemical Potential tem of lower g Cu . Therefore g Cu is called
the chemical potential of copper.
The partial molar Gibbs energy of com- An important principle of phase equilib-
ponent i, gi , also known as the chemical rium can now be stated. When two or more
potential, m i , is defined as: phases are in equilibrium, the chemical po-
gi = m i = (∂G/∂ni )T, P, n j (1-23) tential of any component is the same in all
phases.
where G is the Gibbs energy of the solu-
tion, ni is the number of moles of compo-
1.4.3 Tangent Construction 1.4.4 Gibbs–Duhem Equation

An important construction is illustrated Differentiation of Eq. (1-27) yields:
in Fig. 1-4. If a tangent is drawn to the
dg l = (XAu dgAu + XCu dgCu )
curve of g l at a certain composition
(XCu = 0.6 in Fig. 1-4), then the intercepts + (gAu dXAu + gCu dXCu) (1-30)
of this tangent on the axes at XAu = 1 and
Comparison with Eq. (1-28) then gives the
XCu = 1 are equal to g Au and g Cu respec-
Gibbs–Duhem equation at constant T and P:
tively at this composition.
To prove this, we first consider that the XAu dgAu + XCu dgCu = 0 (1-31)
Gibbs energy of the solution at constant T
and P is a function of nAu and nCu . Hence:
1.4.5 Relative Partial Properties
⎛ ∂G ⎞ l
⎛ ∂G ⎞ l
dGTl , P = ⎜ ⎟ dnAu + ⎜ ⎟ dnCu The difference between the partial Gibbs
⎝ ∂nAu ⎠ ⎝ ∂nCu ⎠ energy gi of a component in solution and
= g Au dnAu + gCu dnCu (1-25) the partial Gibbs energy gi0 of the same
component in a standard state is called the
Eq. (1-25) can be integrated as follows:
relative partial Gibbs energy (or relative
Gl nAu nCu chemical potential ), Dgi . It is most usual to
∫ dG l = ∫ g Au dnAu + ∫ gCu dnCu choose as standard state the pure compo-
0 0 0 nent in the same phase at the same temper-
G = g Au nAu + gCu nCu
l
(1-26 ) ature. The activity ai of the component rel-
ative to the chosen standard state is then
where the integration is performed at con- defined in terms of Dgi by the following
stant composition so that the intensive equation, as illustrated in Fig. 1-4.
properties g Au and g Cu are constant. This
integration can be thought of as describing Dgi = gi – gi0 = m i – m 0i = R T ln ai (1-32)
a process in which a pre-mixed solution of
constant composition is added to the sys- Note that gi and m i are equivalent symbols,
tem, which initially contains no material. as are gi0 and m 0i , see Eq. (1-23).
Dividing Eqs. (1-26) and (1-25) by From Fig. 1-4, it can be seen that:
(nAu + nCu ) we obtain expressions for the Dgm = XAu DgAu + XCu DgCu
molar Gibbs energy and its derivative:
= R T (XAu ln aAu + XCu ln aCu ) (1-33)
l
g = XAu gAu + XCu gCu (1-27)
The Gibbs energy of mixing can be di-
and vided into enthalpy and entropy terms, as
can the relative partial Gibbs energies:
dg l = gAu dXAu + gCu dXCu (1-28)
Since dXAu = – dXCu , it can be seen that Dgm = Dhm – T D sm (1-34)
Eqs. (1-27) and (1-28) are equivalent to the Dgi = Dhi – T D si (1-35)
tangent construction shown in Fig. 1-4.
These equations may also be rearranged Hence, the enthalpy and entropy of mixing
to give the following useful expression for may be expressed as:
a binary system: Dhm = XAu DhAu + XCu DhCu (1-36)
gi = g + (1 – Xi ) dg/dXi (1-29) D sm = XAu D sAu + XCu DsCu (1-37)
and tangent constructions similar to that of For the reaction, Eq. (1-38), at 1400 K,
Fig. 1-4 can be used to relate the relative DG 0 = – 68.35 kJ (Barin et al., 1977). The
partial enthalpies and entropies Dhi and activity of Cu in the liquid alloy at
Dsi to the integral molar enthalpy of mix- XCu = 0.6 is aCu = 0.43 (Hultgren et al.,
ing Dhm and integral molar entropy of mix- 1973). Substitution into Eq. (1-39) with
ing Dsm respectively. pO2 = 10– 4 bar gives:
dG/dn = DG = – 50.84 kJ
1.4.6 Activity Hence under these conditions the reaction
entails a decrease in the total Gibbs energy
The activity of a component in a solution
and so the copper will be oxidized.
was defined by Eq. (1-32).
Since ai varies monotonically with gi it
1.4.7 Ideal Raoultian Solutions
follows that when two or more phases are
in equilibrium the activity of any compo- An ideal solution or Raoultian solution
nent is the same in all phases, provided that is usually defined as one in which the ac-
the activity in every phase is expressed tivity of a component is equal to its mole
with respect to the same standard state. fraction:
The use of activities in calculations of
aiideal = Xi (1-40)
chemical equilibrium conditions is illus-
trated by the following example. A liquid (With a judicious choice of standard state,
solution of Au and Cu at 1400 K with this definition can also encompass ideal
XCu = 0.6 is exposed to an atmosphere in Henrian solutions, as discussed in Sec.
which the oxygen partial pressure is 1.5.11.)
pO2 = 10– 4 bar. Will Cu2O be formed? The However, this Raoultian definition of
reaction is: ideality is generally only useful for simple
substitutional solutions. There are more
2 Cu (l) + –12 O2 (g) = Cu2O (sol) (1-38) useful definitions for other types of solu-
where the Cu (l) is in solution. If the reactions such as interstitial solutions, ionic so-
tion proceeds with the formation of dn lutions, solutions of defects, polymer solu-
moles of Cu2O, then 2 dn moles of Cu are tions, etc. That is, the most convenient def-
consumed, and the Gibbs energy of the inition of ideality depends upon the solu-
Au – Cu solution changes by tion model. This subject will be discussed
in Sec. 1.10. In the present section, Eq.
– 2 (dG l /dnCu ) dn (1-40) for an ideal substitutional solution
will be developed with the Au – Cu solution
The total Gibbs energy then varies as: as example.
dG/dn = gCu2O – –12 gO2 – 2 (dG l /dnCu ) In the ideal substitutional solution model
it is assumed that Au and Cu atoms are
= gCu2O – –12 gO2 – 2 gCu nearly alike, with nearly identical radii and
0
= (gCu – –12 gO0 2 – 2 gCu
0
) electronic structures. This being the case,
2O
there will be no change in bonding energy
– –12 R T ln pO2 – 2 R T ln aCu or volume upon mixing, so that the en-
= DG 0 + R T ln ( pO–1/2 –2
aCu ) thalpy of mixing is zero:
2
ideal
= DG (1-39) Dhm =0 (1-41)
Furthermore, and for the same reason, the Hence Eq. (1-40) has been demonstrated
Au and Cu atoms will be randomly distrib- for an ideal substitutional solution.
uted over the lattice sites. (In the case of a
liquid solution we can think of the “lattice
1.4.8 Excess Properties
sites” as the instantaneous atomic positions.)
For a random distribution of NAu In reality, Au and Cu atoms are not iden-
gold atoms and NCu copper atoms over tical, and so Au – Cu solutions are not per-
(NAu + NCu ) sites, Boltzmann’s equation fectly ideal. The difference between a solu-
can be used to calculate the configurational tion property and its value in an ideal solu-
entropy of the solution. This is the entropy tion is called an excess property. The ex-
associated with the spatial distribution of cess Gibbs energy, for example, is defined
the particles: as:
(1-42)
S config = kB ln (NAu + NCu ) ! / NAu ! NCu ! gE = Dgm – Dgmideal (1-47)
where kB is Boltzmann’s constant. The con- Since the ideal enthalpy of mixing is zero,
figurational entropies of pure Au and Cu the excess enthalpy is equal to the enthalpy
are zero. Hence the configurational entropy of mixing:
of mixing, DS config, will be equal to S config.
hE = Dhm – Dhm
ideal
= Dhm (1-48)
Furthermore, because of the assumed close
similarity of Au and Cu, there will be no Hence:
non-configurational contribution to the en-
tropy of mixing. Hence, the entropy of gE = h E – T s E
mixing will be equal to S config. = Dhm – T s E (1-49)
Applying Stirling’s approximation, which Excess partial properties are defined simi-
states that ln N ! = [(N ln N) – N ] if N is larly:
large, yields:
giE = Dgi – Dgiideal
DSmideal = S config = − kB ( NAu + NCu ) (1- 43) = R T ln ai – R T ln Xi (1-50)
⎛ NAu NCu ⎞ siE = Dsi – Dsiideal = Dsi + R ln Xi (1-51)
× ⎜ NAu ln + NCu ln ⎟
⎝ NAu + NCu NAu + NCu ⎠
Also:
For one mole of solution, (NAu + NCu ) = N 0,
giE = hiE – T siE
where N 0 = Avogadro’s number. We also
note that (kB N 0 ) is equal to the ideal gas = Dhi – T siE (1-52)
constant R. Hence: Equations analogous to Eqs. (1-33),
(1-44)
DSmideal = – R (XAu ln XAu + XCu ln XCu ) (1-36) and (1-37) relate the integral and
partial excess properties. For example, in
Therefore, since the ideal enthalpy of mix- Au – Cu solutions:
ing is zero:
(1-45) gE = XAu g Au
E E
+ XCu g Cu (1-53)
Dgmideal = R T (XAu ln XAu + XCu ln XCu )
s E = XAu s Au
E E
+ XCu s Cu (1-54)
By comparing Eqs. (1-33) and (1-45) we
Tangent constructions similar to that of
obtain:
Fig. 1-4 can thus also be employed for ex-
Dgiideal = R T ln aiideal = R T ln Xi (1-46) cess properties, and an equation analogous
to Eq. (1-29) can be written: Eq. (1-50), Dgi < Dgiideal. That is, the com-
ponent i is more stable in the solution than
giE = gE + (1 – Xi ) dgE /dXi (1-55) it would be in an ideal solution of the same
The Gibbs – Duhem equation, Eq. (1-31), composition. If g i > 1, then giE > 0 and the
also applies to excess properties: driving force for the component to enter
into solution is less than in the case of an
E E
XAu dg Au + XCu dg Cu =0 (1-56) ideal solution.
In Au – Cu alloys, gE is negative. That is, 1.4.10 Multicomponent Solutions
Dgm is more negative than Dgmideal and so
the solution is thermodynamically more The equations of this section were de-
stable than an ideal solution. We say that rived with a binary solution as an example.
Au – Cu solutions exhibit negative devia- However, the equations apply equally to
tions from ideality. If gE > 0, then the solu- systems of any number of components. For
tion is less stable than an ideal solution and instance, in a solution of components
is said to exhibit positive deviations. A– B – C – D …, Eq. (1-33) becomes:
Dgm = XA DgA + XB DgB + XC DgC
1.4.9 Activity Coefficient + XD DgD + … (1-59)
The activity coefficient of a component
in a solution is defined as: 1.5 Binary Phase Diagrams
g i = ai /Xi (1-57)
1.5.1 Systems with Complete Solid
From Eq. (1-50): and Liquid Miscibility
giE = R T ln g i (1-58) The temperature – composition (T – X )

phase diagram of the CaO – MnO system is
In an ideal solution g i = 1 and giE = 0 for shown in Fig. 1-5 (Schenck et al., 1964;
all components. If g i < 1, then giE < 0 and by Wu, 1990). The abscissa is the composi-
Figure 1-5. Phase dia-

gram of the CaO – MnO
system at P = 1 bar
(after Schenck et al.,
1964, and Wu, 1990).
1.5 Binary Phase Diagrams 15
tion, expressed as mole fraction of MnO, (XMnO = 0.28). As the temperature is de-
XMnO . Note that XMnO = 1 – XCaO . Phase di- creased further, solid continues to precipi-
agrams are also often drawn with the com- tate with the compositions of the two
position axis expressed as weight percent. phases at any temperature being given by
At all compositions and temperatures in the liquidus and solidus compositions at
the area above the line labelled liquidus, a that temperature and with their relative
single-phase liquid solution will be ob- proportions being given by the lever rule.
served, while at all compositions and tem- Solidification is complete at 2030 °C, the
peratures below the line labelled solidus, last liquid to solidify having composition
there will be a single-phase solid solution. XMnO = 0.60 (point C).
A sample at equilibrium at a temperature The process just described is known as
and overall composition between these two equilibrium cooling. At any temperature
curves will consist of a mixture of solid during equilibrium cooling the solid phase
and liquid phases, the compositions of has a uniform (homogeneous) composition.
which are given by the liquidus and solidus In the preceding example, the composition
compositions at that temperature. For ex- of the solid phase during cooling varies
ample, a sample of overall composition along the line APC. Hence, in order for the
XMnO = 0.60 at T = 2200 °C (at point R in solid grains to have a uniform composition
Fig. 1-5) will consist, at equilibrium, of a at any temperature, diffusion of CaO from
mixture of liquid of composition XMnO = the center to the surface of the growing
0.70 (point Q) and solid of composition grains must occur. Since solid-state dif-
XMnO = 0.35 (point P). fusion is a relatively slow process, equi-
The line PQ is called a tie-line or co- librium cooling conditions are only ap-
node. As the overall composition is varied proached if the temperature is decreased
at 2200 °C between points P and Q, the very slowly. If a sample of composition
compositions of the solid and liquid phases XMnO = 0.60 is cooled very rapidly from the
remain fixed at P and Q, and only the rela- liquid, concentration gradients will be ob-
tive proportions of the two phases change. served in the solid grains, with the concen-
From a simple mass balance, we can derive tration of MnO increasing towards the sur-
the lever rule for binary systems: (moles of face from a minimum of XMnO = 0.28 (point
liquid) / (moles of solid) = PR /R Q. Hence, A) at the center. Furthermore, in this case
at 2200 °C a sample with overall composi- solidification will not be complete at
tion XMnO = 0.60 consists of liquid and solid 2030 °C since at 2030 °C the average con-
phases in the molar ratio (0.60 – 0.35) / centration of MnO in the solid particles
(0.70 – 0.60) = 2.5. Were the composition will be less than XMnO = 0.60. These con-
axis expressed as weight percent, then the siderations are discussed more fully in
lever rule would give the weight ratio of Chapter 2 of this volume (Müller-Krumb-
the two phases. haar et al., 2001).
Suppose that a liquid CaO – MnO solu- At XMnO = 0 and XMnO = 1 in Fig. 1-5 the
tion with composition XMnO = 0.60 is liquidus and solidus curves meet at the
cooled very slowly from an initial tempera- equilibrium melting points, or tempera-
ture of about 2500 °C. When the temperatures of fusion of CaO and MnO, which are
ture has decreased to the liquidus tempera- T f0(CaO) = 2572 °C, T f0(MnO) = 1842 °C.
ture 2270 °C (point B), the first solid The phase diagram is influenced by the
appears, with a composition at point A total pressure, P. Unless otherwise stated,
T – X diagrams are usually presented for Since, to a first approximation, Dhf0 and
P = const. = 1 bar. For equilibria involving Dsf0 are independent of T, Dgf0 is approxi-
only solid and liquid phases, the phase mately a linear function of T. If T > Tf0 ,
boundaries are typically shifted only by the then Dgf0 is negative. If T < Tf0 , then Dgf0 is
order of a few hundredths of a degree per positive. Hence, as seen in Fig. 1-6, as T
bar change in P. Hence, the effect of pres- decreases, the g s curve descends relative to
sure upon the phase diagram is generally g l. At 1500 °C, g l < g s at all compositions.
negligible unless the pressure is of the or- Therefore, by the principle that a system
der of hundreds of bars. On the other hand, always seeks the state of minimum Gibbs
if gaseous phases are involved then the ef- energy at constant T and P, the liquid phase
fect of pressure is very important. The ef- is stable at all compositions at 1500 °C.
fect of pressure will be discussed in Sec. At 1300 °C, the curves of g s and g l cross.
1.5.3. The line P1 Q1 , which is the common tan-
gent to the two curves, divides the compo-
sition range into three sections. For compo-
1.5.2 Thermodynamic Origin
sitions between pure Ge and P1 , a single-
of Phase Diagrams
phase liquid is the state of minimum Gibbs
In this section we first consider the ther- energy. For compositions between Q1 and
modynamic origin of simple “lens-shaped” pure Si, a single-phase solid solution is the
phase diagrams in binary systems with stable state. Between P1 and Q1 , a total
complete liquid and solid miscibility. Gibbs energy lying on the tangent line
An example of such a diagram was given P1 Q1 may be realized if the system adopts
in Fig. 1-5. Another example is the Ge – Si a state consisting of two phases with com-
phase diagram in the lowest panel of Fig. positions at P1 and Q1 and with relative
1-6 (Hansen, 1958). In the upper three pan- proportions given by the lever rule. Since
els of Fig. 1-6, the molar Gibbs energies of the tangent line P1 Q1 lies below both g s and
the solid and liquid phases, g s and g l, at g l, this two-phase state is more stable than
three temperatures are shown to scale. As either phase alone. Furthermore, no other
illustrated in the top panel, g s varies with line joining any point on g l to any point on
composition between the standard molar g s lies below the line P1 Q1 . Hence, this line
Gibbs energies of pure solid Ge and of pure represents the true equilibrium state of the
0 (s)
solid Si, g Ge and g 0Si(s), while g l varies system, and the compositions P1 and Q1 are
between the standard molar Gibbs energies the liquidus and solidus compositions at
0 (l)
of the pure liquid components g Ge and 1300 °C.
0 (l)
g Si . As T is decreased to 1100 °C, the points
0 (l)
The difference between g Ge and g 0Si(s) is of common tangency are displaced to
equal to the standard molar Gibbs energy higher concentrations of Ge. For T < 937 °C,
of fusion (melting) of pure Si, Dg f0(Si) = g s < g l at all compositions.
(g 0Si(l) – g 0Si(s) ). Similarly, for Ge, Dg f0(Ge) = It was shown in Fig. 1-4 that if a tangent
0 (l) 0 (s)
(g Ge – g Ge ). The Gibbs energy of fusion of is drawn to a Gibbs energy curve, then the
a pure component may be written as: intercept of this tangent on the axis at Xi = 1
is equal to the partial Gibbs energy or
Dgf0 = Dhf0 – T Dsf0 (1-60)
chemical potential g i of component i. The
where Dhf0 and Dsf0 are the standard molar common tangent construction of Fig. 1-6
enthalpy and entropy of fusion. thus ensures that the chemical potentials of
Figure 1-6. Ge – Si phase diagram at

P = 1 bar (after Hansen, 1958) and
Gibbs energy composition curves at
three temperatures, illustrating the com-
mon tangent construction (reprinted
from Pelton, 1983).
Ge and Si are equal in the solid and liquid If we rearrange Eq. (1-61), subtracting
phases at equilibrium. That is: the Gibbs energy of fusion of pure Ge,
l s Dg f0(Ge) = (g Ge
0 (l) 0 (s)
– g Ge ), from each side, we
g Ge = gGe (1-61)
get:
l s
g Si =g Si (1-62) l 0 (l) s 0 (s)
(g Ge – g Ge ) – (gGe – g Ge )
This equality of chemical potentials was 0 (l)
= – (g Ge 0 (s)
– g Ge ) (1-63)
shown in Sec. 1.4.2 to be the criterion for
phase equilibrium. That is, the common Using Eq. (1-32), we can write Eq. (1-63)
tangent construction simultaneously mini- as:
mizes the total Gibbs energy and ensures l
Dg Ge s
– DgGe = – Dg f0(Ge) (1-64)
the equality of the chemical potentials,
thereby showing that these are equivalent or
l s
criteria for equilibrium between phases. R T ln aGe – R T ln aGe = – Dg f0(Ge) (1-65)
l
where aGe is the activity of Ge (with re- Dgf0 = 0 when T = Tf0. Calculated phase dia-
spect to pure liquid Ge as standard state) in grams for Dsf0 = 3, 10 and 30 J/mol K are
the liquid solution on the liquidus, and aGes
shown in Fig. 1-7. A value of Dsf0 ≈ 10 is
is the activity of Ge (with respect to pure typical of most metals. However, when the
solid Ge as standard state) in the solid solu- components are ionic compounds such as
tion on the solidus. Starting with Eq. (1- ionic oxides, halides, etc., Dsf0 can be sig-
62), we can derive a similar expression for nificantly larger since there are several
the other component: ions per formula unit. Hence, two-phase
l s “lenses” in binary ionic salt or oxide phase
R T ln aSi – R T ln aSi = – Dg f0(Si) (1-66)
diagrams tend to be “fatter” than those
Eqs. (1-65) and (1-66) are equivalent to the encountered in alloy systems. If we are
common tangent construction. considering vapor – liquid equilibria rather
It should be noted that absolute values of than solid – liquid equilibria, then the
Gibbs energies cannot be defined. Hence, shape is determined by the entropy of
0 (l) 0 (l)
the relative positions of g Ge and g Si in vaporization, Dsv0 . Since Dsv0 is usually an
Fig. 1-6 are completely arbitrary. However, order of magnitude larger than Dsf0 , two-
this is immaterial for the preceding discus- phase (liquid + vapor) lenses tend to be
sion, since displacing both g 0Si(l) and g 0Si(s) by
0 (l)
the same arbitrary amount relative to g Ge
0 (s)
and g Ge will not alter the compositions of
the points of common tangency.
It should also be noted that in the present
discussion of equilibrium phase diagrams
we are assuming that the physical dimen-
sions of the single-phase regions in the
system are sufficiently large that surface
(interfacial) energy contributions to the
Gibbs energy can be neglected. For very
fine grain sizes in the sub-micron range,
however, surface energy effects can notice-
ably influence the phase boundaries.
The shape of the two-phase (solid + liq-
uid) “lens” on the phase diagram is deter-
mined by the Gibbs energies of fusion,
Dgf0 , of the components and by the mixing
terms, Dg s and Dg l. In order to observe
how the shape is influenced by varying
Dgf0 , let us consider a hypothetical system
A– B in which Dg s and Dg l are ideal Raoul-
tian (Eq. (1-45)). Let T f0(A) = 800 K and
T f0(B) = 1200 K. Furthermore, assume that
Figure 1-7. Phase diagram of a system A – B with
the entropies of fusion of A and B are equal
ideal solid and liquid solutions. Melting points of A
and temperature-independent. The enthalp- and B are 800 and 1200 K, respectively. Diagrams
ies of fusion are then given from Eq. (1-60) are calculated for entropies of fusion ∆Sf(A)
0
= ∆Sf(B)
0
=
by the expression Dhf0 = Tf0 Dsf0 since 3, 10 and 30 J/mol K.
very wide. For equilibria between two solid common tangent construction then gives
solutions of different crystal structure, the the equilibrium vapor and liquid composi-
shape is determined by the entropy of tions. The phase diagram depends upon the
solid – solid transformation, which is usu- Gibbs energies of vaporization of the com-
ally smaller than the entropy of fusion by ponents Dg v (Zn) and Dg v (Mg) as shown in
approximately an order of magnitude. Fig. 1-8.
Therefore two-phase (solid + solid) lenses To generate the isothermal pressure –
tend to be very narrow. composition (P – X ) phase diagram in the
lower panel of Fig. 1-8 we require the
1.5.3 Pressure – Composition Phase Gibbs energies of vaporization as functions
Diagrams of P. Assuming monatomic ideal vapors
and assuming that pressure has negligible
Let us consider liquid – vapor equilib-
effect upon the Gibbs energy of the liquid,
rium with complete miscibility, using as an
we can write:
example the Zn – Mg system. Curves of gv
and g l can be drawn at any given T and P, Dg v(i) = Dg0v(i) + R T ln P (1-67)
as in the upper panel of Fig. 1-8, and the
where Dg v(i) is the standard Gibbs energy
of vaporization (when P = 1 bar), which is
given by:
Dg0v(i) = Dh0v(i) – T Ds0v(i) (1-68)
For example, the enthalpy of vaporization

of Zn is Dh0v (Zn) = 115 300 J/mol at its nor-
mal boiling point of 1180 K (Barin et al.,
1977). Assuming that Dh0v is independent
of T, we calculate from Eq. (1-68) that
Ds0v (Zn) = 115 300 /1180 = 97.71 J/mol K.
From Eq. (1-67), Dg v (Zn) at any T and P is
thus given gy:
(1-69)
Dg v (Zn) = (115 300 – 97.71 T ) + R T ln P
A similar expression can be derived for the

other component Mg.
At constant temperature, then, the curve
of gv in Fig. 1-8 descends relative to gl as
the pressure is lowered, and the P – X phase
diagram is generated by the common tan-
gent construction. The diagram at 1250 K
in Fig. 1-8 was calculated under the as-
sumption of ideal liquid and vapor mixing
(gE (l) = 0, gE (v) = 0).
Figure 1-8. Pressure – composition phase diagram
of the Zn – Mg system at 1250 K calculated for ideal P – X phase diagrams involving liquid –
vapor and liquid solutions. Upper panel illustrates solid or solid – solid equilibria can be cal-
common tangent construction at a constant pressure. culated in a similar fashion through the fol-
lowing general equation, which gives the that the solid state is the stable state for
effect of pressure upon the Gibbs energy both pure components. However, in this
change for the transformation of one mole system gE (l) < gE (s), so that g s presents a
of pure component i from an a-phase to a flatter curve than does g l and there exists a
b-phase: central composition region in which gl < gs.
P Hence, there are two common tangent
b
Dga→ b = Dga0 → b + ∫ (vi − via ) dP (1- 70 ) lines, P1 Q1 and P2 Q2 . Such a situation
P =1 gives rise to a phase diagram with a mini-
where Dg 0aÆ b is the standard (P = 1 bar) mum in the two-phase region, as observed
Gibbs energy of transformation, and vib and in the Na2CO3 – K 2CO3 system (Dessu-
via are the molar volumes. reault et al., 1990) shown in Fig. 1-10. At a
composition and temperature correspond-
ing to the minimum point, liquid and solid
1.5.4 Minima and Maxima
of the same composition exist in equilib-
in Two-Phase Regions
rium.
As discussed in Sec. 1.4.8, the Gibbs en- A two-phase region with a minimum
ergy of mixing Dg m may be expressed as point as in Fig. 1-10 may be thought of as a
the sum of an ideal term Dgmideal and an ex- two-phase “lens” which has been “pushed
cess term gE. As has just been shown in down” by virtue of the fact that the liquid is
Sec. 1.5.2, if Dg ms and Dg ml
for the solid relatively more stable than the solid. Ther-
and liquid phases are both ideal, then a modynamically, this relative stability is ex-
“lens-shaped” two-phase region always re- pressed as gE (l) < gE (s).
sults. However in most systems even ap- Conversely, if gE (l) > gE (s) to a sufficient
proximately ideal behavior is the exception extent, then a two-phase region with a
rather than the rule. maximum will result. Such maxima in (liq-
Curves of g s and gl for a hypothetical uid + solid) or (solid + solid) two-phase re-
system A– B are shown schematically in gions are nearly always associated with the
Fig. 1-9 at a constant temperature (below existence of an intermediate phase, as will
the melting points of pure A and B) such be discussed in Sec. 1.5.10.
Figure 1-9. Isothermal Gibbs energy-composition Figure 1-10. Phase diagram of the K2CO3– Na2CO3
curves for solid and liquid phases in a system A – B in system at P = 1 bar (Dessureault et al., 1990).
which g E(l) > g E(s). A phase diagram of the type of
Fig. 1-10 results.
1.5.5 Miscibility Gaps
If gE > 0, then the solution is thermody-

namically less stable than an ideal solution.
This can result from a large difference in
size of the component atoms, ions or mole-
cules, which will lead to a (positive) lattice
strain energy, or from differences in elec-
tronic structure, or from other factors.
In the Au – Ni system, gE is positive in
the solid phase. In the top panel of Fig. 1-11,
gE (s) is plotted at 1200 K (Hultgren et al.,
1973) and the ideal Gibbs energy of
mixing, Dgmideal , is also plotted at 1200 K.
The sum of these two terms is the Gibbs
energy of mixing of the solid solution,
Dgms, which is plotted at 1200 K as well
as at other temperatures in the central panel
of Fig. 1-11. Now, from Eq. (1-45), Dgmideal
is always negative and varies directly
with T, whereas gE varies less rapidly with
temperature. As a result, the sum Dgms =
Dgmideal + gE becomes less negative as T de-
creases. However, the limiting slopes to the
Dgmideal curve at XAu = 1 and XNi = 1 are both
infinite, whereas the limiting slopes of gE
are always finite (Henry’s Law). Hence,
s
Dgm will always be negative as XAu Æ1
and XNi Æ1 no matter how low the temper-
ature. As a result, below a certain tempera-
s
ture the curve of Dgm will exhibit two neg-
ative “humps”. Common tangent lines Figure 1-11. Phase diagram (after Hultgren et al.,
1973) and Gibbs energy – composition curves of solid
P1 Q1 , P2 Q2 , P3 Q3 to the two humps at dif- solutions for the Au–Ni system at P = 1 bar. Letters
ferent temperatures define the ends of tie- “s” indicate spinodal points (Reprinted from Pelton,
lines of a two-phase solid – solid miscibility 1983).
gap in the Au – Ni phase diagram, which is
shown in the lower panel in Fig. 1-11
(Hultgren et al., 1973). The peak of the gap have less of an effect upon gE (l) in the liq-
occurs at the critical or consolute tempera- uid phase owing to the greater flexibility of
ture and composition, Tc and Xc . the liquid structure to accommodate differ-
When gE (s) is positive for the solid phase ent atomic sizes, valencies, etc. Hence, a
in a system it is usually also the case that solid – solid miscibility gap is often asso-
gE (l) < gE (s) since the unfavorable factors ciated with a minimum in the two-phase
(such as a difference in atomic dimensions) (solid + liquid) region, as is the case in the
which are causing gE (s) to be positive will Au – Ni system.
Below the critical temperature the curve gap at any temperature. Suppose that gE (s)
s
of Dgm exhibits two inflection points, indi- is so positve that Tc is higher than the min-
cated by the letter “s” in Fig. 1-11. These imum in the (solid + liquid) region. The re-
are known as the spinodal points. On the sult will be a phase diagram such as that of
phase diagram their locus traces out the the MgO – CaO system shown in Fig. 1-12
spinodal curve (Fig. 1-11). The spinodal (Doman et al., 1963; Wu, 1990).
curve is not part of the equilibrium phase The lower panel of Fig. 1-12 shows the
diagram, but it is important in the kinetics Gibbs energy curves at 2450 °C. The two
of phase separation, as discussed in Chap- common tangents define two two-phase re-
ter 6 (Binder and Fratzl, 2001). gions. As the temperature is decreased be-
low 2450 °C, the g s curve descends relative
to g l and the two points of tangency P1
1.5.6 Simple Eutectic Systems
and P2 approach each other until, at T =
The more positive gE is in a system, the 2374 °C, P1 and P2 become coincident at
higher is Tc and the wider is the miscibility the composition E. That is, at T = 2374 °C
Figure 1-12. Phase diagram

at P = 1 bar (after Doman et
al., 1963, and Wu, 1990) and
Gibbs energy – composition
curves at 2450 °C for the
MgO – CaO system. Solid
MgO and CaO have the
same crystal structure.
there is just one common tangent line con- the liquid has solidified, and during the re-
tacting the two portions of the g s curve at action the compositions of the three phases
compositions A and B and contacting the gl will remain fixed at A, B and E. For this
curve at E. This temperature is known as reason the eutectic reaction is called an in-
the eutectic temperature, TE , and the com- variant reaction. More details on eutectic
position E is the eutectic composition. For solidification may be found in Chapter 2
temperatures below TE , gl lies completely (Müller-Krumbhaar et al., 2001).
above the common tangent to the two por-
tions of the g s curve and so for T < TE a
1.5.7 Regular Solution Theory
solid – solid miscibility gap is observed.
The phase boundaries of this two-phase re- Many years ago Van Laar (1908) showed
gion are called the solvus lines. The word that the thermodynamic origin of a great
eutectic is from the Greek for “to melt many of the observed features of binary
well” since the system has its lowest melt- phase diagrams can be illustrated at least
ing point at the eutectic composition E. qualitatively by simple regular solution
This description of the thermodynamic theory. A simple regular solution is one for
origin of simple eutectic phase diagrams is which:
strictly correct only if the pure solid com-
gE = X A X B ( w – h T ) (1-72)
ponents A and B have the same crystal
structure. Otherwise, a curve for g s which where w and h are parameters independent
is continuous at all compositions cannot be of temperature and composition. Substitut-
drawn. ing Eq. (1-72) into Eq. (1-29) yields, for
Suppose a liquid MgO – CaO solution of the partial properties:
composition XCaO = 0.52 (composition P1) (1-73)
E 2 E 2
is cooled from the liquid state very slowly gA = XB (w – h T ) , gB = XA (w – h T )
under equilibrium conditions. At 2450 °C Several liquid and solid solutions con-
the first solid appears with composition Q1 . form approximately to regular solution be-
As T decreases further, solidification con- havior, particularly if gE is small. Examples
tinues with the liquid composition follow- may be found for alloys, molecular solu-
ing the liquidus curve from P1 to E and the tions, and ionic solutions such as molten
composition of the solid phase following salts and oxides, among others. (The very
the solidus curve from Q1 to A. The rela- low values of gE observed for gaseous solu-
tive proportions of the two phases at any T tions generally conform very closely to Eq.
are given by the lever rule. At a tempera- (1-72).)
ture just above TE , two phases are ob- To understand why this should be so, we
served: a solid of composition A and a liq- only need a very simple model. Suppose
uid of composition E. At a temperature just that the atoms or molecules of the compo-
below TE , two solids with compositions A nents A and B mix substitutionally. If the
and B are observed. Therefore, at TE , dur- atomic (or molecular) sizes and electronic
ing cooling, the following binary eutectic structures of A and B are similar, then the
reaction occurs: distribution will be nearly random, and the
liquid Æ solid1 + solid2 (1-71) configurational entropy will be nearly
ideal. That is:
Under equilibrium conditions the tempera-
ture will remain constant at T = TE until all gE ≈ Dhm – T S E (non-config) (1-74)
More will be said on this point in Sec. tively weak, then the solution is rendered
1.10.5. less stable, (w – h T ) > 0 and gE > 0.
We now assume that the bond energies Simple non-polar molecular solutions
eAA , e BB and eAB of nearest-neighbor pairs and ionic solutions such as molten salts of-
are independent of temperature and com- ten exhibit approximately regular behavior.
position and that the average nearest- The assumption of additivity of the energy
neighbor coordination number, Z, is also of pair bonds is probably reasonably realis-
constant. Finally, we assume that the en- tic for van der Waals or coulombic forces.
thalpy of mixing results mainly from the For alloys, the concept of a pair bond is, at
change in the total energy of nearest-neigh- best, vague, and metallic solutions tend to
bor pair bonds. exhibit larger deviations from regular be-
In one mole of solution there are (N 0 havior.
Z / 2) neareast-neighbor pair bonds, where In several solutions it is found that
N 0 is Avogadro’s number. Since the distri- | h T | < | w | in Eq. (1-72). That is, gE ≈ Dhm
bution is assumed random, the probability = w XA XB , and to a first approximation
that a given bond is an A–A bond is equal gE is independent of T. This is more often
to X A2 . The probabilities of B – B and A– B the case in non-metallic solutions than in
bonds are, respectively, X B2 and 2 XA XB . metallic solutions.
The molar enthalpy of mixing is then equal
to the sum of the energies of the nearest-
1.5.8 Thermodynamic Origin
neighbor bonds in one mole of solution,
of Simple Phase Diagrams Illustrated
minus the energy of the A–A bonds in XA
by Regular Solution Theory
moles of pure A and the energy of the B – B
bonds in XB moles of pure B: Figure 1-13 shows several phase dia-
grams, calculated for a hypothetical system
Dhm = (N 0 Z /2)
A– B containing a solid and a liquid phase
¥ (X A2 eAA + X B2 e BB + 2 XAB eAB ) with melting points of T f0(A) = 800 K and
– (N Z /2) (XA eAA) – (N 0 Z /2) (XB e BB )
0
T f0(B) = 1200 K and with entropies of fusion
= (N 0 Z ) [eAB – (eAA + e BB )/2] XA XB of both A and B set to 10 J/mol K, which is
= w XA XB (1-75) a typical value for metals. The solid and
liquid phases are both regular with temper-
We now define sAB , sAA and s BB as the ature-independent excess Gibbs energies
vibrational entropies of nearest-neighbor
pair bonds. Following an identical argu- gE (s) = w s XA XB and gE (l) = w l XA XB
ment to that just presented for the bond The parameters w s and w l have been varied
energies we obtain: systematically to generate the various pan-
s E (non-config) (1-76) els of Fig. 1-13.
0 In panel (n) both phases are ideal. Panels
= (N Z ) [sAB – (sAA + s BB )/2] = h XA XB
(l) to (r) exhibit minima or maxima de-
Eq. (1-72) has thus been derived. If A– B pending upon the sign and magnitude of
bonds are stronger than A–A and B – B (gE (l) – gE (s) ), as has been discussed in Sec.
bonds, then (eAB – h AB T ) < [(eAA – h AA T )/2 1.5.4. In panel (h) the liquid is ideal but
+ (e BB – h BB T ) / 2]. Hence, (w – h T ) < 0 positive deviations in the solid give rise to
and gE < 0. That is, the solution is rendered a solid – solid miscibility gap as discussed
more stable. If the A– B bonds are rela- in Sec. 1.5.6. On passing from panel (h) to
Figure 1-13. Topological changes in the phase diagram for a system A–B with regular solid and liquid phases,
brought about by systematic changes in the regular solution parameters ω s and ω l. Melting points of pure A and
B are 800 K and 1200 K. Entropies of fusion of both A and B are 10.0 J/mol K (Pelton and Thompson, 1975).
The dashed curve in panel (d) is the metastable liquid miscibility gap (Reprinted from Pelton, 1983).
panel (c), an increase in gE (s) results in a sultant lowering of the eutectic tempera-
widening of the miscibility gap so that the ture.
solubility of A in solid B and of B in solid Eutectic phase diagrams are often drawn
A decreases. Panels (a) to (c) illustrate that with the maximum solid solubility occur-
negative deviations in the liquid cause a ring at the eutectic temperature (as in Fig.
relative stabilization of the liquid with re- 1-12). However, panel (d) of Fig. 1-13, in
which the maximum solubility of A in the by sufficiently rapid cooling, then a meta-
B-rich solid solution occurs at approxi- stable liquid – liquid miscibility gap is ob-
mately T = 950 K, illustrates that this need served as shown in Fig. 1-13 (d). For exam-
not be the case even for simple regular so- ple, in the Na2O – SiO2 system the flattened
lutions. (or “S-shaped”) SiO2 liquidus heralds the
existence of a metastable miscibility gap of
importance in glass technology.
1.5.9 Immiscibility – Monotectics
In Fig. 1-13 (e), positive deviations in
1.5.10 Intermediate Phases
the liquid have given rise to a liquid – liquid
miscibility gap. The CaO – SiO2 system The phase diagram of the Ag – Mg
(Wu, 1990), shown in Fig. 1-14, exhibits system (Hultgren et al., 1973) is shown in
such a feature. Suppose that a liquid of Fig. 1-15 (d). An intermetallic phase, b¢, is
composition XSiO2 = 0.8 is cooled slowly seen centered approximately about the
from high temperatures. At T = 1815 °C the composition XMg = 0.5. The Gibbs energy
miscibility gap boundary is crossed and a curve at 1050 K for such an intermetallic
second liquid layer appears with a compo- phase has the form shown schematically in
sition of XSiO2 = 0.97. As the temperature is Fig. 1-15 (a). The curve g b¢ rises quite rap-
lowered further, the composition of each idly on either side of its minimum, which
liquid phase follows its respective phase occurs near XMg = 0.5. As a result, the b¢
boundary until, at 1692 °C, the SiO2-rich phase appears on the phase diagram only
liquid has a composition of XSiO2 = 0.99 over a limited composition range. This
(point B), and in the CaO-rich liquid form of the curve g b¢ results from the fact
XSiO2 = 0.74 (point A). At any temperature, that when XAg ≈ XMg a particularly stable
the relative amounts of the two phases are crystal structure exists in which Ag and Mg
given by the lever rule. atoms preferentially occupy different sites.
At 1692 °C the following invariant bi- The two common tangents P1 Q1 and P2 Q2
nary monotectic reaction occurs upon cool- give rise to a maximum in the two-phase
ing: (b¢ + liquid) region of the phase diagram.
(Although the maximum is observed very
Liquid B Æ Liquid A + SiO2 (solid) (1-77)
near XMg = 0.5, there is no thermodynamic
The temperature remains constant at reason for the maximum to occur exactly at
1692 °C and the compositions of the phases this composition.)
remain constant until all of liquid B is con- Another intermetallic phase, the e phase,
sumed. Cooling then continues with pre- is also observed in the Ag – Mg system,
cipitation of solid SiO2 with the equilib- Fig. 1-15. The phase is associated with a
rium liquid composition following the liq- peritectic invariant ABC at 744 K. The
uidus from point A to the eutectic E. Gibbs energy curves are shown schemati-
Returning to Fig. 1-13, we see in panel cally at the peritectic temperature in Fig.
(d) that the positive deviations in the liquid 1-15 (c). One common tangent line can be
in this case are not large enough to produce drawn to gl, g b¢ and g e.
immiscibility, but they do result in a flat- Suppose that a liquid alloy of composi-
tening of the liquidus, which indicates a tion XMg = 0.7 is cooled very slowly from
“tendency to immiscibility”. If the nuclea- the liquid state. At a temperature just above
tion of the solid phases can be suppressed 744 K a liquid phase of composition C and
Figure 1-14. CaO – SiO2 phase diagram at P = 1 bar (after Wu, 1990) and Gibbs energy curves at 1500 °C illus-
trating Gibbs energies of fusion and formation of the stoichiometric compound CaSiO3 .
This reaction occurs isothermally at 744 K

with all three phases at fixed compositions
(at points A, B and C). For an alloy with
overall composition between points A and
B the reaction proceeds until all the liquid
has been consumed. In the case of an alloy
with overall composition between B and C,
the b¢ phase will be the first to be com-
pletely consumed.
Peritectic reactions occur upon cooling
with formation of the product solid (e in
this example) on the surface of the reactant
solid (b¢), thereby forming a coating which
can prevent further contact between the re-
actant solid and liquid. Further reaction
may thus be greatly retarded so that equi-
librium conditions can only be achieved by
extremely slow cooling.
The Gibbs energy curve for the e phase,
g e, in Fig. 1-15 (c) rises more rapidly on ei-
ther side of its minimum than does the
Gibbs energy g b¢ for the b¢ phase in Fig. 1-
15 (a). As a result, the width of the single-
phase region over which the e phase exists
(sometimes called its range of stoichiome-
try or homogeneity range) is narrower than
for the b¢ phase.
In the upper panel of Fig. 1-14 for the
CaO – SiO2 system, Gibbs energy curves at
1500 °C for the liquid and CaSiO3 phases
are shown schematically. g 0.5 (CaSiO3 ) rises
extremely rapidly on either side of its min-
imum. (We write g 0.5 (CaSiO3 ) for 0.5 moles
of the compound in order to normalize to a
Figure 1-15. Ag – Mg phase diagram at P = 1 bar (af- basis of one mole of components CaO and
ter Hultgren at al., 1973) and Gibbs energy curves at
three temperatures.
SiO2 .) As a result, the points of tangency
Q1 and Q 2 of the common tangents P1 Q1
and P2 Q2 nearly (but not exactly) coincide.
a b¢ phase of composition A are observed Hence, the range of stoichiometry of the
at equilibrium. At a temperature just below CaSiO3 phase is very narrow (but never
744 K the two phases at equilibrium are b¢ zero). The two-phase regions labelled
of composition A and e of composition B. (CaSiO3 + liquid) in Fig. 1-14 are the two
The following invariant binary peritectic sides of a two-phase region that passes
reaction thus occurs upon cooling: through a maximum at 1540 °C just as the
Liquid + b¢ (solid) Æ e (solid) (1-78) (b¢ + liquid) region passes through a maxi-
mum in Fig. 1-15 (d). Because the CaSiO3 Eq. (1-78), to form a liquid of composition
single-phase region is so narrow, we refer C and another solid phase, b¢, of composi-
to CaSiO3 as a stoichiometric compound. tion A.
Any deviation in composition from the Another example of an incongruent
stoichiometric 1 : 1 ratio of CaO to SiO2 compound is Ca3Si2O7 in Fig. 1-14, which
results in a very large increase in Gibbs melts incongruently (or peritectically) to
energy. form liquid and Ca2SiO4 at the peritectic
The e phase in Fig. 1-15 is based on the temperature of 1469 °C.
stiochiometry AgMg3 . The Gibbs energy An incongruent compound is always as-
curve, Fig. 1-15 (c), rises extremely rapidly sociated with a peritectic. However, the
on the Ag side of the minimum, but some- converse is not necessarily true. A peritec-
what less steeply on the Mg side. As a re- tic is not always associated with an inter-
sult, Ag is virtually insoluble in AgMg3 , mediate phase. See, for example, Fig. 1-
while Mg is sparingly soluble. Such a 13 (i).
phase with a narrow range of homogeneity For purposes of phase diagram calcula-
is often called a non-stoichiometric com- tions involving stoichiometric compounds
pound. At low temperatures the b¢ phase such as CaSiO3 , we may, to a good approx-
exhibits a relatively narrow range of stoi- imation, consider the Gibbs energy curve,
chiometry about the 1 : 1 AgMg composi- g 0.5 (CaSiO3 ) , to have zero width. All that
tion and can properly be called a com- is then required is the value of g 0.5 (CaSiO3 )
pound. However, at higher temperatures it at the minimum. This value is usually
is debatable whether a phase with such a expressed in terms of the Gibbs energy
wide range of composition should be called of fusion of the compound, Dg 0f (0.5 CaSiO3 )
a “compound”. or the Gibbs energy of formation
From Fig. 1-14 it can be seen that if stoi- Dg 0form (0.5 CaSiO3 ) of the compound from
chiometric CaSiO3 is heated it will melt the pure solid components CaO and SiO2
isothermally at 1540 °C to form a liquid of according to the reaction: 0.5 CaO (sol) +
the same composition. Such a compound is 0.5 SiO2 (sol) = 0.5 CaSiO3 (sol). Both these
called congruently melting or simply a con- quantities are interpreted graphically in
gruent compound. The compound Ca2SiO4 Fig. 1-14.
in Fig. 1-14 is congruently melting. The b¢
phase in Fig. 1-15 is also congruently melt-
ing at the composition of the liquidus/sol- 1.5.11 Limited Mutual Solubility –
idus maximum. Ideal Henrian Solutions
It should be noted with regard to the con-
gruent melting of CaSiO3 in Fig. 1-14 that In Sec. 1.5.6, the region of two solids in
the limiting slopes dT /dX of both branches the MgO – CaO phase diagram of Fig. 1-12
of the liquidus at the congruent melting was described as a miscibility gap. That is,
point (1540 °C) are zero since we are really only one continuous g s curve was assumed.
dealing with a maximum in a two-phase re- If, somehow, the appearance of the liquid
gion. phase could be suppressed, then the two
The AgMg3 (e) compound in Fig. 1-15 is solvus lines in Fig. 1-12, when projected
said to melt incongruently. If solid AgMg3 upwards, would meet at a critical point
is heated it will melt isothermally at 744 K above which one continuous solid solution
by the reverse of the peritectic reaction, would exist at all compositions.
Such a description is justifiable only if g solvent ≈ 1

the pure solid components have the same g solute ≈ g solute
0
= constant (1-79)
crystal structure, as is the case for MgO
and CaO. However, consider the Ag – Mg (Care must be exercised for solutions other
system, Fig. 1-15, in which the terminal than simple substitutional solutions. Henry’s
(Ag) solid solution is face-centered-cubic Law applies only if the ideal activity is defined
and the terminal (Mg) solid solution is hex- correctly, as will be discussed in Sec. 1.10).
agonal-close-packed. In this case, one con- Treating, then, the h.c.p. (Mg) phase
tinuous curve for g s cannot be drawn. Each in the Ag – Mg system (Fig. 1-15 (b)) as a
solid phase must have its own separate Henrian solution we write:
Gibbs energy curve, as shown schemati-
g h.c.p. = (XAg g0Ag(f.c.c.) + XMg g0Mg
(h.c.p.)
)
cally in Fig. 1-15 (b) for the h.c.p. (Mg)
phase at 800 K. In this figure, g0Mg (h.c.p.)
and + R T (XAg ln aAg + XMg ln aMg )
0 (f.c.c.)
g Ag are the standard molar Gibbs ener- = (XAg g0Ag(f.c.c.) + XMg g0Mg
(h.c.p.)
) (1-80)
gies of pure h.c.p. Mg and pure f.c.c. Ag, 0
+ R T (XAg ln (g Ag XAg ) + XMg ln XMg )
while g0Ag(h.c.p.-Mg) is the standard molar
Gibbs energy of pure (hypothetical) h.c.p. 0
where aAg and g Ag are the activity and ac-
Ag in the h.c.p. (Mg) phase. tivity coefficient of silver with respect to
Since the solubility of Ag in the h.c.p. pure f.c.c. silver as standard state. Let us
(Mg) phase is limited we can, to a good ap- now combine terms as follows:
proximation, describe it as a Henrian ideal
solution. That is, when a solution is suffi- g h.c.p. = [XAg (g0Ag(f.c.c.) + R T ln g Ag
0
)
ciently dilute in one component, we can ap- + XMg g0Mg
(h.c.p.)
] (1-81)
proximate g Esolute = R T ln g solute by its value + R T (XAg ln XAg + XMg ln XMg )
in an infinitely dilute solution. That is, if
0
0
Xsolute is small we set g solute = g solute where Since g Ag is independent of composition,
0
g solute is the Henrian activity coefficient at let us define:
Xsolute = 0. Thus, for sufficiently dilute solu- 0
g0Ag(h.c.p.-Mg) = (g0Ag(f.c.c.) + R T ln g Ag ) (1-82)
tions we assume that g solute is independent
of composition. Physically, this means that From Eqs. (1-81) and (1-82) it can be seen
in a very dilute solution there is negligible that, relative to g0Mg (h.c.p.)
and to the hypothet-
0 (h.c.p.-Mg)
interaction among solute particles because ical standard state g Ag defined in
they are so far apart. Hence, each addi- this way, the h.c.p. solution is ideal. Eqs.
tional solute particle added to the solution (1-81) and (1-82) are illustrated in Fig. 1-
0
produces the same contribution to the ex- 15 (b). It can be seen that as g Ag becomes
cess Gibbs energy of the solution and so g E- larger, the point of tangency N moves
E
solute = dG /dnsolute = constant. to higher Mg concentrations. That is, as
From the Gibbs – Duhem equation, Eq. (g0Ag(h.c.p.-Mg) – g0Ag(f.c.c.) ) becomes more posi-
(1-56), if dg Esolute = 0, then dg Esolvent = 0. tive, the solubility of Ag in h.c.p. (Mg) de-
Hence, in a Henrian solution g solute is also creases.
constant and equal to its value in an infi- It must be stressed that g0Ag(h.c.p.-Mg) as de-
nitely dilute solution. That is, g solute = 1 and fined by Eq. (1-82) is solvent-dependent.
the solvent behaves ideally. In summary That is, g0Ag(h.c.p.-Mg) is not the same as, say,
then, for dilute solutions (Xsolvent ≈ 1) g0Ag(h.c.p.-Cd) for Ag in dilute h.c.p. (Cd) solid
Henry’s Law applies: solutions.
Henrian activity coefficients can usually This is the Gibbs Phase Rule. F is called
be expressed as functions of temperature: the number of degrees of freedom or vari-
ance of the system and is the number of pa-
R T ln g i0 = a – b T (1-83)
rameters which can and must be specified
where a and b are constants. If data are lim- in order to completely specify the state of
ited, it can further be assumed that b ≈ 0 so the system.
that R T ln g i0 ≈ constant. Binary temperature – composition phase
diagrams are plotted at a fixed pressure,
usually 1 bar. This then eliminates one de-
1.5.12 Geometry of Binary Phase gree of freedom. In a binary system, C = 2.
Diagrams Hence, for binary isobaric T – X diagrams
The geometry of all types of phase dia- the phase rule reduces to:
grams of any number of components is F=3–P (1-86)
governed by the Gibbs Phase Rule.
Consider a system with C components in Binary T – X diagrams contain single-
which P phases are in equilibrium. The phase areas and two-phase areas. In the sin-
system is described by the temperature, the gle-phase areas, F = 3 – 1 = 2. That is, tem-
total pressure and the composition of each perature and composition can be specified
phase. In a C-component system, (C – 1) in- independently. These regions are thus
dependent mole fractions are required to called bivariant. In two-phase regions,
describe the composition of each phase F = 3 – 2 = 1. If, say, T is specified, then the
(because S Xi = 1). Hence, the total number compositions of both phases are deter-
of variables required to describe the system mined by the ends of the tie-lines. Two-
is [P (C – 1) + 2]. However, as shown in Sec. phase regions are thus termed univariant.
1.4.2, the chemical potential of any compo- Note that the overall composition can be
nent is the same in all phases (a, b, g, …) varied within a two-phase region at con-
since the phases are in equilibrium. That is: stant T, but the overall composition is not a
parameter in the sense of the phase rule.
gia (T, P, X1a , X2a , X3a , …) Rather, it is the compositions of the indi-
= gib (T, P, X1b, X2b, X3b, …) vidual phases at equilibrium that are the
parameters to be considered in counting the
= gig(T, P, X1g, X2g, X3g, …) = … (1-84)
number of degrees of freedom.
where gia (T, P, X1a , X2a , X3a , …) is a func- When three phases are at equilibrium in
tion of temperature, of total pressure, and a binary system at constant pressure,
of the mole fractions X1a , X2a , X3a , … in F = 3 – 3 = 0. Hence, the compositions of
the a phase; and similarly for the other all three phases, as well as T, are fixed.
phases. Thus there are C (P – 1) indepen- There are two general types of three-phase
dent equations in Eq. (1-84) relating the invariants in binary phase diagrams. These
variables. are the eutectic-type and peritectic-type
Let F be the differences between the invariants as illustrated in Fig. 1-16. Let
number of variables and the number of the three phases concerned be called a, b
equations relating them: and g, with b as the central phase as shown
in Fig. 1-16. The phases a, b and g can be
F = P (C – 1) + 2 – C (P – 1) solid, liquid or gaseous. At the eutectic-
F=C–P+2 (1-85) type invariant, the following invariant re-
tectic reaction is l + s2 Æ s1 ; (ii) syntectics

(Fig. 1-13 (k)) in which a = liquid1 , b =
solid, g = liquid2 . The syntectic reaction is
l1 + l2 Æ s; (iii) peritectoids in which a =
solid1 , b = solid2 , g = solid3 . The peritec-
toid reaction is s1 + s3 Æ s2 .
An important rule of construction which
applies to invariants in binary phase dia-
grams is illustrated in Fig. 1-16. This ex-
tension rule states that at an invariant the
extension of a boundary of a two-phase re-
gion must pass into the adjacent two-phase
region and not into a single-phase region.
Examples of both correct and incorrect
constructions are given in Fig. 1-16. To
understand why the “incorrect extensions”
shown are not right consider that the (a + g)
phase boundary line indicates the composi-
tion of the g-phase in equilibrium with the
Figure 1-16. Some geometrical units of binary phase a-phase, as determined by the common
diagrams, illustrating rules of construction.
tangent to the Gibbs energy curves. Since
there is no reason for the Gibbs energy
curves or their derivatives to change dis-
action occurs isothermally as the system is continuously at the invariant temperature,
cooled: the extension of the (a + g) phase boundary
also represents the stable phase boundary
bÆa+g (1-87)
under equilibrium conditions. Hence, for
whereas at the peritectic-type invariant the this line to extend into a region labeled as
invariant reaction upon cooling is: single-phase g is incorrect.
Two-phase regions in binary phase dia-
a+gÆb (1-88)
grams can terminate: (i) on the pure com-
Some examples of eutectic-type invari- ponent axes (at XA = 1 or XB = 1) at a trans-
ants are: (i) eutectics (Fig. 1-12) in which formation point of pure A or B; (ii) at a
a = solid1 , b = liquid, g = solid2 ; the eutectic critical point of a miscibility gap; (iii) at an
reaction is l Æ s1 + s2 ; (ii) monotectics (Fig. invariant. Two-phase regions can also ex-
1-14) in which a = liquid1 , b = liquid2 , g = - hibit maxima or minima. In this case, both
solid; the monotectic reaction is l2 Æ l1 + s; phase boundaries must pass through their
(iii) eutectoids in which a= maximum or minimum at the same point as
solid1 , b = solid2 , g = solid3 ; the eutectoid shown in Fig. 1-16.
reaction is s2 Æ s1 + s3 ; (iv) catatectics in All the geometrical units of construction
which a = liquid, b = solid1 , g = solid2 ; the of binary phase diagrams have now been
catatectic reaction is s1 Æ l + s2 . discussed. The phase diagram of a binary
Some examples of peritectic-type invari- alloy system will usually exhibit several of
ants are: (i) peritectics (Fig. 1-15) in which these units. As an example, the Fe – Mo
a = liquid, b = solid1 , g = solid2 . The peri- phase diagram (Kubaschewski, 1982) is
Figure 1-17. Fe – Mo
phase diagram at P = 1 bar
(Kubaschewski, 1982).
shown in Fig. 1-17. The invariants in this and 2051 °C; a monotectic at 1692 °C; and
system are peritectics at 1540, 1488 and a peritectic at 1469 °C. The compound
1450 °C; eutectoids at 1235 and 1200 °C; Ca3SiO5 dissociates upon heating to CaO
peritectoids at 1370 and 950 °C. The two- and Ca2SiO4 by a peritectoid reaction at
phase (liquid + g) region passes through a 1789 °C and dissociates upon cooling to
minimum at XMo = 0.2. CaO and Ca2SiO4 by a eutectoid reaction at
Between 910 °C and 1390 °C is a two- 1250 °C. Maxima are observed at 2130 and
phase (a + g) g-loop. Pure Fe adopts the 1540 °C. At 1470 °C there is an invariant
f.c.c. g structure between these two temper- associated with the tridymite Æ cristobalite
atures but exists as the b.c.c. a phase at transition of SiO2 . This is either a peritec-
higher and lower temperatures. Mo, how- tic or a catatectic depending upon the rela-
ever, is more soluble in the b.c.c. than tive solubility of CaO in tridymite and cris-
0 (b.c.c.-Fe)
in the f.c.c. structure. That is, g Mo tobalite. However, these solubilities are
0 (f.c.c.-Fe)
< g Mo as discussed in Sec. 1.5.11. very small and unknown.
Therefore, small additions of Mo stabilize
the b.c.c. structure.
In the CaO – SiO2 phase diagram, Fig.
1-14, we observe eutectics at 1439, 1466
1.6 Application of important step in the development of data-

Thermodynamics to Phase bases for multicomponent systems.
Diagram Analysis
1.6.2 Polynomial Representation
1.6.1 Thermodynamic/Phase Diagram of Excess Properties
Optimization
Empirical equations are required to ex-
In recent years the development of press the excess thermodynamic properties
solution models, numerical methods and of the solution phases as functions of com-
computer software has permitted a quanti- position and temperature. For many simple
tative application of thermodynamics to binary substitutional solutions, a good rep-
phase diagram analysis. For a great many resentation is obtained by expanding the
systems it is now possible to perform a excess enthalpy and entropy as polynomi-
simultaneous critical evaluation of avail- als in the mole fractions XA and XB of the
able phase diagram measurements and of components:
available thermodynamic data (calorimet-
hE = XA XB [h0 + h1 (XB – XA ) (1-89)
ric data, measurements of activities, etc.) 2 3
with a view to obtaining optimized equa- + h2 (XB – XA) + h3 (XB – XA) + …]
tions for the Gibbs energies of each phase s E = XA XB [s0 + s1 (XB – XA ) (1-90)
which best represent all the data. These + s2 (XB – XA)2 + s3 (XB – XA)3 + …]
equations are consistent with thermody-
namic principles and with theories of solu- where the hi and si are empirical coeffi-
tion behavior. cients. As many coefficients are used as
The phase diagram can be calculated are required to represent the data in a
from these thermodynamic equations, and given system. For most systems it is a good
so one set of self-consistent equations de- approximation to assume that the coeffi-
scribes all the thermodynamic properties cients hi and si are independent of tempera-
and the phase diagram. This technique of ture.
analysis greatly reduces the amount of ex- If the series are truncated after the first
perimental data needed to fully character- term, then:
ize a system. All data can be tested for
g E = hE – T s E = XA XB (h0 – T s0 ) (1-91)
internal consistency. The data can be inter-
polated and extrapolated more accurately This is the equation for a regular solution
and metastable phase boundaries can be discussed in Sec. 1.5.7. Hence, the polyno-
calculated. All the thermodynamic proper- mial representation can be considered to be
ties and the phase diagram can be repre- an extension of regular solution theory.
sented and stored by means of a small set When the expansions are written in terms
of coefficients. of the composition variable (XB – XA ), as in
Finally, and most importantly, it is often Eqs. (1-89) and (1-90), they are said to be
possible to estimate the thermodynamic in Redlich – Kister form. Other equivalent
properties and phase diagrams of ternary polynomial expansions such as orthogonal
and higher-order systems from the assessed Legendre series have been discussed by
parameters for their binary sub-systems, as Pelton and Bale (1986).
will be discussed in Sec. 1.11. The analysis Differentiation of Eqs. (1-89) and (1-90)
of binary systems is thus the first and most and substitution into Eq. (1-55) yields the
1.6 Application of Thermodynamics to Phase Diagram Analysis 35
following expansions for the partial excess

enthalpies and entropies:
hAE = XB2 S
i=0
hi [(XB – XA )i
– 2 i XA (XB – XA )i – 1] (1-92)
hEB = XA2 S
i=0
hi [(XB – XA )i
+ 2 i XB (XB – XA )i – 1] (1-93)
sAE = XB2 S
i=0
si [(XB – XA )i
– 2 i XA (XB – XA )i – 1] (1-94)
sBE = XA2 S
i=0
si [(XB – XA )i
Figure 1-18. LiF – NaF phase diagram at P = 1 bar
+ 2 i XB (XB – XA )i – 1] (1-95) calculated from optimized thermodynamic parame-
ters (Sangster and Pelton, 1987). Points are experi-
Partial excess Gibbs energies, g Ei , are mental from Holm (1965). Dashed line is theoretical
then given by Eq. (1-52). limiting liquidus slope for negligible solid solubility.
Eqs. (1-89) and (1-90), being based upon
regular solution theory, give an adequate
representation for most simple substitu- each pure component at temperature T is
tional solutions in which deviations from given by:
ideal behavior are not too great. In other
cases, more sophisticated models are re- Dg f0 = Dhf0( Tf ) (1 − T / Tf )
quired, as discussed in Sec. 1.10. T
+ ∫ (cpl − cps ) (1 − 1/ T ) dT (1-96 )
Tf
1.6.3 Least-Squares Optimization
where Dh0f(Tf ) is the enthalpy of fusion at
Eqs. (1-89), (1-90) and (1-92) to (1-95) the melting point Tf , and cpl and c ps are the
are linear in terms of the coefficients. heat capacities of the pure liquid and solid.
Through the use of these equations, all The following values are taken from Barin
integral and partial excess properties (gE , et al. (1977):
hE , s E , g Ei , hEi , sEi ) can be expressed by
linear equations in terms of the one set of Dg f0(LiF) = 14.518 + 128.435 T
coefficients {hi , si }. It is thus possible to + 8.709 ¥ 10– 3 T 2 – 21.494 T ln T
include all available experimental data for – 2.65 ¥ 105 T –1 J/mol (1-97)
a binary phase in one simultaneous linear
Dg f0(NaF) = 10.847 + 156.584 T
least-squares optimization. Details have
been discussed by Bale and Pelton (1983), + 4.950 ¥ 10– 3 T 2 – 23.978 T ln T
Lukas et al. (1977) and Dörner et al. – 1.07 ¥ 105 T –1 J/mol (1-98)
(1980). Thermodynamic properties along the liq-
The technique of coupled thermody- uidus and solidus are related by equations
namic/phase diagram analysis is best illus- like Eqs. (1-64) and (1-65). Taking the
trated by examples. ideal activities to be equal to the mole frac-
The phase diagram of the LiF – NaF tions:
system is shown in Fig. 1-18. Data points
measured by Holm (1965) are shown on R T ln Xil – R T ln Xis + giE (l) – giE (s)
the diagram. The Gibbs energy of fusion of = – Dg f0(i) (1-99)
where i = LiF or NaF. Along the LiF-rich aLiF = 1 with respect to pure solid LiF as
liquidus, the liquid is in equilibrium with standard state. In the Henrian solution at
s
essentially pure solid LiF. Hence, X LiF =1 saturation,
E (s)
and gLif = 0. Eq. (1-99) then reduces to:
aLiF = g 0LiF XLiF = g 0LiF (1 – 0.915) = 1
l E (l) 0
R T ln X LiF +g Lif = – Dg f (LiF) (1-100)
Hence, the Henrian activity coefficient
l
From experimental values of X LiF on the in the NaF-rich solid solution at 649 °C
liquidus and with Eq. (1-97) for Dg f0(LiF) , is g 0LiF = 11.76. Since no solubilities have
E (l)
values of gLif at the measured liquidus been measured at other temperatures, we
points can be calculated from Eq. (1-100). assume that:
Along the NaF-rich solidus the solid so-
lution is sufficiently concentrated in NaF R T ln g 0LiF = R (922) ln (11.76) (1-104)
that Henrian behavior (Sec. 1.5.11) can be = 18 900 J/mol = constant
E (s)
assumed. That is, for the solvent, gNaF = 0. Using the notation of Eq. (1-82):
Hence, Eq. (1-99) becomes:
(l) (s) E (l)
g 0LiF
(s, NaF)
= g 0LiF
(s)
+ 18 900 J/mol (1-105)
R T ln X NaF – R T ln X NaF + gNaF
= – Dg f0(NaF) (1-101) where g 0LiF
(s)
is the standard Gibbs energy of
solid LiF, and g 0LiF (s, NaF)
is the hypothetical
Thus, from the experimental liquidus and standard Gibbs energy of LiF dissolved in
solidus compositions and with the Gibbs solid NaF.
energy of fusion from Eq. (1-98), values of The phase diagram drawn in Fig. 1-18
E (l)
gNaF can be calculated at the measured liq- was calculated from Eqs. (1-97) to (1-104).
uidus points from Eq. (1-101). Complete details of the analysis of the
Finally, enthalpies of mixing, hE , in the LiF – NaF system are given by Sangster and
liquid have been measured by calorimetry Pelton (1987).
by Hong and Kleppa (1976). As a second example of thermodynamic/
Combining all these data in a least- phase diagram optimization, consider the
squares optimization, the following expres- Cd – Na system. The phase diagram, with
sions for the liquid were obtained by Sang- points measured by several authors (Math-
ster and Pelton (1987): ewson, 1906; Kurnakow and Kusnetzow,
hE (l) = XLiF XNaF (1-102) 1907; Weeks and Davies, 1964) is shown in
Fig. 1-19.
¥ [– 7381 + 184 (XNaF – XLiF )] J/mol From electromotive force measurements
s E (l) = XLiF XNaF (1-103) on alloy concentration cells, several au-
¥ [– 2.169 – 0.562 (XNaF – XLiF )] J/mol thors have measured the activity coeffi-
cient of Na in liquid alloys. The data
Eqs. (1-102) and (1-103) then permit all are shown in Fig. 1-20 at 400 °C. From
other integral and partial properties of the the temperature dependence of g ENa =
liquid to be calculated. R T ln g Na , the partial enthalpy of Na in the
For the NaF-rich Henrian solid solution, liquid was obtained via Eq. (1-52). The re-
the solubility of LiF has been measured by sults are shown in Fig. 1-21. Also, hE of the
Holm (1965) at the eutectic temperature liquid has been measured by Kleinstuber
where the NaF-rich solid solution is in (1961) by direct calorimetry. These ther-
equilibrium with pure solid LiF. That is, modynamic data for g ENa , hENa and hE were
1.6 Application of Thermodynamics to Phase Diagram Analysis 37
Figure 1-19. Cd – Na phase diagram at P = 1 bar calculated from optimized thermodynamic parameters (Re-
printed from Pelton, 1988 a). 쑗 Kurnakow and Kusnetzow (1907), 왕 Mathewson (1906), × Weeks and Davies
(1964).
Figure 1-20. Sodium ac-

tivity coefficient in liquid
Cd – Na alloys at 400°C.
Line is calculated from
optimized thermodynamic
parameters (Reprinted
from Pelton, 1988a).
쏔 Hauffe (1940),
쎲 Lantratov and
Mikhailova (1971),
왕 Maiorova et al. (1976),
왔 Alabyshev and
Morachevskii (1957),
쑗 Bartlett et al. (1970).
Eq. (1-106) reproduces the calorimetric

data within 200 J/mol –1. Eqs. (1-52), (1-
58), (1-93) and (1-95) can be used to calcu-
late hENa and g Na . The calculated curves are
compared to the measured points in Figs.
1-20 and 1-21.
For the two compounds, Gibbs energies
of fusion were calculated (Pelton, 1988 a)
so as to best reproduce the measured phase
diagram:
Dg 0f (1/13 Cd11Na2 ) = 6816 – 10.724 T J/g-atom
(1-108)
Dg 0f (1/3 Cd2Na) = 8368 – 12.737 T J/g-atom
(1-109)
The optimized enthalpies of fusion of 6816
and 8368 J/g-atom agree within error lim-
its with the values of 6987 and 7878 J/g-
atom measured by Roos (1916). (See Fig.
1-14 for an illustration of the relation
between the Gibbs energy of fusion of a
compound and the phase diagram.)
The phase diagram shown in Fig. 1-19
was calculated from Eqs. (1-106) to (1-
109) along with the Gibbs energies of fu-
sion of Cd and Na taken from the literature
(Chase, 1983). Complete details of the
analysis of the Cd – Na system are given by
Pelton (1988 a).
Figure 1-21. Partial excess enthalpy of sodium in
It can thus be seen that one simple set of
liquid Cd –Na alloys. Line is calculated from opti-
mized thermodynamic parameters (Reprinted from equations can simultaneously and self-con-
Pelton, 1988a). 쎲 Lantratov and Mikhailova (1971), sistently describe all the thermodynamic
왕 Maiorova et al. (1976), 쑗 Bartlett et al. (1970). properties and the phase diagram of a bi-
nary system.
The exact optimization procedure will
optimized simultaneously (Pelton, 1988 a) vary from system to system depending
to obtain the following expressions for hE upon the type and accuracy of the avail-
and s E of the liquid: able data, the number of phases present, the
extent of solid solubility, etc. A large num-
hE (l) = XCd XNa [– 12 508 + 20 316 (1-106)
ber of optimizations have been published
¥ (XNa – XCd ) – 8714 (XNa – XCd )2 ] J/mol in the Calphad Journal (Pergamon) since
s E (l) = XCd XNa [– 15.452 + 15.186 (1-107) 1977.
¥ (XNa – XCd ) – 10.062 (XNa – XCd )2
– 1.122 (XNa – XCd )3 ] J/mol K
1.7 Ternary and Multicomponent Phase Diagrams 39
1.6.4 Calculation of Metastable Phase and Cd – Bi are found. Lines of constant

Boundaries mole fraction XBi are parallel to the Sn – Cd
edge, while lines of constant XSn and XCd
In the Cd – Na system just discussed, the
are parallel to the Cd – Bi and Bi – Sn edges
liquid exhibits positive deviations from
respectively. For example, at point a in Fig.
ideal mixing. That is, gE (l) > 0. This fact is
1-22, XBi = 0.05, XSn = 0.45 and XCd = 0.50.
reflected in the very flat liquidus in Fig.
Similar equilateral composition triangles
1-19 as was discussed in Sec. 1.5.9.
can be drawn with coordinates in terms of
By simply not including any solid phases
wt.% of the three components.
in the calculation, the metastable liquid
miscibility gap as well as the spinodal
curve (Sec. 1.5.5) can be calculated as 1.7.2 Ternary Space Model
shown in Fig. 1-19. These curves are im-
A ternary temperature – composition
portant in the formation of metallic glasses
“phase diagram” at constant total pressure
by rapid quenching.
may be plotted as a three-dimensional
Other metastable phase boundaries such
“space model” within a right triangular
as the extension of a liquidus curve below a
prism with the equilateral composition tri-
eutectic can also be calculated thermody-
angle as base and temperature as vertical
namically by simply excluding one or more
axis. Such a space model for a simple eu-
phases during the computations.
tectic ternary system A – B – C is illustrated
in Fig. 1-23. On the three vertical faces of
the prism we find the phase diagrams of the
1.7 Ternary and Multicomponent three binary subsystems, A – B, B – C and
Phase Diagrams C – A which, in this example, are all simple
eutectic binary systems. The binary eutec-
This section provides an introduction to tic points are e1 , e2 and e3 . Within the
ternary phase diagrams. For a more de- prism we see three liquidus surfaces de-
tailed treatment, see Prince (1966); Ricci scending from the melting points of pure
(1964); Findlay (1951); or West (1965). A, B and C. Compositions on these sur-
faces correspond to compositions of liquid
in equilibrium with A-, B- and C-rich solid
1.7.1 The Ternary Composition Triangle
phases.
In a ternary system with components In a ternary system at constant pressure,
A – B – C, the sum of the mole fractions is the Gibbs phase rule, Eq. (1-85), becomes:
unity, (XA + XB + XC ) = 1. Hence, there are
F=4–P (1-110)
two independent composition variables. A
representation of composition, symmetri- When the liquid and one solid phase are in
cal with respect to all three components, equilibrium P = 2. Hence F = 2 and the
may be obtained with the equilateral “com- system is bivariant. A ternary liquidus is
position triangle” as shown in Fig. 1-22 for thus a two-dimensional surface. We may
the Bi – Sn – Cd system. Compositions at choose two variables, say T and one
the corners of the triangle correspond to the composition coordinate of the liquid, but
pure components. Along the edges of the then the other liquid composition coordi-
triangle compositions corresponding to the nate and the composition of the solid are
three binary subsystems Bi – Sn, Sn – Cd fixed.
Figure 1-22. Projection of the liquidus surface of the Bi – Sn –Cd system onto the ternary composition triangle
(after Bray et al., 1961 – 1962). Small arrows show the crystallization path of an alloy of overall composition at
point a. (Reprinted from Pelton, 1996.)
Figure 1-23. Perspective view of ternary space

model of a simple eutectic ternary system. e1, e2 , e3
are the binary eutectics and E is the ternary eutectic.
The base of the prism is the equilateral composition
triangle. (Reprinted from Pelton, 1983.)
The A- and B-liquidus surfaces in Fig. 1- pure Cd. (Note that this fact cannot be de-
23 intersect along the line e1 E. Liquids duced from Fig. 1-22 alone.) Therefore, as
with compositions along this line are there- solidification proceeds, the liquid becomes
fore in equilibrium with A-rich and B-rich depleted in Cd, but the ratio XSn /XBi in the
solid phases simultaneously. That is, P = 3 liquid remains constant. Hence, the compo-
and so F – 1. Such “valleys” are thus called sition path followed by the liquid (its crys-
univariant lines. The three univariant lines tallization path) is a straight line passing
meet at the ternary eutectic point E at through point a and projecting to the Cd-
which P = 4 and F = 0. This is an invariant corner of the triangle. This crystallization
point since the temperature and the compo- path is shown on Fig. 1-22 as the line a b.
sitions of all four phases in equilibrium are In the general case in which a solid solu-
fixed. tion rather than a pure component or stoi-
chiometric compound is precipitating, the
crystallization path will not be a straight
1.7.3 Polythermal Projections
line. However, for equilibrium cooling, a
of Liquidus Surfaces
straight line joining a point on the crystal-
A two-dimensional representation of the lization path at any T to the overall compo-
ternary liquidus surface may be obtained as sition point a will extend through the com-
an orthogonal projection upon the base position, on the solidus surface, of the solid
composition triangle. Such a polythermal phase in equilibrium with the liquid at that
projection of the liquidus of the Bi – Sn – Cd temperature.
system (Bray et al., 1961 – 62) is shown in When the composition of the liquid has
Fig. 1-22. This is a simple eutectic ternary reached point b in Fig. 1-22 at T ≈ 435 K,
system with a space model like that shown the relative proportions of the solid Cd and
in Fig. 1-23. The constant temperature liquid phases at equilibrium are given by
lines on Fig. 1-22 are called liquidus iso- the lever rule applied to the tie-line d a b:
therms. The univariant valleys are shown (moles of liquid)/(moles of Cd) = d a /a b.
as heavier lines. By convention, the large Upon further cooling, the liquid composi-
arrows indicate the directions of decreas- tion follows the univariant valley from b to
ing temperature along these lines. E while Cd and Sn-rich solids coprecipitate
Let us consider the sequence of events as a binary eutectic mixture. When the
occurring during the equilibrium cooling liquidus composition attains the ternary eu-
from the liquid of an alloy of overall com- tectic composition E at T ≈ 380 K the invar-
position a in Fig. 1-22. Point a lies within iant ternary eutectic reaction occurs:
the field of primary crystallization of Cd.
liquid Æ s1 + s2 + s3 (1-111)
That is, it lies within the composition re-
gion in Fig. 1-22 in which Cd-rich solid where s1 , s2 and s3 are the three solid
will be the first solid to precipitate upon phases and where the compositions of all
cooling. As the liquid alloy is cooled, the four phases (as well as T ) remain fixed un-
Cd-liquidus surface is reached at T ≈ 465 K til all liquid is solidified.
(slightly below the 473 K isotherm). A In order to illustrate several of the fea-
solid Cd-rich phase begins to precipitate at tures of polythermal projections of liquidus
this temperature. Now, in this particular surfaces, a projection of the liquidus of a
system, Bi and Sn are nearly insoluble in hypothetical system A – B – C is shown in
solid Cd, so that the solid phase is virtually Fig. 1-24. For the sake of simplicity, iso-
Figure 1-24. Projection of

the liquidus surface of a
system A – B – C. The binary
subsystems A – B and C – A
are simple eutectic systems.
The binary phase diagram
B – C is shown in the insert.
All solid phases are assumed
pure stoichiometric compo-
nents or compounds. Small
arrows show crystallization
paths of alloys of composi-
tions at points a and b. (Re-
printed from Pelton, 1983.)
therms are not shown, only the univariant field, since e is a congruent compound,
lines with arrows to show the directions of while the composition of the d phase lies
decreasing temperature. The binary sub- outside of its field since d is incongruent.
systems A – B and C – A are simple eutectic Similarly for the ternary compounds, h is a
systems, while the binary subsystem B – C congruently melting compound while z is
contains one congruent binary phase, e, incongruent. For the congruent compound
and one incongruent binary phase, d, as h, the highest temperature on the h liquidus
shown in the insert in Fig. 1-24. The letters occurs at the composition of h.
e and p indicate binary eutectic and peritec- The univariant lines meet at a number of
tic points. The e and d phases are called bi- ternary eutectics E (three arrows converg-
nary compounds since they have composi- ing), a ternary peritectic P (one arrow en-
tions within a binary subsystem. Two ter- tering, two arrows leaving the point), and
nary compounds, h and z, with composi- several ternary quasi-peritectics P¢ (two
tions within the ternary triangle, as indi- arrows entering, one arrow leaving). Two
cated in Fig. 1-24, are also found in this saddle points s are also shown. These are
system. All compounds, as well as pure points of maximum T along the univariant
solid A, B and C (the “a, b and g” phases), line but of minimum T on the liquidus sur-
are assumed to be stoichiometric (i.e., there face along a section joining the composi-
is no solid solubility). The fields of pri- tions of the two solids. For example, s1 is at
mary crystallization of all the solids are in- a maximum along the univariant E1 P¢3 , but
dicated in parentheses in Fig. 1-24. The is a minimum point on the liquidus along
composition of the e phase lies within its the straight line z s1 h.
Let us consider the events occurring dur- univariant valley until the liquid reaches
ing cooling from the liquid of an alloy of the peritectic composition P. The invariant
overall composition a in Fig. 1-24. The pri- ternary peritectic reaction then occurs iso-
mary crystallization product will be the e thermally:
phase. Since this is a pure stoichiometric
liquid + e + b Æ z (1-113)
solid the crystallization path of the liquid
will be along a straight line passing Since there are three reactants, there are
through a and extending to the composition three possible outcomes: (i) the liquid is
of e as shown in the figure. consumed before either e or b and solidifi-
Solidification of e continues until the cation terminates at P; (ii) e is consumed
liquid attains a composition on the univari- first, solidification then continues along
ant valley. Thereafter the liquid composi- the path P P¢3 ; or (iii) b is consumed first
tion follows the valley towards the point P¢1 and solidification continues along the path
in co-existence with e and z. At point P¢1 P P¢1 . Which outcome occurs depends on
the invariant ternary quasi-peritectic reac- whether the overall composition b lies
tion occurs isothermally: within the tie-triangle (i) e b z; (ii) b z P, or
(iii) e z P. In the example shown, outcome
liquid + e Æ d + z (1-112)
(i) will occur.
Since there are two reactants in a quasi-
peritectic reaction, there are two possible
1.7.4 Ternary Isothermal Sections
outcomes: (i) the liquid is completely con-
sumed before the e phase; in this case, so- Isothermal projections of the liquidus
lidification will be complete at the point P¢1 ; surface do not provide information on the
(ii) e is completely consumed before the compositions of the solid phases at equilib-
liquid. In this case, solidification will con- rium. However, this information can be
tinue with decreasing T along the univari- presented at any one temperature on an iso-
ant line P¢1 E1 with co-precipitation of d and thermal section such as that shown for the
z until, at E, the liquid will solidify eutecti- Bi – Sn – Cd system at 423 K in Fig. 1-25.
cally (liquid Æ d + z + h). To determine This phase diagram is a constant tempera-
whether outcome (i) or (ii) occurs, we use ture slice through the space model of Fig.
the mass balance criterion that, for three- 1-23.
phase equilibrium, the overall composition The liquidus lines bordering the one-
a must always lie within the tie-triangle phase liquid region of Fig. 1-25 are identi-
formed by the compositions of the three cal to the 423 K isotherms of the projection
phases. Now, the triangle joining the com- in Fig. 1-22. Point c in Fig. 1-25 is point c
positions of d, e and z does not contain the on the univariant line in Fig. 1-22. An alloy
point a, but the triangle joining the compo- with overall composition in the one-phase
sitions of d, z and liquid at P¢1 does contain liquid region of Fig. 1-25 at 423 K will
the point a. Hence, outcome (ii) occurs. consist of a single liquid phase. If the over-
An alloy of overall composition b in Fig. all composition lies within one of the two-
1-24 solidifies with e as primary crystal- phase regions, then the compositions of the
lization product until the liquid composi- two phases are given by the ends of the tie-
tion contacts the univariant line. There- line which passes through the overall com-
after, co-precipitation of e and b occurs position. For example, a sample with over-
with the liquid composition following the all composition p in Fig. 1-25 will consist
Figure 1-25. Isothermal section of the Bi – Sn – Cd system at 423 K at P = 1 bar (after Bray et al., 1961–1962).
Extents of solid solubility in Bi and Sn have been exaggerated for clarity of presentation. (Reprinted from Pel-
ton, 1996.)
of a liquid of composition q on the liquidus is somewhat exaggerated in Fig. 1-25 for

and a solid Bi-rich alloy of composition r the sake of clarity of presentation.) Alloys
on the solidus. The relative proportions of with overall compositions rich enough in
the two phases are given by the lever rule: Bi or Sn to lie within the single-phase (Sn)
(moles of liquid)/(moles of solid) = p r /p q, or (Bi) regions of Fig. 1-25 will consist at
where p r and p q are the lengths of the tie- 423 K of single-phase solid solutions. Al-
line segments. loys with overall compositions at 423 K in
In the case of solid Cd, the solid phase is the two-phase (Cd + Sn) region will consist
nearly pure Cd, so all tie-lines of the (Cd + of two solid phases.
liquid) region converge nearly to the corner Alloys with overall compositions within
of the triangle. In the case of Bi- and Sn- the three-phase triangle d c f will, at 423 K,
rich solids, some solid solubility is ob- consist of three phases: solid Cd- and Sn-
served. (The actual extent of this solubility rich solids with compositions at d and f and
liquid of composition c. To understand this

better, consider an alloy of composition a
in Fig. 1-25, which is the same composi-
tion as the point a in Fig. 1-22. In Sec.
1.7.3 we saw that when an alloy of this
composition is cooled, the liquid follows
the path a b in Fig. 1-22 with primary pre-
cipitation of Cd and then follows the uni-
variant line with co-precipitation of Cd and
Sn so that at 423 K the liquid is at the com-
position point c, and two solid phases are in
equilibrium with the liquid.
Figure 1-26. A tie-triangle in a ternary isothermal
section illustrating the lever rule and the extension
1.7.4.1 Topology of Ternary Isothermal rule.
Sections
At constant temperature the Gibbs en-
ergy of each phase in a ternary system is single-phase region, and from each edge of
represented as a function of composition the triangle there extends a two-phase re-
by a surface plotted in a right triangular gion. The edge of the triangle is a limiting
prism with Gibbs energy as vertical axis tie-line of the two-phase region.
and the composition triangle as base. Just For overall compositions within the tie-
as the compositions of phases at equilib- triangle, the compositions of the three
rium in binary systems are determined phases at equilibrium are fixed at the cor-
by the points of contact of a common tan- ners of the triangle. The relative propor-
gent line to their isothermal Gibbs energy tions of the three phases are given by the
curves, so the compositions of phases at lever rule of tie-triangles, which can be de-
equilibrium in a ternary system are given rived from mass balance considerations. At
by the points of contact of a common tan- an overall composition q in Fig. 1-26 for
gent plane to their isothermal Gibbs energy example, the relative proportion of the g
surfaces. A common tangent plane can phase is given by projecting a straight line
contact two Gibbs energy surfaces at an in- from the g corner of the triangle (point c)
finite number of pairs of points, thereby through the overall composition q to the
generating an infinite number of tie-lines opposite side of the triangle, point p. Then:
within a two-phase region on an isothermal (moles of g)/(total moles) = q p /c p if com-
section. A common tangent plane to three positions are expressed in mole fractions,
Gibbs energy surfaces contacts each sur- or (weight of g)/(total weight) = q p /c p if
face at a unique point, thereby generating a compositions are in weight percent.
three-phase tie-triangle. Isothermal ternary phase diagrams are
Hence, the principal topological units of generally composed of a number of these
construction of an isothermal ternary phase topological units. An example for the Al –
diagram are three-phase (a + b + g) tie-tri- Zn – Mg system at 25 °C is shown in Fig.
angles as in Fig. 1-26 with their accompa- 1-27 (Köster and Dullenkopf, 1936). The
nying two-phase and single-phase areas. b, g, d, q, h and z phases are binary inter-
Each corner of the tie-triangle contacts a metallic compounds with small (~ 1 to 6%)
Another important rule of construction,

whose derivation is evident, is that within
any two-phase region tie-lines must never
cross one another.
1.7.5 Ternary Isopleths

(Constant Composition Sections)
A vertical isopleth, or constant composi-
tion section through the space model of the
Bi – Sn – Cd system, is shown in Fig. 1-28.
The section follows the line AB in Fig.
1-22.
The phase fields on Fig. 1-28 indicate
which phases are present when an alloy
Figure 1-27. Ternary isothermal section of the Al – with an overall composition on the line AB
Zn – Mg system at 25°C at P = 1 bar (after Köster and is equilibrated at any temperature. For ex-
Dullenkopf, 1936). (Reprinted from Pelton, 1983.) ample, consider the cooling, from the liq-
uid state, of an alloy of composition a
which is on the line AB (see Fig. 1-22). At
ranges of stoichiometry which can dissolve T ≈ 465 K, precipitation of the solid (Cd)
a limited amount (~ 1 to 6%) of the third phase begins at point a in Fig. 1-28. At
component. The t phase is a ternary phase T ≈ 435 K (point b in Figs. 1-22 and 1-28)
with a single-phase region existing over a the solid (Sn) phase begins to appear. Fi-
fairly extensive oval-shaped central com- nally, at the eutectic temperature TE , the
position range. Examination of Fig. 1-27 ternary reaction occurs, leaving solid (Cd)
shows that it consists of the topological + (Bi) + (Sn) at lower temperatures. The
units of Fig. 1-26. intersection of the isopleth with the univar-
An extension rule, a case of Schreine- iant lines on Fig. 1-22 occurs at points f and
makers’ Rule (Schreinemakers, 1915), see g which are also indicated on Fig. 1-28.
Sec. 1.7.5, for ternary tie-triangles is illus- The intersection of this isopleth with the
trated in Fig. 1-26. At each corner, the isothermal section at 423 K is shown in
extension of the boundaries of the single- Fig. 1-25. The points s, t, u and v of Fig.
phase regions, indicated by the broken 1-25 are also shown on Fig. 1-28.
lines, must either both project into the tri- It is important to note that on an isopleth
angle as at point a, or must both project the tie-lines do not, in general, lie in the
outside the triangle as at point b. Further- plane of the diagram. Therefore, the dia-
more, the angle between these extensions gram provides information only on which
must be less than 180°. For a proof, see phases are present, not on their composi-
Lipson and Wilson (1940) or Pelton tions. The boundary lines on an isopleth do
(1995). not in general indicate the phase composi-
Many published phase diagrams violate tions, only the temperature at which a
this rule. For example, it is violated in Fig. phase appears or disappears for a given
1-27 at the d-corner of the (e + d + t) tie-tri- overall composition. The lever rule cannot
angle. be applied on an isopleth.
Figure 1-28. Isopleth (constant composition section) of the Bi – Sn – Cd system at P = 1 bar following the line
AB at XSn = 0.45 of Fig. 1-22. (Reprinted from Pelton, 1996).
Certain geometrical rules apply to iso- 1.7.5.1 Quasi-Binary Phase Diagrams

pleths. As a phase boundary line is crossed,
one and only one phase either appears or Several of the binary phase diagrams
disappears. This Law of Adjoining Phase in the preceding sections (Figs. 1-5, 1-10,
Regions (Palatnik and Landau, 1964) is il- 1-12, 1-14, 1-18) are actually isopleths
lustrated by Fig. 1-28. The only apparent of ternary systems. For example, Fig. 1-12
exception occurs for the horizontal invari- is an isopleth at constant XO = nO /(nMg +
ant line at TE . However, if we consider this nCa + nO ) = 0.5 of the Mg – Ca – O system.
line to be a degenerate infinitely narrow However, all tie-lines lie within (or virtu-
four-phase region (L + (Cd) + (Bi) + (Sn)), ally within) the plane of the diagram be-
then the law is also obeyed here. cause XO = 0.5 in every phase. Therefore,
Three or four boundary lines meet at in- the diagram is called a quasi-binary phase
tersection points. At an intersection point, diagram.
such as point f or g, Schreinemakers’ Rule
applies. This is discussed in Sec. 1.9. 1.7.6 Multicomponent Phase Diagrams
Apparent exceptions to these rules (such
as, for example, five boundaries meeting at Only an introduction to multicomponent
an intersection point) can occur if the sec- phase diagrams will be presented here.
tion passes exactly through a node (such For more detailed treatments see Palatnik
as a ternary eutectic point) of the space and Landau (1964), Prince (1963), Prince
model. However, these apparent exceptions (1966) and Hillert (1998).
are really only limiting cases (see Prince, For systems of four or more components,
1963 or 1966). two-dimensional sections are usually plot-
ted with one or more compositional vari- at equilibrium for an overall composition
ables held constant. Hence these sections lying on the section. Tie-lines do not, in
are similar to the ternary isopleths dis- general, lie in the plane of the diagram, so
cussed in Sec. 1.7.5. In certain cases, sec- the diagram does not provide information
tions at constant chemical potential of one on the compositions or amounts of the
or more components (for example, at con- phases present. As a phase boundary is
stant oxygen partial pressure) can be use- crossed, one and only one phase appears or
ful. These are discussed in Sec. 1.8. disappears (Law of Adjoining Phase Re-
Two sections of the Fe – Cr –V– C system gions). If temperature is an axis, as in Fig.
(Lee and Lee, 1992) are shown in Figs. 1- 1-29, then horizontal invariants like the
29 and 1-30. The diagram in Fig. 1-29 is a line AB in Fig. 1-29 can appear. These can
T-composition section at constant Cr and V be considered as degenerate infinitely nar-
content, while Fig. 1-30 is a section at con- row phase fields of (C + 1) phases, where C
stant T = 850 oC and constant C content of is the number of components (for isobaric
0.3 wt.%. The interpretation and topologi- diagrams). For example in Fig. 1-29, on the
cal rules of construction of these sections line AB, five phases are present. Three or
are the same as those for ternary isopleths, four phase boundaries meet at intersection
as discussed in Sec. 1.7.5. In fact, the same points at which Schreinemakers’ Rule ap-
rules apply to a two-dimensional constant- plies. It is illustrated by the extrapolations
composition section for a system of any in Fig. 1-29 at points a, b and c and in Fig.
number of components. The phase fields 1-30 at points b, c, n, i and s (see discus-
on the diagram indicate the phases present sions in Sec. 1.9).
Figure 1-29. Section of the Fe – Cr – V – C system at 1.5 wt.% Cr and 0.1 wt.% V at P = 1 bar (Lee and Lee,
1992).
Figure 1-30. Section of

the Fe – Cr – V – C system
at 850 oC and 0.3 wt.% C
at P = 1 bar (Lee and Lee,
1992).
1.7.7 Nomenclature for Invariant + d) is a 1 Æ 3 reaction; and so on. The ter-

Reactions nary peritectic-type 3 Æ 1 reaction (a + b
+ g Æ d) is an invariant reaction in a ter-
As discussed in Sec. 1.5.12, in a binary nary system, a univariant reaction in a qua-
isobaric temperature – composition phase ternary system, a bivariant reaction in a
diagram there are two possible types of quinary system, etc.
invariant reactions: “eutectic-type” (b Æ a
+ g), and “peritectic type” (a + g Æ b). In
1.7.8 Reciprocal Ternary Phase
a ternary system, there are “eutectic-type”
Diagrams
(a Æ b + g + d), “peritectic-type” (a + b +
g Æ d), and “quasiperitectic-type” (a + b A reciprocal ternary salt system is one
Æ g + d) invariants (Sec. 1.7.3). In a consisting of two cations and two anions,
system of C components, the number of such as the Na+, K+/ F –, Cl – system of Fig.
types of invariant reaction is equal to C. A 1-31. The condition of charge neutrality
reaction with one reactant, such as (nNa+ + nK+ = nF – + nCl – ) removes one degree
(a Æ b + g + d + e) is clearly a “eutectic- of freedom. The system is thus quasiter-
type” invariant reaction but in general there nary and its composition can be repre-
is no standard terminology. These reactions sented by two variables, usually chosen
are conveniently described according to the as the cationic mole fraction XK = nK /
numbers of reactants and products (in the (nNa + nK ) and the anionic mole fraction
direction which occurs upon cooling). XCl = nCl /(nF + nCl ), where ni = number of
Hence the reaction (a + b Æ g + d + e) is a moles of ion i. Note that XNa = (1 – XK ) and
2 Æ 3 reaction; the reaction (a Æ b + g XF = (1 – XCl ).
ternary but is an isopleth of the four-com-

ponent Na – K – F – Cl system, and tie-lines
no longer necessarily lie in the plane of the
diagram.
In Fig. 1-31 the cationic and anionic
fractions are plotted as axes of a square.
Compositions corresponding to the four
neutral salts (KF, KCl, NaCl, NaF) are
found at the corners of the square. Edges of
the square correspond to the binary sub-
systems such as NaF – NaCl. A ternary
space model (analogous to Fig. 1-23) can
be constructed with temperature as vertical
axis. The phase diagram of Fig. 1-31 is a
polythermal projection of the liquidus sur-
face upon the composition square.
In this system, three of the binary edges
are simple eutectic systems, while the
NaCl – KCl binary system exhibits a sol-
idus/liquidus minimum. There is a ternary
eutectic at 570 °C in Fig. 1-31 (b). The
NaF – KCl diagonal contains a saddle point
at 648 °C in Fig. 1-31 (b). This saddle point
is a eutectic of the quasibinary system
NaF – KCl. That is, a binary phase diagram
NaF – KCl could be drawn with one simple
eutectic at 648 °C. However, the NaCl – KF
system, which forms the other diagonal, is
not a quasibinary system. If compositions
lying on this diagonal are cooled at equilib-
rium from the liquid, solid phases whose
compositions do not lie on this diagonal
can precipitate. Hence, a simple binary
phase diagram cannot be drawn for the
Figure 1-31. Projection of the liquidus surface of
the Na+, K+/F –, Cl – reciprocal ternary system.
NaCl – KF system.
a) Calculated from optimized binary thermodynamic For systems such as Ca2+, Na+/ F –, SO42–
parameters. in which the ions do not all have the same
b) As reported by Polyakov (1940). charge, composition axes are conveniently
expressed as equivalent ionic fractions
(e.g. YCa = 2 nCa /(2 nCa + nNa )), see Sec.
The assumption has, of course, been 1.9.2.1.
made that the condition (nNa + nK = nF + nCl ) The concept of reciprocal systems can be
holds exactly in every phase. If there is a generalized beyond simple salt systems
deviation from this exact stoichiometry, and is closely related to the sublattice
then the phase diagram is no longer quasi- model (Sec. 1.10.1).
1.8 Phase Diagrams with Potentials as Axes 51
For further discussion and references,

see Pelton (1988 b) and Blander (1964).
1.8 Phase Diagrams

with Potentials as Axes
So far we have considered mainly iso-
baric temperature – composition phase dia-
grams. However there are many other
kinds of phase diagrams of interest in ma-
terials science and technology with pres-
sure, chemical potentials, volume, etc. as
axes. These can be classified into geomet-
rical types according to their rules of con-
struction.
For instance, binary isothermal P – X di-
agrams as in Fig. 1-8 are members of the
same type as binary isobaric T – X diagrams
because they are both formed from the
same topological units of construction.
Other useful phase diagrams of this same
geometrical type are isothermal chemical Figure 1-32. Corresponding phase diagrams for the
potential – composition diagrams for ter- Co – Ni – O system at 1600 K (from Pelton and
nary systems. An example is shown in Thompson, 1973).
the lowest panel of Fig. 1-32 (Pelton and
Thompson, 1975) for the Co – Ni – O
system at T = 1600 K (and at a constant to- The usual isothermal section of the ter-
tal hydrostatic pressure of 1 bar). Here the nary Co – Ni – O system at 1600 K is shown
logarithm of the equilibrium partial pres- in the top panel of Fig. 1-32. There are two
sure of O2 is plotted versus the metal ratio single-phase regions with a two-phase re-
x = nNi /(nCo + nNi ), where ni = number of gion between them. The single-phase areas
moles of i. There are two phases in this are very narrow because oxygen is only
system under these conditions, a solid alloy very slightly soluble in the solid alloy
solution stable at lower pO2 , and a solid so- and CoO and NiO are very stoichiometric
lution of CoO and NiO stable at higher pO2 . oxides. In the central panel of Fig. 1-32
For instance, point a gives pO2 for the equi- this same diagram is shown with the com-
librium between pure Co and pure CoO at position triangle “opened up” by putting
1600 K. Between the two single-phase re- the oxygen corner at infinity. This can be
gions is a two-phase (alloy + oxide) region. done if the vertical axis becomes h = nO /
At any overall composition on the tie-line (nCo + nNi ) with the horizontal axis as
c d between points c and d, two phases will x = nNi /(nCo + nNi ). These are known as
be observed, an alloy of composition d and Jänecke coordinates. It can be seen in Fig.
an oxide of composition c. The lever rule 1-32 that each tie-line, e f, of the isothermal
applies just as for binary T – X diagrams. section corresponds to a tie-line c d of the
log pO2 – x diagram. This underscores the log pO2 – x diagram. For example, the (spi-
fact that every tie-line of a ternary isother- nel + (Fe, Cr) O + alloy) triangle with cor-
mal section corresponds to a constant chem- ners at points a, b and c corresponds to the
ical potential of each of the components. “eutectic-like” or eutecular invariant with
Another example of a log pO2 – x diagram the same phase compositions a, b and c at
is shown for the Fe – Cr – O system at log pO2 ≈ –10.7. We can see that within a
1573 K in the lower panel of Fig. 1-33 three-phase tie-triangle, pO2 is constant.
(Pelton and Schmalzried, 1973). The corre- An example of yet another kind of phase
sponding ternary isothermal section in diagram of this same geometrical type is
Jänecke coordinates is shown in the upper shown in Fig. 1-34. For the quaternary
panel. Each of the invariant three-phase Fe – Cr – O2 – SO2 system at T = 1273 K and
tie-triangles in the isothermal section at constant pSO2 = 10–7 bar, Fig. 1-34 is a
corresponds to an invariant line in the plot of log pO2 versus the molar metal ratio
Figure 1-33. Correspond-

ing phase diagrams for the
Fe–Cr–O system at 1573 K
(Pelton and Schmalzried,
1973). Experimental points
from Katsura and Muan
(1964).
Figure 1-34. Calculated phase diagram of log pO2 versus molar metal ratio at T = 1273.15 K and pSO2 = 10–7 bar
for the Fe – Cr – SO2 – O2 system.
x . Since log pO2 varies as –1/2 log pS2 when geometrical type. Often for ceramic
pSO2 and T are constant, Fig. 1-34 is also a systems, we encounter “binary” phase dia-
plot of log pS2 versus x . grams such as that for the “CaO – Fe2O3”
Plotting T versus x at constant pO2 in the system in Fig. 1-35, which has been taken
Fe – Cr – O system, or at constant pO2 and from Phillips and Muan (1958). How are
pSO2 in the Fe – Cr – SO2 – O2 system, will we to interpret such a diagram? How, for
also result in phase diagrams of this same instance, do we interpret the composition
Figure 1-35. Phase diagram for the

“CaO – Fe2O3” system in air ( pO2 =
0.21 bar) from Phillips and Muan
(1958) (Reprinted by permission of the
American Ceramic Society from Levin
et al., 1964).
axis when applied to the magnetite phase?

In light of the preceding discussion, it can
be seen that such diagrams are really T – x
plots at constant pO2 , where x is the metal
ratio in any phase. The diagram will be dif-
ferent at different oxygen partial pressures.
If pO2 is not fixed, the diagram cannot be
interpreted.
It can be seen that the diagrams dis-
cussed above are of the same geometrical
type as binary T – X diagrams because they
are all composed of the same geometrical
units of construction as in Fig. 1-16. Their
interpretation is thus immediately clear
to anyone familiar with binary T – X dia- Figure 1-36. Pressure-temperature phase diagram of
grams. Chemical potential – composition H2O.
Figure 1-37. Corresponding phase diagrams for the Fe – O system at PTOTAL = 1 bar (after Muan and Osborn,
1965).
diagrams (Figs. 1-32 to 1-34) are useful in of Figs. 1-1 to 1-3 are of this same geomet-
the study of hot corrosion, metallurgical rical type.
roasting processes, chemical vapor deposi- As yet another example of this geomet-
tion, and many aspects of materials pro- rical type of diagram, a plot of R T ln pO2
cessing. versus T for the Fe – O system is shown in
Another important geometrical type of Fig. 1-37 (b). Again, one-, two- and three-
phase diagram is exemplified by P – T phase regions are indicated by areas, lines
phase diagrams for one-component sys- and triple points respectively. Fig. 1-37 (a)
tems, as shown for H2O in Fig. 1-36. In is the binary T – composition phase diagram
such diagrams (see also Chapter 10 by for the Fe – O system. The correspondence
Kunz (2001)) bivariant single-phase re- between Figs. 1-37 (a) and 1-37 (b) is evi-
gions are indicated by areas, univariant dent. Each two-phase line of Fig. 1-37 (b)
two-phase regions by lines, and invariant “opens up” to a two-phase region of Fig.
three-phase regions by triple points. An 1-37 (a). Each tie-line of a two-phase re-
important rule of construction is the exten- gion in Fig. 1-37 (a) can thus be seen to
sion rule, which is illustrated by the broken correspond to a constant pO2 . Triple points
lines in Fig. 1-36. At a triple point, the in Fig. 1-37 (b) become horizontal invari-
extension of any two-phase line must pass ant lines in Fig. 1-37 (a).
into the single-phase region of the third Yet another type of phase diagram is
phase. Clearly, the predominance diagrams shown in Fig. 1-38. This is an isothermal
Figure 1-38. Phase diagram of log pS2 versus log pO2 at 1273 K and constant molar metal ratio nCr /(nFe + nCr) =
0.5 in the Fe – Cr – S2 – O2 system.
section at constant molar metal ratio Table 1-1. Corresponding pairs of potentials f i and
nCr /(nFe + nCr ) = 0.5 for the Fe – Cr – S2 – O2 extensive variables qi .
system. This diagram was calculated ther- fi : T P m1 m 2 … mC
modynamically from model parameters.
qi : S –V n1 n2 … nC
The axes are the equilibrium sulfur and
oxygen partial pressures. Three or four
boundary lines can meet at an intersection
point. Some of the boundary lines on Fig. where U is the internal energy of the
1-38 separate a two-phase region (a + b) system. The corresponding potentials and
from another two-phase region (a + g). extensive variables are listed in Table 1-1.
These lines thus represent the conditions It may also be noted that the corresponding
for three-phase (a + b + g) equilibrium. pairs are found together in the terms of the
general Gibbs – Duhem equation:
S dT – V dP + S ni dm i = 0 (1-115)
1.9 General Phase Diagram
Geometry
1.9.1 General Geometrical Rules
Although the various phase diagrams for All True Phase Diagram Sections
shown in the preceding sections may ap-
The Law of Adjoining Phase Regions ap-
pear to have quite different geometries, it
plies to all true sections. As a phase boun-
can be shown that, in fact, all true phase di-
dary line is crossed, one and only one
agram sections obey the same set of geo-
phase either appears or disappears.
metrical rules. Although these rules do not
If the vertical axis is a potential (T, P,
apply directly to phase diagram projections
m i ), then horizontal invariant lines like the
such as Figs. 1-22, 1-24 and 1-31, such di-
eutectic line in Fig. 1-12 or the line AB in
agrams can be considered to consist of por-
Fig. 1-29 will be seen when the maximum
tions of several phase diagram sections
number of phases permitted by the phase
projected onto a common plane.
rule are at equilibrium. However, if these
By “true” phase diagram we mean one in
are considered to be degenerate infinitely
which each point of the diagram represents
narrow phase fields, then the Law of Ad-
one unique equilibrium state. In the present
joining Phase Regions still applies. This is
section we give the general geometrical
illustrated schematically in Fig. 1-39 where
rules that apply to all true phase diagram
the three-phase eutectic line has been
sections, and we discuss the choices of
“opened up”. Similarly, if both axes are po-
axes and constants that ensure that the dia-
tentials, then many phase boundaries may
gram is a true diagram.
be degenerate infinitely narrow regions.
We must first make some definitions. In
For example, all phase boundaries on Figs.
a system of C components we can define
1-1 to 1-3, 1-36 and 1-37 (b) are degenerate
(C + 2) thermodynamic potentials f i . These
two-phase regions which are schematically
are T, P, m 1 , m 2 , …, m C , where m i is the
shown “opened up” on Fig. 1-40.
chemical potential defined in Eq. (1-23).
All phase boundary lines in a true phase
For each potential there is a corresponding
diagram meet at nodes where exactly four
extensive variable qi related by:
lines converge, as in Fig. 1-41. N phases
f i = (∂U/∂qi )q j ( j ≠ i ) (1-114) (a1 , a2 , …, aN ) where N ≥ 1 are common
1.9 General Phase Diagram Geometry 57
Figure 1-39. An isobaric binary T – X

phase diagram (like Fig. 1-12) with the
eutectic line “opened up” to illustrate that
this is a degenerate 3-phase region.
Figure 1-41. A node in a true phase diagram sec-

Figure 1-40. A potential – potential phase diagram tion.
(like Fig. 1-1 or Fig. 1-36) with the phase boundaries
“opened up” to illustrate that they are degenerate 2-
phase regions.
components was noted by Hillert (1985)
and proved by Pelton (1995). In the case of
to all four regions. Schreinemakers’ Rule degenerate phase regions, all nodes can
states that the extensions of the boundaries still be considered to involve exactly four
of the N-phase region must either both lie boundary lines if the degenerate boundar-
within the (N + 1)-phase regions as in Fig. ies are “opened up” as in Figs. 1-39 and
1-41 or they must both lie within the 1-40.
(N + 2)-phase region. This rule is illustrated An objection might be raised that a
by the extrapolations in Fig. 1-29 at points minimum or a maximum in a two-phase re-
a, b and c and in Fig. 1-30 at points b, c, n, gion in a binary temperature – composition
i and s. The applicability of Schreine- phase diagram, as in Fig. 1-10 or in the
makers’ Rule to systems of any number of lower panel of Fig. 1-16, represents an ex-
ception to Schreinemakers’ Rule. How- Fig. 1-1, all lines are actually two coinci-
ever, the extremum in such a case is not ac- dent ZPF lines.
tually a node where four phase boundaries The ZPF line concept is very useful in
converge, but rather a point where two the development of general algorithms for
boundaries touch. Such extrema in which the thermodynamic calculation of phase di-
two phase boundaries touch with zero agrams as discussed in Sec. 1-12.
slope may occur for a C-phase region in a
phase diagram of a C-component system 1.9.2 Choice of Axes and Constants
when one axis is a potential. For example, of True Phase Diagrams
in an isobaric temperature – composition
In a system of C components, a two-
phase diagram of a four-component sys-
dimensional diagram is obtained by choos-
tem, we may observe a maximum or a min-
ing two axis variables and holding (C – 1)
imum in a four-phase region separating two
other variables constant. However, not all
three-phase regions. A similar maximum
choices of variables will result in a true
or minimum in a (C – n)-phase region,
phase diagram. For example on the P – V
where n > 0, may also occur, but only for
diagram for H2O shown schematically in
a degenerate or special composition path.
Fig. 1-42, at any point in the area where the
For further discussion, see Hillert (1998).
(S + L) and (L + G) regions overlap there
are two possible equilibrium states of the
1.9.1.1 Zero Phase Fraction Lines system. Similarly, the diagram of carbon
activity versus XCr at constant T and P in
All phase boundaries on true phase dia-
the Fe – Cr – C system in Fig. 1-43 (Hillert,
gram sections are zero phase fraction
1997) exhibits a region in which there is no
(ZPF) lines, a very useful concept intro-
unique equilibrium state.
duced by Gupta et al. (1986). There are
In order to be sure that a diagram is a
ZPF lines associated with each phase. On
true phase diagram, we must choose one
one side of its ZPF line the phase occurs,
and only one variable (either fi or qi ) from
while on the other side it does not. For ex-
ample, in Fig. 1-30 the ZPF line for the a
phase is the line a b c d e f. The ZPF line for
the g phase is g h i j k l. For the MC phase the
ZPF line is m n c i o p q. The ZPF line for
M7C3 is r n b h s p k e t, and for M23C6 it is
u d j o s v. These five ZPF lines yield the en-
tire two-dimensional phase diagram. Phase
diagram sections plotted on triangular co-
ordinates as in Figs. 1-25 and 1-27 also
consist of ZPF lines.
In the case of phase diagrams with de-
generate regions, ZPF lines for two differ-
ent phases may be coincident over part
of their lengths. For example, in Fig. 1-12,
line CABD is the ZPF line of the liquid,
while CEBF and DEAG are the ZPF lines Figure 1-42. Schematic P – V diagram for H2O. This
for the a and b phases respectively. In is not a true phase diagram.
selected extensive variables we form the

ratio nFe /(nFe + nCr ). Fig. 1-38 is a plot
of m S2 versus m O2 at constant T, P and
nFe /(nFe + nCr ).
In Fig. 1-28 the selected variables are T,
P, nBi , nSn and nCd , and ratios are formed
from the selected extensive variables as
nCd /(nCd + nBi ) and nSn /(nCd + nBi + nSn )
= XSn . Fig. 1-28 is a plot of T versus
nCd /(nCd + nBi ) at constant P and XSn .
Fig. 1-42, the P – V diagram for H2O, is
not a true phase diagram because P and V
are members of the same conjugate pair.
For the diagram shown in Fig. 1-43, we can
choose one variable from each pair as fol-
Figure 1-43. Carbon activity versus mole fraction lows: T, P, m C , nFe , nCr . However the verti-
of Cr at constant T and P in the Fe – Cr – C system. cal axis is XCr = nCr /(nFe + nCr + nC ). This ra-
This is not a true phase diagram (from Hillert, 1997). tio is not allowed because it contains nC
which is not on the list of chosen variables.
That is, since we have chosen m C to be an
each of the (C + 2) conjugate pairs in Table axis variable, we cannot also choose nC .
1-1. (Also, at least one of these must be an Hence, Fig. 1-43 is not a true phase dia-
extensive variable qi .) From among the gram. A permissible choice for the vertical
n (1 ≤ n ≤ C + 2) selected extensive vari- axis would be nCr /(nFe + nCr ) (see Fig.
ables, (n – 1) independent ratios are then 1-33). Note that many regions of Figs. 1-42
formed. These (n – 1) ratios along with the and 1-43 do represent unique equilibrium
(C + 2 – n) selected potentials are the (C + 1) states. That is, the procedure given here is a
required variables. Two are chosen as axis sufficient, but not necessary, condition for
variables and the remainder are held con- constructing true phase diagrams.
stant. To apply this procedure simply, the com-
As a first example, consider a binary ponents of the system should be formally
system with components A – B. The conju- defined to correspond to the desired axis
gate pairs are (T, S), (P, – V ), (mA , nA ) and variables or constants. For example, in Fig.
(m B , nB ). Let us choose one variable from 1-1 we wish to plot pSO2 and log pO2 as
each pair as follows: T, P, nA , nB . From the axes. Hence we define the components as
selected extensive variables, nA and nB , we Cu – SO2 – O2 rather than Cu – S – O.
form a ratio such as nB /(nA + nB ) = XB . The In several of the phase diagrams in this
resultant phase diagram variables are T, P, chapter, log pi or R T ln pi has been substi-
XB . Choosing any two as axes and holding tuted for m i as axis variable or constant.
the third constant will give a true phase di- From Eq. (1-32), this substitution can
agram as in Fig. 1-6 or Fig. 1-8. clearly be made if T is constant. However,
As a second example, consider Fig. 1-38 even when T is an axis of the phase dia-
for the Fe – Cr – S2 – O2 system. We choose gram as in Fig. 1-37 (b), this substitution is
one variable from each conjugate pair as still permissible since m i0 is a monotonic
follows: T, P, m S2 , mO2 , nFe , nCr . From the function of T. The substitution of ln ai for
m i results in a progressive expansion and Na2O system we would choose as axes

displacement of the axis with increasing T the equivalent cationic and anionic frac-
that preserves the overall geometry of the tions, nNa /(nNa + 2 nCa ) and nCl /(nCl + 2 nO ),
diagram. whose denominators are equal because of
charge neutrality.
1.9.2.1 Tie-lines
1.9.2.2 Corresponding Phase Diagrams
If only potentials (T, P, m i ) are held con-
When only potentials are held constant
stant, then all tie-lines lie in the plane of
and when both axes are also potentials,
the phase diagram section. In this case, the
then the geometry exemplified by Figs. 1-1
compositions of the individual phases at
to 1-3, 1-26 and 1-37 (b) results. Such dia-
equilibrium can be read from the phase
grams were called “type-1 phase diagrams”
diagram, and the lever rule applies as, for
by Pelton and Schmalzried (1973). If only
example, in Figs. 1-6, 1-25, 1-33 or 1-34.
potentials are held constant and one axis is
However, if a ratio of extensive variables,
a potential while the other is a composition
such as a composition, is held constant as
variable, then the geometry exemplified
in the isopleths of Figs. 1-28 to 1-30, then
by Figs. 1-8, 1-12, 1-34, 1-37 (a), and the
in general, tie-lines do not lie in the plane.
lower panel of Fig. 1-33 results. These
If both axes are composition variables
were termed “type-2” diagrams. Finally, if
(ratios of ni ), and if only potentials are held
only potentials are held constant and both
constant, then it is desirable that the tie-
axes are compositions, then a “type-3” dia-
lines (which lie in the plane) be straight
gram as in the upper panels of Figs. 1-32
lines. It can be shown (Pelton and Thomp-
and 1-33 results.
son, 1975) that this will only be the case if
If the fi axis of a phase diagram is re-
the denominators of the two composition
placed by a composition variable that var-
variable ratios are the same. For example,
ies as its conjugate variable qi (ex: qi /q j ,
in the central panel of Fig. 1-32, which is in
qi /(qi + q j )), then the new diagram and the
Jänecke coordinates, the composition vari-
original diagram are said to form a pair of
ables, nCo /(nCo + nNi ) and nO /(nCo + nNi ),
corresponding phase diagrams. For in-
have the same denominator. This same dia-
stance, Figs. 1-37 (a) and 1-37 (b) are cor-
gram can be plotted on triangular coordi-
responding type-1 and type-2 phase dia-
nates as in the upper panel of Fig. 1-32 and
grams, while Fig. 1-33 shows a corre-
such a diagram can also be shown (Pelton
sponding pair of type-2 and type-3 dia-
and Thompson, 1975) to give straight tie-
grams. It is useful to draw corresponding
lines.
diagrams beside each other as in Figs. 1-37
Similarly, in the quasiternary reciprocal
or 1-33 because the information contained
phase diagram of Fig. 1-31 the vertical and
in the two diagrams is complementary.
horizontal axes are nNa /(nNa + nK ) and
nCl /(nCl + nF ). To preserve charge neutral-
1.9.2.3 Theoretical Considerations
ity, (nNa + nK ) = (nCl + nF ), and so the tie-
lines are straight. Generally, in quasiter- A complete rigorous proof that the pro-
nary reciprocal salt phase diagrams, cedure described in this section will always
straight tie-lines are obtained by basing the generate a true phase diagram is beyond
composition on one equivalent of charge. the scope of this chapter. As an outline of
For example, in the CaCl2 – NaCl – CaO – the proof, we start with the generalized
stability criterion: values. For example, in the predominance

(1-116)
diagram of Fig. 1-1, if the components are
(∂f1 /∂q1)f 2 , f 3 , …, qN , qN +1 , qN + 2 , …, qC + 2 ≥ 0
chosen as Cu – SO2 – O2 , then the com-
This equation states that a potential fi al- pound Cu2S is written as Cu2(SO2 )O– 2 ;
ways increases as its conjugate variable qi that is nO2 = –1. This is no problem in Fig.
increases when either f j or q j from every 1-1, since m O2 rather than nO2 was chosen
other conjugate pair is held constant. For from the conjugate pair and is plotted as an
instance, m i of a component always in- axis variable. However, suppose we wish
creases as that component is added to a to plot a diagram of m Cu versus m SO2 at con-
system (that is, as ni is increased) at constant T and P in this system. In this case,
stant T and P, when either the number of the chosen variables would be T, P, m SO2 ,
moles or the chemical potential of every nO2 . Since one of the selected extensive
other component is held constant. In a bi- variables must always be positive, and
nary system, for example, this means that since nO2 is the only selected extensive var-
the equilibrium Gibbs energy envelope is iable, it is necessary that nO2 be positive
always convex, as shown in Fig. 1-6. If the everywhere. For instance, a phase field for
envelope were concave, then the system Cu2S is not permitted. In other words, only
would be unstable and would separate into compositions in the Cu – SO2 – O2 subsys-
two phases, as shown in Fig. 1-11. tem are permitted. A different phase dia-
Consider first a phase diagram with axes gram would result if we plotted m Cu versus
f1 and f2 with f3 , f4 , …, fC +1 and qC + 2 m SO2 in the Cu – SO2 – S2 subsystem with nS2
constant. Such a diagram is always a true always positive. Cu2O would then not ap-
phase diagram. If the potential f1 is now re- pear, for example. That is, at a given m Cu
placed by q1 , the diagram still remains a and pSO2 we could have a low pO2 and a
true phase diagram because of Eq. (1-116). high pS2 in equilibrium with, for example,
The sequence of equilibrium states that oc- Cu2S, or we could have a high pO2 and a
curs as q1 is increased will be the same as low pS2 in equilibrium with, for example,
that which occurs as f1 is increased when Cu2O. Hence the diagram will not be a true
all the other variables (fi or qi ) are held diagram unless compositions are limited to
constant. the Cu – SO2 – O2 or Cu – SO2 – S2 triangles.
A true phase diagram is therefore ob- As a second example, if m SiO and m CO are
tained if the axis variables and constants chosen as variables in the SiO – CO– O
are chosen from the variables f1 , f2 , …, system, then the diagram must be limited
fN , qN +1 , qN +2 , …, qC +1 with qC + 2 held to nO > 0 (SiO – CO – O subsystem) or to
constant. The extensive variables can be nO < 0 (SiO – CO – Si – C subsystem).
normalized as (qi /qC + 2) or by any other in-
dependent and unique set of ratios. 1.9.2.4 Other Sets of Conjugate Pairs
It should be noted that at least one exten-
The set of conjugate pairs in Table 1-1 is
sive variable, qC + 2 , is considered to be con-
only one of many such sets. For example, if
stant across the entire diagram. In practice,
we make the substitution (H = T S + S ni m i )
this means that one of the extensive vari-
in Eq. (1-115), then we obtain another form
ables must be either positive or negative
of the general Gibbs – Duhem equation:
everywhere on the diagram. For certain
formal choices of components, extensive – H d (1/T ) – (V/T ) dP + S ni d ( m i /T ) = 0
composition variables can have negative (1-117)
This defines another set of pairs of conju- case the model is formally the same as that
gate potentials and extensive variables: of a simple substitutional solution, because
(1/T, – H ), (P, – V/T ), ( m i /T, ni ). Choosing the site fractions XMg and XCa of Mg2 + and
one and only one variable from each pair, Ca2 + cations on the cationic sublattice are
we can construct a true phase diagram by numerically equal to the overall component
the procedure described above. However, mole fractions XMgO and XCaO . Solid and liq-
these diagrams may be of limited practical uid MgO – CaO solutions have been shown
utility. This is discussed by Hillert (1997). (Wu et al., 1993) to be well represented by
simple polynomial equations for g E .
As a second example, the intermetallic e-
1.10 Solution Models FeSb phase exhibits non-stoichiometry to-
ward excess Fe. This phase was modeled
In Sec. 1.4.7, the thermodynamic expres- (Pei et al., 1995) as a solution of Fe and
sions for simple ideal substitutional solu- stoichiometric FeSb by assuming two sub-
tions were derived and in Secs. 1.5.7 and lattices: an “Fe sublattice” occupied only
1.6.2, the regular solution model and poly- by Fe atoms and an “Sb sublattice” occu-
nomial extensions thereof were discussed. pied by both Fe and Sb atoms such that, per
For other types of solutions such as ionic gram atom,
mixtures, interstitial solutions, polymeric Dgm = 0.5 R T ( yFe ln yFe + ySb ln ySb )
solutions, etc., the most convenient defini-
+ a yFe ySb (1-118)
tion of ideality may be different. In the
present section we examine some of these where ySb = (1 – yFe ) = 2 XSb is the site frac-
solutions. We also discuss structural order- tion of Sb atoms on the “Sb sites” and a is
ing and its effect on the phase diagram. For an empirical polynomial in ySb .
further discussion, see Pelton (1997).
1.10.1.2 Ionic Solutions
1.10.1 Sublattice Models
Let us take as an example a solution,
The sublattice concept has proved to be solid or liquid, of NaF, KF, NaCl and KCl
very useful in thermodynamic modeling. as introduced in Sec. 1.7.8. If the cations
Sublattice models, which were first devel- are assumed to mix randomly on a cationic
oped extensively for molten salt solutions, sublattice while the anions mix randomly
find application in ceramic, interstitial soon an anionic sublattice, then the molar
lutions, intermetallic solutions, etc. Gibbs energy of the solution can be mod-
eled by the following equation which con-
1.10.1.1 All Sublattices Except One tains an ideal mixing term for each sublat-
Occupied by Only One Species tice:
In the simplest limiting case, only one g = (XNa XCl g 0NaCl + XK XF g 0KF
sublattice is occupied by more than one + XNa XF g 0NaF + XK XCl g 0KCl ) (1-119)
species. For example, liquid and solid + R T (XNa ln XNa + XK ln XK )
MgO – CaO solutions can be modeled by + R T (XF ln XF + XCl ln XCl ) + g E
assuming an anionic sublattice occupied
only by O2 – ions, while Mg2 + and Ca2 + where the factor (XNa XCl ), for example, is
ions mix on a cationic sublattice. In this the probability, in a random mixture, of
1.10 Solution Models 63
finding a Na ion and a Cl ion as nearest accounts for the fact that the number
neighbors. Differentiation of Eq. (1-119) of Na+ – F – and K+ – Cl – nearest-neighbor
gives the following expression for the ac- pairs will be higher than the number of
tivity of NaF: such pairs in a random mixture. This term
is usually not negligible.
R T ln aNaF = – XK XCl DG exchange (1-120)
The phase diagram in Fig. 1-31 (a) was
+ R T ln (XNa XF ) + g ENaF calculated by means of Eqs. (1-119) and
(1-122) solely from optimized excess
where DG exchange is the Gibbs energy
Gibbs energies of the binary systems and
change for the following exchange reaction
from compiled data for the pure salts.
among the pure salts:
Agreement with the measured diagram is
NaCl + KF = NaF + KCl; (1-121) very good.
DG exchange
=g 0
+g 0
–g 0
– g 0KF Eqs. (1-119) and (1-122) can easily be
NaF KCl NaCl
modified for solutions in which the num-
In this example, DG exchange < 0. The salts bers of sites on the two sublattices are not
NaF and KCl are thus said to form the equal, as in MgCl2 – MgF2 – CaCl2 – CaF2
stable pair. The first term on the right of solutions. Also, in liquid salt solutions the
Eq. (1-120) is positive. The members of the ratio of the number of lattice sites on one
stable pair thus exhibit positive deviations, sublattice to that on the other sublattice
and in Fig. 1-31 this is reflected by the flat can vary with concentration, as in molten
liquidus surfaces with widely spaced iso- NaCl – MgCl2 – NaF – MgF2 solutions. In
therms for NaF and KCl. That is, the mix- this case, it has been proposed (Saboungi
ing of pure NaF and KCl is unfavorable be- and Blander, 1975) that the molar ionic
cause it involves the formation of K+ – F – fractions in all but the random mixing
and Na+ – Cl – nearest-neighbor pairs at the terms of these equations should be replaced
expense of the energetically preferable by equivalent ionic fractions. Finally, the
Na+ – F – and K+ – Cl – pairs. If DG exchange is equations can be extended to multicompo-
sufficiently large, a miscibility gap will be nent solutions. These extensions are all dis-
formed, centered close to the stable diago- cussed by Pelton (1988 b).
nal joining the stable pair. For solutions such as liquid NaF – KF –
Blander (1964) proposed the following NaCl – KCl for which DG exchange is not too
expression for g E in Eq. (1-119): large, these equations are often sufficient.
For solutions with larger exchange Gibbs
g E = XNa g ENaCl–NaF + XK g EKCl–KF (1-122) energies, however, in which liquid immis-
E E
+ XF g NaF–KF + XCl g NaCl–KCl cibility appears, these equations are gener-
– XNa XK XF XCl (DG exchange )2/ZRT ally unsatisfactory. It was recognized by
Saboungi and Blander (1974) that this is
where, for example, g ENaCl–NaF is the excess due to the effect of non-randomness upon
Gibbs energy in the NaCl – NaF binary the second nearest-neighbor cation – cation
system at the same cationic fraction XNa as and anion – anion interactions. To take
in the ternary, and where Z is the first coor- account of this, Blander proposed addi-
dination number. That is, g E contains a tional terms in Eq.(1-122). Dessureault and
contribution from each binary system. The Pelton (1991) modified Eqs. (1-119) and
final term in Eq. (1-122) is a first-order (1-122) to account more rigorously for
correction for non-random mixing which non-random mixing effects, with good re-
sults for several molten salt systems with ions occupy both sublattices. The solu-
miscibility gaps. See also section 1.10.4. tion can be represented as (Fe21–+ x Ti3x + )A
4+ 3+
(Ti1– x Tix )B where x is the overall mole
fraction of Ti2O3 . The ions are assumed to
1.10.1.3 Interstitial Solutions
mix randomly on each sublattice so that:
As an example of the application of the
sublattice model to interstitial solutions we Ds ideal = – 2 R [(1 – x) ln (1 – x) + x ln x]
will take the f.c.c. phase of the Fe –V– C (1-124)
system. Lee and Lee (1991) have modeled Deviations from ideal mixing are as-
this solution using two sublattices: a metal- sumed to occur due to interlattice cation –
lic sublattice containing Fe and V atoms, cation interactions according to
and an interstitial lattice containing C
(FeA2+ – TiB4+ ) + (TiA3+ – TiB3+ )
atoms and vacancies, va. The numbers of
sites on each sublattice are equal. An equa- = (FeA2+ – TiB3+ ) + (TiA3+ – TiB4+ )
tion identical to Eq. (1-119) can be written DG = a + b T (1-125)
for the molar Gibbs energy:
where a and b are the adjustable parameters
g = (XFe Xva g 0Feva + XFe XC g 0FeC of the model. The probability that an A – B
+ XV Xva g 0Vva + XV XC g 0VC ) (1-123) pair is an (FeA2+ – TiB3+ ) or a (TiA3+ – TiB4+ )
+ R T (XFe ln XFe + XV ln XV) pair is equal to x (1 – x). Hence, g E =
x (1 – x) (a + b T ).
+ R T (XC ln XC + Xva ln Xva ) + g E
Similar models can be proposed for
where XFe = (1 – XV) and XC = (1 – Xva ) are other ceramic solutions such as spinels,
the site fractions on the two sublattices and pseudobrookites, etc. These models can
“Feva” and “Vva” are simply pure Fe and rapidly become very complex mathemati-
V, i.e., g 0Feva = g 0Fe . An expression for g E as cally as the number of possible species on
in Eq. (1-122), although without the final the lattices increases. For instance, in
non-random mixing term, was used by Lee Fe3O4 – Co3O4 spinel solutions, Fe2+, Fe3+,
and Lee with optimized binary g E parame- Co2+ and Co3+ ions are all distributed over
ters. Their calculated Fe –V– C phase dia- both the tetrahedral and octahedral sublat-
gram is in good agreement with experimen- tices. Four independent equilibrium con-
tal data. The sublattice model has been stants are required (Pelton et al., 1979) to
similarly applied to many interstitial solu- describe the cation distribution even for the
tions by several authors. ideal mixing approximation. This com-
plexity has been rendered much more tract-
able by the “compound energy model”
1.10.1.4 Ceramic Solutions (Sundman and Ågren, 1981; Hillert et al.,
Many ceramic solutions contain two or 1988). This is not actually a model, but is
more cationic sublattices. As an example, rather a mathematical formalism permit-
consider a solution of Ti2O3 in FeTiO3 (il- ting the formulation of various models in
menite) under reducing conditions. There terms of the Gibbs energies, g 0 , of “pseu-
are two cationic sublattices, the A and B docomponents” so that equations similar to
sublattices. In FeTiO3 , Fe2 + ions and Ti 4 + Eq. (1-119) can be used directly.
ions occupy the A and B sublattices, re-
spectively. With additions of Ti2O3 , Ti 3 +
1.10.1.5 The Compound Energy multicomponent solutions. It also provides

Formalism a conceptual framework for treating many
different phases with different structures.
As an example, the model for the
This facilitates the development of com-
FeTiO3 – Ti2O3 solution in Sec. 1.10.1.4
puter software and databases because many
will be reformulated. By taking all com-
different types of solutions can be treated
binations of an A-sublattice species and
as cases of one general formalism.
a B-sublattice species, we define two
real components, (Fe2+ )A (Ti 4 + )BO3 and
(Ti3+ )A (Ti 3+ )BO3 , as well as two “pseudo- 1.10.1.6 Non-Stoichiometric Compounds
components”, (Fe2+ )A (Ti 3+ )BO3– and
3+ 4+
(Ti )A (Ti )BO3 . + Non-stoichiometric compounds are gen-
Pseudocomponents, as in the present ex- erally treated by a sublattice model. Con-
ample, may be charged. Similarly to Eq. sider such a compound A1– d B1+ d . The sub-
(1-119) the molar Gibbs energy can be lattices normally occupied by A and B at-
written oms will be called, respectively, the A-sub-
lattice and the B-sublattice. Deviations
g = (1 – x)2 g 0FeTiO3 + x 2 g 0Ti2O3 (1-126) from stoichiometry (where d = 0) can occur
+ x (1 – x) g FeTiO3– + x (1 – x) g Ti2O3+ – T Ds ideal
0 0
by the formation of defects such as B atoms
on A sites, vacant sites, atoms occupying
Note that charge neutrality is maintained in
interstitial sites, etc. Generally, one type of
Eq. (1-126). The Gibbs energies of the two
defect will predominate for solutions with
pseudocomponents are calculated from the
excess A and another type will predomi-
equation
nate for solutions with excess B. These are
DG = a + b T called the majority defects.
= g 0FeTiO3– + g 0Ti2O3+ – g 0FeTiO3 – g 0Ti2O3 (1-127) Consider first a solution in which the
majority defects are A atoms on B sites and
where DG is the Gibbs energy change of B atoms on A sites: (A1– x Bx )A (A y B1– y )B . It
Eq. (1-125) and is a parameter of the follows that d = (x – y). In the compound
model. One of g 0FeTiO3– or g 0Ti2O3+ may be as- energy formalism we can write, for the mo-
signed an arbitrary value. The other is then lar Gibbs energy,
given by Eq. (1-127). By substitution of
0 0
Eq. (1-127) into Eq. (1-126) it may be g = (1 – x) (1 – y) g AB + (1 – x) y g AA
shown that this formulation is identical to 0
+ x (1 – y) g BB 0
+ x y g BA (1-128)
the regular solution formulation given in + R T [x ln x + (1 – x) ln (1 – x)
Sec. 1.10.1.4. Note that excess terms, g E,
could be added to Eq. (1-126), thereby giv- + y ln y + (1 – y) ln (1 – y)]
ing more flexibility to the model. In the 0
where g AB is the molar Gibbs energy of
present example, however, this was not re- (hypothetical) defect-free stoichiometric
quired. AB. Now the defect concentrations at equi-
The compound energy formalism is de- librium are those that minimize g. There-
scribed and developed by Barry et al. fore, setting (∂g /∂x) = (∂g /∂y) = 0 at con-
(1992), who give many more examples. An stant d, we obtain
advantage of formulating the sublattice (1-129)
model in terms of the compound energy xy ⎛ Dg + Dg 2 ⎞
= exp ⎜ − (1 − x − y ) 1 ⎟
formalism is that it is easily extended to (1 − x ) (1 − y ) ⎝ RT ⎠
0 0
where Dg1 = (g AA – g AB ) and Dg2 = 40)) are observed. This large discrepancy
0 0
(g BB – g AB ) are the Gibbs energies of for- can be attributed to the fact that the indi-
mation of the majority defects and where vidual segments of the polymer molecule
0 0
g BA has been set equal to g AB . At a given have considerable freedom of movement.
composition d = (x – y), and for given val- Flory (1941, 1942) and Huggins (1942)
ues of the parameters Dg1 and Dg2 , Eq. proposed a model in which the polymer
(1-129) can be solved to give x and y, segments are distributed on the solvent
which can then be substituted into Eq. sites. A large polymer molecule can thus be
(1-128) to give g. The more positive are oriented (i.e. bent) in many ways, thereby
Dg1 and Dg2 , the more steeply g rises on ei- greatly increasing the entropy. To a first ap-
ther side of its minimum, and the narrower proximation the model gives an ideal mix-
is the range of stoichiometry of the coming term with mole fractions replaced by
pound. volume fractions in Eq. (1-45):
Consider another model in which the
⎛ XA v A0
majority defects are vacancies on the B- Dg mideal = RT ⎜ XA ln
sublattice and B atoms on interstitial sites. ⎝ XA v A0 + XB v B0
We now have three sublattices with occu- XB v B0 ⎞
pancies (A)A (B1– y vay )B (Bx va1– x )I where + XB ln 0 ⎟
(1-131)
XA v A + XB v A ⎠
0
“I” indicates the interstitial sublattice. The
A-sublattice is occupied exclusively by A Lewis and Randall (1961) have com-
atoms. A vacancy is indicated by va. Stoi- pared the Flory – Huggins equation with
chiometric defect-free AB is represented experimental data in several solutions.
by (A) (B) (va) and (x – y) = 2 d /(1 – d ). In general, the measured activities lie be-
Per mole of A1– d B1+ d , the Gibbs energy is: tween those predicted by Eq. (1-131) and
0 by the Raoultian ideal equation, Eq. (1-45).
g = (1 – d ) {[(1 – x) (1 – y) gABva
0 0 0
A recent review of the thermodynamics
+ (1 – x) y gAvava + x (1 – y) gABB + x y gAvaB of polymer solutions is given by Trusler
+ R T [x ln x + (1 – x) ln (1 – x) (1999).
+ y ln y + (1 – y) ln (1 – y)]} (1-130)
1.10.3 Calculation of Limiting Slopes
Eq. (1-130) is identical to Eq. (1-128) apart
of Phase Boundaries
from the factor (1 – d ), and gives rise to an
equilibrium constant as in Eq. (1-129). From the measured limiting slope
Other choices of majority defects result in (dT/dX )XA =1 of the liquidus at the melting
very similar expressions. The model can point of a pure component A, much infor-
easily be modified to account for other mation about the extent of solid solubility,
stoichiometries Am Bn , for different num- as well as the structure of the liquid, can be
bers of available interstitial sites, etc., and inferred. Similar information can be ob-
its extension to multicomponent solutions tained from the limiting slopes of phase
is straightforward. boundaries at solid-state transformation
points of pure components.
1.10.2 Polymer Solutions Eq. (1-65) relates the activities along the
liquidus and the solidus to the Gibbs en-
For solutions of polymers in monomeric
ergy of fusion:
solvents, very large deviations from simple
l
Raoultian ideal behaviour (i.e. from Eq. (1- R T ln aA – R T ln a As = – Dg0f(A) (1-132)
In the limit XA Æ 1, the liquidus and solidus Agreement with the measured limiting liq-
converge at the melting point Tf (A) . Let us uidus slope is very good, thereby showing
assume that, in the limit, Raoult’s Law, Eq. that the solid solubility of NaF in LiF is not
(1-40), holds for both phases. That is, large.
l
aA = X Al and a As = X As . Furthermore, from In the general case, the solute B may dis-
Eq. (1-60), solve to form more than one “particle”. For
example, in dilute solutions of Na2SO4 in
Dg0f(A) Æ Dhf (A) (1 – T /Tf (A) )
MgSO4 , each mole of Na2SO4 yields two
Finally, we note that moles of Na+ ions which mix randomly
with the Mg2+ ions on the cationic sublat-
lim (ln XA ) = lim (ln (1 – XB )) = – XB tice. Hence, the real mole fraction of sol-
XA Æ 1 XA Æ 1
vent, XA , is (1 – n XB ) where n is the number
Substituting these various limiting values of moles of foreign “particles” contributed
into Eq. (1-132) yields: by one mole of solute. In the present exam-
lim (dX Al /dT – dX As /dT ) ple, n = 2.
XA Æ 1 Eq. (1-133) now becomes:
= Dh0f(A) /R (Tf (A) )2 (1-133)
lim (dX Al /dT – dX As /dT )
XA Æ 1
If the limiting slope of the liquidus,
lim (dX Al /dT ), is known, then the limiting = Dh0f(A) / n R (Tf (A) )2 (1-134)
XA Æ 1
slope of the solidus can be calculated, or The broken line in Fig. 1-44 is the limiting
vice versa, as long as the enthalpy of fusion liquidus slope calculated from Eq. (1-134)
is known. under the assumption of negligible solid
For the LiF – NaF system in Fig. 1-18, solubility.
the broken line is the limiting liquidus It can be shown (Blander, 1964) that Eq.
slope at XLiF = 1 calculated from Eq. (1- (1-134) applies very generally with the fac-
133) under the assumption that there is no tor n as defined above. For example, add-
solid solubility (that is, that dX As /dT = 0). ing LiF to NaF introduces only one foreign
Figure 1-44. Phase dia-

gram of the MgSO4 –
Na2SO4 system calculated
for an ideal ionic liquid so-
lution. Broken line is the
theoretical limiting liquidus
slope calculated for negli-
gible solid solubility taking
into account the ionic nature
of the liquid. Agreement
with the measured diagram
(Ginsberg, 1909) is good.
particle Li +. The F – ion is not “foreign”. or molecule is Zi Xi . Hence, mass balance

Hence, n = 1. For additions of Na2SO4 to equations can be written as
MgSO4 , n = 2 since two moles of Na+ ions
are supplied per mole of Na2SO4 . For Z1 X1 = 2 n11 + n12
CaCl2 dissolved in water, n = 3, and so on. Z 2 X2 = 2 n22 + n12 (1-136)
For C dissolving interstitially in solid Fe, where ni j is the number of i – j bonds in one
n = 1. The fact that the solution is intersti- mole of solution. “Coordination equivalent
tial has no influence on the validity of Eq. fractions” may be defined as:
(1-134). Eq. (1-134) is thus very general
and very useful. It is independent of the so- Y1 = 1 – Y2 = Z1 X1 /(Z1 X1 + Z 2 X2 ) (1-137)
lution model and of the excess properties,
where the total number of pairs in one mole
which become zero at infinite dilution.
of solution is (Z1 X1 + Z 2 X2 )/2. Letting Xi j
An equation identical to Eq. (1-134) but
be the fraction of i – j pairs in solution, Eq.
with the enthalpy of transition, Dh0tr , re-
(1-136) may be written as:
placing the enthalpy of fusion, relates the
limiting phase boundary slopes at a trans- 2 Y1 = 2 X11 + X12
formation temperature of a component. 2 Y2 = 2 X22 + X12 (1-138)
The molar enthalpy and excess entropy of
1.10.4 Short-Range Ordering
mixing are assumed to be directly related
The basic premise of the regular solution to the number of 1– 2 pairs:
model (Sec. 1.5.7) is that random mixing
Dhm – T s E (non-config)
occurs even when gE is not zero. To ac-
count for non-random mixing, the regular = (Z1 X1 + Z 2 X2 ) X12 (w – h T )/4 (1-139)
solution model has been extended though An approximate expression for the config-
the quasichemical model for short-range urational entropy of mixing is given by a
ordering developed by Guggenheim (1935) one-dimensional Ising model:
and Fowler and Guggenheim (1939) and
modified by Pelton and Blander (1984, Dsmconfig = − R ( X1 ln X1 + X2 ln X2 )
1986) and Blander and Pelton (1987). The R
model is outlined below. For a more com- − ( Z1 X1 + Z2 X2 )
2
plete development, see the last two papers
× [ X11 ln ( X11/ Y12 ) + X22 ln ( X22 / Y22 )
cited above, Degterov and Pelton (1996),
Pelton et al. (2000) and Pelton and Char- + X12 ln ( X12 / 2 Y1 Y2 )] (1-140)
trand (2000). The equilibrium distribution is calculated
For a binary system, consider the for- by minimizing Dgm with respect to X12 at
mation of two nearest-neighbor 1– 2 pairs constant composition. This results in a
from a 1–1 and a 2 – 2 pair: “quasichemical” equilibrium constant for
(1–1) + (2 – 2) = 2 (1– 2) (1-135) the reaction, Eq. (1-135):
Let the molar Gibbs energy change for 2
X12 ⎛ − (w − h T ) ⎞
this reaction be (w – h T ). Let the nearest- = 4 exp ⎜ ⎟ (1-141)
X11 X22 ⎝ RT ⎠
neighbor coordination numbers of 1 and 2
atoms or molecules be Z1 and Z 2 . The total When (w – h T ) = 0, the solution of Eqs. (1-
number of bonds emanating from an i atom 138) and (1-141) gives a random distribu-
tion with X11 = Y12, X22 = Y22 and 2 Y1 Y2 , and the following optimized (Kongoli et al.,
Eq. (1-140) reduces to the ideal Raoultian 1998) polynomial:
entropy of mixing. When (w – h T ) be- (1-142)
comes very negative, 1– 2 pairs predomi- (w – h T ) = – (70 017 + 9 T ) – 74 042 YS
nate. A plot of Dhm or s E (non-config) versus – (798 – 15 T ) YS3 + 40 791 YS7 J/mol –1
composition then becomes V-shaped and a
config
plot of Dsm becomes m-shaped, with Far fewer parameters are required than if a
minima at Y1 = Y2 = 1/2, which is the com- polynomial expansion of gE (as in Sec.
position of maximum ordering, as illus- 1.6.2) were used. Furthermore, and more
trated in Fig. 1-45. When (w – h T ) is quite importantly, the model permits successful
negative, the plot of gE also has a negative predictions of the properties of multicom-
V-shape. ponent systems as illustrated in Fig. 1-47,
For Fe – S liquid solutions, the activity where measured sulfur activities in quater-
coefficients of sulfur as measured by sev- nary liquid Fe – Ni – Cu – S solutions are
eral authors are plotted in Fig. 1-46. It is compared with activities calculated (Kon-
clear in this case that the model should be goli et al., 1998) solely from the optimized
applied with Z Fe = Z S . The curves shown in model parameters for the Fe – S, Ni – S
Fig. 1-46 were calculated from the quasi- and Cu – S binary systems. A pair exchange
chemical model with (w – h T ) expanded as reaction like Eq. (1-135) was assumed
for each M – S pair (M = Fe, Ni, Cu), and
an optimized polynomial expansion of
(wMS – h MS T ) as a function of YS , similar to
Eq. (1-142), was obtained for each binary.
Three equilibrium constant equations like
Eq. (1-141) were written, and it was as-
sumed that (wMS – h MS T ) in the quaternary
system was constant at constant YS . No ad-
justable ternary or quaternary parameters
were required to obtain the agreement
shown in Fig. (1-47), although the model
permits the inclusion of such terms if re-
quired.
Silicate slags are known to exhibit such
short-range ordering. In the CaO – SiO2
system, Dhm has a strong negative V-shape,
as in Fig. 1-45, but with the minimum at
XSiO2 = 1/3 which is the composition corre-
sponding to Ca2SiO4 . That is, the ordering
is associated with the formation of ortho-
silicate anions SiO44 –. In the phase diagram,
Fig. 1-14, the CaO-liquidus can be seen to
Figure 1-45. Molar enthalpy and entropy of mixing descend sharply near the composition
curves for a system AB calculated at 1000 °C with
Z1 =Z2 =2 from the quasichemical model for short- XSiO2 = 1/3. The quasichemical model has
range ordering with (ω – η T ) = 0, – 21, – 42, and been extended by Pelton and Blander
– 84 kJ. (1984) to treat silicate slags. The diagram
Figure 1-46. Activity coefficient of sulfur in liquid Fe – S solutions calculated from optimized quasichemical
model parameters and comparison with experimental data (Kongoli et al., 1998).
Figure 1-47. Equilibrium partial pressure of sulfur at 1200 oC over Fe – Ni – Cu – S mattes predicted by the
quasichemical model from binary data (Kongoli et al., 1998) and comparison with experimental data
(Bale and Toguri, 1996).
shown in Fig. 1-14 is thermodynamically model more flexible by permitting the Z i to

calculated (Wu, 1990). vary with composition and by expanding
Many liquid alloy solutions exhibit the (w – h T ) as polynomials in the bond
short-range ordering. The ordering is fractions Xi j rather than the overall compo-
strongest when one component is relatively nent fractions. A merger of the quasichem-
electropositive (on the left side of the peri- ical and sublattice models has also been
odic table) and the other is relatively elec- completed (Chartrand and Pelton, 2000),
tronegative. Liquid alloys such as Alk – Au permitting nearest-neighbor and second-
(Hensel, 1979), Alk – Pb (Saboungi et al., nearest neighbor short-range-ordering to
1985) and Alk – Bi (Petric et al., 1988 a), be treated simultaneously in molten salt so-
where Alk = (Na, K, Rb, Cs), exhibit curves lutions.
of Dhm and Dsm similar to those in Fig.
1-45 with one composition of maximum
1.10.5 Long-Range Ordering
ordering. For example, in the Au – Cs sys-
tem the minima occur near the composition In solid solutions, long-range ordering
AuCs; in Mg – Bi alloys the minima occur can occur as well as short-range ordering.
near the Mg3Bi2 composition, while in In Fig. 1-15 for the Ag – Mg system, a
K – Pb alloys the maximum ordering is at transformation from an a¢ to an a phase is
K 4Pb. shown occurring at approximately 665 K
It has also been observed that certain liq- at the composition Ag3Mg. This is an
uid alloys exhibit more than one composi- order – disorder transformation. Below the
tion of ordering. For example, in K – Te al- transformation temperature, long-range or-
loys, the “excess stability function”, which dering (superlattice formation) is observed.
is the second derivative of Dgm , exhibits An order parameter may be defined which
peaks near the compositions KTe8 , KTe decreases to zero at the transformation
and K 2Te (Petric et al., 1988 b) thereby temperature. This type of phase transfor-
providing evidence of ordering centred on mation is not a first-order transformation
these compositions. The liquid might be con- like those considered so far in this chapter.
sidered as consisting of a series of mutually Unlike first-order transformations which
soluble “liquid intermetallic compounds”. involve a change of state (solid, liquid, gas)
When (w – h T ) is expanded as a polyno- and also involve diffusion over distances
mial as in Eq. (1-142), the quasichemical large compared with atomic dimensions,
model and the polynomial model of Sec. order – disorder transformations, at least at
1.6.2 become identical as (w – h T ) ap- the stoichiometric composition (Ag3Mg in
proaches zero. That is, the polynomial this example), occur by atomic rearrange-
model is a limiting case of the quasichemi- ment over distances of the order of atomic
cal model when the assumption of ideal dimensions. The slope of the curve of
configurational entropy is made. Gibbs energy versus T is not discontinuous
When (w – h T ) is positive, (1–1) and at the transformation temperature. Order-
(2 – 2) pairs predominate. The quasichemi- ing and order – disorder transformations are
cal model can thus also treat such cluster- discussed in Chapter 8 (Inden, 2001).
ing, which accompanies positive devia- A type of order – disorder transformation
tions from ideality. of importance in ferrous metallurgy is the
Recent work (Pelton et al., 2000; Pelton magnetic transformation. Below its Curie
and Chartrand, 2000) has rendered the temperature of 769 °C, Fe is ferromagnetic.
Above this temperature it is not. The trans-

formation involves a change in ordering of
the atomic spins and is not first order. Ad-
ditions of alloying elements will change
the temperature of transformation. Mag-
netic transformations are treated in Chapter
4 (Binder, 2001). See also Miodownik
(1982) and Inden (1982).
1.11 Calculation of Ternary Phase

Diagrams From Binary Data
Among 70 metallic elements 70!/3! 67!
= 54 740 ternary systems and 916 895 qua-
ternary systems are formed. In view of the
amount of work involved in measuring
even one isothermal section of a relatively
simple ternary phase diagram, it is very im-
portant to have a means of estimating ter-
nary and higher-order phase diagrams.
The most fruitful approach to such pre-
dictions is via thermodynamic methods. In
recent years, great advances have been
made in this area by the international Cal-
phad group. Many key papers have been
published in the Calphad Journal.
As a first step in the thermodynamic ap-
proach, we critically analyze the experi-
mental phase diagrams and thermodynamic Figure 1-48. Projection of the liquidus surface of
data for the three binary subsystems of the the KCl – MgCl2 – CaCl2 system.
ternary system in order to obtain a set of a) Calculated from optimized binary thermodynamic
parameters (Lin et al., 1979).
mathematical expressions for the Gibbs b) As reported by Ivanov (1953).
energies of the binary phases, as was dis-
cussed in Sec. 1.6. Next, interpolation pro-
cedures based on solution models are used
to estimate the Gibbs energies of the ter-
nary phases from the Gibbs energies of the (Ivanov, 1953) and estimated (Lin et al.,
binary phases. Finally, the ternary phase di- 1979) liquidus projections of the KCl –
agram is calculated by computer from MgCl2 – CaCl2 system are shown in Fig.
these estimated ternary Gibbs energies by 1-48. The estimated phase diagram was
means of common tangent plane or total calculated from the thermodynamic prop-
Gibbs energy minimization algorithms. erties of the three binary subsystems with
As an example of such an estimation of the Gibbs energy of the ternary liquid ap-
a ternary phase diagram, the experimental proximated via the equation suggested by
1.11 Calculation of Ternary Phase Diagrams From Binary Data 73
Kohler (1960): tal effort involved in determining multi-

component phase diagrams.
g E = (1 – XA )2 g B/C
E
+ (1 – XB )2 g C/A
E
Reviews of various interpolation proce-
+ (1 – XC )2 g A/B
E
(1-143) dures and computer techniques for estimat-
ing and calculating ternary and higher-or-
In this equation, g E is the excess molar der phase diagrams are given by Ansara
Gibbs energy at a composition point in the (1979), Spencer and Barin (1979) and Pel-
E E E
ternary liquid phase and g B/C , g C/A and g A/B ton (1997).
are the excess Gibbs energies in the three Other equations, similar to the Kohler
binary systems at the same ratios XB /XC , Eq. (1-143) in that they are based on exten-
XC /XA and XA /XB as at the ternary point. If sion of regular solution theory, are used to
the ternary liquid phase as well as the three estimate the thermodynamic properties of
binary liquid phases are all regular solu- ternary solutions from the properties of the
tions, then Eq. (1-143) is exact. In the binary subsystems. For a discussion and
general case, a physical interpretation of references, see Hillert (1980). However,
Eq. (1-143) is that the contribution to g E for structurally more complex solutions
from, say, pair interactions between A and involving more than one sublattice or with
B particles is constant at a constant ratio significant structural ordering, other esti-
XA /XB apart from the dilutive effect of the mation techniques must be used. For a re-
C particles, which is accounted for by the view, see Pelton (1997).
term (1 – XC )2 taken from regular solution An example, the calculation of the
theory. phase diagram of the NaCl – KCl – NaF – KF
Ternary phase diagrams estimated in this system in Fig. 1-31, has already been pre-
way are quite acceptable for many pur- sented in Sec. 1.10.1.2.
poses. The agreement between the experi- The quasichemical model for systems
mental and calculated diagrams can be with short-range ordering was discussed
greatly improved by the inclusion of one for the case of binary systems in Sec.
or two “ternary terms” with adjustable 1.10.4. The model has been extended to
coefficients in the interpolation equations permit the estimation of ternary and multi-
for g E. For example, the ternary term component phase diagrams (Pelton and
a XKCl XMgCl2 XCaCl2 , which is zero in all Blander, 1986; Blander and Pelton, 1987;
three binaries, could be added to Eq. (1- Pelton and Chartrand, 2000). Very good re-
143) and the value of the parameter a sults have been obtained in the case of sili-
which gives the “best” fit to the measured cate systems. The liquidus surface of the
ternary liquidus could be determined. This, SiO2 – MgO – MnO system, estimated from
of course, requires that ternary measure- optimized binary data with the quasichem-
ments be made, but only a very few (even ical model for the liquid and under the
one or two in this example) experimental assumption of ideal mixing for the solid
liquidus points will usually suffice rather MgSiO3 – MnSiO3 and Mg2SiO4 – Mn2SiO4
than the larger number of measurements solutions, is shown in Fig. 1-49. Agree-
required for a fully experimental determi- ment with the measured phase diagram
nation. In this way, the coupling of the (Glasser and Osborn, 1960) is within ex-
thermodynamic approach with a few well perimental error limits.
chosen experimental measurements holds
promise of greatly reducing the experimen-
Figure 1-49. Projection of liquidus surface of the SiO2 – MgO – MnO system calculated from optimized binary
parameters with the quasichemical model for the liquid phase.
1.12 Minimization of Gibbs solution phases a, b, g, … . The total Gibbs

Energy energy of the system may be written as:
G = (nA g A0 + nB g B0 + …)
Throughout this chapter it has been
+ (na ga + nb gb + …) (1-144)
shown that phase equilibria are calculated
by Gibbs energy minimization. Computer where nA , nB , etc. are the number of moles
software has been developed in recent of the pure solid or liquidus; g A0 , g B0 , etc.
years to perform such calculations in sys- are the molar Gibbs energies of the pure
tems of any number of components, phases solids or liquids (which are functions of T
and species. and P); na , nb , etc. are the total number of
Consider a system in which several stoi- moles of the solution phases; g a , g b , etc.
chiometric solid or liquid compounds A, B, are the molar Gibbs energies of the solution
C, … could be present at equilibrium along phases (which are function of T, P and
with a number of gaseous, liquid or solid composition). For a given set of constraints
1.12 Minimization of Gibbs Energy 75
(such as fixed T, P and overall composi- has been asked to calculate the equilibrium
tion), the free energy minimization algo- state when 1 mol of SiI 4 is held at 1400 K
rithms find the set of mole numbers nA , nB , in a volume of 104 l. The thermodynamic
etc., na , nb , etc. (some may be zero) as well properties of the possible product species
as the compositions of all solution phases have been automatically retrieved from
which globally minimize G. This is the the database and the total Gibbs energy
equilibrium phase assemblage. Other con- has been minimized by the Solgasmix
straints such as constant volume or a fixed algorithm. At equilibrium there will be
chemical potential (such as constant pO2 ) 2.9254 mol of gas of the composition
may be applied. shown and 0.11182 mol of solid Si will
A discussion of the strategies of such al- precipitate. The total pressure will be
gorithms is beyond the scope of the present 0.0336 bar.
chapter. One of the best known general Although the calculation was performed
Gibbs energy minimization programs is by minimization of the total Gibbs energy,
Solgasmix written by Eriksson (1975) and substitution of the results into the equilib-
constantly updated. rium constants of Eqs. (1-10) to (1-12) will
When coupled to a large thermodynamic show that these equilibrium constants are
database, general Gibbs energy minimiza- satisfied.
tion programs provide a powerful tool for Another example is shown in Table 1-3
the calculation of phase equilibria. Several (Pelton et al., 1990). Here the program has
such expert database systems have been been asked to calculate the equilibrium
developed. They have been reviewed by
Bale and Eriksson (1990).
An example of a calculation performed
Table 1-3. Calculation of equilibrium state when re-
by the F * A * C * T (Facility for the Analy-
actants shown (masses in g) are held at 1873 K at a
sis of Chemical Thermodynamics) expert pressure of 1 atm. Calculations performed by mini-
system, which the author has helped to de- mization of the total Gibbs energy.
velop, is shown in Table 1-2. The program
100. Fe + 0.08 O + 0.4 Fe + 0.4 Mn + 0.3 Si + 0.08 Ar =
0.30793 litre ( 99.943 vol% Ar
+ 0.24987E – 01 vol% Mn
Table 1-2. Calculation of equilibrium state when + 0.24069E – 01 vol% SiO
1 mole SiI4 is held at 1400 K in a volume of 104 l. + 0.82057E – 02 vol% Fe
Calculations performed by minimization of the total + 0.79044E – 07 vol% O
Gibbs energy. + 0.60192E – 08 vol% Si
+ 0.11200E – 08 vol% O2
SiI4 =
+ 0.35385E – 15 vol% Si2 )
2.9254 ( 0.67156 I
(1873.0, 1.00 ,G)
+ 0.28415 SiI2
+ 0.24835E – 01 I2 + 0.18501 gram ( 49.948 wt.% SiO2
+ 0.19446E – 01 SiI4 + 42.104 wt.% MnO
+ 0.59083E – 05 SiI + 7.9478 wt.% FeO)
+ 0.23039E – 07 Si (1873.0, 1.00 ,SOLN 2)
+ 0.15226E – 10 Si2 + 100.99 gram ( 99.400 wt.% Fe
+ 0.21462EE – 11 Si3) + 0.33630 wt.% Mu
(1400.0,0.336E – 01,G) + 0.25426 wt.% Si
+ 0.11182 Si + 0.98375E – 02 wt.% O2)
(1400.0,0.336E – 01,S1, 1.0000) (1873.0, 1.00 ,SOLN 3)
state when 100 g Fe, 0.08 g oxygen, 0.4 g calculate all types of phase diagrams with
Fe, 0.4 g Mn, 0.3 g Si and 0.08 g Ar are any properly chosen variables as axes or
brought together at 1873 K at a total pres- constants.
sure of 1 bar. The database contains data
for a large number of solution phases as
well as for pure compounds. These data 1.13 Bibliography
have been automatically retrieved and the
total Gibbs energy has been minimized. At 1.13.1 Phase Diagram Compilations
equilibrium there are 0.30793 l of a gas
phase, 0.18501 g of a molten slag, and The classic compilation in the field of bi-
100.99 g of a molten steel of the composi- nary alloy phase diagrams is that of Hansen
tions shown. (1958). This work was continued by Elliott
The Gibbs energies of the solution (1965) and Shunk (1969). These compila-
phases are represented as functions of com- tions contain critical commentaries. A non-
position by various solution models (Sec. critical compilation of binary alloy phase
1.10). As discussed in Sec. 1.11, these diagrams is supplied in looseleaf form with
models can be used to predict the thermo- a continual up-dating service by W.G. Mof-
dynamic properties of N-component solu- fatt of the General Electric Co., Schenec-
tions from evaluated parameters for binary tady, N.Y. An extensive non-critical compi-
(and possibly ternary) subsystems stored in lation of binary and ternary phase diagrams
the database. For example, in the calcula- of metallic systems has been edited by
tion in Table 1-3, the Gibbs energy of the Ageev (1959 – 1978). An index to all com-
molten slag phase was estimated by the pilations of binary alloy phase diagrams up
quasichemical model from optimized pa- to 1979 was prepared by Moffatt (1979). A
rameters for the binary oxide solutions. critical compilation of binary phase dia-
grams involving Fe has been published by
Kubaschewski (1982). Ternary alloy phase
1.12.1 Phase Diagram Calculation
diagrams were compiled by Ageev (1959 –
Gibbs energy minimization is used to 1978).
calculate general phase diagram sections From 1979 to the early 1990s, the Amer-
thermodynamically using the zero phase ican Society for Metals undertook a project
fraction line concept (Sec. 1.9.1.1), with to evaluate critically all binary and ternary
data retrieved from databases of model co- alloy phase diagrams. All available litera-
efficients. For example, to calculate the di- ture on phase equilibria, crystal structures,
agram of Fig. 1-30, the program first scans and often thermodynamic properties were
the four edges of the diagram to find the critically evaluated in detail by interna-
ends of the ZPF lines. Each line is then fol- tional experts. Many evaluations have ap-
lowed from beginning to end, using Gibbs peared in the Journal of Phase Equilibria
energy minimization to determine the point (formerly Bulletin of Alloy Phase Dia-
at which a phase is just on the verge of be- grams), (ASM Int’l., Materials Park, OH),
ing present. When ZPF lines for all phases which continues to publish phase diagram
have been drawn, then the diagram is com- evaluations. Condensed critical evalua-
plete. Because, as shown in Sec. 1.9, all tions of 4700 binary alloy phase diagrams
true phase diagram sections obey the same have been published in three volumes
geometrical rules, one algorithm suffices to (Massalski et al., 1990). The ternary phase
1.13 Bibliography 77
diagrams of 7380 alloy systems have also Compilations of properties of solutions

been published in a 10-volume compilation (activities, enthalpies of mixing, etc.) are
(Villars et al., 1995). Both binary and ter- much more difficult to find. Hultgren et al.
nary compilations are available from ASM (1973) present the properties of a number
on CD-ROM. Many of the evaluations of binary alloy solutions. An extensive bib-
have also been published by ASM as liography of solution properties of all types
monographs on phase diagrams involving a of solutions was prepared by Wisniak and
particular metal as a component. Tamir (1978).
Each year, MSI Services (http://www. Thermodynamic/phase diagram optimi-
msiwp.com) publishes The Red Book, zation as discussed in Sec. 1.6.1 has been
which contains abstracts on alloy phase carried out for a large number of alloy, ce-
diagrams from all sources, notably from ramic and other systems. Many of these
the extensive Russian literature. MSI also evaluations have been published in the
provides a CD-ROM with extensive alloy international Calphad Journal, published
phase diagram compilations and reports. since 1977 by Pergamon Press. Several of
Phase diagrams for over 9000 binary, ter- the evaluations in the Journal of Phase
nary and multicomponent ceramic systems Equilibria discussed above include ther-
(including oxides, halides, carbonates, sul- modynamic/phase diagram optimizations,
fates, etc.) have been compiled in the 12- as do a number of the evaluations in Vol. 7
volume series, Phase Diagrams for Ceram- of Phase Diagrams for Ceramists.
ists (1964–96, Am. Ceramic Soc., Colum- Extensive computer databases of the
bus, OH). Earlier volumes were non-criti- thermodynamic properties of compounds
cal compilations. However, recent volumes and solutions (stored as coefficients of
have included critical commentaries. model equations) are available. These in-
Phase diagrams of anhydrous salt sys- clude F * A * C * T (http://www.crct.poly-
tems have been compiled by Voskresen- mtl.ca), Thermocalc (http://www.met.
skaya (1970) and Robertson (1966). kth.se), ChemSage (http://gttserv.lth.rwth-
An extensive bibliography of binary aachen.de), MTS-NPL (http://www.npl.
and multicomponent phase diagrams of all co.uk), Thermodata (http://www.grenet.fr),
types of systems (metallic, ceramic, aque- HSC (http://www.outokumpu.fi), and
ous, organic, etc.) has been compiled by MALT2 (http://www. kagaku.com). Gibbs
Wisniak (1981). energy minimization software permits the
calculation of complex equilibria from the
stored data as discussed in Sec. 1.12 as
1.13.2 Thermodynamic Compilations
well as the thermodynamic calculation of
Several extensive compilations of ther- phase diagram sections. A listing of these
modynamic data of pure substances of and other available databases is maintained
interest in materials science are available. at http://www.crct. polymtl.ca .
These include the JANAF Tables (Chase A bibliographic database known as
et al., 1985) and the compilations of Barin Thermdoc, on thermodynamic properties
et al. (1977), Barin (1989), Robie et al. and phase diagrams of systems of interest
(1978) and Mills (1974), as well as the to materials scientists, with updates, is
series of compilations of the National Insti- available through Thermodata (http://
tute of Standards and Technology (Wash- www.grenet.fr).
ington, D.C.).
1.13.3 General Reading Alper, A. M. (Ed.) (1970–1978), Phase Diagrams –

Materials Science and Technology, Vols. 1 – 5.
The theory, measurement and applica- New York: Academic.
Ansara, I. (1979), Internat. Met. Rev. 238, No. 1, 20.
tions of phase diagrams are discussed in a Bale, C. W. (1990), Canad. Metall. Quart. 29, 263.
great many texts. Only a few can be listed Bale, C. W., Eriksson, N. G. (1990), Canad. Metall.
here. A recent text by Hillert (1998) pro- Quart. 29, 105.
Bale, C. W., Pelton, A. D. (1983), Metall. Trans.
vides a complete thermodynamic treatment B 14, 77.
of phase equilibria as well as solution mod- Bale, C. W., Toguri, J. (1996), private communica-
eling and thermodynamic/phase diagram tion.
Bale, C. W., Pelton, A. D., Thompson W. T. (1986),
optimization. Canad. Metall. Quart. 25, 107.
A classical discussion of phase diagrams Barin, I. (1989), Thermochemical Data of Pure Sub-
in metallurgy was given by Rhines (1956). stances. 2. Weinheim: VCH Verlagsgesellschaft.
Barin, I., Knacke, O., Kubaschewski, O. (1977),
Prince (1966) presents a detailed treatment Thermochemical Properties of Inorganic Sub-
of the geometry of multicomponent phase stances. Berlin: Springer Verlag.
diagrams. A series of five volumes edited Barry, T. I., Dinsdale, A. T., Gisby, J. A., Hallstedt,
B., Hillert, M., Jansson, B., Jonsson, J., Sundman,
by Alper (1970 – 1978) discusses many as- B., Taylor, J. R. (1992), J. Phase Equilib. 13, 459.
pects of phase diagrams in materials sci- Bartlett, H. E. Neethling, A. J., Crowther, P. (1970),
ence. Bergeron and Risbud (1984) give an J. Chem. Thermo. 2, 523.
introduction of phase diagrams, with par- Bergeron, C. J., Risbud, S. H. (1984), Introduction to
Phase Equilibria in Ceramics. Columbus, Ohio:
ticular attention to applications in ceramic Amer. Ceramic Soc.
systems, see also Findlay (1951), Ricci Binder, K. (2001), in: Phase Transformations in Ma-
(1964) and West (1965). terials: Kostorz, G. (Ed.). Weinheim: Wiley-VCH,
Chap. 4.
In the Calphad Journal and in the Jour- Binder, K., Fratzl, P. (2001), in: Phase Transforma-
nal of Phase Equilibria are to be found tions in Materials: Kostorz, G. (Ed.). Weinheim:
many articles on the relationships between Wiley-VCH, Chap. 6.
Blander, M. (1964), Molten Salt Chemistry. New
thermodynamics and phase diagrams. York: Interscience, Chap. 3.
It has been beyond the scope of the Blander, M., Pelton, A. D. (1987), Geochim. et Cos-
present chapter to discuss experimental mochim. Acta 51, 85.
Bray, H. F., Bell, F. D., Harris, S. J. (1961 – 1962),
techniques of measuring thermodynamic J. Inst. Metals 90, 24.
properties and phase diagrams. For the Buckley, R. A. (1970), in: Techniques of Metals Re-
measurement of thermodynamic proper- search, Vol. IV, Part 1: Rapp, R. A. (Ed.). New
York: Interscience.
ties, including properties of solutions, the Chartrand, P., Pelton, A. D. (2000), Metall. Mater.
reader is referred to Kubaschewski and Trans., in press.
Alcock (1979). For techniques of measur- Chase, M. W. (1983), Bull. Alloy Phase Diag. 4, 124.
Chase, M. W., Davies, C. A., Downey, J. R., Frurip,
ing phase diagrams, see Pelton (1996), D. J., McDonald, R. A., Syverud, A. N. (1985),
Raynor (1970), MacChesney and Rosen- JANAF Thermochemical Tables, 3rd ed. New
berg (1970), Buckley (1970) and Hume- York: Amer. Inst. Physics.
Chu, W. F., Rahmel, A. (1979), Rev. High-Temp.
Rothery et al. (1952). Mater. IV, 139.
Degterov, S., Pelton, A. D. (1996), J. Phase Equilib.
17, 488.
Dessureault, Y., Pelton, A. D. (1991), J. Chim. Phys.
1.14 References 88, 1811.
Dessureault, Y., Sangster, J., Pelton, A. D. (1990),
Ageev, N. V. (Ed.) (1959–1978), Phase Diagrams of J. Electrochem. Soc. 137, 2941.
Metallic Systems, Vols. 1–22. Moscow: Acad. Sci. Doman, R. C., Barr, J. B., McNally, R. N., Alper, A.
USSR. M. (1963), J. Am. Ceram. Soc. 46, 313.
Alabyshev, A. F., Morachevskii, A. G. (1957), Z. Dörner, P., Henig, E.-Th., Krieg, H., Lucas, H. L.,
Neorg. Khim. 2, 669. Petzow, G. (1980), Calphad Journal 4, 241.
1.14 References 79
Elliott, R. P. (1965), Constitution of Binary Alloys. Katsura, T., Muan, A. (1964), TMS-AIME 230, 77.
First Supplement. New York: McGraw-Hill. Kellogg, H. H., Basu, S. K. (1960), TMS-AIME 218,
Eriksson, N. G. (1975), Chemical Scripta 8, 100. 70.
Findlay, A. (1951), The Phase Rule, 9th edn., revised Kleinstuber, T. (1961), Ph.D. Thesis, Univ. Munich,
by Campbell, A. N., Smith, N. A. New York: Germany.
Dover. Kohler, F. (1960), Monatsh. Chemie 91, 738.
Flory, P. J. (1941), J. Chem. Phys. 9, 660. Köster, W., Dullenkopf, W. (1936), Z. Metallk. 28,
Flory, P. J. (1942), J. Chem. Phys. 10, 5. 309.
Fowler, R. H., Guggenheim, E. A. (1939), Statistical Kongoli, F., Dessureault, Y., Pelton, A. D. (1998),
Thermodynamics. Cambridge: Cambridge Univ. Metall. and Mat. Trans. 29B, 591.
Press, p. 350. Kubaschewski, O. (1982), Iron-Binary Phase Dia-
Garrels, R. M., Christ, C. L. (1965), Solutions Miner- grams. New York: Springer Verlag.
als and Equilibria. New York: Harper and Row. Kubaschewski, O., Alcock, C. B., (1979), Metallur-
Ginsberg, A. S. (1909), Z. Anorg. Chem. 61, 130. gical Thermochemistry, 5th ed. New York: Per-
Glasser, E. P., Osborn, E. F. (1960), J. Amer. Ceram. gamon Press.
Soc. 43, 136. Kurnakow, N. S., Kusnetzow, A. N. (1907), Z. Anorg.
Guggenheim, E. A. (1935), Proc. Roy. Soc. A 148, Chem. 52, 173.
304. Kunz, M. (2001), in: Phase Transformations in Ma-
Gulbransen, E. A., Jansson, S. A. (1970), Proc. terials: G. Kostorz (Ed.). Weinheim: Wiley-VCH,
Symp. Electrochem. Soc. Fall meeting 1969. New Chap. 10.
York: The Electrochem. Soc., p. 3. Lantratov, M. F., Mikhailova, A. G. (1971), Zh. Prikl.
Gupta, H., Morral, J. E., Nowotny, H. (1986), Scripta Khimii 44, 1778.
Metall. 20, 889. Lee, B.-J., Lee, D. N. (1991), Calphad 15, 293.
Hansen, M. (1958), Constitution of Binary Alloys, Lee, B. J., Lee, D. N. (1992), J. Phase Equilib. 13, 349.
2nd edn. New York: McGraw-Hill. Levin, E. M., Robbins, C. R., McMurdie, H. F. (1964),
Harshe, G., Venkatachalam, S. (1984), J. Metals 36, and 11 supplements to (1996), Phase Diagrams for
34. Ceramists; Columbus, Ohio: Am. Ceramic Soc.
Hauffe, K. (1940), Z. Elektrochem. 46, 348. Lewis, G. N., Randall, M. (1961), Thermodynamics,
Hensel, F. (1979), Adv. Phys. 28, 555. revised by Pitzer, K. S., Brewer, L. New York:
Hillert, M. (1980), Calphad Journal 4, 1. McGraw-Hill.
Hillert, M. (1985), Int. Metall. Rev. 30, 45. Lin, P.-L., Pelton, A. D., Bale, C. W. (1979), J. Am.
Hillert, M. (1997), J. Phase Equilib. 18, 249. Ceram. Soc. 62, 414.
Hillert, M. (1998), Phase Equilibria, Phase Dia- Lipson, H., Wilson, A. J. C. (1940), J. Iron Steel Inst.
grams and Phase Transformations. Cambridge: 142, 122.
Cambridge University Press. Lukas, H. L., Henig, E.-Th., Zimmermann, B.
Hillert, M., Jansson, B., Sundman, B. (1988), Z. (1977), Calphad Journal 1, 225.
Metallk. 79, 81. MacChesney, J. B., Rosenberg, P. E. (1970), in Phase
Holm, J. L. (1965), Acta Chem. Scand. 19, 638. Diagrams – Materials Science and Technology,
Hong, K. C., Kleppa, O. J. (1976), J. Chem. Thermo- Vol. 1: Alper, A. M. (Ed.). New York: Academic,
dyn. 8, 31. Ch. 3.
Huggins, M. L. (1942), Ann. N. Y. Acad. Sci. 43, 1. Maiorova, E. A., Morachevskii, A. G., Kovalenko, S.
Hultgren, R., Desai, P. D., Hawkins, D. T., Gleiser, G. (1976), Elektrokhimiya 12, 313.
M., Kelly, K. K., Wagman, D. D. (1973), Selected Massalski, T. B., Okamoto, H., Subramanian, P. R.,
Values of the Thermodynamic Properties of the Kacprzak, L. (1990), Binary Alloy Phase Dia-
Elements and Binary Alloys. Metals Park, Ohio: grams, 2nd ed. Metals Park, OH: Am. Soc. Metals.
Am. Soc. of Metals. Mathewson, C. H. (1906), Z. Ang. Chem. 50, 180.
Hume-Rothery, W., Christian, J. W., Pearson, W. B. Mills, K. C. (1974), Thermodynamic Data for Inor-
(1952), Metallurgical Equilibrium Diagrams. ganic Sulphides, Selenides and Tellurides. Lon-
London: Inst. Phys. don: Butterworths.
Inden, G. (1982), Bull. Alloy Phase Diag. 2, 412. Miodownik, A. P. (1982), Bull. Alloy Phase Diag. 2,
Inden, G. (2001), in: Phase Transformations in Mate- 406.
rials: G. Kostorz (Ed.). Weinheim: Wiley-VCH, Moffatt, W. B. (1979), The Index to Binary Phase Col-
Chap. 8. lections, Schenectady. N.Y.: General Electric Corp.
Ingraham, T. R., Kellogg, H. H. (1963), TMS-AIME Muan, A., and Osborn, F. (1965), Phase Equilibria
227, 1419. Among Oxides in Steelmaking. Reading, MA.:
Ingraham, T. R., Kerby, R. (1967), Canad. Metall. Addison Wesley.
Quart. 6, 89. Müller-Krumbhaar, H., Kurz, W., Brener, E. (2001),
Ivanov, A. I. (1953), Izv. Sekt. Fiz.-Khim. Anal. Inst. in: Phase Transformations in Materials: G. Kostorz
Obshchei Neorg. Khim. Akad. Nauk SSSR 23, 197. (Ed.). Weinheim: Wiley-VCH, Chap. 2.
Palatnik, L. S., Landau, A. I. (1964), Phase Equilib- Robertson, W. D. (Ed.) (1966), Binary Phase Dia-
ria in Multicomponent Systems. NY: Holt, Rine- grams of Halide Salts, U. S. At. En. Comm. Con-
hart and Winston. tract AT (30-1) 2723, Washington: Clearinghouse
Pehlke, R. D. (1973), Unit Processes of Extractive for Federal Scientific and Techn. Info.
Metallurgy. New York: Elsevier. Robie, R. A., Hemingway, B. S., Fisher, J. R. (1978),
Pei, B., Björkman, B., Sundman, B., Jansson, B. Thermodynamic Properties of Minerals and Re-
(1995), Calphad 19, 1. lated Substances at 298.15 K and 1 Bar Pressure
Pelton, A. D. (1983), in: Physical Metallurgy: Cahn, and at Higher Temperatures. Washington: U.S.
R. W., Haasen, P. (Eds.). Amsterdam: Elsevier, Government Printing Office.
Chap. 7. Roos, G. D. (1916), Z. Anorg. Chem. 94, 329.
Pelton, A. D. (1988 a), Bull. Alloy Phase Diag. 9, 41. Saboungi, M.-L., Blander, M. (1974), High Temp.
Pelton, A. D. (1988 b), Calphad Journal 12, 127. Sci. 6, 37.
Pelton, A. D. (1995), J. Phase Equilib. 16, 501. Saboungi, M.-L., Blander, M. (1975), J. Am. Ceram.
Pelton, A. D. (1996), in: Physical Metallurgy: Cahn, Soc. 58, 1.
R. W., Haasen, P. (Eds.). Amsterdam, North-Hol- Saboungi, M.-L., Herron, S. J., Kumar, R. (1985),
land, Chap. 6. Ber. Bunsenges. Phys. Chem. 89, 375.
Pelton, A. D. (1997), in: Advanced Physical Chemis- Sangster, J., Pelton, A. D. (1987), J. Phys. Chem.
try for Process Metallurgy: Sano, N., Lu, W.-K, Ref. Data 16, 509.
Riboud P. (Eds.). New York, Academic, Chap. 3. Schenk, H., Frohberg, M. G., Nunninghof, R. (1964),
Pelton, A. D., Bale, C. W. (1986), Metall. Trans. Arch. Eisenhütten 35, 269.
A 17, 1057. Schreinemakers, F. A. H. (1915), Proc. K. Akad.
Pelton, A. D., Blander, M. (1984), Proc. AIME Symp. Wetenschappen, Amsterdam (Section of Sciences)
on Molten Salts and Slags. Warrendale, Pa.: The 18, 116.
Metall. Soc. AIME, p. 281. Shatynski, S. R. (1977), Oxidn. Metals 11, 307.
Pelton, A. D., Blander, M. (1986), Metall. Trans. Shunk, F. A. (1969), Constitution of Binary Alloys.
B 17, 805. Second Supplement. New York: McGraw-Hill.
Pelton, A. D., Chartrand, P. (2000), Metall. Mater. Spencer, P. J., Barin, I. (1979), Mater. Eng. Appl. 1,
Trans., in press. 167.
Pelton, A. D., Degterov, S. A., Eriksson, G., Robelin, Stringer, J., Whittle, D. P. (1977), Rev. Int. Htes.
C., Dessureault, Y. (2000), Metall. Mater. Trans., Temp. Réfract. 14, 6.
B 31, 651. Sundman, B., Ågren, J. (1981), J. Phys. Chem. Solids
Pelton, A. D., Schmalzried, H. (1973), Metall. Trans. 42, 297.
4, 1395. Trusler, J. P. M. (1999), in: Chemical Thermodynam-
Pelton, A. D., Schmalzried, H., Sticher, J. (1979), ics: Letcher, T. (Ed.). Abingdon, Oxon, UK:
J. Phys. Chem. Solids 40, 1103. Blackwell Science, Chap. 16.
Pelton, A. D., Thompson, W. T. (1975), Prog. Solid Van Laar, J. J. (1908), Z. Phys. Chem. 63, 216; 64,
State Chem. 10, part 3, 119. 257.
Pelton, A. D., Thompson, W. T., Bale, C. W., Erik- Villars, P., Prince, A., Okamoto, H. (1995), Hand-
son, N. G. (1990), High Temp. Sci. 26, 231. book of Ternary Alloy Phase Diagrams. Metals
Petric, A., Pelton, A. D., Saboungi, M.-L. (1988 a), Park, OH: Am. Soc. Metals.
J. Electrochem. Soc. 135, 2754. Voskresenskaya, N. K. (Ed.) (1970), Handbook of
Petric, A., Pelton, A. D., Saboungi, M.-L. (1988 b), Solid-Liquid Equilibria in Systems of Anhydrous
J. Chem. Phys. 85, 5070. Inorganic Salts, Vol. 1, TT69-55079/1, AEC-tr-
Phillips, B., Muan, A. (1958), J. Am. Ceram. Soc. 41, 69831/3, UC-4, Washington: U.S. At. En. Comm.
448. and Nat’l. Sci. Foundation.
Pilgrim, R. F., Ingraham, T. R. (1967), Canad. Weeks, J. R. Davies, H. A. (1964), AEC Report,
Metall. Quart. 6, 333. Conf. 660712-1, BNL-10372.
Polyakov, V. D. (1940), Izv. Sektora Fiz.-Khim Anal- West, D. R. F. (1965), Ternary Equilibrium Dia-
iza Inst. Obshchei Neorg. Khim. Akad. Nauk SSSR grams. New York: McMillan.
13, 308. Wisniak, J. (1981), Phase Diagrams: A Literature
Prince, A. (1963), Metall. Rev. 8, 213. Source Book, Vols. 1 and 2. New York: Elsevier.
Prince, A. (1966), Alloy Phase Equilibria. Amster- Wisniak, J., Tamir, A. (1978), Mixing and Excess
dam: Elsevier. Thermodynamic Properties: A Literature Source
Raynor, G. V. (1970), in: Physical Metallurgy, 2nd Book. New York: Elsevier.
edn: Cahn, R. W. (Ed.). Amsterdam: North Hol- Wu. P. (1990), Ph. D. thesis, Ecole Polytechnique,
land, Chap. 7. Montreal.
Rhines, F. N. (1956), Phase Diagrams in Metallurgy. Wu, P., Eriksson, G., Pelton, A. D. (1993), J. Am.
New York: McGraw-Hill. Ceram. Soc. 76, 2065.
Ricci, J. E. (1964), in: Molten Salt Chemistry: Yokokowa, H. (1999), J. Phase Equilib. 20, 258.
Blander, M. (Ed.). New York: Interscience, Chap. 4.
2 Solidification
Heiner Müller-Krumbhaar
Institut für Festkörperforschung, Forschungszentrum Jülich, Germany
Wilfried Kurz
Département des Matériaux, EPFL, Lausanne, Switzerland
Efim Brener
Institut für Festkörperforschung, Forschungszentrum Jülich, Germany
List of Symbols and Abbreviations . . . . . . . . . . . . . . . . . . . . . . . . . . 82

2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
2.2 Basic Concepts in First-Order Phase Transitions . . . . . . . . . . . . 86
2.2.1 Nucleation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
2.2.2 Interface Propagation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
2.2.3 Growth of Simple Crystal Forms . . . . . . . . . . . . . . . . . . . . . . . 91
2.2.4 Mullins – Sekerka Instability . . . . . . . . . . . . . . . . . . . . . . . . . 93
2.3 Basic Experimental Techniques . . . . . . . . . . . . . . . . . . . . . . . 95
2.3.1 Free Growth . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
2.3.2 Directional Growth . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
2.4 Free Dendritic Growth . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
2.4.1 The Needle Crystal Solution . . . . . . . . . . . . . . . . . . . . . . . . . 101
2.4.2 Side-Branching Dendrites . . . . . . . . . . . . . . . . . . . . . . . . . . 108
2.4.3 Experimental Results on Free Dendritic Growth . . . . . . . . . . . . . . 115
2.5 Directional Solidification . . . . . . . . . . . . . . . . . . . . . . . . . . 120
2.5.1 Thermodynamics of Two-Component Systems . . . . . . . . . . . . . . . 121
2.5.2 Scaled Model Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
2.5.3 Cellular Growth . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
2.5.4 Directional Dendritic Growth . . . . . . . . . . . . . . . . . . . . . . . . 131
2.5.5 The Selection Problem of Primary Cell Spacing . . . . . . . . . . . . . . 135
2.5.6 Experimental Results on Directional Dendritic Growth . . . . . . . . . . . 139
2.5.7 Extensions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
2.6 Eutectic Growth . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152
2.6.1 Basic Concepts. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152
2.6.2 Experimental Results on Eutectic Growth . . . . . . . . . . . . . . . . . . 158
2.6.3 Other Topics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163
2.7 Summary and Outlook . . . . . . . . . . . . . . . . . . . . . . . . . . . 164
2.8 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165

ISBN: 3-527-30256-5
82 2 Solidification

a distance
A area
A amplitude
Â differential operator
aI , aK constants
b arbitrary parameter
B̂ self-adjoint differential operator
DC miscibility gap
C, C ¢ constants
CE eutectic concentration
CL , CS , Ca , etc. concentration
cp specific heat at constant pressure
d atomistic length, capillary length
D diffusion constant
d0 capillary length
Di transport coefficients, diffusion constants
DT thermal diffusion coefficient
E1 (P) exponential integral
erfc error function complement
F Helmholtz energy
f scaling function
g Gibbs energy density
DG change in Gibbs energy
Ĝ Gibbs energy per surface element
G ({x i }) Gibbs energy
GT constant temperature gradient
J probability current
K curvature
k segregation coefficient, wave number
L natural scale
l external length, diffusion length
l˜ thickness of layer, diffusion length
Lm latent heat of melting
ma, mb liquidus or solidus slope
Ni particle number of species i
P pressure, Péclet number
ᏼ principal value
P ({x i }) probability of configuration
q wave number, inverse length
Q min minimal work
r radius of nucleus, radial distance
R (S) local radius of curvature at point S
R0 radius of curvature
rc critical radius
S position on the interface, entropy
DS change in entropy
T temperature
t time
Dt change in time
T0 reference temperature
TE eutectic temperature
TI interface temperature
Tm melting temperature
u dimensionless temperature field or concentration field
U energy
V speed
Va maximal speed, absolute stability
Vc critical velocity
vR growth rate
w probability of fluctuation
x space coordinates
Xi , Xj extensive variables
Yi , Yj intensive variables
Z partition function
z space coordinate
b interface kinetics
b4 4-fold anisotropy of the kinetic coefficient
g surface tension or surface free energy
G ratio of S – L interface energy to specific melting entropy
G2 Green’s function integral
D supercooling (negative temperature field u at infinity)
d local concentration gap, surface tension anisotropy
e relative strength of anisotropy of capillary length
z (x, t) deviation of the interface
h coordinate
J angle of orientation relative to the crystallographic axes
Q function of (x, t)
k (s) mobility
l interface spacing, wavelength
L ( s, e ) solvability function
l2 wavelength of side branch
lf wavelength of the fastest mode
ls stability length
m chemical potential
n0 effective kinetic prefactor
x variable
84 2 Solidification
xi position vectors
Dxi small normal displacement
r Ivantsov radius
s scaling function
t variable
W atomic volume, atomic area
DLA diffusion limited aggregation

DS directional solidification
f faceted
l.h.s. left-hand side
nf non-faceted
PVA pivalic acid
r.h.s. right-hand side
SCN succinonitrate
WKB Wenzel – Kramers – Brillouin technique for singular pertubations
2.1 Introduction 85
2.1 Introduction mean? An intuitive argument goes as fol-

lows: the patterns formed at the solid – liq-
This chapter on solidification discusses uid interface and in both adjacent phases
the basic mechanisms of the liquid – solid during the solidification process result
phase transformation. In particular, we ad- from the competition of two “forces”, one
dress the phenomena of cellular and den- being stabilizing for homogeneous struc-
dritic patterns formed by the interface be- tures, the other being destabilizing. The
tween liquid and solid, as the interface, or stabilizing force here clearly must be asso-
solidification front, advances into the liq- ciated with the intrinsic atomistic length d,
uid during the solidification process. since we have argued that it is related to the
The atomistic processes of the liquid – length of correlation or coherence inside
solid transformation are still not well un- the material or at the interface. In contrast,
derstood, so we will use a phenomenologi- we must associate the external length l with
cal level of description. The processes a destabilizing force. Again this is a quite
on very large scales, such as casting or natural assumption, as the phase transfor-
welding, depend greatly on the experi- mation or destabilization of the nutrient
mental equipment and are discussed, for phase is induced by the experimental envi-
example, by Flemings (1991) and Mordike ronment.
(1991). It is now obvious that the result of such a
We will therefore restrict our attention to competition of “forces” should be ex-
phenomena occurring on some important pressed by the product d l of the two repre-
intermediate length scales. There is a natu- sentative quantities rather than by the sum,
ral scale L of the order of micrometers (or since the latter would change the relative
up to millimeters) that gives a measure for importance of the two lengths when their
the microcrystalline structures formed dur- values become very different.
ing the solidification process. In its sim- Admittedly, these arguments look a little
plest form this natural length is the geomet- overstressed considering the many parame-
ric mean L ~ 앀앙 d l of a microscopic intrinsic ters controlling the details of a solidifica-
length d defined by typical material proper- tion process. Note, however, that nothing
ties and an external length l defined by the has been said so far about the precise rela-
macroscopic arrangement of the experi- tion of d, l and L to any specific process,
mental equipment. The intrinsic correlation nor have we defined the proportionality
lengths in liquids and solids near the freez- factor. In principle, d and l could also enter
ing point are rather short, of the order of with different exponents but fortunately
atomic size, or several Ångströms, because things are not usually that complicated
solidification is a phase transition of first right from the start.
order. By contrast, the experimental equip- If we are still courageous enough to
ment gives external length scales in the make one more step on this slightly unsafe
range of centimeters to meters, such that ground, we may finally assume that the ex-
we consequently arrive at the mentioned ternal length scale l is related (destabilizing
scale of micrometers. force!) to the speed V of the solidification,
Assuming for the moment that only two which gives a length when combined with
lengths are important, why should we ex- a diffusion constant D for heat or matter as
pect L to be given by the geometric mean l ~ D/V. From this we immediately obtain a
rather than, for example, the arithmetic relation between the speed of phase change
86 2 Solidification
V and the length scale L of the resulting pler than if we attempted to carry along all
pattern: prefactors. The second and more important
reason is that the scaled form allows for a
V L2 ≈ constant (2-1)
much simpler comparison of experiments
Surprisingly enough, we have arrived at for different sets of parameters which usu-
about the most celebrated equation for pro- ally appear only in some combinations in
cesses of dendritic, directional and eutectic the equations, thereby leading to cancella-
solidification, without even defining any of tions and compensations.
these processes! Two remarks, therefore, Section 2.2 gives a quick summary of the
may be in order here. First, detailed studies ingredients for a theory, starting at nuclea-
of the different solidification schemes in tion, then deriving boundary conditions for
recent years have attempted to extract the a propagating interface between two
common and universal aspects of these phases, and finally discussing some gen-
processes. Such common features indicate eral aspects of the diffusion equation with a
a common basis of rather general nature, as propagating boundary. This is followed by
outlined above. Second, we have of course an introduction to basic experimental tech-
ignored most of the specific aspects of each niques in Sec. 2.3. In Sec. 2.4, the case of a
individual process. In directional solidifi- simple solid growing in a supercooled melt
cation, for example, a band of possible is discussed in some detail, leading to the
wavelengths for stationary patterns are present understanding of dendritic growth.
found and up to now it is not clear if and In Sec. 2.5 the technically important pro-
how a specific wavelength from that band cess of directional solidification is pre-
is finally selected. The assumption of just sented. The evolution of cellular patterns
two independent length scales in many above a critical growth rate can in principle
cases is also a rather gross simplification of be understood without any knowledge of
the actual situation. We will therefore leave dendritic growth. Actually, the parameter
this line of general arguments and look at range for simple sinusoidal cells is very
some concrete models that are believed to narrow so that we usually operate in the
capture at least some essentials for the fas- range of deep cells or even dendrites,
cinating patterns produced during solidifi- which suggests our sequence of presenta-
cation. tion. Finally, these concepts are extended
Some remarkable progress has been in Sec. 2.6 to alloys with a eutectic phase
achieved in the theoretical treatment of diagram and the resulting complex phe-
these phenomena during the recent years. nomena. As this field is currently in rapid
In the list of references, we have concen- theoretical development, our discussion
trated our attention on recent developments here necessarily is somewhat preliminary.
since there are some good reviews on older The chapter is closed by a summary with
work (for example, Langer, 1980 a; Kurz complementary remarks.
and Fisher, 1998).
An experimentalist may finally wonder
why we have expressed most of the equa- 2.2 Basic Concepts
tions in a non-dimensionally scaled form in First-Order Phase Transitions
rather than writing all material parameters
down explicitly at each step. One reason is The different possible phases of a mate-
that the equations then appear much sim- rial existing in thermodynamic equilibrium
2.2 Basic Concepts in First-Order Phase Transitions 87
must be discriminated by some quantity in larities or discontinuities in some of the

order to formulate a theory. Such quantities quantities, such as the order parameter or
are called “order parameters” and should the associated chemical potential, but they
correspond to extensive thermodynamic will still leave the temperature as a contin-
variables. The difference between a solid uous function of the position in space.
and a liquid is defined by the shear mod-
ulus, which changes discontinuously at the
2.2.1 Nucleation
phase transition. This definition describes
the difference in long-range orientational A particular consequence of the well-de-
correlations between two distant pairs of fined local equilibrium is the existence of
neighboring atoms. Normally, we use in- well-defined metastable states, correspond-
stead the more restrictive concept of trans- ing to a local, but not global, minimum in
lational order as expressed through two the free energy. But so far we have ne-
point correlation functions, or Bragg peaks, glected thermal fluctuations. The probabil-
in scattering experiments. ity of a fluctuation (or deviation from the
Although these different order-parame- average state) of a large closed system is
ter concepts pose a number of subtle ques-
w ~ exp (DS) (2-2)
tions, particularly in two dimensions where
fluctuations are very important, we will where DS is the change in entropy of the
simply assume in this chapter that there system due to the fluctuation (see, e.g.,
is some quantity which discriminates be- Landau and Lifshitz, 1970). Defining Qmin
tween a solid and a liquid in a unique way. as the minimal work necessary to create
Such an order parameter may be the den- this change in thermodynamic quantities,
sity, for example, which usually changes we obtain
during melting, or the composition in a
two-component system. We would like to DS = Qmin /T0 (2-3)
stress, however, that these are just auxiliary with T0 being the average temperature of
quantities that change as a consequence of the system. Note that this holds even for
the solid – liquid transition but which are large fluctuations, as long as the change of
not the fundamental order parameters in extensive quantities in the fluctuation re-
the sense of symmetry arguments. For a gion is small compared with the respective
more general discussion see the Chapter by quantities in the total system.
Binder (Binder, 2001) and literature on or- Considering this system as a metastable
der – disorder transitions (Brazovsky, 1975; liquid within which a fluctuation has
Nelson, 1983). formed a small solid region, and assuming,
First-order transitions are characterized furthermore, that the liquid is only slightly
by a discontinuous change of the order pa- metastable, we arrive at the well-known re-
rameter. All intrinsic length scales are sult (Landau and Lifshitz, 1970) in three
short, typically of the size of a few atomic dimensions for a one-component system:
diameters. We may therefore assume local
thermal equilibria with locally well-de- 4 p r3
Qmin = − [m L ( P ) − m S ( P )] + 4 p r 2 g
fined thermodynamic quantities like tem- 3W
perature, etc., and smooth variations in (2-4)
these quantities over large distances. Inter- which together with Eqs. (2-2) and (2-3)
faces in such systems will represent singu- gives the probability for the reversible for-
88 2 Solidification
mation of a spherical solid nucleus of ra- The extremal value of Eqs. (2-4) and (2-5)
dius r within a slightly supercooled liquid. with respect to the variation of r gives the
Here W is the atomic volume, P the pres- critical radius
sure, m L > m S are the chemical potentials of
liquid and solid in a homogeneous system, 2g
rc = W (3-dim) (2-6 a)
and g is the (here isotropic) surface ten- Dm
sion, or surface free energy. or
A few remarks should be made here.
g
When deriving Eq. (2-4), we use the con- rc = W (2-dim) (2-6 b)
Dm
cept of small deviations from equilibrium,
while Eqs. (2-2) and (2-3) are more gener- so that for r < rc , the nucleus tends to
ally valid (Landau and Lifshitz, 1970). In shrink, while for r > rc it tends to grow and
the estimation of the range of validity of at r = rc it is in an unstable equilibrium.
Eq. (2-4), it is apparent, however, that it The same thermal fluctuations causing
should be applicable to even very large such a nucleus to appear also produce devi-
supercoolings for most liquids, since the ations from the average spherical shape.
thermal transport is either independent of, This leads to power-law corrections in the
or faster than, the kinetics of nucleus for- prefactor of Eq. (2-5) or logarithmic cor-
mation (Ohno et al., 1990). The range of rections in the exponent (Voronkov, 1983;
validity of Eq. (2-4) is then typically lim- Langer, 1971).
ited by the approach to the “spinodal” re- So far these considerations have dealt
gion, where metastable states become un- with static aspects only. Since the fluctua-
stable, even when fluctuations are ignored tions vary locally with time, Eq. (2-5) may
(see the Chapter by Binder and Fratzl, 2001). be interpreted as the rate at which such
Assuming, therefore that we are still in fluctuation occurs, and, consequently, with
the range of well-defined metastable states, r = rc we obtain the rate for the formation
we may write Eqs. (2-2) and (2-3) as of a critical nucleus which, after appear-
w = n0 exp (– DG/T0 ) (2-5) ance, is assumed to grow until the new
phase fills the whole system. This is the
identifying the change in Gibbs energy DG classical nucleation theory. A very elegant
by Eq. (2-4), with an undetermined prefac- formulation was given by analytic continu-
tor n0 . Here we do not discriminate ation into the complex plane (Langer,
between surface tension and surface free 1971). Further additions include the defini-
energy, despite the fact that the first is a tion of the prefactor in Eq. (2-5) and a more
tensorial quantity, and the latter only scalar detailed analysis of the kinetics near r = rc
(although it may be anisotropic, which is (Zettlemoyer, 1969, 1976).
ignored here). A difference between sur- Considering the many uncertainties en-
face tension and surface free energy arises tering from additional sources such as the
when the system does not equilibrate range of atomic potentials and the change
between surface and bulk so that, for exam- of atomic interaction in the surface, we will
ple, the number of atoms in the surface ignore all these effects by absorbing them
layer is conserved. Throughout this chap- into the effective kinetic prefactor n0 in Eq.
ter, we will assume perfect local equilibra- (2-5), to be determined experimentally.
tion in this respect, and we may then ignore
the difference.
2.2.2 Interface Propagation face. We now assume local equilibrium to

exist on length scales a (i.e. i Æ i ± 1) such
An interface between two regions in
that the probability current is zero
space of different order parameters (solid
vs. liquid) will be treated in this chapter as ∂P
Vi P = Di (2-11)
a jump discontinuity and as an object of in- ∂x i
finitesimal thickness. In this section we de-
rive a local equation of motion for the inter- and with Eqs. (2-7) and (2-8), we immedi-
face, which will serve as a boundary condi- ately arrive at
tion in the remaining part of the chapter. ∂x i D ∂G ({x i })
For simplicity, consider a one-dimen- =− i (2-12)
∂t T ∂x i
sional interface in a two-dimensional sys-
tem. Ignoring the atomic structure, assume Taking the continuum limit a Æ 0, we ob-
that the interface is a smooth continuous tain the final form
line. Marking points at equal distances a on ∂x ( S ) dGˆ
this line, we may then define velocities of = − k (S) (2-13)
∂t dx ( S )
the points in the normal direction as
This is the time-dependent Ginzburg – Lan-
Dx i dau equation (Burkhardt et al., 1977). Here
Vi = (2-7)
Dt S denotes a position on the interface, ∂x (S)
where D x i is the small normal displace- is the normal displacement, d/d x (S) the
ment of point i. If we assume these points variational derivative, and k (S) the “mo-
to be kept at fixed positions x i for the mobility”, which may depend on position and
ment, a (restricted) Gibbs energy G ({x i }) orientation. Ĝ is the Gibbs energy per sur-
may be assigned to this restricted interface. face element.
The probability of this configuration is We will now make some explicit as-
sumptions about G in order to arrive at an
P ({x i }) = Z –1 exp (– G/T ) (2-8) explicit equation of motion. Let n̂S be the
with temperature T in units of kB K and Z normal direction on the interface, g (n̂S ) the
the partition function. Since the total prob- interface free energy, J the angle of orien-
ability is conserved, we obtain the continu- tation relative to the crystallographic axes,
ity equation R (S) the local radius of curvature at point S
on the interface, and AL, S the areas cov-
∂P + div J = 0 ered by liquid or solid. The Gibbs free en-
∂t
(i ) (2-9)
ergy of the total solid – liquid system with
interface is
with J = {Ji } as the probability current in
i-space Gtot = ∫ dS g + ∫ d AL gL ( x )
S Liquid
∂P
Ji = Vi P − Di (2-10) + ∫ d AS gS ( x ) (2-14)
∂x i Solid
and the divergence taken in the same ab- from which the variational derivative in
stract space. The first term in Eq. (2-10) is Eq. (2-13) is formally obtained by
a drift, the second term the constitutive re-
lation with transport coefficients Di , and ⎡ dGˆ ⎤
dGtot = ∫ dS ⎢ ⎥ dx ( S ) (2-15)
the derivative is taken normal to the inter- S ⎣ dx ⎦
90 2 Solidification
where d x (S) is a small arbitrary displace- infinite solid or liquid without influences
ment of the interface in the normal direc- from curvature terms. For a single-compo-
tion. Explicitly this is written as nent system, m = g W, where m is the chemi-
cal potential of the respective phase. The
dGtot = ∫ {g d (dS ) + dS dg } generalization to three dimensions adds an-
S
other curvature term into Eq. (2-21), which
− ∫ dS d x ( S ) [ g L ( S ) − gS ( S )] (2-16) then corresponds to two curvatures and an-
S
gles in the two principal directions (for iso-
with Gibbs energy densities g given here tropic g , the curvature 1/R is simply
per atomic “volume” (or area) W. As usual, changed to 2/R).
we have to extract terms d x out of the Two useful observations can be made at
terms d (dS) and d g . Assuming g to depend this stage. There is a solution with V^ = 0 if
only on local orientation, the term {…} in Eq. (2-21) is zero. For fi-
nite radius of curvature R, this corresponds
dg precisely to the critical nucleus, Eq. (2-6 b),
dg (J ) = dJ (2-17)
dJ but now with anisotropic g (J ). This equa-
and with tion therefore determines the shape of the
(2-18) critical nucleus in agreement with the Wulff
d d dJ d d 1 construction (Wulff, 1901). Second, for
dJ = dx ; = ; J =−
dS dS dS dJ dS R very large mobility k (J ) the deviation
from equilibrium {…} may be very small
we obtain in order to produce a desired normal veloc-
dg 1 d ity V^ . We will use this simplification of
dg (J ) = − dx (2-19) equilibrium at the interface
dJ R dS
1
The other term simply gives Dg ≅ (g + g ′′ ) ( 2-dim) (2-22)
R
1 wherever possible, but we will comment on
d ( dS ) = dS dx (2-20)
R the changes due to finite k (J ). The double
Incorporating Eqs. (2-19) and (2-20) into prime means derivatives as in Eq. (2-21).
Eq. (2-16) and integrating Eq. (2-19) by In many cases this seems to be justified
parts, we obtain through comparison with by experimental conditions. On the other
Eq. (2-15) hand, very little is known quantitatively
(2-21) about k (J ). A last point to mention here is
∂x ⎧ 1⎛ d g ⎞⎫
2
our assumption that the interfaces are not
≡ V⊥ = k (J ) ⎨Dg ( S ) − ⎜ g + ⎟⎬
∂t ⎩ R⎝ dJ 2 ⎠ ⎭ faceted at equilibrium. If they are faceted,
as crystals typically are at low tempera-
as an explicit local equation for the ad- tures in equilibrium with their vapor, the
vancement of an interface in two dimen- situation is considerably more complicated
sions with normal velocity V^ , anisotropic and not completely understood (Kashuba
kinetic coefficient k (J ), surface Gibbs en- and Pokrovsky, 1990).
ergy g (J ), and jump Dg = gL – gS of Gibbs It is now generally believed that faceted
energy density at position S along the inter- surfaces undergo a kinetic roughening tran-
face (Burkhardt et al., 1977). Here, the sition even at small driving forces, so that a
Gibbs energy density g corresponds to an rough surface is present under growth con-
ditions. The concept outlined above then where cp is the specific heat of the liquid at
generally applies in an effective fashion. constant pressure. If, instead of tempera-
Actually, facets may persist over experi- ture or heat, a second chemical component
mentally relevant length scales, which are is diffusing, a similar transformation to
not covered by this analysis in the kinetic the same dimensionless equations can be
region. A summary of theoretical develop- made. This is described in Section 2.5.2.
ments can be found in the article by Krug We would like to stress the importance of
and Spohn (1992). such a scaled representation as it allows us
To complete this section on the basic to compare at a glance experimental situa-
theoretical ingredients, we will now dis- tions with different sets of parameters.
cuss the influence of diffusive transport of In this dimensionless form, the equation
heat or matter on the propagation of a of motion is
solid – liquid interface (see also Langer,
∂
1980 a). u = DT ∇ 2 u (2-25)
∂t
2.2.3 Growth of Simple Crystal Forms and the boundary conditions are
a) Flat interface u = u• < 0 for z Æ • (2-26 a)

The simplest model for a solidifying u=0 at interface (2-26 b)
system consists of two half-spaces filled So far we have not specified how the inter-
with the liquid and solid of a one-compo- face motion is coupled to the equation of
nent material of invariant density and sep- motion. Obviously, the latent heat Lm gen-
arated by a flat interface. The interface is erated during this freezing process at a rate
approximately in equilibrium (Eq. (2-22)) proportional to V has to be carried away
at melting temperature Tm but advancing at through the diffusion field. This requires a
a speed V in the positive z-direction into the continuity equation at the interface
liquid. During this freezing process, latent
heat Lm has to be transported into the su- V = – DT n̂ · —u (2-27)
percooled liquid, the solid remaining at Tm .
The equation of motion is then the ther- and since in the model here only the z-axis
mal diffusion equation is important
∂ ∂
T ( x, t ) = DT ∇ 2 T ( x, t ) (2-23) V = − DT u ( z, t ) at interface (2-28)
∂t ∂z
with thermal diffusion coefficient DT and The growth rate is therefore proportional to
appropriate boundary conditions. At infin- the gradient of the diffusion field at the
ity in the supercooled liquid the tempera- interface in the liquid. As the interface is
ture T• < Tm is prescribed, and Tm is the moving, we conveniently make a coordi-
temperature at the interface. It is now very nate transformation from {z, t} for {z¢, t ¢}
convenient to replace the temperature field into a frame of reference moving along
T by a dimensionless u through the trans- with the interface at z¢ = 0:
formation t
z ′ = z − ∫ V (t ) dt
T ( x, t ) − Tm 0 (2-29)
u ( x, t ) = (2-24)
( Lm / c p ) t′ = t
92 2 Solidification
Performing that transformation, we obtain rate V (or l) is not specified, but seems to
the following for the diffusion equation be arbitrary. This degeneracy is practically
(after having again dropped the primes for eliminated by other effects such as inter-
convenience): face kinetics (Collins and Levine, 1985) or
the density difference between solid and
1 ∂u 2 ∂u
= ∇2 u + (2-30) liquid (Caroli et al., 1984), so that in prac-
DT ∂t l ∂z tice a well-defined velocity will be selected.
which defines a diffusion length
b) Growth of a sphere
l ∫ 2 DT /V (2-31)
The influence of the surface tension g
The interface is now always at z = 0, which when it is incorporated through Eq. (2-21)
makes the boundary condition Eq. (2-26 b) is most easily understood by looking at a
and the continuity Eq. (2-28) definite. spherical crystal. This will not lead to sta-
We can obtain a stationary solution by tionary growth. In order to make the analy-
combining Eq. (2-30) with Eqs. (2-26) and sis simple, we will invoke the so-called
(2-28) when ∂u /∂t = 0: “quasistationary approximation”, by set-
ting the left-hand side in Eq. (2-25) equal
u = exp ⎛ − ⎞ − 1; z Û 0
2z
(2-32) to zero. The physical meaning is that the
⎝ l ⎠
diffusion field adjusts itself quickly to a
This equation is consistent with the as- change in the boundary structure which is,
sumption that u = 0 (T = Tm ) everywhere in however, still evolving in time because the
the solid. continuity equation is velocity-dependent.
This equation describes the diffusion field Of course, this approximation reproduces
ahead of the interface. It varies exponen- the stationary solutions precisely (if they
tially from its value at the interface toward exist), and in addition, it exactly identifies
the value far inside the liquid, so that the the instability of a stationary solution as
diffusion field has a typical “thickness” of l. long as it is not of the Hopf type (i.e., the
Note that we have not used Eq. (2-26 a) critical eigenvalue is not complex). It is
as a boundary condition at infinity, but find generally assumed that the approximation
from Eq. (2-32) that u = – 1 implies the so- is good as long as the diffusion length is
called “unit supercooling”, which corre- large compared with other lengths of the
sponds to evolving pattern.
In a spherical coordinate system, the
T (z = •) = Tm – (Lm /cp ) (2-33) equation of motion then becomes
This basically says that the difference in 1 ∂u ⎧ ∂2 2 ∂⎫
melting enthalpy Lm between liquid and ≈0=⎨ 2 + ⎬u (2-34)
DT ∂t ⎩ ∂r r ∂r ⎭
solid must be compensated by a tempera-
ture difference in order to globally con- as a simple Laplace equation.
serve the energy of the system during this The interface is at radius R0 (t) and is ad-
stationary process. In other words, if Eq. vancing with time. The continuity equation
(2-33) is not fulfilled by Eq. (2-26 a), the requires
process cannot run with a stationary profile
of the thermal field, Eq. (2-32). On the dR0 ∂u
vR ≡ = − DT (2-35)
other hand, a particular value of the growth dt ∂r r = R0 ( t )
Concerning the boundary condition, we The growth rate therefore is

(2-43)
note that for a simple substance the differ-
dR D ⎛ 2 d0 ⎞ ⎛ b DT ⎞
ence in slopes of the chemical potentials vR ≡ 0 = T ⎜D − ⎟ ⎜1 + ⎟
for solid and liquid at the melting point dt R0 ⎝ R0 ⎠ ⎝ R0 ⎠
(Landau and Lifshitz, 1970) is Lm /Tm , giv-
ing and it is found that for large radii R0 (t) the
growth rate is proportional to the super-
D m = – Lm (T – Tm )/Tm (2-36) cooling D :
With Eqs. (2-21) and (2-24), the boundary DT ⎛ D ⎞
VR ≈ D ⎜ 1 − b T + …⎟ (2-44)
condition in the case of isotropic g be- R0 ⎝ R0 ⎠
comes
It can also be seen that the kinetic coeffi-
u (interface) = – d0 K – b V (2-37) cient b becomes less important in the limit
with curvature of large radii and correspondingly small
growth rates. This in many cases justifies
⎧2 / R, 3-dim our previous equilibrium approximation,
K=⎨ (2-38)
⎩1/ R, 2-dim Eq. (2-22).
We have not discussed here the solidifi-
capillary length cation of a binary mixture. As the details
d0 = g Tm cp /L2m (2-39) of the phase diagram will become impor-
tant later, we will postpone this topic to the
and interface kinetics section on directional solidification. We
b = k –1 Tm cp /L2m (2-40) would like to mention, however, that the
representation of the diffusion field in di-
The generalization to non-isotropic g fol- mensionless form, found in Eq. (2-24), has
lows from Eq. (2-21) and comments. the virtue that several results can be carried
Note that in Eq. (2-37) both curvature over directly from the thermal case to com-
and non-equilibrium occur at the boundary positional diffusion, although there are
as opposed to the simple case in Eq. (2- some subtleties in the boundary conditions
26 b), shown above. At infinity we finally which make the latter more difficult to an-
impose alyze.
u• = u (•, t) = – D (2-41)
as the now arbitrary dimensionless super- 2.2.4 Mullins – Sekerka Instability
cooling. This equation, together with Eq. We now combine the considerations of
(2-24), is the definition of D . Secs. 2.2.3 a) and b) to study the question
The solution to Eq. (2-34) with Eqs. (2- of whether an originally flat interface will
35), (2-37), and (2-41) in the liquid is then remain flat during the growth process. The
simply (2-42) results indicate that a flat interface moving
R ⎛ 2d ⎞ ⎛ b DT ⎞ into a supercooled melt will become rip-
u (r ) = − D + 0 ⎜ D − 0 ⎟ ⎜1 + ⎟ pled (Mullins and Sekerka, 1963, 1964).
r ⎝ R0 ⎠ ⎝ R0 ⎠
The basic equations are almost exactly
where we have used K = 2/r as the curva- the same as in the previous section a) Eqs.
ture of the spherical surface in three dimen- (2-30), (2-27), (2-26 a), but instead of the
sions. boundary condition Eq. (2-26 b), we now
94 2 Solidification
have to consider deviations from a flat where we have linearized the exp (…) with
interface as we move parallel to the inter- respect to z . Instead of the velocity in Eq.
face in the x-direction. This is provided by (2-27), we must now use Vz = 2 D/l + ∂z /∂t.
Eq. (2-37) as the boundary condition, With this equation and the same lineariza-
where we set b = 0 for simplicity. tion as before
Denoting the deviation of the interface (2-49)
ˆ 2 ˆ
W k z k = − D ( 2 / l ) z k + Dq uˆk + D′ q ′ uˆk′
from z = 0 as z (x, t), we designate the cur-
vature for small amplitudes simply as is obtained for the conservation law, Eq.
K = – ∂ 2 z /∂x 2, where, in agreement with (2-27) (and for small values of z , practi-
Eq. (2-38), the sign of K allows for the cur- cally Eq. (2-28)). Eliminating ûk here using
vature to be positive for a solid protrusion Eq. (2-48) and eliminating q using Eq. (2-47)
into liquid. We follow here the notation of in the limit k l Ⰷ1 or k ≈ q ≈ q¢, we obtain
Langer (1980 a).
In a quasistationary approximation, for W k = V ( q − 2 / l ) − ( Dq + D′ q ′ ) d0 k 2
the sake of simplicity we set the l.h.s. in ≈ k V [1 − 12 (1 + D′ / D) l d0 k 2 ] (2-50 )
Eq. (2-30) equal to zero, which of course
allows for the stationary solution, Eq. (2- This formula describes the basic mecha-
32). It is not difficult to treat the fully time- nism of diffusion-controlled pattern forma-
dependent problem here, but the modification in crystal growth. Althoug in general,
tions do not change the results substan- diffusion tends to favor homogeneous
tially. We now perturb that solution by structures, in the present case it works in
making a small sinusoidal perturbation of the opposite direction! This is easy to
the flat interface: understand; the foremost points of a sinu-
soidal deformation of the interface can dis-
z (x, t) = zˆk exp (i k x + W k t) (2-45) sipate the latent heat of freezing by a larger
Similarly, a perturbation of the diffusion space angle into the liquid than the points
field is made in the liquid and in the solid inside the bays. The latter points will there-
fore be slowed down. The tip points will be
u ( x , z, t ) = exp ⎛ − ⎞ − 1
2z accelerated in comparison with the average
(2-46 )
⎝ l ⎠ rate of advancement.
+ uˆk exp (i k x − q z + W k t ); The formula therefore consists of a de-
u ′ ( x , z, t ) = uˆk′ exp (i k x + q ′ z + W k t ) stabilizing part leading to positive Wk and
a stabilizing part controlled by the capillar-
where the unprimed form is for the liquid, ity d0 . The stabilization is most efficient
and the primed values are for the solid. at large k values or short wavelengths; at
Inserting Eq. (2-46) into Eq. (2-30) with longer wavelengths, the destabilization due
∂u /∂t = 0, we obtain to diffusion into the supercooled liquid
dominates.
– 2 q /l + q 2 – k 2 = 0
The dividing line is marked by the com-
2 q¢/l ¢ + q¢ 2 – k 2 = 0 (2-47) bination of d0 l, which is most conveniently
expressed as
Replacing z in Eq. (2-46) by z (x, t) from Eq.
(2-45), we can insert this into the boundary l S = 2 p /kS = 2 p d0 l (2-51)
condition Eq. (2-37) for b = 0 to obtain
assuming D = D¢. This will be called the
uˆk′ = − 2 zˆk / l + uˆk = − d0 k 2 zˆk (2-48) “stability length”, which gives a measure
2.3 Basic Experimental Techniques 95
of the typical lengths for diffusion-gener- interested in nucleation experiments is

ated ripples on the interface, since the referred to reviews by Perepezko et al.
fastest-growing mode has a wavelength (1987, 1996).
-
l f = ÷3 l S . This is the first explicit example There are two essentially different situa-
of the fundamental scaling relation Eq. tions of solidification, or, more generally,
(2-1) mentioned in the introduction to this of phase transformation (Fig. 2-1):
chapter.
i) free (undercooled or equiaxed) growth,
ii) constrained (columnar on directional)
growth.
2.3 Basic Experimental Techniques
In the first case (Fig. 2-1 a), the melt
Before discussing the various growth undercools, before transforming into a
models, a short overview of selected ex- crystal, until nucleation sets in. The cru-
periments that have contributed to our cible that contains the melt must be less ef-
present understanding will be given. fectively catalytic to crystallization than
An exhaustive treatment of the subject is the heterogeneous particles in the melt.
not intended here; only some of the charac- The crystals then grow with an interface
teristic techniques will be described. No temperature above the temperature of the
specific reference will be given to the surrounding undercooled liquid. Heat is
many solidification processes available. carried into the liquid. The temperature
The main emphasis will be on growth of gradient at the solid – liquid interface in the
crystals, because direct evidence on nucle- liquid is therefore negative and approxi-
ation mechanisms is generally not avail- mately zero in the solid owing to the in-
able. There is an exception; the original itially small crystal size relative to the ther-
technique developed by Schumacher et al. mal boundary layer.
(1998). These authors analysed quenched- In the second case of constrained growth
in nuclei (inoculants) in glassy Al alloys in (Fig. 2-1 b), the temperature gradient is
the transmission electron microscope. In positive in both liquid and solid, and the
this state, nucleation kinetics are very slow first solid is either formed in contact with
and in-situ observation of the phases and the chill mold or is already present, as in
and their structure, composition and crystal surface treatment by lasers, for example,
orientation can be measured. The reader where growth occurs epitaxially after re-
Figure 2-1. Growth in (a) under-

cooled melt (free growth) and (b)
superheated melt (directional
growth). The arrows on the out-
side of the mold represent the heat
flux, and the arrows at the
solid/liquid interface, the growth
direction.
96 2 Solidification
melting. In case i), nucleation is essential ferent, and some of its important elements
in controlling the microstructure (the grain will be described below.
size), while in case ii), nucleation is of mi-
nor importance.
Typical technical processes with respect 2.3.1 Free Growth
to these two classes of solidification are as Slow growth rates
follows:
In free growth, the undercooling DT or
i) casting into low conductivity molds
the temperature of the melt T0 = (Tm – DT )
(ceramics), producing small temperature
is imposed on the system, and the growth
gradients, or adding inoculating agents or
rate V and the microstructural scale are the
stirring the melt, thereby producing many
dependent variables. For low undercool-
heterogeneous or homogeneous nuclei;
ings, and therefore slow growth rates, all
ii) processes with high heat flux imposed
three variables can be measured precisely
through strong cooling of the solid such as
if the substance is transparent. The crystal-
continuous casting, welding, laser treat-
lization of non-transparent metals, how-
ment, or through heating of the melt and
ever, is the major issue in solidification
cooling of the solid, such as in single crys-
studies relevant to technical applications.
tal growth or Bridgman type directional so-
Instead of investigating the crystalliza-
lidification experiments (see also Flem-
tion of metallic systems directly, suitable
mings, 1991).
model substances have to be found. These
In free growth, the undercooling DT of are generally organic “plastic crystals”
the melt is given and controls the growth which, like metals, have simple crystal
rate V and the scale (spacing l or tip radius structures and low melting entropies (Jack-
R) of the forming microstructure. In direc- son, 1958). One of the substances which
tional growth, the rate of advance of the behaves very similarly to metals and also
isotherms is imposed by heat flux leading has well characterized properties is succi-
to an imposed growth rate (constrained nonitrile (SCN) (Huang and Glicksman,
growth). This, in turn, controls the inter- 1981). Numerous results have been ob-
face temperature (undercooling) and the tained with this material by Glicksman and
microstructural scales. These three vari- co-workers. Their careful experimental ap-
ables (temperature, growth rate, and micro- proach not only produced the most precise
structural size) have to be measured experi- measurements known at that time but has
mentally for a quantitative comparison of also stimulated new ideas about possible
theory and experiment. A number of mate- mechanisms of structure formation through
rial parameters of the alloy system have to the discrepancy found between the obser-
be known, such as the solid – liquid interfa- vations and the predictions of previous the-
cial energywith its anisotropy, the diffusion ories.
coefficient, the stable and metastable phase Fig. 2-2 shows the experimental setup
diagram, etc. developed by Glicksman et al. (1976) to
The experimental techniques are divided study free dendritic solidification. After
into two classes; those which produce rela- zone melting, the material is introduced
tively small growth rates and those which into the growth chamber (C), which is then
aim for high solidification rates. The corre- slowly undercooled with the heaters (A)
sponding experimental setup is quite dif- and (B) in order to avoid premature crystal-
cooling, are measured with the aid of pho-

tographs. In order to choose a proper pro-
jection plane of observation, the entire
equipment can be translated, rotated and
tilted at (D). A series of similar experi-
ments have been performed at low tem-
peratures with rare gases (Bilgram et al.,
1989).
In this kind of experiment, it is important
to avoid thermal or solutal convection, as
this transport mechanism will change the
results and make them difficult to compare
with diffusional theory. Glicksman et al.
(1988) have shown the effect of thermally
driven convection on the growth morphol-
ogy of pure SCN dendrites. Fig. 2-3 clearly
shows a strong deviation of the growth rate
at undercoolings lower than 1 K, where
convection is believed to accelerate den-
drite growth. Here V0 is the dendrite tip
rate of pure diffusion-controlled growth.
Recent experiments on free dendritic
growth of succinonitrile in space under mi-
Figure 2-2. Equipment for free growth of organic crogravity conditions in the undercooling
dendrites (Glicksman et al., 1976). A and B, control
range of 0.064 to 1.844 K confirmed this
heaters; C, growth chamber; D, tilting and rotating
device; F, tank cover; G and H, zone-refining tubes. effect clearly (Koss et al., 1999).
Another type of slow dendrite growth in
undercooled media has been analyzed by
lization. By careful adjustment of heater Trivedi and Laorchan (1988). These au-
(A), growth starts there, and a crystal thors made interesting in situ observations
grows through the orifice in chamber (C). during the crystallization of glasses. Even
From then on, the dendrite can grow freely if the driving forces in these systems are
into the undercooled liquid and its shape, very large, growth is heavily restricted due
size and growth rate, depending on super- to slow diffusion and attachment kinetics.
Figure 2-3. Ratio of measured

growth rate, V, to predicted
rate from diffusion theory, VD ,
as a function of undercooling
(Glicksman et al., 1988).
98 2 Solidification
Fast growth rates 2.3.2 Directional growth

In order to explore solidification behav- Slow growth rates
ior under very large driving forces, Herlach
Understanding of solidification improved
et al. (1993) and Herlach (1994, 1997)
in the 1950s when the need for better semi-
systematically measured dendrite growth
conductor materials stimulated research
rates in a large number of highly under-
using directional growth techniques such
cooled alloys. Nucleation is avoided by
as zone melting and Bridgman growth.
levitation melting in an ultrahigh vacuum
Later, directional casting became an impor-
environment. Undercooling of more than
tant topic of research for the production
300 K leading to growth rates of up to
of single crystal turbine blades. Fig. 2-4
70 m/s has been reached. Such large under-
shows the essentials of two techniques:
coolings have been obtained by several au-
Bridgman type and chill plate directional
thors before, but growth rates have not
solidification. The first of these processes
been measured. Interesting results in Ni –
(Fig. 2-4 a) has the advantage of being able
Sn alloys have also been obtained by high-
to produce a constant growth rate and a
speed cinematography of highly under-
constant temperature gradient over a con-
cooled samples. These results were pro-
siderable length and to allow for a certain
duced by Wu et al. (1987) by encapsulating
uncoupling of these two most important
the melt in glass. They tried to measure
variables. It was this latter advantage
tip radii from the photograph. In this case,
which, early on in solidification research,
however, only the thermal images of the
provided much insight into the mecha-
dendrites could be seen, their tips being
nisms of growth. With one growth rate, but
controlled by solute diffusion. The radii
varying temperature gradient (or vice
from the thermal images are therefore
versa), plane-front, cellular or dendritic
believed to be much larger than the real
morphologies could be grown, and their
radii.
evolution studied. Important concepts
(constitutional undercooling, cell growth,
Figure 2-4. Basic methods

of directional solidification;
(a) Bridgman type furnace
(a) (b) and (b) directional casting.
etc.) have been developed with the aid of

observations made with the Bridgman type
of equipment. The experiments by Hunt et
al. (e.g., Burden and Hunt, 1974) deserve
special attention, as the quality of the
measurements was of a very high standard.
Furthermore, most of the work on direc-
tional eutectics and their growth mecha-
nisms was performed with the aid of this
technique. In eutectic solidification, the
microstructure after transformation has the
same size (interphase spacing l) as that
found at the growing interface and thus al-
lows for direct conclusions to be drawn
concerning the growth process. This im-
portant variable can therefore be easily de-
termined in non-transparent metals. This,
however, is not possible with the corre-
sponding quantity of the dendrite, the tip
radius R, as discussed earlier.
One disadvantage of the Bridgman ex- Figure 2-5. Schematic diagram of growth cell (a, b)
periment is the need for a small diameter for observation of directional solidification of trans-
due to heat flux constraints. This is avoided parent substances under a microscope (Esaka and
Kurz, 1985). 1, solid; 2, liquid; 3, thermocouple;
in the process shown in Fig. 2-4 b (direc-
4, cell; 5, low melting glass; 6, araldite seal; 7 and 8,
tional casting), but in this case a separation heaters; 9, cooler; 10, drive mechanism; 11, micro-
of the variables V and G is not possible. scope. In (c) the temperature distribution in the
Therefore, this method is less interesting growth cell is shown.
for scientific purposes. However it is used
extensively for directional casting of single
crystal turbine blades (Versnyder and A thin thermocouple incorporated into
Shank, 1970). the alloy allows for measurement of inter-
For in-situ observation of microstructure face temperature (Fig. 2-6) and tempera-
information, we can also use plastic crys- ture gradient in liquid and solid (Esaka and
tals in an experimental arrangement resem- Kurz, 1985; Somboonsuk and Trivedi,
bling a two-dimensional Bridgman appara- 1985; Trivedi and Somboonsuk, 1985; Tri-
tus, as shown in Fig. 2-5. Two glass slides vedi and Kurz, 1986).
containing the organic analogue are moved
over a heating and cooling device produc-
Fast growth rates
ing a constant temperature gradient GT .
This can be controlled to certain limits by There have been attempts to increase the
the temperature difference and distance growth rate in Bridgman type experiments.
between heater and cooler (Somboonsuk The first to reach rates of the order of sev-
and Trivedi, 1985; Trivedi and Somboon- eral mm/s were Livingston et al. (1970).
suk, 1985; Akamatsu et al., 1995; Ginibre The best way to reach much higher rates is
et al., 1997; Akamatsu and Faivre, 1998). through laser resolidification using a stable
100 2 Solidification
solidification microstructure and its growth

rate as well as the bath temperature and the
growth rate in rapid undercooled growth.
In both cases, the measurement of the inter-
face temperature is not possible for the
time being and must be evaluated through
theory alone.
2.4 Free Dendritic Growth

The most popular example of dendriti-
cally (= tree-like) growing crystals is given
by snowflakes. The six primary arms on
each snowflake look rather similar but there
is an enormous variety of forms between
different snowflakes (Nakaya, 1954). From
our present knowledge, the similarity of
arms on the same snowflake is an indica-
Figure 2-6. Array of dendrites approaching a fine tion of similar growth conditions over the
thermocouple for measurement of the tip temperature
and temperature gradient. Diameter of the bead ap-
arms of a snowflake, the variation of struc-
proximately 50 µm (Esaka and Kurz, 1985). ture along each arm being an indication of
a time variation of external growth condi-
tions.
high-powered laser. In this case, a small In order to formulate a theory, some spe-
melt pool with very steep temperature gra- cific assumptions about these environmen-
dients is created (for a remelted layer of tal conditions must be made. Unfortu-
100 mm thickness, the temperature gradi- nately, the growth of a snowflake is an
ent GT is of the order of 5000 K /mm). The extremely complicated process, involving
microstructure is then constrained to fol- strongly anisotropic surface tension and
low closely the heat flow, which is perpen- kinetics and the transport of heat, water va-
dicular to the isotherms (Fig. 2-7). There- por, and even impurities. Therefore, we
fore, knowing the angle q between the di- will initially drastically simplify the model
rection of growth and the direction and rate assumptions. In the same spirit, a number
of movement of the laser beam alloys for of precise experiments have been per-
the growth rate to be obtained locally, even formed to identify quantitatively the most
in the electron microscope (Zimmermann important ingredients for this mechanism.
et al., 1989). Note that the growth rate in- In its simplest form, dendritic growth re-
creases from zero at the interface with the quires only a supercooled one-component
substrate up to a maximum at the surface. liquid with a solid nucleus inside, so that
The only unknown which would be ex- the heat released at the solid – liquid inter-
tremely difficult to measure is the interface face during growth is transported away into
temperature. the liquid. This is precisely the condition
Consequently, in the case of rapid direc- given in Sec. 2.2.3 concerning a sphere
tional growth we can determine the post- growing into the supercooled, and there-
2.4 Free Dendritic Growth 101
Figure 2-7. Schematic diagrams of a la-

ser trace (a, b) and of the local interface
velocity (c). Vb is the laser scanning ve-
locity and Vs is the velocity of the solid/
liquid interface which increases from zero
at the bottom of the trace to a maximum
at the surface. W and D are the width and
depth of the trace, respectively, and q is
the angle between the growth direction
and the scanning direction.
fore metastable, liquid. A relatively straight- Saito, 1990; Xiao et al., 1988). For these
forward stability analysis (Mullins and Se- processes, some specific assumptions are
kerka, 1963, 1964; Langer, 1980 a) shows made about the incorporation of atoms into
that the solid tends to deviate from spheri- the advancing interface which are not eas-
cal form as soon as its radius R has become ily carried over into the notion of surface
larger than seven times the critical radius tension, etc. Another line of closely related
Rc = 2 d0 /D . A time-dependent analysis has problems concerns the viscous flow of two
also been carried out (Yokoyama and Ku- immiscible liquids (Saffmann and Taylor,
roda, 1988). During the further evolution, 1958; Brener et al., 1988; Dombre and Ha-
these deformations evolve into essentially kim, 1987; Kessler and Levine, 1986 c). We
independent arms, the primary dendrites. will briefly refer to this in Sec. 2.5.
The growth of such dendrites is a very
widespread phenomenon, as will become
2.4.1 The Needle Crystal Solution
clearer in Sec. 2.5 on directional solidifica-
tion. There are also close relations to pro- In this subsection, we will look at an iso-
cesses called diffusion-limited aggregation lated, needle-shaped crystal growing under
(DLA) (for a review see Meakin, 1988; for stationary conditions into a supercooled
relation to crystal growth, see Uwaha and melt. A stationary condition, of course, only
holds in a frame of reference moving at ve- lattice, but they are not directly related
locity V in the positive z-direction. For a one- (Burkhardt et al., 1977). For the moment,
component crystal, the latent heat of freez- we will assume b = 0, i.e., the interface ki-
ing must be diffused away into the liquid. netics should be infinitely fast. Even then,
The process is therefore governed by the the curvature K in Eq. (2-54) is a compli-
heat diffusion Eq. (2-25) cated function of the interface profile. The
simplest approximation for the moment,
2 ∂u
0 = ∇2 u + (2-52) therefore, is to ignore both terms on the
l ∂z r.h.s. of the boundary condition, which cor-
where we have used a dimensionless form, responds to setting the surface tension to
Eq. (2-30), for a stationary pattern together zero. The boundary condition is then sim-
with the definition of the diffusion length, ply a constant (= 0 in our notation).
Eq. (2-31). If the diffusion constants in the A second-order partial differential equa-
crystal and the liquid are different, we tion such as the diffusion equation can be
would have to use two diffusion lengths. decomposed in the typical orthogonal coor-
The continuity Eq. (2-27) at the interface is dinate systems, and we therefore obtain a
for normal velocity closed form solution for the problem with a
boundary of parabolic shape: a rotational
V^ = – DT n̂ · —u L + DT n̂ · —u S (2-53) paraboloid in three dimensions and a
simple parabola in two dimensions. This is
where the subscripts L and S denote gra-
the famous “Ivantsov” solution (Ivantsov,
dients taken on the liquid side and on the
1947).
solid side of the interface. This is known as
The straightforward way to look at this
the “two-sided” model (Langer and Turski,
problem is as a coordinate transformation
1977) and, because of identical diffusion
from the cartesian {x, z} frame, where z is
constants, as the “symmetrical” model. The
the growth direction, to the parabolic coor-
generalization to different diffusion con-
dinates {x , h }:
stants is simple.
The boundary condition at the interface is x = (r – z)/r ; h = (r + z)/r
------
u I = – d0 K – b V (2-54) where r is the radial distance ÷x 2 + z 2 from
the origin, and r is a constant. After trans-
and at infinity forming the differential operators in Eqs.
u• = – D (2-55) (2-52) and (2-53) to {x , h } (Langer and
Müller-Krumbhaar, 1977, 1978), it can be
as introduced before in Eqs. (2-37) and (2- seen immediately that h (x ) = h S = 1 for the
41), respectively. Note that a constant like interface is a solution to the problem con-
D may be added on the right-hand side of firming the parabolic shape of the inter-
both equations without changing the results face, with r being the radius of curvature at
apart from this additive constant in the u- the tip.
field, as frequently appears in the litera- This Ivantsov radius, r , is now related to
ture. The definition of D follows easily the supercooling D and the diffusion length
from Eqs. (2-24) and (2-41). l by the relation
In general, the capillary length d0 and
the kinetic coefficient b are anisotropic be- ⎧ P eP E1 ( P ) 3-dim (2-56 a)
D=⎨ P
cause of the anisotropy of the crystalline ⎩ p P e erfc ( P ) 2-dim (2-56 b)
which for small D Ⰶ1 gives face tension effects. Before we consider

surface tension explicitly we now give an
⎧ P ( − ln P − 0.5772…) 3-dim integral formulation for the problem using
D≈⎨
⎩ pP 2-dim Green’s functions, which has proved to be
more convenient for analytical and numer-
and for D Æ1 asymptotically ical calculations (Nash and Glicksman,
1974).
⎧1 − 1/ P 3-dim
D≈⎨ The value of a temperature field u (x, t)
⎩1 − 1/ 2 P 2-dim in space and time is obtained by multiply-
where the dimensionless Péclet number P ing the Green’s function of the diffusion
is defined as the relation equation with the distribution for sources
and sinks for heat and integrating the prod-
P = r /l = r V/ 2 D (2-57) uct over all space and time. In our case, this
between tip radius and growth rate. E1 (P) explicitly gives
(2-59)
is the exponential integral t
∞
u ( x , z, t ) = ∫ dt ′ ∫ d x ′ G [ x − x ′, z − z ( x ′, t ′ )
e−t −∞
E1 ( P ) = ∫ t
dt (2-58) ⎧ ∂z ⎫
P + V (t − t ′ ), t − t ′ ] ⎨V + ⎬
⎩ ∂t′ ⎭
and erfc is the complement of the error
function. (Langer, 1987 b), with the Green’s function
Eq. (2-56) may be interpreted as an ex-
1 ⎛ | x |2 + z 2 ⎞
pression of supercooling in terms of the G ( x , z, t ) = exp ⎜− ⎟
Péclet number. This explanation is impor- ( 4 p D t )3 / 2 ⎝ 4 Dt ⎠
tant because the following considerations (2-60)
of surface tension give only very small cor- for diffusion in an infinite three-dimen-
rections to the shape of the needle crystal. sional medium (symmetrical case), while z
Therefore, Eq. (2-56) will be also valid is the z coordinate of the interface. The
with non-zero surface tension at typical ex- source term {…} in this equation is obvi-
perimental undercoolings of D Ó 10–1. An ously the interface in the frame of refer-
important consequence is that the scaling ence moving at velocity V in the z-direc-
results derived below then become inde- tion. This equation already contains the
pendent of the dimensionality (2 or 3), if conservation law or the continuity equation
the supercooling is expressed through the at the interface. Furthermore, it is valid
Péclet number. everywhere in space and, in particular, at
The basic result of Eqs. (2-57) and (2- the interface z = z (x, t), where the l.h.s. of
48) is that the growth rate of this parabolic Eq. (2-59) is then set equal to Eq. (2-54).
needle is inversely proportional to the tip Here u is assumed to vanish at infinity, so
radius, but no specific velocity is selected. D must be added on the l.h.s.
For experimental comparison, we make a In the two-dimensional case and for
fit of the tip shape to a parabola. The tip stationary conditions, this can be rewritten
radius of that parabola is then compared as
with the Ivantsov radius r. The actual tip
radius will be different from the Ivantsov d0
D− K {z} = P G2 {P, x , z ( x )} (2-61)
radius of the fit parabola, because of sur- r
(Langer, 1987 b), with For convenience we combine some param-

∞ +∞ eters into a dimensionless quantity s :
1 dy
G2 {P, x , z ( x )} =
2p
∫ y
∫ dx ′ (2-62)
2 Dd0 d0
0 −∞ s= = V (2-68)
⎛ P ⎞ r2 V 2 D P2
× exp ⎜ − {( x − x ′ )2 + [z ( x ) − z ( x ′ ) + y ]2 }⎟
⎝ 2y ⎠ so that the l.h.s. of Eq. (2-67) becomes
In addition, it was shown (Pelce and Po- d
meau, 1986) that with the parabolic Ivant- − K = − s AK (2-69)
rP
sov solution
the curvature as usual being
z ( x ) → z IV = − 12 x 2 (2-63)
∂ 2 z / ∂x 2
we can obtain K {z} = − (2-70)
[1 + ( ∂z / ∂x )2 ]3 / 2
P G2 {P, x , z IV ( x )} = D
It should be clear at this point that the pa-
which is independent of x, with D coming rameter s in Eq. (2-68) plays an important
from Eq. (2-56 b). Note that from Eqs. role, because it multiplies the highest de-
(2-59) to (2-61) we have replaced x by x r rivative in Eqs. (2-61) and (2-67). More
etc., which mirrors the scaling form of specifically, we can expect that the result-
Eq. (2-57). ing structure z (x) of the interface depends
We are now ready to consider a non-zero on the material properties and experimental
capillary length d0 Eq. (2-39), which we conditions only through this parameter s
generalize to be anisotropic: (within the model assumptions), which be-
comes a function of P and e
d0 Æ d = d0 (1 – e cos 4 J ) (2-64)
s = s (P, e ) (2-71)
where J is the angle between the interface-
normal and the z-axis (2-dim, 4-fold an- The importance of the paramter s was
isotropy), and e > 0 is the relative strength recognized in an earlier stability analysis
of that anisotropy. This form arises from (Langer and Müller-Krumbhaar, 1978,
the assumption 1980) of the isotropic case. It turned out
later, however, that the anisotropy is essen-
g = g 0 (1 + d cos 4 J ) (2-65)
tial in determining the precise value of s .
for anisotropy surface tension, which gives This is crucial as e Æ 0 implies s Æ 0, i.e.,
through Eq. (2-22) e = 15 d. Note that it is no stationary needle solutions exist without
the stiffness g ≤ which dominates the be- anisotropy.
havior, not g directly. Along the Ivantsov We will now briefly describe the analy-
parabola, the angle J is related to x: sis of Eq. (2-67). The details of this singu-
(2-66) lar perturbation theory are somewhat in-
8e x 2 volved, and we therefore have to omit them
d = d0 A ( x ); A ( x ) = 1 − e +
(1 + x 2 )2 here. The basic method was formulated by
Kruskal and Segur (1985), and the first
and the deviation from z IV (x) can be ex-
scaling relations for dendritic growth were
pressed as
(2-67) obtained for the boundary-layer model
d (Ben Jacob et al., 1983, 1984). A good in-
− K (z ) = G2 {P, x , z} − G2 {P, x , z IV}
rP troduction to the mathematical aspects can
be found in the lectures by Langer (1987 b), such that

on which the following presentation is +∞
Z˜ ( x )
based. The most mathematically sound L (s , e ) ≡ ∫ dx ′ =0 (2-78)
(nonlinear) solution seems to have been −∞ (1 + x 2 )3 / 4
given by Ben Amar and Pomeau (1986).
For convenience, we have sketched a In fact, this is already a sufficient condition
slightly earlier linearized version here, for the solvability of the inhomogeneous
while the nonlinear treatment leads to basi- equation, but it is not very simple.
cally the same result. A solution for Z̃ (x) can be found by a
Looking for a solution to Eq. (2-67) in WKB technique, for which we refer to the
linear approximation, we start by expand- literature (Kessler et al., 1987, 1988; Lan-
ing to first order in ger, 1987 b; Caroli et al., 1986 a, b). The
result for the solvability condition, Eqs.
z1 (x) = z (x) – z IV (x) (2-72) (2-71) and (2-78), is that the parameter s
should depend on anisotropy e as
In the limit of the small Péclet number, and
with the substitution s ≈ s 0 e 7/4 (2-79)
z1 (x) = (1 + x 2 )3/4 Z (x) (2-73) in the limit P Æ 0, e Æ 0, with some con-

stant prefactor s 0 of order unity.
gives Eq. (2-79) is the solution for the needle-
(B̂ + Â) · Z (x) = s /(1 + x 2 )3/4 (2-74) shaphed crystal with capillary anisotropy
e > 0, together with Eq. (2-68). Note again
where B̂ is a self-adjoint differential opera- that e is the anisotropy of capillary length,
tor which differs by a factor from surface-ten-
sion anisotropy Eq. (2-65). Formally, there
d2 (1 + x 2 )1/ 2
Bˆ = s + + 0 (s ) (2-75) is not just one solution but infinitely many,
dx 2 A( x) corresponding to slow, fat needles which
and are dynamically unstable. Only the fastest
of these needle solutions appears to be
(1 + x 2 )3 / 4 stable against tip-splitting fluctuations and
Aˆ ⋅ Z ( x ) = (2-76)
2 p A( x) may thus represent an ‘observable’ needle
+∞
( x + x ) (1 + x 2 )3 / 4 crystal, as expressed by Eq. (2-79). For
′ ′
×ᏼ ∫ dx ′ ( x − x ′ ) [1 + 1 ( x + x ′ )2 ] Z ( x ′ ) practical comparison, experimental data is
−∞ 4 best compared with numerical solutions of
with ᏼ denoting the principal value. Eq. (2-61), because the applicability of Eq.
The integral kernel in Eq. (2-76) is anti- (2-79) seems to be restricted to very small
symmetric apart from a prefactor A (x)–1. values of e (Meiron, 1986; Ben Amar and
An analytic solution to Eq. (2-74) has not Moussallam, 1987; Misbah, 1987). This will
yet been found. A necessary condition to be discussed further in the next section.
be fulfilled by the present inhomogeneity is
that it should be orthogonal to the null- Needle-crystal solution
eigenvectors Z̃ (x) of the adjoint homoge- in three-dimensional dendritic growth
neous problem:
The theory of dendritic growth becomes
(B̂ + Â+ ) · Z (x) = 0 (2-77) extremely difficult, however, for three-di-
mensional (3D) anisotropic crystals. A cubic symmetry corresponds to m = 4 and

simple extrapolation of the 2D case, where A 4 = 1/96 is only numerically small. There-
the surface energy is averaged in the azi- fore, the shape correction, Eq. (2-80), in
muthal direction (axisymmetric approach, units of the tip radius of curvature, depends
Ben Amar, 1998; Barbieri and Langer, mostly on the crystalline symmetry and is
1989), is very important in order to have almost independent of the material and
some qualitative predictions. But any phys- growth parameters.
ical anisotropy will give rise to a non- Stability theory for the 3D dendritic
axisymmetric shape of the crystal. A nu- growth against tip-splitting modes has been
merical approach to the non-axisymmetric developed by Brener and Mel’nikov (1995).
problem was presented by Kessler and Le- An important aspect of Eq. (2-80) is that
vine (1988) who pointed out the following the shift vector r m cos mf grows at a faster
aspect of the problem. In either the 2D or rate than the underlying Ivantsov solution.
the axisymmetric case, selection of the This means that only the tip region, where
growth velocity follows from the solvabil- the anisotropy correction is still small, can
ity condition of smoothness of the dendrite be described by the usual approximation
tip. In the 3D non-axisymmetric case a (Kessler and Levine, 1988; Ben Amar and
solvability condition must be satisfied for Brener, 1993), a linearization around the
each of the azimuthal harmonics. Kessler Ivantsov paraboloid. This is the crucial dif-
and Levine made several approximations ference between the 3D non-axisymmetric
and performed only a two-mode calcula- case and the 2D case. In the latter, small
tion, but the crucial point of their analysis anisotropy implies that the shape of the se-
is that they found enough degrees of free- lected needle crystal is close to the Ivant-
dom to satisfy all solvability conditions. sov parabola everywhere; in the former,
More recently, an analytic theory of strong deviations from the Ivantsov para-
three-dimensional dendritic growth has boloid appear for any anisotropy.
been developed by Ben Amar and Brener Several important questions arise. How
(1993). In the framework of asymptotics is the tail of the dendrite to be described? Is
beyond all orders, they derived the inner it possible to match the non-axisymmetric
equation in the complex plane for the non- shape (Eq. (2-80)) in the tip region to the
axisymmetric shape correction to the Ivant- asymptotic shape in the tail region? What
sov paraboloid. The solvability condition is the final needle-crystal solution? The an-
for this equation provides selection of both swers to these questions have been given
the stability parameter s µ e 7/4 and the by Brener (1993).
interface shape. The selected shape can be The basic idea is that the non-axisym-
written as metric shape correction, generated in the
tip region, should be used as an ‘initial’
r2
z (r, f ) = − + ∑ Am r m cos m f (2-80) condition for a time-dependent two-dimen-
2 sional problem describing the motion of
where all lengths are reduced by the tip ra- the cross-section of the interface in the tail
dius of curvature r. Solvability theory (Ben region. In this reduced description, the role
Amar and Brener, 1993) predicts that the of time is played by the coordinate z for
numbers Am are independent of the aniso- steady-state growth in the z direction. The
tropy strength a , in the limit of small a . deviation from the isotropic Ivantsov solu-
For example, the first non-trivial term for tion remains small only during the initial
period of the evolution. This initial devia- portant intermediate asymptotic exists if
tion can be handled by a linear theory start- the size of the 2D pattern is still much
ing from a mode expansion which takes the smaller than a diffusion length (D t)1/2.
same form as Eq. (2-80) with, in principle, This 2D Laplacian problem (with a fixed
arbitrary coefficients. These amplitudes flux from the outside) was solved, both nu-
then have to be chosen according to the merically and analytically, by Almgren et
predictions of selection theory (Ben Amar al. (1993) who were interested in aniso-
and Brener, 1993) in order to provide the tropic Hele – Shaw flow. They found that
matching to the tip region. As ‘time’ goes after some transition time the system
on, the deviation increases owing to the shows an asymptotic behavior which is in-
Mullins – Sekerka instability and a nonlin- dependent of the initial conditions and in-
ear theory must take over. We can guess volves the formation of four well-devel-
what the long-time behavior of the system oped arms. The length of these arms in-
will be. Indeed, this two-dimensional prob- creases in time as t 3/5 and their width in-
lem is precisely the same as that which creases as t 2/5. The basic idea that explains
leads to two-dimensional dendritic struc- these scaling relations is that the stability
tures. Four well-developed arms (for cubic parameter s 2 = 2 D d0 /(r 22 v2 ) is supposed to
symmetry) grow with a constant speed in be equal to s 2* (e ) even though both v2 and
the directions favored by the surface en- r 2 depend on time. Moreover, the growing
ergy anisotropy. Each arm has a parabolic self-similar shape of the arms was deter-
shape, its growth velocity, v2 = 2 D P22 (D ) mined. In terms of the dendritic problem it
¥ s 2* (e )/d0 , and the radius of curvature of reads (Brener, 1993)
its tip, r 2 = d0 /(P2 s 2* ), are given by 2D se- 1/ 5 2/3
⎛ s*⎞ ⎛ x ⎞
lection theory, where the anisotropic sur- y ( x , z ) = ( 5 | z | / 3) 2/5
⎜ *⎟ ⎜x ⎟
face energy again plays a crucial role. The ⎝s2 ⎠ ⎝ tip ⎠
Péclet number P2 = r 2 v2 /2 D is related to 1
ds
the undercooling D by the 2D Ivantsov for- × ∫ (2-83 a)
mula, which for small D gives x / x tip s 2/3
1 − s4
P2 = (D ) = D2/p (2-81) where the tip position x tip of the arm is
The selected stability parameter s 2 = given by
d0 /(P2 r 2 ) depends on the strength of the x tip (z) = (5 |z |/ 3)3/5 (s 2*/s * )1/5 (2-83 b)
anisotropy e , and s 2* (e ) µ e 7/4 for small e
as was explained above. Replacing t by The ratio s 2* (e )/s * (e ) is independent of e
|z |/v and reducing all lengths by r, we can in the limit of small e. This means that
present the shape of one-quarter of the the shape, Eq. (2-83), in the tail region is
interface (except very close to the dendritic almost independent of the material and
backbone) in the form growth parameters, as well as shape, Eq.
(2-80), in the tip region (if all lengths are
v2 y 2 r
x ( y, z ) = | z | − (2-82) reduced by r ).
v 2 r2 Recent experiments on 3D dendrites
This asymptotic describes the strongly an- (Bisang and Bilgram, 1995) and numerical
isotropy interface shape far behind the tip simulations (Karma and Rappel, 1998) are
and does not match, for small D , the shape in very good agreement with these theoret-
described by Eq. (2-80). In this case an im- ical predictions.
2.4.2 Side-Branching Dendrites The dimensionless supercooling is given

as
This section provides a summary of the
present understanding of dendritic growth. ⎧( Tm − T∞ ) /( Lm c −p 1 )
⎪thermal
It is centered around numerical simulations ⎪
D=⎨ (2-85)
of isolated side-branching dendrites in a
⎪(m eq − m ∞ ) /( DC ∂m / ∂ C )
one-component system where heat diffu- ⎪⎩chemical
sion is the relevant dynamical process.
Alternatively, it also describes dendritic The capillary length is then
growth from a two-component system at
essentially constant temperature. In the lat- ⎧Tm c p L−m2
⎪thermal
ter case, we should also discuss the phase ⎪ (2-86)
d = [g (J ) + g ′′ (J )] ⎨
diagram; this will be covered later in the −2 −1
⎪DC (∂m / ∂ C )
section on directional solidification. For ⎪⎩chemical
many typical cases of growth from a dilute
solution however, the information con- The quantity ∂m /∂C is not easily measured,
tained in this section should be sufficient. but for small DC Ⰶ1 of a dilute solution, it
We start with the definition of the model can be related to the slope of the liquidus
resulting from the set of Eqs. (2-52) to (2- line dT /dC at T0 = Tm by
55). The dynamics come from the conser-
∂m Lm dT
vation law, Eq. (2-53), at the interface. As ≈ (2-87)
in the previous section, we use here the ∂C Tm DC dC
form of dimensionless units introduced in
(Mullins and Sekerka, 1963, 1964; Langer,
Sec. 2.2.3 for the case of heat diffusion.
1980 a). Note that the chemical capillary
The case of chemical diffusion (tempera-
length can be several orders of magnitude
ture then being assumed constant = T0 ) can
larger than the thermal length.
be treated by the same equations. The nor-
The boundary condition, Eq. (2-54), then
malization is described in Sec. 2.5.2. For
simply changes to
convenience, we will simply summarize
here the basic formulas for both cases. uI = D – d K – b V (2-88)
In contrast to Eqs. (2-37) and (2-54), we
where now the anisotropic capillary length
will now normalize the following equation
d is used. The kinetic coefficient may also
to obtain u = 0 at infinity, which results
depend on concentration (Caroli et al.,
from adding the constant D to the field-var-
1988), which we ignore here. Far away
iable u in all equations. Then, the diffusion
from the interface in the liquid, the bound-
field becomes
ary condition becomes
⎧( T − T∞ ) /( Lm c p−1 )
⎪⎪thermal diffusion u• = 0 (2-89)
u=⎨ (2-84) For the chemical case, we may practically
⎪(m − m ∞ ) /( DC ∂m / ∂ C ) ignore diffusion in the solid.
⎪⎩chemical diffusion
The diffusion Eq. (2-52) and the conser-
where m is the chemical potential differ- vation law, Eq. (2-53), remain unchanged,
ence between solute and solvent and DC and the diffusion length is defined as be-
(0 ≤ DC ≤ 1) is the miscibility gap at the op- fore as l = 2 DT /V, with V being the average
erating temperature T0 . velocity of the growing dendrite.
There are, of course, important differ- uid interface. In Fig. 2-8, we show a typical
ences between two and three dimensions, dendrite with side branches resulting from
as a three-dimensional needle crystal is not the time-dependent calculations (Saito et
necessarily rotationally symmetric around al., 1987, 1988) (compare with the experi-
its axis. Snow crystals, for example, show mental result by Glicksman et al. (1976) in
large anisotropies in directions orthogonal Fig. 2-9). The profile is symmetric around
to the growth direction of the primary den- the axis by definition of the calculation. An
dritic needle (Yokoyama and Kuroda, approximately parabolic tip has been
1988). In the immediate neighborhood of formed from which side branches begin to
the tip, however, the deviation from this ra- grow further down the shaft (only the early
tional symmetry is often small. Therefore, stage of side-branch formation was consid-
we may work with this two-dimensional ered). They have a typical distance which,
model by using an “effective” capillary however, is not strictly regular.
length. The scaling relations given below As a first result, the scaling relation, Eq.
are expected to be insensitive to this apart (2-68), was checked using the Péclet num-
from a constant prefactor of order unity in ber from Eq. (2-56 b). Experimentally, this
the s (e )-relation (Kessler and Levine, requires the anisotropic capillary length
1986 b, d; Langer, 1987 a). and the supercooling to be varied indepen-
The numerical simulations were per- dently. In Fig. 2-10, the scaled numerical
formed for a two-dimensional crystal – liq- results are shown as symbols for two dif-
Figure 2-8. Free dendrite in stationary growth com-

puted in quasistationary approximation for the two-
dimensional case. Capillary anisotropy was e = 0.1
(Saito et al., 1988). The parameter-dependence of the Figure 2-9. Dendrite tip in pure succinonitrile (SCN)
growth rate, tip radius and sidebranch spacing is con- at small undercoolings and inscribed parabola for
sistent with analytical scaling results from solvability measuring the tip radius (Huang and Glicksman,
theory of the needle crystal. 1981).
growth rate V through the Péclet number,

Eq. (2-56), depending on supercooling. Eq.
(2-68) should then give a constant, al-
though arbitrary, value of s. This scaling
result was confirmed experimentally in
the 1970s, before the full theory existed
(Langer and Müller-Krumbhaar, 1978,
1980). At that time, it was assumed (“mar-
ginal stability” hypothesis) that a universal
value of s ≈ 0.03 was determined by a dy-
Figure 2-10. Scaling parameter s (e ) for free den- namic mechanism independent of aniso-
dritic growth depending on capillary anisotropy e tropy. The results for the needle crystal, to-
and for two-dimensional supercoolings D. Average gether with these numerical simulations,
capillary length is d0 , diffusion constant D, and now show that s depends on anisotropy e
Péclet number P. Comparison of numerical results
(circles and squares, Fig. 2-8, one-sided model) with
as shown in Fig. 2-10. Experimental tests
solvability results: (a) one-sided model (Misbah, on the e -dependence (Sec. 2.4.3) are still
1987), (b) two-sided model (Ben Amar and Moussal- rather sparse and do not quite fit that pic-
lam, 1987). The agreement is excellent, the indepen- ture, for reasons not well understood.
dence upon supercooling is seen to work at least up So far we have only looked at the rela-
to D = 0.5.
tion between growth rate, anisotropy and
supercooling. We will now consider the
size of the dendrite, which is approxi-
ferent supercoolings and compared with mately parabolic, and which can probably
the results (full lines) for the stationary be characterized by the radius of curvature
needle crystal. The upper line corresponds at its tip.
to the model with diffusion in the liquid This is a subtle point, as the tip radius
only (Misbah, 1987), as used in the numer- cannot easily be measured directly. As an
ical simulation here. The lower line corre- alternative, we can try to fit a parabola to
sponds to the symmetrical model with the observed dendrite in the tip region. The
equal diffusion in liquid and solid (Ben tip radius of this fitted parabola should be
Amar and Moussallam, 1987). Ap- interpreted as the Ivantsov radius r , which
parently, the two results look the same, turns out to be slightly larger than the true
apart from a factor of approximately two tip radius R of the dendrite. The deviation
in s . Note that in unscaled form (i.e., mul- of R for r does not depend on supercooling
tiplying by P 2 ) the data for the two super- D but on anisotropy e. This is shown in Fig.
coolings would differ by about two orders 2-11, where a comparison is made between
of magnitude! the dynamic numerical simulations (Saito
From an experimental point of view, it is et al., 1988) and the needle crystal solution
better to use Fig. 2-10 rather than Eq. (2- in the limit of small Péclet number (Ben
79) for comparison, as the range of validity Amar and Moussallam, 1987). It can be
of Eq. (2-79) seems to be restricted to seen that there is excellent agreement and
rather small values of e. For unknown ma- that the actual tip radius R becomes smaller
terial parameters such as diffusion con- than the Ivantsov radius r at increasing e.
stant, capillary length and anisotropy, we We now can relate the growth rate V and
can still check the scaling relation of the the tip radius R or the Ivantsov radius r in
Figure 2-11. Tip radius of free dendrite over Ivant- Figure 2-12. Numerical scaling result confirming
sov radius plotted versus anisotropy as a function of V R 2 = const. for free dendritic growth independent
dimensionless supercooling. The numerical results of supercooling D, depending on anisotropy e only
(see also Fig. 2-8, 2-10) are consistent with the pre- (Saito et al., 1988).
dictions from the needle solution (Ben Amar and
Moussallam, 1987).
order to check the scaling form Eq. (2-68) only. This is precisely the relation, Eq. (2-
involving the radius rather than the Péclet 1), derived from qualitative considerations
number. The Ivantsov parabola and its ra- in the introduction to this chapter. This re-
dius r basically originate from a global lation has been confirmed by the analysis
conservation law for the quantity (heat) re- of many experiments (Huang and Glicks-
leased at the interface, while the tip radius man, 1981).
R is a local geometric quantity. In practical The final point to be discussed here con-
experiments, by fitting a parabola to the cerns the side branches and their origin,
tip, we can interpolate between these two spacing and amplitudes. This issue is theo-
numbers, the result depending on how far retically not completely resolved, because
down the shaft the fitting parabola is used. none of the available analytical approxima-
Using the actual radius R rather than the tions can correctly handle the long-wave-
Ivantsov radius r , perfect scaling can be length limit of side-branch perturbations.
seen in Fig. 2-12 with respect to supercool- Moreover, the subject is somewhat tech-
ing D , even up to the very large value of nically involved. Therefore, we will only
D = 0.5. Since for smaller supercoolings, summarize the main arguments below and
D Ù 0.1, the difference between r and R be- refer to the above-mentioned numerical
comes negligible, as shown in Fig. 2-11, simulations (Saito et al., 1988) for compar-
and we may safely use Eq. (2-68) as ison with experiments.
An important quantity which charac-
2 Dd0
V R2 ≈ V r 2 = = constant (2-90) terizes the stability of flat moving interface
s (e ) ripples is the so-called stability length
independent of supercooling D , to interpret
l S = 2 p l d0 (2-91)
experiments and to make predictions. The
term “constant” here means that the prod- where d0 is the capillary length, and l the
uct V R 2 depends on material parameters diffusion length. Perturbations of wave-
lengths l longer than l S will grow, while mental anisotropy of the material succino-
shorter wavelengths will decay with time. nitrile (Huang and Glicksman, 1981),
This quantity characterizes the competition e ≈ 0.1, gave the same s -value as the theo-
between the destabilizing diffusion field retical concept, and since e cannot be var-
through l against the stabilizing surface ied easily, there was no discrepancy.
tension through d0 . A derivation of this To summarize these results, it appears
Mullins – Sekerka instability was given in that the scaling relation, Eq. (2-92), shown
Sec. 2.2.4. in Fig. 2-13 from the numerical solution of
It is natural to assume that this length the model in two dimensions, is in agree-
scale is related to the formation of side ment with the experimental results.
branches. A direct estimate of the typical We will now give a somewhat qualitative
wavelength l 2 of the side branches is explanation of the mechanism of side-
branch formation as far as this can be de-
l 2 ⲏ l S = 2 p l d0 = 2 p r s (2-92) duced from the theoretical approaches. A
The remarkable result of the numerical linear stability analysis (Langer and Mül-
simulation is shown in Fig. 2-13. Appar- ler-Krumbhaar, 1978, 1980; Kessler and
ently, the ratio l 2 /l S is a constant of ap- Levine, 1986 a; Barber et al., 1987; Bar-
proximately 2.5, which is independent of bieri et al., 1987; Bensimon et al., 1987;
supersaturation and anisotropy. This seems Caroli et al., 1987; Kessler et al., 1987;
to be in quite good agreement with experi- Pelce and Calvin, 1987) indicates that the
ments (Glicksman et al., 1976; Dougherty relevant modes for side-branch formation
et al., 1987; Honjo et al., 1985; Huang and in the frame of reference moving with the
Glicksman, 1981). tip consist of an almost periodic sinusoidal
The experimental comparison was made, wave travelling from the tip down the shaft
in fact, with an older theoretical concept such that they are essentially stationary in
(Langer and Müller-Krumbhaar, 1978), the laboratory frame of reference (Langer
which did not correctly consider aniso- and Müller-Krumbhaar, 1982; Deissler,
tropy. By accident, however, the experi- 1987). The amplitude of these waves is not
constant in space, but first grows exponen-
tially in the tip region (Barbieri et al.,
1987; Caroli et al., 1987). The exponential
increase of that envelope in the tip region
depends on the “wavelength” of the oscil-
latory part (Bouissou et al., 1990).
In contrast to the earlier analysis by
Langer and Müller-Krumbhaar, all these
modes are probably stable, so that without
a triggering source of noise, they would
decay, and a smooth needle crystal would
result. Some driving force in the form
of noise due to thermal or hydrodynamic
Figure 2-13. Scaled sidebranch spacing l = l 2 , nor-
fluctuations is needed to generate side
malized with Ivantsov radius R0 and s (e )1/2, plotted
versus capillary anisotropy for two supercoolings. branches, but apparently this is usually
No dependence on e or D is found, as expected (Saito present. Estimates of the strength of these
et al., 1988). fluctuations (Langer and Müller-Krumb-
haar, 1982; Barbieri et al., 1987; Langer, tual non-axisymmetric shape of the needle
1987 a) are still somewhat speculative. crystal, defined by Eqs. (2-83 a, b), was
Given such a small noise at the tip, the given by Brener and Temkin (1995). They
exponentially increasing envelope over the found that the root-mean-squared ampli-
side branches into the direction of the tail tude for the side branches generated by
then amplifies that noise so that the side thermal fluctuations is
branches become visible. This happens
over a range of about two to ten side- ⎧ 2 x 3/ 2 ⎫
〈x12 ( Z , Y )〉1/ 2 ~ Q exp ⎨ 1 / 2 ⎬ (2-93)
branch spacings. The actual selected wave- ⎩ 3 3s | z | ⎭
length of the side branches in that tip re-
gion (assuming a white noise, triggering all where the function x (z) is given by the
modes equivalently), according to these underlying “needle” solution, Eq. (2-83 b),
–
considerations, is defined by the mode with and the fluctuation strength Q is given by
–2
the largest amplitude at a distance of about Langer (1987 a), Q = 2 kB T 2 cp D/(L2m v r 4 ).
one “wavelength” away from the tip. This The root-mean-squared amplitude for
is the product of the average amplitude due the side branches increases with the dis-
to noise at the tip and the amplification fac- tance from the tip, | z |. This amplitude
tor from the envelope. grows exponentially as a function of
Langer (1987 a) described the side- (|z | 2/5/s 1/2 ). The important result is that
branching deformation as a small (linear) the amplitude of the side branches for the
perturbation moving on a cylindrically anisotropic needle grows faster than for
symmetric needle crystal (Ivantsov parabo- the axisymmetric paraboloid shape. In
loid). The noise-induced wave packets the latter case x (z) = 2 |z |1/2 and the ampli-
generated in the tip region grow in amplitude grows exponentially as a function
tude, spread and stretch as they move down of (|z |1/4/s 1/2 ). This effect resolves the
the sides of the dendrite producing a train puzzle that experimentally observed side
of side branches. In the linear approxima- branches have much larger amplitudes than
tion, the amplitude grows exponentially can be explained by thermal noise in the
and the exponent is proportional to |z | 1/4. framework of the axisymmetric approach.
These results are in approximate, qualita- Agreement with experiment now is indeed
tive agreement with available experimental very good (Bisang and Bilgram, 1995).
observations (Huang and Glicksman, 1981; Far down from the tip the side-branching
Dougherty et al., 1987; Bisang and Bil- deformations grow out of the linear regime
gram, 1995), but experimentally observed and eventually start to behave like den-
side branches have much larger amplitudes drites themselves. It is clear that the
than explicable by thermal noise in the branches start to grow as free steady-state
framework of the axisymmetric approach. dendrites only at distances from the tip
This means that either the thermal fluctua- which are of the order of the diffusion
tion strength is not quite adequate to pro- length which, in turn, is much larger than
duce visible side-branching deformations, the tip radius r in the limit of small P. This
or agreement with experiment would re- means that there is a large range of z,
quire at least one more order of magnitude 1 Ⰶ|z | Ⰶ 1/P, where the side branches al-
in the exponential amplification factor. ready grow in the strongly nonlinear re-
The description of the side-branching gime, but they do not yet behave as free
problem, which takes into account the ac- dendrites. We can think of some fractal ob-
ject where the length and thickness of the

dendrites and the distance between them
increase according to some power laws
with the distance |z | from the tip. The den-
drites in this object interact, owing to the
competition in the common diffusion field.
Some of them die and some continue to
grow in the direction prescribed by the an-
isotropy. This competition leads to the
coarsening of the structure in such a way
that the distance between the surviving
dendrites l (z) is adjusted to be of the same
order of magnitude as the length of the den-
drites, l (z). The scaling arguments give
l (z) ~ l (z) ~ |z | (Brener and Temkin, 1995).
The morphology measurements on SCN
crystals yield a good quantitative agree-
ment with this linear law (Li and Becker- Figure 2-14. Dimensionless growth rate Ṽ = V d0 /2 D
mann, 1998). versus dimensionless undercooling D. The scaling
We now summarize the presently estab- quantity for the full curve (Langer and Müller-
lished findings for free dendritic growth Krumbhaar, 1977, 1978) was taken as s = 0.025 (co-
incidentally in agreement with the anisotropy of suc-
with respect to their experimental signifi- cinonitrile). For references to the experimental points
cance. A discussion of additional effects see Langer (1980 a). Excellent agreement between
such as faceting will be given in Sec. 2.5.7 theory (solid line) and experiment is found.
in the context of directional solidification.
For a given material with fixed D, d0 and
e, the growth rate V depends upon super- can be described by remarkably simple
cooling D through Eq. (2-68), and with “scaling laws”. The whole dendritic struc-
Péclet number P taken from Eq. (2-56). ture with side-branches looks like a fractal
The dimensionless parameter s is given in object on a scale smaller than the diffusion
Fig. 2-10. This is demonstrated for various length and as a compact object on a scale
materials in Fig. 2-14. The size or tip ra- larger than the diffusion length (Brener et
dius of the dendrite is related to its velocity al., 1996).
by Eq. (2-90) and can be taken from Fig. The new steady-state growth structures
2-12. The typical wavelength of the side that have been identified are the so-called
branches is then given by Eq. (2-92) and “doublons” in two dimensions (Ihle and
can be taken from Fig. 2-13. This provides Müller-Krumbhaar, 1994; Ben Amar and
all the basic information that should be Brener, 1995), first observed in the form of
valid in the tip region. a doublet cellular structure in directional
Beyond the understanding of steady- solidification (Jamgotchian et al., 1993),
state growth of the tip, the major new con- and the “triplon” in three dimensions (Abel
cept that has emerged over the last few et al., 1997). Both structures have been
years is that complex pattern formation shown to exist without crystalline aniso-
processes occurring on the much larger tropy, unlike conventional dendrites. The
scale of an entire dendrite grain structure doublon has the form of a dendrite split
into two parts about its central axis with a his co-workers made an essential contribu-
narrow liquid groove between the two tion to our understanding of dendritic
parts, and triplons in three dimensions are growth in pure undercooled melts (Glicks-
split into three parts. For a finite aniso- man et al., 1976; Huang and Glicksman,
tropy, however, these structures only exist 1981). This research was initially con-
above a critical undercooling (or supersat- cerned mainly with highly purified succi-
uration for the isothermal solidification of nonitrile (SCN). It was extended to cyclo-
an alloy), such that standard dendrites hexanol (Singh and Glicksman, 1989), wa-
growing along ·100Ò directions are indeed ter (Fujioka, 1978; Tirmizi and Gill, 1989),
the selected structures in weakly aniso- rare gases (Bilgram et al., 1989), and to
tropy materials at low undercoolings, in other pure substances with a crystal aniso-
agreement with most experimental obser- tropy different from SCN such as pivalic
vations in organic and metallic systems. acid (PVA) (Glicksman and Singh, 1989).
From a broad perspective, the existence of Work on free growth of alloys includes
doublons and triplons is of fundamental NH4Cl – H2O (Kahlweit, 1970; Chan et al.,
importance because it has provided a basis 1978), NH4Br – H2O (Dougherty and Gol-
on which to classify the wide range of pos- lub, 1988), SCN with acetone and argon
sible growth morphologies that can form as (Glicksman et al., 1988; Chopra et al.,
a function of undercooling and anisotropy 1988), PVA – ethanol (Dougherty, 1990),
(Brener et al., 1996). and others.
We have so far ignored the influence of The specific merit of the work of Glicks-
the kinetic coefficient b in Eq. (2-88). This man et al. was that the systems for which
omission is not likely to be very important they characterized all the properties, in-
for low growth rates, but for fast growth cluding surface energy, diffusion coeffi-
rates, as in directional solidification, b cient, phase diagram etc., have been exam-
should be taken into account. We will re- ined. This led to clear evidence in the mid-
turn to this point in Sec. 2.5. 1970s that the theory of that time (using
extremum arguments for the operating
point of the tip) was not able to describe
2.4.3 Experimental Results
the results quantitatively.
on Free Dendritic Growth
At the same time, Müller-Krumbhaar
The answer to the question of whether and Langer worked on precisely the same
dendritic growth is diffusion-controlled problem and proposed a theory based on
or controlled by anisotropic attachment ki- the stability of the growing dendrite tip,
netics, was sought by Papapetrou (1935), called the marginal stability criterion (Lan-
who was probably the first to make system- ger and Müller-Krumbhaar, 1977, 1978).
atic in situ experiments on free dendritic Most of the existing experimental data
growth. He examined dendritic crystals could be fitted using this criterion. Despite
of transparent salts (KCl, NaCl, NH4Br, the fact that this theory incorrectly ignored
Pb(NO3)2 , and others) under a microscope the important role of anisotropy (as we
in aqueous solutions and proposed that the know now), it inspired a number of new
tip region should be close to a paraboloid experiments and also attracted the interest
of rotational symmetry. of other physicists.
Many years later, the extensive and As has been said before, today’s theory
systematic experiments by Glicksman and is consistent with the older approximate
theory if we allow for a s (e ) value that var- perturbed tip of PVA is longer, with a
ies with the anisotropy of the capillary sharper delineation of the crystallographic
length. The corresponding central equation orientations. Glicksman and Singh (1989)
for dendritic growth (Eq. (2) in Kurz and found that PVA has a ten-fold larger sur-
Trivedi, 1990) should therefore still apply. face energy anisotropy than SCN (see Ta-
ble 2.1). The tip radii and growth rates as a
function of undercooling for both sub-
Pure substances (thermal dendrites) stances scale well when using the values
Fig. 2-15 shows dendrites of two differ- 0.22 and 0.195 for s , respectively (Fig. 2-
ent transparent materials with cubic crystal 16). According to solvability theory, the
structure: face-centered cubic PVA and great difference in the anisotropy constant
body-centered cubic SCN (Glicksman and e should make a larger difference in s (e )
Singh, 1989). Qualitatively, both dendrites (compare with Fig. 2-10). The reason for
look similar, but their branching behavior this discrepancy is not known, and we have
shows some important differences. The un- to leave this point to future research.
(a) (b)
Figure 2-15. Dendrite morphologies of two transparent materials with small melting entropies and cubic crys-
tal structures (plastic crystals); (a) pivalic acid (PVA) and (b) succinonitrile (SCN) (Glicksman and Singh,
1989).
Table 2-1. Experimentally determined dendrite tip quantities.
System Growth s* R2 V l 2 /R d Reference

type [µm3/s]
Thermal dendrites Pure

Succinonitrile (SCN) Free 0.0195 3 0.005 Huang and
Glicksman (1981)
Pivalic acid (PVA) Free 0.022 7 0.05 Glicksman and
Singh (1986, 1989)
Cyclohexanol Free 0.027 Singh and
Glicksman (1989)
Solutal dendrites Alloy

NH4Br – 49 wt.% H2O Free 0.081 ± 0.02 18 ± 3 5.2 0.016 ± 0.004 Dougherty and
Gollub (1988)
SCN – 1.3 wt.% ACE Directional 0.032 * 1300 2.1 ± 0.2 Esaka and
Kurz (1985)
SCN – 4 wt.% ACE Directional 0.037 * 441 ± 30 2.2 ± 0.3 Somboonsuk
et al. (1984)
CBr4 – 7.9 wt.% C2Cl6 Directional 0.044 * 978 ± 8 3.18 Seetharaman
et al. (1988)
C2Cl6 – 89.5 wt.% CBr4 Directional 0.038 * 124 ± 13 3.47 Seetharaman
et al. (1988)
Thermal and solutal Alloy

dendrites
SCN – ACE Free See reference Chopra et al.
(1988)
SCN – argon Free See reference Chopra et al.
(1988)
PVA – 1 wt.% ethanol Free 0.05 ± 0.02 6±1 0.006 ± 0.002 Dougherty
(1991)
PVA – 2/4 vol.% ethanol Free 0.032 ± 0.006 35 6.8 Bouissou et al.
(1989)
* Due to differences in the definitions of s * these values, as given in the corresponding literature, are smaller
by a factor of 2 with respect to the values used by Dougherty and Gollub (1988) and defined in this paper. The
values given here have been obtained by multiplying the original data by a factor of 2 in order to compare with
the same (one-sided) model. (See also Fig. 2-10.)
Koss et al. (1999) have shown that even behind the tip is delayed up to about seven
in microgravity environment there is a radii in PVA. This is quite consistent with
small but significant difference between the recent calculations discussed in Sec.
transport theory and experiment. Again 2.5.4. The ratio of initial secondary arm
this has to be left to the future. spacing l 2 over tip radius R is also indi-
The secondary branch formation which cated in Table 2-1.
starts in SCN at a distance of three tip radii
Free alloy growth

(thermal and solutal dendrites)
In the free dendritic growth of alloys, an
interesting observation has been made by
various authors. For constant undercool-
ing, the growth rate first increases when
small amounts of a second substance are
added to a pure material, then reaches a
maximum, and finally drops and converges
with the pure solutal case. Early experi-
ments in this area by Fujioka and Linde-
meyer were first successfully analyzed by
Langer (1980 c). Fig. 2-17 shows some re-
sults on SCN – ACE alloys from Chopra et
(a) al. (1988). The increase in V is accompa-
nied by a decrease in the tip radius, which
sharpens due to the effect of solute. The
experimental findings can be compared
to two models: Karma and Langer (1984)
(broken line) and Lipton et al. (1987) (full
line). Both models provide at least qualita-
tively good predictions of the observed be-
havior. In their more recent calculations,
Ben Amar and Pelce (1989) concluded that
the simple model by Lipton et al. (1987)
is consistent with their more rigorous ap-
proach.
Table 2-1 gives a summary of represen-
tative results of in situ experiments con-
cerning the dendrite tip.
Large undercoolings
Interesting experiments have also been
performed with pure and alloyed systems
under large driving forces, which reach
values beyond unit undercooling (for ex-
(b)
ample by Wu et al. (1987) and especially by
Figure 2-16. Effect of undercooling on (a) tip radius Herlach et al. (1990 – 1999), see Table 2-2).
and (b) growth rate of the two organic materials Some of these are reproduced in Fig.
shown in Fig. 2-15 (a = D). The experimental results
2-18 together with predictions from IMS
superimpose as they are plotted with respect to dimen-
sionless parameters (Glicksman and Singh, 1989). (Ivantsov – marginal stability) theory (Lip-
ton et al., 1987; Trivedi et al., 1987; Boet-
tinger et al., 1988) (with s (e ) = 0.025) and
including interface attachment kinetics.
(a) (b)
Figure 2-17. Effect of dimensionless composition at constant undercooling of 0.5 K (2.1% of unit undercool-
ing) on (a) dimensionless growth rate and (b) on dimensionless tip radius for free dendritic growth in SCN – ace-
tone alloys (Lipton et al., 1987). Points: experiments (Chopra et al., 1988); solid line: LGK model (Lipton et al.,
1987); interrupted line: Karma and Langer (1984) model.
These results are specifically of interest be- Table 2-2. Dendritic growth velocity measurements in
cause two different techniques, one for the highly undercooled melts; comparison between experi-
ment and theory (Herlach and coworkers, 1999–2000).
determination of the diffusive speed and the
other for the undercooling – growth rate re- Metals and alloys
lationship of the dendrites, have been cou- Co – Pd Volkmann et al. (1998)
pled in order to make the analysis as free Co – V Tournier et al. (1997)
from adjustable parameters as possible. Up Cu Li et al. (1996)
Ni Eckler and Herlach (1994)
to undercoolings of 200 K there is reason-
Ni – Al Barth et al. (1994),
able agreement between experiment and Assadi et al. (1998)
theory using the best fit for the measured Ni – B Eckler et al. (1991 a, 1992, 1994)
diffusive speed and the liquid diffusivity, Ni – C Eckler et al. (1991 b)
VD = 26 m/s and DL = 2.7 ¥ 10– 9 m2/s (Ar- Ni – Si Cochrane et al. (1991)
Ni – Zr Schwarz et al. (1997),
nold et al., 1999). The lower V (∆T )-curve
Arnold et al. (1999)
is for a constant (equilibrium) value of the
distribution coefficient k (V) = ke showing Intermetallics
the importance of the appropriate velocity CoSi Barth et al. (1995)
FeAl Barth et al. (1997)
dependence of the distribution coefficient. FeSi Barth et al. (1995)
At higher undercoolings other phenom- Ni3 Al Assadi et al. (1996)
ena take over. Nix Sny Barth et al. (1997)
One example is the grain refinement be- NiTi Barth et al. (1997)
yond a certain undercooling. This structure Nix Tiy Alz Barth et al. (1997)
has been explained by dendrite fragmenta- Semiconductors
tion due to morphological instability of the Ge Li et al. (1995 a, 1996)
fine dendrite trunks (Schwarz et al., 1990; Ge – Cu Li and Herlach (1996)
Ge – Si Li et al. (1995 b)
Karma, 1998).
Ge – Sn Li and Herlach (1996)
is the cold atmosphere above it, in casting

iron in a foundry, it is the cold sand mold,
into which the heat flow is directed.
At first, it may seem surprising to think
that anything interesting should happen at
the solid – liquid interface during this pro-
cess. In contrast to the situation described
earlier, in Secs. 2.4 and 2.2.4, the solid in a
casting process is cold and the liquid is hot,
so that we would expect the interface to be
stable against perturbations.
However, so far we have just considered
the solidification of a one-component ma-
terial, while in reality a mixture of materi-
als is almost always present, even if one of
the components is rather dilute. If, there-
fore, we assume that material diffusion is
the rate-determining (slow) mechanism,
while heat diffusion is much faster, the ori-
gin of a destabilization of the flat interface
can be easily understood on a qualitative
basis. We may consider one of the two
components of the liquid as an “impurity”,
which, instead of being fully incorporated
Figure 2-18. Dependence on total bath undercooling
of Ni – 1 wt.% Zr alloy. (a) Dendrite growth velocity,
into the solid, is rejected at the interface.
as measured (dots), and results from Ivantsov – mar- Such excess impurities have to be diffused
ginal stability dendrite growth theory using the val- away into the liquid in much the same way
ues VD = 26 m/s and Dl = 2.7 ¥ 10– 9 m2/s. The lower as latent heat has to be carried away in the
curve in (a) is for local equilibrium partition. (b) The case of a pure material as a rate-determin-
calculated dendrite tip radius, (c) the computed inter-
face compositions and (d) the individual undercool-
ing mechanism. Accordingly, precisely the
ing contributions: thermal undercooling DTt , consti- same destabilization and subsequent for-
tutional undercooling DTc , curvature undercooling mation of ripples and dendrites should oc-
DTr , and attachment kinetic undercooling DTk (Ar- cur.
nold et al., 1999). Based on these qualitative arguments
we can expect the following modification
of the Mullins – Sekerka instability (Sec.
2.2.4) to occur in the present situation of
2.5 Directional Solidification directional solidification. The diffusion of
material together with capillary effects pro-
Directional solidification is the most fre- duces a spectrum for the growth rates or the
quent way in which a material changes its decay rates similar to Eq. (2-50), while the
state from solid to liquid. The necessary re- temperature field acts as a stabilizer, inde-
moval of the latent heat of freezing usually pendent of the curvature of the interface,
occurs in a direction prescribed by the lo- when a constant term (independent of V )
cation of heat sinks: for a freezing lake, it inside the brackets of Eq. (2-50) is sub-
2.5 Directional Solidification 121
tracted. At low solidification rates, the flat The Helmholtz energy F is then
interface is stable; above a critical speed, it
F (T, V, Ni , …) = U – T S ;
becomes unstable against the formation of
ripples, cells and dendrites. dF = – S dT – P dV + S m i dNi (2-97)
i
In the next section, a few thermody-
and the often-used Gibbs energy G is
namic questions related to interface prop-
erties in two-component systems are con- G (T, P, Ni , …) = U – T S + P V = S m i dNi ;
sidered, before describing patterns in direc- i
dG = – S dT + V dP + S m i dNi (2-98)
tional solidification. i
At atmospheric pressure in metallurgical

2.5.1 Thermodynamics applications, the differences between F and
of Two-Component Systems G often can even be neglected, but gener-
ally the Gibbs energy, Eq. (2-98), is most
There is a vast amount of literature avail- frequently used. It follows immediately that
able on the thermodynamics of solidifica- the chemical potentials m i are defined as
tion and on multi-component systems (for
example Callen, 1960; Baker and Cahn, ⎛ ∂G ⎞
⎜ ⎟ = mi (2-99)
1971). Despite this fact, to further the clar- ⎝ ∂Ni ⎠ T, P
ity of presentation, we would like to at least
sketch the tools that may be used to gener- The thermodynamic equilibrium for a sys-
alize some approximations which will be tem is defined by the minimum of the re-
made in the next sections. spective thermodynamic potential with re-
The fundamental law of thermodynam- spect to all unconstrained internal parame-
ics defines entropy as a total differential in ters of the system. If the system consists
relation to energy and work: of two subsystems a and b in contact with
each other, then in thermal equilibrium the
dU = T dS – P dV + S m i dNi (2-94) temperatures, pressures, and the chemical
i
potentials for each particle type i must be
with energy U, entropy S, volume V, pres- equal:
sure P, particle numbers Ni for each species
Ta = Tb , Pa = Pb , m i, a = m i, b (2-100)
and chemical potential m i . The energy is a
homogeneous function of the extensive For the case under consideration we have a
variables solid phase a (with assumed low concen-
(2-95)
tration of B atoms) and a liquid phase b
U (b S, b V, b Ni , …) = b U (S, V, Ni , …)
(with higher concentration of B atoms).
with an arbitrary scale parameter b > 0. For simplicity, we further assume that the
With the differentiation rule d (X Y ) = atomic volumes of both species are the
X dY + Y dX, other thermodynamic poten- same and unchanged under the solid – liq-
tials Ũ are obtained from U by Legendre uid transformation.
transformations For a system with a curved interface
between a solid and a liquid of different
Ũ = U – S Xj Yj (2-96) compositions, the chemical potentials can
j
be calculated as follows. Assuming that NA
where Xj are some extensive variables, and particles of solvent and NB particles of
Yj the corresponding intensive variables. solute are given, there will be an unknown
number of NS particles in the solid and NL lar, the spatial continuity of chemical po-
particles in the liquid, whose composition tentials together with continuous tempera-
is still undetermined. Define the number n ture and pressure guarantees local thermal
as the number of B particles in the solid. equilibrium, which we will assume to hold
Keeping NS and n initially fixed, the Gibbs in most of the following discussions.
potential is obtained as We now come to the discussion of a typ-
ical phase diagram for a two-component
G (NA , NB , NS , n) = NS gS (CS ) + NL gL (CL ) system (Fig. 2-19). The vertical axis de-
+ 4 p R2 g (2-101) notes the temperature, the horizontal axis
the relative concentration of “solute” in a
where g is the surface free energy density, “solvent”, or, more generally, B atoms rel-
R is the radius of the solid sphere ative to A atoms.
NS = –43 p R 3 (unit atomic volume), and gS At high temperatures, T > T0 , the system
and gL are free energy densities for homo- is liquid, regardless of concentration. As-
geneous solids or liquids at concentrations suming a concentration C• to be given in-
CS and CL . Removing the constraints on NS itially, we lower the temperature to T1 . At
and n, we obtain thermodynamic equilib-
rium by minimizing G with respect to NS
and n, so that G = G (NA , NB ). Together
with the chemical potentials from Eq. (2-
99), this gives
g
m − m0 = − K (2-102)
( CL0 − CS0 )
with curvature K = 2/R and m = m B – mA be-

ing the difference in chemical potentials
between solute and solvent. The values m 0 ,
C L0 , C S0 correspond to equilibrium at g = 0
as a reference, around which g (C ) was lin-
earized. Here g was assumed to be inde-
pendent of curvature and concentrations,
but this can easily be incorporated (for ef- Figure 2-19. Typical solid – liquid phase diagram for
a two-component system with the possibility of eu-
fects of surface segregation, for example). tectic growth. C̃ L , C̃S are the liquidus and solidus
This equation is the boundary condition lines with a solid – liquid two-phase region in
for the chemical potential on the surface of between. For a relatively low concentration C• of so-
a solid sphere of surface tension g coexist- lute in the liquid, a single solid phase at the same
ing with a surrounding liquid of higher composition CS0 = C• can show stationary growth. In
directional solidification, a positive temperature gra-
concentration CL of B atoms. From a prac- dient ∂T /∂z is assumed to be given perpendicular to
tical point of view, the formation in terms the solid – liquid interface and to advance in + z-di-
of chemical potentials does not look very rection. The advancing interface chooses its position
convenient, as they are not directly mea- to be at temperature T0 , the concentration in the liq-
surable. From a theoretical point of view, uid at the interface is then at CL0. Ahead of the inter-
face (z = 0), the concentration profile C (z) decays
this has advantages, as the chemical poten- towards C• at z = • (dashed curve). As long as
tial is the generalized force controlling C (z) > C̃ L (T (z)) with T (z) = T0 + z ∂T /∂z, a flat inter-
matter flow and phase changes. In particu- face remains stable.
this temperature, we first hit the liquidus concentration profile ahead of the interface
line C̃L (T ), and the system begins to solid- which decays exponentially from C L0 at the
ify, producing a solid of very low concen- interface to C• far away from the interface.
tration marked by the solidus line C̃S (T ). But why should it decay to C• (or why
When the temperature is slowly lowered, should the interface choose a temperature
solidification becomes complete at T0 . At position such that C S0 = C• )?
lower temperatures, the whole system is The answer is quite simple, and again
solid. was given in similar form in Sec. 2.2.3, Eq.
The region between C̃S (T ) and C̃L (T ) is (2-33) for the pure thermal case: if C S0 were
the two-phase region: if we prepare a not identical to C• , then during the solid-
system at a concentration between C S0 and ification process there would be either a to-
C L0 at high temperatures and then quickly tal increase (or decrease) of concentration
quench it to T0 , the system starts to sepa- – which clearly is impossible – or at least
rate into a solid phase of concentration C S0 the concentration profile could not be sta-
and a liquid phase of C L0 . In practice, this is tionary.
a very slow process, with lengths varying This is a rather strict condition, which
with time t approximately as t 1/3 (Lifshitz we can reformulate as follows: if we im-
and Slyozov, 1961; Wagner, 1961). pose a fixed temperature gradient ∂T /∂z
In the case of directional solidification, a and move this at fixed speed V0 over a
thermal gradient in the system defines a di- system of concentration C• at infinity in
rection such that the liquid is hot and the the liquid (toward the liquid in the positive
solid is cold. A flat interface may then be z-direction), then the interface will choose
present at a position in space at tempera- a position such that its temperature is at T0 ,
ture T0 . For equilibrium between solid and the concentration in the solid will be C S0
liquid at that temperature, the concentra- (averaged parallel to the interface), and the
tion in the solid must be at C S0 = C• , and in liquid concentration at a flat interface will
the liquid at C L0 = C̃L (T0 ). We now assume be at C L0 . This follows from global conser-
that the liquid at infinity has concentration vation of matter together with the imposed
C• . Clearly there must be a decrease in stationary solidification rate.
concentration as we proceed from the inter- As a final point, we can even derive a
face into the liquid. In order to maintain condition for the stability of the interface.
such an inhomogeneous concentration, the The concentration profile in the liquid will
interface must move toward the liquid. decay exponentially with distance z away
In other words, when the liquid of com- from the interface as
position C L0 freezes, the solid will only
CL (z) = (C L0 – C• ) e – 2 z /l + C• (2-104)
have a concentration C S0 . The difference in
concentrations by analogy to Eq. (2-32). Since we assume
DC = C L0 – C S0 (2-103) the temperature gradient
is not incorporated but is pushed forward GT = ∂T /∂z > 0 (2-105)

by the advancing interface and must be car- to be fixed, the temperature varies linearly
ried away through diffusion into the liquid. with distance z from the interface. This may
This is equivalent to the latent heat gener- be written (by GT = (T – T0 )/z) as
ated by a pure freezing solid, discussed in (2-106)
Sec. 2.2.3. Therefore, we expect a spatial CL (T ) = (C L0 – C• ) e – 2 (T – T0 ) / (l GT ) + C•
and incorporated into Fig. 2-19 as a 2.5.2 Scaled Model Equations

dashed-dotted line. Note that the diffusion
length l is again defined as l = 2 D/V0 , with A theoretical analysis of practical situa-
D being the diffusion constant of solute tions of directional solidification suffers –
atoms in the solvent, and V0 the interface among other problems – from the many rel-
velocity imposed by the advancement rate evant parameters entering the description.
of the temperature gradient. The usual way to proceed in such cases is
From Eq. (2-106), it is obvious that the to scale out as many parameters as pos-
dashed-dotted concentration line in Fig. 2- sible, writing the problem in dimensionless
19 converges very quickly to C• for high variables. We have done this already in the
solidification speeds V0 . As long as that discussion of free dendritic growth by in-
concentration line is fully in the liquid re- troducing the dimensionless diffusion field
gion of the phase diagram, nothing specific u; in hydrodynamic applications, it is com-
happens. But if the dashed-dotted line mon practice to use Reynolds and Rayleigh
partly goes through the two-phase region numbers (Chandrasekhar, 1961).
between C̃S (T ) and C̃L (T ), the liquid in For our present problem, we will pro-
front of the interface is supercooled! This ceed in an analogous way. The first step
implies the possibility of an instability of is to express all experimental parameters
the solid – liquid interface, which is com- (wherever possible) in length units (i.e.,
pletely analogous to the discussion in Sec. diffusion length, capillary length, etc.). For
2.2.4. presenting results, we divide these lengths
A sufficient condition for stability of the by the thermal length introduced below, as
interface in directional solidification is this is a macroscopic length which will ap-
therefore proximately set the scale at the onset of the
instability.
V dC˜ L Directional solidification involves chem-
DC < (2-107)
D GT dT ical diffusion of material together with heat
diffusion. As heat diffusion is usually
so that the dashed-dotted curve remains faster by several orders of magnitude, we
outside the two-phase region (Mullins and may often assume constant temperature
Sekerka, 1963; Langer, 1980 a). Here we gradients to exist in the liquid and in the
have assumed that material diffusion in the solid. Furthermore, we will also assume
solid can be ignored. In fact, in practical that there are equal thermal diffusivities in
situations, violation of this condition typi- liquid and solid, which is often the case
cally means “instability” of the interface, within a few percent, but this has to be
so that cellular or dendritic patterns are checked for concrete applications. The dif-
formed. The reason for this latter conclu- fusion field to be treated dynamically then
sion is that the effect of stabilization due to corresponds to the chemical concentration.
capillarity (or surface tension) is rather It is clear from the discussion in the pre-
weak for typical experiments at threshold. vious section that for a flat interface mov-
In summary, in this section we have ing at constant speed there is a concentra-
derived both the boundary condition – in tion jump DC = C L0 – C S0 across the inter-
terms of chemical potential – for a curved face, while in the liquid and in the solid,
interface and a basic criterion for interface the term C• is approached asymptotically
stability during directional solidification. because of the condition of stationary
movement, together with the global con- with segregation coefficient k (equilibrium
servation of matter. We therefore normalize values assumed) defined as
the diffusion field in the liquid to
dT dT
k= (2-113)
C (Liquid) − C∞ dC˜ L dC˜ S
uL = (2-108)
DC
through the slopes of the liquidus and sol-
so that it varies from one to zero in the pos- idus lines. When they intersect at {Tm ,
itive z-direction from the interface at z = 0 C = 0} this is equivalent to the conventional
to z = •. If the interface is not at position definition k = CS /CL , but in the above for-
z = 0 but at z , we must require uL = 1 – z /l T , mulation, k = 1 may also be true for a con-
because at a distance stant jump in concentrations, independent
of temperature.
DC dT
lT = (2-109) The conservation law at the interface
GT dC˜ L z = z finally becomes
the liquidus concentration has reached the V^ {1 + (1 – k) (uL – 1)}
asymptotic value. This is the thermal = – DL n̂ · —uL + DS n̂ · —uS (2-114)
length which we assume to be fixed by the
thermal gradient GT , the concentration where V^ is the interface velocity in direc-
jump DC, and the liquidus line C̃ (T ), tion n̂ normal to the interface. For k = 1, the
which is here assumed to be a straight line brackets {…} give 1, corresponding to a
in the T vs. C diagram. constant concentration jump, while for
The equation of motion in quasi-station- k = 0, they give uL , since for a solid in Eq.
ary approximation then becomes, in anal- (2-112), uS = 0.
ogy to Eq. (2-52) with l = 2 DL /V This standard model for directional so-
lidification (Saito et al., 1989) therefore
1 2 ∂uL consists of Eqs. (2-108) to (2-114), to-
∂ t uL ≈ 0 = ∇ 2 uL + (2-110)
DL l ∂z gether with an additional diffusion equa-
tion as Eq. (2-110) inside the solid phase.
This equation applies equivalently to the
The open point is finally the relation of
solid but with a different chemical diffu-
the capillary length d (Eq. (2-111)) with ex-
sion length l ¢ due to different chemical
perimentally measurable material parame-
diffusion constants. The boundary condi-
ters. As a first step, we interpret the u-field,
tion in analogy to Eq. (2-88) obviously be-
Eq. (2-108), as a scaled form of the chemi-
comes
cal potential m (see Sec. 2.5.1) near T0
uL (z ) = 1 – d K – z (x, t)/l T – b V^ (2-111)
m − m∞
uL = (2-115)
where we now have D = 1 as the first term DC ( ∂m / ∂C˜ L )
on the r.h.s., with curvature K being posi-
assuming that linearizing m around its equi-
tive for the tip of a solid nose pointing into
librium value at the liquidus line C̃L (T0 ) is
the liquid. The capillary length d is dis-
sufficient to describe its dependence upon
cussed below and interface kinetics with
C. By the definitions in Eq. (2-102) to-
b ≠ 0 will be discussed in Sec. 2.5.4. The
gether with Eq. (2-111), we now obtain the
solid boundary condition is simply
capillary length in the form given in Eq.
uS = k (uL – 1) (2-112) (2-86) for the chemical case. Here we have
generalized to anisotropic g as derived all parameters is that qualitatively different

in Eq. (2-21). It can finally be related to behavior always corresponds to different
measurable quantities using the Clausius – ratios of length scales or time scales rather
Clapeyron equation for the latent heat L m than some differences in absolute mea-
of freezing of a solution at Tm sures, and, consequently, this presentation
allows for a more intuitive formulation of
L m = – Tm DC (d m /dT )coex (2-116)
results.
where (d m /dT )coex is the slope of the coex-
istence line when m is plotted against T. 2.5.3 Cellular Growth
Together with
A plane interface between the solid
⎛ dm ⎞ ⎛ ∂m ⎞ dC˜ L ∂m phase and the liquid phase of a two-compo-
=⎜ ⎟ + (2-117)
⎝ dT ⎠ coex ⎝ ∂C ⎠ dT ∂T nent system tries to locate its position in a
thermal gradient so that the chemical po-
this gives for the chemical capillary length,
tentials of both components are continuous
d, in the limit of small DC
across the interface. Under stationary
Tm growth conditions, that position corre-
d = [g + g ′′ (J )] (2-118) sponds to a temperature, such that the con-
DC Lm | ∂T / ∂C˜ L |
centration in the solid (solidus line of the
Certain approximations used here, such as phase diagram) is equal to the concentra-
the linearization involved in Eq. (2-115) or tion in the distant liquid. This growth mode
neglecting of ∂m /∂T in Eqs. (2-118), may persists for velocities up to a critical veloc-
not be safe for the case of a large segrega- ity, above which the interface undergoes a
tion coefficient k ≈ 1 or when DC is large Mullins – Sekerka instability toward cel-
(Langer, 1980 a). In most practical applica- lular structures. A necessary condition for
tions, however, this is a minor source of er- this instability to occur follows from Eq.
ror in comparison with other experimental (2-107), which can be written in terms of
uncertainties. Furthermore, the concentra- chemical diffusion length l = 2 D/V and
tion jump DC in Eq. (2-118) is kept fixed, thermal diffusion length lT , Eq. (2-109), as
while in reality it should correspond to the
l/lT ⱗ 2 (2-119)
temperature-dependent difference in con-
centration between the liquidus and solidus The 2 comes from the specific definition of
lines. Both for slow and fast growth rates, l, and the inequality for instability is only
however, this only gives a minor correc- approximate because minor surface tension
tion, and we will ignore its effect in order effects have not yet been considered here.
to facilitate comparison with free dendritic Incorporation of surface tension reveals
growth. that the instability first occurs for a critical
In summary, with this model we now wavelength l c larger than the stability
--
have all the ingredients to discuss some ba- length l s= 2 p ÷d l. Slightly above the criti-
sic features of directional solidification by cal pulling speed Vc , the interface makes
analytical and numerical tools. The presen- periodic structures of finite amplitude. This
tation in the scaled form may not seem at was analyzed by Wollkind and Segel
first to be the most convenient means of di- (1970), and for other specific cases, by
rect comparison with experiments. Its great Langer and Turski (1977). A more general
advantage over an explicit incorporation of treatment was given by Caroli et al. (1982).
The result of these investigations is that curve does not define such a normal bifur-
in a diagram of pulling speed V versus cation but rather an inverse bifurcation.
wavelength l there exists a closed curve of This means that immediately above Vc
neutral stability, Fig. 2-20. At fixed V, a large amplitude cells with deep grooves are
small amplitude perturbation of the inter- formed. A time sequence of the evolution
face at a wavelength on that curve neither of a sinusoidal perturbation into elongated
grows nor decays. Perturbations at wave- cells at 1% above Vc due to inverse bifurca-
lengths outside that curve decay, inside the tion is shown in Fig. 2-21. This can be
curve they grow to some finite amplitude. understood theoretically (Wollkind and
This is similar to periodic roll patterns in Segel, 1970; Langer and Turski, 1977; Ca-
the Rayleigh – Benard system of a fluid roli et al., 1982) by means of an amplitude
heated from below (Chandrasekhar, 1961), equation valid near Vc :
but here a maximal speed, Va , is present,
above which a flat interface is absolutely ∂A ⎛ V − Vc ⎞
⎟ A − a1 | A | A + …
2
=⎜ (2-120)
stable. For normal alloys, this speed is very ∂t ⎝ Vc ⎠
high, while for liquid crystals, it is more
easily accessible in controlled experiments where A is the (possibly complex) ampli-
(Bechhoefer et al., 1989). The diffusion tude of a periodic structure exp (i k x) with
length at Va is of the order of the capillary k = 2 p /lc . The coefficient a1 is called the
length. Landau coefficient. If it is positive, we have
At low speeds, but slightly above the
critical velocity, Vc sinusoidal “cells” will
be formed for systems with segregation co-
efficients k near unity. For small segrega-
tion coefficients, however, the neutral
Figure 2-20. Neutral stability curve for a flat solid –

liquid interface in directional solidification (sche-
matic). The dependence of the growth rate (pulling
speed) V on the wavelength l of the interface pertur- Figure 2-21. Time evolution of an interface from si-
bation is approximately V ~ l – 2, both for the short nusoidal to cellular structure slightly above the criti-
and long wavelength part of the neutral stability cal threshold Vc for the case of inverse bifurcation. A
curve, Vc and Va are the lower critical velocity and secondary instability quickly leads to a halving of the
upper limit of absolute stability, respectively. wavelength.
-----
a normal bifurcation with | A | ~ ÷V – Vc , has been made (Brattkus and Misbah,
while for a1 < 0, the third order term does 1990). A phase-diffusion equation has been
not stabilize the pattern but allows very derived describing the temporal evolution
large amplitudes leading to elongated cells of a pattern without complete periodic vari-
(Fig. 2-21), which will be stabilized by ation of the interface. The basic idea is to
some higher-order effects. replace the periodic trial form exp (i k x)
A second phenomenon is usually asso- by a form exp (i Q (x, t)) so that q (x, t) =
ciated with this inverse bifurcation, namely, ∂Q /∂x is now no longer a constant but is
the splitting of the wavelength l c Æ l c /2. slowly varying in space along the interface
Qualitatively, this is understandable from and evolving with time. We can derive a
nonlinear corrections since the squaring of nonlinear phase diffusion equation
the original pattern ~ exp (i k x) produces
∂t q = ∂x {D̃ (q) ∂ x q} (2-122)
terms ~ exp (i 2 k x). This effect has clearly
been observed in experiments (de Che- with a diffusion coefficient D̃ (q) depend-
veigne et al., 1986). ing in a complicated way on q. The proce-
We will discuss some aspects of the very dure is well known in hydrodynamics and
high speed region in Sec. 2.5.7 but devote it is associated there with the so-called
the main part of the following discussions Eckhaus instabilility (Eckhaus, 1965). This
to the most interesting region for practical instability eventually causes an (almost)
purposes, which is not too close to the periodic spatial structure to lose or gain
upper and lower bounds of the growth rate one “period”, thereby slightly changing the
Va and Vc . average wavelength. In directional solidifi-
Approximating by straight lines the neu- cation, the result (Brattkus and Misbah,
tral curve of the logarithmic plot Fig. 2-20 1990) is shown in Fig. 2-22, where velocity
in the intermediate velocity region, we find is plotted against wavenumber in a small
for both the small and the large l limits the interval above the critical velocity. The full
relation line is the neutral curve, the full triangles
mark the Eckhaus boundary of phase
V l 2 ≈ constant (2-121)
stability. A periodic (sinusoidal) pattern is
Again we have recovered the form of Eq. stable against phase slips only inside the
(2-1) mentioned in the introduction as a region surrounded by triangles, thereby al-
scaling law where l here is the cell spac- lowing for an Eckhaus band of stationary
ing. This suggests that the cellular pattern periodic solutions with a substantially re-
formed in actual experiments would also duced spread in wavenumbers as compared
follow this behavior. Unfortunately, this to the linear stability results. Note also that
problem has not yet been settled to a satis- the results for phase stability (dashed line)
factory degree from a theoretical point of from the amplitude equation only hold in
view. This is partly due to the difficulty of an extremely small region above Vc , while
finding good analytical approximations to only 20% above Vc , it shows no overlap
the cellular structures, which makes nu- with the result from the present analysis
merical calculations necessary to a large (triangles). The short-wavelength branch
degree. We will return to this point in Sec. has a very complicated structure, while the
2.5.5. long-wavelength branch far from the
For small amplitude cells obtainable threshold scales as l Eck ~ V –1/2, again like
under normal bifurcation, some progress the neutral curve. This also seems to be in
Figure 2-22. Stability diagram V vs. 2 p/l near the lower critical threshold for a flat moving interface in direc-
tional solidification. The solid line is the neutral stability curve Fig. 2-20, the dotted curve is the most danger-
ous mode, the dashed curve is the limit of the Eckhaus stability from the amplitude equation. The triangles mark
the Eckhaus limit as obtained from the full nonlinear analysis (Brattkus and Misbah, 1990), with stable cellular
interfaces possible only inside that region. The band width of possible wavelengths for cellular interfaces ac-
cordingly is smaller by a factor of ≈ 0.4, as compared to the band width given by the neutral (linear) stability
curve.
agreement with experimentally observed narrow, similar to Fig. 2-21 (Ungar and
results, as discussed later (Billia et al., Brown, 1984 a, b, 1985 a, b, Karma, 1986,
1987, 1989; Somboonsuk et al., 1984; Kessler and Levine, 1989, McFadden and
Esaka and Kurz, 1985; Eshelman and Tri- Coriell, 1984, Pelce and Pumir, 1985).
vedi, 1987; Faivre et al., 1989; Kurowsky,
1990).
At higher velocities, the cells quickly be-
come elongated (Fig. 2-23) with deep
grooves forming bubbles. This was first
obtained through numerical calculations by
Ungar and Brown (1984 a, b, 1985 a, b).
Calculations with a dynamical code in
quasi-stationary approximation (Saito et
al., 1989) confirmed the stability of these
structures with respect to local deforma-
tions and short-wavelength perturbations.
The long-wavelength Eckhaus stability has
not been investigated yet for these cells.
All calculations were made in two dimen-
sions, which are believed to be appropriate
for experiments of directional solidifica-
tion in a narrow gap between glass plates.
At higher velocities and wavelengths (or Figure 2-23. Computed example for a deep cellular
cell sizes) not much smaller than the dif- interface at V ≈ 5 Vc with bubble formation at the bot-
fusion length, the grooves become very tom of the groove (Saito et al., 1990).
If the velocity is fixed and the wavelength For comparison with experiments, it is
l is reduced significantly below the diffu- useful to draw a V vs. l diagram (Fig. 2-
sion length l, the Saffmann – Taylor limit is 24). Here the full line is again the neutral
reached (Brener et al., 1988; Dombre and curve, the broken line is the most danger-
Hakim, 1987; Kessler and Levine, 1986 c), ous (or most unstable) mode, and the dot-
which is equivalent to a low-viscosity fluid ted line denotes the relation l = l , where the
being pushed into a channel of width l diffusion length is equal to the imposed
filled with a high-viscosity fluid. The low- wavelength. The asterisks mark some de-
viscosity fluid forms a finger just like the tailed numerical investigations (Saito et al.,
solid in directional solidification. Near the 1989). The asterisk furthest to the left is
tip, the width of the finger l f corresponds to close to the above-mentioned Saffmann –
Taylor limit. At slightly higher wave-
lf = D l
lengths, where l < l still holds, we are in a
with cell spacing l , where D < 1 is the ac- scaling region, where the radius of curva-
tual supercooling at the tip (remember that ture at the tips of the cells is about 1/5 of
D = 1 for a flat interface at z = 0, and D = 0 the cell spacing, as also found experimen-
for a flat interface at z = l T ). This serves tally (Kurowsky, 1990). All these consider-
to verify the consistency of numerical ations give sufficient confidence that the
calculations (Saito et al., 1989). An even numerical calculations may also provide
more detailed analysis was carried out by insight into the mechanism of directional
Mashaal et al. (1990). solidification for the most interesting case
Figure 2-24. “Phase”-diagram log (V ) vs. log (l) for interface patterns in directional solidification. The solid
and dashed curves denote the neutral stability curve, and the dashed-dotted curve the most dangerous mode.
Asterisks mark fixed parameter values discussed hereafter. The lower critical threshold here is Vc ≈ 1, lc ≈ 0.5 for
velocity and cell spacing. For other parameters see text. At low pulling speeds and high wavelengths cellular
patterns with narrow grooves are found (a). At very short wavelengths and moderate speeds cellular patterns
with wide grooves are found, consistent with theories for viscous fingering. At high pulling speeds, such that
the cell spacing l is significantly wider than the diffusion length l, side-branching dendritic arrays are formed
(c) (Saito et al., 1990). The speeds are still much smaller than the absolute stability limit Va .
of dendritic arrays formed at higher growth other, until finally they may behave like in-
rates. dividual isolated dendrites.
A few words on numerical methods and In order to test this hypothesis, a series
system parameters may be in order before of numerical experiments were performed
we discuss the dendritic region. The nu- at a fixed cell spacing and increasing pull-
merical code is equivalent to the one used ing velocity (Saito et al., 1989). The nu-
for the free dendritic case with the modifi- merical parameters of the model were rep-
cation that it is necessary to integrate over resentatively taken to correspond to steel
several cells to arrive safely outside the with Cr – Ni ingredients (Lesoult, 1980). In
diffusion length. Furthermore, in principle, dimensionless units, the critical velocity
diffusion has to be considered in both the and wavelength for the plane-front in-
liquid and the solid. Since the diffusion co- stability were Vc = 1.136, l c = 0.514. The
efficient for material in the solid is usually anisotropy of the capillarity length was not
much lower than it is in the liquid, it is known and was taken as e = 0.1 to allow for
found that diffusion in the solid alloy can comparison with the previously mentioned
usually be neglected on time scales for the calculations on the free dendritic case. The
formation of cells. For long durations, of cellular wavelength was fixed to l = 0.36,
course, microsegregation takes place, and corresponding to the asterisks at increasing
solid diffusion then becomes important velocity and constant l in Fig. 2-24.
(Kurz and Fisher, 1998). At the lowest velocity still below the
A more serious difficulty in directional l = l dividing line, rounded cells were ob-
solidification is the large number of param- served; the tip was not well approximated
eters defining the system. We will concen- by a parabola. At higher speeds (V = 12) the
trate here on typical parameter values used parabolic structure of the tip became vis-
in experiments performed for some trans- ible, Fig. 2-25, and at even higher speeds
parent materials between glass plates. Sev- (V = 20) the dendritic structure with side
eral tests and specific calculations also branches was fully developed, Fig. 2-26.
done for alloys, however, indicate that a We can now compare the resulting tip ra-
large part of the results can be carried over dius with the predictions from free den-
to these more relevant situations from a dritic growth. Note that in the present case
metallurgical point of view without qual- the velocity is fixed rather than the super-
itative changes. cooling, so that the dendrite now uses a
supercooling corresponding to the given
velocity. This means that the tip of the den-
2.5.4 Directional Dendritic Growth
drite is no longer at a position in the tem-
The diagram in Fig. 2-24 showing veloc- perature-gradient field like a flat interface,
ity versus l in logarithmic form indicates but has advanced toward the warmer liq-
that qualitatively different behavior may be uid.
expected depending on whether the diffu- Fig. 2-27 contains the ratio of the tip ra-
sion length is larger or smaller than the cell dius divided by the radius from scaling, Eq.
spacing. In the previous section, we dis- (2-68) (where the Péclet number P was
cussed the first case. When the diffusion used in the original form as the ratio of tip
length becomes smaller than the cell spac- radius to diffusion length). Furthermore,
ing we expect that the individual cells be- this figure gives the ratio of the tip radius
come more and more independent of each to the Ivantsov radius, which comes from
Figure 2-25. Transition from needle-shaped to dendritic cells at increasing pulling speeds. V = 4 is below the
dotted line in Fig. 2-24, and V = 12 above it. Parabolas adjusted to the tip radius are not a good fit to the profiles.
the Péclet number through the relation for

the supercooling at the tip, Eq. (2-56).
The data are instantaneous measurements
rather than time-averaged measurements.
It is obvious from Fig. 2-27 that the scaling
relation, Eq. (2-68), holds very well at
rather low speeds, where neighboring cells
still interact substantially through the dif-
fusion field, while the relation from the
Ivantsov formula for the Péclet number
only holds at higher velocities. The obvious
reason for the latter deviation at low veloc-
ities is that the Ivantsov relation represents
a global conservation law for an isolated
parabolic structure, which is clearly not
valid when several cells are within a diffu-
sion length.
The observation that the scaling relation,
Eq. (2-68), is very robust obviously has to
do with the fact that it results from a solv-
ability condition at the tip of the dendrite,
which is only very weakly influenced by
Figure 2-26. Time sequence of a dendritic array at
deformations further down the shaft.
V = 40, e = 0.1, corresponding to point (c) in Fig. 2-
24. The starting structure corresponds to V = 12, sim- In the same study, it was also confirmed
ilar to Fig. 2-25. The cellular array quickly converges that the side branches fulfilled precisely
to a stationary side-branching mode of operation. the same scaling relation (Fig. 2-13) as the
Figure 2-27. Ratios R tip /R x of tip radius computed numerically (Fig. 2-24) over two theoretical predictions,
where R x is either the Ivantsov radius (circles) or the radius from solvability theory (asterisks). See also Fig.
2-12. The result is in nearly perfect agreement with the solvability theory down to very low speeds in the cellular
region. The Ivantsov radius (for free growth) is not a good approximation there, as the diffusion fields of neigh-
boring cells strongly overlap. At high speeds, essentially free dendritic growth is confirmed (Saito et al., 1990).
free dendrites at relatively low speeds of Up to this point these investigations

V = 12 shown in Fig. 2-25. In this case, the were done at a constant anisotropy of
side branches are just beginning to show e = 0.1 of the capillarity length. In Fig.
up, while the tip is not very noticeably par- 2-28 the normalized supercooling D has
abolic. been plotted against velocity V, where
Figure 2-28. Supercooling at the tip of a cell or dendrite vs. pulling speed as obtained from numerical simula-
tion. For a flat interface at low speeds, the global conservation law forces D = 1, then it decreases until l ≈ l , and
finally approaches the slowly increasing relation D (V ) obtained for the free dendritic case (see also Fig. 2-40).
The expected dependence on capillary anisotropy e is also recovered.
D = 1 for a flat interface at stationary

growth. Two sets of data for e = 0.1 and
e = 0.2 are shown. If we increase the pull-
ing speed above the critical value V ≈ 1, the
supercooling at the tip of the cellular pat-
terns first decreases, because the forward
bulges come into a range of higher tem-
perature. At intermediate velocity, D goes
through a minimum and finally approaches
the broken lines corresponding to the scal-
ing relation, Eq. (2-68), together with the
Ivantsov relation, Eq. (2-56). At intermedi-
ate velocities, the supercooling D is above
the corresponding curve meaning that the
Péclet number, and therefore, the Ivantsov
radius is larger than expected from the free
dendritic scaling. Figure 2-29. Pronounced parabolic directional den-
drites at speed V = 40, e = 0.2, as used in Fig. 2-28.
This is in agreement with Fig. 2-27 Note that the tip-radius here is about 0.03 and the
shown above. The minimum of the D ver- short-wavelength limit of neutral stability 0.05 in
sus V relation is in the range where the dif- units of the cellular spacing. This is qualitatively
fusion length is comparable to the cell consistent with experimental observations of large
spacing, as expected from Fig. 2-24. interdendritic spacings in units of tip radii. Tip split-
ting was only observed at much lower values of cap-
As a final example, Fig. 2-29 shows a illary anisotropy.
dendritic array at the relatively high veloc-
ity V = 40 at anisotropy e = 0.2. As in free
dendritic growth, the structure appears
isotropy b 4 of the kinetic coefficient, a
sharper than the structure in Fig. 2-26 at
scaling relation similar to Eq. (2-68) was
smaller anisotropy.
derived by Brener and Melnikov (1991):
The opposite case of extremely small an-
isotropies has not yet been analyzed in ⎛ 2 DL b ⎞
9/ 2
great detail, and it is rather unclear what bV = sb ⎜ ⎟ P11/ 2 b47 / 2 (2-123)
⎝ d0 ⎠
happens both from a theoretical and an ex-
perimental point of view. It is likely that at with a constant prefactor sb ≈ 5 and with
zero anisotropy e = 0, the cells will tend to Péclet number P as used before in Eq. (2-
split if the cell spacing becomes much 68). The scaling relation, Eq. (2-123), con-
larger than the diffusion length, which af- sists of several non-trivial power laws;
fords the possibility for chaotic dynamics only the one with P = R/l relating tip radius
at high speeds. However, this is still specu- to velocity has been confirmed (Classen
lative. et al., 1991). With regard to the general
Let us take a quick look at the kinetic co- agreement between analytical and numeri-
efficient b in Eq. (2-111). As can be con- cal results obtained so far, however, there
cluded from its multiplication by V^ , b be- is little doubt that these scaling results (and
comes more and more important at high others given by Brener and Melnikov,
growth rates. For the free dendritic case 1991) will also hold for the dendritic re-
with kinetic coefficient b and 4-fold an- gion l ⲏl in directional solidification.
A final point to be kept in mind is that Kurz and Fisher (1998) and references
the tip supercooling D in directional solid- therein).
ification is not small, as required by the ap- In a rigorous sense the answer is still
proximations used for the derivation of the negative, but at least arguments can be
scaling relation. On the other hand, P = 1 given for the existence of some boundaries
corresponds to a supercooling as large as on the wavelengths (or cell spacings)
D ≈ 0.75; 0.6 for 2-dim and 3-dim, respec- which can be estimated with the use of sim-
tively, and the scaling relations can be ex- plifications.
pected to hold over a large range of veloc- The situation here shows some similarity
ities, as already indicated from the other to the formation of hydrodynamic periodic
free dendritic case, Fig. 2-12. roll patterns (Newell and Whitehead, 1969;
In summary, these investigations have Kramer et al., 1982; Riecke and Paap,
shown that there appears to be a smooth 1986). In a laterally infinite system, a
transition from cellular to dendritic struc- whole band of parallel rolls is present
tures. The dendritic growth laws are very above the threshold for roll formation, the
well represented by the scaling relations so-called Eckhaus band. This was men-
for the free dendritic case. This scaling tioned in Sec. 2.5.3 for directional solidifi-
should hold in the region cation.
The reason for the stability of these rolls
d0 < R ≈ l < l < l T (2-124)
is that an infinitesimal perturbation is not
where l is the primary cell spacing. It was sufficient to create or annihilate a roll, but
proposed by Karma and Pelce (1989 a, b) a perturbation must exceed a threshold
that the transition from cells to dendrites value before such an adjustment can occur.
could occur via an oscillatory instability, In directional solidification, the situation
for which the present investigations under is different insofar as the envelope over the
quasi-stationary approximation have shown tips of the cells could make a smooth defor-
no evidence. A fully time-dependent calcu- mation of very long wavelength, thereby
lation is possible in principle with Green’s building up enough deformation energy so
function methods (Strain, 1989). that a cell could be created or annihilated at
isolated points. One indication for such a
process is the oscillatory instability of cells
2.5.5 The Selection Problem
postulated by Karma and Pelce (1989) and
of Primary Cell Spacing
Rappel and Brener (1992).
An important question from an engineer- The only hard argument for the selection
ing point of view appears to be the follow- of a unique wavelength comes from an
ing: suppose we know all the material pa- analysis of a spatially modulated thermal
rameters and the experimentally controlla- gradient acting on a cellular pattern of
ble parameters like thermal gradient and small amplitude (normal bifurcation)
pulling speed for a directional solidifica- which imposes a ramp on the pattern
tion process – can we then predict the dis- (Misbah, 1989; Misbah et al., 1990). The
tance between the cells and dendrites? idea originally proposed for the hydrody-
A positive answer to this question is de- namic case (Kramer et al., 1982) is to have
sirable because the mechanical properties a periodically varying thermal gradient
of the resulting alloy are improved with a parallel to the interface, which keeps the
decrease in the primary cell spacing (see interface flat in some regions and allows
annihilation of cells is therefore likely to

be a discrete process.
A natural mechanism for the local reduc-
tion of cell spacings (or creation of a new
cell) is either a nucleation in one of the
grooves (the liquid is supercooled), or even
more likely, the formation of a new cell out
of a side branch in such a groove. Alterna-
Figure 2-30. Numerical study of cellular wave-
length selection at the interface by introducing a tively, tip splitting of a cell may give the
ramp in the thermal gradient field. A high thermal same result (Fisher and Kurz, 1978, 1980).
gradient on the side approximately normal to the The opposite mechanism for the increase
interface keeps the interface flat, the smaller gradient of cell spacing (or annihilation of an exist-
in the center allows for cells to develop. At fixed
ing cell) could occur through the competi-
ramp profile a unique cell spacing is selected in the
center, starting from different initial conditions (Mis- tion of neighboring cells for the diffusion
bah, 1990, unpublished). field, such that one cell finally moves at a
slightly slower speed than the neighboring
cells and, consequently, will be supressed
relative to the position of the moving front.
for the formation of cells in between These two mechanisms have been con-
(Fig. 2-30). For such a specific setup it was jectured by many authors in the past. Some
shown (Misbah et al., 1990) that a unique progress has been made recently by the
wavelength must be selected in the center confirmation of the scaling relations in
of the small-gradient area. The reason for the dendritic region. It seems, therefore,
this special construction is that it allows for worthwhile to reformulate those conjec-
the formation of cells at arbitrarily small tures with the help of these scaling rela-
amplitudes (and therefore small pinning tions. Let us first consider the short-wave-
forces) in the region of strong thermal gra- length l (cell spacing) argument. Assume
dient. that we are at dendritic growth speeds, Eq.
In general, however, the boundary con- (2-124), ignoring here kinetic coefficients.
ditions on the other sides of the cells, due The solidification front then looks like an
to the container walls, are not well speci- array of individual dendrites which only
fied and typically will not provide such a weakly interact with each other through the
ramp structure (see Misbah (1989), how- diffusion field l < l .
ever, for growth in a rotating vessel). For The solidification front z = 0 will be
the time being, we can therefore try to at understood here as a smooth envelope
least find some boundary similar to the touching all the dendrite tips, so that defor-
Eckhaus band for the limits of large and mations of the front have a smallest wave-
small wavelengths in the cell spacing. length l equal to the cell spacing. There are
It is not easy to extend the correspond- now basically two “forces” acting on de-
ing analysis of small-amplitude cells formations ∂z (x, t)/∂t of that front. If some
(Brattkus and Misbah, 1990) to cells with of the tips are trailing a little behind the
deep grooves, as these essentially infinite others, they will be screened through the
grooves present a kind of topological con- diffusion field of the neighboring tips,
straint on the number of longitudinal cells as in the conventional Mullins – Sekerka
in a given lateral interval. The creation or instability, but now without a stabilizing
surface tension interacting between neigh- comes important there. Of course, a num-
boring tips. Taking this into account, the ber of rough approximations were used
destabilizing force is Fd ≈ W̃k(d) z˜k with specifically in the treatment of the destabi-
W̃k(d) = V |k | for a sinusoidal perturbation of lizing force, but this argument should at
amplitude z˜k of a plane interface without least qualitatively capture the competition
surface tension moving at velocity V and mechanism between neighboring den-
wavenumber k. The maximum lies at drites. A more detailed analysis (Warren
k = 2 p/l . The actual area under this pertur- and Langer, 1990) is quite promising at
bation z˜k contained in the solid cells is large velocities in comparison with experi-
smaller by a factor ≈ 2 R/l . We thus arrive ments (Somboonsuk et al., 1984) (see also
at a maximal destabilizing force of Kessler and Levine, 1986 c; Bechhoefer
and Libchaber, 1987).
Fd ≈ W (d) z l ; W (d) ≈ 4 p V R /l2 (2-125) Let us now look at the large-wavelength
corresponding to a depression or enhance- limit l . The initial growth conditions are
ment of every second dendrite. assumed to be just as before, Eq. (2-124),
On the other hand, each of these individ- but now at possibly large cell spacings l . In
ual dendrites knows its operating point, the numerical calculations, it was found
and through the given velocity its super- (Saito et al., 1989) that for fixed cell spac-
cooling at the tip. We approximate this ing l at increasing velocity, a tail instabil-
by the asymptotic form of Eq. (2-56) ity occurs (Fig. 2-31). A side branch in the
D ≈ 1 – 1/P since basically the variation
of D with P enters below, even through P
may not be very large compared to unity.
Capillary effects do not seem to be very
important in this region and are thus ne-
glected here for simplificity. By definition,
D = 1 – z tip /l T , and the two expressions for
D can be evaluated: z tip ≈ l T l /R. The stabi-
lizing force Fs ≈ W (s) z l follows from the
obvious relation W (s) = dVtip /dz tip as
VR
Fs ≈ W ( s ) z l ; W ( s ) ≈ − 2 (2-126)
lT l
As expected, the sign of this “force” is op-
posite to the destabilizing force, Eq. (2-
125). Setting the sum of the two equal to
zero, we expect an instability to occur first Figure 2-31. After a sudden increase in the growth
at cell spacings rate in the dendritic region at fixed cell spacing, tail
instability occurs. One of the side branches near the
tip produces a proturberance in the forward direction,
l ⱗ lT l (2-127) which then becomes a new primary cell (the imposed
for segregation coefficients around one. mirror symmetry is not present in reality of course).
In accordance with the stability of dendritic cells
We cannot say much about small segrega- against tip splitting (compare Fig. 2-29), this tail in-
tion coefficients because the nonlinearity stability appears to be an important selection mecha-
in Eq. (2-114) replacing Eq. (2-56) be- nism for primary cell spacing. See also Fig. 2-43.
groove between two dendrites splits off a between the two critical values and again
ternary branch, which then moves so fast in recovers Eq. (2-1) by noting that l ~ V –1
the strong supersaturation in the groove and therefore V l 2 ≈ constant. The results,
that it finally becomes a new primary Eqs. (2-127) and (2-128), seem to be in
branch. In the plot, the imposed mirror agreement with experiments (Somboonsuk
symmetry is, of course, artificial, but it et al., 1984; Kurowsky, 1990; Kurz and
even acts opposite to this effect, making Fisher, 1981, 1998) concerning the scaling
the process more plausible in reality. In with respect to diffusion length l and ther-
fact, this is also observed experimentally mal length l T . The limit k Æ 0 for the seg-
(Esaka and Kurz, 1985), in particular when regation coefficient as a singular point is
the solid consists of slightly misaligned re- not reliably tractable here.
gions separated by grain boundaries, so The previously given relation l ~ l 1/4
that the growth direction of two neighbor- (Hunt, 1979; Trivedi, 1980; Kurz and
ing dendrites is slightly divergent. Our ba- Fisher, 1981) seems to be valid in an inter-
sic assumption now is that this tail instabil- mediate velocity region (Fig. 2-28) where
ity occurs when the intersection of para- D does not vary significantly, so that
bolic envelopes over neighboring dendrites z tip /lT ≈ 1/2 (see also Sec. 2.6.1).
occurs at a point z ⱗ0, where, theoretically, A serious point is the neglect of surface
D > 1. In this case, there is no need for tension and anisotropy in these derivations.
long-range diffusion around a side branch, In the experiments analyzed so far the rela-
for its dynamics become local. Of course, tion V l 2 ≈ const. seems to hold approxi-
this assumption ignores geometrical com- mately, but what happens for the capillary
petition between neighboring side branches anisotropy e going to zero? Numerically,
to a certain extent, but for the moment tip splitting occurs at lower velocities for
there seems to be no better argument to smaller e . In a system with anistropy e = 0
hand. (and zero kinetic coefficient) the structures
Taking into account the point that neigh- probably show chaotic dynamics at veloc-
boring parabolas with tip position at z tip > 0 ities where the diffusion length l is smaller
cannot intersect further down to the cold than the short wavelength limit of the neu-
side than at z = 0, we obtain z tip = l T l /R tral stability curve (Fig. 2-24), but this is
from the two relations for D, just as in the rather speculative (Kessler and Levine,
previous case of small wavelengths. But 1986 c).
now we must use the parabolic relation In considering whether the tail instabil-
z tip = l2/8 R for the intersection of two pa- ity (large l) or the competition mechanism
rabolas of radius R at z = 0, which are a dis- (small l) will dominate in casting pro-
tance l apart. The tail instability is then ex- cesses, we tend to favor the former. If the
pected to occur for solidification front consists of groups of
dendrites slightly misoriented against each
l ⲏ lT l (2-128)
other due to small-angle grain boundaries,
again with a prefactor roughly of the order cells will disappear at points where the lo-
of unity. In comparison with Eq. (2-127), it cal growth directions are converging and
can be seen that in both cases the same new cells will appear through the tail in-
scaling relation results. The scaling with stability at diverging points at the front.
the inverse growth rate l follows the neu- To summarize, the most likely scaling
tral stability curve at velocities safely in behavior of the primary cell spacing l , de-
pending on pulling velocities, follows Eqs. Mullins – Sekerka (MS) instabilities (also
(2-1) and (2-128) as a consequence of the called limit of constitutional undercool-
arguments presented in this section. This ing), to rates where plane front growth
conclusion is supported by a number of ex- again appears at velocities above Va , the
periments (Billia et al., 1987; Somboonsuk absolute stability limit.
et al., 1984; Kurowsky, 1990; Esaka and Morphological instabilities. The onset
Kurz, 1985), but more work remains to be of plane front interface instability is ob-
done. served to start at defects such as grain
boundaries, subgrain boundaries and dislo-
cations, forming a more or less pronounced
depression at the intersection with the
on Directional Dendritic Growth
solid – liquid boundary, as shown in Fig.
Since 1950, in situ experiments on direc- 2-32 (Fisher and Kurz, 1978). It is inher-
tional solidification (DS) of transparent ently difficult to quantitatively observe the
model systems have been performed (Kof- break-down because the growth rates are
ler, 1950). However, it was some time be- small, and a long period of time is required
fore such experiments were specifically to reach steady state.
conceived to support microstructural mod- The amplitude and wavelength of the
els developed in the 1950s and early 1960s. perturbations as a function of V developing
The work of Jackson and Hunt (1966) is a in systems with small distribution coeffi-
milestone in this respect (see also Hunt et cients k (of the order of 0.1) are shown in
al., 1966). Their experimental approach to Fig. 2-33. In CBr4 – Br2 (de Cheveigne et
dendritic growth has been developed fur- al., 1986) and SCN – ACE (Eshelman and
ther by several groups: Esaka and Kurz Trivedi, 1987), the bifurcation is of a sub-
(1985), Trivedi (1984), Somboonsuk et al. critical type, i.e., there are two critical
(1984), Somboonsuk and Trivedi (1985), growth rates, one for increasing growth
Eshelman et al. (1988), Seetharaman and rate, Vc+ , and another lower value for de-
Trivedi (1988), Seetharaman et al. (1988), creasing growth rate, Vc– . Therefore, a peri-
de Cheveigne et al. (1986), Akamatsu et al. odic interface shape with infinitesimally
(1995), Akamatsu and Faivre (1998), and small amplitudes cannot form in these
others. Substantial progress has been made systems. As has been discussed in Sec.
during these years, especially due to results 2.5.3, only systems with k near unity will
obtained by Faivre and coworkers in very give supercritical (normal) bifurcation
thin (~ 15 µm) cells that constrain the pat- with a single, well-defined critical growth
terns to two dimensions. This research is rate Vc . The evolution of the wavelength
still producing interesting new insights into for two temperature gradients (70 and
the dynamics of interface propagating dur- 120 K/cm) is shown in Fig. 2-33 b. At the
ing crystallization. onset of instability the experimentally de-
The specific interest of DS is that growth termined wavelength is smaller than the
morphologies can be studied not only for critical wavelength by a factor of 2 – 3 com-
dendritic growth but also for cellular and pared with linear stability analysis. In-
plane front growth. We will discuss these creasing the rate above the threshold leads
phenomena in the sequence of their appear- to a decrease in the wavelength proportional
ance when the growth rate is increased to V – 0.4 (de Cheveigne et al., 1986; Kurow-
from Vc , the limit of first formation of sky, 1990). Once stability has started, the
j J
(a) (b)
Figure 2-32. Morphological instabilities of a planar solid/liquid interface as seen on an inclined solidification
front between two glass plates (arrows). Photograph (a) was taken at an earlier stage than (b). The beginning of
the breakdown at defects such as dislocations, subgrain boundaries, or grain boundaries intersecting with the
solid/liquid interface is evident. The widths of the photographs correspond to 100 µm.
(a) (b)
Figure 2-33. Amplitude, A, and wavelength, l, of a periodic deformation of the solid/liquid interface of
CBr4 – Br2 solution versus pulling speed (de Cheveigne et al., 1986). Diagram (a) shows the hysteresis between
appearance and disappearance of perturbations (typical for an inverse bifurcation) for a temperature gradient of
12 K/mm. The lines in diagram (b) represent the calculated neutral stability curves for the two gradients indi-
cated. Open circles are experimental results for 7 K/mm, and crosses, for 12 K/mm. See also Fig. 2-21.
structure evolves to a steady-state cellular or Cells and dendrites. Losert et al. (1998 a)
dendritic growth mode. Which one of these developed an interesting technique in
structures will finally prevail is a question of which a spatially periodic UV laser pulse is
the growth conditions. directed onto the solidification front of the
transparent succinonitrile – coumarin sys- stabilities, cells, dendrites) appear to have

tem. These experiments allowed for a their own wavelength or array spacing.
systematic investigation of the dynamic se- Owing to competition between neighboring
lection and stability of cellular structures. crystals, the mean spacing of large ampli-
Through an increase in V (or C• , or a de- tude cells seems to be always larger than
crease in GT ), a columnar dendritic struc- that of the initial perturbations of the plane
ture can be formed out of a cellular array front, and the mean trunk distance (primary
(Fig. 2-34). All three morphologies (in- spacing l ) of the dendrites, larger than the
Figure 2-34. Time evolution of the solid/liquid interface morphology when accelerating the growth rate from
0 to 3.4 µm/s at a temperature gradient of 6.7 K/mm. Magnification 41 ¥. (a) 50 s, (b) 55 s, (c) 65 s, (d) 80 s,
(e) 135 s, (f) 740 s (Trivedi and Somboonsuk, 1984).
spacing between smooth cells. The reason

for this change in typical spacing is not yet
clear. We will come back to this subject
later. Before we do so, some relevant obser-
vations on tip growth need to be discussed.
The tip is the “head” of the dendrite
where most of the structural features are in-
itiated. Fig. 2-35 shows a dendrite tip of
succinonitrile (SCN) with 1.3 wt.% ace-
tone (ACE) in an imposed temperature gra-
dient, GT = 16 K/cm and a growth rate,
V = 8.3 µm/s. The smooth tip of initially
parabolic shape (Fig. 2-35 b) is clearly vis-
ible. In contrast of free thermal dendrites
(Fig. 2-9), in DS of alloys the secondary
instabilities start forming much closer to
the tip. The imposed temperature gradient
also widens the dendrite along the shaft
relative to a parabola fitted to the tip. This
effect increases with an increasing temper-
ature gradient (Esaka, 1986).
(a)
Figure 2-35. Dendrite tip of

SCN – 1.3 wt.% acetone so-
lution in directional growth.
(a) For V = 8.3 µm/s and
G = 1.6 K/mm; (b) parabola
fitting the tip growing
at V = 33 µm/s and G =
(b) 4.4 K/mm (Esaka and Kurz,
1985; Esaka, 1986).
The sequence of steady-state growth These observations are summarized in

morphologies, from well-developed large Fig. 2-37. Three areas of growth can be dif-
amplitude cells to well-developed den- ferentiated for this alloy:
drites, is shown in Fig. 2-36. Besides the
i) at low speeds, cells are found showing
information on the form and size of the cor-
no side branches and a non-parabolic tip;
responding growth morphologies, this fig-
ii) at intermediate rates (over a factor 5 in
ure also contains indications specifying the
V ), cellular dendrites are formed with weakly
diffusion length l = 2 D/V and the ratio of
developed secondary arms, and they show
the half spacing over tip radius. The char-
an increasingly sharpened parabolic tip;
acteristic diffusion distance decreases more
iii) at large rates dendritic arrays grow
rapidly than the primary spacing of the den-
with well-developed side branches and a
drites (Fig. 2-36). When l Û l the ratio l /2 R
tip size much smaller than the spacing.
of directionally solidified SCN–1 wt.% ACE
alloys is between 5.5 and 6, in agreement It is difficult at present to judge the influ-
with numerical calculations (Sec. 2.5.3). ence of the width of the gap of the experi-
Figure 2-36. Cellular and den-

dritic growth morphologies in
SCN – 1.3 wt.% acetone; thermal
gradient 8 K/mm < G < 10.5 K/mm
(Esaka, 1986). The growth rate V
in µm/s, the diffusion length
(2 D/V ) in mm, and the ratio of the
primary trunk spacing to the tip
diameter are as follows: A = 1.6,
1.6, 2.0; B = 2.5, 1.0, 2.5; C = 8.3,
0.3, 5.5; D = 16, 0.16, 6.0; E = 33,
0.08, 7.5; F = 83, 0.03, 9.0.
Figure 2-37. Primary trunk spacing, diffusion

length and tip radius of SCN – 1.3 wt.% ace-
tone dendrites as a function of the growth rate
for a temperature gradient of 9.7 K/mm. The
various points correspond to the conditions
and microstructures given in Fig. 2-36 (Esaka,
1986). In region A no side arms are observed
and the tips are of non-parabolic shape; in B
the tip becomes parabolic and some side arms
appear; in C well-developed side arms and a
parabolic tip are the sign of isolated tips.
mental cell, which had approximately the et al., 1995). With the ·100Ò direction
same size as the diffusion distance, when oriented along the heat flow direction,
l = l. The corresponding growth rate also characteristic cellular/dendritic arrays such
marks the transition from two- to three-di- as shown in Fig. 2-36 are obtained. With
mensional growth of dendrites, as can be other orientations, where growth is ren-
seen in Fig. 2-36. Under conditions A to C, dered “effectively isotropic” a “seaweed”
no secondary arms are observed perpendic- structure is observed, the tips of which
ular to the plane of observation, while conform what has been called on the basis of
ditions D to F show well-developed 3 D theoretical predictions a doublon (Ihle and
dendrites even far behind the tip. There- Müller-Krumbhaar, 1994).
fore, the gap might somewhat influence the The formation of doublons has also been
values of the transition rates but not the suggested to play an important role in the
qualitative behavior of the transition. As formation of “feathery” grains in technical
mentioned above, the theory (in quasi-sta- aluminum alloys. Recent detailed observa-
tionary approximation) only indicates a tions with electron back-scattered diffrac-
very gradual change in morphology. It is tion (EBSD) combined with optical and
interesting to compare Fig. 2-36 with Figs. scanning electron microscopy (Henry et
2-25 and 2-26. It can be seen that both the al., 1998) have clearly shown that feathery
theoretical and the experimental ap- grains are made of ·110Ò columnar den-
proaches show qualitatively the same be- drites, whose primary trunks are aligned
havior, even if the material constants used along and split in their center by a (111)
are not the same. coherent twin plane. The impingement of
The critical role of crystalline anisotropy secondary ·110Ò side arms gives rise to in-
in interface dynamics has been demon- coherent wavy twin boundaries. The switch
strated experimentally in directional growth from ·100Ò to ·110Ò growth direction was
of transparent CBr4 – C2Cl6 alloys grown as attributed to the small anisotropy of the
single crystals in thin samples (Akamatsu solid – liquid interfacial energy of alumi-
num, which can be changed by the addition The ratio l 2 /R obtained by Esaka and
of solute (Henry et al., 1998). Kurz (1985) for SCN – 1.3 wt.% ACE is in-
The scaling of the initial side-branch dependent of growth rate according to the
spacing l 2 with respect to the tip radius is precision of the measurements and takes a
shown in Fig. 2-38. Both quantities scale value of 2.1 ± 0.2 (see Fig. 2-39 and Table
closely with L 2 V = const. or L 2 C• = const. 2-1). This is in good agreement with the
(where L ~ R or l 2 ). Fig. 2-38 b also shows measurement of Somboonsuk and Trivedi
the quantity l p , the distance from the tip (1985) who found a value of 2.0 on the
down the shaft, where the first signs of tip same system over a wide range of growth
perturbations in the SCN – ACE system can rates, compositions, and temperature gra-
be observed. Here l p is of the same order dients (Trivedi and Somboonsuk, 1984).
as l 2 . This ratio increases with crystal anisotropy
(a)
Figure 2-38. Characteristic dimensions

(tip radius, R, initial secondary arm spac-
ing, l 2 , and distance of appearance of
secondary instabilities, l p ) as a function
of growth rate (a), and as a function of
(b) the alloy concentration (b) (Esaka, 1986).
Figure 2-39. Ratio of initial arm spac-

ing to tip radius as a function of growth
rate (Esaka and Kurz, 1985).
in agreement with numerical calculations the cellular growth region is characterized

(Figs. 2-26, 2-29, 2-13) and decreases with by a decrease in undercooling, while to-
increasing temperature gradient. On the ward the dendritic region, the undercooling
other hand, its value is larger in the case of increases again (Fig. 2-28).
free dendritic growth (Table 2-1). One of the characteristics of directional
The tip undercooling of the dendrite is a solidification of cells/dendrites is the for-
measure of the driving force necessary for mation of array structures with a primary
its growth at the imposed rate V. Fig. 2-40 trunk spacing l . For cellular and cellu-
shows the variation of the tip temperature lar – dendritic structures the growth direc-
with V, the undercooling being defined by tion/crystal orientation relationship plays
the difference between Tm and the tip tem- an important role in the establishment of
perature. During an increase in growth rate, stable or unstable array patterns with a
Figure 2-40. Tip temperatures of

the dendrites of Fig. 2-36 as a
function of growth rate (Esaka and
Kurz, 1985). See also Fig. 2-28.
characteristic spacing, as has been shown

by Akamatsu and Faivre (1998). The
growth-rate dependence of dendritic pat-
terns has been shown in the log – log plot of
Fig. 2-37. We can see that the slope of l vs.
V is initially lower than the slope of R vs.
V. If we take the three points around
V (l = l ), where the tip temperature is ap-
proximately constant, the slope is about
0.25, while a mean slope through all mea-
surements shown gives 0.4. This is fully
consistent with data by Kurowsky (1990).
Furthermore, it is in agreement with our ar-
guments in Sec. 2.5.5.
Most of the measurements give a rate ex-
ponent somewhere between the two limit-
ing values, 0.25 – 0.5. Taking the lower Figure 2-41. Normalized primary trunk spacing, l,
value expressed by temperature gradient as a function of growth rate for various alloy systems
(Somboonsuk et al., 1984). Note that the variation
and concentration gives a relationship
with growth rate at these high velocities is in agree-
l ~ GT– 0.5 C•0.25 (Hunt, 1979; Kurz and ment with the discussion in Sec. 2.5.5.
Fisher, 1981; Trivedi, 1984). Therefore, a
normalized spacing l4 G T2 V /k DT0 is plot-
ted vs. V in Fig. 2-41, showing that differ-
ent materials behave similarly except for a
constant factor (note that the solidus – liq-
uidus interval of an alloy, DT0 , is propor-
tional to C• ).
Primary spacings, however, are not
uniquely defined but form a rather wide
distribution. This is shown in Fig. 2-42 for
one superalloy which was directionally so-
lidified under different conditions. This be-
havior can be understood by examining the
mechanism of wavelength (spacing) reduc-
tion through tail instability, which is a
complicated process. A series of competi-
tive processes between secondary and
tertiary arms in a region behind the tip (of
the order of one primary spacing) finally
causes one tertiary branch to grow through
and to become a new primary trunk (Esaka Figure 2-42. Distribution of nearest dendrite – den-
drite separations (primary trunk spacings) measured
et al., 1988), as shown in Figs. 2-43 and
on a transverse section of a directionally solidified
2-31. The range of stability of dendritic Ni-base superalloy (Quested and McLean, 1984).
arrays has been analysed by Losert et al. Solid lines for 16.7 µm/s and broken lines for
(1998 b). Hunt and Lu (1996) approached 83.3 µm/s.
Fig. 2-7, the interface may be driven to ve-

locities of several m/s. Under such high
rates, the structures become extremely
fine. Primary spacings as small as 20 nm
have been measured in Al – Fe alloys (Gre-
maud et al., 1990). Fig. 2-44 represents
measured primary spacings (black squares),
measured secondary spacings (open
squares), and the calculated tip radius of
the dendrites, using Ivantsov’s solution
and the solvability-scaling criterion (Kurz
et al., 1986, 1988). l and l 2 vary as V – 0.5,
Figure 2-43. Mechanism for the formation of a new as does the tip radius when the Péclet num-
primary trunk by repeated branching of side arms at a ber of the tip is not too large. At large
grain boundary (gb). The increasing spacing at the gb Péclet numbers (or small diffusion dis-
allows a ternary branch to develop and to compete in tances), capillary forces become dominant,
its growth with other secondary branches (hatched
which is the reason for the limit of absolute
arms) (Esaka et al., 1988).
stability, Va . Ludwig and Kurz (1996 a) de-
termined cell spacing and amplitude close
to absolute stability in succinonitrile –
this problem by numerical modeling using
argon alloys and found for both l 3/2 V =
a minimum undercooling criterion for
const.
the selection of the lower limit of array
Since the classical paper by Mullins and
growth.
Sekerka (1964), it has been known that
plane front growth should also be observed
Effects at high growth rates
at interface rates where the diffusion length
As the final topic in this section, we dis- reaches the same order as the capillary
cuss some interesting effects which have length. This critical rate, called the limit of
been observed at very high growth rates. absolute stability, can be calculated from
In laser experiments of the type shown in linear stability analysis (for temperature
Figure 2-44. Experimentally determined

primary spacings (black squares), secon-
dary arm spacings (open squares) and
calculated tip radius (line) for Al – 4 wt.%
Fe alloy rapidly solidified by laser treat-
ment. The minimum in tip temperature
or the maximum in undercooling is due
to the decreasing curvature undercooling
when the dendrite approaches the limit
of absolute stability, Va (Gremaud et al.,
1990).
gradients which are not too high) approxi-

mately as
Va = DT0 D /k G
Here G is the ratio of solid – liquid interface
energy to specific melting entropy. Typical
limits are of the order of m/s. The precise
value depends also on the effect of a vary-
ing solute diffusion coefficient (due to the
large undercoolings), on the variations of k
and DT0 with the growth rate associated
with the loss of local equilibrium, and, fi-
nally, on interface kinetics, which cannot
be ignored. According to an extensive
study of several alloy systems, it may be
stated that at Va we generally do not ob-
serve a simple plane front growth but
rather an oscillating interface, which pro-
duces bands of plane front and cellular
dendritic morphology. Fig. 2-45 a shows
Figure 2-45. Transmission electron micrographs of
such a transition from columnar dendritic
the banded structure in laser-remelted alloys. (a)
grains to bands (Carrard et al., 1992). General view of the abrupt transition from columnar
Fig. 2-45 b shows more details of the eutectic grains (lower part) to a banded structure
banded structure. It is visible that the dark (upper part) in a eutectic Al – 33 wt.% Cu alloy with
band is cellular/dendritic and the clear a solidification rate, V = 0.5 m/s. (b) Enlarged view
of the dark and light bands in a cellular dendritic
band structure-less (supersaturated plane
Al – 4 wt.% Fe alloy with V = 0.7 m/s (Carrard et al.,
front growth). The possibility of chaotic 1992).
interface motion was shown by Misbah et
al. (1991), and oscillatory motion of a
plane interface with time was suggested by to work under heat flow limited conditions
Corriell and Sekerka (1983) and Temkin allowing a growth rate of not more than
(1990). The full theoretical analysis of the some mm/s. In order to reach absolute
banding problem has been given by Karma stability at these low rates the composition
and Sarkissian (1993), see also Kurz and (and DT0 ) had to be kept very small (typi-
Trivedi (1996). Growth rates much higher cally below 100 mK).
than Va are needed in order to definitely
produce an absolutely stable plane solid –
2.5.7 Extensions
liquid interface (Fig. 2-46).
Under steady-state growth conditions In this section, we summarize a few im-
using a transparent organic system, Ludwig provements on the theory of solidification
and Kurz (1996 b) observed the onset of ab- regarding various effects that are important
solute stability but no sign of banding. The in the practical experimental situation.
latter was due to the fact that the experi- They are observed when experimental pa-
ments had to be undertaken with highly di- rameters are outside the range of the simple
luted alloys. This was necessary as one had models considered in this chapter, or when
smallest possible number of experimen-

tal parameters. Despite this simplification,
however, the results appear to be of general
importance.
The discussion of free dendritic growth
has been restricted so far to small Péclet
numbers or low supercoolings. In direc-
tional solidification, however, we are usu-
ally at moderate or even high Péclet num-
bers (Ben Amar, 1990; Brener and Melni-
kov, 1990). It was shown by Brener and
Melnikov (1990) that in this case a devia-
tion from dendritic scaling occurs if P e 1/2
increases approximately beyond unity.
At high growth rates, furthermore, kinetic
coefficients can no longer be ignored at
the interface. A proposed scaling relation
(Brener and Melnikov, 1991) was con-
firmed numerically (Classen et al., 1991),
and combined anisotropy of surface ten-
sion and kinetic coefficient was treated an-
alytically (Brener and Levine, 1991; Ben-
Jacob and Garik, 1990). A general treat-
ment of kinetic coefficients on interface
stability was given by Caroli et al. (1988).
For eutectic growth, this was formulated by
Geilikman and Temkin (1984). For higher
anisotropies than considered so far, we en-
counter facets on the growing crystals.
This was analyzed for single dendrites
Figure 2-46. Experimental velocities at the transi- (Adda Bedia and Ben Amar, 1991; Maurer
tions from dendrites to bands (squares) and from
et al., 1988; Raz et al., 1989; Yokoyama
bands to microsegregation-free structures (circles)
for the (a) Al – Fe, (b) Al – Cu and (c) Ag – Cu sys- and Kuroda, 1988) and for directional so-
tems. The triangles indicate experimentally deter- lidification (Bowley et al., 1989).
mined transitions from dendrites to microsegregation- In our treatment of directional solidifica-
free structure, and the crosses the values of the abso- tion, only one diffusion field was treated ex-
lute stability limit. The full and dashed lines are the
plicitly, namely the compositional diffusion.
theoretical predictions of the model for dilute and con-
centrated solutions, respectively (Carrard et al., 1992). If a simple material grows dendritically
(thermal diffusion), small amounts of impur-
ities may become a matter of concern. This
additional factors influence the growth of was reconsidered by Ben Amar and Pelce
the solid. (1989), confirming the previous conclusion
The models in this chapter are minimal (Karma and Langer, 1984; Karma and Kot-
in the sense that they were intended to cap- liar, 1985; Lipton et al., 1987) that impur-
ture the essence of a phenomenon with the ities may increase the dendritic growth rate.
A subject of appreciable practical impor- al. (1988). Other sources of convective in-
tance concerns the late stages of growth, stability (Corriell et al., 1980; Sahm and
where coarsening of the side branch struc- Keller, 1991) cannot be discussed here
tures occurs together with segregation in any detail, as the literature is too exten-
(Kurz and Fisher, 1998). If elastic forces sive in order to be covered within this
are not of primary importance, it is now chapter.
generally accepted (see Chapter 6, Sec.
6.4.1) that a relation L* = A + B t 1/3 (Huse,
Phase-field models
1986) is a good representation of typical
length scales L* varying with time t during In recent years, the phase-field metho-
diffusional coarsening processes. This con- dology has achieved considerable impor-
firms the classical Lifshitz – Slyozov – tance in modeling and numerically simulat-
Wagner theory (Lifshitz and Slyozov, ing a range of phase transitions and com-
1961; Wagner, 1961). The result, however, plex growth structures that occur during
is not specific about the geometric details, solidification. Phase-field models have
for example, in directional solidification, been applied to a wide range of materials
and it also assumes constant temperature. such as pure melts, simple binary alloys,
We have only briefly mentioned the tran- eutectic and peretectic phase transitions
sition from dendritic arrays back to a plane and grain growth structures in situations as
front during directional solidification at diverse as dendritic growth and rapid solid-
very high speeds. The importance of ki- ification.
netic coefficients was demonstrated by In the phase-field formulation a mathe-
Brener and Temkin (1989), and recently, matically sharp solid – liquid interface is
the possibility of chaotic dynamics in this smeared out or regularized and treated as
region similar to those described in the Ku- a boundary layer, with its own equation
ramoto – Sivashinsky equation was sug- of motion. The resulting formulation no
gested by Misbah et al. (1991). longer requires front tracking and the im-
A richness of dynamic phenomena was position of boundary conditions, but must
obtained in a stability analysis of eutectics be related to the sharp interface model by
(Datye et al., 1981). The possibility of an asymptotic analysis. In fact, there are
tilted lamellar arrays in eutectics was dem- many ways to prescribe a smoothing and
onstrated by Caroli et al. (1990) and, simi- dynamics of the sharp interface consistent
larly, for directional solidification at high with the original sharp-interface model. So
speeds by Levine and Rappel (1990). This there is no unique phase-field model, but
was observed in experiments on nematic rather a family of related models. The first
liquid crystals (Bechhoefer et al., 1989) phase-field model was developed by Lan-
and in eutectics (Faivre et al., 1989). ger (1986) in ad hoc manner. Subsequently,
Finally, the density difference between models have been placed on a more secure
liquid and solid should have some marked basis by deriving them within the frame-
influence on the growth mode (Caroli work of irreversible thermodynamics (Pen-
et al., 1984, 1989). For dendritic growth, rose and Fife, 1990; Wang et al., 1993).
forced flow was treated in some detail by Caginalp (1989) presented for the first time
Ben Amar et al. (1988), Ben Amar and Po- the relation between phase-field models
meau (1992), Bouissou and Pelce (1989), and sharp interface models or free boun-
Bouissou et al. (1989, 1990) and Rabaud et dary problems by considering the sharp
interface limit in which the interface thick- Elder et al., 1994; Wheeler et al., 1996) and
ness is allowed to tend to zero. peritectic reactions (Lo et al., 2000).
However, the technique also has some
disadvantages. The first is related to the ef-
fective thickness of the diffuse interface of 2.6 Eutectic Growth
alloys (1– 5 nm) which is three to four or-
ders of magnitude smaller than the typical 2.6.1 Basic Concepts
length scale of the microstructure. Since
the interface must spread over several Eutectic growth is a mode of solidifica-
points of the mesh, this limits considerably tion for a two-component system. Operat-
the size of the simulation domain. The sec- ing near a specific point in the phase dia-
ond problem arises from the attachment ki- gram, it shows some unique features (Kurz
netics term, which plays a significant role and Fisher, 1998; Lesoult, 1980; Hunt and
in the phase-field equation, unlike micro- Jackson, 1966; Elliot, 1983).
structure formation of metallic alloys at The crucial point in eutectic growth is
low undercooling. These two factors have that the solidifying two-component liquid
so far limited phase-field simulations of at a concentration near CE (Fig. 2-19) can
alloy solidification with realistic solid- split into two different solid phases. The
state diffusivities to relatively large super- first phase consists of a high concentration
saturations. Recent mathematical and com- of A atoms and a low concentration of B at-
putational advances, however, are rapidly oms; the second solid phase has the oppo-
changing this picture. Some of the recent site concentrations. These two phases ap-
advances include: 1) a reformulated asymp- pear alternatively as lamellae or as fibers of
totic analysis of the phase-field model one phase in a matrix of the other phase.
for pure melts (Karma and Rappel, 1996, One condition for the appearance of a
1998) that has (i) lowered the range of ac- eutectic alloy is apparently a phase dia-
cessible undercooling by permitting more gram (as sketched in Fig. 2-19) with a tem-
efficient computations with a larger width perature TE . The two-phase regions meet at
of the diffuse interface region (as com- (TE , CE ), and the two liquidus lines inter-
pared to the capillary length), and (ii) made sect before they continue to exist as meta-
it possible to choose the model parameters stable liquidus lines (dash-dotted line) at
so as to make the interface kinetics vanish; temperatures somewhat below TE . This is a
2) a method to compensate for the grid an- material property of the alloy (Pb – Sn for
isotropy (Bösch et al., 1995); 3) an adaptive example). The other condition is the ex-
finite element method formulation that re- perimental starting condition for the con-
fines the zone near the diffuse interface and centration in the high-temperature liquid.
that has been used in conjunction with the Assume that we are moving a container
reformulated asymptotics to simulate 2D filled with liquid at concentration C• in a
dendritic growth at low undercooling (Pro- thermal gradient field where the solid is
vatas et al., 1998); 4) the implementation cold, the liquid is hot, and the solidification
of the method for fluid flow effects (Diep- front is proceeding towards the liquid in
ers et al., 1997; Tonhardt and Amberg, the positive z-direction. As a stationary so-
1998; Tong et al., 1998), and 5) the extension lution, we find a similar condition to the
of the technique to other solidification phe- simple directional solidification case; that
–
nomena including eutectic (Karma, 1994; is the concentration in the solid CS aver-
2.6 Eutectic Growth 153
aged across the front must be equal to C• to Diffusion of excess material not incorpo-
maintain global mass conservation together rated into one of the lamellae does not have
with a stationary concentration profile near to continue up to infinity but only to the
the front. neighboring lamella, which has the oppo-
Assume now that the liquid concentra- site composition relative to C• . Assuming
tion at infinity, C• , is close to CE and the again for simplicity a symmetric phase dia-
temperature at the interface is at TE . The gram, we may write this flux balance as
solid may split into two spatially alternat-
DC˜
ing phases now on the equilibrium (bino- V CE (1 − k ) = D (2-129)
dal) solidus lines C̃S (TI ), one located near (l / 2)
C Æ 0, and one located near C Æ1 at where the left-hand side describes the flux
TI < TE . To see what happens, let us look J of material to be carried away from each
at the situation with the concentration in lamellar interface during growth at veloc-
the liquid C• being precisely at eutectic ity V, which then must be equal to the
composition C• = CE . At TI < TE , there may diffusion current (concentration gradient)
now be alternating lamellae formed of over a distance l /2 to the neighboring la-
solid concentration C̃Sa and C̃Sb (Fig. 2-47), mella (across the liquid, as we ignore diffu-
the corresponding metastable liquidus con- sion in the solid). DC̃ = C̃L(a) – C̃L(b) is the
centrations being at C̃La and C̃Lb . For both concentration difference in the liquid at the
solid phases now the liquid concentration interfaces of the lamellae, and k < 1 is the
C• = CE is in the metastable two-phase segregation coefficient.
regions at TI : C̃Sa < C• < C̃La , C̃Lb < C• < C̃Sb . With the help of the liquidus slopes
In principle, we have to consider the pos- dT/dC̃L as material parameters, we can ex-
sibility of metastable solid phases (Tem- press the undercooling DTI = TE – TI as
kin, 1985), which we ignore here for sim-
plicity. dT
DTI = DC˜ 2 (2-130)
dC˜ L
(symmetrical phase diagram assumed

about CE) and thus
dT l
DTI = 2 CE (1 − k ) (2-131)
dC˜ L l
Again, l = 2 D/V is the diffusion length.

(a) (b) So far we have ignored the singular
points on the interface where the liquid and
Figure 2-47. Sketch of symmetrical eutectic phase
diagram (a) and eutectically growing lamellar cells two lamellae a and b meet (Fig. 2-47). At
(b), where a is the solid phase with concentration this triple point in real space (or three-
C̃Sa, and b the solid phase with concentration C̃Sb. The phase junction), the condition of mechani-
average temperature at the interface is around TI be- cal equilibrium requires that the surface
low the eutectic temperature TE . The liquid at the
tension forces exerted from the three inter-
interface is at metastable concentrations (dashed
lines in (a)). The solid – liquid interfaces in (b) are faces separating a, b and the liquid cancel
curved and they meet with a – b interfaces at three- to zero. In the simplest version, this condi-
phase coexistence points. tion defines some angles Ja , J b of the two
solid – liquid interfaces relative to a flat imum undercooling would serve as the op-
interface. Accordingly, the growth front erating point of the system with spacing
will have local curvature. What is worse, given by Eq. (2-134).
there is no guarantee that the a – b solid – This stationary calculation was extended
solid interfaces are really parallel to the by Datye and Langer (1981) to a dynamic
growth direction. Making a symmetry as- stability analysis, where the solid – solid –
sumption for simplicity again, we can liquid triple points could move parallel and
ignore this problem for the moment. The perpendicular to the local direction of
curvature K of the growth front at each la- growth, coupled however to the normal di-
mella is then proportional to l–1. This cur- rection of the local orientation of the front.
vature requires, through the Gibbs – Thom- It was found that the marginal stability co-
son relation, Eq. (2-102), another reduc- incided exactly with the point of minimum
tion, DT K , of the interface temperature be- undercooling.
low T I : The basic model equations for eutectic
growth in a thermal gradient field can be
DT K = TE d K (2-132)
written in scaled form as follows (Brattkus
where d is an effective capillary length, de- et al., 1990; Caroli et al., 1990). Neglecting
pending on surface tensions, with K now diffusion in the solid (one-sided model)
taken as K ~ l–1. and assuming a single diffusion coefficient
The total reduction of temperature below D for solute diffusion near eutectic concen-
TE during eutectic growth can thus be writ- tration CE (Fig. 2-47), we assume a sym-
ten as (2-133) metric phase diagram about CE for simplic-
⎧ l d⎫ ity.
DT = DTI + DTK = TE ⎨aI + aK ⎬ We define a relative (local) concentra-
⎩ l l⎭
tion gap as
with dimensionless constants aI and aK .
Plotting this supercooling as a function DCa
d= (2-135)
of lamellar spacing, we find a minimal DC
supercooling ∂DT /∂l = 0 at
with DCa = CE – C̃Sa (T ), DCb = C̃Sb(T ) – CE
l = aK / aI ld (2-134) and DC = DCa + DCb .
The dimensionsless composition is then
Again, this is the fundamental scaling rela-
defined as
tion conjectured in Eq. (2-1) and encoun-
tered in various places in this chapter, C − CE
u= (2-136)
where the origin is a competition between DC ( TE )
driving force and stabilizing forces, most
intuitively expressed in Eq. (2-133). and (in contrast to the previous definitions)
A more thorough theoretical analysis of we express lengths and times in units of a
eutectic growth was given in the seminal diffusion length l˜ and time t̃
paper by Jackson and Hunt (1966), which l˜ = D/V , t˜ = D/V 2 (2-137)
is still a standard reference today. One ba-
sic approximation in this paper was to av- (Note that l˜ here differs by a factor of 2
erage the boundary conditions on flux and from previous definitions.) Restricting our
temperature over the interface. This led to attention to two dimensions corresponding
Eq. (2-133), and it was argued that the min- to lamellar structures, we now define the
diffusion equation as and are equivalent for the b phase, with

ma = |dT /dC̃La | as the absolute liquidus
∂u ∂ slope, GT as the fixed thermal gradient and
= ∇2 u + u (2-138)
∂t ∂z L a as the specific latent heat. At infinity,
z Æ •, the boundary condition to Eq. (2-
The conservation law at the interface z =
138) is u• , depending on the initial concen-
z (x, t) is
tration in the liquid. This then forms a
·
– n̂ · ⵜu = D (x, t) (1 + z ) nz (2-139) closed set of equations.
Ahead of the eutectic front, the diffusion
with the unit vector n̂ normal to the inter- field can be thought of as containing two
face and
(2-140) ingredients: a diffusion layer of thickness l˜
associated with global solute rejection and
⎧d for x in a / l -front regions
D( x , t ) = ⎨ modulations due to the periodic structure
⎩d − 1 for x in b / l -front regions of the solid of the extent l (l Ⰶl˜ ). When
the amplitude of the front deformations is
The Gibbs – Thomson relation for the
small compared to these lengths, the aver-
boundary becomes
(2-141) aging approximation by Jackson and Hunt
(1966) (and also by Datye and Langer,
⎧ z
⎪⎪− da K − l in a / l -regions 1981) seems justified.
Ta
u (z , t ) = ⎨ This point was taken up by Caroli et al.
z
⎪− db K − in b / l -regions (1990), who found that the approximation
⎪⎩ lT b is safe only in the limit of large thermal
gradients
(kinetic coefficients were considered by
Geilikman and Temkin (1984). Finally, and GT Ⰷ V m (DC)/D. (2-145)
this is a new condition in comparison with This approximation appears difficult to
the simple directional solidification, the reach experimentally, though.
triple point where three interfaces meet In an attempt to shed some light on
should be in mechanical equilibrium wavelength selection, Datye et al. (1981) and
g aL + g bL + g ab = 0 (2-142) Langer (1980 b) considered finite ampli-
tude perturbations of the local wavelength
with the surface tension vectors g oriented (Fig. 2-48). This type of approach was used
so that each vector points out of the triple in a somewhat refined version by Brener et
point and is tangent to the corresponding al. (1987). They derived an approximate
interface. As before, K is the local curva- potential function for wavelengths l and
ture of the interface, being positive for a argued that under finite amplitude of noise,
solid bulging into the liquid. The dimen- the wavelength selected on average is de-
sionless capillary and thermal lengths asso- fined by a balance in the creation rates and
ciated with the a-phase are the annihilation rates of lamellae. In other
words, if lamellae disappear through sup-
1 g aL TE pression by neighboring lamellae and ap-
da = (2-143)
l˜ ma ( DC ) La pear through nucleation, then an equal rate
of these processes leads to a selection of
1 ma ( DC ) –
lT a = (2-144) an average spacing l , because both depend
l˜ GT on l . The operating point was found in a
Figure 2-48. Turbulent behav-

ior of lamellar spacings with
time in a simplified model of
eutectic growth (Datye et al.,
1981).
limited interval near the wavelength cor- system, revealing the similarities in the
responding to minimal supercooling (or structure of solution space between eutec-
maximal velocity in an isothermal process) tic and dendritic growth by showing the
and, accordingly, is described by Eq. (2- existence of a discrete set of solutions
134). for each parameter set (Kassner and Mis-
More recent extensions of the theory bah, 1991 b). Axisymmetric solutions were
(Coullet et al., 1989) gave indications that shown to cease to exist beyond a certain
the orientation of the lamellae (under iso- wavelength by merging in a fold singular-
tropic material parameters) are necessarily ity. For larger wavelengths, extended tilted
parallel to the growth direction of the front states appear. They were characterized in
but may be tilted and travel sideways at detail, both numerically (Kassner and Mis-
some specific angles (Caroli et al., 1990; bah, 1991 c) and analytically (Misbah and
Kassner and Misbah, 1991). Finally, it was Temkin, 1992; Valance et al., 1993). The
found (Kassner and Misbah, 1991 b) that supercritical nature of the tilt bifurcation
the standard model of eutectic solidifica- was demonstrated and a prediction was
tion has an intrinsic scaling structure made as to how extended tilted domains
could be produced rather than the localized
l ~ l d f (l / lT ) (2-146)
structures found so far. This prediction was
with a scaling function f depending only on borne out by subsequent experiments by
l /l T so that l ~ V –1/2 for 2 l ⱗ l T or for high the Faivre group (Faivre and Mergy,
enough velocities, while at lower velocities 1992 a). Amplitude equations were derived
the exponent should be smaller: l ~ V – 0.3. that explained how a supercritical bifurca-
This is in good agreement with the argu- tion could give rise to localized tilted do-
ments given for ordinary directional solid- mains and predicted a dynamical wave-
ification and also explains a lot of experi- length selection mechanism (Caroli et al.,
mental data (Lesoult, 1980). 1992; Kassner and Misbah, 1992 a) as well
Numerical simulations of the full contin- as describing the influence of crystalline
uum model of eutectic growth were subse- anisotropy (Kassner and Misbah, 1992 a).
quently carried out by Kassner and Misbah Again, this was verified in detail experi-
(1990, 1991 a, b, c). They gave a fairly mentally (Faivre and Mergy, 1992 b). A
complete description of the steady-state similarity equation was derived both
structures to be expected in the eutectic for axisymmetric (Kassner and Misbah,
1991 b) and tilted states (Kassner and Mis- objection was countered (Baumann et al.,
bah, 1992 b), giving the corrections to the 1995 b) by pointing out that there is quite a
aforementioned scaling law. Anomalous number of steady states on the high-wave-
asymmetric cells bearing resemblance to length side of this point and that the ques-
asymmetric cells in experiments on dilute tion of irrelevance for pattern formation is
alloys (Jamgotchian et al., 1993) were a question of time-scales. Since these be-
found in the numerical solution of the eu- come slow near the accumulation point, the
tectic growth problem and described ana- relevance of a density of additional states
lytically (Kassner et al., 1993). cannot be easily discarded. It should be
To conclude the discussion of steady kept in mind that tilted lamellar states are
states, it was shown numerically that a also unstable with respect to phase diffu-
number of periodicity-increasing bifurca- sion near the tilt bifurcation (Fauve et al.,
tion exist, greatly enlarging the space of 1991). Nevertheless, they survive experi-
possible stationary solutions (Baumann et mental time-scales, because phase diffu-
al., 1995 a). These results strongly suggest sion is very slow.
that the question of wavelength selection The first dynamic simulation of the con-
should be considered from a different point tinuum model of eutectic growth (in the
of view. It is known that a band of steady- quasi-stationary approximation) appears to
state solutions exists in directional solidifi- have been done by Kassner et al. (1995),
cation beyond the instability threshold and but the only extensive study so far is due to
a similar statement holds for eutectic (even Karma and Sarkissian (1996). This beauti-
though there is no threshold in this case). ful work was done in close collaboration
Moreover, careful experiments (Faivre, with experimentalists and thus validated
1996) have clarified that the observed quantitively. It pinpointed a number of new
wavelength depends on the history of the short-wavelength oscillatory instabilities
sample. Therefore, there is no wavelength and made quantitative predictions for the
selection in the strict mathematical sense. CBr4 – C2Cl6 system allowing these in-
This is at least true in the absence of noise, stabilities to be reproduced in experiments
i.e. for the deterministic equations of mo- (Faivre, 1996; Ginibre et al., 1997). There-
tion. However, the investigations of Bau- fore, a pretty complete picture exists today
mann et al. (1995 a) render it plausible that of the possible stationary and oscillatory
a whole attractor of steady-state solutions patterns in lamellar eutectic growth and
with accumulation points exists. One of their stability range.
these accumulation points is the point of A recent development is the stability
marginal stability. Since the temporal evo- analysis of ternary eutectics by Plapp and
lution of any structure in the vicinity of this Karma (1999), based on an extension of the
attractor will become slow, it seems likely Datye – Langer theory. It shows how the
that observed wavelengths will correspond presence of a third component in the sys-
to points of increased density of this attrac- tem leads to the formation of eutectic colo-
tor. There have been objections (Karma nies; an effective description of the eutectic
and Rappel, 1994) that many of the newly front on length scales much larger than the
found steady-state solutions are on the low- lamellar spacing is provided, which yields
wavelength side of the point of marginal a simple means of calculating an approxi-
stability, hence they are unstable and there- mate stability spectrum.
fore irrelevant for pattern formation. This
2.6.2 Experimental Results DSf values are prone to form atomically

on Eutectic Growth smooth interfaces, which lead to the forma-
tion of macroscopically faceted appear-
Eutectic growth was a subject of much ance.
interest to experimentalists in the late In the case of nf/nf eutectics, volume
1960s and early 1970s. Substantial re- fractions (of one eutectic phase) of less
search has been motivated by the possibil- than 0.3 lead generally to fibers, while
ity of developing new high-temperature at volume fractions between 0.3 and 0.5,
materials. The in situ directional solidifica- lamellar structures prevail. The micro-
tion of two phases of very different proper- structures of nf/nf eutectics (often simple
ties is an interesting method of producing metal/metal systems) are considered regu-
composite materials with exceptional prop- lar and those of nf/f eutectics (mostly the
erties. However, since these materials could above-mentioned casting alloys) are con-
not outperform the more conventional, di- sidered irregular. Fig. 2-49 shows schemat-
rectionally solidified dendritic superalloys ically the morphology of the growth front
in the harsh environment of a gas turbine, in both cases. It can be easily understood
the interest dropped. Therefore, most of the that growth in nf/nf eutectics is much more
research on eutectics was performed before of a steady-state type than it is in f/nf eu-
1980 (for a review, see Kurz and Sahm, tectics.
1975; Elliott, 1983). One exception is the
ongoing research concerning casting of eu- Regular structures
tectic alloys such as cast iron (Fe – C or
Applying a criterion such as growth at
Fe – Fe3C eutectic) and Al – Si.
the extremum to the solution of the capil-
Casting alloys are generally inoculated
and solidify in equiaxed form (free eutectic
growth) (see Flemings, 1991). This fact,
however, does not make any substantial
difference to their growth behavior because
growth is solute diffusion controlled in
nearly all cases owing to the high concen-
tration of the second element. The models
described above therefore apply to both di-
rectional and free solidification.
The different alloys can be classified
into four groups of materials (Kurz and
Fisher, 1998): lamellar or fibrous systems,
and non-faceted/non-faceted (nf/nf) or
non-faceted/faceted (nf/f) systems.
The distinction between f and nf growth
behavior can be made with the aid of the
melting entropy. Small entropy differences
DSf between liquid and solid (typical for
metals and plastic crystals such as SCN, Figure 2-49. Solid/liquid interface morphologies of
PVA, etc.) lead to nf growth with atomi- (a) regular (nf/nf) eutectics and of (b) irregular (nf/f)
cally rough interfaces. Materials with large eutectics during growth.
lary-corrected diffusion equations (Jackson eutectoid systems. (Eutectoids are “eutec-

and Hunt, 1966), Eq. (2-133), we obtain for tics” with the liquid parent phase replaced
nf/nf eutectics the well-known relation- by a solid.) Often a l3 V relationship is
ships (Eq. (2-134)) found in these systems (Fig. 2-50 b) over
l2 V = C some range of the variables (see Eq. (2-
146)).
DT 2 /V = 4 C¢
In general, it may be said that the field of
where C and C¢ are constants. Fig. 2-50 eutectic growth is under-represented in ma-
shows that this behavior has been observed terials research, and many more careful
globally in many eutectic systems, some of studies are needed before a clearer insight
them having been studied over five orders into their growth can be gained. Trivedi
of magnitude in velocity. The situation is and coworkers began such research in the
much less clear when it comes to analyzing early 1990s, and some of their results are
(a)
(b)
Figure 2-50. Lamellar spacings of (a) directionally solidified eutectics and (b) directionally transformed eutec-
toids as a function of growth rate (Kurz and Sahm, 1975).
presented here. Fig. 2-51 indicates that in manent creation and movement of faults
eutectics the spacings are also not at all (see below). This process is three-dimen-
uniquely defined. There is a rather wide sional and cannot be realistically simulated
distribution around a mean value for each in two-dimensional calculations.
rate (Trivedi et al., 1991). The operating In Fig. 2-52, the mean spacings are plot-
range of eutectics is determined by the per- ted as points, and the limits of the distribu-
Figure 2-51. Eutectic spacing distribution curves as a function of velocity for Pb – Pd eutectics (Trivedi et al.,
1991).
Figure 2-52. A comparison of the ex-

perimental results on the interlamellar
spacing variation with velocity for
CBr4 – C2Cl6 with the theoretical values
(solid lines) for two marginally stable
spacings (Seetharaman and Trivedi,
1988).
tion are given by the extension of the Irregular structures

bars. From the calculated range of stability,
The above relationship for eutectic spac-
which was discussed by Jackson and Hunt
ing and undercooling as a function of
(1966), it can be seen that the minimum
growth rate are also useful in the case of ir-
of the experimental values coincides with
regular systems such as the nf/f casting al-
the theoretical prediction (see also Sec.
loys Fe – C or Al – Si (Fig. 5-49 b). Jones
2.6.1). This, however, does not provide
and Kurz (1981) introduced a factor, j ,
definitive proof of this prediction, due to
which is equal to the ratio of the mean
the fact that several physical parameters
spacing, ·l Ò, of the irregular structure to
of the system are not precisely known. On
the spacing at the extremum. This leads to
the other hand, it is clear that eutectic spac-
the following relationships:
ings do not explore the upper range of
stability (catastrophic breakdown), at least ·l Ò2 V = j 2 C
not in nf/nf systems. Some other mecha- ·DT Ò2/V = [j + (1/j )] 2 C ¢
nism limits the spacing at its upper bound.
The adjustment of spacings is a rapid pro-
Faults
cess, and its rate increases when the spac-
ing increases (Seetharaman and Trivedi, Defects in the ideal lamellar or fibrous
1988). structure are an essential ingredient of eu-
Figure 2-53. Eutectic fault struc-

tures in directionally solidified
Al – CuAl2 alloy (Double, 1973).
tectic growth. They allow the two-phase Oscillations

crystal to rotate into crystallographically
Periodic oscillations have been observed
(energetically) favorable orientations (Ho-
as a morphological instability in several
gan et al., 1971) and to adapt its spacing to
systems grown under various conditions
the local growth conditions. In lamellar
(Hunt, 1987; Carpay, 1972; Zimmermann
systems, there are different types of faults
et al., 1990; Gill and Kurz, 1993, 1995).
(Double, 1973): single or extended faults,
These morphological instabilities form in
with or without net mismatch (Fig. 2-53).
off-eutectic alloys even at growth rates
They mostly represent sub-boundaries of a
of several cm/s, as is shown in Fig. 2-55.
eutectic grain and could develop through
polygonization of dislocations which form
because of the stresses created at the inter-
phase boundaries of the composite. Fig. 2-
54 shows distributions of fault spacings, L ,
for different growth rates indicated by dif-
ferent lamellar spacings, l (Riquet and Du-
rand, 1975). In the case of non-faceted fi-
brous structures, the faults are formed by
simple fiber branches or terminations.
(a)
(b)
Figure 2-54. Eutectic fault spacing distribution Figure 2-55. Periodic oscillations in hypoeutectic
curves for Al – CuAl2 directionally solidified with Al – CuAl2 eutectic under rapid laser resolidification
different growth rates and spacing values as indi- conditions: (a) experimental observation and (b) sim-
cated (Riquet and Durand, 1975). ulation (Zimmermann et al., 1990; Karma, 1987).
There is good correspondence of the ob-

served structures with the results of theo-
retical modeling by Datye and Langer
(1981) and by Karma (1987). In recent ex-
periments a complete stability diagram for
the lamellar CBr4 – C2Cl6 eutectic has been
determined by Ginibre et al. (1997) (Fig.
2-56 a). It shows the observed structures
for the variables dimensionless spacing
(L ) and concentration (h ). In the limits
1 < L < 2 the basic state is stable, but over
a decreasing composition range when
L = l /l min is increased. Outside this re-
gime, three different states and combina-
tions of these have been found; tilt mode
(T ), 1 l oscillations (period-preserving os-
(a)
cillatory or optical mode) and 2 l oscilla-
tions (period-doubling oscillatory mode,
Fig. 2-56). These results are in good quan-
titative agreement with numerical results
by Karma and Sarkissian (1996).
As was the case with dendrites, eutectics
also form extremely fine microstructures
when they are subject to rapid solidifica-
tion. A value of l = 15 nm (each phase is
some 20 atoms wide) seems to be the mini-
mum spacing that can be achieved with
growth rates of the order of 0.2 m/s (Zim-
mermann et al., 1989). In highly under-
cooled alloys anomalous eutectic struc-
tures have been found which, as with den- (b)
drites, seem to be the result of the large
capillary forces involved with such fine
scales (Goetzinger et al., 1998). Figure 2-56. Stability diagram of directionally so-
lidified CBr4 – C2Cl6 alloys. (a) Symbols: experimen-
tally observed periodic patterns (Ginibre et al.,
2.6.3 Other Topics 1997). Lam. Term.: lamellar termination. (b) Curves:
numerically calculated diagram (Karma and Sarkis-
Peritectics sian, 1996).
A wide spectrum of microstructures can
be found in peritectic alloys. Recently this
has produced a great deal of interest (Kerr smaller distribution coefficient, more com-
and Kurz, 1996). In directional solidifica- plex microstructures can form in the two-
tion, when the G/V ratio is high enough, phase region. New phases may appear dur-
i.e., in the range of the limit of constitu- ing transient or steady-state growth via nu-
tional undercooling of the phase with the cleation at the solidification front. The ac-
tual microstructure selection process is bines. The interested reader is referred to

thus controlled by nucleation of the phases, Rappaz (1989), Rappaz et al. (2000), Wang
and by growth competition between the nu- et al. (1995), Wang and Beckermann (1996),
cleated grains and the pre-existing phase Beckermann and Wang (1996), Boettinger
under non-steady-state conditions. In this et al. (2000).
case nucleation in the constitutionally
undercooled zone ahead of the growth
front has to be taken into account in order 2.7 Summary and Outlook
to determine the microstructure selection
(Ha and Hunt, 1997; Hunziker et al., 1998). During the recent years, very substantial
Further, new structures have been observed improvements in our understanding of the
which are controlled by convection. More pattern-forming processes in solidification
detailed discussions of this subject can be have been achieved. Although the basic
found in Karma et al. (1998), Park and Tri- model equations have been known for sev-
vedi (1998), Trivedi et al. (1998), Van- eral decades, it is only during recent years
dyoussefi et al. (2000), Mazumder et al. that the mathematical and numerical tools
(2000). were extended to allow for a reliable analy-
sis of the complicated expressions. In addi-
tion, careful experiments have been per-
Phase/microstructure selection
formed, mostly on model substances, which
Phase and microstructure selection is of have provided an impressive amount of
utmost importance to applications of solid- precise, quantitative data. This combined
ification theory as it controls the properties effort has basically solved the problem of
of materials. Nucleation and growth have free dendritic growth with respect to veloc-
to be modeled in order to make predictions ity selection and side branch formation.
of structures as a function of composition In the process of directional solidifica-
and the solidification variables; growth tion, a consistent picture is now emerging,
rate and temperature gradient for direc- relating the growth mode to free dendritic
tional solidification, undercooling for free growth and, at the same time, to viscous
growth. The interested reader is referred to a fingering and growth in a channel. At very
recent overview by Boettinger et al. (2000) high growth rates that approach the limit
where these phenomena are discussed. of absolute stability, the situation is still
somewhat unclear, for non-equilibrium
effects like kinetic coefficients then be-
Micro/macro modeling
come of central importance. These quan-
The last decade has experienced signifi- tities are difficult to determine experimen-
cant progress in numerical modeling of tally. Furthermore, the selection of the pri-
combined macroscopic and microscopic mary spacings of the growing array of cells
solidification phenomena. Nucleation and and dendrites is still subject to discussion.
growth models have been implemented in One such point of contention, for example,
2D and 3D heat flow and fluid flow pro- is the typically observed increase in spac-
grams providing a substantial improvement ing when moving from cellular to dendritic
of our analytical tools for optimized and structures in model substances.
new materials processes such as single In eutectic growth, the situation is even
crystal casting of superalloys for gas tur- less understood, for good reason. The
2.8 References 165
three-phase junctions at the solid – liquid Barber, M. N., Barbieri, A., Langer, J. S. (1987),
Phys. Rev. A 36, 3340.
interface enter as additional conditions and Barbieri, A., Langer, J. S. (1989), Phys. Rev. A 39,
further details of the phase diagram be- 5314.
come important. The dynamics of the sys- Barbieri, A., Hong, D. C., Langer, J. S. (1987), Phys.
Rev. A 35, 1802.
tem seem to show a richer structure than Barth, M., Wei, B., Herlach, D. M., Feuerbacher, B.
ordinary directional solidification. The se- (1994), Mater. Sci. Eng. A 178, 305.
lection of spacings between the different Barth, M., Wei, B., Herlach, D. M. (1995), Phys. Rev.
B 51, 3422.
solid phases in materials of practical im- Barth, M., Wei, B., Herlach, D. M. (1997), Mater.
portance occurs through three-dimensional Sci. Eng. A 226 – 228, 770.
defect formation. In addition, nucleation Baumann, R., Kassner, K., Misbah, C., Temkin, D.
(1995 a), Phys. Rev. Lett. 74, 1597.
and faceting of the interfaces should be Baumann, R., Kassner, K., Misbah, C., Temkin, D.
considered. (1995 b), Phys. Rev. Lett. 75, 2445.
A number of problems common to all Bechhoefer, J., Libchaber, A. (1987), Phys. Rev. Lett.
35, 1393.
of these growth modes have only been Bechhoefer, J., Simon, A. J., Libchaber, A., Oswald,
touched upon so far. These problems in- P. (1989), Phys. Rev. A 40, 3974.
clude, for example, the redistribution of Beckermann, C., Wang, C. Y. (1996), Metall. Mater.
Trans. 27 A, 2784.
material far behind the tip regions, the Ben Amar, M. (1990), Phys. Rev. A 41, 2080.
treatment of elastic effects, and the interac- Ben Amar, M. (1998), Physica D 31, 409.
tion with hydrodynamic instabilities due to Ben Amar, M., Brener, E. (1993), Phys. Rev. Lett. 71,
589.
thermal and compositional gradients. Ben Amar, M., Brener, E. (1995), Phys. Rev. Lett. 75,
In summary, we expect the field to re- 561.
main very active in the future, as it is at- Ben Amar, M., Moussallam, B. (1987), Physica 25 D,
tractive from a technological point of view. 155.
Ben Amar, M., Pelce, P. (1989), Phys. Rev. A 39, 4263.
It will certainly provide some surprises and Ben Amar, M., Pomeau, Y. (1968), Europhys. Lett. 2,
new insights into the general concepts of 307.
pattern formation in dissipative systems. Ben Amar, M., Pomeau, Y. (1992), in: Solids Far
from Equilibrium: Godreche, C. (Ed.). Cambridge:
Cambridge University Press.
Ben Amar, M., Bouissou, Ph., Pelce, P. (1988), J.
Crystal Growth 92, 97.
2.8 References Ben Jacob, E., Garik, P. (1990), Nature 343, 523.
Ben Jacob, E., Goldenfeld, N., Langer, J. S., Schön,
Abel, T., Brener, E., Müller-Krumbhaar, H. (1997), G. (1983), Phys. Rev. Lett. 51, 1930.
Phys. Rev. E 55, 7789. Ben Jacob, E., Goldenfeld, N., Langer, J. S., Schön,
Adda Bedia, M., Ben Amar, M. (1991), Phys. Rev. A G. (1984), Phys. Rev. A 29, 330.
43, 5702. Bensimon, D., Pelce, P., Shraiman, B. I. (1987), J.
Akamatsu, S., Faivre, G. (1998), Phys. Rev. E 58, Phys. France 48, 2081.
3302. Bilgram, J. H., Firmann, M., Hürlimann, E. (1989), J.
Akamatsu, S., Faivre, G., Ihle, T. (1995), Phys. Rev. Crystal Growth 96, 175.
E 51, 4751. Billia, B., Jamgotchian, H., Capella, L. (1987), J.
Almgren, R., Dai, W.-S., Hakim, V. (1993), Phys. Crystal Growth 82, 747.
Rev. Lett. 71, 3461. Billia, B., Jamgotchian, H., Capella, L. (1989), J.
Arnold, C. B., Aziz, M. J., Schwarz, M., Herlach, Crystal Growth 94, 987.
D. M. (1999), Phys. Rev. B 59, 334. Binder, K. (2001), in: Phase Transformations in Ma-
Assadi, H., Barth, M., Greer, A. L., Herlach, D. M. terials: Kostorz, G. (Ed.), Weinheim: Wiley-VCH,
(1996), Mat. Sci. Forum 215 – 216, 37. Chap. 4.
Assadi, H., Barth, M., Greer, A. L., Herlach, D. M. Binder, K., Fratzl, P. (2001), in: Phase Transforma-
(1998), Acta Mater. 46, 491. tions in Materials: Kostorz, G. (Ed.), Weinheim:
Baker, J. C., Cahn, J. (1971), in: Solidification: Hu- Wiley-VCH, Chap. 6.
ghel, T. A., Bolling, G. F. (Eds.). Metals Park, OH, Bisang, U., Bilgram, J. H. (1995), Phys. Rev. Lett. 75,
USA: American Society of Metals. 3898.
Boettinger, W. J., Coriell, S. R., Trivedi, R. (1988), Caroli, B., Caroli, C., Roulet, B., Faivre, G. (1989),
in: Rapid Solidification Processing: Principles J. Crystal Growth 94, 253.
and Technologies IV: Mehrabian, R., Parrish, P. A. Caroli, B., Caroli, C., Roulet, B., Brattkus, K. (1990),
(Eds.), Baton Rouge, LA: Claitor’s, p. 13. J. Phys. France 51, 1865.
Boettinger, W. J., Coriell, S. R., Greer, A. L., Karma, Caroli, B., Caroli, C., Fauve, S. (1992), J. Phys. I
A., Kurz, W., Rappaz, M., Trivedi, R. (2000), Acta (Paris) 2, 281.
Mater. 48, 43. Carpay, F. M. A. (1972), in: Handbuch der Mikrosko-
Bösch, A., Shochet, O., Müller-Krumbhaar, H. pie in der Technik, Vol. VI: Freund, H. (Ed.).
(1995), Z. Physik B97, 367. Frankfurt: Umschau, p. 355.
Bouissou, Ph., Pelce, P. (1989), Phys. Rev. A 40, Carrard, M., Gremaud, M., Zimmermann, M., Kurz,
6673. W. (1992), Acta Metall. Mater. 40, 983.
Bouissou, Ph., Perrin, B., Tabeling, P. (1989), Phys. Chan, S.-K., Reimer, H.-H., Kahlweit, M. (1978), J.
Rev. A 40, 509. Crystal Growth 43, 229.
Bouissou, Ph., Chiffaudel, A., Perrin, B., Tabeling, P. Chandrasekhar, C. (1961), Hydrodynamics and Hy-
(1990), Europhys. Lett. 13, 89. drodynamic Stability. Oxford. Clarendon.
Bowley, R., Caroli, B., Caroli, C., Graner, P., No- de Cheveigne, S., Guthman, C., Lebrun, M. M.
zieres, P., Roulet, B. (1989), J. Phys. France 50, (1986), J. Physique 47, 2095.
1377. Chopra, M. A., Glicksman, M. E. Singh, N. B.
Brattkus, K., Misbah, C. (1990), Phys. Rev. Lett. 64, (1988), Metall. Trans. 19 A, 3087.
1935. Classen, A., Misbah, C., Müller-Krumbhaar, H.
Brattkus, K., Caroli, B., Caroli, C., Roulet, B. (1990), (1991), Phys. Rev. A 43, 6920.
J. Phys. France 51, 1847. Cochrane, R. F., Greer, A. L., Eckler, K., Herlach,
Brazovsky, S. (1975), Sov. Phys. JETP 41, 85. D. M. (1991), Mater. Sci. Eng. A 133, 698
Brener, E. (1993), Phys. Rev. Lett. 71, 3653. Collins, J. B., Levine, H. (1985), Phys. Rev. B 31,
Brener, E. A., Levine, H. (1991), Phys. Rev. A 43, 6119.
883. Corriell, S., Sekerka, R. F. (1983), J. Crystal Growth
Brener, E. A., Melnikov, V. I. (1990), J. Phys. France 61, 499.
51, 157. Corriell, S., Cordes, M., Boettinger, M., Sekerka, R.
Brener, E. A., Melnikov, V. I. (1991), Adv. Phys. 40, (1980), J. Crystal Growth 49, 13.
53. Coullet, P., Goldstein, R. E., Gunaratne, G. H.
Brener, E. A., Temkin, D. E. (1989), Europhys. Lett. (1989), Phys. Rev. Lett. 63, 1954.
10, 171. Datye, V., Langer, J. S. (1981), Phys. Rev. B 24,
Brener, E. A., Temkin, D. (1995), Phys. Rev. E 51, 4155.
351. Datye, V., Mathur, R., Langer, J. S. (1981), J. Stat.
Brener, E. A., Geilikman, M. B., Temkin, D. E. Phys. 29, 1.
(1987), Sov. Phys. JETP 65, 93. Deissler, R. J. (1987), Physica 22 D, 233.
Brener, E. A., Geilikman, M. B., Temkin, D. E. Diepers, H.-J., Beckermann, C., Steinbach, I. (1997),
(1988), Sov. Phys. JETP 67, 1002. in: Solidification Processing 1997: Beech, J.,
Brener, E. A, Müller-Krumbhaar, H., Temkin, D. Jones H. (Eds.), Sheffield, UK: University, p. 426.
(1996), Phys. Rev. E 54, 2714. Dombre, T. Hakim, V. (1987), Phys. Rev. A 36, 2811.
Burden, M. H., Hunt, J. D. (1974), J. Crystal Growth Double, D. D. (1973), Mater. Sci. Eng. 11, 325.
22, 99. Dougherty, A. (1991), J. Crystal Growth 110, 501.
Burkhardt, T., Kroll, D., Müller-Krumbhaar, H. Dougherty, A., Gollub, J. P. (1988), Phys. Rev. A 38,
(1977), J. Crystal Growth 38, 13. 3043.
Caginalp, G., (1989), Phys. Rev. A 39, 5887. Dougherty, A., Kaplan, P. D., Gollub, J. P. (1987),
Callen, H. B. (1960), Thermodynamics. New York: Phys. Rev. Lett. 58, 1652.
Wiley. Eckhaus, W. (1965), Studies in Nonlinear Stability
Caroli, B., Caroli, C., Roulet, B. (1982), J. Phys. Theory; Springer Tracts in Natural Philosophy 6.
France 43, 1767. Berlin: Springer.
Caroli, B., Caroli, C., Roulet, B. (1984), J. Crystal Eckler, K., Cochrane, R. F., Herlach, D. M., Feuer-
Growth 66, 575. bacher, B. (1991 a), Mater. Sci. Eng. A 133, 702.
Caroli, B., Caroli, C., Roulet, B., Langer, J. S. Eckler, K., Herlach, D. M., Hamerton, R. G., Greer,
(1986 a), Phys. Rev. A 33, 442. A. L. (1991 b), Mater. Sci. Eng. A 133, 730.
Caroli, B., Caroli, C., Misbah, C., Roulet, B. Eckler, K., Cochrane, R. F., Herlach, D. M., Feuer-
(1986 b), J. Phys. France 47, 1623. bacher, B., Jurisch, M. (1992), Phys. Rev. B 45,
Caroli, B., Caroli, C., Roulet, B. (1987), J. Phys. 5019.
France 48, 1423. Eckler, K., Herlach, D. M. (1994), Mater. Sci. Eng.
Caroli, B., Caroli, C., Roulet, B. (1988), in: Solid A 178, 159.
State Phenomena: Kubin, L. P., Martin, G. Eds.). Eckler, K., Herlach, D. M., Aziz, M. J. (1994), Acta
Aedermannsdorf, CH: Trans. Tech. Publ. Metall. Mater. 42, 975.
2.8 References 167
Elliot, R. (1983), Eutectic Solidification Processing Hogan, L. M., Kraft, R. W., Lemkey, F. D. (1971),
– Crystalline and Glassy Alloys. London: Butter- Adv. Mater. Res. 5, 83.
worths. Honjo, H., Ohta, S., Sawada, Y. (1985), Phys. Rev.
Esaka, H. (1986), Sc.D. thesis, EPFL, Lausanne, CH. Lett. 55, 841.
Esaka, H., Kurz, W. (1985), J. Crystal Growth 72, Huang, S. C., Glicksman, M. E. (1981), Acta Metall.
578. 29, 701; 717.
Esaka, H. Kurz, W., Trivedi, R. (1988), in: Solidifica- Hunt, J. D. (1979), in: Solidification and Casting of
tion Processing 1987. London: The Institute of Metals. London: TMS, p. 1.
Metals, p. 198. Hunt, J. D. (1987), private communication.
Eshelman, M. A., Trivedi, R. (1987), Acta Metall. 35 Hunt, J. D., Jackson, K. A. (1966), Trans. Metall.
2443. Soc. AIME 236, 843.
Eshelman, M., Seetharaman, V., Trivedi, R. (1988), Hunt, J. D., Jackson, K. A., Brown, H. (1966), Rev.
Acta Metall. 36, 1165. Sci. Instr. 37, 805.
Faivre, G. (1996), J. Crystal Growth 166, 29. Hunt, J. D., Lu, S.-Z. (1996), Metall. Mater. Trans.
Faivre, G., Mergy, J. (1992a), Phys. Rev. A 45, 7320. 27A, 611.
Faivre, G., Mergy, J. (1992b), Phys. Rev. A 46, 963. Hunziker, O., Vandyoussefi, M., Kurz, W. (1998),
Faivre, G., de Cheveigne, S., Guthmann, C., Kurow- Acta Mater. 46, 6325.
sky, P. (1989), Europhys. Lett. 9, 779. Huse, D. (1986), Phys. Rev. B 34, 7845.
Fauve, S., Douady, S., Thual, O. (1991), J. Phys. II 1, Ihle, T., Müller-Krumbhaar, H. (1994), Phys. Rev.
311. E 49, 2972.
Fisher, D. J., Kurz, W. (1978), unpublished. Ivantsov, G. P. (1947), Dokl. Akad. Nauk SSSR 58,
Fisher, D. J., Kurz, W. (1980), Acta Metall. 28, 777. 567.
Flemings, M. C. (1991), in: Materials Science and Jackson, K. A. (1958), in: Liquid Metals and Solidifi-
Technology Vol. 15, Processing of Metals and Al- cation. Cleveland: ASM, p. 174.
loys: Cahn, R. W. (Ed.). Weinheim: VCH Verlags- Jackson, K. A., Hunt, J. D. (1966), Trans. Metall.
gesellschaft, p. 1. Soc. AIME 236, 1129.
Fujioka, T. (1978), Ph.D. Thesis, Carnegie-Mellon Jamgotchian, H., Trivedi, R., Billia, B. (1993), Phys.
Univ. Pittsburgh (unpublished). Rev. E 47, 4313.
Geilikman, M. B., Temkin, D. (1984), J. Crystal Jones, H., Kurz, W. (1981), Z. Metallkde 72, 792.
Growth 67, 607. Kahlweit, M. (1970), J. Crystal Growth 6, 125.
Gill, S., Kurz, W. (1993), Acta Metall. Mater. 41, Karma, A. (1986), Phys. Rev. Lett. 57, 858.
3563. Karma, A. (1987), Phys. Rev. Lett. 59, 71.
Gill, S., Kurz, W. (1995), Acta Metall. Mater. 43, Karma, A. (1994), Phys. Rev. E 49, 2245.
139. Karma, A. (1998), Intern. J. Non-Equilib. Process.
Ginibre, M., Akamatsu, S., Faivre, G. (1997), Phys. 11, 201.
Rev. E 56, 780. Karma, A., Langer, J. S. (1984), Phys. Rev. A 30,
Glickman, M. E., Singh, N. B. (1986), in: Rapidly 3147.
Solidified Powder Alloys: Fine, M. E. Stark, E. A. Karma, A., Kotliar, B. G. (1985), Phys. Rev. A 31,
(Eds.). Philadelphia: ASTM, p. 44. 3266.
Glickman, M. E. Singh, N. B. (1989), J. Crystal Karma, A., Pelce, P. (1989 a), Phys. Rev. A 39, 4162.
Growth 98, 277. Karma, A., Pelce, P. (1989 b), Europhys. Lett. 9, 713.
Glicksman, M., Shaefer, R. J., Ayers. J. D. (1976), Karma, A., Rappel, W.-J. (1994), Phys. Rev. Lett. 75,
Metall, Trans. A 7, 1747. 2444.
Glickman, M. E., Singh, N. B., Chopra, M. (1988), Karma, A., Rappel, W.-J. (1996), Phys. Rev. E 53,
J. Crystal Growth 92, 543. 3017.
Goetzinger, R., Barth, M., Herlach, D. M. (1998), Karma, A., Rappel, W.-J. (1998), Phys. Rev. E 57,
Acta Mater. 46, 1647. 4323.
Gremaud, M. Carrard, M., Kurz, W. (1990), Acta Karma, A., Sarkissian, A. (1993), Phys. Rev. E 47,
Metall. Mater. 39, 1431. 513A.
Ha, H. P., Hunt, J. D. (1997), in: Proc. 4th Dec. Int. Karma, A., Sarkissian, A. (1996), Metall. Trans. 27,
Conf. on Solidification Proc. (SP’97) Beech, J., 635.
Jones, H. (Eds.), Sheffield, UK: University, p. 444. Karma, A., Rappel, W. J., Fuh, B. C., Trivedi, R.
Henry, S., Minghetti, T., Rappaz, M. (1998), Acta (1998), Metall. Trans. 29 A, 1457.
Mater. 46, 6431. Kashuba, A., Pokrovsky, V. (1990), Z. Physik 78, 289.
Herlach, D. M. (1994), Mater. Sci. Eng. A 179 – 180, Kassner, K., Misbah, C. (1990), Phys. Rev. Lett. 65,
147. 1458.
Herlach, D. M. (1997), Mater. Sci. Eng. A 226 – 228, Kassner, K., Misbah, C. (1991 a), Phys. Rev. Lett. 66,
348. 445.
Herlach, D. M., Cochrane, R. F., Egry, I., Fecht, H. J., Kassner, K., Misbah, C. (1991 b), Phys. Rev. Lett.
Greer, A. L. (1993), Intern. Mater. Rev. 38, 273. A 44, 6513.
Kassner, K., Misbah, C. (1991 c), Phys. Rev. A 44, Landau, L. D., Lifshitz, E. M. (1970), Statistische
6533. Physik, Berlin: Akademie-Verlag.
Kassner, K., Misbah, C. (1992 a), Phys. Rev. A 45, Langer, J. S. (1971), Ann. Phys. 65, 53.
7372. Langer, J. S. (1980 a), Rev. Mod. Phys. 52, 1.
Kassner, K., Misbah, C. (1992 b), J. Phys. Rev. A 25, Langer, J. S. (1980 b), Phys. Rev. Lett. 44, 1023.
3213. Langer, J. S. (1980 c), Physico-Chem. Hydrodyn. 1,
Kassner, K., Valance, A., Misbah, C., Temkin, D. 41.
(1993), Phys. Rev. E 48, 1091. Langer, J. S. (1986, in: Directions in Condensed Mat-
Kassner, K., Misbah, C., Baumann, R. (1995), Phys. ter. Grinstein, G., Mazenko, G. (Eds.). Singapore:
Rev. E 51, 2751. World Scientific, p. 164.
Kerr, H. W., Kurz, W. (1996), Int. Mater. Rev. 41, Langer, J. S. (1987 a), Phys. Rev. Lett. A 36, 3350.
129. Langer, J. S. (1987 b), in: Le hasard et la matiere/
Kessler, D., Levine, H. (1986 a), Phys. Rev. Lett. 57, Chance and matter: Souletie, J., Vannimenus, J.,
3069. Stora, R. (Eds.). Amsterdam: Elsevier Science
Kessler, D., Levine, H. (1986 b), Phys. Rev. A 33, Publ.
3352. Langer, J. S., Müller-Krumbhaar, H. (1977), J. Crys-
Kessler, D., Levine, H. (1986 c), Phys. Rev. A 33, tal Growth 42, 11.
3325. Langer, J. S., Müller-Krumbhaar, H. (1978), Acta
Kessler, D., Levine, H. (1986 d), Phys. Rev. A 36, Metall. 26, 1681; 1689; 1697.
4123. Langer, J. S., Müller-Krumbhaar, H. (1980), Acta
Kessler, D., Levine, H. (1988), Acta Metall. 36, Metall. 29, 145.
2693. Langer, J. S., Müller-Krumbhaar, H. (1982), Phys.
Kessler, D., Levine, H. (1989), Phys. Rev. A 39, 3041. Rev. A 27, 499.
Kessler, D. Koplik, J., Levine, H. (1987), in: Pattern, Langer, J. S., Turski, L. A. (1977), Acta Metall. 25,
Defects and Microstructures in Nonequilibrium 1113.
Systems: Walgraef, D. (Ed.). Dordrecht: Martinus Lesoult, G. (1980), Ann. Chim. Fr. 5, 154.
Nijhoff. Lesoult, G. (1990), private communication.
Kessler, D., Koplik, J., Levine, H. (1988), Adv. Phys. Levine, H., Rappel, W. (1990), unpublished.
37, 255. Li, D. Eckler, K., Herlach, D. M. (1995b), Europhys.
Kofler, A. (1950), Z. Metallkde 41, 221. Lett. 32, 223.
Koss, M. B., La Combe, J. C., Tennenhouse, L. A., Li, D. Eckler, K., Herlach, D. M. (1996), J. Crystal
Glicksman, M. E., Winsa, E. A. (1999), Metall. Growth 160, 59.
Mater. Trans. 30 A, 3177. Li, D. Herlach, D. M. (1996), Phys. Rev. Lett. 77,
Kramer, L., Ben Jacob, E., Brand, H., Cross, M. 1801.
(1982), Phys. Rev. Lett. 49, 1891. Li, D., Volkmann, T., Eckler, K., Herlach, D. M.
Krug, J., Spohn, H. (1992), in: Solids Far from (1995 a), J. Crystal Growth 152, 101
Equilibrium: Growth, Morphology and Defects: Li, Q., Beckermann, C. (1998), Phys. Rev. E 57, 1.
Godreche, C. (Ed.). Cambridge: Cambridge Uni- Lifshitz, I. M., Slyozov, V. V. (1961), J. Phys. Chem.
versity Press, p. 479. Solids 19, 35.
Kruskal, M., Segur, H. (1985), Asymptotics Beyond Lipton, J., Glicksman, M. E., Kurz, W. (1987), Met-
All Orders in a Model of Dendritic Crystals, all. Trans. 18 A, 341.
A.R.A.P. Tech. Memo. Lipton, J., Kurz, W., Trivedi, R. (1987 a), Acta Met-
Kurowsky, P. (1990), Thesis, GPS-Univ. Paris VII. all. 35, 957.
Kurz, W., Fisher, D. J. (1981), Acta Metall. 29, 11. Livingston, J. D., Cline, H. E., Koch, E. F., Russel, R.
Kurz, W., Fisher, D. J. (1998), Fundamentals of So- R. (1970), Acta Metall. 18, 399.
lidification. Aedermannsdorf, CH: Trans. Tech. Lo, T. S., Karma, A., Plapp, M. (2001), Phys. Rev.
Publ. E63, 1504.
Kurz, W., Trivedi, R. (1990), Acta Metall. Mater. 38, Losert, W., Mesquita, O. N., Figueiredo, J. M. A.,
1. Cummins, H. Z. (1998b), Phys. Rev. Lett. 81, 409.
Kurz, W., Trivedi, R. (1991), Metall. Trans. 22 A, Losert, W., Stillman, D. A., Cummins, H. Z., Kop-
3051. czynski, P., Rappel, W.-J. Karma, A. (1998 a),
Kurz, W., Trivedi, R. (1996), Metall. Mater. Trans. Phys. Rev. E 58, 7492.
27A, 625. Ludwig, A., Kurz, W. (1996a), Scripta Mater. 35,
Kurz, W., Sahm, P. R. (1975), Gerichtet erstarrte eu- 1217
tektische Werkstoffe. Berlin: Springer Verlag. Ludwig, A., Kurz, W. (1996b), Acta Mater. 44, 3643
Kurz, W., Giovanola, B., Trivedi, R. (1986), Acta Mashaal, M., Ben Amar, M., Hakim, V. (1990), Phys.
Metall. 34, 823. Rev. A 41, 4421.
Kurz, W., Giovanola, B., Trivedi, R. (1988), J. Crys- Maurer, J., Bouissou, P., Perrin, B., Tabeling, P.
tal Growth 91, 123. (1988), Europhys. Lett. 8, 67.
2.8 References 169
Mazumder, P., Trivedi, R., Karma, A. (2000), Metall. Rappel, W.-J., Brener, E. A. (1992), J. Phys. I France
Mater. Trans. A31, 1233. 2, 1779.
McFadden, G. B., Coriell, S. R. (1984), Physica Raz, E., Lipson, S. G., Polturak, E. (1989), Phys. Rev.
12 D, 253. A 40, 1088.
Meakin, P. (1988), in: Phase Transitions and Critical Riecke, H., Paap, H. G. (1986), Phys. Rev. A 33, 547.
Phenomena, Vol. 12: Domb, C., Lebowitz, J. Riquet, J. P., Durant, F. (1975), Mat. Res. Bull. 10,
(Eds.). London: Academic, p. 336. 399.
Meiron, D. I. (1986), Phys. Rev. A 33, 2704. Saffmann, P. G., Taylor, G. I. (1958), Proc. R. Soc.
Misbah, C. (1987), J. Phys. France 48, 1265. London Ser. A 254, 312.
Misbah, C. (1989), J. Phys. France 50, 971. Sahm, P. R., Keller, M. H. (1991), in: Materials
Misbah, C., Temkin, D. (1992), Phys. Rev. A 46, Science and Technology, Vol. 15, Processing of
4497. Metals and Alloys. Weinheim: VCH Verlagsge-
Misbah, C., Müller-Krumbhaar, H., Saito, Y. (1990), sellschaft, p. 539.
J. Crystal Growth 99, 156. Saito, Y., Goldbeck-Wood, G. Müller-Krumbhaar, H.
Misbah, C., Müller-Krumbhaar, H., Temkin, D. E. (1987), Phys. Rev. Lett. 58, 1541.
(1991), J. Physique I 1, 585. Saito, Y., Goldbeck-Wood, G. Müller-Krumbhaar, H.
Mordike, B. L. (1991), in: Materials Science and (1988), Phys. Rev. Lett. A 38, 2148.
Technology Vol. 15, Processing of Metals and Al- Saito, Y., Misbah, C., Müller-Krumbhaar, H. (1989),
loys: Cahn, R. W. (Ed.). Weinheim: VCH Verlags- Phys. Rev. Lett. 63, 2377.
gesellschaft, p. 111. Schumacher, P., Greer, A. L., Worth, J., Evans, P. V.,
Mullins, W. Sekerka, R. (1963), J. Appl. Phys. 34, Kearns, M. A., Fisher, P., Green, A. H. (1998),
323. Mater. Sci. Technol. 14, 394.
Mullins, W. Sekerka, R. (1964), J. Appl. Phys. 35, Schwarz, M., Arnold, C. B., Aziz, M. J., Herlach,
444. D. M. (1997), Mater. Sci. Eng. A 226–228, 420.
Nakaya, U. (1954), Snow Crystals. Cambridge: Har- Schwarz, M., Karma, A., Eckler, K., Herlach, D. M.
vard Univ. Press. (1990), Phys. Rev. Lett. 56, 324.
Nash, G. E., Glicksman, M. (1974), Acta Metall. 22, Seetharaman, V., Trivedi, R. (1988), Metall. Trans.
1283. 19 A, 2955.
Nelson, D. (1983), in: Phase Transitions and Critical Seetharaman, V. Eshelman, M., Trivedi, R. (1988),
Phenomena: Domb, C., Lebowitz, J. (Eds.). New Acta Metall. 36, 1175.
York: Academic Publ. Singh, N. B., Glicksman, M. E. (1989), J. Crystal
Newell, A. C., Whitehead, J. A. (1969), J. Fluid. Growth 98, 534.
Mech. 38, 279. Somboonsuk, K., Trivedi, R. (1985), Acta Metall. 33,
Ohno, K., Trinkaus, H., Müller-Krumbhaar, H. 1051.
(1990), J. Crystal Growth 99, 68. Somboonsuk, K., Mason, J. T., Trivedi, R. (1984),
Papapetrou, A. Z. (1935), Z. f. Kristallograph. 92, 89. Metall. Trans. A15, 967.
Park, J. S., Trivedi, R. (1998), J. Crystal Growth 187, Strain, J. (1989), J. Comput. Phys. 85, 342.
511. Temkin, D. E. (1985), Kristallografia 30, 1055.
Pelce, P., Clavin, P. (1987), Europhys. Lett. 3, 907. Temkin, D. E. (1990), unpublished.
Pelce, P., Pomeau, Y. (1986), Studies Appl. Math. 74, Tirmizi, S. H., Gill, W. N. (1989), J. Crystal Growth
1283. 96, 277.
Pelce, P., Pumir, A. (1985), J. Crystal Growth 73, 357. Tong, X., Beckermann, C., Karma, A. (1998), in:
Penrose, O., Fife, P. G. (1990), Physica D 43, 44. Modeling of Casting, Welding, and Advanced So-
Perepezko, J. H., Uttormark, M. J. (1996), Metall. lidification Conference VIII: Thomas, B. G., Beck-
Mater. Trans. 27 A, 533. ermann, C. (Eds.), Warrendale, PA: TMS, p. 613.
Perepezko, J. H., Mueller, B. A., Ohsaka, K. (1987), Tonhardt, R., Amberg, G. (1998), J. Crystal Growth
in: Undercooled Alloy Phases: Collings, E. W., 194, 406.
Koch, C. C. (Eds.). Warrendale: TMS, p. 289. Tournier, S., Barth, M., Herlach, D. M., Vinet B.
Plapp, M., Karma, A. (1999), Phys. Rev. E 60, 6865. (1997), Acta Mater. 45, 191.
Provatas, N., Goldenfeld, N., Dantzig, J. (1998), Trivedi, R. (1980), J. Crystal Growth 49, 219.
Phys. Rev. Lett. 80, 3308. Trivedi, R. (1984), Metall. Trans. 15 A, 977.
Quested, P. N., McLean, M. (1984), Mater. Science Trivedi, R., Karma, A., Lo, T. S., Mazumder, P.,
Eng. 65, 171. Plapp, M., Park, J. S. (1998), in: Zermatt Work-
Rabaud, M., Couder, Y., Gerard, N. (1988), Phys. shop on “Solidification Microstructures”, CD,
Rev. A37, 935. Swiss Federal Institute of Technology Lausanne,
Rappaz, M. (1989), Inter. Mater. Rev. 34, 93. 1015 Lausanne EPFL, Switzerland. This CD (PC
Rappaz, M., Desbiolles, J.-L., Gandin, C.-A., Henry, format) may be obtained from the Laboratory
S., Smoroz, A. (2000), Mater. Sci. Forum 329 – of Physical Metallurgy, EPFL through: arlette.
330, 389. blatti@epfl.ch
Trivedi, R., Kurz, W. (1986), Acta Metall. 34, 1663. Wagner, C. (1961), Z. Elektrochem. 65, 581.
Trivedi, R., Laorchan, V. (1988), Acta Metall. 39, Wang, C. Y., Beckermann, C. (1996), Metall. Mater.
1941; 1951. Trans. 27 A, 2765.
Trivedi, R., Lipton, J., Kurz, W. (1987), Acta Metall. Wang, C. Y., Ahuja, S. Beckermann, C., de Groh III,
35, 965. H. C. (1995), Metall. Trans. 26 B, 111.
Trivedi, R., Mason, J. T., Verhoeven, J. D., Kurz, W. Wang, S.-L., Sekerka, R. F., Wheeler, A. A., Murray,
(1991), Metall. Trans. 22 A, 2523. B. T., Corriel, S. R., Braun, R. J., McFadden, G. B.
Trivedi, R., Somboonsuk. K. (1984), Mater. Sci. Eng. (1993), Physica D 69, 189.
65, 65. Wheeler, A. A., McFadden, G. B., Boettinger, W. J.
Trivedi, R., Somboonsuk, K. (1985), Acta Metall. 33, (1996), Proc. Royal Soc. London A 452, 495.
1061. Warren, J. A., Langer, J. S. (1990), Phys. Rev. A42,
Ungar, L. H., Brown, R. A. (1984 a), Phys. Rev. B 29, 3518.
1367. Wollkind, D., Segel, L. (1970), Philos. Trans. R. Soc.
Ungar, L. H., Brown, R. A. (1984 b), Phys. Rev. B 30, London 268, 351.
3993. Wu, Y., Piccone, T. J., Shiohara, Y., Flemings, M. C.
Ungar, L. H., Brown, R. A. (1985 a), Phys. Rev. B 31, (1987), Metall. Trans. 18, 915; 925.
1367. Wulff, G. (1901), Z. Kristallogr. Mineral. 34, 449.
Ungar, L. H., Brown, R. A. (1985 b), Phys. Rev. B 31, Xiao, R. F., Alexander, J. I. D., Rosenberger, F.
5931. (1988), Phys. Rev. A38, 2447.
Uwaha, M., Saito, Y. (1990), J. Crystal Growth 99, Yokoyama, E., Kuroda, T. (1988), in: Dynamics of
175. Ordering Processes in Condensed Matter: Ko-
Valance, A., Misbah, C., Temkin, D., Kassner, K. mura, S., Furukawa, H. (Eds.). New York: Plenum
(1993), Phys. Rev. E 48, 1924. Publ., p. 95.
Vandyoussefi, M., Kerr, H. W., Kurz, W. (2000), Acta Zettlemoyer, A. (Ed.) (1969), Nucleation I. New
Mater. 48, 2297. York: Dekker.
Versnyder, F. L., Shank, M. E. (1970), Mater. Sci. Zettlemoyer, A. (Ed.) (1976), Nucleation II. New
Eng. 6, 213. York: Dekker.
Volkmann, T., Wilde, G., Willnecker, R., Herlach, D. Zimmermann, M., Carrard, M., Kurz, W. (1989),
M. (1998), J. Appl. Phys. 83, 3028. Acta Metall. 37, 3305.
Voronkov, V. V. (1983), in: Modern Theory of Crys- Zimmermann, M., Karma, A., Carrard, M. (1990),
tal Growth: Chernov, A. A., Müller-Krumbhaar, Phys. Rev. B 42, 833.
H. (Eds.). Heidelberg: Springer Verlag. p. 75.
3 Diffusion Kinetics in Solids
Graeme E. Murch
Department of Mechanical Engineering, University of Newcastle, N.S.W. 2308, Australia

3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175
3.2 Macroscopic Diffusion . . . . . . . . . . . . . . . . . . . . . . . . . . . 175
3.2.1 Fick’s Law . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175
3.2.2 Types of Diffusion Coefficients . . . . . . . . . . . . . . . . . . . . . . . 178
3.2.2.1 Tracer or Self-Diffusion . . . . . . . . . . . . . . . . . . . . . . . . . . . 178
3.2.2.2 Impurity and Solute Diffusion . . . . . . . . . . . . . . . . . . . . . . . . 178
3.2.2.3 Chemical or Interdiffusion . . . . . . . . . . . . . . . . . . . . . . . . . . 178
3.2.2.4 Intrinsic or Partial Diffusion Coefficients . . . . . . . . . . . . . . . . . . 179
3.2.2.5 Surface Diffusion Coefficients . . . . . . . . . . . . . . . . . . . . . . . . 180
3.2.3 Phenomenological Equations of Irreversible Thermodynamics . . . . . . . 181
3.2.3.1 Tracer Diffusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182
3.2.3.2 Chemical Diffusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 184
3.2.3.3 Einsteinian Expressions for the Phenomenological Coefficients . . . . . . 185
3.2.3.4 Relating Phenomenological Coefficients to Tracer Diffusion Coefficients . 185
3.2.4 Short-Circuit Diffusion . . . . . . . . . . . . . . . . . . . . . . . . . . . 186
3.3 Microscopic Diffusion . . . . . . . . . . . . . . . . . . . . . . . . . . . 189
3.3.1 Random Walk Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189
3.3.1.1 Mechanisms of Diffusion . . . . . . . . . . . . . . . . . . . . . . . . . . 189
3.3.1.2 The Einstein Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192
3.3.1.3 Tracer Correlation Factor . . . . . . . . . . . . . . . . . . . . . . . . . . 193
3.3.1.4 Impurity Correlation Factor . . . . . . . . . . . . . . . . . . . . . . . . . 196
3.3.1.5 Correlation Factors for Concentrated Alloy Systems . . . . . . . . . . . . 199
3.3.1.6 Correlation Factors for Highly Defective Systems . . . . . . . . . . . . . 201
3.3.1.7 The Physical or Conductivity Correlation Factor . . . . . . . . . . . . . . 204
3.3.1.8 Correlation Functions (Collective Correlation Factors) . . . . . . . . . . . 206
3.3.2 The Nernst – Einstein Equation and the Haven Ratio . . . . . . . . . . . . 211
3.3.3 The Isotope Effect in Diffusion . . . . . . . . . . . . . . . . . . . . . . . 213
3.3.4 The Jump Frequency . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 216
3.3.4.1 The Exchange Frequency . . . . . . . . . . . . . . . . . . . . . . . . . . 217
3.3.4.2 Vacancy Concentration . . . . . . . . . . . . . . . . . . . . . . . . . . . . 218
3.4 Diffusion in Materials . . . . . . . . . . . . . . . . . . . . . . . . . . . 219
3.4.1 Diffusion in Metals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 220
3.4.1.1 Self-Diffusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 220
3.4.1.2 Impurity Diffusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 220
3.4.2 Diffusion in Dilute Alloys . . . . . . . . . . . . . . . . . . . . . . . . . . 221

ISBN: 3-527-30256-5
172 3 Diffusion Kinetics in Solids
3.4.2.1 Substitutional Alloys . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 221

3.4.2.2 Interstitial Alloys . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 224
3.4.3 Diffusion in Concentrated Binary Substitional Alloys . . . . . . . . . . . 226
3.4.4 Diffusion in Ionic Crystals . . . . . . . . . . . . . . . . . . . . . . . . . . 227
3.4.4.1 Defects in Ionic Crystals . . . . . . . . . . . . . . . . . . . . . . . . . . . 227
3.4.4.2 Diffusion Theory in Ionic Crystals . . . . . . . . . . . . . . . . . . . . . 228
3.5 Experimental Methods for Measuring Diffusion Coefficients . . . . . . 231
3.5.1 Tracer Diffusion Methods . . . . . . . . . . . . . . . . . . . . . . . . . . 231
3.5.2 Chemical Diffusion Methods . . . . . . . . . . . . . . . . . . . . . . . . 232
3.5.3 Diffusion Coefficients by Indirect Methods . . . . . . . . . . . . . . . . . 233
3.5.3.1 Relaxation Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 233
3.5.3.2 Nuclear Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 234
3.5.4 Surface Diffusion Methods . . . . . . . . . . . . . . . . . . . . . . . . . 234
3.6 Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 235
3.7 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 235

a lattice parameter
b1 , b2 solvent enhancement factors
B1 , B2 solute enhancement factors
C, Ci concentration: particles per unit volume of species i
cA , cB , ci , cv , cp mole fractions of metal A, B, species i, vacancies, paired interstitials
D diffusion coefficient (in m2 s –1)
D¢ short-circuit diffusion coefficient
D̃ collective diffusion coefficient or interdiffusion coefficient
D* tracer diffusion coefficient
DAI , DBI , etc. intrinsic or partial diffusion coefficient of metal A, B, etc.
Dl lattice diffusion coefficient
Ds diffusion coefficient derived from the ionic conductivity
D0 pre-exponential factor
e electronic charge
Evf energy of vacancy formation
E b, E B binding energy of impurity to dislocation (impurity-vacancy binding energy)
f, f I ( fc ) tracer correlation factor, physical correlation factor
Fvf Helmholtz free energy for vacancy formation
g coordination
Gm Gibbs free energy of migration
Hvf enthalpy of vacancy formation
Hm enthalpy of migration
HR Haven ratio
J flux of atoms
k Boltzmann constant
K equilibrium constant
L phenomenological coefficients
l spacing between grain boundaries (average grain diameter)
l distance between dislocation pinning points
m mass
–
PAV vacancy availability factor
Q activation energy for diffusion
R vector displacement
DRi total displacement of species i
r, rq jump distance of atom, of charge q
Svf entropy of vacancy formation
Sm entropy of migration
t time
T temperature
TM.Pt. temperature of melting point
u mobility
·vÒ average drift velocity
V volume
x distance, coordinate
Xi , Xv driving force on species i, vacancies
Zi number of charges on i
g activity coefficient
G, Gq , Gi jump frequency (of charge q; of species i)
n vibration frequency
s ionic conductivity
w exchange frequency
CASCADE computer code

DEVIL computer code
erf Gaussian error function
PPM path probability method
SIMS secondary ion mass spectrometry
3.2 Macroscopic Diffusion 175
“The elementary diffusion process is tance) in one direction of a certain species

so very fundamental and ubiquitous of atom, a flux J of atoms of the same spe-
in the art and science of dealing with cies is established down the concentration
matter in its condensed phase that it gradient. The law relating flux and concen-
never ceases to be useful but, at the tration gradient is Fick’s First Law, which,
same time, is a problem which is for an isotropic medium or cubic crystal
never really solved. It remains impor- can be expressed as
tant by any measure.”
∂C
D. Lazarus, 1984 J=−D (3-1)
∂x
The proportionality factor or “coefficient”
3.1 Introduction of ∂C/∂x is termed the “diffusion coeffi-
cient” or less commonly the “diffusivity”.
Many phenomena in materials science The recommended SI units for D are m2 s–1
depend in some way on diffusion. Common but much of the literature is still in the
examples are sintering, oxidation, creep, older c.g.s. units cm2 s–1. The negative sign
precipitation, solid-state chemical reac- in Eq. (3-1) arises because the flux is in the
tions, phase transformations, and crystal opposite direction to the concentration gra-
growth. Even thermodynamic properties dient. This negative sign could, of course,
and structure are sometimes dependent on have been absorbed into D, but it is more
diffusion, or rather the lack of it. Many of convenient for D to be a positive quantity.
these phenomena are the subjects of other When an external force such as an
contributions in this series. This contribu- electric field also acts on the system a more
tion is concerned with the fundamentals of general expression can be given:
the diffusion process itself.
∂C
The depth of subject matter is generally J=−D + 〈v 〉 C (3-2)
introductory, and no prior knowledge of ∂x
solid-state diffusion is assumed. Where
where ·vÒ is the average velocity of the
possible, the reader is directed to more de-
center of mass arising from the external
tailed texts, reviews, and data compila-
force on the particles. The first term in Eq.
tions.
(3-2) is thus the diffusive term, and the sec-
ond term is the drift term. Note the inde-
pendence of these terms. The external force
3.2 Macroscopic Diffusion here is assumed to be applied gradually so
that the system moves through a series of
3.2.1 Fick’s Laws
equilibrium states. When the force is sud-
Although diffusion of atoms, or atomic denly applied the system can be thrown out
migration, is always occurring in solids at of equilibrium. These matters and the types
temperatures above absolute zero, for mac- of external force are considered by Flynn
roscopically measurable diffusion a gradi- (1972) in a general consideration of diffu-
ent of concentration is required. In the sion under stress.
presence of such a concentration gradient By itself, Fick’s First Law (Eq. (3-1)) is
∂C/∂x (where C is the concentration in, say, not particularly useful for diffusion mea-
particles per unit volume and x is the dis- surements in the solid state since it is virtu-
ally impossible to measure an atomic flux solution, C (x, t), of the diffusion equation
unless steady state is reached. Further, be- has been established, the diffusion coeffi-
cause solid-state diffusivities are generally cient itself is obtained as a parameter by
small, the attainment of steady state in fitting the experimental C (x, t) to the ana-
a macroscopic specimen can take a very lytical C (x, t). In the following we shall fo-
long time. In a few cases, where the solid- cus on some analytical solutions for some
state diffusivity is high, for example car- well-known initial and boundary condi-
bon diffusion in austenite (Smith, 1953), tions used in experimental diffusion stud-
the steady-state flux and the concentration ies.
gradient can be measured and the diffusiv- Let us first examine some solutions for
ity obtained directly from Eq. (3-1). Eq. (3-5). In a very common experimental
In order to produce a basis for measuring arrangement for “self” and impurity diffu-
the diffusion coefficient, Eq. (3-1) is usu- sion a very thin deposit of amount M of ra-
ally combined with the equation of con- dioactive isotope is deposited as a sand-
tinuity: wich layer between two identical samples
of “infinite” thickness. After diffusion for a
∂J ∂C
=− (3-3) time t the concentration is described by
∂x ∂t
M
to give Fick’s Second Law: C ( x, t ) = exp ( − x 2 / 4 Dt ) (3-8)
2 p Dt
∂C ∂ ⎛ ∂C ⎞
= ⎜D ⎟ (3-4) which is illustrated in Fig. 3-1. If the de-
∂t ∂x ⎝ ∂x ⎠ posit is left as a surface layer rather than a
sandwich, C (x, t) is doubled.
If the diffusion coefficient is independent
In another experimental arrangement,
of concentration and therefore position,
the surface concentration Cs of the diffus-
then Eq. (3-4) reduces to
ing species is maintained constant for time
∂C ∂2 C t, perhaps by being exposed to an atmo-
=D 2 (3-5) sphere of it. Again, the substrate is thick or
∂t ∂x
mathematically infinite. The solution of the
The second-order partial differential equa- diffusion equation is
tion Eq. (3-5) (or (3-4)) is sometiems called
the “diffusion equation”. Eq. (3-2) can also C ( x , t ) − Cs
= erf ( x / 2 Dt ) (3-9)
be developed in the same way to give C0 − Cs
∂C ∂ ⎛ ∂C ⎞ ∂ where C0 is the initial or background con-

= ⎜D ⎟− ( 〈v 〉 C ) (3-6)
∂t ∂x ⎝ ∂x ⎠ ∂x centration of the diffusing species in the
substrate and erf is the Gaussian error func-
and, if ·vÒ and D are independent of C, then tion defined by
z
∂C ∂2 C ∂C 2
∫ exp ( − u ) du
2
= D 2 − 〈v 〉 (3-7) erf z = (3-10)
∂t ∂x ∂x p 0
In order to obtain a solution to the diffu- This function is now frequently available
sion equation, it is necessary to establish as a scientific library function on most
the initial and boundary conditions. Once a modern computers. Accurate series expan-
Figure 3-1. Time evolution of Eq. (3-8). Figure 3-2. Time evolution of Eq. (3-12).
sions can be found in mathematical func- diffusion coefficient can be expressed as

tion handbooks. (Boltzmann, 1894; Matano, 1933)
The total amount of diffusant S taken up C′
(or, in fact, lost, depending on the relative 1 ⎛ ∂x ⎞
D (C ′ ) = − ⎜ ⎟ ∫ x dC (3-13)
values of Cs and C (x, t)) from the substrate 2 t ⎝ ∂C ⎠ C ′ C1
is given by
where
S (t) = [2 /C0 – Cs )] /A (D t /p)1/2 (3-11) C0
where A is the surface area of the sample. ∫ x dC = 0 (3-14)

C1
A further common experimental ar-
rangement is the juxtaposition of two “infi- A very detailed description of the use of
nite” samples, one of which has a uniform this analysis is given by Borg and Dienes
concentration C0 and the other a concentra- (1988).
tion C1 of the diffusing species. After time In many practical situations of interdif-
t the solution is fusion the relevant phase diagram traversed
C ( x , t ) − C0 1 will ensure that diphasic regions or new
= [1 − erf ( x / 2 Dt )] (3-12) phases will appear. This does not imply in-
C1 − C0 2
cidentally that all equilibrium phases will
This gives a time evolution of the concen- thus appear. The Boltzmann – Matano anal-
tration profile as shown in Fig. 3-2. ysis can still be applied within single phase
On many occasions where this experi- regions of the concentration profile. The
mental arrangement is used, we are con- growth of a new phase, provided it is dif-
cerned with diffusion in a chemical compo- fusion controlled, can usually be described
sition gradient. The relevant diffusion co- by a parabolic time law. These matters are
efficient (see Sec. 3.2.2.3) is frequently de- dealt with in detail by Philibert (1991).
pendent on concentration, so Eq. (3-12) is The solutions of the diffusion equation
then inappropriate, and a solution to Eq. given here are among the more commonly
(3-4) where D = D (C ) must be sought. A encountered ones in solid-state diffusion
wellknown technique for this is the graphi- studies. Numerous others have been given
cal integration method, usually called the by Carslaw and Jaeger (1959) and Crank
Boltzmann – Matano analysis. The general (1975).
solution for the concentration-dependent
3.2.2 Types of Diffusion Coefficients 3.2.2.2 Impurity and Solute Diffusion

For the non-specialist, the meaning and In order to measure the impurity diffu-
significance of the various diffusion coeffi- sion coefficient, the tracer is now the im-
cients used can be confusing because of in- purity and is different chemically from the
consistent terminology in the literature. An host. However, the concentration of impur-
attempt will be made here to clarify the sit- ity must be sufficiently low that there is not
uation as much as possible. a chemical composition gradient. Strictly,
of course, the tracer impurity should be
3.2.2.1 Tracer or Self-Diffusion permitted to diffuse into the sample already
containing the same concentration of im-
Consider a well-annealed sample of a
purity. In practice, the concentration of
pure metal. Although atoms are diffusing
tracer impurity is normally kept extremely
about in the sample at a rate depending on
low, thereby making this step unnecessary.
temperature, from a macroscopic point of
Because the impurity is always in stable
view nothing appears to be happening. In
solid solution (unless implanted), is often
order to observe diffusion macroscopically
termed the solute and the impurity diffu-
we must impose a concentration gradient.
sion coefficient is sometimes also termed
For the case of a pure metal, a radioactive
the solute diffusion coefficient at infinite
tracer of the same metal is used. The result-
dilution. However, the terminology solute
ing diffusion coefficient is termed the
diffusion coefficient is often “reserved” for
tracer diffusion coefficient with the symbol
dilute alloys, where, in addition, we often
D*. Because the tracer is chemically the
measure the solvent diffusion coefficient.
same as the host, this diffusion coefficient
In the context of those experiments both
is also termed the self-diffusion coefficient,
solute and solvent diffusion coefficients
although sometimes this terminology is re-
frequently depend on solute content (see
served for the tracer diffusion coefficient
Sec. 3.4.2). In all of these experiments, as
divided by the tracer correlation factor, f
in self-diffusion, the chemical composition
(see Sec. 3.3.1.3). The same idea is readily
of the sample must remain essentially un-
extended to alloys and compounds. How-
changed by the diffusion process, other-
ever, care must always be taken that a
wise it is a chemical diffusion experiment.
chemical composition gradient is not un-
wittingly imposed. For example, in order to
3.2.2.3 Chemical or Interdiffusion
measure the oxygen self-diffusion coeffi-
cient in a nonstoichiometric compound So far, we have discussed diffusion coef-
such as UO2 + x , we can permit 18O (a stable ficients which are measured in the absence
isotope that can be probed later by nuclear of chemical composition gradients. Chemi-
analysis or Secondary Ion Mass Spectrom- cal diffusion is the process where diffusion
etry) to diffuse in from the gas phase, pro- takes place in the presence of a chemical
vided that the partial pressure of oxygen is composition gradient. It is the diffusion co-
already in chemical equilibrium with the efficients describing this process which
composition of the sample. Alternatively, a generate the greatest amount of confusion.
layer of U18O2 + x can be deposited on the It is helpful to look at several examples.
surface of a sample of UO2 + x provided that Consider first diffusion in a pseudo-one-
it has exactly the same chemical composi- component system. One example is the dif-
tion as the substrate. fusion of an adsorbed monolayer onto a
clean section of surface. Another is the dif- of its practical significance the interdiffu-
fusion between two metal samples differ- sion coefficient is the one often quoted in
ing only in their relative concentrations of metal property data books (Brandes, 1983;
a highly mobile interstitial species such as Mehrer, 1990; Beke 1998, 1999). See also
H. A further example is diffusion between the following section for further discussion
two nonstoichiometric compounds, e.g., of D̃.
Fe1– d O and Fe1– d ¢ O. In all of these cases
only one species of atom is involved in the
3.2.2.4 Intrinsic or Partial Diffusion
diffusion process which brings the system
Coefficients
to a common composition (which is why it
can be considered a pseudo-one-compo- In contrast to the pseudo-one-component
nent system). Often the process can be pic- systems described above where the dif-
tured as the interdiffusion of vacant sites fusion rates of the atoms and vacant sites
and atoms. The diffusion coefficient de- are necessarily equal, in the substitutional
scribing this process is usually called the binary alloy the individual diffusion rates
chemical diffusion coefficient, sometimes of A and B are not generally equal since the
the interdiffusion coefficient, and occasion- corresponding self-diffusion coefficients
ally, the collective diffusion coefficient. are not. In the interdiffusion experiment
Generally the symbol used is D̃. For these this implies that there is a net flux of atoms
pseudo-one-component systems, the pre- across any lattice plane normal to the diffu-
ferred name is chemical diffusion coeffi- sion direction. If the number of lattice sites
cient. In general, the chemical diffusion co- is conserved, each plane in the diffusion re-
efficient does not equal the self-diffusion gion must then shift to compensate. This
coefficient because of effects arising from shift with respect to parts of the sample
the gradient of chemical composition, see outside the diffusion zone, say the ends of
Eq. (3-99). the sample, is called the Kirkendall effect.
Chemical diffusion in binary substitu- This effect can be measured by observing
tional solid solutions is frequently called the migration of inert markers, usually fine
interdiffusion. In a typical case pure metal insoluble wires which have been incorporat-
A is bonded to pure metal B and diffusion ed into the sample before the experiment.
is permitted at high temperature. Although The assumption is that the wires follow the
both A and B atoms move, only one con- motion of the lattice in their vicinity.
centration profile, say of A, is established The intrinsic diffusion coefficients of A
(the profile from B contains no new infor- and B, DAI and DBI , are defined with refer-
mation). The resulting diffusion coefficient ence to the fluxes of A and B relative to the
which is extracted from the profile, by the local lattice planes:
Boltzmann – Matano analysis (e.g., see Sec.
∂CA
3.2.1) is termed the interdiffusion coeffi- JA′ = − DAI (3-15)
cient and is given the symbol D̃. Not infre- ∂x
quently, the diffusion coefficient for this and
situation is loosely called the chemical dif- ∂CB
fusion coefficient or the mutual diffusion JB′ = − DBI (3-16)
∂x
coefficient. This single diffusion coeffi-
cient is sufficient to describe the concentra- The diffusion coefficients are sometimes
tion profile changes of the couple. Because also termed partial diffusion coefficients.
If v is the velocity (Kirkendall velocity) 3.2.2.5 Surface Diffusion Coefficients

of the lattice plane measured with respect
Surface diffusion refers to the motion of
to parts of the sample outside the diffusion
atoms, sometimes molecules, over the sur-
zone then the fluxes with respect to it are
face of some substrate. The diffusing spe-
JA = JA¢ + v CA (3-17) cies can be adsorbed atoms, e.g. impurity
metal atoms on a metal substrate (this is
and normally referred to as hetero-diffusion) or
JB = JB¢ + v CB (3-18) of the same species as the substrate (this is
usually referred to as self-diffusion).
where CA and CB are the concentrations at As a field, surface diffusion has evolved
the lattice planes. It should be noted that somewhat separately from solid-state dif-
the interdiffusion coefficient is measured fusion, perhaps because of the very differ-
in this frame of reference. Since ∂CA /∂x ent techniques employed. This separate-
= – ∂ CB /∂x, we easily deduce that ness has resulted in some inconsistencies in
nomenclature between the two fields. It is
∂CA
v = ( DAI − DBI ) (3-19) appropriate here to discuss these briefly.
∂x For both self- and hetero-diffusion it is
and usual to refer to the motion of atoms for
short distances, where there is one type of
∂CA
JA = − ( cA DBI + cB DAI ) (3-20) site, as “intrinsic diffusion”. When the con-
∂x centration of diffusion species is very low
where cA and cB are the mole fractions of the relevant diffusion coefficient is called
A and B. The term in parentheses in Eq. a “tracer” diffusion coefficient. This is not
(3-20) is the interdiffusion coefficient D̃ : in fact the diffusion coefficient obtained
with radioactive tracers but a single parti-
D̃ = cA DBI + cB DAI (3-21) cle diffusion coefficient. (A single particle
can be followed or traced.) At higher con-
The interdiffusion coefficient is seen to be
centrations, the relevant diffusion coeffi-
a weighted average of the individual (in-
cient is called a chemical diffusion coeffi-
trinsic) diffusion coefficients of A and B.
cient. Thus “intrinsic” diffusion is formally
With the aid of Eqs. (3-19) and (3-21) the
the same as diffusion in the pseudo-one-
intrinsic diffusion coefficients can be de-
component system, as described in Sec.
termined if D̃ and v are known. The compo-
3.2.2.3.
sition they refer to is the composition at the
When the surface motion of atoms ex-
inert marker. Note that if there is no marker
tends over long distances, and many types
shift, Eq. (3-19) implies that the intrinsic
of site are encountered (this is macroscopic
coefficents are equal. The relationship of
diffusion in contrast with the microscopic
the intrinsic diffusion coefficient to the
or “intrinsic” diffusion discussed above)
tracer (self) diffusion coefficient is ex-
the relevant diffusion coefficient is called
plored in Sec. 3.2.3.2.
a mass transfer diffusion coefficient. Fur-
From the form of Eq. (3-21) when
ther discussion of these and relationships
cA Æ 0, it can be seen that D̃ Æ DAI . It
among these diffusion coefficients are re-
should also be noted that in this limit DAI
viewed by Bonzel (1990).
reduces to the impurity diffusion coeffi-
cient of A in B.
3.2.3 Phenomenological Equations where Q k* is called the “heat of transport”

of Irreversible Thermodynamics for species k (Manning, 1968). This can
An implication of Fick’s First Law (Eq. be further related to the heat of transport
(3-1)) is that once the concentration gradi- derived from the actual heat flow. This
ent for species i reaches zero, all net flow subject is dealt with further by Manning
for species i stops. Although frequently (1968) and Philibert (1991). Sometimes it
correct, this is rather too restrictive as a is convenient to discuss the flux in an
condition for equilibrium. In general, net electric field occurring simultaneously with
flow for species i can cease only when all diffusion in a chemical potential gradient.
direct or indirect forces on species i are A simple example would be diffusion from
zero. This is conveniently handled by pos- a tracer source in an electric field – the so-
tulating linear relations between each flux called Chemla experiment (Chemla, 1956).
and all the driving forces. We have for flux In such a case, we write Xi as
i in a system with k components Xi = – grad m i + Z i e E (3-23)
J i = ∑ L i k X k + Li q X q (3-22) Now the phenomenological coefficient
k
L iq (Eq. (3-22)) describes the phenomenon
where the Ls are called phenomenological of thermal diffusion, i.e., the atom flow re-
coefficients and Xi is the driving force on sulting from the action of the Xq , i.e., the
component i and is written as – grad m i , temperature gradient (Soret effect). There
where m i is the chemical potential of spe- is also an equation analogous to Eq. (3-22)
cies i. Xi can also result from an external for the heat flow itself:
driving force such as an electric field. In
this case Xi = Z i e E, where Z i is the number Jq = ∑ Lqk Xk + Lq q Xq (3-24)
k
of changes on i, e is the electronic charge,
and E is the electric field. For ionic con- L qk is a heat flow phenomenological coef-
ductors Z i is the actual ionic valence. For ficient, i.e., the heat flow which accompa-
alloys Z i is an “effective” valence which is nies an atom flow (the Dufour effect). L qq
usually designated by the symbol Z i*. Z i* refers directly to the thermal conductivity.
consists of two parts, the first, Z iel , repre- The other phenomenological coeffi-
sents the direct electrostatic force on the cients, L ik in Eq. (3-22), are concerned
moving ion (Z iel is expected to be the ion’s with the atomic transport process itself.
nominal valence), and the second, Z iwd , The off-diagonal coefficients are con-
accounts for the momentum transfer be- cerned with “interference” between the at-
tween the electron current and the diffusing oms of different types. This may arise from
atom. A comprehensive review of all as- interactions between A and B atoms and/or
pects of Z i* and the area of electromigra- the competition of A and B atoms for the
tion has been given by Huntington (1975). defect responsible for diffusion, see Sec.
Xq is a driving force resulting from a tem- 3.3.1.8. A most important property of the
perature gradient (if present). Xq is given phenomenological coefficients is that they
by – T –1 grad T. When referring to diffu- are independent of driving force. Further,
sion in a temperature gradient it is usual to the matrix of L coefficients is symmetric:
let L iq be expressed as this is sometimes called Onsager’s theorem
n or the reciprocity condition. That is,
Li q = ∑ Qk* Li k L ij = Lji (3-25)
k =1
The set of equations represented by Eq. e.g., the tracer correlation factor and the
(3-22) are normally called the “phenomen- vacancy-wind effect, can be conveniently
ological equations” or Onsager equations expressed in terms of the Ls. This can
and were postulated as a central part of the sometimes aid in understanding the nature
theory of irreversible processes. Details of of these complex correlations. A most im-
the theory as it applies to diffusion can be portant contribution to this was made by
found in many places but see especially the Allnatt (1982), who related the Ls directly
reviews by Howard and Lidiard (1964), to the microscopic behavior – see Sec.
Adda and Philibert (1966), and Allnatt and 3.3.1.8.
Lidiard (1993). In the next sections we shall restrict our-
The phenomenological coefficients are selves to isothermal diffusion and focus on
sometimes said to have a “wider meaning” the relations which can be derived among
than quantities such as the diffusion coeffi- the Ls and various experimentally acces-
cients or the ionic conductivity. The wider sible transport quantities.
meaning comes about because the pheno-
menological coefficients do not depend on 3.2.3.1 Tracer Diffusion
driving force but only on temperature and
composition. In principle, armed with a full Let us first consider tracer diffusion in a
knowledge of the Ls, the technologist pure crystal. The general strategy is (1) to
would have the power, if not to control the describe the flux of the tracer with the On-
diffusive behavior of the material, at least sager equations (Eq. (3-22)) and (2) to de-
to predict the diffusive behavior no matter scribe the flux with Fick’s First Law (Eq.
what thermodynamic force or forces or (3-1)) and then equate the two fluxes to
combination thereof were acting. find expressions between the diffusion co-
Unfortunately, the experimental determi- efficient and the phenomenological coeffi-
nation of the Ls is most difficult in the solid cients. We shall deal with this in detail to
state. This is in contrast, incidentally, to show the typical procedure (see, for exam-
liquids, where, by use of selectively perme- ple, Le Claire (1975)). We can identify two
able membranes, measurement of the Ls is components, A and its tracer A*. We write
possible. The reader might well ask then for the fluxes of the host and tracer atoms,
why the Ls were introduced in the first respectively,
place when they are essentially not amen- JA = LAA XA + LAA* XA* (3-26)
able to measurement! There are several
and
reasons for this. First, relations can be de-
rived between the Ls and the (measurable) JA* = LA*A* XA* + LA*A XA (3-27)
tracer diffusion coefficients – we will dis-
Strictly, the vacancies should enter as a
cuss this further in Sec. 3.2.3.4. Further,
component but it is unnecessary here be-
the Onsager equations provide a kind of ac-
cause of the lack of a vacancy gradient.
counting formalism wherein an analysis
Since the tracer atoms and host atoms
using this formalism ensures that once the
mix ideally (they are chemically identical),
components and driving forces have all
it can easily be shown that the driving
been identified nothing is overlooked, and
forces are
that the whole is consistent; this will be
discussed in Secs. 3.2.3.1 and 3.2.3.2. k T ∂cA
XA = − (3-28)
Finally, many “correlations” in diffusion, cA ∂x
and an expression for the Haven ratio HR (see

k T ∂cA* also Sec. 3.3.2).
XA* = − (3-29) We consider the system of A and A* in
cA* ∂x
an electric field. We will assume that the
where ci is the mole fraction. This leads A and A* are already mixed. We will now
immediately to let XA = XA* = Z e E, where Z is the number
(3-30) of charges carried by each atom. From Eq.
k T ∂cA* k T ∂cA (3-27) the flux of A* is
JA* = − LA*A* − LA*A
cA* ∂x cA ∂x
JA* = ZA* e E (LA*A* + LA*A) (3-36)
It is convenient to introduce the quantity The drift mobility uA* is related to the flux
Ci (= N ci /V ) which is the number of atoms by
of type i per unit volume and N is the num-
ber of entities (A + A*), and V is the vol- JA* = CA* uA* E (3-37)
ume. Since there must be a symmetry con-
and so
dition ∂cA /∂x = – ∂ cA* /∂x, we now find that
ZA* eV ⎛ LA*A* + LA*A ⎞
∂C k TV ⎛ LA*A LA*A* ⎞
uA* = ⎜ ⎟ (3-38)
JA* = A* − N ⎝ cA* ⎠
⎜ ⎟ (3-31)
∂x N ⎝ cA cA* ⎠
The mobility uA is equal to uA* since A and
However, Fick’s First Law, Eq. (3-1), states A* are chemically identical.
that By means of the so-called Nernst – Ein-
stein equation (Eq. (3-130)), the mobility
∂CA*
JA* = − DA* (3-32) can be converted to a dimensionally correct
∂x diffusivity, Ds , i.e.,
and so the tracer diffusion coefficient is
kTV
given by Ds = ( LA*A* + LA*A ) (3-39)
N cA*
k TV ⎛ LA*A* LA*A ⎞
DA* = ⎜ − ⎟ (3-33) As is discussed in detail in Sec. 3.3.2, Ds
N ⎝ cA* cA ⎠ does not have a meaning in the sense of
When cA* Æ 0, which is the usual situation Fick’s First Law (Eq. (3-1)). Its meaning is
when tracers are used experimentally, we the diffusion coefficient of the assembly of
find that ions as if the assembly itself acts like a sin-
gle (hypothetical) particle. It is conven-
DA* = k T V LA*A* /cA* N cA* Æ 0 (3-34) tionally related to the tracer diffusion coef-
ficient by the Haven ratio HR which is de-
Since the fluxes JA and JA* are always
fined as:
equal but opposite in sign, then going
through the above procedure, but now for DA*
HR ≡ (3-40)
JA , leads to the relation Ds
LAA + LA*A LAA* + LA*A* Using the equation for DA* we find, after
= (3-35)
cA cA* letting cA* Æ 0, that
Now let us relate the diffusion coefficient LA*A*

HR = (3-41)
to the ionic conductivity in order to obtain LA*A* + LA*A
For a pure crystal and where the vacancy Following the same kind of procedure as
concentration is very low, HR can be iden- before (see Sec. 3.2.3.1), but recognizing
tified directly with the tracer correlation that generally A and B do not ideally mix,
factor f (see Sec. 3.3.2) and we find that rather than by Eq. (3-28), Xi is
now given by
LA*A*
HR = f = (3-42) k T ∂ci ⎛ ∂ ln g ⎞
LA*A* + LA*A Xi = − ⎜1 + ⎟ (3-49)
ci ∂x ⎝ ∂ ln c ⎠
Note that if LA*A = 0, i.e., there is no inter-
ference between tracer and host, then where g is the activity coefficient of either A
HR = 1. More generally, when the defect or B. We find that the diffusion coefficient
concentration is high and there are corre- of, say A, actually an intrinsic diffusion co-
lations in the ionic conductivity (see Sec. efficient (see Sec. 3.2.2.4), is given by
3.3.1.7), we find that k T V ⎛ LAA LAB ⎞ ⎛ ∂ ln g ⎞
DAI = ⎜ − ⎟ ⎜1 + ⎟ (3-50)
LA*A* N ⎝ cA cB ⎠ ⎝ ∂ ln c ⎠
HR = f / f I = (3-43)
LA*A* + LA*A If we want to find a relation between DAI
and f I is the physical correlation factor. and the tracer diffusion coefficient, say
DA* , we would need to introduce A* for-
mally into the phenomenological equations
3.2.3.2 Chemical Diffusion which would now become (Stark, 1976;
For chemical diffusion between A and B Howard and Lidiard, 1964)
when, say, the vacancy mechanism is oper- JA = LAA XA + LAA* XA* + LAB XB (3-51)
ating (see Sec. 3.3.1.1), the Onsager equa-
tions are written as (Howard and Lidiard, JA* = LA*A* XA* + LA*A XA + LA*B XB (3-52)
1964) JB = LBB XB + L BA XA + L BA* XA* (3-53)
JA = LAA XA + LAB XB (3-44) These equations can be developed to give
JB = LBB XB + L BA XA (3-45) k TV ⎛ LA*A* LA*A ⎞
DA* = ⎜ − ⎟ (3-54)
Strictly, the vacancies should enter here as N ⎝ cA* cA ⎠
another species, so that we would write and
JA = LAA XA + LAB XB + LAV XV (3-46) k TV ⎛ LAA + LA*A LAB + LA*B ⎞
DAI = ⎜ − ⎟
JB = LBB XB + L BA XA + L BV XV (3-47) N ⎝ cA cB ⎠
and, because of conservation of lattice ⎛ ∂ ln g ⎞

× ⎜1 + ⎟ (3-55)
sites, ⎝ ∂ ln c ⎠
JV = – (JA + JB ) (3-48) When these two equations are combined
The usual assumption is that the vacancies we find that
are always at equilibrium and XV = 0. For ⎛ ∂ ln g ⎞
this to happen the sources and sinks for va- DAI = DA* ⎜ 1 + ⎟ (3-56)
⎝ ∂ ln c ⎠
cancies, i.e., the free surface, dislocations,
or grain boundaries, must be effective and ⎡ cA* ⎛ LA*A LAB ⎞ ⎤
× ⎢1 + ⎜ − ⎟⎥
sufficiently numerous. ⎣ LA*A* cA ⎝ cA* cB ⎠ ⎦
This is in fact the so-called Darken equa- placement of the system of species i as if
tion (Darken, 1948), relating the intrinsic that system were a particle itself. The diag-
diffusion coefficient to the tracer diffusion onal phenomenological coefficients derive
coefficient but with the addition of the term from correlations of the system of species i
in brackets. This term is sometimes called with itself, i.e., self-correlations. This does
the “vacancy-wind-term” and does not not mean tracer correlations here: they are
vary much from unity. Note that it results correlations in the random walk of an atom.
from non-zero cross terms LA*A and LAB . The correlations here are in the random
The approach from the phenomenological walk of the system “particle”. The off-diag-
equations has told us only of the formal ex- onal terms derive from interference of the
istence of this term, but an evaluation of it system of species i with the system of spe-
requires consideration of the detailed mi- cies j. In effect it is mathematically equiva-
croscopic processes which generate the lent to two systems i and j being treated
correlations contained in the cross terms. like two interfering “particles”.
This is considered in Sec. 3.3.1.8. Eq. (3-57) has been very useful for cal-
These examples suffice to show how the culating the L ij by means of computer sim-
phenomenological equations are useful in ulation – see the pioneering calculations
presenting a consistent and unified picture by Allnatt and Allnatt (1984). Much of that
of the diffusion process, no matter how com- material has been reviewed by Murch and
plex. For further information on the subject, Dyre (1989). A brief discussion is given in
the reader is directed to the classic review by Sec. 3.3.1.8.
Howard and Lidiard (1964) and also many
other treatises, such as those by Adda and 3.2.3.4 Relating Phenomenological
Philibert (1966), Le Claire (1975), Stark Coefficients to Tracer Diffusion
(1976), Kirkaldy and Young (1987), Phili- Coefficients
bert (1991), and Allnatt and Lidiard (1993).
For the case of multi-component alloys,
Manning (1968, 1970, 1971) derived expres-
3.2.3.3 Einsteinian Expressions for sions relating the phenomenological coeffi-
the Phenomenological Coefficients cients to tracer diffusion coefficients. The ex-
An important development in the area of pressions were developed for a particular
solid-state diffusion was the fact that the model for concentrated alloys, the so-called
phenomenological coefficients can be ex- random alloy model. In this model the atomic
pressed directly in terms of atomistic Ein- components are randomly distributed over
steinian formulae (Allnatt, 1982). the available sites, the vacancy mechanism is
assumed and the exchange frequencies of the
Lij = lim lim (6 V k T t ) −1 atomic components with the vacancies de-
V→ ∞ t→ ∞
pend only on the nature of the atomic compo-
× 〈 DR (t ) ⋅ DR( j ) (t )〉
(i )
(3-57) nents and not on their environment.
where DR i (t) is the total displacement of For a binary system the relations are
species i in time t, V is the volume, and the N
Dirac brackets denote a thermal average A. LAA = cA DA* (3-58)
kT
It is important to note that DR(i) is the sum
⎡ (1 − f0 ) cA DA* ⎤
of the displacements of the individual par- × ⎢1 + ⎥
ticles of type i. In effect, DR(i) is the dis- ⎢⎣ f0 ( cA DA* + cB DB* ) ⎥⎦
N (1 − f0 ) cA DA* cB DB* A most important consequence of Eqs.

LAB = (3-59)
kT f0 ( cA DA* + cB DB* ) (3-58) and (3-59) is the Darken – Manning
expression which relates the interdiffusion
where f0 is the tracer correlation factor in coefficient D̃, the tracer diffusion coeffi-
the lattice of either pure component, see cients DA* and DB* and the thermodynamic
Sec. 3.3.1.3. The relation for L BB is ob- factor (1 + ∂ ln g /∂c):
tained from LAA by interchanging B with A.
Lidiard (1986) showed that Eqs. (3-58) ˜ = ( cB DA* + cA DB* ) ⎛⎜ 1 + ∂ ln g ⎞⎟ S (3-60)
D
and (3-59) can in fact be obtained without ⎝ ∂c ⎠
recourse to the random alloy model, on the where the vacancy wind factor S is given by
basis of two microscopic assumptions
which, although intuitive in nature, are in- (1 − f0 ) cA cB ( DA* − DB* )2
S =1+
dependent of any microscopic model ex- f0 (cA DA* + cB DB* ) (cB DA* + cA DB* )
cept for the inclusion of the tracer correla- (3-60 a)
tion factor in the pure lattice f0 , which is
dependent on mechanism and structure. Eq. (3-60) is not only appropriate for
Lidiard’s findings immediately suggested random alloys, but also for ordered alloys/
that Eqs. (3-58) and (3-59) are probably ap- intermetallic compounds with B1 and B2
propriate for a much wider range of alloys structures. Recent computer simulations
than are reasonably represented by the ran- have also shown that Eq. (3-60) is a good
dom alloy model itself. Bocquet (1987) approximation for ordered alloys/intermet-
also found relations of the same form for allic compounds with L12 , D03 , and A15
the random alloy when interstitial mecha- structures (Murch and Belova, 1998).
nisms are operating. With Monte Carlo Further discussion on relations between
computer simulation, Zhang et al. (1989 a) phenomenological coefficients and tracer
and Allnatt and Allnatt (1991) explored the diffusion coefficients, and impurity diffu-
validity of Eqs. (3-58) and (3-59) in the sion coefficients, can be found in the re-
context of the interacting binary alloy views by Howard and Lidiard (1964), Le
model described in Sec. 3.3.1.5. It was Claire (1975), and Allnatt and Lidiard
found that Eqs. (3-58) and (3-59) apply (1987, 1993).
very well except, perhaps surprisingly, at
compositions approaching impurity levels.
3.2.4 Short-Circuit Diffusion
The breakdown there is readily traced to
the fact that these equations have built into It is generally recognized that the rate at
them the requirement that w4 = w0 (see Sec. which atoms migrate along grain boundar-
3.3.1.4 for the impurity frequency nota- ies and dislocations is higher than that
tion). For many cases w3 also equals w0 through the lattice. From an atomistic point
and so this is equivalent to a condition of of view it is very difficult to discuss, in a
no vacancy-impurity binding. This is not precise way, the diffusion events in such
surprising given that the equations origi- complex and variable situations. Much of
nated with the random alloy model where the understanding has been gained from
this requirement is automatically fulfilled. computer simulations using the Molecular
Eqs. (3-58) and (3-59) have also been Dynamics and Lattice Statics methods;
derived for B1 and B2 ordered alloys (Be- see, for example, Kwok et al. (1984) and
lova and Murch, 1997). Mishin (1997). In this section we shall dis-
cuss short-circuit diffusion from a pheno- volume diffusion, then this condition
menological point of view. D¢/D > 106 is usual for temperatures below
Fisher (1951) was probably the first to about half the melting temperature. This is
introduce a distinct diffusion coefficient why experiments where lattice diffusion
for material migrating along the short-cir- only is of interest tend to be made at tem-
cuit path. For grain boundaries, each one is peratures above about half the melting tem-
conceived as a slab of width d for which perature.
the average diffusion coefficient is D¢. For Impurities are often bound to short-cir-
dislocations, each one is conceived as a cuit paths, in which case the Hart Equation
pipe of diameter 2 a for which the average (Eq. (3-60)) is written as
diffusion coefficient is D¢.
D¢ = ᑠ D¢ exp (Eb /k T ) + (1 – ᑠ ) Dl (3-62)
For conventional diffusion experiments
it has been usual to distinguish three dis- where Eb is the binding energy of the im-
tinct regimes, depending on the magnitude purity and the diffusion coefficients refer
---
of the lattice diffusion length ÷Dl t, where to impurities. The Hart Equation can be
Dl is the lattice diffusion coefficient and t is shown to be fairly accurately followed
--- ---
time (Harrison, 1961). when l /÷Dl t ≤ 0.3 and also for l /÷Dl t ≥ 100
---
Regime A: ÷Dl t is much larger than the (this corresponds to Regime C – see the
spacing l between the short-circuit paths. next paragraph) (Murch and Rothman,
For grain boundaries l is the average diam- 1985; Gupta et al., 1978).
eter of the grains and for dislocations l is Regime C: It is convenient to discuss the
---
the distance between dislocation pinning other limit out of sequence. When ÷Dl t is
points. Diffusion from adjacent short cir- much smaller than the distances l between
cuits overlap extensively. This condition the short-circuit paths, we have Regime C
is met for small-grained materials or very kinetics. In this instance, all material
long diffusion times. Hart (1975) proposed comes down the short-circuit paths and the
that an effective diffusion coefficient Deff measured diffusion coefficient is given en-
can be introduced which still satisfies solu- tirely by D¢.
tions to Fick’s Second Law and can be Regime B: In this intermediate case, it is
---
written as assumed that ÷Dl t is comparable to l so
that material which is transported down a
Deff = ᑠ D¢ + (1 – ᑠ ) Dl (3-61)
short-circuit path and which diffuses out
where ᑠ is the fraction of all sites which into the lattice is unlikely to reach another
belong to the short-circuit paths. Eq. (3-61) short-circuit path.
is usually called the Hart Equation. Le There are various solutions of Fick’s
Claire (1975) has given a rough calculation Second Law available to cope with tracer
using Eq. (3-61). For typical dislocation diffusion in the presence of short-circuit
densities in metals, ᑠ = 10– 8. The disloca- paths, but space prevents us from dealing
tion contribution to a measured or effective with these in any detail: they are discussed
D will then exceed 1% when D¢/D > 106. at length by Kaur, Mishin and Gust (1995).
Because the activation energy for lattice Suzuoka (1961) and Le Claire (1963) have
diffusion in metals is about 34 TM. Pt. , given a solution for the grain boundary
where TM.Pt. is the melting temperature problem for the usual case when there is a
and the activation energy for short-circuit finite amount of tracer originally at the sur-
diffusion is generally about half that of face, see Eq. (3-8). It is found, among other
things, that and provided near-exact solutions to Fick’s

5/ 3 Second Law for both an isolated disloca-
⎛ d ln C ⎞ 4 Dl ⎞ 1/ 2
=⎛ ( D′ d ) −1 0.66 (3-63) tion pipe and arrays of dislocation pipes.
⎝ dx 6 / 5 ⎠ ⎝ t ⎠ The profiles generally are not unlike the
Hence the product D¢ d can be found by de- grain boundary ones except that a linear re-
termining the slope of the linear region gion in a plot of ln C versus x is now found,
(penetrations reached by grain boundary i.e.,
diffusion) in a plot of ln C versus x 6/5 (not d ln C A (a)
the usual x 2 ) and also with a knowledge of =− (3-64)
dx [( D′ / Dl − 1) a 2 ]1/ 2
Dl itself. It can be difficult, however, to ob-
tain the two diffusion coefficients in one where A is a slowly varying function of
experiment; the practicalities of this and a/(D t)1/2. An example of such a tracer pen-
alternatives are discussed by Rothman etration plot with a clear contribution from
(1984). An example of a tracer penetration dislocation pipe diffusion (x dependence)
plot with a clear contribution from grain is shown in Fig. 3-4.
boundary diffusion (x 6/5 dependence) is For further details on grain boundary
shown in Fig. 3-3. diffusion we refer to reviews by Peterson
Le Claire and Rabinovitch (1984) have (1983), Mohan Rao and Ranganathan
addressed the dislocation pipe problem (1984), Balluffi (1984), Philibert (1991),
Figure 3-3. Tracer concentration

profiles of 203Pb into polycrystal-
line Pb showing a contribution
from grain boundaries (Gupta and
Oberschmidt, 1984).
3.3 Microscopic Diffusion 189
erence to the discrete atomic events which

give rise to macroscopically observable
diffusion. To describe the atomic events, a
wholly separate theory, termed “Random
Walk Theory” has been formulated. This
theory is based on the premise that macro-
scopic diffusion is the net result of many
individual atomic jumps. The theory at-
tempts to relate the quantities such as the
diffusion coefficients, the ionic conductiv-
ities, or, more generally, the L ij , in terms of
lattice and atomic characteristics, notably
jump frequencies. Although originally
quite precise in use (Howard, 1966), the
term “Random Walk Theory” is now used
rather loosely to describe any mathematical
approach that focuses on the sequence of
jumps of atoms in the solid state. Inextri-
cably linked with this theory is the exis-
tence of correlated random walks of atoms,
Figure 3-4. Tracer concentration profiles of 22Na
into single crystal NaCl showing a contribution from in other words, walks where there is an ap-
dislocations (Ho, 1982). parent memory between jump directions.
Much of random walk theory is concerned
with describing such correlations.
Kaur et al. (1995), Mishin et al. (1997). For
dislocation pipe diffusion we refer to Le 3.3.1.1 Mechanisms of Diffusion
Claire and Rabinovitch (1984) and Phili-
bert (1991). As a prelude to a discussion of random
walk theory, in this section we will briefly
discuss the common mechanisms of solid-
state diffusion.
3.3 Microscopic Diffusion
3.3.1 Random Walk Theory Interstitial Mechanism
We have already seen in Sec. 3.2 that the In the interstitial mechanism, see Fig.
phenomenological or continuum theory of 3-5, sometimes called the direct intersti-
diffusion is set up in terms of quantities tial mechanism, the atoms jump from one
such as the diffusion coefficients, the ionic interstice to another without directly in-
conductivities, or more generally, the phen- volving the remainder of the lattice. Since
omenological coefficients, L ij . For the ma- the interstitial atom does not need to “wait”
terials technologist this theory provides a to be neighboring to a defect in order to
perfectly suitable formalism to describe the jump (in a sense it is always next to a va-
macroscopic features arising from the dif- cancy), diffusion coefficients for atoms mi-
fusion of atoms in solids. By its very na- grating with this mechanism tend to be
ture, this macroscopic theory makes no ref- fairly high. As would be expected, atoms
small and their contribution to diffusion is

unimportant in most cases. However, for
plastically deformed or irradiated metals
the concentration of interstitials (besides
vacancies) can be appreciable. It should be
Figure 3-5. Interstitial diffusion mechanism. noted that the interstitial thus formed is not
located on an interstice, but in a dumbbell
split configuration. The migration possibil-
such as H, N, O, and C diffuse in this way ities of dumbbells have been discussed by
in metals. It should also be noted that with- Schilling (1978).
out the defect being required to affect the As a result of measurements of the
jump, no defect concentration term and Haven ratio (see Sec. 3.3.2), the interstis-
therefore defect formation energy enters tialcy mechanism appears to be highly
the activation energy for diffusion; see Sec. likely for silver diffusion in AgBr (Friauf,
3.3.4. 1957). Note that with the interstitialcy
mechanism, say the collinear version, a
Interstitialcy Mechanism tracer atom moves a distance r whereas the
charge apparently moves a distance 2 r.
In the interstitialcy mechanism, see Fig. This needs to be taken into account in the
3-6, sometimes called the indirect intersti- interpretation of the Haven ratio. The inter-
tial mechanism, two atoms, one an intersti- stitialcy mechanism probably occurs rela-
tial and the other an atom on a regular lat- tively frequently in ionic materials, espe-
tice site, move in unison. The interstitial cially those with open lattices such as the
atom moves to a regular site, whereas the anion lattice in the fluorite structure or
regular site atom moves to an interstice. those which are highly defective. Nonethe-
Collinear and non-collinear versions are less, there is often so much attendant local
possible depending on constraints imposed relaxation around the interstitial that far
by the lattice. more complex quasi-interstitialcy mecha-
Because the interstitial formation energy nisms could well operate. An example is
is generally very high in metals, the equi- oxygen diffusion in UO2 + x , where the ex-
librium concentration of interstitials is very cess oxygen is located as di-interstitials
with relaxation of two oxygen atoms from
regular sites to form two new interstitials
and two new vacancies. The actual mecha-
nism of oxygen transport is not known
but a quasi-interstitialcy mechanism is un-
doubtedly responsible (Murch and Catlow,
1987).
We should also mention that the interstit-
ialcy mechanism appears to be very impor-
tant in self-diffusion in silicon and possibly
certain substitutional solutes in silicon are
also transported via this mechanism (Frank
et al., 1984).
Figure 3-6. Interstitialcy diffusion mechanism.
Vacancy Mechanism utors to tracer diffusion in alkali halides

(Bénière et al., 1976).
The most important diffusion mecha-
nism of all is undoubtedly the vacancy
Interstitial-Substitutional Mechanism
mechanism, shown in Fig. 3-7. A substitu-
tional solute atom or an atom of the solvent On occasion, solute atoms may dissolve
diffuses simply by exchanging places with interstitially and substitutionally. These
the vacancy. There can be attractive or re- solute atoms may diffuse by way of the
pulsive interactions of the solute with the dissociative mechanism (Frank and Turn-
vacancy, which can have a profound in- bull, 1956) and/or the kickout mechanism
fluence on the diffusion coefficient of the (Gösele et al., 1980). In both mechanisms
solute and to a lesser extent the solvent the interstitial solute diffuses rapidly by the
by way of correlation effects. This is dis- interstitial mechanism. In the dissociative
cussed in Secs. 3.3.1.4 and 3.4.2. The va- mechanism the interstitial combines with a
cancy mechanism is definitely the pre- vacancy to form a substitutional solute. At
ferred mechanism for metals and alloys for a later time this substitutional can dissoci-
both host and substitutional solutes. In ate to form a vacancy and an interstitial so-
most other materials the vacancy appears lute (really a Frenkel defect). The anoma-
to play the major diffusion role except lously fast diffusion of certain solutes, e.g.,
when the concentration of interstitials pro- Cu, Ag, Au, Ni, Zr, and Pd in Pb, appears
duced by nonstoichiometry or doping (in to have a contribution from the dissocia-
oxides) or irradiation (in metals) is so high tive mechanism (Warburton and Turnbull,
that contributions from interstitialcy or 1975; Bocquet et al., 1996). In the kickout
similar mechanisms become dominant. mechanism, on the other hand, the solute
Aggregates of vacancies such as the interstitial uses the interstitialcy mecha-
divacancy or trivacancy can also contribute nism to involve the regular site lattice. In
to diffusion. These appear to be important the process a host interstitial is formed and
at high temperatures in metals where their the interstitial solute then occupies a sub-
contribution has been largely inferred from stitutional site. This process can be re-
curvatures of the Arrhenius plot (log D vs. versed at a later stage. The kickout mecha-
1/T ); see, for example, the f.c.c. metals nism appears to operate for rapid diffusion
(Peterson, 1978) (see Sec. 3.4.1.1). Bound of certain foreign atoms, such as Au in Si
vacancy pairs, i.e., a cation vacancy bound (Frank et al., 1984).
to an anion vacancy, are important contrib-
Exchange Mechanism
The exchange mechanism, in which two
neighboring atoms exchange places, has in
the past been postulated as a possible diffu-
sion mechanism. The existence of the Kir-
kendall effect in many alloy interdiffusion
experiments (this implies that the respec-
tive intrinsic diffusion coefficients are un-
equal, which is not possible with the ex-
change mechanism, see Sec. 3.2.2.4) and
Figure 3-7. Vacancy mechanism. the very high theoretical activation ener-
gies in close-packed solids suggest that this plus higher order terms. In Eq. (3-65), C (x0 )
mechanism is unlikely. It may, however, is the concentration of the diffusing atoms
occur in liquids and in quenched-liquids at x0 , ·XÒ is the mean displacement or drift,
such as amorphous alloys (Jain and Gupta, and ·X 2 Ò is the mean squared displacement.
1994). Ring versions of the exchange The Dirac brackets · Ò denote an average
mechanism certainly have lower theoreti- over all possible paths taken in time t.
cal activation energies but require substan- In a situation where diffusion properties
tial cooperation among the atoms, which do not depend on position, e.g., diffusion
seems unlikely. of tracer atoms in a chemically homogene-
ous system, the third term in Eq. (3-65) is
Surface Diffusion Mechanisms zero. The term containing the drift ·XÒ is
A number of mechanisms for surface also zero because in a chemically homoge-
diffusion on metals (by far the most studied neous system the probability that, say, a
class of material) have been postulated. tracer atom migrates some distance + X in
They include activated hopping of ad- time t (starting from x0 ) equals the prob-
sorbed atoms from one surface site to an- ability that an atom migrates a distance – X.
other, where the jump distance is simply Similarly, other odd moments are zero.
the distance between sites. Similarly, a va- Provided that other higher-order terms can
cancy in a terrace, i.e., surface vacancy, be neglected, Eq. (3-65) reduces to
can also move in the same sort of way ∂C 〈 X 2 〉
within the terrace. At low temperatures and Jx = (3-66)
∂x 2 t
rough surfaces, exchange between an ad-
sorbed atom and an atom in the substrate is By comparison with Fick’s First Law, Eq.
predominant. At high temperatures non-lo- (3-1), Eq. (3-66) immediately gives
calized diffusion and surface melting are
also possible. The mechanisms for surface 〈X2 〉
Dx* = (3-67)
diffusion are discussed further in the book 2t
edited by Vu Thien Binh (1983) and in the Eq. (3-67) is called the Einstein Equation,
reviews by Rhead (1989) and Bonzel probably the single most important equa-
(1990). tion in the theory of diffusion kinetics. The
superscript * indicates that the diffusion
3.3.1.2 The Einstein Equation coefficient refers to tracer atom diffusion
Let us consider atoms (of one chemical in a chemically homogeneous system.
type) diffusing in their concentration gradi- Equations for Dy* and Dz* have the same
ent in the x direction. It can readily be form as Eq. (3-67). For three-dimensional
shown (see, for example, Adda and Phili- isotropic crystals the tracer diffusion coef-
bert, 1966; Manning, 1968; Le Claire, ficient is the same in every direction and
1975) that the net flux of atoms across a
〈 R2 〉
given plane x0 is given by D* = (3-68)
6t
〈 X 〉 ∂C 〈 X 2 〉
J x = C ( x0 ) − where R is the vector displacement of an
t ∂x 2 t atom in time t. For two-dimensional situa-
∂C ∂ 〈 X 2 〉 tions the factor 6 in Eq. (3-68) is replaced
− C ( x0 ) (3-65)
∂x ∂C 2 t by 4.
3.3.1.3 Tracer Correlation Factor

Either Eq. (3-67) or (3-68) can provide
the starting point for a discussion of corre-
lation effects in the diffusion walk. For
convenience we shall use Eq. (3-68). Let us
consider the atomistic meaning of the dis-
placement R. It is simply the sum of n suc-
cessive jump vectors, r1 , r2 , r3 , …, rn
n
R = ∑ ri (3-69)
i =1
The squared displacement R2 is then simply

n n −1 n − i
R2 = ∑ ri2 + 2 ∑ ∑ ri ⋅ ri + j (3-70)
i =1 i =1 j =1
The average of the squared displacement

·R2 Ò equals of course the sum of the aver-
ages and we have
n n −1 n − i
〈 R2 〉 = ∑ 〈 ri2 〉 + 2 ∑ ∑ 〈 ri ⋅ ri + j 〉 (3-71)
i =1 i =1 j =1
Figure 3-8. (a) Correlation effects arising from the
For a complete random walk, i.e., where vacancy mechanism – see text. (b) Correlation ef-
each direction is unrelated to the previous fects arising from the interstitialcy mechanism (see
one, the second term in Eq. (3-71) is zero text).
because for any product ri · ri + j there will
always be another of opposite sign. We
now have probability of doing this is in fact exactly
n 2/g where g is the lattice coordination
〈 R2 〉 = ∑ 〈 ri2 〉 (3-72) (Kelly and Sholl, 1987). On the other hand,
i =1 the tracer has a reduced probability of con-
Because in most cases atoms require the tinuing in the direction of the first jump,
assistance of point defects in order to move since this requires the vacancy to migrate
about (see Sec. 3.3.1.1), there is generally to point A. In other words, there is a mem-
an unavoidable memory or correlation ef- ory or correlation between directions of
fect between jump directions. In order to tracer jumps. In no way does this imply
appreciate this, let us focus on Fig. 3-8 a that the vacancy somehow favors the tracer
where the vacancy mechanism is shown. In atom. In fact, in this example atoms sur-
this figure one atom symbol is shown filled rounding the vacancy jump randomly with
to indicate that it is a tracer and can be fol- the vacancy so that the vacancy itself
lowed. Let us assume that the tracer and the moves on an uncorrelated random walk
vacancy have just exchanged places. Be- with no memory or correlation whatsoever
cause the vacancy is still neighboring to the between its jump directions. It should be
tracer the next jump of the tracer is quite noted that weaker correlations in the tracer
likely to cancel out the previous jump. The jumps also come about as long as the va-
cancy remains in the vicinity of the tracer It should be noted that the limit n Æ • is
and can approach the tracer from a non- applied to Eq. (3-74) to ensure that all pos-
random direction. sible correlations are included.
Let us consider Fig. 3-8 b, where the col- There have been numerous publications
linear interstitialcy mechanism is assumed. concerned with the calculation of the tracer
Let us assume that the tracer is in the inter- correlation factor. The earlier work has been
stitial position. The first jump (a pair of reviewed in detail by Le Claire (1970),
atoms move) will take the tracer immedi- more recent work has been covered by All-
ately to a regular lattice site and a new natt and Lidiard (1993). Table 3-1 gives
atom, formerly at a regular lattice site, some values of f for various mechanisms
takes up a position as the interstitial. The and lattices. It should be noted from Table
pair of atoms which next moves possibly 3-1 that the correlation factor for the inter-
again involves the tracer atom and in doing stitial mechanism is unity. For this mecha-
so will cancel out the previous tracer jump. nism the interstitials, which are considered
Like the vacancy mechanism, the memory to be present at a vanishingly small concen-
or correlation between successive tracer tration, move on an uncorrelated random
atom jump directions comes about purely walk, very much like the vacancy in the va-
because of the proximity of the defect. Un- cancy-assisted diffusion mechanism. Accord-
like the vacancy diffusion mechanism, ingly, the second term in Eq. (3-71) drops out
however, in this case the correlation comes and f = 1. If the interstitial concentration is
about only for consecutive pairs of tracer increased, the interstitials impede one an-
jumps of the type interstitial site Æ regular other and correlations are introduced; as a
lattice site Æ interstitial site. result, the tracer correlation factor de-
In the presence of correlation between creases from unity. In fact, it continues to
jump directions, the sum of the dot prod- decrease until only one interstice is left va-
ucts ·ri · ri + j Ò no longer averages out to cant. The situation now is identical to that
zero. A convenient way of expressing these for vacancy-assisted diffusion. The varia-
correlations quantitatively is to form the tion of f with vacancy (or vacant interstice)
ratio of the actual ·R2 Ò to the ·R2 Ò result- concentration in the f.c.c. lattice is shown
ing from a complete random walk, i.e., in Fig. 3-9. Of course, in the unphysical sit-
·R2 Òrandom : uation where the interstitials do not “see”
one another, i.e., multiple occupancy of a
〈 R2 〉 site is permitted, the interstitials continue
f= (3-73)
〈 Rrandom
2
〉 to move on a complete random walk at all
concentrations, and f always equals unity.
where the quantity f is called the tracer cor-
For cubic lattices all the jumps are of the
relation factor or simply the correlation
same length. Then we have that | ri | = r :
factor. Sometimes the terminology “corre-
n
lation coefficient” is used, but this is to be
discouraged. With Eqs. (3-71) and (3-72) ∑ 〈 ri2 〉 = n r 2 (3-75)
i =1
we have
n −1 n − i Then we have, with Eqs. (3-68), (3-71),
2 ∑ ∑ 〈 ri ⋅ ri + j 〉 and (3-75)
i =1 j =1
f = lim 1 + (3-74)
n→∞ n n r2 f G r2 f
∑ 〈 ri2 〉 D* =
6t
=
6
(3-76)
i =1
Table 3-1. Some correlation factors (at infinitely low defect concentrations) from Le Claire (1970), Manning
(1968), and Murch (1982 d).
Lattice Mechanism f
Honeycomb Vacancy 1/3

Square planar Vacancy 1/(p – 1)
Triangular Vacancy 0.56006
Diamond Vacancy 1/2
B.c.c. Vacancy 0.72714
Simple cubic Vacancy 0.65311
F.c.c. Vacancy 0.78146
F.c.c. Divacancy 0.4579 ± 0.0005
All lattices Interstitial 1
NaCl structure Collinear interstitialcy 2/3
CaF2 structure (F) Non-collinear interstitialcy 0.9855
CaF2 structure (Ca) Collinear interstitialcy 4/5
CaF2 structure (Ca) Non-collinear interstitialcy 1
The tracer correlation factor itself can be

expressed as
〈 R2 〉
f = lim (3-77)
n→∞ n r2
This equation has been very useful for di-
rect computer simulation calculations of f;
see the review by Murch (1984 a).
We see that f normally acts to decrease
the tracer diffusion coefficient from its ran-
dom walk ( f = 1) value. The inclusion of f
in the expression for D* is necessary for a
Figure 3-9. Tracer correlation factor vs. vacancy
concentration for non-interacting vacancies in the
complete description of the atomic diffu-
f.c.c. lattice; after Murch (1975). sion process. From Table 3-1, however, it
can be seen that for many 3 D lattices f
really only decreases D* by some 20 –
where G is the jump frequency. The jump 30%. This is not much more than the preci-
frequency is further discussed in Sec. 3.3.4. sion routinely obtainable in measurements
This partitioning of the diffusion coeffi- of the tracer diffusion coefficient (Roth-
cient into its uncorrelated (G r 2/6) and cor- man, 1984) and, given the difficulty in cal-
related ( f ) parts is very basic to random culating the jump frequency G (see Sec.
walk theory. Other partitionings are cer- 3.3.4), may not appear to be particularly
tainly possible, e.g., only jumps which are significant. There are, however, many rea-
not immediately cancelled contribute to the sons why a study of correlation effects in
jump frequency, i.e., an “effective jump” diffusion is sufficiently important that it
frequency, but the partitioning here seems has consumed the energies of many re-
to be the most natural. searchers over almost a 40-year period.
First, as can be seen from Table 3-1, the impurity. We shall deal with this subject in
tracer correlation factor is quite sensitive to the following section.
the mechanism of diffusion operating. Al-
though f, by itself, is not measurable, mea- 3.3.1.4 Impurity Correlation Factor
surements of the Haven ratio (see Sec.
We first consider that the impurity con-
3.3.2) and the isotope effect (see Sec. 3.3.3)
centration is sufficiently dilute that there is
(which are closely related to f ) in favorable
not a composition-dependent impurity dif-
cases can throw considerable light on the
fusion coefficient. Accordingly, each im-
mechanism(s) of diffusion that are operat-
purity atom is considered to diffuse in pure
ing. Identification of the diffusion mecha-
host. The jump frequency of the impurity is
nism is surely the most important ingredi-
given by w 2 whereas the jump frequency of
ent in understanding the way atoms mi-
the host atoms is given by w 0 . The signifi-
grate in solids and how it can be controlled.
cance of the subscripts will be apparent
Much of our discussion so far has been
later. The vacancy mechanism is assumed.
concerned with pure solids with few de-
The “tracer” now is the impurity. The
fects. There are, however, a large number
result for the impurity correlation factor f 2
of solids where the apparent defect concen-
in the f.c.c. lattice as a function of w 2 /w 0
tration can be fairly high, e.g., highly non-
is shown in Fig. 3-10. When w 2 < w 0 (upper
stoichiometric compounds, fast ion con-
curve), the impurity motion becomes
ductors, and certain concentrated intersti-
considerably decorrelated since, after an
tial solid solutions. In such cases the de-
impurity/vacancy exchange, the vacancy
fects interact and correlation effects tend to
does not trend to remain in the vicinity of
be magnified and become highly tempera-
the impurity. When the impurity is next
ture dependent. Then the apparent activa-
tion energy for tracer diffusion includes an
important contribution directly from the
correlation factor. Clearly, an understand-
ing of f is a very important part of the
understanding of the diffusion process in
such materials. We shall deal further with
this subject in Sec. 3.3.1.6. In ordered
binary alloys, the correlation factor can
become very small since atoms which
make a jump from the “right” lattice to the
“wrong” lattice tend to reverse, i.e., cancel
that jump. As a result, the tracer diffusion
coefficient can become abnormally small,
largely because of this strong memory ef-
fect. We shall deal further with this subject
in Sec. 3.3.1.5.
Another important case of correlation ef-
Figure 3-10. Impurity correlation factor in the f.c.c.
fects is impurity diffusion. The correlation
lattice as a function of w 2 /w 0 : points by computer
factor for the impurity is highly dependent simulation (Murch and Thorn, 1978), solid lines
on the relative jump frequencies of the im- from the formalism of Manning (1964); after Murch
purity and host atoms in the vicinity of the and Thorn (1978).
approached by the vacancy, it will tend

to be from a random direction, with the
result that the impurity correlation factor
f 2 Æ 1.0, thereby signifying a less corre-
lated random walk.
Conversely, when w 2 > w 0 (lower curve),
the impurity and vacancy tend to continue
exchanging places in a particular configu-
ration. Many impurity jumps are thereby
effectively cancelled and f 2 Æ 0.0, i.e.,
the jumps are more correlated. Of course,
when w 2 = w 0 the “impurity” is now a
tracer in the host, and in effect it can be
considered to have a different “color” from Figure 3-11. Five-frequency model for impurity dif-
fusion in the f.c.c. lattice (see text), figure taken from
the host. The impurity correlation factor
Murch and Thorn (1978).
now is the tracer correlation factor, which
in this example is 0.78146 (see Table 3-1).
The diffusion coefficient can be consid- nearest neighbor position of the impurity,
ered to be the product of an uncorrelated i.e., the reverse of w 3 . All other host jumps
and a correlated part (Eq. (3-76)). In this occur with the frequency w 0 .
example, when w 2 > w 0 , the reduced im- We have mentioned “associative” and
purity correlation factor acts to reduce the “dissociative” jumps of the vacancy. These
impurity diffusion coefficient from that ex- are directly related to the impurity – va-
pected on the basis of w 2 alone. Similarly, cancy binding energy EB at the first nearest
when w 2 < w 0 the raised impurity correla- neighbor separation by
tion factor acts to increase the impurity dif- w 4 /w 3 = exp (– EB /k T ) (3-78)
fusion coefficient from that expected on
the basis of w 2 alone. Note that EB is negative for attraction
In general, the presence of the impurity between the impurity and the vacancy.
atom in the host also influences the host Manning (1964) has shown that the im-
jump frequencies in the vicinity of the impurity correlation factor f 2 is given rigor-
purity so that they differ from w 0 . For the ously for the five-frequency model by
f.c.c. lattice the usual model adopted is the f 2 = (2 w1 + 7 F w 3 )/(2 w 2 + 2 w1 + 7 F w 3 )
so-called five-frequency model and is de- (3-79)
picted in Fig. 3-11. The frequency w1 is
where F is the fraction of dissociating va-
the host frequency for host atom/vacancy
cancies that are permanently lost from a
jumps that are both nearest neighbors to the
site and are uncompensated for by return-
impurity. This jump is often called the “ro-
ing vacancies. F is given to a very good ap-
tational jump” because in effect the host can
proximation by
rotate around the impurity. The frequency
w 3 refers to a “dissociative” jump, i.e., a 7F = 7 (3-80)
host atom jump which takes the vacancy ⎡ 10 a + 180.5a + 927 a + 1341a ⎤
4 3 2
away from the impurity. Finally w 4 (not −⎢ 4 ⎥
⎣ 2 a + 40.2 a + 254 a + 597 a + 436 ⎦
3 2
shown) is the “associative” jump, which is
a host jump that brings the vacancy to the and a = w 4 /w 0 .
Other more accurate expressions for F nearest neighbor by means of the assump-
have been reviewed by Allnatt and Lidiard tions w 3 = w 3¢ = w ≤3 and w 5 = w 6 = w 0 . This
(1987, 1993). For many practical purposes also implies that w 4 = w 4¢ = w ≤4 .
the extra rigor is probably unwarranted. Model I leads to the expression (Man-
Extension to the case where there are va- ning, 1964; Le Claire, 1970)
cancy – impurity interactions at second
7w 3′ F
nearest neighbors has been considered by f2 = (3-84)
Manning (1964). 2 w 2 + 7w 3′ F
In the b.c.c. lattice the second nearest with F given by
neighbor is close to the nearest neighbor
(within 15% in fact). In a formal sense we 2 a 2 + 5.1817 a + 2.476
7F = (3-85)
need to consider dissociative jumps for a a + 0.8106
vacancy, jumping from the first to the sec- and a = w 3 /w 4¢ .
ond, third (not the fourth), and fifth nearest Model II leads to the expression
neighbors: these are denoted by w 3 , w 3¢ ,
and w ≤3 . 7w 3 F
f2 = (3-86)
The corresponding reverse associative 2 w 2 + 7w 3 F
jumps are w 4 , w 4¢ , and w ≤4 . We also need to
where F is given by
consider dissociative jumps from the sec- (3-87)
ond nearest neighbor to the fourth (w 5 ) and 3a + 33.43a + 97.38 a + 66.06
3 2
the reverse of this (w 6 ). There is no rota- 7F =

a 3 + 8.68 a 2 + 18.35a + 9.433
tional, i.e., w1 , jump.
The binding energy EB1 between the im- and a = w 4 /w 0 .
purity and the vacancy at first nearest Correlation factors for impurity diffu-
neighbor is expressible as (see, for exam- sion via vacancies in the simple cubic lat-
ple, Bocquet et al., 1996) tice (Manning, personal communication,
cited by Murch, 1982 a) and the diamond
w 4¢ /w 3¢ = w ≤4 /w ≤3 = exp (– EB1 /k T ) (3-81) structure have also been calculated (see
The binding energy EB2 between the im- Manning, 1964), as well as the h.c.p. struc-
purity and the vacancy at second nearest ture (see Huntington and Ghate (1962) and
neighbors can be expressed as Ghate (1964)). Correlation factors for im-
purity diffusion by interstitialcy jumps
w 6 /w 5 = exp (– EB2 /k T ) (3-82) have also been reported for the f.c.c. lattice
We also have that and the AgCl structure (Manning, 1959).
In many cases of impurity diffusion the
w 4 w 4 /w 5 w 3 = w 4¢ /w 3¢ (3-83) expression for the impurity correlation fac-
In order to proceed it has been usual to re- tor is of the form
duce the large number of jump frequencies u
f2 = (3-88)
by making certain assumptions. There are w2 + u
two basic models. In the first, usually
called Model I, it is assumed that w 6 = w 0 = where u contains only the host frequencies;
w 4¢ = w ≤4 . This implies that w 3¢ = w ≤3 and e.g., for vacancy diffusion in the f.c.c. lat-
w 3 w 5 = w 3¢ w 4 . In the second, Model II, tice u is given by
on the other hand, the impurity – vacancy w3
u = w1 + F (w 4 /w 0 ) (3-89)
interaction is in effect restricted to the first 2
This form for the impurity correlation fac- troduced which limit the number of jump
tor is especially relevant to the isotope ef- frequencies but only as a result of some
fect in diffusion (see Sec. 3.3.3) where for loss of realism.
self-diffusion the isotopes are tracers of the The first of these is the “random alloy”
host but, being isotopes, actually have model introduced by Manning (1968,
slightly different jump frequencies from 1970, 1971). The random alloy is of con-
the host and can be classified as “impur- siderable interest because, despite its sim-
ities”. We might also mention that for im- plicity, it seems to describe fairly well the
purity diffusion experiments the isotopes diffusion behavior of a large number of al-
can additionally be of the impurity. loys. The atomic components (two or
As a closing remark for this section, it more) are assumed to be ideally mixed and
should be remembered that because of its the vacancy concentration is assumed to be
mathematical form, the impurity correla- very low. The atomic jump frequencies,
tion factor will be temperature dependent. e.g., wA and w B , in a binary alloy are expli-
Over a fairly small temperature range the citly speficied and neither changes with
temperature dependence can frequently be composition or environment. In the very
approximated by an Arrhenius-like expres- dilute limit this model formally corre-
sion, i.e., sponds, of course, to specifying a host
jump frequency w 0 and an impurity jump
f 2 ≈ f 20 exp (– Q¢/k T ) (3-90)
frequency w 2 with all other host jump fre-
where Q¢ is some activation energy for the quencies equal to w 0 . However, it should
correlation process (this has no particular be appreciated for this model that physi-
physical meaning), k is the Boltzmann con- cally, although not mathematically, the
stant, and f 20 is a “constant”. It should be jump frequencies wA and w B are conceived
noted that Q¢ will be unavoidably included to be “average” jump frequencies. The
in the activation energy for the entire im- average jump frequency of the vacancy, wv ,
purity diffusion process and is not neces- was postulated to be given by
sarily unimportant.
wv = cA wA + cB w B (3-91)
3.3.1.5 Correlation Factors for where cA and cB are the atomic fractions
Concentrated Alloy Systems of A and B. Manning (1968, 1970, 1971)
finds that
The impurity correlation factors dis- (3-92)
S + ( 2 − 3 x0 + cB ) w B /w A + ( cA − x0 )
cussed in the previous section are probably fA =
appropriate for impurity concentrations up 2 {S + (1 − x0 ) w B /w A + x0 }
to about 1 at.%. At higher concentrations where
the impurities can no longer be considered
independent in the sense that the correla- S = {[(cB – x 0 ) wB /wA + (cA – x 0 )]2
tion events themselves are independent. + 4 x 0 (1 – x 0 ) wB /wA }1/2 (3-92 a)
This presents a special difficulty because
there is no easy way of extending, say, the and
five-frequency model into the concentrated x0 = 1 – f (3-92 b)
regime without rapidly increasing the num-
ber of jump frequencies to an unworkable and f is the correlation factor for vacancy
number. As a result, models have been in- diffusion in the pure metal A or B.
Early Monte Carlo calculations indi- not given explicitly as in the random alloy
cated that Eq. (3-92) was a very good ap- model but are expressed in the following
proximation over a wide range of wA /w B way in terms of the interaction energies
but more recently it has been found that (bonds) for a given atom in a given config-
this was largely illusory (Belova and uration:
Murch, 2000). Eq. (3-92) is actually only a
reasonable approximation when the ex- wi = n i exp (– Ui /k T )
change frequency ratio is within about an ¥ exp [(gi Eii + g j Ejj )/k T ], (3-94)
order of magnitude of unity. A much better i = A, B i( j
approximation to the problem is that given
by Moleko et al. (1989), but unfortunately where Eij < 0, n i is the lattice vibration fre-
the equations are much more cumbersome quency, Ui is the reference saddle point en-
to use. ergy in the absence of interactions, gi is the
A different approach to diffusion in con- number of atoms of the same type which
centrated systems has been taken by Kiku- are nearest neighbors of the given atom and
chi and Sato (1969, 1970, 1972). They de- g j is the number of atoms of the other type
veloped their Path Probability Method which are nearest neighbors of the given
(PPM) to cope specifically with the prob- atom. Other forms of wi are also possible
lems of diffusion in concentrated systems. but have not been developed in this con-
The method can be considered to be a time- text.
dependent statistical mechanical approach. In the PPM a path probability function
They started with the so-called binary alloy is formulated and maximized, a process
analogue of the Ising antiferromagnet which is said to be analogous to mini-
model. In this model, sometimes called the mizing the free energy in equilibrium sta-
“bond” model, interactions EAA , EBB , and tistical mechanics. The first calculations
EAB are introduced between nearest neigh- were made using what was called a time-
bor components of the type A-A, B-B, and averaged conversion. The results were in
A-B, respectively. Equilibrium properties poor agreement with later Monte Carlo
of this model are well known; see for ex- computer simulation results. Subsequently,
ample, Sato (1970). It is convenient to fo- substantial improvements were made
cus on the ordering energy E defined by (Sato, 1984) and the PPM results are now
in fairly good agreement with computer
E = EAA + EBB – 2 EAB (3-93)
simulation results. A typical comparison of
(note that in the literature there are other the earlier results with Monte Carlo results
definitions of the ordering energy which from Bakker et al. (1976) is shown in Fig.
differ from this one by either a negative 3-12. Note the bend at the order/disorder
sign or a factor of 2). When E > 0, there is transition. The lower values of the correla-
an ordered region in the b.c.c. and s.c. lat- tion factors on the ordered side can be
tices (the f.c.c. lattice is more complicated ascribed generally to the higher probability
but has not been investigated by the PPM). of jump reversals as an atom which jumps
The ordered region is symmetrically placed from the “right” sublattice to the “wrong”
about cA = 0.5. When E < 0, there is a two- sublattice tends to reverse the jump while
phase region at lower temperatures (this the vacancy is still present. This cancella-
side has not been investigated by the PPM). tion of jumps obviously leads to small val-
The exchange frequencies wA and wB are ues of f.
mechanism, which we shall now briefly

discuss. This sequence is interspersed, of
course, with jump reversals.
It became clear at an early stage that dif-
fusion in fully ordered structures poses cer-
tain difficulties. A purely random walk of
the vacancy will lead to large amounts of
disorder and inevitably to a great increase
in the lattice energy of the solid. Hunting-
ton (see Elcock and McCombie, 1958)
seems to have been the first to suggest that net
migration of atoms in, say, the B2 structure
Figure 3-12. Arrhenius plot of the tracer correlation could occur by way of a specific sequence
factor for B in the alloy A 0.6 B0.4 for various values of
of six vacancy jumps with only a relatively
U = (EAA – EBB )/E. Points from Bakker et al. (1976),
lines from Kikuchi and Sato (1969, 1970, 1972); af- slight increase in energy. This sequence is
ter Murch (1984 a). The abscissa is in units of E/k. illustrated in Fig. 3-13 a for the ordered
square lattice (Bakker, 1984), and sche-
matic energy changes for the sequence are
Although there is no particular reason shown in Fig. 3-13 b. The starting and fin-
for assuming that the correlation factor ishing configurations have the same energy
should follow an Arrhenius behavior, e.g., but there has been a net migration of atoms.
Eq. (3-90), these results nonetheless show The correlation factor for the six-jump
it quite well. The contribution to the total cycle sequence alone has been calculated
activation energy is difficult to determine by Arita et al. (1989). The six-jump cycle
directly because Ui is a “spectator quan- sequence is contained in the earlier path
tity” in these calculations and the vacancy probability method calculations and the
formation energy has not been calculated Monte Carlo computer simulations (Bak-
in the same work, although the latter can be ker, 1984). The correlation factors calcu-
determined separately in this model (Lim lated in those calculations are statistical av-
et al., 1990). In a recent application of this erages over all possible sequences in addi-
model to self-diffusion in b -CuZn (Belova tion to the jump reversals which tend to
and Murch, 1998), the contribution to the predominate.
total activation energy for, say, Cu tracer Further detailed discussion of the corre-
diffusion was variously estimated at 22 – lation factors in concentrated alloy systems
40%. It is clear that for a complete analysis can be found in the detailed reviews by
of diffusion in such materials the assess- Bakker (1984), Mehrer (1998), Murch and
ment of the contribution of the tracer corre- Belova (1998).
lation factor to the activation energy is es-
sential.
3.3.1.6 Correlation Factors for
We have mentioned that cancellation of
Highly Defective Systems
jumps leads to small values of f. There are,
however, special sequences of jumps in the Many solids, such as nonstoichiometric
ordered structure which lead to effective compounds, intercalation compounds, and
diffusion. The most important of these se- fast ion conductors, appear to have a very
quences is the so-called six-jump cycle high concentration of defects. Manning
extremes, provided the vacancies are ran-

domly distributed, roughly linear behavior
of the correlation factor is found (Fig. 3-9).
When the vacancies are not randomly
distributed, which is usually the case be-
cause of atom – atom repulsion, the behav-
ior of the correlation factor becomes more
complicated. The calculations that have
been performed are for the so-called lattice
gas model. In this, the atoms occupy dis-
crete sites of a rigid lattice, typically a
nearest neighbor interaction being speci-
fied between the atoms. For nearest neigh-
bor attractive interactions, at low tempera-
tures a two-phase region develops symmet-
rically about the concentration 0.5. Con-
versely, with nearest neighbor repulsion an
ordered region centered about a concentra-
tion 0.5 develops for the simple lattices
such as honeycomb, square planar, or sim-
ple cubic. The face centered cubic and tri-
angular lattices are rather more compli-
cated, but have not been investigated. We
shall focus on repulsion here because it is
the most likely. In a manner similar to that
for the concentrated alloy, the exchange
frequency of a given atom with a vacancy
is written for a given configuration as
Figure 3-13. (a) Six-jump cycle in the ordered square
lattice: the upper figure shows the path of the va- w = n exp (– U/k T ) exp (gnn fnn /k T ) (3-95)
cancy, the lower figure shows the displacements re-
sulting from the complete cycle, after Bakker (1984). where gnn is the number of atoms which are
(b) Schematic representation of energy changes dur- nearest neighbors to the given atom, U is
ing the six-jump vacancy cycle, after Bakker (1984).
the activation energy for diffusion of an
isolated atom, n is the vibration frequency,
and fnn is the atom – atom interaction en-
(1968) was probably the first to note that the ergy. It can be seen that nearest-neighbor
correlation factor could increase if more repulsion between atoms works to diminish
vacancies are present. In effect, the extra the activation energy. Other forms of the
vacancies decorrelate the reverse jump of a exchange frequency are also possible, but
tracer atom; see the discussion in Sec. most calculations have used this one.
3.3.1.3. As more vacancies are added we Obviously lattice gas models cannot
reach the limit of a single atom remaining generally be very realistic in their diffusion
(for convenience, the tracer) and the corre- behavior. The interest in them comes about
lation factor is unity as befitting a complete primarily because they exhibit behavior
random walk of an interstitial. Between the which is rich in physics and likely to occur
to a greater or lesser extent in real materi- from the right lattice to the wrong lattice it
als. Surprisingly, some materials such as tends to reverse that jump, thereby giving
interstitial solid solutions are described low values of the correlation factor. As in
fairly well by the lattice gas model even the ordered alloy, there are sequences of
with only nearest neighbor interactions. jumps which lead to long-range diffusion.
This must be a result of short-range inter- By and large these sequences are based
actions being by far the most important. around interstitialcy progressions, but
Sato and Kikuchi (1971) have employed rather than two atoms moving in unison as
the PPM to good effect in the lattice gas in the actual interstitialcy mechanism, here
model. Extensive calculations have been they are separated in time.
made, especially for the honeycomb lattice. The temperature dependence of f in the
Equally extensive Monte Carlo calcula- lattice gas, like the alloy, is fairly strong.
tions have also been made both for this lat- Again, it is usual to write f in an Arrhenius
tice and many others; see the review by fashion, e.g., Eq. (3-90), but there is no
Murch (1984 a). A typical result is shown known physical reason for assuming that f
in Fig. 3-14. The pronounced minimum in must really take this form. Fig. 3-15 shows
the correlation factor is due to the effects of the behavior in the square planar lattice
the ordered structure. This ordered struc- gas. The Arrhenius plot is in fact curved
ture consists of atoms and vacant sites ar- above and below the order/disorder tem-
ranged alternately. The wide composition perature, though in an experimental study
variability of the ordered structure at this this would be overlooked because of a
temperature (indicated by the arrows) is ac- much smaller temperature range. This acti-
commodated by vacant sites or “intersti- vation energy will be included in the ex-
tials” as appropriate. When an atom jumps perimental tracer diffusion activation en-
ergy. This is likely to be a significant con-
tribution which (like in the ordered alloy)
cannot be ignored.
The subject of tracer correlation effects
in defective materials with high defect con-
Figure 3-14. PPM results for the dependence of the Figure 3-15. Arrhenius plot of the tracer correlation
tracer correlation factor on ion site fraction ci in the factor gained from Monte Carlo simulation of a
honeycomb lattice with nearest neighbor repulsion. square planar lattice gas with 50% particles and 50%
T * = k T /fnn ; after Murch (1982 d). vacant sites (Zhang and Murch, 1990).
tent is dealt with at length in a number of

places; see, for example, Murch (1984 a)
and Sato (1989).
3.3.1.7 The Physical or Conductivity

Correlation Factor
In Sec. 3.3.1.3 we have seen that tracer
correlation effects are conventionally em-
bodied in the so-called tracer correlation
factor. In this way the correlation factor ap-
pears as a correction factor in the random
walk expression for the tracer diffusion co-
efficient Eq. (3-76).
In 1971 Sato and Kikuchi showed that Figure 3-16. PPM results for the dependence of the
the ionic conductivity should also include a physical correlation factor on ion site fraction ci in
correlation factor, now called the physical the honeycomb lattice with nearest neighbor repul-
or conductivity correlation factor and sion and site inequivalence. T * = k T /fnn and the site
given the symbol f I , sometimes f c . In the difference a priori is 5 fnn ; after Murch (1982 d).
usual hopping model expression for the
d.c. conductivity s (0), we have
s (0) = C (Z e)2 Gq rq2 f I /6 k T (3-96) Historically, the atomistic meaning of f I
has taken quite some time to determine,
where C is the concentration of charge car- principally because the calculations (path
riers per unit volume, Z is the number of probability and Monte Carlo methods)
charges, e is the electronic charge, Gq is the were based on the calculation of a flow of
jump frequency, rq is the jump distance of charge.
the charge, k is the Boltzmann constant, It is now known that the physical corre-
and T is the absolute temperature. lation factor can be expressed as
It should be noted that f I does not be-
come nontrivial, i.e., 71, until a relatively f I = lim 〈 DR2 〉 / n r 2 (3-97)
n→∞
high defect concentration is present and
ion – ion interactions and/or trapping sites where DR is the total displacement of the
are present. A model containing these was system after a total of n jumps in time t.
explored by Sato and Kikuchi (1971) in This differs from Eq. (3-77) only in the im-
their pioneering work. Their results for f I in portant points that DR refers to the dis-
the lattice gas model of nearest neighbor placement of the entire system and not a
interacting atoms diffusing on a honey- single particle. DR is simply the sum of the
comb lattice with inequivalent sites ar- individual particle displacements R occur-
ranged alternately are shown in Fig. 3-16. ring in time t:
This work was the starting point for the DR = ∑ Ri (3-98)
development of the new area of correla- all
particle
tion effects in ionic conductivity. This area
has been reviewed by Murch and Dyre We can now see that whereas f encom-
(1989). passes correlation effects of a single (tracer)
particle, f I encompasses correlation effects

of the entire system in a collective sense.
The correlation factor f I of the system is
manifested in the d.c. ionic conductivity,
Eq. (3-96). It is also manifested in chemi-
cal diffusion (see below), but not in tracer
diffusion.
What is sufficient to make f I 71? First,
a relatively high concentration of defects
is required (see Fig. 3-16). Next, inequality
of lattice sites (and therefore a variable Figure 3-17. Arrhenius plot of the physical correla-
jump frequency), or mutual interactions tion factor gained from Monte Carlo simulation of a
among the ions, or differences in acces- square planar lattice gas with 50% particles and 50%
vacant sites (Zhang and Murch, 1990).
sibilities of ions: all are sufficient, with a
high concentration of defects, to give a
nontrivial value for f I .
The importance of f I in materials with duction. However, f I also occurs in the ex-
high defect concentrations cannot be over- pression for the chemical diffusion coeffi-
stated. For example, in a calculation of f I cient in the one-component system, i.e.,
in a model for CeO2 doped with Y2O3
(to obtain a high concentration of anion va- D̃ = G r 2 f I (1 + ∂ ln g /∂ ln c)/6 (3-99)
cancies), it was found that at 455 K f I or, with Eq. (3-76)
changed from unity at low Y 3+ content to
only 0.05 at high Y 3+ content, correspond- D* ⎛ ∂ ln g ⎞
= f I ⎜1 + ⎟
ing to 14% of the anion sites being vacant f ⎝ ∂ ln c ⎠
(Murray et al., 1986). This very low value
of f I came about primarily because of the where g is the activity coefficient of the
trapping effects of the Y 3+ ions. The phys- particles with the term in parentheses be-
ical correlation factor is also temperature coming the “thermodynamic factor”. This
dependent, but there is no known physical equation applies to interstitial solid solu-
reason why f I should take an Arrhenius tions and intercalation compounds such as
form. Fig. 3-17 shows a typical result for a TiS2 intercalated with Li where the atomic
lattice gas above and below the order/dis- mobility on the defective lattice is rate
order transition. Both regions are slightly determining. It does not apply to ionic
curved in fact, although this could be over- conductors, however, since – for change
looked in an experiment over a small tem- balance reasons – compositional changes
perature range. The activation energy asso- are controlled by atomic mobility or hole-
ciated with f I will be included in the overall electron hopping on another lattice. For
activation energy for d.c. ionic conduction. example, in the oxygen ion conductor
This contribution cannot be ignored. In the calcia stabilized zirconia, compositional
model for CeO2 doped with Y2O3 , the con- changes of the oxygen ion vacancy concen-
tribution is of the order of 30% very high tration are probably controlled by the
Y 3+ contents. dopant calcium ion mobility on the cation
Most of the contributions to the applica- lattice.
tion of f I have centered around ionic con-
3.3.1.8 Correlation Functions where DRA is the displacement of the

(Collective Correlation Factors) system of particles of type A, i.e., DRA is
the vector sum of the displacements of all
There are a number of other correlation
the atoms of type A and nA is the total
phenomena in diffusion. In order to discuss
number of jumps of the species A. Thus the
these in a unified fashion, say for a binary
group of particles of type A in the system is
system, it is convenient first to introduce
treated itself like a (hypothetical) particle,
correlation functions or collective correla-
and fAA is its correlation factor.
tion factors. They represent the correlated
Similarly,
parts of the phenomenological coefficients
(i)
(see Sec. 3.4). For the diagonal coefficients fAB = ·DRA · DRB Ò/nA r 2 i = A, B (3-104)
we write (Allnatt and Allnatt, 1984) (i)
Thus fAB expresses correlations between
L ii = L0ii fii i = A, B (3-100) the vector sum of the displacements of all
where L0ii is the uncorrelated part of L ii , the atoms of type A with the corresponding
i.e., quantity for atoms of type B.
(0)
An alternative way of looking at the fij
L ii = Ci Gi r 2/6 k T i = A, B (3-101) is in terms of the drifts in a driving force.
where Ci is the number of species i per unit Let us consider a “thought experiment”
volume, Gi is the jump frequency of species for a binary system AB where an external
i, and r is the jump distance, k is the Boltz- driving force Fd is directly felt only by
mann constant, and T is the absolute tem- the A atoms. It is straightforward to show
perature. For the off-diagonal terms (we (Murch, 1982 a) that fAA is given by
note the Onsager condition LAB = L BA ), we 2 k T 〈 XA 〉
write fAA = (3-105)
Fd nA r 2
(0) (A) (0) (B)
LAB = LAA fAB = LBB fAB (3-102)
where ·XA Ò is the average drift distance of
This partitioning of the L ij into these parts the A atoms in the driving force and nA is
might appear arbitrary. The choice comes the number of jumps of a given A atom
about largely by analogy with the tracer in time t. However, although only the A
diffusion coefficient which is partitioned as atoms feel the force directly, the B atoms
a product of an uncorrelated part, contain- feel it indirectly because they are compet-
ing the jump frequency etc., and a corre- ing for the same vacancies as the A atoms.
lated part, containing the tracer correlation Accordingly, there is also a drift of the
factor (see Eq. (3-76) and associated dis- B atoms, smaller than the drift of the A
cussion). (i)
atoms. The correlation function fAB is re-
Some physical understanding of the fij is lated to this drift by
appropriate here. There are two equivalent
expressions for the fij . First, Allnatt (1982) (i ) 2 k T 〈 XB 〉
fAB = (3-106)
and Allnatt and Allnatt (1984) showed that Fd Gi t r 2
the fij have Einsteinian forms reminiscent
of the expression for the tracer correlation where ·XB Ò is the average drift distance of
factor (see Eq. (3-76)). Thus they write for the B atoms.
a binary system A, B
fAA = ·DRA2 Ò/nA r 2 (3-103)
Calculation of the Correlation Functions and

Calculations have been made for various (A)
fAB = 2 cA cB w A w B fA fB (3-109)
impurity models, the random alloy model −1
⎛ ⎞
and interacting or (bond) models for alloys. × ⎜ cA w A M0 ∑ ck w k fk ⎟
For impurity systems, the area has been re- ⎝ k = A,B ⎠
viewed exhaustively by Allnatt and Lidiard
where M0 = 2 f (1 – f ) and f is the tracer cor-
(1987, 1993). Here we shall restrict our-
relation factor in the pure lattice of either
selves to a discussion of the results for the
component.
five-frequency impurity model in the f.c.c.
The correlation functions have also been
lattice; see Fig. 3-11 and associated text for
calculated in interacting bond models (see
details of the model.
Sec. 3.3.1.5), especially with the PPM and
Inspection of Eq. (3-103) indicates that
by computer simulation. Some results of
when the A atoms (say) are the infinitely
the latter are shown in Figs. 3-18 and 3-19
dilute impurities in B, i.e., cA Æ 0, the
for the simple cubic lattice with ordering
system of atoms of type A reduces to a
between A and B.
single A atom. Accordingly the diagonal
A brief discussion of the behavior of
correlation function fAA reduces to the im-
these quantities is appropriate here. First,
purity correlation factor, fA . We have al-
with respect to fAA , as cA Æ 0, the A atoms
ready given Manning’s (1964) expression
behave like impurities in the B matrix, and
for fA for the five-frequency model in the
fAA does in fact reduce to the impurity cor-
f.c.c. lattice (see Eq. (3-78)). Manning
relation factor fA . At cA ≈ 0.5, the minima
(1968) showed that the cross correlation
(A) are further manifestations of the prepon-
function fAB can be calculated by a careful
derance of jump reversals in the diffusion
analysis of the various impurity jump tra-
(A) process in the ordered alloy. As cA Æ 1, all
jectories in a field. He found that fAB is
curves converge on unity and the correla-
given by
(A)
fAB = 2 fA (3-107)
F ) (w 4 − w 0 )w 4−1 ] −
w 3 [3 − 7 (1 − 2w1
×
2w1 + 7 F w3
where F is the fraction of dissociating va-

cancies that are permanently lost from a
site and are uncompensated for by return-
ing vacancies and is given by Eq. (3-80).
Let us move on to the binary random
alloy model described in detail in Sec.
3.3.1.5; explicit expressions have been de-
rived by Manning (1968). He found that
fAA is given by
⎡ ⎛ ⎞⎤ Figure 3-18. Monte Carlo results for the diagonal
fAA = fA ⎢1 + 2 cA w A fA ⎜ M0 ∑ ck w k fk ⎟ ⎥ correlation factor fAA in the simple cubic alloy as a
⎢⎣ ⎝ k = A,B ⎠ ⎥⎦ function of cA at various values of the ordering en-
(3-108) ergy; after Zhang et al. (1989 a).
current there are correlations or interfer-

ence between the two ionic conductivities.
When there are two ionic species and the
vacancy mechanism is operating, the ex-
pressions for the d.c. ionic conductivities
are (Murch and Dyre, 1989)
(3-110)
e 2 ZA2 CA GA r 2 ⎛ ZB ( A ) ⎞
sA = ⎜ fAA + f ⎟
6kT ⎝ ZA AB ⎠
and
(3-111)
e 2
ZB2 CB GB r 2 ⎛ ZA ( B) ⎞
sB = ⎜ fBB + fAB ⎟
Figure 3-19. Monte Carlo results for the off-diago- 6kT ⎝ ZB ⎠
A
nal correlation factor f AB in the simple cubic alloy as
a function of cA at various values of the ordering en- where e is the electronic charge, ZA (B) is the
ergy; after Zhang et al. (1989 a). number of charges on A (B), CA (B) is the
concentration of A (B), GA (B) is the jump
frequency of A (B) and r is the jump dis-
tion effects contained in fAA disappear en-
(A) tance.
tirely. Next, let us discuss fAB . As cA Æ 0,
Sometimes the bracketed terms are called
again A behaves like an impurity in B. The
(A) the binary conductivity correlation factors;
physical meaning of fAB has been dis-
they are, in fact, formally binary analogues
cussed earlier, see Eq. (3-106). At this limit
of f I in Eq. (3-96). Eqs. (3-110) and (3-111)
the B atoms have maximum interference
become
on A. Similar to fAA , minima develop at
cA ≈ 0.5 as a result of jump cancellations. sA = e 2 Z A
2
CA Gi r 2 f IA /6 k T (3-112)
As cA Æ 1, all curves converge on zero as 2
sB = e Z B2 CB Gi r 2 f IB /6 k T (3-113)
the B atoms, now in the minority, no longer
have much effect on the diffusion of A. For those readers familiar with Man-
With the correlation functions in hand, ning’s (1968) treatment of impurity ionic
we can form the various correlation terms conductivity, f IA (where A is the impurity
that commonly occur in ionic conductivity in B) is expressed as
and chemical diffusion. We shall restrict
⎛ Z ⎞
ourselves to binary systems. Unary sys- f IA = fA ⎜ 1 + B 〈 n p 〉⎟ (3-114)
tems are dealt with in Sec. 3.3.1.7, where it ⎝ ZA ⎠
is seen that the physical correlation factor
where fA is the impurity correlation factor
is really a correlation function for a one-
and ·n p Ò is a complex kinetic parameter
component system.
introduced by Manning. Eq. (3-114) as en-
visaged by Manning (1968) applies only to
Correlations in Ionic Conductivity
situations where the vacancy concentration
(Binary Systems)
is very low. Often the term in parentheses
When ionic conductivity occurs in a main Eq. (3-114) is loosely called a vacancy-
terial where two or more ionic species wind factor, although this terminology is
sharing the same sublattice (and therefore frequently applied to the whole of f IA . In
competing for the same defects) carry the the special case where the impurity is a
tracer of the host and where, of course,

ZA = Z B , the term in parentheses reduces to
fA–1 with the result that f IA = 1 and there are
no correlations in the ionic conductivity.
However, it is emphasized that this result
applies only where the defect concentration
is very low. For high concentrations of de-
fects which in general cannot be randomly
distributed there are still residual correla-
tions arising from non-ideal effects. These
are encompassed in f I (see Eq. (3-96)).
For the five-frequency model, see Sec.
3.3.1.4, for certain combinations of the
jump frequencies it is possible to make f IA
(Eq. (3-114)) negative. Specifically this
can occur when the vacancy and impurity Figure 3-20. PPM results for the quantities f IA and
are tightly bound together, i.e., when f IB as a function of concentration r A in the two-com-
w4 > w3 . Vacancies that are bound to im- ponent conductor with 20% vacancies at half the or-
purities can be transported around the im- der – disorder temperature. The dashed line shows the
case where the development of long-range order is
purity by the host atom flux. The probabil-
artificially suppressed; after Murch and Dyre (1989).
ity of the impurity moving upfield can then
be larger than the probability of moving
downfield. The impurity may then actually
move upfield, opposite to its “expected” stronger as the dimensionality is lowered
direction. For further details, see Manning (in fact all correlation effects do), and may
(1975). It is also possible in the five-fre- be a major contributor to the so-called
quency model to make f IA for the impurity mixed-alkali effect. This effect is charac-
exceed unity, even when Z B = ZA . terized by a substantial decrease in the d.c.
For concentrated interacting systems, the ionic conductivity at intermediate mixed
calculation of the f IA (and f IB ) is of some compositions without any obvious physical
interest, especially in the case of mixed fast cause. This subject is dealt with further by
ion conductors such as Na,K b -alumina. Murch and Dyre (1989) and Sato (1989).
An example of the results of calculation of
f IA and f IB , by way of the PPM, is shown in Correlations in Chemical Diffusion
Fig. 3-20. This lattice gas model contains (Binary Systems)
two species A and B (ZA = Z B ) with 20% The usual equation of practical interest
vacancies on a honeycomb lattice. The for chemical diffusion in a binary system is
model approximates the fast ion conductor the Darken equation. This equation relates
Na,K b -alumina, although it is probably the intrinsic diffusion coefficient of a par-
also a reasonable description of some sili- ticular component to its tracer diffusion
cate glasses. Note the minima in f IA and coefficient. The original Darken equation
f IB . These are caused by what is called a (which neglects correlations) is written as
“percolation difficulty” in the flow created
by ordered arrangements. Physically, many ⎛ ∂ ln g A ⎞
DAI = DA* ⎜ 1 + ⎟ (3-115)
jumps are reversed. This effect seems ⎝ ∂ ln cA ⎠
where gA is the activity coefficient of A. where f is the tracer correlation factor for
However, it is straightforward to show that diffusion in the pure crystal, i.e., pure A or
the rigorous equation is Eq. (3-56) which pure B. It turns out that the maximum in rA
can be rewritten in terms of the correlation in this model occurs when cA Æ 1. In the
functions (Le Claire, 1975; Murch, 1982 a; extreme case when the jump frequency of
Stark, 1976): A is much larger than that of B, then
⎛ ∂ ln g A ⎞ rA Æ f0–1. Therefore, when f0 = 0.78146 (va-
DAI = DA* ⎜ 1 + ⎟ cancy diffusion on the f.c.c. lattice, see Ta-
⎝ ∂ ln cA ⎠
ble 3-1), rA can only enhance chemical dif-
⎡⎛ (A) cA ⎞ ⎤ fusion by a factor of 1.28.
× ⎢⎜ fAA − fAB ⎟ fA ⎥ (3-116)
⎣⎝ cB ⎠ Manning’s equation (Eq. (3-117)) seems
⎦
to have more general significance than
The fA in the denominator is the correlation application to the random alloy. Murch
factor for the actual composition of the al- (1982 a) showed that this equation performs
loy and not some impurity limit. In the lit- very well indeed for an interacting alloy
erature the term derived from correlations model that can exhibit short and long-range
[ ] is sometimes called a vacancy-wind fac- order. This suggests that this equation can
tor and is given the symbol rA . As it turns be used almost with impunity. This subject
out, however, this term cannot vary much arises again in the “Manning relations”
from unity since the original Darken equa- which relate the phenomenological coeffi-
tion is reasonably well obeyed. The behav- cients to the tracer diffusion coefficients
ior of rA is not especially transparent. The (see Sec. 3.2.3.4).
vacancy flux in chemical diffusion is al- The subject of the experimental check of
ways in the same direction as that of the rA has been discussed by Bocquet et al.
slower moving species (and opposite to (1996). They pointed out that in most cases
that of the faster moving species). The va- the Manning formulation for rA and rB
cancy-wind effect always tends to provide (Eq. (3-117)) improves the agreement be-
a given atom with an enhanced probability tween experimental and calculated values
of flowing in a direction opposite to that of of the Kirkendall shift and the ratio of
the vacancy flow. When A is less mobile the intrinsic diffusion coefficients DAI /DBI .
than B, the vacancy-wind factor rA effec- However, the individual experimental val-
tively makes DAI even smaller. Conversely, ues of DAI and DBI often tend to be quite a
when A is more mobile than B, the va- bit higher than the calculated values. The
cancy-wind factor rA effectively makes DAI experimental Kirkendall shift also has a
even larger. But how much? This can be tendency to be higher than the calculated
partially answered in the following. value. Carlson (1978) points out that the
Although we have given Manning’s problem could be due to the random alloy
(1968) expressions for the individual fij assumption in the Manning formulation.
and the correlation factor fA for the random However, the success of the Manning for-
alloy (Eqs. (3-108) and (3-109)), it may not mulation for the interacting alloy model
be immediately obvious that rA is in fact mentioned above seems to vindicate the
given in this model by (see also Sec. random alloy assumption and so the reason
3.2.3.4) for the discrepancy probably should be
f DB* + cA ( DA* − DB* ) sought elsewhere.
rA = (3-117)
f ( cA DA* + cB DB* )
3.3.2 The Nernst – Einstein Equation When the external force is the result of an
and the Haven Ratio electric field E we have
The Nernst – Einstein equation relates Fd = Z e E (3-123)
the d.c. ionic conductivity to a diffusion
where Z is the number of charges (ionic
coefficient. Probably no other equation in
valence) and e is the electronic charge.
diffusion has generated more misunder-
The mobility u is defined as the velocity
standing than this one. Let us consider the
per unit field and so we have
standard derivation (see, for example,
Batchelor (1976)), and discuss its implica- u Ze
= (3-124)
tions in detail. ˜ kT
D
We consider a pseudo-one-component
system in a situation where the flux result- In the solid-state diffusion literature Eq.
ing from an applied force on the particles (3-124) is generally called the Nernst – Ein-
(which are completely noninteracting) ex- stein relation. Because the d.c. ionic con-
actly counterbalances the flux due to diffu- ductivity is related to the mobility by s =
sion. That is, from Eq. (3-2) C Z e u, Eq. (3-124) can be rewritten as
dC s C Z 2 e2
〈v 〉 C = D
˜ (3-118) = (3-125)
dx D̃ kT
It is important to note that the diffusion co- More generally, interactions are present
efficient here refers to a chemical composi- between the particles and it can be shown
tion gradient and is most definitely con- that the general form of the Nernst – Ein-
ceived as a chemical diffusion coefficient, stein equation is in fact (Murch, 1982 b)
not a tracer or self-diffusion coefficient.
Lack of appreciation of this fact leads to s C Z 2 e 2 ⎛ ∂ ln c ⎞
= ⎜ ⎟ (3-126)
misunderstandings and inconsistencies. ˜
D k T ⎝ ∂m ⎠
The external force Fd is a result of a po-
where m is the chemical potential of the
tential so that
particles and c is the site fraction.
df Now let us discuss this equation in detail
Fd = − (3-119)
dx by exploring some particular cases. When
At equilibrium the distribution of com- the distribution of particles is completely
pletely noninteracting particles follows a ideal, meaning that the particles do not feel
Boltzmann distribution such that one another, not even site blocking, the
thermodynamic factor drops out of Eq. (3-
C (x) = C0 exp [– f (x)/k T ] (3-120) 126) and s /D̃ reduces to Eq. (3-125). In this
very special case, and only in this case, the
Eq. (3-120) must be the solution of Eq. (3-
tracer diffusion coefficient D* equals D̃, so
120) at steady state. We then have that
that
dC C df C Fd
=− = (3-121) s C Z 2 e2
dx k T dC k T = (3-127)
D* kT
With Eq. (3-118) we find that
When the particles are ideally distributed
〈v 〉 Fd but subject to the condition that no more
= (3-122)
D̃ k T than one particle can occupy one site, then
the thermodynamic factor in Eq. (3-126) i.e., formally the same as Eq. (3-129). This
equals (1 – c)–1. However, for this situation route to Eq. (3-132) is clearly a case of two
the tracer diffusion coefficient is related to wrongs ending up making a right. How-
D̃ by (see, for example, Murch (1982 c)) ever, the real danger lies in the fact that Eq.
D* = D̃ (1 – c) f (3-128) (3-132) obtained in this way blinds us to its
limitations, limitations which are clearly
where f is the tracer correlation factor, so stated in the derivation of Eq. (3-129).
that In practice, Eq. (3-132) has often been
s C Z 2 e2 used to describe situations where the par-
= (3-129) ticles are interacting and many sites are
D* kT f
vacant, such as in fast ion conductors. In
Ionic solids having virtually a perfect lat- such cases, we cannot necessarily expect a
tice of particles (charge carriers) fall into meaningful interpretation. What is the cor-
this category and Eq. (3-129) is appropriate rect way to proceed? We already have the
to such solids. Similarly, when the particles Nernst – Einstein equation to cover the situ-
are extremely dilute, Eq. (3-129) is again ation of interacting particles (Eq. (3-126)).
appropriate. However, this normally cannot be applied
In the solid-state diffusion literature we to real materials because local charge neu-
very often see Eq. (3-125) used directly trality prevents composition gradients be-
to calculate another diffusion coefficient, ing set up in ionic conductors by the con-
sometimes called the “charge” diffusion co- ducting ions themselves. What can be done
efficient and given the symbol Ds . Thus we is to return once again to Eq. (3-130) and use
encounter the following equation, which is it purely as a definition of Ds , recognizing
also called the Nernst – Einstein equation at the same time that Ds has no Fickian
s C Z 2 e2 meaning. Eq. (3-129) is then being used
= (3-130) purely as a means of changing s to a quan-
Ds kT
tity which has the dimensions of a diffusion
Ds is dimensionally correct but it does not coefficient. It is clear that it would be inap-
correspond to any diffusion coefficient that propriate in these circumstances to call this
can actually be measured by way of Fick’s equation the Nernst – Einstein equation.
laws for a solid system. Recall that the We can now define the Haven ratio,
identical Eq. (3-125) requires that the parti- which is simply the ratio of D* to Ds :
cles are completely noninteracting for it to
be meaningful. We frequently then see the D*
HR ≡ (3-133)
following equation relating D* and Ds Ds
D* = f Ds (3-131) In view of what has been said above about
with Ds often being called a self-diffusion Ds , it is appropriate to ask whether HR has
coefficient or the diffusion coefficient of any physical meaning. This can be partially
the (random walking) charge carriers. Eq. answered by examining hopping models
(3-131) is generally given as if it is self-ev- for conduction. It can be shown quite gen-
ident or similar to Eq. (3-73). Eqs. (3-130) erally for hopping models using the va-
and (3-131) are then combined to give cancy mechanism that
s C Z 2 e2 f f
= (3-132) HR = = (3-134)
D* kT f fI f + c g
where f I is the physical or conductivity cor- Much less satisfactorily interpreted are
relation factor and g is a two-particle corre- fast ion conductors where the defect con-
lation factor (see Sec. 3.3.1.7). For specific centration is high and f I must be included
models of interacting particles f I ≤ 1 (see in the analysis of HR . A review of HR for
Murch and Dyre (1989) for a detailed re- such materials has been provided by Murch
view). At the limits of an almost full or (1982 d). Further comments on the subject
empty lattice of charge carriers, f I Æ 1 and can be found in the review on correlation
HR = f. This is compatible, of course, with effects in ionic conductivity by Murch and
Eq. (3-131) for these conditions. Dyre (1989). An introduction to the subject
Accordingly, a measurement of the may be found in the book by Philibert
Haven ratio (obtained by measuring s and (1991).
D*, preferably in the same sample) can in
some cases give f alone and therefore the 3.3.3 The Isotope Effect in Diffusion
mechanism of diffusion can be exposed.
The isotope effect, sometimes called the
However, the interstitialcy mechanism adds
mass effect, is of considerable importance
a minor complication because the “charge”
in diffusion because it provides one of the
moves two jump distances whereas the
two experimental means of gaining access
tracer moves only one (see Fig. 3-6). This
to the tracer correlation factor, f (the other
is easily taken care of in the analysis so that
is the Haven ratio; see Sec. 3.3.2). Since f
HR is now written generally as (see for ex-
depends on mechanism (see Sec. 3.3.1.3), a
ample, Murch (1982 d))
2
measurement of the isotope effect can, in
f ⎛r⎞ principle, throw light on the diffusion
HR = ⎜ ⎟ (3-135)
fI ⎝ rq ⎠ mechanism operating.
The measurement of the isotope effect
where r and rq are the distances moved by
depends on accurately measuring a small
the tracer and charge respectively. Pro-
difference between the diffusion coeffi-
vided we focus on cases where f I = 1 (either
cients of two tracers a and b (Peterson,
almost empty or full lattice of charge car-
1975). This small difference can be related
riers), then HR still has a unique value for
to f in the following way. For isotropic
each mechanism. The classic example of
tracer and impurity diffusion, the diffusion
the value of HR in identifying the mecha-
coefficients of a and b can be written as
nism is Ag motion in AgBr (Friauf, 1957).
It was found that HR varied from 0.5 at low Da = A wa fa (3-136)
temperature to 0.65 at high temperature. Db = A wb fb (3-137)
Frenkel defects (see Sec. 3.4.4) are ex-
pected and therefore a temperature depen- where A contains geometrical terms and
dence of HR . But HR is not consistent with defect concentrations which do not depend
a combination of vacancy and direct inter- on the atom/defect exchange frequency w.
stitial jumps. This would require that HR The correlation factors fa and fb have the
varies from 0.78 to unity. The lower values mathematical form of impurity correlation
of HR obtained require a contribution from factors because the tracers (i.e. isotopes) in
interstitialcy jumps. It was found that both real experiments are formally impurities in
collinear and noncollinear interstitialcy the sense that their jump frequencies differ
jumps were required to fit the behavior of from the host atoms. Most impurity corre-
HR . lation factors for impurity diffusion in oth-
erwise pure and nearly perfect crystals From Eqs. (3-140) and (3-141), we find
have the mathematical form that
fa = u /(wa + u) (3-138) ( Da − Db )/ Db
f DK = (3-142)
( mb / ma )1/ 2 − 1
fb = u /(wb + u) (3-139)
Eq. (3-142) refers to a process where only
where u contains exchange frequencies
one atom jumps. The more general case of
other than the tracer. After taking the ratio
n atoms jumping simultaneously, e.g., the
of Eqs. (3-136) and (3-137) and eliminat-
interstitialcy mechanism (where n = 2) (see
ing u and fb (say) from Eqs. (3-136), (3-
Fig. 3-6), is described by
138), and (3-139), we find that
(Da – Db ) /Db = f [(wa – wb ) /wb ] (3-140) f DK (3-143)
[( Da − Db )/ Db ]
At this point, the formal distinction be- =
[( mb + ( n – 1) m0 ) /( ma + ( n – 1) m0 )]1 / 2 – 1
tween fa , fb and the tracer correlation fac-
tor f, the latter referring to a hypothetical where m0 is the average mass of the non-
tracer with the same jump frequency as the tracers. It is common to refer to f DK in
host, can be dropped since they are very Eqs. (3-142), (3-143) as the “isotope ef-
similar in numerical value. Knowledge of fect” E (n). Because f and DK are ≤ 1 then
the experimental jump frequencies, wa and this also applies to E (n). Accordingly, al-
wb , is usually not available, of course. Ac- though a measurement of the isotope effect
cordingly, they are expressed in terms of may not uniquely determine the mecha-
quantities which are known, such as the nism, it can be invaluable in showing
masses of the tracers. which mechanisms (through their values of
Application of classical statistical me- f and n) are not consistent.
chanics to describe the dynamics of the There are some mechanisms for which
jumping process leads to the result the appropriate correlation factor does not
have the simple form of Eq. (3-136), for
⎡⎛ mb ⎞ 1/ 2 ⎤ example a mechanism which has several
(wa − w b )/w b = DK ⎢⎜ ⎟ − 1⎥ (3-141) jump frequencies. Nonetheless, it is always
⎢⎣⎝ ma ⎠ ⎥⎦
possible to derive equations equivalent
where ma and mb are the masses of the to Eqs. (3-142), (3-143) for such cases (see
tracers a and b, DK is the fraction of the the review by Le Claire (1970) for details).
total kinetic energy, associated with the In the discussion above, there has per-
whole motion in the jump direction at the haps been the implication that the theory
saddle point of a jump, that actually be- described here applies only to tracer diffu-
longs to the diffusing atom. The conversion sion in “simple” materials, e.g., pure met-
of jump frequencies to masses has unfortu- als, and alkali metal halides. Certainly the
nately led to the introduction of this new theory was originally developed with these
quantity DK, about which detailed informa- cases in mind, but it can be used, with cau-
tion is difficult to obtain. If the remainder tion, in other more complex situations. One
of the lattice is not involved in the jump, of the earliest of these was application to
DK = 1 (this is the upper limit for DK). impurity diffusion. In this case the tracers
More usually, there is a certain amount of a and b now refer to isotopes of an impur-
coupling between the diffusing atom and ity which is chemically different from the
the remainder of the lattice and DK < 1. host. When the mechanism is not in any
doubt, and frequently it is not to the experi-

enced researcher in the field, a measure-
ment of the isotope effect for self-diffusion
yields DK. If the impurity is substitutional,
and provided that DK is assumed to be the
same as for self-diffusion, then an isotope
measurement for the impurity can give the
correlation factor for impurity diffusion.
Since the impurity correlation factor is a
function of local jump frequencies, e.g., the
five-frequency model in the f.c.c. lattice
(see Fig. 3-11), then a knowledge of f com-
bined with other knowledge of the same
jump frequencies can lead to a knowledge
of some of their ratios (see, for example, Figure 3-21. Arrhenius plot of Monte Carlo results
for the tracer correlation factor in the simple cubic
Rothman and Peterson (1967), Bocquet lattice. 첸 represents the actual tracer correlation fac-
(1972), and Chen and Peterson (1972, tor; 쑗 represents the tracer correlation factor forced
1973)). to follow the impurity form; after Zhang et al.
The “impurity form” of the tracer corre- (1989 b).
lation factor is also implied for Manning’s
theory for vacancy diffusion in the random
alloy model (see Sec. 3.3.1.5). Therefore, countered at low temperature in fast ion
isotope effect measurements in concen- conductors (Zhang and Murch, 1997). The
trated alloys which are reasonably well implication of these and other findings is
described thermodynamically by the ran- that the isotope effect measured in ordered
dom alloy model can be interpreted along materials does not contain the usual tracer
traditional lines. Similarly, it has been correlation factor but some other correla-
shown by Monte Carlo computer simulation factor defined only by way of the im-
tion that the impurity form of f is also valid purity form, Eq. (3-136). This does not
for vacancy diffusion in a lattice containing mean that isotope effect measurements in
a high concentration of randomly distrib- ordered materials are intrinsically without
uted vacancies, up to 50% in fact (Murch, meaning, but simply that they cannot be
1984 b). easily interpreted.
When order is increased among the com- The actual measurement of the isotope
ponents, the impurity form for f even for effect in diffusion requires the very accu-
vacancy diffusion is increasingly not fol- rate measurement of (Da – Db )/Db . This
lowed. For example, in the stoichiometric quantity typically lies between 0.0 and
binary alloy AB with order, the correlation 0.05. Because of the inaccuracies in mea-
factor, forced to follow the impurity form, surements of D it is not feasible to measure
diverges from the actual f below the or- Da and Db in separate experiments. Nor-
der/disorder temperature (Zhang et al., mally the isotopes a and b are co-de-
1989 b) (see Fig. 3-21). A very similar situ- posited in a very thin layer and permitted
ation arises for the case of ordered atoms to diffuse simultaneously into a thick sam-
diffusing by the vacancy mechanism on a ple. The geometry of the experiment per-
highly defective lattice such as is often en- mits Eq. (3-8) to be used for each isotope.
With a little manipulation it is soon found “wait” until a defect arrives. The probabil-
that ity of a defect, say a vacancy, arriving at a
particular neighboring site to a given atom
⎛C ⎞ is simply the fractional vacancy concentra-
ln ⎜ a ⎟ = const. − ln Ca [( Da − Db )/ Db ]
⎝ Cb ⎠ tion cv . However, in certain situations such
(3-144) as alloys showing order, or fast ion conduc-
This means that by plotting ln (Ca /Cb ) vs. tors showing relatively high defect order-
ln Ca , (Da – Db )/Db can in fact be obtained ing, the availability of vacancies to the
very accurately, avoiding errors due to time atoms can be enhanced or depressed from
and temperature of the diffusion anneal and the random mixing value which cv signi-
errors in sectioning. The separation of the fies. This topic is discussed further in Secs.
isotopes is normally achieved by half-life, 3.4.2 and 3.4.3.
energy spectroscopy, or the use of different The probability of a given atom being
kinds of radiation. Details of how to carry next to a vacancy is thus g cv where g is the
out careful experiments in this area can be coordination number. The jump frequency
found in the very comprehensive review of G can we decomposed to
experimental techniques in diffusion by
G = g cv w (3-145)
Rothman (1984). A more detailed review
of the isotope effect in diffusion has been where w is usually called the “exchange”
provided by Peterson (1975); see also Le frequency to signify the exchange between
Claire (1970). an atom and a neighboring vacant site and
to distinguish it from the actual jump fre-
quency G . It is permissible at low defect
3.3.4 The Jump Frequency
concentrations to call w the defect jump
In Sec. 3.3.1.2 we showed that a tracer frequency, since the defect does not have to
diffusion coefficient is, in essence, a prod- “wait” for an atom.
uct of an uncorrelated part containing In, say, f.c.c. crystals where the jump
–
the jump frequency and a correlated part distance r is given by a /÷2, where a is the
containing the correlation factor, see Eq. lattice parameter and g = 12, the tracer dif-
(3-76). The Arrhenius temperature depen- fusion coefficient from Eq. (3-76) can now
dence of the diffusion coefficient arises be written as
largely from the jump frequency, although
D* = 12 cv w (a 2/2) f /6 = cv w a 2 f (3-146)
some temperature dependence of the corre-
lation factor in some situations, e.g., al- The same equation is also valid for b.c.c.
loys, cannot be ignored (see Secs. 3.3.1.4 crystals. Other examples have been given
to 3.3.1.6). In this section we will study the by Le Claire (1975).
make-up of the atomic jump frequency. In many solids, such as metals, alloys
First, for an atom to jump from one site and ionic crystals, cv is said to be “intrin-
to another, the defect necessary to provide sic” with an Arrhenius temperature depen-
the means for the jump must be available. dence. The vacancy formation enthalpy is
For the interstitial diffusion mechanism contained in the measured activation en-
there is essentially always a vacancy thalpy for the tracer diffusion process (see
(really a vacant interstice) available except Eq. (3-149)). In some solids, notably ionic
at high interstitial concentrations. For other crystals and certain intermetallic com-
mechanisms, however, the atom has to pounds, apart from the inevitable tempera-
ture-dependent intrinsic concentration, the In Sec. 3.3.4.2 we show that the defect
vacancy concentration can also be manipu- concentration in many cases depends on
lated by doping or by changing the degree temperature in an Arrhenius fashion (Eq.
of nonstoichiometry. In such cases this va- (3-159)). In such cases, the atomic jump
cancy concentration is sometimes said to frequency G, being the product of w , the
be “extrinsic”, and this contribution to the defect concentration, and the coordination
total vacancy concentration can easily number, is written as from Eq. (3-145)
swamp the intrinsic vacancy contribution. G = g n– exp [(S m + S f )/k]
v
The extrinsic vacancy concentration is in-
¥ exp [– (H m + Hvf )/k T ] (3-149)
dependent of temperature. However, in the
m f
case of a change in stoichiometry, tempera- The sum of H and H is more accurately
ture independence of cv requires adjust- called the activation enthalpy. We empha-
ment of the external partial pressure (see size again that the activation enthalpy
Sec. 3.4.4). measured in a diffusion experiment can, for
some solids such as ordered alloys, contain
other contributions such as from correla-
3.3.4.1 The Exchange Frequency
tion effects or defect availability terms.
With the defect immediately available, The enthalpy of migration and the en-
the atom can jump when it acquires suffi- thalpy of defect formation are now rou-
cient thermal energy from the lattice for it tinely calculated by Lattice or Molecular
to pass over the energy barrier between its Statics methods; see Mishin (1997) and
present site and the neighboring site. The references therein for examples of such
probability of the atom having this thermal calculations in grain boundaries.
energy is given by the Boltzmann probabil- Although it is straightforward to couch
ity exp (– G m/k T ), where G m (a Gibbs en- the argument above concerning w in statis-
ergy) is the barrier height, k is the Boltz- tical mechanical terms, this would still be
mann constant, and T is the absolute tem- inadequate because we have neglected the
perature. The attempt frequency, i.e., the participation of the remainder of the lattice,
number of times per second the atom on its especially the neighbors of the jumping
site is moving in the direction of the neigh- atom, in the jump process. A more satisfac-
boring site, is the mean vibrational fre- tory treatment has been provided by Vine-
quency n–. Accordingly, the “jump rate”, yard (1957). This puts n– on a sounder basis
i.e., number of jumps per second, w , is and gives physical significance to the en-
given by tropy of migration S m.
Vineyard (1957) took a classical statisti-
w = n– exp (– G m/k T ) (3-147)
cal mechanical approach by considering
Because G m is a Gibbs energy, the ex- the phase space of the system comprising
change frequency can be partitioned as all of its N atoms and one vacancy. Each
point in this phase space represents an
w = n– exp (S m/k) exp (– H m/k T ) (3-148)
atomic configuration of the system. All
where S m is the entropy of migration and possible configurations are represented.
H m is the enthalpy of migration. H m is We are concerned with two neighboring
sometimes loosely called the “activation” energy minima in this phase space. These
enthalpy, but this can lead to misunder- minima are centered on two points, P and
standings, as we shall now see. Q. Point P is the phase space point repre-
senting an atom adjacent to a vacancy; all where ni0 is the frequency of the ith normal
atoms are in their equilibrium positions. mode with the system constrained to move
Point Q is the corresponding point after the normal to the direction joining point P to
atom – vacancy exchange. Between the the saddle point.
points P and Q there is a potential energy There have been many further considera-
barrier. In order to jump, the atom must tions of the detailed dynamics of the jump
pass over this barrier at its lowest point, a process, the most notable of these being the
saddle point. work carried out by Jacucci, Flynn and co-
Vineyard (1957) was able to calculate workers (see reviews by Jacucci (1984)
the jump rate, i.e., the number of crossings and Pontikis (1990)).
per unit time at the “harmonic approxima- For general purposes it is often sufficient
tion”. This entailed expanding the potential to focus on Eq. (3-148) and to note that the
energy at point P and also at the saddle Debye frequency is an adequate represen-
point in a Taylor series to second order to tation of the mean vibration frequency n–.
give for w
⎛ N ⎞ 3.3.4.2 Vacancy Concentration
⎜∏ ni ⎟
i =1
w = ⎜ N ⎟ exp ( − U / k T ) (3-150) The change in Helmholtz energy FVf as-
⎜ n ′⎟ sociated with the formation of one vacancy
⎜∏ i⎟
⎝ i =1 ⎠ is given by
where the ni are normal frequencies for vi- FVf = EVf – T SVf (3-154)
bration of the system at point P, ni¢ are the
where EVf and SVf are the energy and en-
normal frequencies at the saddle point and
tropy of formation, respectively. The for-
U is the difference in potential energy
mation process itself is conceived to be the
between the saddle point and the equilib-
removal of an atom from the interior of the
rium configuration point, P.
crystal to the surface. The entropy part has
It is possible to transform Eq. (3-150)
two contributions: a vibrational or thermal
into the form of Eq. (3-148) by writing f
part S vib arising from the fact that atoms
N
close to the vacancy have a different vibra-
∏ ni tional frequency from those far from the
i =1
N −1
= n exp ( S m / k T ) (3-151) f
vacancy, and a configurational part S config
∏ n i′ which is usually thought to be an ideal mix-
i =1
ing entropy, at least for pure metals – but it
where the entropy of migration can be is rather more complicated for alloys. The
identified with configurational part for pure metals is eas-
3N ily found from elementary classical statisti-
Sm = k ∑ ln (n i0 /n i′ ) (3-152) cal mechanics (see, for example, Peterson
i =2 (1978)). The number of different ways W
and of putting n vacancies and N atoms on N + n
3N sites assuming indistinguishability within
∏ ni each group is
i =1
n= (3-153)
3N ( N + n )!
∏ n i0 W=
N! n!
(3-155)
i=2
3.4 Diffusion in Materials 219
The configurational entropy is given by ation techniques with computer codes such
as DEVIL and CASCADE (these tech-
f ⎡ ( N + n )! ⎤
Sconfig = k ln W = k ln ⎢ ⎥ (3-156) niques are described in the book edited by
⎣ N! n! ⎦ Catlow and Mackrodt, 1982) and lattice
which, with Stirling’s approximation ln N ! dynamics and Monte Carlo methods (see,
= N ln N – N for large N, results in for example, Jacucci, 1984).
Vacancy concentrations in metals are con-
⎛ n ⎤⎞
= − k ⎜ N ln ⎡⎢
N ⎤
f
Sconfig ⎥ + n ln ⎡⎢ ⎟ ventionally measured by quenching, dila-
⎝ ⎣N + n⎦ ⎣ N + n ⎥⎦⎠ tometry, and positron annihilation. An intro-
(3-157) duction to these techniques has been pro-
The total change in free energy when n va- vided by Borg and Dienes (1988), and a de-
cancies are produced is tailed review is also available (Siegel, 1978).
n FVf = n EVf – T (n Sconfig

f
) (3-158)
After substitution of Eq. (3-157) and put- 3.4 Diffusion in Materials
ting the derivative ∂F/∂n = 0 to obtain the
equilibrium number of vacancies, we soon Probably every type of solid has, at one
find that time, been investigated for its diffusion be-
(3-159) havior. It is impossible in the space here to
n
= cv = exp ( Svib / k ) exp ( − EV / k T )
f f cover even the major findings. In this sec-
N+n tion we propose to discuss, at an introduc-
where cv is the site fraction of vacancies. tory level, diffusion in metals and alloys,
Note that the configurational term has dis- and, as an example of ionic crystals,
f
appeared. The vibrational contribution S vib oxides. Our emphasis here is on the usual
is considered to be small and the leading theoretical framework for a description of
term in Eq. (3-159) is usually ignored, and diffusion in these materials rather than on
data compilations or reviews. Diffusion
cv = exp (– EVf /k T ) (3-160)
data compilations for metals and alloys can
For low pressures we can also assume (not be found in the Smithells Metals Reference
always safely) that the energy of formation Book (Brandes, 1983) and in the extensive
EVf can be approximated by the enthalpy of compilations in Landolt-Börnstein (Meh-
formation H Vf . rer, 1990). The older tracer diffusion data
Divacancies are handled in a similar way up to 1970 on metals and oxides has been
(see, for example, Peterson (1978)). Fren- collected by Askill (1970) and data on
kel and Schottky defects in ionic crystals oxides up to 1970 on metals and oxides
(see Sec. 3.4.4) can also be analysed along has been collected by Askill (1970) and
similar lines (Kofstad, 1972). Alloys, how- data on oxides up to 1980 by Freer (1980).
ever, present a special problem because of Extensive compilations of diffusion data
uncertainties about reference states includ- in nonmetallic solids have been published
ing the typical surface site where an atom in Landolt – Börnstein (Beke, 1998 1999).
is to be symbolically placed (see, for exam- The journal Defect and Diffusion Form,
ple, Lim et al. (1990) and Foiles and Daw formerly Diffusion and Defect Data, regu-
(1987)). larly publishes abstracts of all diffusion-re-
Calculations of defect formation ener- lated papers and extensive indices are reg-
gies are frequently handled by lattice relax- ularly provided.
3.4.1 Diffusion in Metals diffusion measurements have been made

over a very wide temperature range, there
3.4.1.1 Self-Diffusion is a slight curvature in the Arrhenius plots.
We have seen (Sec. 3.3.4) that the tracer This has been variously attributed to a con-
diffusion coefficient in solids normally fol- tribution to diffusion from divacancies (see,
lows an Arrhenius form. This is usually for example, Peterson (1978)), or from va-
written empirically as cancy double jumps (Jacucci, 1984). Be-
cause of the high activation energy, how-
D = D 0 exp (– Q /R T ) (3-161) ever, the latter is really only a candidate for
explaining curvature very close to melting
where D 0 is called the pre-exponential fac- temperature and not over a wide tempera-
tor, or sometimes the “frequency factor”, Q ture range.
is the activation energy for diffusion, R is There is still controversy over the cause
the ideal gas constant, and T is the absolute of the curvature in Ag and other f.c.c. met-
temperature. In most pure metals self-dif- als, with one view favoring a temperature-
fusion is characterized by a pre-exponen- dependent activation energy and diffusion
tial D 0 which falls in the range 10 – 3 to by single vacancies (see the review by
5 ¥10 – 6 m2 s –1. This corresponds to activa- Mundy (1992)), and the other tending to fa-
tion entropies which are positive and of the vor, in part, a contribution from divacan-
order of k, the Boltzmann constant. The ac- cies at high temperatures (Seeger, 1997).
tivation energy Q is given fairly closely The behavior of self-diffusion in b.c.c.
(± ≈ 10%) by Q = 34 TM.Pt. , where TM.Pt. is metals is quite varied. Some b.c.c. metals
the melting point. These values lead to a such as Cr show linear Arrhenius plots.
value of the self-diffusion coefficient at the The alkali metals show slightly curved
melting point of about 10 –12 m2 s –1. plots. Finally, there is a large group of
Careful dilatometric and quenching ex- “anomalous” metals including b -Zr, b -Hf,
periments on a large number of f.c.c. met- b -Ti, g -U, and e -Pu, which show a strong
als indicate the vacancy as being the defect curvature and show values of D0 and Q
responsible for “normal” diffusion in f.c.c. both of which are anomalously low. There
metals. For vacancy diffusion the tracer has been considerable controversy here
diffusion coefficient is written for cubic also. Again, the monovacancy mechanism
lattices as and a temperature-dependent activation en-
D* = g cv w r 2 f /6 (3-162) ergy have been strongly supported (Mundy,
1992) but this view alone does not seem to
where g is the coordination, cv is the va- be entirely consistent with the nature of the
cancy concentration, w is the vacancy – curvature. Other processes such as ring
atom exchange frequency, r is the jump mechanisms and Frenkel pair formation/re-
distance, and f is the tracer correlation combination may be operative (Seeger,
factor. The decomposition of cv into its 1997).
enthalpy/entropy parts is given by Eq.
(3-159). The decomposition of w is given
3.4.1.2 Impurity Diffusion
in Eq. (3-148). The tracer correlation factor
is discussed in Sec. 3.3.1.3. The appropriate diffusion coefficient for
For some f.c.c. metals, e.g., silver (Roth- an impurity A (at infinite dilution) in a host
man et al., 1970; Lam et al., 1973), where of B when the vacancy mechanism is oper-
ating is (for cubic lattices) interstitials and partly as substitutionals

(see Sec. 3.3.1.1). For a full discussion of
DA = g pAV wA r 2 fA /6 all aspects of impurity diffusion we refer to
= DA0 exp (– QA /R T ) (3-163) Le Claire (1978).
where g is the coordination, pAV is the va-
cancy availability factor to the impurity, 3.4.2 Diffusion in Dilute Alloys
wA is the impurity – vacancy exchange fre-
quency, r is the jump distance, and fA is the 3.4.2.1 Substitutional Alloys
impurity correlation factor (for details see Alloys containing something less than
Sec. 3.3.1.4). Again, as for self-diffusion in about 1– 2% solute concentrations are con-
metals, in many cases we can discern “nor- sidered to be sufficiently dilute that the dif-
mal” behavior wherein the Arrhenius plots fusion of solute atoms can be considered in
are linear and DA0 and QA do not differ terms of isolated atoms or isolated group-
greatly from the self-diffusion values. As ings of atoms such as pairs. For many bi-
pointed out by Le Claire (1975, 1978), the nary dilute alloys, measurements of the
relative values are determined principally solute diffusion coefficient D2 (measured
by DQ = QA – QB , where QB is the self-dif- as a tracer diffusion coefficient) have been
fusion activation energy for the host. When made as a function of solute concentration
DA > DB , the impurity is a fast diffuser and (we use the subscript 2 to denote the solute
DQ is negative. It is often found that this is and 0 to denote the solvent). It is usual to
correlated to a situation where the valence represent the solute diffusion coefficient
of the impurity is greater than that of the empirically as
host. The converse is true when DA < DB .
The similarity between impurity and host D2 (c2) = D2 (0) (1+ B1 c2 + B2 c22 …) (3-165)
diffusivities suggest the vacancy mecha- where D2 (0) is the solute diffusion coeffi-
nism for both. For this mechanism it can cient for c2 Æ 0, c2 is the atomic fraction of
easily be shown that (see, for example, Le solute and B1 , B2 , … are termed solute en-
Claire, 1978) hancement factors. D2 (0) is also called the
impurity diffusion coefficient (at infinite
DQ = (HAm – HBm ) + EB – Q ¢ (3-164)
dilution) depending on the context of the
where HAm, HBm are the activation enthal- experiment.
pies of migration for the impurity and host, Similarly, solvent diffusion coefficients
EB is the vacancy – impurity binding en- D0 , also measured as tracer diffusion coef-
thalpy, and Q ¢ is the activation enthalpy ficients, are represented as
arising from the temperature dependence D0 (c2) = D0 (0) (1 + b1 c2 + b2 c22 …) (3-166)
of fA (see Eq. (3-90)). This is normally a
fairly small contribution. and b1 , b2 , … are termed solvent enhance-
There are “anomalous” impurity diffus- ment factors.
ers which have diffusion coefficients much Of interest are the solute and solvent en-
higher than the host. Well-known examples hancement factors. Naturally, also of inter-
are the diffusion of noble metals in Pb. As est is the composition (a) of values of the
has been discussed in conjunction with the solute and solvent diffusion coefficients
substitional-vacancy diffusion mechanism, and (b) of values of their respective activa-
the impurities probably exist partly as tion energies.
In Eq. (3-166) D2 (0) arises from isolated w 0 , the solvent exchange frequency far
solute atoms, while the term containing B1 from the solute. For the f.c.c. lattice the re-
arises from pairs of solute atoms which are sult for b1 is (Howard and Manning, 1967)
sufficiently close that the solute jump fre-
w 4 ⎡ c 1 w1 c ⎤
quency differs from the isolated solute – va- b1 = − 18 + ⎢ 4 + 14 2 ⎥ (3-168)
cancy exchange frequency w 2 . B2 arises w 0 ⎣ f0 w 3 f0 ⎦
from triplets of solute atoms.
where c 1 and c 2 are termed partial correla-
The solute enhancement factor B1 has
tion factors and are known functions of the
been calculated for the f.c.c. lattice by ex-
ratios w 2 /w 1 , w 1 /w 3 and w 4 /w 0 , and f0 is
tending the five-frequency model (see Sec.
the correlation factor in the pure lattice.
3.3.1.4) to include three new frequencies.
Eq. (3-168) can be re-expressed as
These frequencies describe solute jumps
which create, i.e., associate, a new solute ⎡ c w c w ⎤
pair (w 23 ), dissociate a solute pair (w 24 ), b1 = − 18 + ⎢ 4 1 1 + 14 2 3 ⎥
⎣ f w0 f0 w 0 ⎦
and reorient an existing pair (w 21). Pairs of
solute atoms do not occur randomly when × exp( − E2B / k T ) (3-169)
solute atoms interact. When an interaction thereby showing the effect of an altered va-
energy E 22 between solute atoms is defined cancy concentration near a solute by virtue
it is straightforward to show that of the solute – vacancy binding energy E 2B .
w 24 exp (– E 22 /k T ) = w 23 exp (– E 2B /k T ) There are similar relations for the b.c.c.
lattice. For the impurity model described
where E 2B is the impurity – vacancy bind- in Sec. 3.3.1.4 (Model I), b1 is given by
ing energy. If it is assumed that the impur- (Jones and Le Claire, 1972)
ity correlation factor does not depend on
u1
solute concentration (Stark, 1972, 1974), b1 = − 20 + 14
then B1 is given by (Bocquet, 1972; Le f
w3 n 1
Claire, 1978) +6 exp( − E2B / k T ) (3-170)
w0 f
B1 = − [6 + 12 exp( − E22 / k T )] (3-167)
− exp( − E2B / k T ) and for Model II (see Sec. 3.3.1.4)
⎡ w w ⎤ u2
× ⎢ 4 21 exp( − E22 / k T ) − 14 23 ⎥ b1 = − 20 + 6
⎣ w2 w2 ⎦ f
w n
If it is assumed that E 22 = 0 and that + 14 3 2 exp( − E2B / k T ) (3-171)
w 21 = w 23 = w 2 then B1 is reduced to w0 f
where u 1 , n1 (w 2 /w 3¢ , w 3 /w 3¢ ) and u 2 , n 2
B1 = 18 [exp (– E 2B /k T ) – 1] (3-167 a)
(w 2 /w 3 , w 4 /w 0 ) are called mean partial
There have been no calculations along sim- correlation factors and are known functions
ilar lines of B2 . of the frequencies given.
The solvent enhancement factor b1 can It is possible to express b1 in terms of the
be calculated by noting the number of jump ratio of the solute and solvent diffusion co-
frequencies close to the solute. For in- efficients. This aids in the discussion of the
stance, in the five-frequency model for the numerical values taken by the enhance-
f.c.c. lattice we need to count the number ment factors. For the f.c.c. lattice, b1 is
of solvent frequencies which differ from given by Le Claire (1978).
4 f0 D2 ⎡ 4 c 1 (w 1 /w 3 ) + 14 c 2 ⎤ sion for B1 . These comments are borne out

b1 = − 18 +
1 − f2 D0 ⎢⎣ f0 ( 4 w 1 /w 3 ) + 14 F ⎥⎦ by the data in Table 3-2.
(3-172) The relative values of the diffusion coef-
ficients of solute and solvent are dictated
where F has been given in Sec. 3.3.1.4 (see largely by the difference in activation ener-
Eq. (3-80)) and f0 is the tracer correlation gies for solute and solvent diffusion rather
factor in the pure lattice. The term in brack- than by differences in the pre-exponen-
ets is roughly unity and the impurity corre- tial factors. The difference DQ = Q2 – Q0
lation factor f2 is about 0.5. The greater the (where Q 2 and Q 0 are the activation ener-
solute diffusion coefficient is compared gies for solute and solvent respectively) of-
with the solvent diffusion coefficient, then ten seems to be closely related to the differ-
the more likely it is that b1 is greater than ence in valencies of solute and solvent, Z 2
zero. On the other hand, for slow solute and Z 0 . When DQ is negative (fast solute
diffusers b1 will probably be negative but diffusion), Z 2 > Z 0 . On the other hand,
not less then – 18. A few examples taken when DQ is positive Z 2 < Z 0 . These matters
from Le Claire (1978) are given in Table are discussed further in a detailed review
3-2. by Le Claire (1978). The reader is also di-
The solvent enhancement factor b2 re- rected to a recent commentary (Le Claire,
sults from the change in the solvent jump 1992).
frequencies in the vicinity of pairs of An approximate approach which acts as
solute atoms. Expressions for b2 for the a transition between the models described
f.c.c. lattice have been given by Bocquet above and the concentrated alloy models of
(1972). Sec. 3.4.3 is the complex model initiated
As noted by Le Claire (1978), the ex- first by Dorn and Mitchell (1966) and de-
pressions for b1 and B1 are similar in form. veloped by Faupel and Hehenkamp (1986,
Provided that the frequency ratios in one 1987). When impurities (B) have a positive
equation do not greatly differ from the excess charge, vacancy – impurity com-
other, as might be expected from relatively plexes of one vacancy and i impurity atoms
weak perturbations caused by the solute, are likely to form. Assuming that the bind-
then b1 and B1 tend to have the same sign ing free energy (– GBi ) is independent of
and roughly comparable magnitudes, apart the configuration of the impurity atoms,
from the exp (– E 22 /k T ) term in the expres- Dorn and Mitchell (1966) wrote for the
Table 3-2. Some values of solute and solvent enhancement factors, from Le Claire (1978).
Alloy system Lattice D2 /D0 b1 B1 T (°C)

(Solvent – solute)
Ag – Pd f.c.c. 0.04 – 8.2 – 7.5 730

Ag – Au f.c.c. 3.7 7.0 5.5 730
Ag – Cd f.c.c. 5.3 13.0 8.5 730
Ag – In f.c.c. 8.4 37 30 730
Ag – Tl f.c.c. 0.26 – 1.2 – 0.56 730
b -Ti – V b.c.c. 1.26 – 4.3 – 2.5 1400
V – Ti b.c.c. 1.66 16 27 1400
a -Fe – Si b.c.c. 1.64 20.4 12.4 1427
mole fraction of vacancies in the ith com- enhanced in a nonlinear way (Hehenkamp
plex cvi et al., 1980; Hehenkamp and Faupel, 1983).
Moreover, independent measurements of
⎛ g⎞
cvi = cv ( 0 ) ⎜ ⎟ c0g − i c2i exp( GBi / k T ) (3-173) the vacancy concentration on alloying have
⎝ 0⎠
shown a linear dependence between nor-
where cv (0) is the mole fraction of vacan- malized vacancy concentration and the
cies in the pure metal (A) and c0 and c2 are normalized solvent diffusion coefficient
the mole fractions of solvent (A) and solute (Hehenkamp et al., 1980; Hehenkamp and
(B) respectively, and g is the coordination. Faupel, 1983). The nonlinear enhancement
Bérces and Kovács (1983) modified Eq. (3- of solvent diffusion upon alloying is thus
173) to take into account possible configu- apparently due to a corresponding nonlin-
rations of the complexes. ear increase in the vacancy concentration
Within this complex model framework on alloying. These findings are very nicely
Faupel and Hehenkamp (1986, 1987) have described by the complex model.
introduced average effective jump fre- Despite its obvious limitations, the com-
A plex model can be useful quantitatively up
quencies per solvent (A) atom ·w eff Òi and
average effective jump frequencies per to 5 at.% of solute and qualitatively even
B higher; see Le Claire (1992).
solute (B) atom ·w eff Òi in the ith complex
(the latter specifically includes correlation
effects – which are always important for 3.4.2.2 Interstitial Alloys
the solute). The normalized solvent diffu-
sion coefficient can now be written as Elements such as H, N, O, and C dis-
solve interstitially in metals and diffuse by
g −1
D0 ( c2 ) 〈w eff
A
〉 i ⎛ g − 1⎞ the interstitial mechanism. The diffusion
= c0g −1 − ∑ ⎜ ⎟
D0 ( 0 ) i =1 f0 w 0 ⎝ i ⎠ coefficients here are often measured by re-
laxation techniques or outgassing. Tracer
× c0g −1 − i c2i exp( GBi / k T ) (3-174)
diffusion is difficult to measure in the case
Similarly, the normalized solute diffusion of N and O because of the lack of suitable
coefficient can be written as radioisotope; however, other techniques
are available (see Sec. 3.5.1).
D0 ( c2 ) g
i 〈w eff
B
〉i ⎛ g⎞
= c0g −1 − ∑ ⎜ ⎟ (3-175) For interstitials at infinite dilution, the
D0 ( 0 ) i = 2 g f2 w 2 ⎝ i ⎠ interstitials move independently (by the
× c0g −1 c2i −1 exp[( GBi − GB1 )/ k T ] interstitial mechanism, see Fig. 3-5). The
vacancy concentration (vacant interstice) is
These equations for the normalized solute unity, the correlation factor is unity (a com-
and solvent diffusion coefficients can be plete random walk), and the tracer diffu-
expanded in terms of c2 to give explicit ex- sion coefficient is given simply by (from
pressions for the various solute and solvent Eqs. (3-76) and (3-145))
enhancement factors by way of comparison
D* = g w 0 r 2/6 (3-176)
with Eqs. (3-165) and (3-166); see Le
Claire (1992). where w 0 is the interstitial/vacant interstice
It has been found, in a number of careful “exchange” frequency.
studies in f.c.c. systems (Ag – Sb, Ag – Sn) In a number of cases the interstitial con-
where the impurities have excess positive centration can become sufficiently high for
charge, that solvent diffusion on alloying is the interstitials to interfere with one an-
other and this interference may have an im- (McKee, 1981)

portant effect on diffusion. An example is
D̃ = –13 r 2 [12 k0 + 12 u4 /u3 ]
C in austenite where the carbon interstitials
can fill roughly 8% of the octahedral voids. ¥ [4 u1 + 7 u3 – 12 k0 + 7 (u4 – k0 ) u3 /u4 ]
The two theoretical approaches taken to ¥ c {1 + c [1 + 12 (1 – u4 /u3 )]} (3-178)
cope with the problem are reminiscent of
where r is the jump distance. Eq. (3-178)
the situation found in the modeling of sub-
contains a thermodynamic factor.
stitutional alloys (see Sec. 3.4.2.1). In the
The above model, where the frequencies
first, the exchange frequencies for the so-
are specified explicitly, is probably only
lute are specified. In the second, the ex-
valid for 1– 2% of interstices occupied. At
change frequencies are specified indirectly
higher concentrations many more frequen-
by way of interaction energies.
cies are required with the result that this
For the first approach, McKee (1980 a,
approach becomes unwieldy. For the sec-
b) and Le Claire (1981) conceived a four-
ond approach a lattice gas model is used to
frequency model for interstitial solute dif-
represent the solute interstitials and their
fusion in the f.c.c. lattice (the octahedral
lattice of interstices. High concentrations
voids of the f.c.c. lattice). Two distinct spe-
of interstitials can in principle be handled
cies are considered: the isolated interstitial
with this model. In this rigid model the at-
with a jump frequency of k0 , and paired
oms are localized at sites and pairwise
interstitials, which rotate with frequency
interactions among atoms are specified.
u1 , dissociate with frequency u3 , and asso-
The exchange frequency of a solute atom to
ciate with frequency u4 . The following ex-
a neighboring site can be written as (Sato
pression for the tracer diffusion coefficient
and Kikuchi, 1971)
of the interstitial solute can then be derived
(Le Claire (1981) using Howard’s (1966) w = n exp (gnn fnn /k T ) exp (– U/k T ) (3-179)
random walk method)
although many other choices are possible.
⎡ cp ⎛ In Eq. (3-179) n is the vibration frequency,
D* = 4 r 2 ⎢ k0 + ⎜ 4 u1 + 14 u3 (3-177)
⎣ 6c ⎝ gnn is the number of occupied interstices
that are nearest neighbors to a solute, fnn is
( 2 u1 − 3 u3 )2 u ⎞⎤ the solute – solute “binding” energy (posi-
− − 12 k0 − 7 k0 3 ⎟ ⎥
( 2 u1 + 7 u3 ) u4 ⎠ ⎦⎥ tive or negative), and U is the migration en-
ergy for an isolated solute atom. In effect
where cp is the site fraction concentration
the solute neighbors can assist or impede
of paired interstitials and c is the site frac-
the diffusion process depending on the sign
tion of interstitials. Using the pair associa-
of fnn .
tion method, McKee (1980 a, b) derived a
By its nature this model requires a statis-
similar result valid for weak binding of the
tical mechanical approach for its analysis.
paired interstitials. Explicit expressions for
One method used is the PPM (Sato and
the tracer correlation factor can be found in
Kikuchi, 1971; Sato, 1989, see also Sec.
the original papers.
3.3.1.5); another is the Monte Carlo com-
It was also possible to determine an ex-
puter simulation method (Murch, 1984 a).
pression for the chemical diffusion coeffi-
It is convenient to decompose the tracer
cient, D̃, relating to diffusion of the solute
diffusion coefficient into
in its own concentration gradient. In the
weak binding limit the expression for D̃ is D* = g r 2 V W f exp (– U/k T )/6 (3-180)
where V is the vacancy availability factor some loss of realism. The first of these is
(the average availability of neighboring va- the random alloy model introduced by
cant interstices to an interstitial solute Manning (1968, 1970, 1971). The second
atom, and is a more general quantity than is the interacting bond model which has
cv ), and W represents the effect of the envi- been extensively developed by Kikuchi
ronment on the jump frequency. It is in ef- and Sato (1969, 1970, 1972).
fect the statistical average of the first term In the random alloy model, the atomic
in Eq. (3-179). The same lattice gas model components are assumed to be mixed ran-
has also been applied extensively to fast domly and the vacancy mechanism is as-
ion conductors (see, for example, Murch sumed. The atomic jump frequencies wA
(1984 a)). and wB for the two components A and B
Chemical diffusion can readily be ex- are specified and these do not change with
pressed along similar lines (Murch, 1982 c) composition or environment. The tracer
as diffusion coefficient of, say, A is given by
D̃ = g r 2 V W n f I (3-181) * = g wA r 2 cv fA /6
DA (3-182)
¥ exp (– U/k T ) (∂m /k T/∂ ln c)/6 where g is the coordination, r is the jump
distance, and cv is the vacancy site fraction
where m is the chemical potential of the
(see Sec. 3.3.4.2). The only theoretical re-
solute and f I is the physical correlation fac-
quirement here is to calculate the tracer
tor (see Sec. 3.3.1.7). The physics for f I
correlation factor fA ; this has been de-
are not contained in the first approach (Eq.
scribed in Sec. 3.3.1.5 on correlation ef-
(3-178)).
fects in diffusion. The random alloy model
These approaches seem to describe C
performs very well in describing the diffu-
tracer and chemical diffusion in g -Fe fairly
sion behavior of many alloy systems which
well (McKee, 1980 a, b, 1981; Murch and
are fairly well disordered. This subject is
Thorn, 1979) but other applications have
discussed extensively in the review by
not been made because of the lack of suit-
Bakker (1984) to which we refer. A much
able data, especially for D*.
more accurate method for dealing with fA
Quantum effects are an important ingre-
has been given by Moleko et al. (1989).
dient in the description of H diffusion in
For alloys which exhibit order, Man-
metals; see the reviews by Völkl and Ale-
ning’s theory, based on the random alloy
feld (1975), Fukai and Sugimoto (1985),
model, still performs reasonably well in
and Hempelmann (1984).
describing certain aspects of diffusion be-
havior, notably correlation effects which
3.4.3 Diffusion in Concentrated Binary can be expressed in terms of the tracer dif-
Substitional Alloys fusion coefficients, see Sec. 3.3.1.8 and the
review by Murch and Belova (1998). A
Dilute alloy models, see Sec. 3.4.2.1, wholly different approach pioneered by
cannot be extended very far into the con- Kikuchi and Sato (1969) is the PPM (see
centrated regime without rapidly increas- Sec. 3.3.1.5) in which the tracer diffusion
ing the number of jump frequencies to an coefficient is expressed in terms of quan-
unworkable level. As a result, models have tities which are statistically averaged over
been introduced which limit the number of the atomic configurations encountered in
jump frequencies but only as a result of the alloy. The tracer diffusion coefficient
of, say, A is now written as from w – because the configurations change

A
D* = g w – r 2 p– –f /6 (3-183)
with temperature. We also have contribu-
A A Av A
tions from p–Av and the tracer correlation
–
where wA is an averaged exchange fre- factor. In order to make sense of the experi-
quency which includes the effect of the en- mental activation energies, we need to
vironment on the jump frequency, p–Av is the model the system in question; there seems
vacancy availability factor, i.e., the prob- to be no alternative. Few detailed applica-
ability of finding a vacancy next to an A tions of this model have yet been made.
–
atom (see below), and f A is the tracer corre- One detailed application has been made to
lation factor. We have discussed the corre- b -CuZn (Belova and Murch, 1998). Fur-
–
lation factor f A in Sec. 3.3.1.5, to which we ther details can be found in the review by
refer. The quantity w – is the average of the
A Bakker (1984), see also the review by
exchange frequency for an A atom with a Murch and Belova (1998).
vacancy in particular configuration: Interdiffusion in binary concentrated al-
wA = nA exp (– UA /k T ) (3-184) loys has been dealt with largely in Secs.
3.2.3.2 and 3.3.1.8. Some recent calcula-
¥ exp [(gA EAA + gB EAB )/k T]
tions (which use the “bond model” de-
where nA is the vibration frequency, UA is scribed above) have been performed by
the migration energy (referred to some ref- Zhang et al. (1988) and Wang and Akbar
erence) for a jump, gA is the number of A (1993), see also the review by Murch and
atoms that are nearest neighbors to a given Belova (1998).
A atom (itself next to a vacancy), gB is the
number of B atoms which are likewise
3.4.4 Diffusion in Ionic Crystals
nearest neighbors to the A atom, and EAA
and EAB are nearest neighbor interactions In this section diffusion in ionic crystals
(assumed negative here). This equation is (exemplified here by oxides) will be briefly
essentially the binary analogue of Eq. (3- discussed. A more detailed discussion has
179). We see that the neighbors of the atom been provided by Kofstad (1972) and
A can increase or decrease the apparent mi- Schmalzried (1995). Diffusion in other
gration energy. ionic crystals, especially alkali metal ha-
The use of cv alone in the expression for lides and silver halides, has been reviewed
*
DA in Eq. (3-182) implies that either com- by, for example, Fredericks (1975), Laskar
ponent “sees” the vacancy equally. This (1990, 1992), Monty (1992).
obviously cannot generally be true. The in-
troduction of the quantity p–Av in Eq. (3-
3.4.4.1 Defects in Ionic Crystals
183) (and p–Bv ) is a recognition that vacan-
cies are somewhat apportioned between the In order to discuss diffusion in ionic
two atomic components. This is not a small crystals, we need first to discuss at an ele-
effect, particularly in alloys which show mentary level the types of defects which
separate sublattices. arise. Of almost overriding concern in
The overall activation energy for diffu- ionic crystals is the requirement of charge
sion is now rather complicated. Apart from neutrality.
the usual contribution from the vacancy Let us first consider stoichiometric crys-
formation energy and reference migration tals, say the oxide MO, and “intrinsic” de-
energy, we have a complicated contribution fect production. The Schottky defect (actu-
ally a pair of defects) consists of a vacant where MM is a metal atom (M) on a metal
anion site and a vacant cation site. It arises site (M), M ••
i is an effectively doubly posi-
as a result of thermal activation and not tively charged metal ion on interstitial i site
through interaction with the atmosphere. and V≤ M is an effectively doubly negatively
The Schottky defect generation is written charged metal ion vacancy. We have as-
as a chemical reaction, i.e., sumed double charges here purely for illus-
0 [ V≤ •• trative purposes.
M + VO (3-185)
The equilibrium constant for the Frenkel
where 0 refers to a perfect crystal, V≤ M is a defect reaction can be written as
vacant (V) metal (M) site, the primes refer
to effective negative charges (with respect KF = [M •• M]
i ] [V≤ (3-189)
to the perfect crystal), VO•• is a vacant (V)
provided that the defect concentration is
oxgen (O) site and the dots refer to effec-
low. This equation is often called the Fren-
tive positive charges. This is the Kröger-
kel product. Again, the equilibrium con-
Vink defect notation (see, for example,
stant KF can be expressed as exp (– GFf /k T )
Kofstad (1972)).
where GFf is the Gibbs energy of formation
We can write an equilibrium constant Ks
of the Frenkel defect, which again can be
for the reaction in the following form, valid
partitioned into its enthalpy H Ff and en-
for low defect concentrations:
tropy SFf parts.
Ks = [V≤ ••
M ] [VO ] (3-186)
where the brackets [ ] indicate concentra-
tions. Eq. (3-186) is usually called the 3.4.4.2 Diffusion Theory in Ionic Crystals
Schottky product. Ks can be expressed in
In most mechanisms of diffusion except
the usual way
the interstitial mechanism, an atom must
Ks = exp (– Gsf /k T ) (3-187) “wait” for a defect to arrive at a nearest
where Gsf is the Gibbs energy of formation neighbor site before a jump is possible (see
of the Schottky defect, which can be parti- Sec. 3.3.4). Thus the jump frequency in-
tioned into its enthalpy Hsf and entropy Ssf cludes a defect concentration term, e.g., cv ,
parts. the vacancy concentration. Let us examine
The other type of defect occurring in the an example for diffusion involving the
stoichiometric ionic crystal is the Frenkel Frenkel defect. Although both an intersti-
defect. The Frenkel defect (actually a pair tial and a vacancy are formed, in oxides
of defects) consists of a cation interstitial one of them is likely to be much more mo-
and cation vacancy or an anion interstitial bile, i.e., to have a lower migration energy,
and anion vacancy. In the latter case it has than the other. (In the case of stoichiomet-
also been called an anti-Frenkel defect, al- ric UO2 , for example, theoretical calcula-
though this nomenclature is now relatively tion of migration energies suggests a much
uncommon. Like the Schottky defect, the lower migration energy for the oxygen va-
Frenkel defect is thermally activated. The cancy than for the interstitial (by either
Frenkel defect generation is also written as interstitial or interstitialcy mechanisms)
a chemical reaction, for example for cat- (Catlow, 1977).) At the stoichiometric
ionic disorder: composition,
MM [ M •• i + V≤ M (3-188) [M •• M]
i ] = [V≤ (3-190)
so that the cation vacancy concentration is Mn1– y O. As an example we will deal with
given by an anion excess accommodated by metal
f vacancies.
M ] = cv = exp (– G F / 2 k T )
[V≤ (3-191)
The degree of nonstoichiometry and
= exp (– H Ff /2 k T ) exp (SFf /2k) (3-192) therefore the defect concentration accom-
panying it are functions of temperature and
The measured activation enthalpy for dif- partial pressure of the components. The de-
fusion will be the migration enthalpy plus fects produced in this way are sometimes
half the Frenkel defect formation enthalpy. said to be “extrinsic”, but this terminology
The reader might well ask how we know is probably to be discouraged since they
we are dealing with a Frenkel defect and are still strictly intrinsic to the material. In
not a Schottky defect, and if we do know it an oxide it is usual, at the temperatures of
is the Frenkel defect then how do we know interest, to consider only the partial pres-
it is the vacancy mechanism that is operat- sure of oxygen since it is by far the more
ing and not the interstitialcy mechanism? volatile of two components. In carbides,
In general, we have to rely on independent where the temperatures of diffusion of
information, principally computer calcula- interest are much higher, the metal partial
tions of defect formation and migration en- pressure can be comparable to the carbon
thalpies, but structural information such as partial pressure and either can be manipu-
is provided by neutron diffraction and ther- lated externally.
modynamic information is also useful. The simplest chemical reaction generat-
Another process of interest here is the in- ing nonstoichiometry is
trinsic “ionization” process whereby an
electron is promoted from the valence band –12 O2 (g) [ OO
¥ ¥
+ VM (3-194)
to the conduction band leaving behind a where OO ¥
is a neutral (¥) oxygen ion on an
hole in the valence band. In the Kröger- oxygen site and VM ¥
is a neutral metal va-
Vink notation, we write for the intrinsic cancy. Physically, this corresponds to oxy-
ionization process (the electrons and holes gen being adsorbed on the surface to form
may be localized on the metal atoms) ions and more lattice sites, thereby effec-
0 [ e¢ + h• (3-193) tively making metal vacancies. These va-
cancies diffuse in by cation – vacancy ex-
All ionic crystals are capable, in princi- change until the entire crystal is in equilib-
ple, of becoming nonstoichiometric. The rium with the atmosphere. Excess oxygen
limit of nonstoichiometry is largely dic- nonstoichiometry corresponds also to oxi-
tated by the ease with which the metal ion dation of the metal ions. Here in this exam-
can change its valence and the ability of the ple, the holes produced are conceived to be
structure to “absorb” defects without re- associated very closely with the metal va-
verting to some other structure and thereby cancies. The holes can be liberated, in
changing phase. Nonstoichiometry can be which case the reaction can now be written
achieved by either 1) an anion deficiency,
–12 O2 (g) [ V¢M + OO
¥
+ h• (3-195)
which is accommodated by either anion va-
cancies, e.g., UO2 – x , or metal interstitials, If the hole is then localized at a metal ion,
e.g., Nb1+ y O2 or 2) an anion excess, which we can consider this as the chemical equiv-
is accommodated by either anion intersti- alent of having M 3+ ions formally present
tials, e.g., UO2 + x or metal vacancies, e.g., in the sublattice of M 2+ ions.
Assuming reaction Eq. (3-194) is the gen partial pressure will expose the actual
¥
preferred one, we find that the equilibrium charge state of the vacancy, i.e., VM or V¢M .
constant for this reaction is In practice, this kind of differentiation
between charge states and even defect
[VM× ]
K1 = 1/ 2 (3-196) types does not often occur unambiguously.
pO 2 The result of plotting log D* versus log pO2
thereby immediately showing that the often does not show precisely one slope
metal vacancy concentration depends on and often there is also some curvature.
the partial pressure of oxygen in the fol- These effects can sometimes be associated
lowing way: with contributions to the diffusion from the
¥
¥
three types of vacancy, VM , V¢M , and V≤M,
[VM ] = K1 p1/2
O2 (3-197) e.g., Co1– d O (Dieckmann, 1977). Probably
Accordingly, the metal vacancy concentra- the most successful and convincing appli-
tion is directly proportional to p1/2 cation of this mass-action law approach has
O2 .
Assuming that the reaction in Eq. (3- made the defect types in Ni1– d O compat-
195) is the preferred one, we find that the ible with data from tracer diffusion, electri-
equilibrium constant for this reaction is cal conductivity, chemical diffusion, and
deviation from stoichiometry/partial pres-
[ h•] [O O× ] [VM
′] sure (Peterson, 1984).
K2 = 1/ 2
(3-198)
pO2 At higher partial pressures of oxygen,
which in some cases such as Co1– d O,
¥
The concentration of [OO ] is essentially Mn1– d O, and Fe1– d O can result in large
constant and is usually absorbed into K2 . deviations from stoichiometry, we can log-
The condition of electrical neutrality re- ically expect that defect interactions could
quires that play an important role. First, the concentra-
[h• ] = [V¢M ] (3-199) tions in the mass-action equations should
be replaced by activities. This so greatly
thereby showing that the metal vacancy complicates the analysis that progress has
concentration depends on the partial pres- been fairly slow. The retention of concen-
sure of oxygen in the following way: trations rather than activities may be per-
[V¢M ] = K21/2 p21/4 (3-200) missible, however, because there is some
evidence that non-ideal effects tend to can-
Accordingly, the metal vacancy concentra- cel in a log D/log pO2 plot (Murch, 1981).
tion is directly proportional to p1/4
O2 . At high defect concentrations, defect clus-
Let us assume that the migration of tering to form complex defects which
metal ions will principally be via vacancies themselves might move or act as sources
in the nonstoichiometric region, since they and sinks for more mobile defects, is pos-
are the predominant defect. Since the tracer sible. Two very well-known examples,
diffusion coefficient of the metal ions de- identified by neutron scattering, are the
pends directly on vacancy concentration Koch – Cohen clusters in Fe1– d O (Koch
(see Sec. 3.3.4), then the tracer diffusion and Cohen, 1969) and the Willis cluster in
coefficient will depend on oxygen partial UO2 + x (Willis, 1978). Although attempts
¥
pressure in the same way that either [VM ] have been made to incorporate such defects
or [V¢M ] does. A measurement of the tracer into mass-action equations, it is probably
diffusion coefficient as a function of oxy- fair to say that the number of adjustable pa-
3.5 Experimental Methods for Measuring Diffusion Coefficients 231
rameters thereby resulting and the lack of stoichiometry (cation vacancies) in the cat-
any precise independent information on the ion sublattice) is generally well understood
mobility or lifetime of the defect complexes except at large deviations from stoichiome-
make any conclusions reached rather spec- try (see the review by Peterson (1984)).
ulative. New statistical mechanical ap- Other oxides such as MgO and Al2O3 , al-
proaches have considerable promise, how- though apparently simple, are less well
ever (see the review by Murch (1995)). understood because of the low intrinsic de-
Somewhat analogous to changing the defect population and the presence of extrin-
fect concentration, i.e., deviation from stoi- sic defects coming from impurities (Subba-
chiometry by changing the oxygen partial Rao, 1985). An understanding of oxygen
pressure, is the doping of ionic crystals diffusion in oxides has been hampered by
with ions of a different valence from the the lack of suitable radioisotopes, but
host metal ion in order to produce defects. methods using 18O and secondary ion mass
As an example, consider the solubility of spectrometry (SIMS) have rapidly changed
an oxide ᑧ2O3 in an oxide MO2 . The im- the situation (Rothman, 1990). Ionic con-
plication in the stoichiometry ᑧ2O3 is that ductivity (where possible) (see Sec. 3.3.2
the ᑧ ions have a valence of + 3, compared for the relation to diffusion) has tradition-
with + 4 for M in the oxide MO2 . For ally been the measurement which has given
charge neutrality reasons the doped oxide much more information on oxygen move-
adopts either oxygen vacancies, metal in- ment; see, for example, the review by
terstitials, or some electronic defect, the Nowick (1984).
choice being dictated by energetics. Let us
consider the first of these:
ᑧ2O3 Æ VO•• + 2 ᑧ M
¢ + 3 OO (3-201) 3.5 Experimental Methods for
Measuring Diffusion Coefficients
where ᑧ M ¢ is a ᑧ ion on a M site. The site
fraction of oxygen vacant sites is not quite In this section we briefly discuss the
directly proportional to the dopant site more frequently used methods of measur-
fraction because extra normal oxygen sites ing diffusion coefficients.
are created in the process. The diffusion
coefficient of oxygen here is proportional
3.5.1 Tracer Diffusion Methods
to the concentration of vacant oxygen sites.
For very large deviations from stoichiom- By far the most popular and, if per-
etry, or high dopant concentrations, those formed carefully, the most reliable of the
extrinsic defects greatly outnumber the in- experimental methods for determining
trinsic defects. However, at low dopant con- “self” and impurity diffusion coefficients
centrations (or small deviations from stoi- is the thin-layer method. Here, a very thin
chiometry), all the various mass-action laws layer of radiotracer of the diffusant is de-
must be combined. This complicates the posited at the surface of the sample. The
analysis, and space prevents us from dealing deposition can be done by evaporation,
with it here. For a more detailed discussion electrochemical methods, decomposition
of this and the foregoing, we refer to Kof- of a salt, sputtering, etc. The diffusant is
stad (1972) and Schmalzried (1995). permitted to diffuse for a certain time t at
Generally, cation diffusion in the oxides high temperature, “high” being a relative
of the transition metals (which show non- term here. If the thickness of the layer de-
---
posited is much smaller than ÷D t , then Eq. hard radiation the method can give com-
(3-8) describes the time evolution of the parable accuracy to the counting of sec-
concentration – depth profile. After diffu- tions.
sion, the concentration – depth profile in A few elements do not have convenient
the sample is established by sectioning and radioisotopes. The diffusion part of the ex-
counting the radioactivity in each section. periment is still performed in much the
A number of techniques are available for same way but now with stable isotopes. In
sectioning. For thick sections ≥ 3 µm, me- some cases, e.g., 18O, the diffusant source
chanical grinding is the standard method. is in the gas phase, although a thin source
Microtomes can be used for sections of of oxide containing 18O can be deposited in
about 1 µm, electrochemical methods for some cases. After diffusion, nuclear reac-
≥ 5 nm, and sputtering for ≥ 1 nm. The dif- tion analysis of 18O can be used (see, for
fusion coefficient is obtained from the example, the review by Lanford et al.
slope of the ln C (x, t) versus x 2 plot (see (1984)), in order to establish the concentra-
Eq. (3-8)). Fig. 3-22 shows an exemplary tion profile. Much more popular recently,
tracer concentration profile of this type especially for oxygen, is SIMS (see, for ex-
(Mundy et al., 1971). ample, the reviews by Petuskey (1984),
A variant of the method is the “residual Kilner (1986) and Manning et al. (1996)).
activity” or “Gruzin (1952) method”. Excellent detailed (and entertaining) ex-
Rather than counting the activity in each positions and critiques of all methods
section, the activity remaining in the sam- available for measuring tracer diffusion co-
ple is determined. This requires an integra- efficients in solids have been provided by
tion of Eq. (3-8) and a knowledge of the ra- Rothman (1984, 1990).
diation absorption characteristics of the
material. At the limits of very soft or very
3.5.2 Chemical Diffusion Methods
The chemical or interdiffusion coeffi-
cient can be determined in a variety of
ways. For binary alloys, the traditional
method has been to bond together two sam-
ples of different concentrations. Interdiffu-
sion is then permitted to occur for a time t
and at high temperature. The concentration
profile can be established by sectioning
and chemical analysis, but since about the
late 1960s the use of an electron micro-
probe has been the usual procedure for ob-
taining the concentration profile directly. If
the compositions of the starting samples
are relatively close, and the interdiffusion
coefficient is not highly dependent on com-
position, then analysis of the profile with
Figure 3-22. A tracer concentration profile for self- Eq. (3-12) can lead directly to the interdif-
diffusion in potassium at 35.5 °C (Mundy et al., fusion coefficient D̃ at the average compo-
1971). sition. The more usual procedure is to use
3.5 Experimental Methods for Measuring Diffusion Coefficients 233
the Boltzmann –Matano graphical integra- of 18O (exchanging with 16O) can be moni-
tion analysis (see Eqs. (3-13) and (3-14)) to tored in the gas phase with a mass spec-
obtain the interdiffusion coefficient and its trometer in order to lead to the tracer diffu-
composition dependence. If fine insoluble sion coefficient of oxygen (see, for exam-
wires (to act as a marker) are also incorpo- ple, Auskern and Belle (1961)). A draw-
rated at the interface of the two samples, back of methods which do not rely on sec-
then the marker shift or Kirkendall shift tioning is the necessary assumption of
with respect to the original interface can rapid gas/surface reactions.
be measured. This shift, in association
with the interdiffusion coefficient, can be
3.5.3 Diffusion Coefficients
used to determine the intrinsic diffusion
by Indirect Methods
coefficients of both alloy components at
the composition of the marker (see Sec. There are a number of methods for ob-
3.2.2.4). taining diffusion coefficients that do not
The rate of absorption or desorption of depend on solutions of Fick’s Second Law.
material from the sample can also be used Unfortunately space does not permit a de-
to determine the chemical diffusion coeffi- tailed discussion of these. We shall men-
cient. This method is useful only where one tion them here for completeness and direct
component is fairly volatile, such as oxy- the reader to the appropriate sources. Often
gen in some nonstoichiometric oxides and these methods extend the accessible tem-
hydrogen in metals. The surface composi- perature range for diffusion measurements.
tion is normally assumed to be held con- In all cases specific atomic models need to
stant. The concentration profile can be de- be introduced in order to extract a diffusion
termined by sectioning or electron micro- coefficient. Inasmuch as there are inevita-
probe analysis and can be analysed with bly approximations in a model, the result-
the aid of Eq. (3-9) to yield the interdiffu- ing diffusion coefficients may not be as re-
sion coefficient. The Boltzmann – Matano liable as those obtained from a concentra-
analysis can be used if there is a composition profile. In some cases additional corre-
tion dependence. lation information can be extracted, e.g.,
The amount of material absorbed or de- nuclear methods. The diffusion coefficients
sorbed from the couple, e.g., weight found from these methods are essentially
change, can be used with Eq. (3-11) to “self” diffusion coefficients but chemical
yield the interdiffusion coefficient. Eqs. (3- diffusion coefficients for interstitial solutes
---
9) and (3-11) refer to situations when ÷D t are obtained from the Gorsky effect.
is very small compared with the geometry
of the sample, i.e., diffusion into an infinite
3.5.3.1 Relaxation Methods
sample is assumed. When the sample must
be assumed to be finite, e.g., fine particles, In the relaxation methods net atomic mi-
other, generally more complex, solutions gration is due to external causes such as
are required; see Crank (1975). A closely stress of magnetic field (Nowick and Berry,
related experiment can be performed where 1972). The best known phenomena are (a)
the gas phase is chemically in equilibrium the Gorsky effect (Gorsky, 1935); for a re-
with the solid, but is dosed with a stable view see Alefeld et al. (1970) and for an in-
isotope. An example is 18O, a stable iso- troduction see Borg and Dienes (1988); (b)
tope of oxygen. The absorption by the solid the Snoek effect (Snoek, 1939); for an ex-
position see Wert (1970) and for an intro- Most applications have been to solids in-
duction see Borg and Dienes (1988); and corporating the rather ideal isotope 57Fe.
(c) the Zener effect (Zener, 1947, 1951); See Mullen (1984) for a detailed discus-
for an introduction see Bocquet et al. sion of MBS and Vogl and Feldwisch (1998)
(1996). for several recent examples of applica-
tions.
3.5.3.2 Nuclear Methods
3.5.4 Surface Diffusion Methods
A number of nuclear methods have be-
come of increasing importance for deter- The techniques for measuring surface
mining diffusion coefficients in solids. The diffusion are generally quite different from
first of these is quasielastic neutron scat- solid-state techniques. For short distance,
tering (QNS), which has often been used microscopic or “intrinsic” diffusion (see
for situations where the diffusion coeffi- Sec. 3.2.2.5), the method of choice is the
cients are larger than about 10–11 m2 s –1 field ion microscope. Much elegant work
and the diffusing species exhibits a reason- has been carried out with this technique by
ably high scattering cross-section. Most Erlich and co-workers; see for example Er-
of the applications have been to hydrogen lich and Stott (1980) and Erlich (1980).
(protonic) in metals (see, for example, With this method single atoms can be im-
Janot et al., 1986). For a detailed introduc- aged and followed. Another method, this
tion to the subject see Lechner (1983) and time using the field electron microscope
Zabel (1984). The second method is nu- (Lifshin, 1992), correlates fluctuations of
clear magnetic resonance (NMR). NMR is an emission current from a very small area
especially sensitive to interactions of the with density fluctuations arising from
nuclear moments with fields produced by surface diffusion in and out of the probe
their local environments. Diffusion of a nu- area (Chen and Gomer, 1979). Other meth-
clear moment can cause variations in these ods include quasi-elastic scattering of
fields and can significantly affect the ob- low energy He atoms (formally analogous
served resonance. In particular, diffusion to quasi-elastic neutron scattering) and re-
affects a number of relaxation times in laxation measurements, making use of dep-
NMR. In favorable cases diffusion coeffi- osition of the adsorbate in a non-equilib-
cients between 10–18 and 10–10 m2 s –1 are rium configuration, followed by annealing
accessible. For a detailed introduction to which permits relaxation to equilibrium.
the subject see Stokes (1984) and Heitjans Techniques useful for following the relaxa-
and Schirmer (1998). We make special tion process at this microscopic level in-
mention of pulsed field gradient (PFG) clude pulsed molecular beam combined
NMR, which has been found to be espe- with fast scanning IR interferometry
cially useful for studying anomalous diffu- (Reutt-Robey et al., 1988) and work func-
sion (Kärger et al., 1998). Finally, we mention measurements (Schrammen and Hölzl,
tion Mössbauer spectroscopy (MBS), which 1983).
shows considerable promise for under- Many methods exist for long distance or
standing diffusion processes in solids. The macrosopic diffusion. When the diffusing
general requirement is that the diffusion species is deposited as a source and the ap-
coefficient of Mössbauer active isotope propriate geometries are known then scan-
should be larger than about 10–13 m2 s –1. ning for the concentration profile followed
3.7 References 235
by processing with the appropriate solution 3.7 References

to Fick’s Second Law (the Diffusion Equa-
tion) Eq. (3-5) will give the mass transfer Adda, Y., Philibert, J. (1966), La Diffusion dans les
diffusion coefficient. Radioactive tracers Solides. Paris: Presses Universitaires de France.
Alefeld, G. J., Völkl, J., Schaumann, G. (1970),
can certainly be used for this (Gjostein, Phys. Status Solidi 37, 337.
1970). The concentration profile (in the Allnatt, A. R. (1982), J. Phys. C 15, 5605.
case of hetero-diffusion) can also be deter- Allnatt, A. R., Allnatt, E. L. (1984), Phil. Mag. A 49,
625.
mined using scanning SIMS, local XPS, Allnatt, A. R., Allnatt, E. L. (1991), Phil. Mag. A 64,
scanning Auger, scanning EM and scan- 341.
ning STM (Bonzel, 1990). Of interest when Allnatt, A. R., Lidiard, A. B. (1987), Rep. Prog.
Phys. 50, 373.
the atoms are weakly adsorbed is laser Allnatt, A. R., Lidiard, A. B. (1993), Atomic Trans-
induced thermal desorption (LITD) (Vis- port in Solids. Cambridge: Cambridge University
wanathan et al., 1982). The field electron Press.
Arita, M., Koiwa, M., Ishioka, S. (1989), Acta
microscope can be used to image the diffu- Metall. 37, 1363.
sion front of adatoms migrating into a re- Askill, J. (1970), Tracer Diffusion Data for Metals,
gion of clean surface (Gomer, 1958). Fi- Alloys, and Simple Oxides. New York: IFI /
Plenum.
nally, the “capillary method”, probably the Auskern, A. B., Belle, J. (1961), J. Nucl. Mater. 3,
best known of the macroscopic methods, 267.
starts with a perturbed surface, usually in a Bakker, H. (1984), in: Diffusion in Crystalline Sol-
ids: Murch, G. E., Nowick, A. S. (Eds.). New
periodic way, say by sinusoidal grooving. York: Academic, p. 189.
There is now a driving force to minimize Bakker, H., Stolwijk, N. A., van der Meij, L., Zuu-
the surface Gibbs energy. The time depen- rendonk, T. J. (1976), Nucl. Metall. 20, 96.
Balluffi, R. W. (1984), in: Diffusion in Crystalline
dence of the decreasing amplitude of, say, Solids: Murch, G. E., Nowick, A. S. (Eds.). New
the sinusoidal profile, can be processed to York: Academic, p. 320.
give the mass transfer diffusion coefficient; Batchelor, G. K. (1976), J. Fluid Mech. 74, 1.
Beke, D. L. (Ed.) (1998), Diffusion in Semiconduc-
see Bonzel (1990). tors and in Non-Metallic Solids, Vol. 33 (Sub-
The reader is referred to a number of the volume A: Diffusion in Semiconductors), Landolt-
fine reviews in the area, e.g., Rhead Börnstein New Series, Berlin: Springer.
Beke, D. L. (Ed.) (1999), Diffusion in Semiconduc-
(1989), Bonzel (1990), and the book edited tors and in Non-Metallic Solids, Vol. 33 (Sub-
by Vu Thien Binh (1983). The first of these volume B: Diffusion in Non-Metallic Solids),
relates surface diffusion to a number of Landolt-Börnstein New Series, Berlin: Springer.
Belova, I. V., Murch, G. E. (1997), Phil. Mag. A 75,
technologically important processes such 1715.
as thin film growth, sintering, and cataly- Belova, I. V., Murch, G. E. (1998), Acta Mater. 46,
sis. The second is a comprehensive compi- 849.
Belova, I. V., Murch, G. E. (2000), Phil. Mag. A 80,
lation of surface diffusion data on metals 1469.
and a detailed survey of experimental Bénière, M., Chelma, M., Bénière, F. (1976), J. Phys.
methods. Chem. Solids 37, 525.
Bérces, G., Kovács, I. (1983), Phil. Mag. A 48, 883.
Binh, Vu Thien (Ed.) (1983), Surface Mobilities on
Solid Materials, NATO-ASI Series B, Vol. 86.
New York: Plenum Press.
3.6 Acknowledgements Bocquet, J. L. (1972), Rep. CEA-R-4292. Saclay
(France): Commissariat à l’Energie Atomique.
I thank my colleagues in diffusion from Bocquet, J. L. (1987), Res. Mech. 22, 1.
whom I have learnt so much. This work Bocquet, J. L., Brébec, G., Limoge, Y. (1996), in:
Physical Metallurgy, 4th edn.: Cahn, R. W., Haa-
was supported by the Australian Research sen, P. (Eds.). Amsterdam: North Holland, p. 535.
Council. Boltzmann, L. (1894), Ann. Phys. 53, 960.
Bonzel, H. P. (1990), Diffusion in Metals and Alloys, Gomer, R. (1958), Discussions Faraday Soc. 28, 23.
in: Landolt-Börnstein, Vol. 13: Mehrer, H. (Ed.). Gorsky, W. S. (1935), Z. Phys. Sowjetunion 8, 457.
Berlin: Springer-Verlag. Gösele, U., Frank, W., Seeger, A. (1980), Appl. Phys.
Borg, R. J., Dienes, G. J. (1988), An Introduction to 23, 361.
Solid State Diffusion. New York: Academic. Gruzin, P. L. (1952), Dokl. Acad. Nauk. SSSR 86,
Brandes, E. A. (Ed.) (1983), Smithells Metal Refer- 289.
ence Book, 6th edn. London: Butterworths. Gupta, D., Oberschmidt, J. (1984), in: Diffusion in
Carlson, P. T. (1978), Met. Trans. A 9, 1287. Solids: Recent Developments: Dayananda, M. A.,
Carslaw, H. S., Jaeger, J. C. (1959), Conduction of Murch, G. E. (Eds.). Warrendale: TMS/AIME,
Heat in Solids. New York: Oxford University Press. p. 121.
Catlow, C. R. A. (1977), Proc. Roy. Soc. London, Ser. Gupta, D., Campbell, D. R., Ho, P. S. (1978), In: Thin
A 353, 533. Films – Interdiffusion and Reactions: Poates, J.
Catlow, C. R. A., Mackrodt, W. C. (1982), Computer M., Tu, K. N., Mayer, J. W. (Eds.). New York: Wi-
Simulation of Solids (Lecture Notes in Physics, ley, p. 61.
Vol. 166). Berlin: Springer. Harrison, L. G. (1961), Trans. Faraday Soc. 57,
Chemla, M. (1956), Ann. Phys. Paris 13, 959. 1191.
Chen, J. R., Gomer, R. (1979), Surface Science 79, Hart, E. W. (1957), Acta Metall. 5, 597.
413. Hehenkamp, Th., Faupel, F. (1983), Acta Metall. 31,
Chen, W. K., Peterson, N. L. (1972), J. Phys. Chem. 691.
Solids 33, 881. Hehenkamp, Th., Schmidt, W., Schlett, V. (1980),
Chen, W. K., Peterson, N. L. (1973), J. Phys. (Paris) Acta Metall. 28, 1715.
34, C 9. Heitjans, P., Schirmer, A. (1998), in: Diffusion in
Crank, J. (1975), The Mathematics of Diffusion. New Condensed Matter: Kärger, J., Heitjans, P., Haber-
York: Oxford University Press. landt, R. (Eds.). Wiesbaden: Vieweg, p. 116.
Darken, L. S. (1948), Trans. AIME 175, 184. Hempelmann, R. (1984), J. Less-Common Met. 101,
De Bruin, H. J., Murch, G. E., Bakker, H., Van der 69.
Meij, L. P. (1975), Thin Solid Films 25, 47. Ho, Y. K. (1982), Thesis. London: Imperial College.
Dieckmann, R. (1977), Z. Phys. Chem. N. F. 107, Howard, R. E. (1966), Phys. Rev. 144, 650.
189. Howard, R. E., Lidiard, A. B. (1964), Rep. Prog.
Dorn, J. E., Mitchell, J. B. (1966), Acta Metall. 14, Phys. 27, 161.
70. Howard, R. E., Manning, J. R. (1967), Phys. Rev.
Elcock, E. W., McCombie, C. W. (1958), Phys. Rev. 154, 561.
109, 605. Huntington, H. B. (1975), in: Diffusion in Solids: Re-
Erlich, G. (1980), J. Vac. Sci. Technol. 17, 9. cent Developments: Nowick, A. S., Burton, J. J.
Erlich, G., Stott, K. (1980), Ann. Rev. Phys. Chem. (Eds.). New York: Academic, p. 303.
31, 603. Huntington, H. B., Ghate, P. B. (1962), Phys. Rev.
Faupel, F., Hehenkamp, Th. (1986), Phys. Rev. B 34, Lett. 8, 421.
2116. Jacucci, G. (1984), in: Diffusion in Crystalline Sol-
Faupel, F., Hehenkamp, Th. (1987), Acta Metall. 35, ids: Murch, G. E., Nowick, A. S. (Eds.). New
771. York: Academic, p. 431.
Fisher, J. C. (1951), J. Appl. Phys. 22, 74. Jain, H., Gupta, D. (Eds.) (1994), Diffusion in Amor-
Flynn, C. P. (1972), Point Defects and Diffusion. Ox- phous Materials. Warrendale: TMS.
ford: Clarendon Press. Janot, C., Petry, W., Richter, D., Springer, T. (Eds.)
Foiles, S. M., Daw, M. S. (1987), J. Mater. Res. 2, 5. (1986), Atomic Transport and Defects in Materials
Frank, F. C., Turnbull, D. (1956), Phys. Rev. 104, by Neutron Scattering. Berlin: Springer-Verlag.
617. Jones, M. J., Le Claire, A. D. (1972), Phil. Mag. 26,
Frank, W., Gösele, U., Mehrer, H., Seeger, A. (1984), 1191.
in: Diffusion in Crystalline Solids: Murch, G. E., Kaur, I., Mishin, Y., Gust, W. (1995), Fundamentals
Nowick, A. S. (Eds.). New York: Academic, p. 64. of Grain and Interphase Boundary Diffusion. Chi-
Fredericks, W. J. (1975), in: Diffusion in Solids: Re- chester: Wiley.
cent Developments: Nowick, A. S., Burton, J. J. Kärger, J., Fleischer, G., Roland, U. (1998), in: Diffu-
(Eds.). New York: Academic, p. 381. sion in Condensed Matter: Kärger, J., Heitjans, P.,
Freer, R. (1980), J. Mater. Sci. 15, 803. Haberlandt, R. (Eds.). Wiesbaden: Vieweg, p. 40.
Friauf, R. J. (1957), Phys. Rev. 105, 843. Kelly, S. W., Scholl, C. A. (1987), J. Phys. C 20,
Fukai, Y., Sugimoto, H. (1985), Adv. Phys. 34, 263. 5293.
Ghate, P. B. (1964), Phys. Rev. A 113, 1167. Kikuchi, R., Sato, H. (1969), J. Chem. Phys. 51, 161.
Gjostein, N. A. (1970), in: Techniques of Metals Re- Kikuchi, R., Sato, H. (1970), J. Chem. Phys. 53,
search, Vol. IV, Part 2: Bunshah, R. F. (Ed.). New 2707.
York: Wiley, p. 405. Kikuchi, R., Sato, H. (1972), J. Chem. Phys. 57, 4962.
3.7 References 237
Kilner, J. A. (1986), Mat. Sci. Forum 7, 205. Matano, C. (1933), Jpn. Phys. 8, 109.
Kirkaldy, J. S., Young, D. J. (1987), Diffusional in McKee, R. A. (1980 a), Phys. Rev. B 21, 4269.
the Condensed State. Brookfield: The Institute of McKee, R. A. (1980 b), Phys. Rev. B 22, 2649.
Metals. McKee, R. A. (1981), Phys. Rev. B 23, 1609.
Koch, F., Cohen, J. B. (1969), Acta Crystallogr., Mehrer, H. (1990), Diffusion in Metals and Alloys,
Sect. B 25, 275. Landolt-Börnstein New Series, Berlin: Springer.
Kofstad, P. (1972), Nonstoichiometry, Diffusion and Mehrer, H. (1998), in: Diffusion in Condensed Mat-
Electrical Conductivity in Binary Metal Oxides. ter: Kärger, J., Heitjans, P., Haberlandt, R. (Eds.).
New York: Wiley. Wiesbaden: Vieweg, p. 1.
Kwok, T., Ho, P. S., Yip, S. (1984), Phys. Rev. B 29, Mishin, Y. (1997), Defect Diffusion Forum 143 –147,
5363. 1357.
Lam, N. Q., Rothman, S. J., Mehrer, H., Nowicki, L. Mishin, Y., Herzig, C., Bernardini, J., Gust, W.
J. (1973), Phys. Status Solidi B 57, 225. (1997), Int. Mat. Reviews 42, 155.
Lanford, W. A., Benenson, R., Burman, C., Wielun- Mohan Rao, M., Ranganathan, S. (1984), Mat. Sci.
ski, L. (1984), in: Nontraditional Methods in Dif- Forum 1, 43.
fusion: Murch, G. E., Birnbaum, H. K., Cost, J. R. Moleko, L. K., Allnatt, A. R., Allnatt, E. L. (1989),
(Eds.). Warrendale: TMS/AIME, p. 155. Phil. Mag. A 59, 141.
Laskar, A. L. (1990), in: Diffusion in Materials: Monty, C. (1992), in: Diffusion in Solids: Unsolved
Laskar, A. L., Bocquet, J. L., Brébec, G., Monty, Problems: Murch, G. E. (Ed.). Zürich: Trans.
C. (Eds.). Dordrecht: Kluwer, Academic, p. 459. Tech. Publications, p. 259.
Laskar, A. L. (1992), in: Diffusion in Solids: Un- Mullen, J. G. (1984), in: Nontraditional Methods in
solved Problems: Murch, G. E. (Ed.). Zürich: Diffusion, Murch, G. E., Birnbaum, H. K., Cost, J.
Trans. Tech. Publications, p. 207. R. (Eds.). Warrendale: TMS/AIME, p. 59.
Lechner, R. E. (1983), in: Mass Transport in Solids: Mundy, J. N. (1992), in: Diffusion in Solids: Un-
Bénière, F., Catlow, C. R. A. (Eds.). London: solved Problems: Murch, G. E. (Ed.). Zürich:
Plenum, p. 169. Trans. Tech. Publications, p. 1.
Le Claire, A. D. (1963), Br. J. Appl. Phys. 14, 351. Mundy, J. N., Miller, T. E., Porte, R. J. (1971), Phys.
Le Claire, A. D. (1970), in: Physical Chemistry – An Rev. B 3, 2445.
Advanced Treatise, Vol. 10: Eyring, H., Henderson, Murch, G. E. (1975), J. Nucl. Mater. 57, 239.
D., Jost, W. (Eds.). New York: Academic, p. 261. Murch, G. E. (1981), J. Phys. Chem. Solids 42, 227.
Le Claire, A. D. (1975), in: Treatise in Solid State Murch, G. E. (1982 a), Phil. Mag. A 46, 575.
Chemistry, Vol. 4, Reactivity of Solids: Hannay, N. Murch, G. E. (1982 b), Phil. Mag. A 45, 685.
B. (Ed.). New York: Plenum, p. 1. Murch, G. E. (1982 c), Solid State Ionics 6, 295.
Le Claire, A. D. (1978), J. Nucl. Mater. 69/70, 70. Murch, G. E. (1982 d), Solid State Ionics 7, 177.
Le Claire, A. D. (1981), Phil. Mag. A 43, 911. Murch, G. E. (1984 a), in: Diffusion in Crystalline
Le Claire, A. D. (1992), in: Diffusion in Solids: Un- Solids: Murch, G. E., Nowick, A. S. (Eds.). New
solved Problems: Murch, G. E. (Ed.). Zürich: York: Academic, p. 379.
Trans. Tech. Publications, p. 19. Murch, G. E. (1984 b), Phil. Mag. A 49, 21.
Le Claire, A. D., Rabinovitch, A. (1984), in: Diffu- Murch, G. E. (1995), Rad. Effects and Defects in Sol-
sion in Crystalline Solids: Murch, G. E., Nowick, ids 134, 1.
A. S. (Eds.). New York: Academic, p. 259. Murch, G. E., Belova, I. V. (1998), in: Diffusion
Lidiard, A. B. (1986), Acta Metall 34, 1487. Mechanisms in Crystalline Materials: Mishin, Y.,
Lifshin, E. (1992), Characterization of Materials, Vogl, G., Cowern, N., Catlow, R., Farkas, D.
Part I, Materials Science and Technology: R. W. (Eds.). MRS Symposium Proceedings 527. War-
Cahn, P. Haasen, E. J. Kramer (Eds.), Vol. 2A. rendale: MRS, p. 135.
Weinheim: VCH. Murch, G. E., Catlow, C. R. A. (1987), J. Chem. Soc.,
Lim, S. H., Murch, G. E., Oates, W. A. (1990), Phil. Faraday Trans. II 83, 1157.
Mag. B 61, 337. Murch, G. E., Dyre, J. C. (1989), CRC Crit. Rev.
Manning, J. R. (1959), Phys. Rev. 113, 1445. Solid State Mat. Sci. 15, 345.
Manning, J. R. (1964), Phys. Rev. 136, 1758. Murch, G. E., Rothman, S. J. (1979), Phil. Mag. A 43,
Manning, J. R. (1968), Diffusion Kinetics for Atoms 229.
in Crystals. Princeton: Van Nostrand. Murch, G. E., Rothman, S. J. (1985), Diffusion De-
Manning, J. R. (1970), Met. Trans. 1, 499. fect Data 42, 17.
Manning, J. R. (1971), Phys. Rev. B 4, 1111. Murch, G. E., Thorn, R. J. (1978), Phil. Mag. A 38,
Manning, J. R. (1975), in: Mass Transport in Ceram- 125.
ics: Cooper, A. R., Heuer, A. H. (Eds.). New York: Murch, G. E., Thorn, R. J. (1979), J. Phys. Chem.
Plenum, p. 75. Solids 40, 389.
Manning, P. S., Sirman, J. D., Kilner, J. A. (1996), Murray, A. D., Murch, G. E., Catlow, C. R. A.
Solid State Ionics 93, 125. (1986), Solid State Ionics 18/19, 196.
Nowick, A. S. (1984), in: Diffusion in Crystalline Schrammen, P., Hölzl, J. (1983), Surface Science
Solids: Murch, G. E., Nowick, A. S. (Eds.). New 130, 203.
York: Academic, p. 143. Seeger, A. (1997), Defect Diffusion Forum 143 –147,
Nowick, A. S., Berry, B. S. (1972), Anelastic Relax- 21.
ation in Crystalline Solids. New York: Academic. Siegel, W. (1978), J. Nucl. Mater. 69/70, 117.
Peterson, N. L. (1975), in: Diffusion in Solids: Re- Smith, R. P. (1953), Acta Metall. 1, 578.
cent Developments: Nowick, A. S., Burton, J. J. Snoek, J. L. (1939), Physica 6, 591.
(Eds.). New York: Academic, p. 116. Stark, J. P. (1972), J. Appl. Phys. 43, 4404.
Peterson, N. L. (1978), J. Nucl. Mater. 69/70, 3. Stark, J. P. (1974), Acta Metall. 22, 1349.
Peterson, N. L. (1983), Int. Met. Rev. 28, 65. Stark, J. P. (1976), Solid State Diffusion. New York:
Peterson, N. L. (1984), Mater. Sci. Forum 1, 85. Wiley.
Petuskey, W. T. (1984), in: Nontraditional Methods Stokes, H. T. (1984), in: Nontraditional Methods in
in Diffusion: Murch, G. E., Birnbaum, H. K., Cost, Diffusion, Murch, G. E., Birnbaum, H. K., Cost, J.
J. R. (Eds.). Warrendale: TMS/AIME, p. 179. R. (Eds.). Warrendale: TMS/AIME, p. 39.
Philibert, J. (1991), Atom Movements: Diffusion and SubbaRao, E. C. (1985), Diffusion Defect Data 41, 1.
Mass Transport in Solids, Paris: Les Editions de Suzuoka, T. (1961), Trans. Jpn. Inst. Met. 2, 25.
Physique. Vineyard, G. H. (1957), J. Phys. Chem. Solids 3, 121.
Pontikis, V. (1990), in: Diffusion in Materials: Viswanathan, R., Burgess, Jr., D. R., Stair, P. C.,
Laskar, A. L., Bocquet, J. L., Brébec, G., Monty, Weitz, E. (1982), J. Vac. Sci. Technol. 20, 605.
C. (Eds.). Dordrecht: Kluwer, Academic, p. 37. Vogl, G., Feldwisch, R. (1998), in: Diffusion in Con-
Reutt-Robey, J. E., Doren, D. J., Chabel, Y. J., densed Matter: Kärger, J., Heitjans, P., Haberlandt,
Christman, S. B. (1988), Phys. Rev. Lett 61, 2778. R. (Eds.), Wiesbaden: Vieweg, p. 144.
Rhead, G. E. (1989), Int. Materials Reviews 34, 261. Völkl, J., Alefeld, G. (1975), in: Diffusion in Solids:
Rothman, S. J. (1984), in: Diffusion in Crystalline Recent Developments: Nowick, A. S., Burton, J. J.
Solids: Murch, G. E., Nowick, A. S. (Eds.). New (Eds.). New York: Academic, p. 231.
York: Academic, p. 1. Wang, C. C., Akbar, S. A. (1993), Acta Metall.
Rothman, S. J. (1990), in: Diffusion in Materials: Mater. 41, 2807.
Laskar, A. L., Bocquet, J. L., Brébec, G., Monty, Warburton, W. K., Turnbull, D. (1975), in: Diffusion
C. (Eds.). Dordrecht: Kluwer, Academic, p. 269. in Solids: Recent Developments: Nowick, A. S.,
Rothman, S. J., Peterson, N. L. (1967), Phys. Rev. Burton, J. J. (Eds.). New York: Academic, p. 171.
154, 552. Wert, C. A. (1970), J. Phys. Chem. Solids 31, 1771.
Rothman, S. J., Peterson, N. L., Robinson, J. T. Willis, B. T. M. (1978), Acta Crystallogr., Sect. A 34,
(1970), Phys. Status Solidi 39, 635. 88.
Sato, H. (1970), in: Physical Chemistry – An Ad- Zabel, H. (1984), in: Nontraditional Methods in Dif-
vanced Treatise, Vol. 10: Eyring, H., Henderson, fusion, Murch, G. E., Birnbaum, H. K., Cost, J. R.
D., Jost, W. (Eds.). New York: Academic, p. 579. (Eds.). Warrendale: TMS/AIME, p. 1.
Sato, H. (1970), in: Nontraditional Methods in Diffu- Zener, C. (1947), Phys. Rev. 71, 34.
sion, Murch, G. E., Birnbaum, H. K., Cost, J. R. Zener, C. (1951), J. Appl. Phys. 22, 372.
(Eds.). Warrendale: TMS/AIME, p. 203. Zhang, L., Oates, W. A., Murch, G. E. (1988), Phil.
Sato, H. (1989), in: Superionic Conductors: Laskar, Mag. A 58, 937.
A. R., Chandra, S. (Eds.). New York: Academic, p. Zhang, L., Oates, W. A., Murch, G. E. (1989 a), Phil.
439. Mag. B 60, 277.
Sato, H., Kikuchi, K. (1971), J. Chem. Phys. 55, Zhang, L., Oates, W. A., Murch, G. E. (1989 b), Phil.
677/702. Mag. A 59, 171.
Schilling, W. (1978), J. Nucl. Mater. 69/60, 465. Zhang, L., Murch, G. E. (1997), Defect Diffusion
Schmalzried, H. (1995), Chemical Kinetics of Solids, Forum. 141, 59.
Weinheim: VCH.
4 Statistical Theories of Phase Transitions
Kurt Binder
Institut für Physik, Johannes Gutenberg-Universität Mainz,

Mainz, Federal Republic of Germany

4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 245
4.2 Phenomenological Concepts . . . . . . . . . . . . . . . . . . . . . . . . 246
4.2.1 Order Parameters and the Landau Symmetry Classification . . . . . . . . . 246
4.2.2 Second-Order Transitions and Concepts about Critical Phenomena
(Critical Exponents, Scaling Laws, etc.) . . . . . . . . . . . . . . . . . . . 260
4.2.3 Second-Order Versus First-Order Transitions; Tricritical
and Other Multicritical Phenomena . . . . . . . . . . . . . . . . . . . . . 269
4.2.4 Dynamics of Fluctuations at Phase Transitions . . . . . . . . . . . . . . . 277
4.2.5 Effects of Surfaces and of Quenched Disorder on Phase Transitions:
a Brief Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 279
4.3 Computational Methods Dealing with the Statistical Mechanics
of Phase Transitions and Phase Diagrams . . . . . . . . . . . . . . . . 283
4.3.1 Models for Order – Disorder Phenomena in Alloys . . . . . . . . . . . . . 283
4.3.2 Molecular Field Theory and its Generalization
(Cluster Variation Method, etc.) . . . . . . . . . . . . . . . . . . . . . . . 288
4.3.3 Computer Simulation Techniques . . . . . . . . . . . . . . . . . . . . . . 293
4.4 Concepts About Metastability . . . . . . . . . . . . . . . . . . . . . . . 298
4.5 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 303
4.6 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 304

ISBN: 3-527-30256-5
240 4 Statistical Theories of Phase Transitions

A prefactor of Friedel interaction
a lattice spacing
ai nearest-neighbor distance
Â, Â¢ critical amplitude of specific heats
B critical amplitude of the order parameter
c concentration
C specific heat
C phenomenological coefficient
c lattice spacing in z-direction
c cluster of geometric configuration
Dc concentration difference
Ĉ, Ĉ¢ critical amplitudes of ordering susceptibilities
ccoex concentration at coexistence curve
ci concentration of lattice site i
cijk … elastic constants
d dimensionality
D̂ critical amplitude at the critical isotherm
d* marginal dimensionality
E electric field
ei random vectors
el (k, x) phonon polarization vector
F [F (x)] Helmholtz energy function in operator form
f factor of order unity
F Helmholtz energy
DF* Helmholtz energy barrier
F (j ) Helmholtz energy functional
F̃ scaling function of the Helmholtz energy
f, f I, …, f IV, f coefficients
F0 background Helmholtz energy of disordered phase
fcg coarse-grained Helmholtz energy
fm n quadrupole moment tensor
Freg background term of the Helmholtz energy
g constant factor in the structure function
G reciprocal lattice vector
G (x) order parameter correlation function
G̃ (x / x ) scaling function of the correlation function
Ĝ critical amplitude of the correlation function
G (x , x ) order parameter correlation function
g nc multispin correlation function
H ordering field
H magnetic field
Ᏼ Hamiltonian
H̃ scaled ordering field
h, hc uniform magnetic field, critical value

HR random field (due to impurities)
i number and index for different lattice sites
J energy gained when two like atoms occupy neighboring sites
J exchange interaction
J̃ Fourier transform of the exchange interaction
Jij exchange interaction between spins at sites i and j (in cases where it is ran-
dom)
Jm strength of the magnetic interaction
Jnn nearest-neighbor interaction
Jnnn next-nearest-neighbor interaction
k wavevector of phonon
K1 , K2 phenomenological coefficients
kF Fermi wavenumber
L coarse-graining length
ᏸ Lagrangian
M period (in Fig. 4-10)
M magnetization
Ml mass of an atom
mm sublattice magnetization
Ms spontaneous magnetization
N degree of polymerization
na lattice sites in state a
p pressure
P polarization
p dielectric polarization
P (J) statistical distribution of random variable J
P ({Si }) probability that a configuration {Si } occurs
p+ probability for spin up
p– probability for spin down
pb parameter in Landau equation (bicritical points)
Q normal coordinate
qa primitive vectors of the reciprocal lattice
qmax value of q where c (q) is at a maximum
q* nonzero wavenumber
qEA ordering parameter for the Ising spin glass
r phenomenological coefficient in the Landau expansion
R effective range of interaction
R ratio of Jnn and Jnnn
R* critical radius of a droplet
R il site of atom l in unit cell i
r, r ¢ coefficient in the Landau equation
rc radius of polymer coil
S entropy
S (q) structure factor

S (q, w ) scattering function
Si , Sj unit vectors in direction of the magnetic moment
t time
t reduced temperature 1 – T/T0 (in Sec. 4.2.2)
T0 temperature where r changes sign
T1 temperature where metastable ordered phase vanishes
Tb bicritical point
tcr reduced temperature at crossover
TL Lifshitz point
Tl (x) transition temperature of superfluid 4He
Tt temperature of tricritical point
u phenomenological coefficient
u coefficient in Landau equation
U enthalpy
ul displacement vector
Ud surface of a d-dimensional sphere
v coefficient in the Landau equation
v ( |x |) Friedel potential
V total volume of the thermodynamic system
V (xi – xj ) pairwise interaction of two particles
Vab interaction between atoms A and B
w coefficient in the Landau equation
W action in classical mechanics
x, xn , xm , xa position vector and its components
z dynamical critical exponent
z coordinate (of position vector)
z coordination number
Z partition function
ᒗ scaling variable
a1m , a2 notation for phases in phase diagram

am magnetic phase
an nonmagnetic phase
a (x) short-range order parameter for metal alloys
a, a ¢ critical exponents related to specific heat
b exponent related to the order parameter
Ĝ critical amplitude of the susceptibility
g, g ¢ critical exponents related to the susceptibility
G0 phenomenological rate factor
d critical exponent at the critical isotherm
dmn Kronecker symbol
e binding energy of a particle
e ik strain tensor
h exponent related to the decay of correlations at Tc
l wavelength
l label of the phonon branch
L factor changing the length scale
m magnetic moment per spin
Dm chemical potential difference
m coex chemical potential at phase coexistence
n exponent related to the correlation length
x correlation length
xi concentration difference (Eq. (4-126))
r density
r (x) charge density distribution function
t characteristic time
j phase shift in Friedel potential
F order parameter
F (x) order parameter density
Fi concentration difference between two sublattices
F ms order parameter of metastable state
Fs spinodal curve
c (q) wavevector-dependent susceptibility
c el dielectric tensor
cT response function of the order parameter
cT isothermal susceptibility
cT staggered susceptibility
y concentration difference between two sublattices
y I, II ordering parameter for D03 structure
ya amplitude of mass density waves
w characteristic frequency (for fluctuations)
w̃ scaling function
AF antiferromagnetic
cg coarse-grained (as index)
coex coexistent (as index)
CV cluster variation method
DAG dysprosium aluminum garnet
ESR electron spin resonance
LRO long-range order
MC Monte Carlo (method)
MD molecular dynamics
MFA mean field approximation
MFA molecular field approximation
ms metastable state (as index)

NMR nuclear magnetic resonance
P paramagnetic
sc simple cubic
SRO short-range order
4.1 Introduction 245
4.1 Introduction (Tsakalakos, 1984; Gonis and Stocks, 1989;

Turchi and Gonis, 1994), magnetic materi-
This chapter deals with the general theo- als (De Jongh and Miedema, 1974; Ausloos
retical concepts of phase transitions of ma- and Elliott, 1983), ferroelectrics (Blinc and
terials and their basis in statistical thermo- Zeks, 1974; Jona and Shirane, 1962), liquid
dynamics. This field has seen extensive crystals (Pershan, 1988; De Gennes, 1974;
scientific research over many decades, and Chandrasekhar, 1992), polymeric materials
a vast amount of literature exists, e.g., the (Flory, 1953; De Gennes, 1979; Riste and
statistical theory of critical point phenom- Sherrington, 1989; Binder, 1994; Baus et
ena fills a series of monographs that so far al., 1995), etc., or on particular types of
consists of 17 volumes (Domb and Green, phase transitions, such as commensurate –
1972 – 1976; Domb and Lebowitz, 1983 – incommensurate transitions (Blinc and Le-
1997). Introductory texts require whole vanyuk, 1986; Fujimoto, 1997), multicriti-
books (Stanley, 1971; Ma, 1976; Patashin- cal transitions (Pynn and Skjeltorp, 1983),
skii and Pokrovskii, 1979; Yeomans, 1992), first-order transitions (Binder, 1987 a),
and so does a comprehensive account of martensitic transformations (Nishiyama,
the Landau theory of phase transitions 1979; Salje, 1990), glass transitions
(Tolédano and Tolédano, 1987). (Jäckle, 1986; Angell and Goldstein, 1986;
This chapter therefore cannot give an ex- Cusack, 1987; Zallen, 1983; Zarzycki,
haustive description of the subject; instead 1991), percolation transitions (Stauffer and
what is intended is a tutorial overview, Aharony, 1992), melting (Baus, 1987),
which gives the flavor of the main ideas, wetting transitions (Dietrich, 1988; Sulli-
methods, and results, with emphasis on the van and Telo da Gama, 1985), metal – insu-
aspects which are particularly relevant for lator transitions (Mott, 1974; Friedmann
materials science, and hence may provide a and Tunstall, 1978), etc. Other approaches
useful background for other chapters in this focus on particular ways of studying phase
book. Hence this chapter will contain transitions: from the random-phase approx-
hardly anything new for the specialist; for imation (Brout, 1965) and the effective-
the non-specialist it will give an initial or- field theory (Smart, 1966) to advanced
ientation and a guide for further reading. techniques such as field-theoretical (Amit,
To create a coherent and understandable 1984) or real space renormalization (Burk-
text, a necessarily arbitrary selection of hardt and van Leeuwen, 1982), and com-
material has been made which reflects the puter simulation studies of phase transi-
author’s interests and knowledge. Reviews tions applying Monte Carlo methods
containing complementary material have (Binder, 1979, 1984 a; Mouritsen 1984;
been written by De Fontaine (1979), who Binder and Heermann, 1988; Binder and
emphasizes the configurational thermody- Ciccotti, 1996) or the molecular dynamics
namics of solid solutions, and by Khacha- techniques (Ciccotti et al., 1987; Hockney
turyan (1983), who emphasizes the theory and Eastwood, 1988; Hoover, 1987; Binder
of structural transformations in solids and Ciccotti, 1996), etc. For particular
(which, from a different perspective, are models of statistical mechanics, exact de-
discussed by Bruce and Cowley, 1981). scriptions of their phase transitions are
There are also many texts which either con- available (Baxter, 1982). Of particular
centrate on phase transformations of par- interest also are the dynamics of critical
ticular types of materials, such as alloys phenomena (Enz, 1979) and the dynamics
of first-order phase transitions (Gunton are ubiquitous, e.g., diamond is the meta-
et al., 1983; Koch, 1984; Binder, 1984 b, stable modification of solid carbon whereas
1989; Haasen et al., 1984; Zettlemoyer, graphite is the stable phase, some com-
1969; Abraham, 1974; Gunton and Droz, ments about the statistical mechanics of
1983; Herrmann et al., 1992). The latter metastability are made in Sec. 4.4.
subject is treated in two other chapters
in this book (see the Chapters by Wagner
et al. (2001) and Binder and Fratzl (2001)) 4.2 Phenomenological Concepts
and will not be considered here.
In this chapter, we shall briefly discuss In this section, the main facts of the the-
the statistical thermodynamics of phase ory of phase transitions are summarized,
transitions on a phenomenological macro- and the appropriate terminology intro-
scopic level (Sec. 4.2), i.e., the Landau the- duced. Rather than aiming at completeness,
ory of first- and second-order transitions, examples which illustrate the spirit of the
critical and multicritical phenomena, and main approaches will be discussed, includ-
also the dynamics of fluctuations at phase ing a discussion of critical phenomena and
transitions, and the effects of quenched scaling laws.
disorder in solids. The second part (Sec.
4.3) is devoted to the complementary “mi-
4.2.1 Order Parameters and the Landau
croscopic” approach, where we start from
Symmetry Classification
“model Hamiltonians” for the system under
consideration. The statistical mechanics of Table 4-1 lists condensed matter systems
such models naturally leads to the consider- that can exist in several phases, depending
ation of some computational methods, i.e., on external thermodynamics parameters
methods based on the molecular field the- such as pressure, p, temperature, T, electric
ory and its generalizations, and methods or magnetic fields (E, H). We assume that
based on computer simulation techniques. an extensive thermodynamic variable can
Since in solid materials metastable phases be identified (i.e., one that is proportional
Table 4-1. Order parameters for phase transitions in various systems.
System Transition Order parameter
Liquid – gas Condensation/evaporation Density difference Dr = r l – rg

(2) (1)
Binary liquid mixture Unmixing Composition difference Dc = c coex – c coex
Nematic liquid Orientational ordering –12 ·3 (cos2q ) – 1Ò
Quantum liquid Normal fluid a suprafluid ·y Ò (y : wavefunction)
Liquid – solid Melting/crystallization rG (G = reciprocal lattice vector)
Magnetic solid Ferromagnetic (Tc ) Spontaneous magnetization M

Antiferromagnetic (TN ) Sublattice magnetization Ms
(2) (1)
Solid binary mixture Unmixing Dc = c coex – c coex
AB Sublattice ordering y = (Dc – DcI )/2
II
Dielectric solid Ferroelectric (Tc) Polarization P

Antiferroelectric (TN ) Sublattice polarization Ps
Molecular crystal Orientational ordering Ylm (J, j)
4.2 Phenomenological Concepts 247
to the volume of the system) which distin-

guishes between these phases (examples
are also given in Table 4-1), called the “or-
der parameter”. We shall denote the order
parameter as F, and the conjugate thermo-
dynamic variable, the “ordering field”, as
H. Using the thermodynamic potential, F,
which has as “natural variables” a field, H
(which is an “intensive” thermodynamic
variable, i.e., independent of the volume),
and the temperature, T, we have
F = – (∂F/∂H )T (4-1)
the other derivative being the entropy,
S = – (∂F/∂T )H (4-2)
As examples, consider a ferromagnet, where
the order parameter is the magnetization,
M, or a ferroelectric, where the order pa-
rameter is the dielectric polarization, P, Figure 4-1. The fluid-magnet analogy. On varying
the chemical potential m at m (T )
coex , the density r jumps
M = – (∂F/∂H)T (4-3 a) from the value at the gas branch of the gas – liquid co-
(1)
existence curve (rgas = r coex ) to the value at the liquid
P = – (∂F/∂E)T (4-3 b) branch of the coexistence curve (r liquid = r coex(2)
) (top
left). Similarly, on varying the (internal) magnetic
It is clear that such thermodynamic rela- field H, the magnetization M jumps from the nega-
tions can be written for any material, but a tive value of the spontaneous magnetization (– Ms ) to
quantity qualifies as an “order parameter” its positive value (top right). While this first-order
when a particular value of the “ordering liquid – gas transition occurs at a curve m coex (T ) in
the m – T-plane ending in a critical point (m c , Tc )
field” exists where the order parameter ex- where the transition is then of second order, the curve
hibits a jump singularity between two dis- where phases with positive and negative spontaneous
tinct values (Fig. 4-1). This means that for magnetization can coexist simply is H = 0 (T < Tc )
these values of the ordering field a first-or- (middle). The order parameter (density difference
der phase transition occurs, where a first Dr, or spontaneous magnetization Ms ) vanishes ac-
cording to a power law near Tc (bottom).
derivative of the thermodynamic potential
F exhibits a singularity. At this transition,
two phases can coexist; e.g., at the liq- be characterized by the same critical expo-
uid – gas transition for a chemical potential nents (see below), there is also an impor-
m = m coex (T ), two phases with different tant distinction that in the magnetic prob-
density coexist; and in a ferromagnet at lem, the Hamiltonian possesses a symme-
zero magnetic field, phases with opposite try with respect to the change of sign of the
sign of spontaneous magnetization can co- magnetic field; reversing this sign and also
exist. Although the fluid-magnet analogy reversing the sign of the magnetization
(Fig. 4-1) goes further, since the first-order leave the Hamiltonian invariant. Because
lines in the ( m , T ) or (H, T ) plane in both of this symmetry, the transition line must
cases end in critical points which may even occur at H = 0. Conversely, if the system
at H = 0 is in a monodomain state with cannot be measured directly, they never-

either positive or negative spontaneous theless provide a useful conceptual frame-
magnetization, this symmetry is violated; work.
this is described as “spontaneous symmetry Another problem which obscures the
breaking”. No such obvious symmetry, on analogy between different phase transitions
the other hand, is identified for the liq- is the fact that we do not always wish to
uid – gas transition of fluids. Consequently, work with the corresponding statistical en-
the curve m coex (T ) is a nontrivial function sembles. For a liquid – gas transition, we
in the m – T-plane, and no simple symmetry can control the chemical potential via the
operation acting on the gas phase atoms ex- fluid pressure, and thus a grand-canonical
ists that would transform this phase into a ensemble description makes sense. How-
liquid, or vice versa. Similar lack of sym- ever, for the binary mixture (AB), the or-
metry between the phases is noted for un- dering field would be a chemical potential
mixing transitions in binary fluid or solid difference D m = mA – m B between the spe-
mixtures, where the order parameter is a cies. In an ensemble where D m is the exter-
(2) (1)
concentration difference, Dc = c coex – c coex , nally controlled variable A atoms could
cf. Table 4-1, whereas solid binary mix- transform into B, and vice versa. Experi-
tures which exhibit sublattice ordering do mentally, of course, we do not usually have
possess a symmetry. The order parameter in mixtures in contact with a reservoir with
an alloy such as brass (b-CuZn) is the dif- which they can exchange particles freely,
ference between the relative concentration instead mixtures are kept at a fixed relative
of the two sublattices, y = (Dc II – Dc I )/2. concentration c. (Such an equilibrium with
However, the two sublattices physically are a gas reservoir can only be realized for
completely equivalent; therefore, the Ha- interstitial alloys such as metal hydrides or
miltonian possesses a symmetry against the oxides.) Thus the experiment is described
interchange of the two sublattices, which by a canonical ensemble description, with
implies that y changes sign, just as the (T, c) being the independent variables, al-
(idealized!) ferromagnet does for H = 0 though the grand-canonical ensemble of
(Fig. 4-1). In this example of an alloy mixtures is still useful for analytical theo-
which undergoes an order – disorder transi- ries and computer simulation. Now, as is
tion where the permutation symmetry obvious from Fig. 4-1, in a canonical en-
between the two sublattices is spontane- semble description of a fluid, the first-
ously broken, the “ordering field” conju- order transition shows up as a two-phase
gate to the order parameter is a chemical coexistence region. Here, at a given den-
potential difference between the two sub- sity, r , with rgas < r < r liquid , the relative
lattices, and hence this ordering field is amounts of the two coexisting phases are
not directly obtainable in the laboratory. given by the lever rule if interfacial contri-
The situation is comparable to the case butions to the thermodynamic potential can
of simple antiferromagnets, the order pa- be neglected. The same is true for a binary
rameter being the “staggered magnetiza- mixture with concentration c inside the two-
(1) (2)
tion” (= magnetization difference between phase coexistence region, c coex < c < ccoex .
the sublattices), and the conjugate ordering These different statistical ensembles also
field would change sign from one sublat- have pronounced consequences on the dy-
tice to the other (“staggered field”). Al- namic properties of the considered sys-
though the action of such fields usually tems: in binary mixtures at constant con-
centration, and also in fluids at constant

density, the order parameter is a conserved
quantity, whereas for a fluid at constant
chemical potential, the order parameter is
not conserved. The latter situation occurs,
for instance, for fluid – gas transitions in
physisorbed monolayers at surfaces, which
can exchange molecules with the surround-
ing gas phase which is in equilibrium with
this adsorbed layer.
An important distinction to which we
turn next is the order of a phase transition.
In the examples shown in the upper part of
Fig. 4-1, a first derivative of the appropri-
ate thermodynamic potential has a jump
singularity and therefore such transitions
are called first-order transitions. However,
if we cool a ferromagnet down from the
paramagnetic phase in zero magnetic field,
the spontaneous magnetization sets in con-
tinuously at the critical temperature Tc
(lower part of Fig. 4-1). Similarly, on cool-
ing the alloy b-CuZn from the state where Figure 4-2. Schematic variation with temperature T
it is a disordered solid solution (in the plotted for several quantities near a critical point Tc :
body-centered cubic phase), sublattice or- specific heat CH (top), ordering “susceptibility” c T
(middle part), and correlation length x of order pa-
dering sets in continuously at the critical
rameter fluctuations (bottom). The power laws which
temperature (Tc = 741 K, see Als-Nielsen hold asymptotically in the close vicinity of Tc are in-
(1976)). Whereas the first derivatives of dicated.
the thermodynamic potential at these con-
tinuous phase transitions are smooth, the
second derivatives are singular, and there- In this context (see also Fig. 4-1), a, a ¢, b ,
fore these transitions are also called sec- g , and g ¢ are critical exponents, while Â, Â¢,
ond-order transitions. For example, in a B̂, Ĉ, and Ĉ¢ are called critical amplitudes.
ferromagnet the isothermal susceptibility Note that B̂ and b refer to the spontaneous
c T and the specific heat typically have magnetization, the order parameter (Fig. 4-1)
power law singularities (Fig. 4-2)
Ms = Bˆ (1 − T / Tc )b (4-6)
c T ≡ − ( ∂ F/ ∂H )T
2 2 T → Tc
⎧Cˆ ( T / Tc − 1) − g , T > Tc Behavior of the specific heat such as de-

= ⎨ −g′
(4-4) scribed by Eq. (4-5) immediately carries
ˆ
⎩ ′ (1 − Tc / T ) , T < Tc
T → Tc C
over to systems other than ferromagnets
CH ≡ − T ( ∂ 2 F/ ∂T 2 ) H = 0 such as antiferromagnets, the liquid –gas
system near its critical point, and brass
⎧ Aˆ ( T / Tc − 1) − a , T > Tc near its order – disorder transition; we must
= ⎨ −a′
(4-5)
ˆ
⎩ ′ (1 − T / Tc ) , T < Tc
T → Tc A remember, however, that H then means the
appropriate ordering field. In fact, this is parameter F as scalar quantities; although

also true for Eq. (4-4), but then the physical this is correct for the gas – fluid transition
significance of c T changes. For a two-sub- and for the unmixing transition of binary
lattice antiferromagnet, the ordering field mixtures, the uniaxial ferro- or antiferro-
is a “staggered field”, which changes sign magnets, uniaxial ferro- or antiferroelec-
between the two sublattices, and hence is trics, etc., and for order – disorder transi-
thermodynamically conjugate to the order tions in alloys or mixed crystals (solid so-
parameter of the antiferromagnet. Al- lutions), when only two sublatties need to
though such a field normally cannot be ap- be considered, there are also cases where
plied in the laboratory, the second deriva- the order parameter must have a vector or
tive, c T (in this case it is called “staggered tensor character. Obviously, for an iso-
susceptibility”) is experimentally access- tropic ferromagnet or isotropic ferroelec-
ible via diffuse magnetic neutron scatter- tric the order parameter in the three-dimen-
ing, as will be discussed below. Similarly, sional space is a three-component vector
for the ordering alloy b-CuZn, H stands for (see Eq. (4-3)). It also makes sense to con-
a chemical potential difference between the sider systems where a planar anisotropy is
two sublattices; the response function c T is present, such that M or E (Eq. (4-3)) must
again physically meaningful and measures lie in a plane and hence a two-component
the peak intensity of the diffuse scattering vector applies as an order parameter. How-
of X-rays or neutrons. (This scattering ever, for describing antiferromagnetic or-
peak occurs at the superlattice Bragg spot dering and for order – disorder transitions
characteristic of the sublattice ordering with many sublattices multicomponent or-
considered.) Finally, in the gas – liquid der parameters are also needed, and the
transition considered in Fig. 4-1, H is the number of components of the order param-
chemical potential and c T the isothermal eter, the so-called “order parameter dimen-
compressibility of the system. sionality”, is dictated by the complexity of
As will be discussed in more detail in the structure, and has nothing to do with
Sec. 4.2.2, the divergences of second deriv- the spatial dimension. This is best under-
atives of the thermodynamic potential at a stood by considering specific examples.
critical point (Eqs. (4-4), (4-5), Fig. 4-2) Consider, for example, the ordering of
are linked to a diverging correlation length Fe – Al alloys (Fig. 4-3): whereas in the dis-
of order-parameter fluctuations (Fig. 4-2). ordered A2 phase Fe and Al atoms are ran-
Hence any discussion of phase transitions domly distributed over the available lattice
must start with a discussion of the order pa- sites, consistent with the considered con-
rameter. The Landau theory (Tolédano and centration (although there may be some
Tolédano, 1987) which attempts to expand short-range order), in the ordered B2 phase
the thermodynamic potential in powers of (the FeAl structure) the bcc lattice is split
the order parameter, gives a first clue to the into two inter-penetrating simple cubic (sc)
question of whether a transition is of sec- sublattices, one taken preferentially by Fe,
ond or first order (see Sec. 4.2.3). the other by Al atoms. This is the same
We first identify the possible types of or- (one-component) ordering as in b-CuZn.
der parameters, since this will distinctly af- What is of interest here is the further “sym-
fect the nature of the Landau expansion. In metry breaking” which occurs for the D03
Eqs. (4-1) and (4-4) to (4-6) we have structure (realized in the Fe3 Al phase):
treated the ordering field H and the order here four face-centered cubic (f.c.c.) sub-
d) is preferentially occupied by Al, all

other sublattices being preferentially occu-
pied by iron. However, the role of the sub-
lattices (a, c) and (b, d) can be inter-
changed; hence a two-component order pa-
B2 rameter is needed to describe the structure,
namely (see for example, Dünweg and
Binder (1987))
y I = ma − mc + mb − md
y II = − ma + mc + mb − md
mm ≡ (1 / N ) ∑ (2 ci − 1) (4-7)
iŒm
where ci is the concentration of lattice site i

and mm the sublattice “magnetization” in
pseudo-spin language. As can example of
an ordering with m = 8 components, we can
consider the f.c.c. antiferromagnet MnO:
D03 as Fig. 4-4 shows, the magnetic structure
consists of an antiferromagnet arrangement
of ferromagnetically ordered planes. If the
ordering were uniaxially anisotropic, we
Figure 4-3. Body-centered cubic lattice showing the

B2 structure (top) and D03 structure (bottom). The
top part shows assignments of four sublattices, a, b,
c, and d. In the A2 structure, the average concentra-
tions of A and B atoms are identical on all four sub-
lattices whereas in the B2 structure the concentra-
tions at the b and d sublattices are the same (example:
stoichiometric FeAl, with Fe on sublattices a and c;
Al on sublattices b and d). In the D03 structure, the
concentrations at the a and c sublattice are still the
same, whereas the concentration at sublattice b dif-
fers from the concentration at sublattice d (example:
stoichiometric Fe3Al, with Al on sublattice b; all
other sublattices taken by Fe). From Dünweg and
Binder (1987). Figure 4-4. Schematic view of the MnO structure,
showing the decomposition of the fcc lattice into two
ferromagnetic sublattices with opposite orientation
lattices a, b, c, d must be distinguished. Al- of the magnetization (indicated by full and open cir-
though the concentrations on sublattices a cles, respectively). Each sublattice is a stack of par-
allel close-packed planes (atoms form a triangular
and c are still equivalent, a further symme- lattice in each plane). Note that there are four equiv-
try breaking occurs between the sublattices alent ways in which these planes can be oriented in
b and d, such that only one sublattice (e.g., an fcc crystal.
would have m = 4 because of the four pos-

sible ways of orienting the planes (and
there could be eight kinds of different do-
mains coexisting in the system). Since for
MnO the order parameter can take any or-
ientation is a plane, it has two independent
components, and hence the total number of
order parameter components is m = 8 (Mu-
Figure 4-5. Schematic view of a lattice plane of a
kamel and Krinsky, 1976). An even larger para-H2 crystal, indicating the orientational order of
number of components is needed to de- the ellipsoid H2 molecules.
scribe the ordering of solid 3He. In view of
this, it is clear that even m Æ •, which is
also called the spherical model (Berlin and see Grazhdankina (1969)). In this example
Kac, 1952; Joyce, 1972) is a useful limit to it is clear that the spontaneous magnetiza-
consider from the theoretical point of view. tion, M, is the “primary order parameter”
Apart from this m-vector model for the whereas the tetragonal distortion (c/a – 1),
order parameter, Table 4-1 illustrates the where c is the lattice spacing in the lattice
need to consider order parameters of tenso- direction where the magnetization direc-
rial character. This happens in molecular tion occurs and a is the lattice spacing
crystals such as para-H2 (Fig. 4-5), N2 , O2 perpendicular to it, is a “secondary order
and KCN, in addition to liquid crystals. parameter”. For purely structural phase
Whereas the atomic degrees of freedom transitions where all considered degrees
considered for ferro- or antiferromagnetic of freedom are atomic displacements, the
order are magnetic dipole moments, for proper distinction between primary and
ferro- or antiferroelectric order electric di- secondary order parameters is much more
pole moments, the degree of freedom of the subtle (Tolédano and Tolédano, 1987).
molecules in Fig. 4-5 which now matters is We first formulate Landau’s theory for
their electric quadrupole moment tensor: the simplest case, a scalar order parameter
density F (x). This density is assumed to
⎛ 1 3 ⎞ be small near the phase transition and
fmn = ∫ dx r ( x ) ⎜ xm xn − ∑ xl2 dmn ⎟ (4-8)
⎝ 3 l =1 ⎠ slowly varying in space. It can be obtained
by averaging a microscopic variable over a
where r (x) is the charge density distribu- suitable coarse-graining volume Ld in d-di-
tion function of a molecule, x = (x1 , x2 , x3 ), mensional space. For example, in an aniso-
and dmn is the Kronecker symbol. tropic ferromagnet the microscopic vari-
Proper identification of the order param- able is the spin variable F i = ±1 pointing in
eter of a particular system therefore needs a the direction of the magnetic moment at
detailed physical insight, and often is com- lattice site i; in an ordering alloy such as b-
plicated because of coupling between dif- CuZn the microscopic variable is the dif-
ferent degrees of freedom, e.g., a ferromag- ference in concentration of Cu between the
netic material which is cubic in the para- two sublattices I and II in unit cell i, F i =
magnetic phase may become tetragonal as ciII – ciI ; and in an unmixing alloy such as
the spontaneous magnetization develops, Al – Zn, the microscopic variable is the
owing to magnetostrictive couplings (for a concentration difference between the local
more detailed discussion of this situation, Al concentration at lattice site i (ci = 1 if it
is taken by Al and ci = 0 otherwise) and the the change of sign of the order parameter
critical concentration, F i = ci – ccrit . In all for H = 0 and thus odd powers such as
of these cases F (x) is considered to be an F 3 (x) do not occur; this is true for magnets
order parameter field defined in continuum (no direction of the magnetization is pre-
space: ferred without magnetic field in a ferro-
magnet) and for sublattice ordering of al-
F ( x ) = ∑ Fi / Ld (4-9)
i Œ Ld
loys such as b-CuZn (since whether the Cu
atoms preferentially occupy sublattices a, c
x being the center of gravity of the “vol-
in Fig. 4-3 or sublattices b, d is equivalent),
ume” Ld . The appropriate magnitude of the
but it is not true in general (e.g., third-order
linear dimension L of the coarse-graining
terms do occur in the description of the
cell will be discussed later; obviously it
Cu3Au structure, or in the ordering of rare
must be much larger than the lattice spac-
gas monolayers adsorbed on graphite in
ing in order for the continuum description –
the ÷3 structure, as will be discussed be-
to make sense. Then a Helmholtz energy
low).
functional F [F (x)] is assumed:
In order to understand the meaning and
1 F use of Eq. (4-10), consider first the fully
F [F ( x )] = 0 (4-10)
kB T kB T homogeneous case, —F (x) ∫ 0, F (x) ∫ F 0 ;
then F [F ] is the standard Helmholtz en-
+ ∫ dx ⎧⎨ r F 2 ( x ) + uF 4 ( x )
1 1
ergy function of thermodynamics, which
⎩2 4
needs to be minimized with respect to F in
H 1 ⎫
− F (x) + [ R ∇F ( x )]2 ⎬ order to determine the thermal equilibrium
kB T 2d ⎭ state, and Ú dx = V the total volume of the
where F0 is the background Helmholtz en- system. Thus,
ergy of the disordered phase, and r, u, and (4-12)
1 ⎛ ∂F ⎞
R are phenomenological constants. (In fact, ⎜ ⎟ = r F0 + u F0 = 0
3
R can be interpreted as the effective range kB T V ⎝ ∂F0 ⎠ T

H =0
of interaction between the atomic degrees
which is solved by
of freedom F i , as will be seen below.) Ob-
viously, Eq. (4-10) is the Taylor series-type F0 = 0 , T > Tc
expansion of F [F (x)] in powers of F (x) F 0 = ± (– r/u) 1/2
(4-13)
and —F (x), where just the lowest order = ± (r¢/kB u)1/2 (Tc /T – 1)1/2 , T < Tc
terms were kept. This makes sense if both
the coefficients u and R 2 are positive con- Hence Eqs. (4-10) and (4-12) indeed yield
stants at Tc , whereas the essential assump- a second-order transition as T is lowered
tion which defines F as playing the role of through Tc . For T < Tc , a first-order transi-
an order parameter of a second-order phase tion as a function of H occurs, since F 0
transition is that the coefficient r changes jumps from (– r/u)1/2 to – (– r/u)1/2 as H
its sign at the transition as the variable of changes sign. This behavior is exactly that
interest (the temperature in the present shown schematically in Fig. 4-1, with
case) is varied, b = 1/2, B̂ = (r¢/kB u)1/2 and F 0 (H = 0) = MS .
If u < 0 in Eq. (4-10), however, we must
kB Tr = r¢ (T – Tc ) (4-11) not stop the expansion at fourth order but
Note also that in Eq. (4-10) we have as- rather must include a term –16 v F 6 (x) (as-
sumed a symmetry in the problem against suming now v > 0). Whereas in the second-
Figure 4-6. Schematic variation of the Helmholtz energy in the Landau model at transitions of (a) second or-
der and (b) first order as a function of the (scalar) order parameter F. Cases (a) and (b) assume a symmetry
around F = 0, whereas case (c) allows a cubic term.
order case F (F ) has two minima for T < Tc this value, i.e.,
which continuously merge as T Æ Tc and
only one minimum at F = 0 remains for [ F (F0 ) − F ( 0 )]/(V kB T )
= F02 ⎛ + F02 + F04 ⎞ = 0

T > Tc (Fig. 4-6 a) F [F ] now has three min- r u v
(4-15)
ima for T0 < T < Tc , and the temperature T0 ⎝2 4 6 ⎠ T = Tc
where r changes sign (r = r¢ (T – T0 ) now) With some simple algebra Eqs. (4-14) and
differs from the phase transition tempera- (4-15) yield
ture Tc where the order parameter jumps
discontinuously from zero for T > Tc to Tc = T0 + 3 u2/(32 r¢v) (4-16)
(F 0 )Tc = ± (3 u/4 v)1/2, see Fig. 4-6 b. and the “stability limit”, where the mini-
These results are found by analogy with mum describing the metastable ordered
Eq. (4-12) from phase in the disordered phase above Tc dis-
1 ⎛ ∂F ⎞ appears, is given by
⎜ ⎟
kB T V ⎝ ∂F0 ⎠ T T1 = T0 + u2/(8 r¢v) (4-17)
H =0
= F0 ( r + uF02 + v F04 ) = 0 (4-14) The significance of such metastable states
as described by the Landau theory for first-
which is solved by order transitions and the associated stabil-
ity limits T0 (for the disordered phase at
F02 = − u /( 2 v ) + (u / 2 v )2 − r / v
T < Tc ) and T1 will be discussed further in
(choosing the minus sign of the square root Sec. 4.4.
would yield the maxima rather than the The alternative mechanism by which a
minima in Fig. 4-6 b). On the other hand, first-order transition arises in the Landau
we know that [F – F (0)]/(V kB T ) = 0 in the theory with a scalar order parameter is
disordered phase, and Tc can be found (Fig. the lack of symmetry of F against a sign
4-6 b) from the requirement that the free change of F. Then we may add a term
energy of the ordered phase then is equal to –13 w F 3 to Eq. (4-10), with another pheno-
menological coefficient, w. For u > 0, F (F ) only in mean-field theory, but lacks any
may have two minima (Fig. 4-6 c); again fundamental justification in statistical me-
the transition occurs when the minima are chanics. Schemes such as those shown in
equally deep. For r = r¢ (T – T0 ) this hap- Fig. 4-6 make sense for a local “coarse-
pens when grained free energy function” only (which
depends on the length scale L introduced in
Tc = T0 + 8 w 2/(81 u r¢) (4-18)
Eq. (4-9)), see Sec. 4.2.2, but not for the
the order parameter jumping there from global free energy.
F 0 = – 9 r/w to F 0 = 0. Again a stability After this digression we return to the
limit of ordered state in the disordered generalization of the Landau expansion,
phase occurs, i.e. Eq. (4-10), the case where the order param-
eter has vector or tensor character. How
T1 = T0 + w 2/(4 u r¢) (4-19)
can we find which kinds of term appear in
At this point, an important caveat should be the expansion?
emphasized: free energy curves involving Basically, there are two answers to this
several minima and maxima as drawn in question. A general method, which follows
Fig. 4-6 are used so often in the literature below, is a symmetry classification based
that many researchers believe these con- on group theory techniques (Landau and
cepts to be essentially rigorous. However, Lifshitz, 1958; Tolédano and Tolédano,
general principles of thermodynamics indi- 1987). A very straightforward alternative
cate that in thermal equilibrium the ther- approach is possible if we consider a par-
modynamic potentials are convex func- ticular model Hamiltonian Ᏼ (for a brief
tions of their variables. In fact, F (F 0 ) discussion of some of the most useful mod-
should thus be convex as a function of F 0 , els of statistical mechanics for studying
which excludes multiple minima! For Fig. phase transitions, see Sec. 4.3.1). We can
4-6 a and b this means that for T < Tc in then formulate a microscopic mean-field
states with – F 0 < F < F 0 (where F 0 is the approximation (MFA), such as the effec-
solution of Eqs. (4-13) or (4-14), respective field approximation of magnetism
tively) the thermal equilibrium state is (Smart, 1966) or the Bragg-Williams ap-
not a pure homogeneous phase: rather, the proximation for order – disorder phenomena
minimum free energy state is given by the in alloys (De Fontaine, 1979), where we
double-tangent construction to F (F ) and then expand the MFA Helmholtz energy di-
this corresponds to a mixed phase state (the rectly.
relative amounts of the coexisting phases As an example of this approach, we con-
are given by the well known lever rule). sider a specific model of a ternary alloy
Now it is standard practice, dating back to where each lattice site i may be taken by ei-
van der Waals’ interpretation of this equa- ther an A, a B, or a C atom, assuming con-
tion of state for fluids, to interpret the part centrations cA = cB = cC = 1/3, and assuming
of F (F ) in Fig. 4-6 which lies above the that an energy J is achieved if two neigh-
F (F ) given by the double-tangent con- boring sites are taken by the same kind of
struction as a metastable state provided atom. This is a special case of the q-state
that c T = (∂2 F/ ∂F 2 )T < 0, whereas states Potts model (Potts, 1952; Wu, 1982), the
with c T < 0 are considered as intrinsically Hamiltonian being
unstable states. As will be seen in Sec. 4.4,
Ᏼ Potts = − ∑ J dSi S j , Si = {1, 2, …, q} (4-20)
this notion is intrinsically a concept valid 〈 ij 〉
The states 1, 2, and 3 of the Potts spin stand in the expansion:

for the three kinds of atomic species A, B,
and C here and ·ij Ò denotes a sum over F zJ
=− − ln 3
nearest neighbor pairs in the lattice. In the V kB T 6 kB T
MFA, we construct the Helmholtz energy ⎛ zJ ⎞ 2
F = U – T S simply by expressing both en- + 3 ⎜1 − ⎟ (F1 + F2 + F1 F2 )
2
⎝ 3 kB T ⎠
thalpy U and entropy S in terms of the frac- 9
tions na of lattice sites in states a . The en- + (F12 F2 + F1 F22 ) + … (4-24 )
2
tropy is simply the entropy of randomly
mixing these species, and, using Stirlings As expected, there is a temperature T0
equation, this yields the standard expres- (= z J/3 kB ) where the coefficient of the
sion (see elementary textbooks on statisti- quadratic term changes sign.
cal thermodynamics): Of course, for many phase transitions a
3 specific model description is not available,
S=−V ∑ na ln na (4-21) and even if a description in terms of a
a =1 model Hamiltonian is possible, for compli-
In the enthalpy term, MFA neglects correla- cated models the approach analogous to
tions in the occupation probability of neigh- Eqs. (4-20) and (4-24) requires tedious cal-
boring sites. Hence the probability of find- culation. Clearly the elegant but abstract
ing a nearest neighbor pair in a state is sim- Landau approach based on symmetry prin-
ply na2 , and in a lattice with coordination ciples is preferable when constructing the
number z there are z /2 pairs per site. Hence Landau expansion. This approach starts
from the observation that usually the disor-
3
zJV
U=− ∑ na2 (4-22) dered phase at high temperatures is more
2 a =1 “symmetric” than the ordered phase(s) oc-
and thus curring at lower temperature. Recalling
the example described in Fig. 4-3, referring
3 3
F zJ to Fe – Al alloys: in the high temperature
=− ∑ na2 + ∑ na ln na (4-23)
V kB T 2 kB T a =1 a =1 A2 phase, all four sublattices a, b, c, and d
are completely equivalent. This permuta-
We could directly minimize F with respect tion symmetry among sublattices is broken
3
in the B2 phase (FeAl structure) where the
to na , subject to the constraint ∑ na = 1,
a =1 concentration on sublattices a, c differs
since each site should be occupied. In order from the concentration on sublattices b, d.
to make contact with the Landau expan- A further symmetry breaking occurs when
sion, however, we rather expand F in terms we go from the B2 phase to the D03 phase
of the two order parameter components (Fe3Al structure), where the concentration
F1 = n1 – 1/3 and F2 = n2 – 1/3 (note that all on sublattice b differs from that on sublat-
ni = 1/3 in the disordered phase). Hence we tice d. In such cases the appropriate struc-
recognize that the model has a two-compo- ture of the Landau expansion for F in terms
nent order parameter and there is no sym- of the order parameter F is found from the
metry between F i and – F i . So cubic terms principle that F must be invariant against
in the expansion of F are expected and do all symmetry operations of the symmetry
occur, whereas for a properly defined order group G0 describing the disordered phase.
parameter, there cannot be any linear term In the ordered phase, some symmetry ele-
ments of G0 fall away (spontaneously This approach also carries over to cases
broken symmetry); the remaining symme- where the order parameter is a tensor. For
try elements form a subgroup G of G0 . example, for elastic phase transitions
Now the invariance of F must hold separ- (Cowley, 1976; Folk et al., 1976) the order
ately for terms F k of any order k and this parameter is the strain tensor {e ik}. Apply-
requirements fixes the character of the ing the summation convention (indices oc-
terms that may be present. curring twice in an expression are summed
Rather than formulating this approach over), the Landau expansion is
systematically, which would require a
F [e ik (x)]
lengthy and very mathematical exposi-
tion (Tolédano and Tolédano, 1987), we = F0 + Ú dx ( –12 ciklm e ik e lm
rather illustrate it with a simple example. + –13 ciklmrs
(3)
e ik e lm e rs
Suppose a cubic crystal exhibits a transi- + –14 ciklmrsuv
(4)
e ik e lm e rs e uv
tion from a para-electric to a ferroelectric
+ … + gradient terms) (4-27)
phase, where a spontaneous polarization
P = (P1 , …, Pn ), n = 3, appears. F is then Here the ciklm are elastic constants and
(3) (4)
given as follows: ciklmrs and ciklmrsuv analogous coefficients
of higher order (“anharmonic”) terms. For
= 0 + ∫ dx ⎡⎢ c el−1 P 2
F [ P ( x )] F 1
most elastic transitions symmetry permits
kB T kB T ⎣2 some nonzero ciklmrs (3)
and hence leads to
1 ⎛ n 4 n ⎞ first-order transitions. Examples of such
+ ⎜ u ∑ Pm + u ′ ∑ Pm2 Pj2 ⎟ + … systems are the “martensitic transitions” in
4 ⎝ m =1 m < j =1 ⎠
Nb3Sn and V3Si; there the elastic distortion
R2 n ⎤ jumps at the transition from zero to a very
+ ∑ (∇Pm )2 + …⎥ (4-25)
small value (e ik ≈ 10– 4 ) and a Landau ex-
2d m =1 ⎦
pansion makes sense. Note that the Landau
Whereas the quadratic term of a general theory is not very useful quantitatively for
dielectric medium would involve the in- transitions which are very strongly first or-
verse of the dielectric tensor, ∑ ( c el−1 )ij Pi Pj , der, since in that case high-order terms in the
ij
this term is completely isotropic for cubic Taylor expansion are not negligible. More
crystals. Inversion symmetry requires in- about martensitic phase transformations can
variance against P Æ – P and hence no be found in the Chapter by Delaey (2001).
third-order term occurs. The fourth-order Just as in a ferromagnetic transition the
term now contains the two “cubic invari- inverse magnetic susceptibility c –1 van-
( )
ants” ∑ Pi4 and ∑ Pi2 . Here we invoke
2 ishes at Tc (Eq. (4-4)) and in a ferroelectric
i i transition the inverse dielectric susceptibil-
the principle that all terms allowed by sym- ity c el–1 vanishes at Tc (Eq. (4-25)), in a
metry will actually occur. Now Eq. (4-25) “ferroelastic” transition one of the elastic
leads, in the framework of the Landau the- constants ciklm vanishes at T0 . For example,
ory, to a second-order transition if both in KCN such a softening is observed for the
elastic constant c44 . Note that this material
u>0 and u + u¢ > 0 (4-26)
has already been mentioned as an example
whereas otherwise we have a first-order of a phase transition where the order pa-
transition (then terms of sixth order are rameter is the electric quadrupole moment
needed in Eq. (4-25) to ensure stability). tensor (Eq. (4-8)), describing the orienta-
tion of the dumbbell-shaped CN–-ions: (mediated via acoustic phonons) is more

at the same time, it can be considered as important than their direct electric quadru-
an example for an elastic transition, where pole – quadrupole interaction. The elastic
the order parameter is the strain tensor character of the phase transition is therefore
(emn (x))! Such an ambiguity is typical of an intrinsic phenomenon for this material.
many first-order transitions, because vari- We briefly discuss the appropriate con-
ous dynamical variables (such as a local struction of the order parameter (compo-
dielectric polarization P (x), a local quadru- nents) for order – disorder transitions. Fig.
pole moment fmn (x) and the strain emn (x) 4-3 (and Eq. (4-7)) illustrated how we can
may occur simultaneously in a crystal and visualize the ordered structure in real
are coupled together. In a Landau expan- space, and apply suitable symmetry opera-
sion, such couplings are typically of the bi- tors of the point group. However, it is often
quadratic energy – energy coupling type, more convenient to carry through a corre-
i.e., various types of mixed fourth-order sponding discussion of the ordering in re-
terms occur, such as P 2 (x) emn em ¢ n ¢ umn m ¢ n ¢ . ciprocal space rather than in real space (re-
Although this is also true for second-order member that the ordering shows up in
transitions, there the “primary order pa- superlattice Bragg spots appearing in the
rameter” is distinct from the fact that the reciprocal lattice in addition to the Bragg
associate inverse response function c T–1 spots of the disordered phase). For exam-
vanishes at Tc . For the “secondary order ple, consider rare gas monolayers adsorbed
parameters” the corresponding inverse re- on graphite: at low temperatures and pres-
–
sponse function stays finite. With first- sures, the adatoms form a ÷3 structure
order transitions, the inverse response commensurate with the graphite lattice.
–
function of the primary order parameter This ÷3 structure can be viewed as a trian-
would vanish at the hypothetical tempera- gular lattice decomposed into three sublat-
ture T0 (stability limit of the disordered tices, such that the adatoms preferentially
phase), which, however, cannot normally occupy one sublattice. Mass density waves
be reached in a real experiment, see Sec. are taken as an order parameter (Bak et al.,
4.4. K + CN– is a good example of such a 1979; Schick, 1981):
complicated situation with three local or- 3
der parameters coupled together, since the r ( x ) = ∑ [ya exp( i q ⋅ x )
CN– ion also has a dipole moment. After a =1
+ y − a exp( − qa ⋅ x )] (4-28)
the first-order transition at T = 110 K from
the cubic “plastic crystal” phase (where Here the qa are the three primitive vectors
there is no long-range orientational order, associated with the reciprocal lattice of the
the CN – dumbbells can rotate) to the tet- rare gas monolayer (a being the lattice
ragonally distorted, orientationally ordered spacing of the triangular lattice)
phase, a second first-order transition oc-
curs at T = 80 K to an antiferroelectric 2p ⎛ 1 ⎞
q1 = ⎜ 0, ⎟
phase where the dipole moments order. In a ⎝ 3⎠
this system, only a microscopic theory 2p ⎛ 1 1 ⎞
(Michel and Naudts, 1977, 1978; De Raedt q2 = ⎜− , − ⎟
a ⎝ 2 2 3⎠
et al., 1981; Lynden-Bell and Michel, 1994)
could clarify that the elastic interaction 2p ⎛ 1 1 ⎞
q3 = ⎜ ,− ⎟ (4-29)
between the quadrupole moments fmn (x) a ⎝ 2 2 3⎠
and ya and y– a are the (complex!) order structures in terms of „spin density waves“
parameter amplitudes. In constructing the (Villain, 1959; Kaplan, 1959). We shall
free energy expansion with the help of Eq. outline the connection between this ap-
(4-28), note that the periodicity of the proach and the MFA, generalizing the ap-
underlying graphite lattice allows invariant proach of Eqs. (4-21) – (4-23) slightly, in
“umklapp” terms (the phase factors of Sec. 4.3. Here we only mention that these
third-order terms add up to a reciprocal lat- concepts are closely related to the descrip-
tice vector of the graphite lattice). Keeping tion of structural transitions in solids,
only those terms in the Landau expansion, where the order parameter can often be
–
and which are nonzero in the ÷3 structure, considered as a “frozen phonon”, i.e., a dis-
with the real order parameter components placement vector wave (Bruce and Cowley,
––
(i = ÷–1) 1981). Note also that the approach of con-
3 centration waves is not restricted to solids,
F1 = ∑ (ya − y − a ) /( 2 3 ) but can also be used to describe meso-
a =1 phases in fluid block-copolymer melts
3
(Leibler, 1980; Fredrickson and Helfand,
F2 = ∑ (ya − y − a ) /( 2 i 3 ) (4-30)
a =1 1987; Fredrickson and Binder, 1989; Bin-
der, 1994), see Fig. 4-7, in liquid crystal-
the expression line polymers, etc.
F 1 As a final remark in this section we men-
= − r (F12 + F22 ) (4-31) tion that not for all phase transitions in sol-
V kB T 2
ids there does exist a group – subgroup rela-
1 1
+ w (F13 − 3F1 F22 ) + u (F12 + F22 )2 tionship between the two groups G1 and G2
3 4 describing the symmetry of the phases co-
is obtained. This is equivalent to the result existing at the transition. These phases can-
for the three-state Potts model, Eq. (4-24) not be distinguished by an order parameter
(Alexander, 1975), after the quadratic form which is zero in one phase and becomes
is diagonalized. A three-dimensional ana- nonzero in the other. Such transitions must
log of this order (where the planes exhibit- be of first order. Examples of this situation
–
ing ÷3 structure are stacked together to are well known for structural phase transi-
form a hexagonal lattice) occurs in the tions, e.g., the tetragonal – orthorhombic
intercalated compound C6Li (Guerard and transition of BaTiO3 or the “reconstruc-
Herold, 1975; Bak and Domany, 1979). tive” transition from calcite to aragonite
Similarly to the description of the den- (Guymont, 1981). For the so-called “non-
sity modulation in the superstructure of disruptive transitions” (Guymont, 1981),
adsorbed layers (in two dimensions) or the new structure can still be described in
interstitial compounds (in three dimen- the framework of the old structure (i.e., its
sions) in terms of mass density waves of symmetry elements can be specified, the
the adsorbate (or interstitial, respectively), Wyckoff positions can be located, etc.).
the superstructure ordering in binary alloys Landau-type symmetry arguments still
can be described in terms of concentration yield information on the domain structures
waves (Khachaturyan, 1962, 1963, 1973, arising in such phase transitions (Guymont,
1983; De Fontaine, 1975, 1979). The same 1978, 1981). The tetragonal – orthorhombic
concept was used even earlier to describe transition of BaTiO3 is considered to be an
the magnetic ordering of helimagnetic spin example of such a non-disruptive transi-
ond-order transition vanishes continuously

(Fig. 4-1). We consider the accompanying
critical singularities (Fig. 4-2), in the
framework of Landau theory (Eq. (4-10)).
For a proper understanding of the critical
fluctuations of the order parameter, we
must no longer restrict the treatment to the
homogeneous case —F (x) ∫ 0 as was done
in Eqs. (4-12) and (4-13); we now wish to
consider, for example, the response to an
inhomogeneous, wavevector-dependent or-
dering field H (x):
H (x) = Hq exp (i q · x) (4-32)
Although for a magnetic transition such a

Figure 4-7. (a) Chemical architecture of a diblock field cannot be directly applied in the la-
copolymer. A diblock copolymer consists of a poly- boratory, the action of such fields can be
merized sequence of A monomers (A-block) cova-
indirectly probed by appropriate scattering
lently attached to a similar sequence of B monomers.
(b) The microphase separation transition occurs measurements, e.g., magnetic neutron scat-
when a compositionally disordered melt of copoly- tering. This is because the scattering can
mers, which are in random coil configurations (right) be viewed as being due to the inhomogene-
transforms to a spatially periodic, compositionally ous (dipolar) field that the magnetic mo-
inhomogeneous phase (left) on lowering the temper-
ment of the neutron exerts on the probe.
ature. For nearly symmetric copolymers the ordered
phase has the lamellar structure shown. Since the Therefore, the scattering intensity for a
wavelength of the concentration wave here is of the scattering vector q is related to the wave-
order of the coil gyration radius, which may be of the vector-dependent susceptibility (see Kittel
order of 100 Å for high molecular weight, the order- (1967)). Similarly, by the scattering of
ing occurs on a “mesoscopic” rather than micro-
X-rays the wavevector-dependent response
scopic scale (Å) and hence such phases are called
“mesophases”. From Fredrickson and Binder (1989). function to the sublattice ordering field can
be measured in ordering alloys such as b-
CuZn (see Als-Nielsen (1976) for a discus-
tion, while “reconstructive transitions” and sion of such scattering experiments in vari-
martensitic transformations of the type of ous systems).
the f.c.c. – b.c.t. transition in Fe are “disrup- Hence in order to deal with Eq. (4-10)
tive” (see the Chapter by Delaey (2001)). we now have to minimize a functional
We shall not go into detail about such prob- rather than a simple function as was done
lems here, however. in Eq. (4-12). In order to do so, we note
that the problem of minimizing Ú dx f (y ,
—y ) in Eq. (4-10) is analogous to the prob-
4.2.2 Second-Order Transitions
lem of minimizing the action in classical
and Concepts about Critical Phenomena
mechanics, W = Ú dt ᏸ (x, ẋ ), where ᏸ is the
(Critical Exponents, Scaling Laws, etc.)
Lagrangian and ẋ = dx/dt, the velocity of
We now return to the case where at a a particle at point x (t = time). Just as in
critical point the order parameter of a sec- classical mechanics, where this problem
is solved in terms of the Euler – Lagrange For T < Tc , using Eq. (4-13) the term 5 v F04
equation near Tc is still negligible, and hence
( ∂ᏸ / ∂x ) −
d
[ ∂ᏸ / ∂ ( dx / dt )] = 0 −1 1
cT = = (4-37)
dt 2 kB T r 2 r ′ ( Tc − T )
we conclude here that x = R2 /( − 2 rd ) = ( R / 2 r ′ d ) (T − Tc ) −1/ 2
( ∂ f / ∂y ) − ∇ [ ∂ f / ∂ (∇y )] = 0 Eqs. (4-36) and (4-37) simply imply the
in order that the Helmholtz energy func- well known Curie – Weiss law, i.e., the ex-
tional is a minimum. Eq. (4-10) thus yields ponents g, g ¢, defined in Eq. (4-4), are, in
the framework of the Landau theory,
r F ( x ) + uF 3 ( x ) + v F 5 ( x ) g = g¢ = 1 (4-38)
R2 2 H (x)
− ∇ F (x) = (4-33) Comparing Eqs. (4-35) and (4-37) we also
d kB T
conclude that
Using
n = n ¢ = 1/2 (4-39)
F (x) = F 0 + DF (x) = F 0 + DFq exp (i q · x) The physical meaning of the correlation
Eqs. (4-32) and (4-33) yield, on linearizing length x is easily recognized if the well
in DFq for small Hq , the wavevector-de- known fluctuation relationship
pendent susceptibility 1
c (q) = ∑ exp( i q ⋅ x )
DFq kB T x
c (q) ≡ × [ 〈F ( 0 ) F ( x )〉 − F02 ] (4-40)
Hq
−1 is expanded to second order in q as
1 ⎡ R2 2 ⎤
⎢ r + 3 u F0 + 5 v F0 + d q ⎥
2 4
=
kB T ⎣ ⎦ 1 ⎧
c (q) ≈ ⎨∑ [ 〈F ( 0 ) F ( x )〉 − F0 ]
2
cT kB T ⎩ x
= (4-34 )
1 + x 2 q2 1 ⎫
− ∑ ( q ⋅ x )2 [ 〈F ( 0 ) F ( x )〉 − F02 ]⎬ (4-41)
From Eq. (4-34) we can simply read off
2 x ⎭
the temperature dependence of the suscep- which can be written as
tibility c T (Eq. (4-4)) and the correlation
1
length x of order parameter fluctuations, c (q) = ∑ [〈F ( 0 ) F ( x )〉 − F02 ]
which is generally expected to behave as kB T x
⎧ q2
⎧xˆ ( T / Tc − 1) − n , T > Tc × ⎨1 − ∑ x 2 [〈F ( 0 ) F ( x )〉 − F02 ]
x =⎨ˆ −n′
(4-35) ⎩ 2d x
⎩x ′ (1 − Tc / T ) , T < Tc ⎫
∑ [〈F ( 0 ) F ( x )〉 − F02 ]⎬ (4-42)
Above Tc , the terms 3 u F02 + 5 v F04 = 0, x ⎭
hence Comparing Eqs. (4-34) and (4-42), we find
1 1 that in terms of the order parameter corre-
cT = = (4-36 )
kB T r r ′ ( T − Tc ) lation function G (x),
x = R2 /( rd ) = ( R / r ′ d ) (Tc − T ) −1/ 2 G( x ) = 〈F ( 0 ) F ( x )〉 − F02 (4-43)
and c T and x 2 are related to the zeroth and Tc . For T < Tc , we have instead, from Eqs.
second moment of the correlation function: (4-10) and (4-13)
F − F0
= F02 ⎛ + F02 ⎞
1 r u
cT = ∑ G (x) (4-44)
⎝2 4 ⎠
kB T x kB T V
− r2 (1 − T / Tc )2
1 ≈ − r′2 (4-52)
2
x = ∑ x G (x) ∑ G (x)
2
(4-45) =
4u 4 kB2 u
2d x x
which implies C = r¢ 2 (T /Tc ) / (4 kB u) for
Eq. (4-44) clearly shows that the critical di-
T < Tc . This jump singularity of the specific
vergence of c T occurs because the correla-
heat at Tc instead of the power law in Eq.
tions G (x) become long ranged: whereas
(4-5) is formally associated with vanishing
off Tc the correlation function for large | x|
critical exponents:
decays exponentially,
a = a¢ = 0 (4-53)
ln G (x) Æ – | x|/ x , | x| Æ • (4-46)
The question must now be asked whether
at Tc we have a power-law decay of the cor- this description of critical phenomena is
relation function: accurate in terms of the Landau theory. The
G (x) = Ĝ | x| – (d – 2 + h ) , T = Tc (4-47) free energy functional F [F (x)] in Eq.
(4-10) should be considered as an effective
The exponent describing this critical decay Hamiltonian from which a partition func-
has been defined such that tion Z can be obtained so that the true
h=0 (4-48) Helmholtz energy becomes:
in the Landau theory, whereas in general F = – kB T ln Z (4-54)

h ⫽ 0, as will be discussed below. = – kB T ln Ú d [F (x)] exp {– F [F (x)]/kB T}
In order to make our collection of critical It is natural to expect that the main contri-
exponents complete, we also consider the bution to the functional integral comes
critical isotherm from the region where the integrand is larg-
F0 = D̂ H 1/d , T = Tc (4-49) est, i.e., the vicinity of the point where
F [F (x)]/kB T has its minimum. If we as-
Using Eq. (4-33) for a uniform field we sume that the distribution over which the
conclude that average in Eq. (4-54) is a delta function at
F0 = (kB Tu)–1/3 H 1/3 (4-50) the value F 0 yielding the minimum, i.e.,
we assume that fluctuations of the order
i.e., D̂ = (kB Tu)–1/3 and parameter make a negligible contribution
d = 1/3 (4-51) to the functional integral, then the mini-
mum of F is equivalent to the minimum of
in the framework of the Landau theory. F (F ). In general, however, this is not true,
Finally, we turn to the specific heat, Eq. as the effects of fluctuations modify the
(4-5), which behaves rather pathologically critical behavior drastically, and the Lan-
in the Landau theory: for T > Tc and H = 0, dau theory does not hold for systems such
Eq. (4-10) just implies F = F0 , i.e., the as b-CuZn.
specific heat associated with the ordering A first hint of the conditions for which
described by F is identically zero above the Landau theory is valid can be obtained
by considering the effect of fluctuations (4-46). Replacing the right inequality in

within the context of the Landau theory it- Eq. (4-57) by the equality, L = x , as done in
self. The criterion Ginzburg (1960) sug- Eq. (4-55), gives therefore the maximum
gests is that the Landau theory is valid if permissible choice for L, for which the
the mean-square fluctuation of the order fluctuations in Eq. (4-55) are smallest. Even
parameter in a correlation volume is small then, however, it turns out that Eq. (4-55)
in comparison with the order parameter typically is not fulfilled for T close to Tc . In
square itself: order to see this, we write Eq. (4-55) in
terms of the local variable F i and denote
·[F (x) – F 0 ]2 ÒL = x O F 02 (4-55)
the number of lattice sites i contained
Here L is the coarse-graining length intro- Ld for the choice L = x as N (t), where
duced in Eq. (4-9) in order to transform the t ∫ 1 – T /Tc . Then Eq. (4-55) becomes
microscopic lattice description of ordering
[ ∑ Fi − 〈Fi 〉 ] ∑ Fi
2 2
phenomena in solids to a continuum de- O
i Œ Vx ( x ) i Œ Vx ( x )
scription. On the lattice level, the local or-
(4-58)
der parameter F i shows a rapid variation
from one lattice site to the next, and the where we have denoted the volume of size
mean-square fluctuation is very large. For L = x centered at x as Vx (x). Making use of
example, for an Ising model of an aniso- the translational invariance of correlation
tropic ferromagnet, functions and ·F i Ò = Ms (t), the spontane-
(4-56) ous order parameter which is independent
Ᏼ Ising = − J ∑ Si S j − H ∑ Si , Si = ± 1 of i, Eq. (4-58) becomes
〈 i, j 〉 i
Where J is the exchange interaction, the N (t ) ∑ [〈Fi = 0 Fi 〉 − Ms2 (t )]

i Œ Vx ( x )
sum ·i, j Ò extends once over all nearest
O N 2 (t ) Ms2 (t ) (4-59)
neighbor pairs, and Si points in the direc-
tion of the local magnetic moment at lattice If the sum over correlations in Eq. (4-59)
site i, we have F i = Si , i.e., F i2 ∫ 1. In this were to extend over all space, it would sim-
case ·[F i – ·F i Ò]2 Ò = 1 – Ms2 is never small ply be the “susceptibility” kB T c (t) (see
in comparison with the square of the spon- Eq. (4-42)). Since the sum does contain
taneous magnetization Ms2 near Tc . Carry- just the volume region over which the F i s
ing out the local averaging defined in Eq. are strongly correlated with each other, the
(4-9) reduces this local fluctuation in a sum clearly is of the order of f kB T c (T ),
“volume” Ld. This averaging should, in where f < 1 is a factor of order unity, which
principle, be carried out over a length scale should have no critical (vanishing or diver-
L that is much larger than the lattice spac- gent) temperature dependence as T Æ Tc.
ing a but much smaller than the correlation Hence in the inequality Eq. (4-59) this fac-
length x , tor may also be omitted, and we conclude
(Als-Nielsen and Laursen, 1980) that
aOLOx (4-57)
Whereas for a study of critical phenomena
c (t) O N (t) O Ms2 (t) (4-60 a)
it is permissible to average out local effects Making use of N (t) = [x (t)/a]d , a being the
on the scale of a lattice spacing, relevant lattice spacing, we obtain:
spatial variations do occur on the scale of
x , as we have seen in Eqs. (4-34) and const. O [x (t)]d Ms2 (t) c –1 (t) (4-60 b)
where in suitable units the constant on the and Laursen, 1980), the essential part of
left-hand side of the equality is of order the magnetic Hamiltonian is a magneto-
unity. Using Eqs. (4-4), (4-6), and (4-35) static dipole – dipole interaction,
we obtain (anticipating g = g ¢ and n = n ¢)
3 z2 − x2
– n d +2 b + g Ᏼ = − m2 ∑ Si S j , Si = ± 1 (4-64)
const. O t (4-61) i≠ j | x |5
Inserting the Landau values for the critical where m is the magnetic moment per spin
exponents b = 1/2, n = 1/2, g = 1 and know- and z the coordinate of x in the direction of
ing the critical amplitude of x ~ R, the inter- the uniaxial anisotropy. Unlike the iso-
action range, Eq. (4-60) implies tropic Ising Hamiltonian in Eq. (4-56), here
the sign of the interaction depends on the
const. O (R/a)d t (d – 4)/2 (4-62) direction in the lattice. Therefore, fluctua-
This condition for the validity of the Lan- tions display an essential anisotropy: in-
dau theory for d < 4 always breaks down stead of the isotropic result (Eq. (4-34)) of
as t Æ 0 (T Æ Tc ). In fact, for d < 4 close the well known Ornstein-Zernike type,
enough to Tc a regime occurs where fluctu- the wavevector-dependent susceptibility
ations dominate the functional integral (Eq. c (q) = S (q)/kB T (S (q) = structure factor)
(4-54)). The “crossover” from the mean- becomes anisotropic:
field regime (where the Landau description S (q) ~ [x – 2 + q2 + g (qz /q)2 ]–1 (4-65 a)
is essentially appropriate) to the non-mean-
field regime occurs at a reduced tempera- where qz is the z-component of q and g is a
ture distance t = tcr , where Eq. (4-62) is constant. Whereas Eq. (4-32) implies that
treated as an equality: for d = 3, S (q x = 1) = S (q = 0)/2, the equation q x = 1
which for Eq. (4-32) defines a sphere of ra-
tcr ~ (R/a)– 6 (4-63) dius x –1 in q-space now defines an aniso-
tropic surface. From
Hence systems with a large but finite range
of interaction behave in an essentially Lan- x – 2 = q2 + g (qz /q)2 (4-65 b)
dau-like manner. An example of such a be-
we obtain an object having the shape of a
havior is the unmixing critical point in
flat disc, with radius 1/x in the qx – qy plane
polymer fluid mixtures with a high degree
but maximum extension of order 1/x 2 in
of polymerization N : since each coil has a
the qz direction. This object, defined by
radius rc ~ N 1/2, the monomer density in-
S (q) = –12 max {S (q)}, can be interpreted as
side the sphere taken by a polymer coil is
the Fourier transform of the correlation
only of order r ~ N/rc3 ~ N – 1/2, which im-
volume. This implies that the correlation
plies that each coil interacts with N 1/2
volume in real space is a long ellipsoid,
neighbor coils, and the effective interaction
with linear dimensions x in the qx , qy di-
volume (R/a)d in Eqs. (4-62) and (4-63)
rection but with linear dimension x 2 in the
should be taken as N 1/2, i.e., tcr ~ N – 1 (De
qz direction. In d-dimensions, this argu-
Gennes, 1979; Binder 1984 c, 1994).
ment suggests for uniaxial dipolar systems:
At this point, we emphasize that using
N (t) = [x (t)/a]d in Eq. (4-60) is not valid in N (t) = [x (t)/a]d +1 (4-66)
systems with long-range anisotropic inter-
which yields in Eq. (4-60)
actions, such as uniaxial dipolar magnets.
For a system such as LiTbF4 (Als-Nielsen const. O t – n (d +1) + 2 b + g = t – (d – 3)/2 (4-67)
where in the last equality, the Landau ex- well over a wide temperature range (Salje,
ponents have been used. Comparing Eqs. 1990).
(4-61) and (4-67) shows that d-dimensional For all other systems we do not expect
uniaxial dipolar systems somehow corre- the Landau critical exponents to be an ac-
spond to (d + 1)-dimensional systems with curate description of the critical singular-
isotropic short-range forces. Therefore the ities. This is borne out well by the exact so-
marginal dimension d*, above which the lution for the two-dimensional Ising model
Landau theory predicts the values of criti- (Onsager, 1944; McCoy and Wu, 1973),
cal exponents correctly, is d* = 3 for uniax- where the critical exponents have the values
ial dipolar systems, unlike the standard iso-
a = 0 (log), b = 1/8, g = 7/4,
tropic short-range case where d* = 4.
For elastic phase transitions in cubic d = 15, n = 1, h = 1/4 (4-71)
crystals where the combination of elastic Obviously, these numbers are a long way
constants c11 – c12 softens as the critical from the predictions of the Landau theory.
point is appproached, such as the system Note that a = 0 in Eq. (4-71) has the mean-
PrAlO3 , the structure factor S (q) has the ing C ~ | log | t ||, unlike the jump singular-
form: (4-68) ity of the Landau theory.
S (q) ~ [x – 2 + q2 + A (qz /q)2 + B (q^ /q)2 ]– 1 Even in the marginal case of uniaxial di-
polar ferromagnets (or uniaxial ferro-
where q = (qz , q^ ) and A and B are con- electrics; see, e.g., Binder et al. (1976) for
stants. In this case the correlation vol- a discussion), the Landau theory is not
ume defined from the condition S (q) = completely correct; it turns out (Larkin and
–12 max {S (q)} in real space is a flat disk Khmelnitskii, 1969; Aharony, 1976) that
with radius x 2 and diameter x , i.e., for elas- the power laws have the Landau form but
tic systems: are modified with logarithmic correction
N (t) = [x (t)/a]d +2 (4-69) factors:
In this case, the Ginzburg criterion, Eq. cT = Ĝ± t – 1 | ln t |1/3, tÆ0 (4-72 a)
(4-60), yields 1/3
C = Â± | ln t | , tÆ0 (4-72 b)
– n (d +2) + 2 b + g – (d – 2)/2
const. O t =t (4-70) Ms = B̂ (– t)1/2 | ln (– t)|1/3, t Æ 0 (4-72 c)
instead of Eqs. (4-62) or (4-67), respec- Ms | Tc = D̂ H 1/3 1/3
| ln H | , HÆ0 (4-72 d)
tively. Since the marginal dimensionality
for such systems is d* = 2, three-dimen- Although for uniaxial ferroelectrics the ex-
sional systems should be accurately de- perimental evidence in favor of Eq. (4-72)
scribed by the Landau theory, and this is is still scarce, convincing experimental ev-
what is found experimentally for PrAlO3. idence does exist for dipolar ferromagnets
Examples for elastic phase transitions such as LiTbF4 (see, e.g., Als-Nielsen and
which are of second order are scarce, as is Laursen (1980) for a review).
understandable from the symmetry consid- Eq. (4-72) results from including fluctu-
erations which lead to first-order transi- ation contributions to the functional inte-
tions in most cases (Cowley, 1976; Folk gral in Eq. (4-54) systematically, which can
et al., 1976). However, for many first-order be done in a most powerful way be renor-
ferroelastic transitions the Landau descrip- malization group theory (Aharony, 1973,
tion also fits the experimental data very 1976; Fisher, 1974; Wilson and Kogut,
1974; Ma, 1976; Amit, 1984; Yeomans, “universality”. Again we wish to convey
1992). A description of this theory is be- only the flavor of the idea to the reader,
yond the scope of this chapter; we only rather than to present a thorough discus-
mention that it is this method which yields sion. Consider, for example, the decay of
highly accurate approximations for the val- the correlation function of order parameter
ues of critical exponents of three-dimen- fluctuations at the critical point, Eq. (4-47):
sional systems. For example, for the uniax- changing the length scale from r to r¢ = L r
ial magnets (as described by the Ising Ha- would change the prefactor Ĝ but leave the
miltonian, Eq. (4-56), experimental exam- power-law invariant. The physical inter-
ples being the antiferomagnets MnF2 or pretation of this fact is the “fractal struc-
FeF2) or the ordering alloy b-brass, the ex- ture” of critical correlations (Mandelbrot,
ponents are predicted to be (LeGuillou and 1982): just as a fractal geometric object
Zinn-Justin, 1980): looks the same on every length scale, the
pattern of critical fluctuations looks the
a ≈ 0.110, b ≈ 0.325, g ≈ 1.240,
same, irrespective of the scale. Slightly off
d ≈ 4.82, n ≈ 0.63, h ≈ 0.032 (4-73) Tc , then, there should be only one relevant
Whereas within the Landau theory the “or- length scale in the problem, the correlation
der parameter dimensionality” n does not length x , which diverges as T Æ Tc , and de-
matter, as far as the values of the critical tails such as the precise behavior of the
exponents are concerned, n does matter if correlation functions for interatomic dis-
we go beyond the Landau theory. For ex- tances should not matter. As a result, it is
ample, for isotropic magnets as described plausible that the correlation function
by the well known Heisenberg model of G (x, x ) should depend only on the ratio of
magnetism, the two lengths x and x , apart from a scale
factor:
Ᏼ Heis = − ∑ Jij Si ⋅ S j − m H ∑ Siz (4-74)
i≠ j G (x, x ) = Ĝ x – (d – 2 + h ) G̃ (x /x ) (4-76)
Si being a unit vector in the direction of the Obviously, the condition G̃ (0) = 1 for the
magnetic moment at lattice site i, we have “scaling function” G̃ (x /x ) ensures that Eq.
(LeGuillou and Zinn-Justin, 1980): (4-76) crosses over smoothly to Eq. (4-47)
a ≈ – 0.116, b ≈ 0.365, g ≈ 1.391, as T Æ Tc , where x Æ ∞.
d ≈ 4.82, n ≈ 0.707, h ≈ 0.034 (4-75) From the “scaling hypothesis” in Eq.
(4-76) we immediately derive a “scaling
Again these results are in fair agreement law” relating the exponents g , n and h. Us-
with available experimental data, such as ing Eq. (4-44) and denoting the surface of
the isotropic antiferromagnet RbMnF3 (see a d-dimensional unit sphere as Ud (Ud = 4 p
Als-Nielsen (1976) for a review of experi- in d = 3) we obtain
ments on critical point phenomena). Note
1 G˜
that the negative value of a in Eq. (4-75) cT = ∑ G (x) ≈ ∫ G ( x , x ) dx
implies that the specific heat has a cusp of kB T x kB Tc
finite height at Tc . ∞
Ud Gˆ 1−h
The renormalization group theory also = ∫x G˜ ( x / x ) dx
kB Tc 0
provides a unifying framework and justifi-
∞
cation for two important concepts about 2 −h Ud Gˆ 1−h
=x ∫ᒗ G˜ (ᒗ) dᒗ (4-77)
critical phenomena, namely “scaling” and kB Tc 0
Since the integral over ᒗ = x /x in Eq. (4-77) cancel out. On the other hand, this also
yields a constant, we conclude yields
c T ~ x (2 – h ) ~ | t | – n (2– h ) ~ | t | – g F | T = Tc ~ H 1/(1 + g /b ) = H1/d ,
cf. Eqs. (4-4) and (4-35). Thus the scaling i.e., d = 1 + g /b (4-83)
relation results in
The scaling assumptions also imply that
g = n (2 – h ) (4-78) exponents above and below Tc are equal,
i.e, a = a ¢, g = g¢, and n = n¢. Now there is
Eq. (4-76) indicates that G (x, x ) depends
still another scaling law which results from
on the two variables x and x in a rather spe-
considering all degrees of freedom inside a
cial form, namely, it is a homogeneous func-
correlation volume x d to be highly corre-
tion. A similar homogeneity assumption is
lated with each other, while different corre-
also true for the singular part of the free
lation volumes can be considered as essen-
energy, (4-79) tially independent: this argument suggests
F = Freg + t 2 – a F̃ (H̃ ), H̃ = H Ĝ t – g – b/B̂ that the singular part of the Helmholtz en-
ergy per degree of freedom can be written
where the singular temperature dependence
as Fsing ≈ [1/Ns (t)] ¥ const., since indepen-
for H̃ = 0 is chosen to be compatible with
dent degrees of freedom do not contribute
Eq. (4-5), Freg is a background term which
any singular free energy. Since
is analytic in both T and H even for T = Tc ,
while F̃ (H̃ ) is another “scaling function”. Ns (t) ~ x d ~ |t | – dn
At this point, we present Eq. (4-79) as a
postulate, but it should be emphasized that we conclude, by comparison with Eq.
both Eqs. (4-76) and (4-79) can be justified (4-79), that
from Eq. (4-54) by the renormalization dn = 2 – a (4-84)
group approach.
Combining Eqs. (4-1) and (4-79) we ob- Note that this so-called “hyperscaling rela-
tain: tion” with the Landau-theory exponents is
(4-80)
Gˆ ( 2 − a − g − b ) ˜ ˜ only true at the marginal dimension d* = 4,
F = − ( ∂F / ∂H )T = − t F′ ( H ) whereas the other scaling relationships ob-
Bˆ
viously are fulfilled by Landau exponents
Defining – (Ĝ /B̂) F̃ ¢ (H̃ ) ∫ B̂ M̃ (H̃ ) with independent of the system dimensionality.
M̃ (0) = 1, we obtain: However, all scaling relationships (includ-
ing Eq. (4-84)) are satisfied for the two-
F = B̂t b M̃ (H̃), b = 2 – a – g – b (4-81)
and three-dimensional Ising model (Eqs.
since for H̃= 0, Eq. (4-80) must reduce to (4-71) and (4-73)) and the three-dimen-
Eq. (4-6). Taking one more derivative we sional Heisenberg model, Eq. (4-75).
find Let us consider the two-dimensional
cT = (∂F /∂H )T = Ĝ t – g M̃¢(H̃) (4-82) Heisenberg model. For this model the ef-
fects of statistical fluctuations are so strong
hence Eq. (4-79) is compatible with Eq. that they destabilize the ordering alto-
(4-4), as it should be. The condition that gether: no spontaneous ordering exists in
Eq. (4-81) reduces to the critical isotherm d = 2 for any system with order parameter
for t Æ 0 requires that M̃ (H̃ Æ •) ~ dimensionality n ≥ 2, and a critical point
H̃ b /(g + b ), in order that the powers of t occurs only at zero temperature, Tc = 0, for
n > 2. Thus d = 2 is the “lower critical di- model of magnetism, Eq. (4-74), since the
mensionality” d1 for n ≥ 2, whereas for ferroelectric ordering would also be truly
n = 1 we have d1 = 1: quasi-one-dimensional isotropic. However, in the presence of a
orderings are always unstable. The case nonvanishing “cubic anisotropy”, u¢ ≠ 0, in
n = 2, d = 2 is very special: a phase transi- general a different universality class re-
tion still occurs at a nonzero Tc , the so- sults. The effect of such higher order invar-
called “Kosterlitz – Thouless” transition iants in the Landau expansion are particu-
(Kosterlitz and Thouless, 1973); for T < Tc larly drastic again in systems with reduced
we have Ms (T ) ∫ 0 while at the same time dimensionality; e.g., the model with n = 2
x (T ) = •, and the correlation function but cubic anisotropy no longer exhibits a
shows an algebraic decay, Eq. (4-47), with Kosterlitz – Thouless transition, but rather
a temperature-dependent exponent h . At a nonzero order parameter is stabilized
first sight, this behavior may appear fairly again. Owing to the “marginal” character
esoteric, but in fact it is closely related of the cubic anisotropy in d = 2, however,
to the “roughening transition” of crystal the behavior is not Ising-like but rather the
surfaces (Weeks, 1980). Such roughening pathological case of a “universality class”
transitions have been observed for high-in- with “nonuniversal” critical exponents oc-
dex crystal faces of various metals (Sala- curs. The latter then do depend on “micro-
non et al., 1988). The power-law decay of scopic” details, such as the ratio of the
correlations, Eq. (4-47), can be related to interaction strengths between nearest and
the behavior of the height – height correla- next nearest neighbors in the lattice (Krin-
tion function of a crystal surface in the sky and Mukamel, 1977; Domany et al.,
rough state (Weeks, 1980). As is well 1978; Swendsen and Krinsky, 1979; Lan-
known, the roughness or flatness of crystal dau and Binder, 1985). Again such prob-
surfaces has a profound effect on their ad- lems are not purely academic, but relevant
sorption behavior, crystal growth kinetics, for order – disorder transitions in chemi-
etc. (Müller-Krumbhaar, 1977). sorbed layers on metal surfaces, such as O
The other important concept about criti- on W (110) (see Binder and Landau (1989)
cal phenomena is “universality”: since the for a review of the theoretical modeling of
only important length scale near a critical such systems).
point is provided by the correlation length One important consequence of univer-
which is much larger than all “micro- sality is that all gas – fluid critical points in
scopic” lengths such as lattice spacing and three-dimensional systems, all critical
interaction range, it is plausible that “de- points associated with unmixing of fluid or
tails” on the atomic scale do not matter, and solid binary mixtures, and all order – disor-
systems near a critical point behave in the der critical points involving a single-com-
same way provided that they fall in the ponent order parameter (such as b-brass)
same “universality class”. It turns out that belong to the same universality class as the
universality classes (for systems with three-dimensional Ising model. All these
short-range interactions!) are determined systems not only have the same critical ex-
by both spatial dimensionality d and order ponents, but also the scaling functions
parameter dimensionality n, and in addi- M̃ (H̃ ), G̃ (ᒗ) etc. are all universal. Also,
tion, the symmetry properties of the prob- critical amplitude ratios are universal,
lem: e.g., Eq. (4-25) for n = 3 and u¢ = 0 such as Ĉ/Ĉ +, (Eq. (4-4), Â/Â¢ (Eq. (4-5)),
falls in the same class as the Heisenberg or more complicated quantities such as
D̂ Ĝ B̂d –1. The theoretical calculation and netization (due to the coupling F 2 (x) M (x)
experimental estimation of such critical in the Hamiltonian, as mentioned above).
amplitude ratios are discussed by Privman However, if we now consider the variation
et al. (1991). of F with M, combining both laws we get
An interesting problem concerns the (assuming a > 0)
critical singularities associated with nonor-
dering “fields”. These singularities occur
F ~ [Mc (T ) – M ]b /(1– a ) (4-85)
because the ordering field F (x) considered Considering now the phase transition in the
only in Eqs. (4-10) and (4-54) couples to space of {M, T} as independent thermody-
another quantity: e.g., for an antiferromag- namic parameters, we then find that on
net in a magnetic field the order parameter crossing the critical line Mc (T ) the expo-
F (the sublattice magnetization) couples to nent describing the vanishing of the order
the uniform magnetization M, to the lattice parameter is b /(1 – a ) rather than b. This
parameters, to the electron density, etc. effect generally occurs when the critical
Since the Hamiltonian is invariant against line depends on extensive (rather than in-
interchange of the sublattices which im- tensive) thermodynamic variables and is
plies a sign change of F , this coupling is called “Fisher renormalization” (Fisher,
quadratic in F 2. Whereas in mean field 1968). This is very common for order – dis-
theory the resulting critical behavior of the order phenomena in adsorbed layers at
nonordering “field” is then proportional to fixed coverage, or in alloys at fixed con-
·F Ò2 ~ t 2 b , taking fluctuations into account, centration, for unmixing critical points in
the actual critical behavior is proportional ternary mixtures, etc. Other exponents also
to ·F 2 Ò ~ t (1– a ), i.e., an energy-like singu- become “renormalized” similarly, e.g., g is
larity. Such energy-like singularities are replaced by g /(1 – a ), etc. A detailed anal-
predicted for the electrical resistivity r el at ysis (Fisher, 1968) shows that no such
various phase transitions (Fisher and “renormalization” of critical exponents oc-
Langer, 1968; Binder and Stauffer, 1976 a), curs for systems with a fixed concentration
for the refractive index n r (Gehring and c if the slope of the critical line vanishes,
Gehring, 1975; Gehring, 1977), etc. Look- dTc (c)/dc = 0, for the considered concen-
ing for anomalies in the temperature deriv- tration. This is approximately true for b-
ative dr el (T )/dT, dn r (T )/dT etc., which brass, for instance.
should exhibit specific-heat-like singular-
ities, is often a more convenient tool for lo-
4.2.3 Second-Order Versus First-Order
cating such phase transitions than measure-
Transitions; Tricritical and Other
ments of the specific heat itself.
Multicritical Phenomena
An important phenomenon occurs when
a nonordering field which is coupled to the An important question for any phase
order parameter in this way, such as the transition is deciding a priori whether it
magnetization in an antiferromagnet in an should be a second- or first-order transi-
external field, is held fixed. Suppose we tion; general principles are sought in an-
first study the approach to criticality by let- swering this question, such that there
ting the uniform magnetic field h tend to its would be no need for specific experimental
critical value hc (T ): we then have the law data.
F ~ [hc (T ) – h]b for the order parameter In the Landau theory, general symme-
and M – Mc (T ) ~ [hc (T ) – h]1– a for the mag- try conditions exist which allow second-
order transitions (Lifshitz, 1942). We here by Baxter (1973), these models have sec-
merely state them, without even attempting ond-order transitions. The critical expo-
to explain the group-theoretical language nents for these models are now believed
(see, e.g., Tolédano and Tolédano, 1987, to be known exactly, e.g., (q = 3) a = 1/3,
for a thorough treatment): b = 1/9, g = 13/9 and (q = 4) a = 2/3,
b = 1/12, g = 7/6 (Den Nijs, 1979; Nienhuis
(i) The order parameters F transform as
et al., 1980). As mentioned in Sec. 4.2.1, an
a basis of a single irreducible repre-
experimental example for the three-state
sentation X of the group G0 character- –
Potts model is the ÷3 superstructure of var-
izing the symmetry properties of the
ious adsorbates on graphite. Convincing
disordered phase.
experimental evidence for a specific heat
(ii) The symmetric part of the representa-
divergence of He4 on grafoil described by
tion X3 should not contain the unit
a = 1/3 was presented by Bretz (1977). The
representation.
system O on Ru (001) in the p (2 ¥ 2) struc-
(iii) If the antisymmetric part of X2 has a
ture (Piercy and Pfnür, 1987) falls into the
representation, the wavevector q asso-
class of the four-state Potts models and
ciated with X is not determined by
again the data are in reasonable agreement
symmetry. In this case q is expected
with the predicted critical exponents.
to vary continuously in the ordered
In d = 3 dimensions, however, the three-
phase.
states Potts model has a weak first-order
If these conditions are met, a transition transition (Blöte and Swendsen, 1979) so
can nevertheless be first order, because a that the Landau rule (ii) is not violated.
fourth-order term can be negative (see Sec. This is of relevance for metallurgical sys-
4.2.1). If they are not met, the transition tems such as the CuAu ordering (Fig. 4-8)
must be first order, according to the Lan- on the f.c.c. lattice, which, according to
dau rules. Domany et al. (1982), belongs to the class
The first of these rules essentially says of the three-states Potts model, while the
that if in the ordered phase two quantities Cu3Au structure belongs to the class of the
essentially independent of each other (not four-state Potts model. Another example of
related by any symmetry operation, etc.) the three-state Potts model is the structural
play the role of order parameter compo- transition of SrTiO3 stressed in the [111]
nents F1 , F2 , there is no reason why in the direction (Bruce and Aharony, 1975). The
quadratic term r1 (T ) F12 + r2 (T ) F22 of the order – disorder transition for all systems
Landau expansion the coefficients r1 (T ), described by these types of ordering are al-
r2 (T ) should change their sign at the same ways of first order, so there is no contradic-
temperature. Thus, if F1 , F2 are “primary” tion with this Landau rule, except for the
order parameter components, they should transition to a charge density wave state in
appear via a first-order transition. 2H-TaSe2 (Moncton et al., 1977) which ap-
The second condition essentially implies parently is second order although there is a
the absence of third-order terms (F 3 ) in the third-order invariant (Bak and Mukamel,
Landau expansion. It turns out, however, 1979). It is possible, of course, that the
that in d = 2 dimensions there are well- transition is very weakly first order so that
known counter-examples to these rules, experimentally it could not be distin-
namely the Potts model (Potts, 1952) with guished from second order. This remark
q = 3 and q = 4 (see Eq. (4-20)); as shown also holds for other examples of apparent
alternating magnetic moment direction be-

tween adjacent sheets (cf. Figs. 4-4 and
4-8 a). Experimentally first-order transi-
tions are known for FeO (Roth, 1958), TbP
(Kötzler et al., 1979) (these systems have
order parameter dimensionality n = 4), UO2
(Frazer et al., 1965; this is an example with
n = 6), MnO (Bloch and Mauri, 1973), NiO
(Kleemann et al., 1980) (examples with
n = 8), etc. The standard phenomenological
understanding of first-order transitions in
these materials invoked magnetostrictive
couplings (Bean and Rodbell, 1962) or
crystal field effects (in the case of UO2 ; see
Blume, 1966), which make the coefficient
u in Eq. (4-10) negative and thus produce a
free energy of the type shown in Fig. 4-6 b.
It has been suggested that the first-order
character of the phase transition in these
materials is a fundamental property due
to the large number n of order-parameter
components and the symmetry of the
Hamiltonian: renormalization group ex-
Figure 4-8. Five crystallographic superstructures on
the face-centered cubic (fcc) lattice and their equiva- pansions discussed by Mukamel et al.
lent antiferromagnetic superstructures if in a binary (1976 a, b) and others suggested that all
alloy A-atoms (full circles) are represented as “spin such transitions are “fluctuation-induced
up” and B atoms (open circles) are represented as first-order transitions”, i.e., all antiferro-
“spin down”. Case (a) refers to the CuAuI structure,
magnets with n ≥ 4 order parameters must
(b) to the A2B2 structure, (c) to the A3B structure of
Cu3Au type, (d) to the A3B structure of Al3Ti type, have first-order transitions. However, there
and (e) the A2B structure as it occurs in Ni2V or seem to be many counter-examples of cu-
Ni2Cr. In case (a) the labels 1, 2, 3, and 4 indicate the bic n ≥ 4 antiferromagnets with phase tran-
decomposition of the fcc lattice into four interpene- sitions of apparently second order, such as
trating simple cubic (sc) sublattices.
BiTb (Kötzler, 1984), GdTb (n = 4; Hulli-
ger and Siegrist, 1979), and GdSb (n = 8;
McGuire et al., 1969). The reasons for
violations of the Landau rules discussed by this discrepancy between renormalization
Tolédano and Pascoli (1980), Tolédano group predictions and experiments are not
(1981), and Tolédano and Tolédano (1987). clear.
There are many transitions which the There are various other cases where
above Landau symmetry criteria would “fluctuation-induced first-order transitions”
permit to be of second order but which are occur, as reviewed in detail by Binder
actually observed to be first order. Exam- (1987 a). Here we mention only the exam-
ples are type I or type II antiferromagnetic ple of the phase transition from disordered
structures, consisting of ferromagnetic block copolymer melts to the lamellar mes-
(100) or (111) sheets, respectively, with an ophase (Fig. 4-7 b). For this system the
free-energy functional does not attain its which implies a specific heat divergence
minimum at wavevector q = 0 in reciprocal C ~ (– t)–1/2, i.e., a t = 1/2. We note that this
space but on a surface given by the equa- set of exponents also satisfies the scaling
tion | q | = q* = 2 p/l , l being the wave- laws in Eqs. (4-78), (4-81), (4-83), and
length of the lamellar pattern. This reflects (4-84). Moreover, using this set of expo-
the degeneracy that there is no preferred di- nents in the Ginzburg criterion, Eqs. (4-58)
rection and hence the lamellae may be to (4-61), we find that, with the Landau ex-
oriented in any direction. Owing to the ponents for a tricritical point, this condi-
large “phase space” in q-space where the tion is marginally fulfilled, unlike the case
inverse of the wavevector-dependent sus- of ordinary critical points (Eq. (4-62)). It
ceptibility goes “soft” (c –1 ( | q | = q*) Æ 0 turns out that d* = 3 is the marginal dimen-
as T Æ Tc ), the mean-square amplitude of sion for tricritical points, and the mean-
the local fluctuation of the order parameter field power laws are modified by logarith-
·F 2 Ò – ·F Ò2 would diverge as T Æ Tc , mic correction terms similar to those noted
which is physically impossible. It can be for dipolar systems (Eq. (4-72)). Experi-
shown that this difficulty is avoided be- mental examples for tricritical points in-
cause the strong local order parameter fluc- clude strongly anisotropic antiferromagnets
tuations turn u negative near Tc (Brazov- in a uniform magnetic field (e.g., FeCl2 ,
skii, 1975; Fredrickson and Helfand, 1987; (Dillon et al., 1978), and dysprosium alu-
Fredrickson and Binder, 1989), producing minum garnet (DAG) (Giordano and Wolf,
a free energy, as shown in Fig. 4-6 b, for the 1975), see Fig. 4-9 for schematic phase
case of “symmetric” diblock copolymers diagrams), systems undergoing structural
where mean field theory would predict a phase transitions under suitable applied pres-
second-order transition. Experiments seem sure, such as NH 4Cl which has a tricritical
to confirm the first-order character of the point at Tt = 250 K and pt = 128 bar (Yelon et
transition (Bates et al., 1988). al., 1974) or the ferroelectric KDP at Tt =113
For certain systems, for which u > 0 and K and pt = 2.4 kbar (Schmidt, 1978), etc. A
thus a second-order transition occurs, it is model system which has been particularly
possible by variation of a non-ordering carefully studied is He3 –He4 mixtures: the
field to change the sign of the coefficient u transition temperature Tl (x) of the normal
at a particular point. This temperature Tt , fluid–superfluid He4 is depressed with in-
where u vanishes in Eq. (4-14), is called a creasing relative concentration x until, at a
tricritical point. From Eq. (4-14) we then tricritical point Tt = Tl (x t ), the transition be-
immediately find comes first order. This then implies a phase
(4-86)
separation between a superfluid phase with
F0 = (– r /v)1/4 = (r ¢/v)1/4 (Tt /T – 1)1/4 a lower He3 content and a normal fluid He3-
Hence the tricritical order parameter expo- rich phase. Most common, of course, are
nent b t = 1/4, while Eqs. (4-32) to (4-48) re- tricritical phenomena in fluid binary mix-
main unchanged and thus g t = 1, n t = 1/2, tures arising from the competition of gas –
h t = 0. The critical isotherm, however, be- liquid transitions and fluid – fluid phase
comes v F 05 = H/kB T, i.e., d t = 5. From Eq. separations in these systems (Scott, 1987).
(4-15) we finally find Just as the vanishing coefficient u in Eq.
(4-10) leads to a multicritical point, the tri-
F (F0 ) − F ( 0 ) t − t 1/ 2
=− ⎛ ⎞
critical point, another multicritical point is
(4-87)
V kB T 3⎝ v ⎠ associated with the vanishing of the coeffi-
The peak height of c (q) for |q | = qmax is

then described by
Gˆ
c ( qmax ) = (4-90)
t0 − K12 /( 4 K2 )
which implies that the actual critical point
occurs at Tc = TL [1 + K12/(4 K2 )]. An exam-
ple of such an ordering characterized by a
Figure 4-9. Schematic phase diagrams of antiferro- nonzero wavenumber q*= qmax is the meso-
magnets with uniaxial anisotropy in an applied uni- phase formation in block copolymer melts
form magnetic field H|| applied parallel to the easy (Fig. 4-7) (see Leibler (1980)). Related phe-
axis: Case (a) shows the case of weak anisotropy,
nomena also occur in crystals exhibiting
case (b) the case of intermediate anisotropy, where in
addition to the antiferromagnetic ordering of the spin “incommensurate superstructures”. Usu-
components in the direction of the easy axis, a spin- ally c (q) is then not isotropic in q-space; in
flop ordering of the transverse components also oc- uniaxial systems, Eqs. (4-88) to (4-90) then
curs. In case (a) both transitions T|| (H|| ), and T^ (H|| ) apply only if q is parallel to this preferred
are of second order and meet in a bicritical point. For
direction. The superstructure described by
intermediate strength of the anisotropy the line
T^ (H|| ) does not end at the bicritical point, but rather the wavelength l = 2 p/qmax is determined
in a critical end-point at the first-order transition line. by the coefficients K1 and K2 and thus in
Then a tricritical point also appears where the anti- general it is not a simple multiple of the lat-
ferromagnetic transition T|| (H|| ) becomes first order. tice spacing a, but (l /a) is an irrational
For very strong anisotropy, the spin-flop phase disap-
number. This is what is meant by the term
pears altogether.
incommensurate. Examples of such incom-
mensurate structures are “helimagnetic
cient of the [—F (x)]2 term; since this term structures” such as VF2 , MnAu2 , Eu, Ho
was also introduced as the lowest-order and Dy (Tolédano and Tolédano, 1987).
term of a systematic expansion, higher-or- Note that in these systems the reason for
der terms such as [—2 F (x)]2 must then be the incommensurate structure is not the
included in Eq. (4-10). This problem is existence of a negative coefficient in front
most conveniently discussed in the frame- of the [—F (x)]2 term but the existence of
work of the wavevector-dependent suscep-
⎛ ∂F − ∂F + ⎞
tibility c (q), Eq. (4-34), which including terms such as ⎜F + − F− ⎟ , where
the contribution resulting from a term ⎝ ∂z ∂z ⎠
[—2 F (x)]2 can be written as (t0 = 1 – TL /T ) F + = F 0 cos (p z /c), F – = F 0 sin (p z /c), c
being the lattice spacing in the z-direction.
Gˆ0 Such terms involving linear terms in —F (x)
c (q) = (4-88)
t0 + K1 q 2 + K2 q 4 are allowed by symmetry in certain space
where K1 and K2 are phenomenological co- groups and are called “Lifshitz invariants”.
efficients. If K1 < 0, the divergence of c (q) Moving from lattice plane to lattice plane
does not occur for t0 = 0, q = 0 but at a value in the z-direction in such a helimagnetic
qmax found from: spin structure, the spin vector (pointing
perpendicular to z) describes a spiral struc-
d ture like a spiral staircase. Of course, such
(t0 + K1 q 2 + K2 q 4 ) = 0
dq incommensurate superstructures (also
⇒ qmax = ( − K1 / 2 K2 )1/ 2 (4-89) called “modulated phases”) are not restrict-
ed to magnetic systems, and many exam- such as pressure, magnetic field (for mag-
ples of dielectric incommensurate phases netic structures), concentration (for alloys;
have been identified (e.g., NaNO2 (Tani- see Fig. 4-10), etc.). Theoretically under
saki, 1961), Rb2ZnCl 4 , BaMnF4 , thiourea certain conditions even a staircase with an
[SC(NH2 )2 ], see Tolédano and Tolédano infinite number (mostly extremely small!)
(1987)). Whereas Rb2ZnCl 4 and BaMnF4 of steps can be expected (“devil’s stair-
also involve the existence of Lifshitz invar- case”, see Selke (1988, 1989, 1992) for a
iants, thiourea is an example of the case discussion and further references). We also
discussed in Eqs. (4-88) to (4-90). Various emphasize that the modulation need not in-
examples of modulated superstructures ex- volve only one direction in space (one
ist in metallic alloys; see Selke (1988, wavevector), but can involve several wave-
1989, 1992) and De Fontaine and Kulik
(1985) for reviews of the pertinent theory
and experimental examples such as Al3Ti
(Loiseau et al., 1985) and Cu3Pd (Broddin
et al., 1986).
Whereas for TeTc the ordering of the
incommensurate phase can be described in
terms of a sinusoidal variation of the local
order parameter density F (x), the nonlin-
ear terms present in Eq. (4-10) at lower
temperatures imply that higher-order har-
monics become increasingly important.
Rather than a sinusoidal variation, the
structure is then better described in terms
of a periodic pattern of domain walls (or
“solitons”), and we talk about a “multi-
soliton lattice” or “soliton staircase”. An-
other important fact about modulated
phases is that the wavevector characteriz-
ing the modulation period is not fixed at
qmax , but varies with temperature or other
parameters of the problem. This is ex-
pected from the third of the general Landau
rules formulated at the beginning of Sec.
4.2.3. Usually this variation of q stops at Figure 4-10. Period M (in units of the lattice spacing
some commensurate value where a “lock- in the modulation direction) for the alloy Al3 – xTi1+ x
as a function of the annealing temperature, as ob-
in transition” of the incommensurate struc- tained from visual inspection of high-resolution elec-
tures occurs. In the incommensurate re- tron microscopic images. Each of the fifteen different
gime rational values of the modulation commensurate superlattices observed is composed of
(“long-period superstructures”) may have antiphase domains of length one or two (1-bands or
an extended regime of stability, leading to 2-bands, as indicated by sequences ·211Ò, ·21Ò etc.)
based on the L12 structure. Very long annealing times
an irregular staircase-like behavior of the were needed to produce these long-period super-
modulation wavelength l as a function of structures supposedly at equilibrium. From Loiseau
temperature (or other control parameters et al. (1985).
vectors. An example of the case of a two- While experimental examples for Lif-
rather than one-dimensional modulation is shitz points are extremely scarce, e.g., the
LiKSO4 (Pimenta et al., 1989). structural transition in RbCaF3 under (100)
Here we discuss only the special case stress (Buzaré et al., 1979), more common
of Eqs. (4-88) – (4-90) briefly where, by multicritical points are bicritical points
changing external control parameters, it where two different second-order transi-
is possible to reach a multicritical point tion lines meet. Consider, for example, the
where K1 vanishes, the so-called “Lifshitz generalization of Eq. (4-10) to an n-compo-
point” (Hornreich et al., 1975). From Eq. nent order-parameter field, but with differ-
(4-88) we then find that for T = Tc (= TL ) we ent coefficient ri for each order-parameter
have c (q) ~ q – 4, and comparing this with component F i :
the result following from a Fourier trans-
1 F
formation of Eq. (4-47), c (q)Tc ~ q – 2 + h , Ᏺ [F ( x )] = 0 (4-92 )
we conclude that h L = – 2 for the Lifshitz kB T kB T
+ ∫ dx ⎧⎨ r1 F12 ( x ) + r2 F22 ( x ) + …
point, whereas for T > Tc we still have 1 1
c (q = 0) = Ĝ0 t0–1, i.e., g L = 1. Since for ⎩ 2 2
K1 = 0 Eq. (4-88) can be rewritten as 1 u
+ rn Fn2 ( x ) + [F 2 ( x )]2
2 4
c (q) = Ĝ0 (1 – TL /T )–1/ [1 + x 4 q 4 ] ⎫
R2
x = [K2 /(1 – TL /T )]1/4 (4-91) + [(∇F1 )2 + … + (∇Fn )2 ]⎬
2d ⎭
we conclude that nL = 1/4 for the correla- For simplicity, the fourth-order term has
tion length exponent at the Lifshitz point. been taken as fully isotropic. If r1 = r2 = …
Also for the Lifshitz point we can ask = rn , we would have the isotropic n-vector
whether the Landau description presented model discussed in Secs. 4.2.2 and 4.2.3.
here is accurate, and whether or not statis- We now consider the case where some of
tical fluctuations modify the picture. The these coefficients differ from each other,
result is that the critical fluctuations are but a parameter p exists on which these co-
very important. A marginal dimensionality efficients depend in addition to the temper-
d* = 8 results in the case of an “isotropic ature. The nature of the ordering will be
Lifshitz point”, where in determined by the term ri F i2 , for which
K1x (∂F /∂x)2 + K1y (∂F /∂y)2 + K1z (∂F /∂z)2 the coefficient ri changes sign at the high-
est temperature. Suppose this is the case
we have simultaneous vanishing of all co- where i = 1 for p < pb , then we have a one-
efficients K1x , K1y , and K1z . Note that the component Ising-type transition at Tc1 ( p)
more common case is the “uniaxial Lifshitz given by r1 ( p, T ) = 0 (the other compo-
point”, where only K1z vanishes while K1x nents F i being “secondary order parame-
and K1y remain nonzero. In this case the ters” in this phase). If, however, for p > pb
multicritical behavior is anisotropic, and the coefficient r2 ( p, T ) = 0 at the highest
we must distinguish between the correla- temperature Tc2 ( p), it is the component F 2
tion length x || describing the correlation which drives the transition as the primary
function decay in the z-direction and the order parameter. The point p = pb , Tc1 ( p) =
correlation length x ^ in the other direc- Tc2 ( p) = Tb is then called a bicritical point.
tions. Eq. (4-91) then holds for x || only, An example of this behavior is found for
whereas Eq. (4-39) still holds for x ^. weakly uniaxial antiferromagnets in a uni-
form magnetic field H|| (Fig. 4-9). A model order parameter, due to the perpendicu-
Hamiltonian for such a system can be writ- lar components Msx and Msy of the stag-
ten as gered magnetization). In the framework of
Eq. (4-92), this would mean n = 3 and
Ᏼ = − ∑ {J (1 − D ) [ Six S jx + Siy S jy ]
〈 i, j 〉
r2 ∫ r3 ≠ r1 , with H|| being the parameter p.
+ J Siz S jz } − H|| ∑ Siz (4-93) Both lines TN|| (H|| ) and TN^ (H|| ) and the
i
first-order line between the two antiferro-
where we assume J < 0 (antiferromagne- magnetic structures join in a bicritical
tism) and for D > 0 the easy axis is the z- point. A well known example of such a be-
axis. For small H|| , we have a uniaxial anti- havior is GdAlO3 with Tb = 3.1242 K (Roh-
ferromagnetic structure and the order pa- rer, 1975). Another example, in our opin-
rameter is the z-component of the stag- ion, is the Fe – Al system where the ferro-
gered magnetization Msz . For stronger paramagnetic critical line joins the second-
fields H|| , however, we have a transition order A2 – B2 transition (Fig. 4-11). In the
to a spin-flop structure (two-component metallurgical literature, the magnetic tran-
Figure 4-11. The Fe – Al phase diagram, as obtained from a mean-field calculation by Semenovskaya (1974)
(left), and experimentally by Swann et al. (1972) (right). The ferro-paramagnetic transition is shown by the
dash-dotted line. The crystallographically disordered (A2) phase is denoted as a n in the nonmagnetic and a m in
the ferromagnetic state. The ordered FeAl phase having the B2 structure (Fig. 4-3) is denoted as a 2 , and the or-
dered Fe3Al phase having the D03 structure (Fig. 4-3) is denoted as a 1n or a 1m , depending on whether it is non-
magnetic or ferromagnetically ordered, respectively. Note that first-order transitions in this phase diagram are
associated with two-phase regions, since the abscissa variable is the density of an extensive variable, unlike the
ordinate variable in Fig. 4-9. From Semenovskaya (1974).
sition of Fe – Al alloys is often disregarded some characteristic time t which diverges

and then the multicritical point where the as T Æ Tc :
A2 – B2 transitions change from second to
t ~ x z ~ |1 – T /Tc | – n z (4-94)
first order is interpreted as a tricritical
point. Bicritical phenomena for structural Many of the concepts developed for static
phase transitions are also known, such as critical phenomena (scaling laws, univer-
SrTiO3 under stress in the [001] direction sality, etc.) can be carried over to dynamic
(Müller and Berlinger, 1975). critical phenomena: “dynamic scaling”
The structure of the phase diagram near (Halperin and Hohenberg, 1967; Ferrell et
bicritical points has been analyzed in detail al., 1967; Hohenberg and Halperin, 1977),
by renormalization group methods (Fisher “dynamic universality classes” (Halperin
and Nelson, 1974; Fisher, 1975 a, b; Muka- et al., 1974, 1976; Hohenberg and Halpe-
mel et al., 1976 a, b), experiment (Rohrer rin, 1977), etc. However, the situation is
and Gerber, 1977) and Monte Carlo simu- much more complicated because the static
lations (Landau and Binder, 1978). Apart universality classes are split up in the dy-
from the fact that the exponents introduced namic case, depending on the nature of the
so far take different values as T Æ Tb , there conservation laws, mode coupling terms in
is one additional exponent, the “crossover the basic dynamic equations, etc. For ex-
exponent” F, describing the singular ap- ample, we have emphasized that aniso-
proach of phase-transition lines towards tropic magnets such as RbMnF3 , ordering
the multicritical point. This behavior, and alloys such as CuZn, unmixing solid mix-
also other more complicated phenomena tures such as Al – Zn, unmixing fluid mix-
(tetracritical points, tricritical Lifshitz tures such as lutidine – water, and the gas –
points, etc.) are beyond the scope of this fluid critical point all belong to the same
chapter. For an excellent introduction to Ising-like static universality class, but each
the field of multicritical phenomena we re- of these examples belongs to a different dy-
fer to Gebhardt and Krey (1979) and for a namic universality class! Thus, in the an-
comprehensive review to Pynn and Skjel- isotropic antiferromagnet, no conservation
torp (1983). law needs to be considered, whereas the
conservation of concentration matters for
all mixtures (where it means the order pa-
4.2.4 Dynamics of Fluctuations rameter is conserved ) and for ordering al-
at Phase Transitions loys (where the order parameter is not con-
served but coupled to a conserved “nonor-
Second-order phase transitions also dering density”). Whereas in solid mix-
show up in the “critical slowing down” of tures the local concentration relaxes simply
the critical fluctuations. In structural phase by diffusion, in fluid mixtures hydrody-
transitions, we speak about “soft phonon namic flow effects matter, and also play a
modes”; in isotropic magnets magnon role at the liquid – gas critical point. In the
modes soften as T approaches Tc from be- latter instance, conservation of energy
low; near the critical point of mixtures the needs to be considered; it does not play a
interdiffusion is slowed down; etc. This role for phase transitions in solid mixtures,
critical behavior of the dynamics of fluctu- of course, where the phonons act as a “heat
ations is again characterized by a dynamic bath” to the considered configurational de-
critical exponent z; we expect there to be grees of freedom.
We do not give a detailed account of crit- to establish its precise value (Wansleben
ical dynamics here, but discuss only the and Landau, 1987).
simplest phenomenological approach (van It is important to realize that not all fluc-
Hove, 1954) for non-conserved order pa- tuations slow down as a critical point is ap-
rameters. We consider a deviation of the or- proached, but only those associated with
der parameter DF (x, t) from equilibrium at long-wavelength order parameter varia-
space x and time t, and ask how this deviations. This clearly expressed in terms of the
tion decays back to equilibrium. The stan- dynamic scaling relationship of the charac-
dard assumption of irreversible thermody- teristic frequency:
namics is a relaxation assumption; we put
w (q) = q z w̃ ( ᒗ), ᒗ = qx (4-99)
∂ ∂Ᏺ
DF ( x, t ) = − G0 (4-95) where w̃ (ᒗ) is a scaling function with the
∂t ∂DF ( x ) properties w̃ (•) = const. and w̃ (ᒗ O 1) ~ ᒗ– z ,
where G0 is a phenomenological rate factor. and, hence, is consistent with Eq. (4-94).
Using Eq. (4-10) and linearizing F (x, t) = Most experimental evidence on dynamic
F0 + DF (x, t) around F0 yields critical phenomena comes from methods
such as inelastic scattering of neutrons or
∂
∂t
{
DF ( x, t ) = − G0 ( r + 3 u F02 ) DF ( x, t ) light, NMR or ESR spectroscopy, and
ultrasonic attenuation. Although the over-
R2 2 ⎫ all agreement between theory and experi-
− ∇ [ DF ( x, t )]⎬ (4-96) ment is satisfactory, only a small fraction
d ⎭
of the various theoretical predictions have
Assuming that DF (x, t) is produced by been thoroughly tested so far. For reviews,
a field H (x) = Hq exp (i q · x) which is see Hohenberg and Halperin (1977) and
switched off at time t = 0, we obtain (see Enz (1979).
also Sec. 4.2.2) Finally, we also draw attention to nonlin-
(4-97) ear critical relaxation. Consider, for exam-
DFq (t ) ple, an ordering alloy which is held well
= exp [ − w ( q ) t ], w ( q ) = G0 / c ( q )
DFq ( 0 ) below Tc in the ordered phase and assume
that it is suddenly heated to T = Tc . The or-
where Eq. (4-34) has been used. Since c (q) der parameter is then expected to decay to-
diverges for q = 0, T Æ Tc , the characteris- wards zero with time in a singular fashion,
tic frequency w (q = 0) ∫ t –1 vanishes as F (t) ~ t – b /n z (Fisher and Racz, 1976). Sim-
w (q = 0) ~ c T–1 ~ x –g /n = x – (2 – h ), and by ilarly, if the alloy is quenched from a tem-
comparison with Eq. (4-94) the result of perature T > Tc to T = Tc , the superstructure
the “conventional theory” is obtained: peak appears and grows in a singular fash-
z=2–h (4-98) ion with the time t after the quench,
S (q, t) ~ t g /zn (Sadiq and Binder, 1984). Re-
Although Eq. (4-98) suggests that there is a lated predictions also exist for the nonlin-
scaling relationship linking the dynamic to ear relaxation of other quantities (ordering
the static exponents, this is not true in gen- energy, electrical resistivity) (Binder and
eral if effects due to critical fluctuations are Stauffer, 1976 a; Sadiq and Binder, 1984).
taken into account. In fact, for noncon- A very important topic is the dynamics
served systems, z slightly exceeds 2 – h and of first-order phase transitions, which will
extensive Monte Carlo studies were needed not be discussed here because it is dis-
cussed elsewhere in this book (see Chap- rameter space of the corresponding multi-
ters by Binder and Fratzl (2001) and by component system show up like a rounding
Wagner et al. (2001)). of the first-order transition.
In solids, however, diffusion of impurity
atoms is often negligibly small and such a
4.2.5 Effects of Surfaces and of Quenched
disorder due to frozen-in, immobile impur-
Disorder on Phase Transitions:
ities is called “quenched disorder” (Brout,
A Brief Overview
1959). Other examples of quenched disor-
It a system contains some impurity at- der in solids are due to vacancies, in addi-
oms which are mobile, we refer to “an- tion to extended defects such as disloca-
nealed disorder” (Brout, 1959). From the tions, grain boundaries, and external sur-
point of view of statistical thermodynam- faces. Quenched disorder has a drastic effect
ics, such mobile impurities act like addi- on phase transitions, as will be seen below.
tional components constituting the system Since the disorder in a system is usually
under consideration. For a second-order assumed to be random (irregular arrange-
transition, small fractions of such addi- ment of frozen-in impurities, for example),
tional components have rather minor ef- we wish to take an average of this random
fects: the transition point may be slightly disorder. We denote this averaging over the
shifted relative to that of the pure material, disorder variables, which we formally de-
and as the concentration of these impurities note here as the set {xa }, by the symbol
is normally strictly conserved, in principle […]av in order to distinguish it from the
the so-called “Fisher renormalization” thermal averaging ·…ÒT implied by statisti-
(Fisher, 1968) of critical exponents is ex- cal mechanics. Thus the average free en-
pected, as discussed in Sec. 4.2.2. For ergy which needs to be calculated in the
small impurity contents, the region around presence of quenched disorder is
the transition point where this happens is
extremely narrow, and hence this effect is F = − kB T [ln Z {xa}]av (4-100 )
not important.
The effect of such impurities on first-or-
[ ( {F }
= − kB T ln tr exp( − Ᏼ {xa ,Fi }/ kB T )
i
) ]av
der transitions is usually more important, whereas in the case of annealed disorder it
e.g., for a pure one-component system, the would be the partition function rather than
melting transition occurs at one well-de- the free energy which is averaged over the
fined melting temperature, whereas in the disorder:
presence of mobile impurities this transi-
Fann = − kB T ln[ Z {xa}]av (4-101)
tion splits up into two points, correspond-
ing to the “solidus line” and “liquidus line”
in the phase diagram of a two-component
( {F } i
)
= − kB T ln tr [exp( − Ᏼ {xa ,Fi }/ kB T )]av
system. This splitting for small impurity = − kB T ln ( tr exp( − Ᏼ {xa ,Fi }/ kB T ))

{F , x }i a
contents is simply proportional to the con-
centration of the impurities of a given type. Eq. (4-101) again shows that annealed dis-
In practice, where a material may contain order just means that we enlarge the space
various impurities of different chemical na- of microscopic variables which need to be
tures and different concentrations which included in the statistical mechanics treat-
often are not known, these splittings of the ment, whereas the nature of quenched aver-
transition line in a high-dimensional pa- aging (Eq. (4-100)) is different.
Experiments can be carried out with just impurity concentration, respectively) is

one probe crystal and no averaging over smaller. Since no ideal long-range order is
many samples as is implied by Eq. (4-100) established at the critical point, this corre-
needs to be performed. This is because we sponds to a rounding of the phase transi-
can think about the parts of a macroscopic tion.
sample as subsystems over which such an In the case n = 1 and d = 3 dimensions,
average as written in Eq. (4-100) is actu- very weak random fields do not destroy
ally performed. This “self-averaging” uniform long-range order (Imbrie, 1984;
property of macroscopically large systems Nattermann, 1998), although long-range
means that the average […]av over the set order is destroyed if HR exceeds a certain
{xa } could also be omitted: it need only be threshold value HRc (T ) which vanishes as
kept for mathematical convenience. T Æ Tc . But the random field drastically
We first discuss the physical effects of changes the nature of both static and dy-
randomly quenched impurities (point-like namic phenomena (Villain, 1985; Grin-
defects!) on phase transitions, assuming stein, 1985). In d = 2 dimensions, arbitrary
the concentrations of these defects to be weak random fields destroy even Ising-
very small. Various cases need to be distin- like (n = 1) order, producing domains of
guished, depending on the nature of the size x D with ln x D ~ (J/HR )2 (see Binder,
local coupling between the defect and the 1983 a).
local order parameter. This coupling may We now turn briefly to the physical real-
be linear (“random field”), bilinear (“ran- ization of “random fields”. For symmetry
dom bond”) or quadratic (“random aniso- reasons, no such defects are expected for
tropy”). As an example, let us consider ferromagnets, but they may be induced in-
magnetic phase transitions described by the directly by a uniform field acting on an
n-vector model of magnetism, with an ex- antiferromagnet (Fishman and Aharony,
change interaction depending on the dis- 1979; Belanger, 1998). Since antiferro-
tance between lattice points ai and aj : magnets in a field can be “translated” into
models for order – disorder phenomena in
Ᏼ = − ∑ J ( ai − a j ) Si ⋅ S j − ∑ Hi Siz (4-102) alloys (the magnetization of the antiferro-
i≠ j i
magnet “translates” into the concentration
In Eq. (4-102) we have assumed that im- of the alloy, see Sec. 4.3.1) or (in d = 2 di-
purities produce a random field Hi which mensions) in adsorbed layers (the magnet-
has the properties ization then “translates” into the coverage
[Hi ]av = 0 , [Hi2 ]av = HR2 (4-103) of the layer), random-field effects are im-
portant for ordering alloys or order – disor-
and Si is an n-component unit vector in the der transitions in monolayers, also. A nice
direction of the local order parameter. Now experimental example was provided by
we can show that for n ≥ 2 arbitrary small Wiechert and Arlt (1993) who showed that
amplitudes of the random field HR (which the transition of CO monolayers on graph-
physically is equivalent to arbitrary small ite near 5 K to a ferrielectric phase is
impurity concentration) can destroy uni- rounded by small dilution with N2 mole-
form long-range order (Imry and Ma, cules, in accord with the theory of random-
1975): the system is broken into an irregu- field effects. Furthermore, random fields
lar arrangement of domains. The mean act on many structural phase transitions
size of the domains is larger as HR (or the owing to impurity effects: ions at low-sym-
metry, off-center positions may produce for a review). Of course, the impurities will
both a random electric field and a random always produce some shift of the critical
strain field. Also, symmetry generally per- temperature, which decreases as p in Eq.
mits random fields to act on tensorial order (4-106) decreases. When p becomes
parameters in diluted molecular crystals smaller than a critical threshold value pc ,
(Harris and Meyer, 1985). For diluted mo- only finite clusters of spins are still mutu-
lecular crystals (such as N2 diluted with Ar, ally connected by nonzero exchange bonds
or KCN diluted with KBr, etc.) it is still un- J > 0, and long-range order is no longer
clear whether the resulting “orientational possible. The phase transition at T = 0 pro-
glass” phase is due to these random fields duced by variation of p in Eq. (4-106) can
or to random bonds (for reviews, see Loidl hence be interpreted purely geometrically
(1989); Höchli et al. (1990); Binder and in terms of the connectivity of finite clus-
Reger (1992); Binder (1998)). ters or an infinite “percolating” network of
The “random bond” Hamiltonian differs spins (Stauffer and Aharony, 1992). This
from Eq. (4-102) by the introduction of dis- transition is again described by a com-
order into the exchange terms rather than pletely different set of exponents.
–
the “Zeeman energy”-type term, For the case where | J | O D J in Eq.
(4-105), a new type of ordering occurs,
Ᏼ = − ∑ Jij Si ⋅ S j − H ∑ Siz (4-104) which is not possible in systems without
i≠ j i
quenched disorder; a transition occurs to a
Whereas in Eq. (4-102) we have assumed state without conventional ferro- or antifer-
the pairwise energy to be translationally in- romagnetic long-range order but to a “spin
variant (the exchange energy J depends glass phase” where the spins are frozen-in
only on the distance ai – aj between the in a random direction (Binder and Young,
spins but not on their lattice vectors ai and 1986; Young, 1998). The order parameter
aj separately), we now assume Ji j to be a of the Ising spin glass was introduced by
random variable, e.g., distributed accord- Edwards and Anderson (1975) as
ing to a Gaussian distribution
– qEA = [·Si Ò2T ] av (4-107)
P (Ji j ) µ exp [– (Ji j – J )2/2 (DJ)2 ] (4-105)
The nature of the phase transition in spin
or according to a two-delta function distri- glasses and the properties of the ordered
bution phase have been the subject of intense re-
P (Ji j ) = p d (Ji j – J) + (1 – p) d (Ji j ) (4-106) search (Binder and Young, 1986). This
great interest in spin glasses can be under-
–
If J o DJ (Eq. (4-105)) or if 1 – p O 1 (Eq. stood because many different systems show
(4-106)), the ferromagnetic order occurring spin glass behavior: transition metals with
–
for Eq. (4-104) if J > 0 or J > 0 is only a small content x of magnetic impurities
weakly disturbed, both for d = 3 and for such as Au1– x Fex and Cu1– x Mnx , diluted
d = 2 dimensions. Following an argument insulating magnets such as Eux Sr1– x S, var-
presented by Harris (1974), we can see that ious amorphous alloys, and also mixed
the critical behavior of the system remains dielectric materials such as mixtures of
unaltered in the presence of such impurities RbH 2PO4 and NH 4 H 2PO4 , where the spin
provided that the specific heat exponent represents an electric rather than magnetic
a < 0, whereas a new type of critical behav- dipole moment. A related random ordering
ior occurs for a ≥ 0 (see Grinstein (1985) of quadrupole moments rather than dipole
moments if found in “quadrupolar glasses” glasses) the critical dynamics again have a
(also called “orientational glasses”) such very different character, and dynamics
as K(CN)1– x Brx and (N2)x Ar1– x , (Loidl, characteristic of thermally activated pro-
1989; Höchli et al., 1990; Binder and cesses with a broad spectrum of relaxation
Reger, 1992; Binder, 1998). However, it times are often observed.
is still debated (Franz and Parisi, 1998) It is also interesting to discuss the effect
whether or not concepts appropriate for the of quenched impurities on first-order tran-
description of spin glasses are also useful sitions (Imry and Wortis, 1979). It is found
for the structural glass transition of under- that typically “precursor phenomena” near
cooled liquids (Jäckle, 1986). the first-order transition are induced, and
Finally we mention systems with ran- the latent heat associated with the transi-
dom anisotropic axes which can be mod- tion in the pure systems can be signifi-
eled by the following Hamiltonian (Harris cantly reduced. It is also possible that such
et al., 1973): impurity effects may completely remove
(4-108) the latent heat discontinuity and lead to a
Ᏼ = − ∑ J ( ai − a j ) Si ⋅ S j − ∑ ( ei ⋅ S j )2 rounding of the phase transition.
i≠ j i
Such rounding effects on phase transi-
where the ei are vectors whose components tions also occur when extended defects
are independent random variables with a such as grain boundaries and surfaces are
Gaussian distribution. This model is also considered. These disrupt long-range order
believed to exhibit destruction of long- because the system is then approximately
range order due to break-up in domains homogeneous and ideal (i.e., defect-free)
similar to the random field systems. It is only over a finite region in space. While
also suggested that spin glass phases may the description of the rounding and shifting
occur in these systems. Again Eq. (4-108) of phase transitions due to finite size ef-
is expected to be relevant, not only for dis- fects has been elaborated theoretically in
ordered magnetic materials but also for detail (Privman, 1990), only a few cases
dielectrics where the spin represents elec- exist where the theory has been tested ex-
tric dipole moments, or displacement vec- perimentally, such as the normal fluid –
tors of atoms at structural phase transitions, superfluid transitions of He4 confined to
etc. pores, or the melting transition of oxygen
Very small fractions of quenched impur- monolayers adsorbed on grafoil where the
ities which do not yet have an appreciable substrate is homogeneous over linear di-
effect on the static critical properties of a mension L of the order of 100 Å (see Marx,
second-order phase transition can already 1989). While at critical points, a rounding
affect the critical dynamics drastically. An and shifting of the transition normally set
example of this behavior is the occurrence in when the linear dimension L and the cor-
of impurity-induced “central peaks” for relation length x are comparable (Fisher,
structural phase transitions in the scattering 1971; Binder, 1987b), at first-order transi-
function S (q, w ) at frequency w = 0 in additions the temperature region DT over
tion to the (damped) soft phonon peaks which the transition is rounded and shifted
(Halperin and Varma, 1976). In the above is inversely proportional to the volume of
systems where the impurities disrupt the the system (Imry, 1980; Challa et al., 1986).
conventional ordering more drastically An important effect of extended defects
(such as random field systems and spin such as grain boundaries and surfaces is
4.3 Computational Methods Dealing with Statistical Mechanics 283
that they often induce precursor effects to Transitions occur where the basic degree of
phase transitions, e.g., the “wetting transi- freedom is the (thermally activated) diffu-
tion” (Dietrich, 1988; Sullivan and Telo da sion process of atoms between various lat-
Gama, 1985) where a fluid layer condens- tice sites. This happens for unmixing alloys
ing at a surface is a precursor phenomenon such as Al – Zn or for ordering alloys such
to gas – fluid condensation in the three- as b-CuZn or the Cu –Au system. The mod-
dimensional bulk volume. Similarly “sur- eling of such systems will be discussed in
face melting” and “grain-boundary melt- Sec. 4.3.1. Many structural transitions are
ing” phenomena can be interpreted as the of a very different nature: we encounter pe-
intrusion of a precursor fluid layer at a sur- riodic lattice distortions where atomic dis-
face (or grain boundary) of a crystal (Di placements are comparable to those of lat-
Tolla et al., 1996). Similarly, at surfaces of tice vibrations. Short wavelength distor-
ordering alloys such as Cu3Au, the effect tions may give rise to “antiferrodistortive”
of the “missing neighbors” may destabilize and “antiferroelectric” ordering, as exem-
the ordering to the extent that “surface-in- plified by the perovskites SrTiO3 , PbZrO3 ,
duced disordering” occurs when the system etc. Long wavelength distortions corre-
approaches the transition temperature of sponding to acoustic phonons give rise to
the bulk (Lipowsky, 1984). Finally, we “ferroelastic” ordering. Sec. 4.3.2 will deal
draw attention to the fact that the local crit- with the mean field treatment of such phase
ical behavior at surfaces differs signifi- transitions where the order parameter is a
cantly from the critical behavior in the bulk phonon normal coordinate, and Sec. 4.3.3
(Binder, 1983 b). Such effects are outside is devoted to numerical methods going be-
the scope of this chapter. yond mean-field theory.
4.3 Computational Methods 4.3.1 Models for Order – Disorder

Dealing with the Statistical Phenomena in Alloys
Mechanics of Phase Transitions Since the diffusive motion of atoms in
and Phase Diagrams substitutional alloys (Fig. 4-12) is so much
slower than other degrees of freedom (such
While the phenomenological theories in as lattice vibrations), we may describe the
Sec. 4.2 yield a qualitative insight into configurational statistics of a substitutional
phase transitions, the quantitative descrip- binary alloy, to which we restrict attention
tion of real materials needs a more detailed for simplicity, by local occupation vari-
analysis. In this section, we complement ables {ci }. If lattice site i is taken by an
the phenomenological theory by a more atom of species B, ci = 1, if it is taken by an
microscopic approach. In the first step, the atom of species A, ci = 0.
essential degrees of freedom for a particu- Neglecting the coupling between these
lar transition are identified and an appro- variables and other degrees of freedom, the
priate model is constructed. In the second Hamiltonian is then:
step, the statistical mechanics of the model
are treated by suitable approximate or nu- Ᏼ = Ᏼ 0 + ∑ [ci c j v BB ( xi − x j )
merical methods. i≠ j
Fig. 4-12 illustrates the modeling of or- + 2 ci (1 − c j ) v AB ( xi − x j ) ( 4-109)
der – disorder phase transitions in solids. + (1 − ci ) (1 − c j ) v AA ( xi − x j )] + …
Figure 4-12. Degree of free-

dom essential for the descrip-
tion of various order – disor-
der transitions in solids. For
further explanations, see text.
where vAA , vAB , and vBB are interactions where Dm is the chemical potential differ-
between pairs of AA, AB, and BB atoms. ence between the two species.
In fact, terms involving three- and four- The same mapping applies for the lat-
body interactions may also occur, but are tice-gas model of fluids, which at the same
not considered here. Also, effects due to time can be considered as a model of ad-
vacancies may easily be included but are sorbed layers on crystalline surface sub-
neglected here. strates (in d = 2 dimensions) (see Binder
As is well known, Eq. (4-109) can be re- and Landau (1989) for a review) or as a
duced to the Ising model, Eq. (4-56), by the model of interstitial alloys, such as hydro-
transformation Si = 1 – 2 ci = ± 1, apart from gen or light atoms such as C, N and O in
a constant term which is of no interest to us metals (Alefeld, 1969; Wagner and Horner,
here. The “exchange interaction” Jij 1974; Alefeld and Völkl, 1978). If we
between spins i and j and “magnetic field” interpret B in Eqs. (4-109) – (4-111) as an
H in Eq. (4-56) are related to the interac- occupied site and A as a vacant site, then
tion parameters of Eq. (4-109) by usually vAB = vAA= 0, such that
(4-112)
J (xi – xj) ∫ Jij = [2 vAB (xi – xj ) (4-110) Ᏼ = Ᏼ 0 + ∑ ci c j v ( xi − x j ) + e ∑ ci
– vAA (xi – xj ) – vBB (xi – xj )]/4 i≠ j i
e being the binding energy which a particle

1⎛ ⎞
H= ⎜ ∑ [v AA ( xi − x j ) − v BB ( xi − x j )]⎟ feels at lattice site i (for adsorbates at sur-
2 ⎝ j (≠ i) ⎠ faces, this is a binding to the substrate, and
− Dm ( 4-111) a similar enthalpy term is expected when
an interstitial is dissolved in a metal, transformation {Si , H} Æ {– Si , – H} . This

whereas for modeling of an ordinary gas – implies a particle–hole symmetry of the
liquid transition we would put e = 0). The phase diagram of the lattice gas model, or a
relationships analogous to Eqs. (4-110) and symmetry against interchange of A and B
(4-111) are then: in the phase diagram of the binary alloy
(4-113)
model. Even in very ideal cases, where
[
H = − (e + m )/ 2 + ∑ v ( xi − x j )/ 4
j (≠ i)
] both partners A and B of a binary alloy
crystallize in the same structure and have
Jij = – v (xi – xj)/4 (4-114) similar lattice spacings, the phase diagram
is in reality not symmetric around the line
where m is the chemical potential and ·ci Ò = 1/2 (see the Cu – Au phase diagram
v (xi – xj) the pairwise interaction of parti- reproduced in Fig. 4-13 as an experimental
cles at lattice sites xi and xj. example). There are various conceivable
Of course, it is straigthforward to gener- reasons for this asymmetry of real phase
alize this approach to binary (A – B) alloys diagrams: (i) the assumption of strictly
including as a third option that a lattice site pairwise potentials (Eq. (4-109)) fails, and
is vacant (V) or to ternary alloys; then terms such as ci cj ck , ci cj ck cl need to be in-
more complicated models result, e.g., the cluded in the effective Hamiltonian; and
Potts (1952) model, Eq. (4-20), or the (ii) the potentials vAA , vAB , and vBB in Eq.
model of Blume et al. (1971). Another gen- (4-109) are not strictly constant but depend
eralization occurs if one species (B) of a
binary alloy is magnetic, e.g., Fe in Fe –Al
alloys (see Dünweg and Binder (1987)).
In the latter case, the Hamiltonian is (the
true magnetic spin variable is now denoted
as s i )
Ᏼ = Ᏼ0 ( 4-115)
+ ∑ {ci c j [v BB ( xi − x j ) − Jm ( xi − x j ) s i ⋅ s j ]
i≠ j
+ 2 ci (1 − c j ) v AB ( xi − x j )
+ (1 − ci ) (1 − c j ) v AA ( xi − x j )} + …
where Jm (xi – xj) denotes the strength of Figure 4-13. Partial phase diagram of copper – gold
the magnetic interaction, which we have alloys in the temperature – concentration plane indi-
assumed to be of the same type as used in cating the existence regions of the three ordered
the Heisenberg model (Eq. (4-102)) with phases Cu3Au, CuAu, and CuAu3 (cf. Fig. 4-8 a and
n = 3 and Hi = 0. c). These phases are separated from each other (and
from the disordered phase occurring at higher tem-
We will not go further into the classifica- peratures) by two-phase coexistence regions. The
tion of the various models here, but note boundaries of these regions are indicated by full and
one general property of both models in broken lines. Region II is a long-period modulated
Eqs. (4-109) and (4-112) which becomes version of the simple CuAu structure occurring in re-
evident when mapping the Ising Hamilton- gion I. Note that for strictly pairwise constant inter-
actions (of arbitrary range!) in a model such as Eq.
ian Eq. (4-56): for H = 0, this Hamiltonian (4-109) the phase diagram should have mirror sym-
is invariant against a change of sign of all metry around the line cAu = 50%. From Hansen
spins. For H ≠ 0, it is invariant against the (1958).
on the average concentration ·cÒ of the al-

loy, since the effective lattice parameter a
and the Fermi wavenumber kF and other
physical characteristics entering the effec-
tive potential of the alloy also depend
on concentration ·cÒ. In fact, both reasons
probably contribute to some extent in real
alloys – the first-principles derivation of
model Hamiltonians such as Eq. (4-109)
from the electronic structure theory for sol-
ids is a challenging problem (Bieber and
Gautier 1984 a, b; Zunger, 1994). Since the
effective interaction typically is oscillatory
in sign (Fig. 4-14) according to the Friedel
form
v (| x |) = A cos (2 | kF || x | + j)/| x | 3 (4-116)
where A and j are constants, small changes
in | kF | and/or the distances lead to rela-
tively large changes in v (| x |) for distant
neighbors. Whereas the interaction Eq.
(4-116) results from the scattering of con-
duction electrons at concentration inhomo-
geneities, in other cases such as H in metals
the effective interactions are due to the
elastic distortions produced by these inter-
stitials in their environment. The resulting
elastic interaction is also of long range Figure 4-14. (a) Interaction potential V lmn = – 2 J (x)
(Wagner and Horner, 1974). (note that x = (l, m, n) –a2 , a being the lattice spacing) vs.
-----------
Since the effective interactions that distance | x | (measured as ÷l 2 + m2 + n2 ) for NiCr0.11
should be used in Eqs. (4-109) to (4-115) at T = 560 °C, as deduced from (b) short-range order
often cannot be predicted theoretically in a parameters a (x) measured by diffuse neutron scatter-
ing. The circles show the results of the high-temper-
reliable way, it may be desirable to extract ature approximation, Eq. (4-121), and crosses the re-
them from suitable experimental data. For sult of the inverse Monte Carlo method. Full and
metallic alloys, such suitable experimental broken curves represent a potential function of the
data are the Cowley (1950) short-range or- Friedel form, Eq. (4-116), with different choices of
der parameters a (x) ∫ a (xi – xj) which de- the Fermi wavevector kF for ·100Ò and ·110Ò direc-
tions, amplitude A and phase F being treated as fit-
scribe the normalized concentration corre- ting parameters. From Schweika and Hauboldt (1986).
lation function, c ∫ ·ciÒ = (1 – M)/2, M ∫ ·SiÒ
〈 ci c j 〉 − c 2 〈 Si S j 〉 − M 2 resentation (Eq. (4-117) for a ferromagnet
a ( xi − x j ) ≡ =
c (1 − c ) 1 − M2 is equivalent to the phenomenological Eq.
(4-117) (4-43)). In the disordered phase of an Ising
which correspond to the normalized spin spin system, it is straightforward to obtain
correlation functions in the Ising spin rep- the wavevector-dependent response func-
tion c (q) describing the response to a (cf. Sec. 4.2.2); a more accurate procedure
wavevector-dependent field (cf. Eqs. (4-32), to deduce J (x) from experimental data on
(4-34), and (4-40)), if we apply the molec- a (x) is the “inverse Monte Carlo method”
ular field approximation (see the next sub- (Gerold and Kern, 1986, 1987; Schweika,
section). The result is (Brout, 1965): 1989), but this method also relies on the as-
(4-118)
sumption that an Ising model description as
1− M 2
S ( q ) = kB T c ( q ) = written in Eq. (4-109) is appropriate. On
1 − [ J ( q )/ kB T ] (1 − M 2 )
˜
the level of the molecular field approxima-
where J̃ (q) is the Fourier transform of the tion (for alloys this is usually referred to as
“exchange interaction” J (x), the Bragg – Williams (1934) approxima-
(4-119) tion) or the Bethe (1935) approximation, it
˜J ( q ) = ∑ J ( xi − x j ) exp [ i q ⋅ ( xi − x j )] is possible to avoid models of the type of
j (≠ i) Eq. (4-109) and include the configurational
degrees of freedom in an electronic struc-
In Eq. (4-118), we have once again invoked
ture calculation (Kittler and Falicov, 1978,
that there is a “fluctuation relation”
1979). Although such an approach sounds
(= static limit of the so-called “fluctuation-
very attractive in principle, the results are
dissipation theorem”) relating c (q) to the
not so encouraging in practice, as shown in
structure factor S (q) which is just the
Fig. 4-15. The results of this method for
Fourier transform of the correlations ap-
Cu3Au are compared with Monte Carlo data
pearing in Eq. (4-117):
(4-120) of a nearest neighbor Ising model (Binder,
S (q) = ∑ exp [ i q ⋅ ( ri − rj )] ( 〈 Si S j 〉 − M 2 ) 1980), with the cluster-variation treatment
j (≠ i) of the same model (Golosov et al., 1973)
and with experimental data (Keating and
Combining now Eqs. (4-118) and (4-120)
Warren, 1951; Moss, 1964; Orr, 1960).
we see that the reciprocal of this diffuse
The conclusion of this subsection is that
scattering intensity S (q) in q-space is sim-
the development of microscopic models for
ply related to the Fourier transformation of
the description of order – disorder phenom-
the interactions as
ena in alloys is still an active area of re-
4 c (1 − c ) 4 J˜ ( q ) search, and is a complicated matter, be-
=1− c (1 − c ) (4-121) cause the validity of the models can only be
S (q) kB T
judged by comparing results drawn from
For the model defined in Eqs. (4-109) to the models with experimental data. How-
(4-111), this expression even is exact in ever, these results also depend on the ap-
an expansion of 1/S (q) in a power series in proximation involved in the statistical-me-
1/T to leading order in 1/T, and is one of chanical treatment of the models (e.g., the
the standard tools for inferring information full and broken curves in Fig. 4-15 refer to
on interactions in alloys from diffuse scat- the same nearest-neighbor Ising model,
tering data (Clapp and Moss, 1966, 1968; whereas the dash-dotted curve refers to a
Moss and Clapp, 1968; Krivoglaz, 1969; different model). In the next subsections,
Schweika, 1994). An example is shown we consider various sophistications of the
in Fig. 4-14. Close to the order – disorder statistical mechanics as applied to various
phase transition, we expect corrections to models for phase-transition phenomena.
the molecular field expression Eq. (4-121) More details about all these problems can
due to the effects of statistical fluctuations be found in the Chapter by Inden (2001).
of lattice gas models is discussed), for mag-

netic transitions (De Jongh and Miedema,
1974), etc. Usually only the source of ex-
perimental data used to extract the interac-
tion parameters is different: e.g., for ferro-
magnets such as EuS, rather than using dif-
fuse magnetic neutron scattering in analogy
with Eq. (4-118) it is more convenient to
extract J̃ (q) from the measurements of spin
wave dispersion curves found from inelas-
tic neutron scattering (Bohn et al., 1980).
4.3.2 Molecular Field Theory

and its Generalization
(Cluster Variation Method, etc.)
The molecular field approximation
(MFA) is the simplest theory for the de-
scription of phase transition in materials;
despite its shortcomings, it still finds wide-
spread application and has been described
in great detail in various textbooks (Brout,
1965; Smart, 1966). Therefore we do not
treat the MFA in full detail here, but rather
Figure 4-15. (a) Order parameters for the A3B struc- indicate only the spirit of the approach.
ture (Fig. 4-8 c) on the fcc lattice: Long-range order
parameter Y (LRO) and short-range order parameter
We start with the Ising ferromagnet, Eq.
– a1 for the nearest-neighbor distance (SRO) vs. tem- (4-56). The exact Helmholtz energy can be
perature, according to the Monte Carlo method found formally from the minimum of the
(Binder, 1980), the cluster variation (CV) method in functional (Morita, 1972)
the tetrahedron approximation (Golosov et al., 1973),
and the Kittler – Falicov (1978, 1979) theory. Data for Ᏺ = ∑ Ᏼ Ising ({Si }) P ({Si })
Cu3Au after Keating and Warren (1951) (LRO) and {Si = ± 1}
after Moss (1964) (SRO). (b) Ordering energy DU
(normalized to zero at Tc in the disordered state) vs.
+ kB T ∑ P ({Si }) ln ( P{Si }) (4-122)
{Si = ± 1}
temperature for the fcc A3B alloy (top) and the fcc
AB alloy (bottom). Theoretical curves are from the where the sums extend over all configura-
same sources as in (a); experimental data were taken tions of the spins in the system and P ({Si})
from Orr (1960) and Orr et al. (1960). is the probability that a configuration {Si}
occurs. Eq. (4-122) thus corresponds to the
thermodynamic relation F = U – TS where
We have concentrated in this subsection the entropy S is written in its statistical
on alloys, but a similar discussion could interpretation. Minimizing Eq. (4-122) for-
have also been presented for order – disor- mally with respect to P yields the canonical
der phenomena in adsorbed monolayers distribution
(see for example, Binder and Landau (1981,
Peq ({Si}) ~ exp [– Ᏼ Ising ({Si})/kB T]
1989), where the modeling of systems such
as H adsorbed on Pd (100) surfaces in terms as desired.
The MFA can now be defined by factor- These probabilities can be expressed in
izing the probability P ({Si}) of a spin con- terms of the multi-spin correlation func-
figuration of the whole lattice into a prod- tions gnc (i) ∫ ·Si Sj 1 … Sjn Ò, where the set of
uct of single-site probabilities pi which vectors xj 1 – xi , …, xjn – xi defines the n-
can take two values: p+ = (1 + M )/2 is the point cluster of type c located at lattice site
probability that the spin at site i is up i. The Helmholtz energy functional to be
and p– = (1 – M )/2 is the probability that it minimized in this cluster variation method
is down, p+ – p– = M is the magnetization. (Kikuchi, 1951; Sanchez and De Fontaine,
Now the expression Jij Si Sj pi pj (cf. Eq. 1980, 1982; Finel, 1994) is a more compli-
(4-56)) summed over the possible values cated approximation of Eq. (4-122) than
p+ and p– simply yields Jij M 2, and hence Eq. (4-123). If the largest cluster consid-
Eq. (4-122) reduces to, using Eq. (4-119) ered exceeds the interaction range, the en-
ergy term in F = U – TS is treated exactly;
1 MFA 1 ˜
Ᏺ = J (q = 0) M 2 − H M (4-123) unlike Eq. (4-123), the entropy is approxi-
N 2 mated. We find
⎡1 + M ⎛ 1 + M ⎞
+ kB T ⎢ ln
⎣ 2 ⎝ 2 ⎠ 1
Ᏺ= ∑ ∑ Jij g2, rj (i ) + kB T ∑ ∑ g nc
1 − M ⎛1 − M⎞⎤ 2 i j i n, c
+ ln
2 ⎝ 2 ⎠ ⎥⎦ 2n
× ∑ pn, c ( k , i ) (4-125)
Minimizing Ᏺ MFA
with respect to M now k =1
yields the elementary self-consistency where the coefficients gnc are combinato-
equation: rial factors depending on the lattice geome-
1 ˜ try and the clusters included in the approx-
M = tanh [ J (q = 0) M + H ] (4-124) imation (Kikuchi, 1951), e.g., in the tetra-
kB T
hedron approximation for the f.c.c. lattice,
As is well known, Eq. (4-124) implies a the sum over c in Eq. (4-125) includes the
second-order transition at Tc ∫ J̃ (q = 0)/kB (nearest-neighbor) tetrahedron, the near-
with the same exponents as in the Landau est-neighbor triangle, the nearest-neighbor
theory. pair, and the single site.
Clearly, in factorizing P ({Si}) into a Assuming the ordered structure to be
product of single-site probabilities and known, the symmetry operations of the as-
solving only an effective single-site prob- sociated group can be applied to reduce
lem, we have disregarded correlation in the the number of variational parameters in
probabilities of different sites. A system- Eq. (4-125). In the MFA, there is a single
atic improvement is obtained if we approx- non-linear self-consistent equation (Eq.
imate the probability of configurations not (4-124)) or a set of equations involving the
just by single-point probabilities but by us- order-parameter components if a problem
ing “cluster probabilities”. We consider more complicated than the Ising ferromag-
probabilities pnc (k, i) that a configuration k net is considered. In the CV method, a
of the n spins in a cluster of geometric con- much larger set of coupled non-linear
figuration c occurs (c may be a nearest- equations involving the short-range order
neighbor pair, or a triangle, tetrahedron, etc.). parameters gnc (i) is obtained when we min-
Note k = 1, … , 2n for Ising spins whereas imize Eq. (4-125). Therefore, whereas the
k = 1, … , qn for the q-state Potts model. simple MFA is still manageable for a wide
variety of systems (Brout, 1965; Smart, priori, it is often convenient to introduce

1966), the CV method is essentially re- Fourier transformations
stricted to Ising-type problems relevant for
1
phase transitions in metallic alloys (De X (q) = ∑ x i exp [ − i q ⋅ x i ] (4-130)
Fontaine, 1979; Finel, 1994). We discuss N i
the merits of the various approaches in the which yield:

next section.
N
At this point, we return to the formula- DᏲ MFA = ∑ f˜ ′′ ( q ) | X ( q )|2
tion of the MFA for the case of an arbitrary 2 q
type of ordering, rather than the simple N
+ f0′′′ ∑ X ( q ) X ( q ′ ) X ( q ′′ )
transition from paramagnetic to ferromag- 3! q, q ′, q ′′
netic considered in the above treatment × d ( q + q ′ + q ′′ − G )
(Eqs. (4-123) and (4-124)) of the Ising N IV
model (Eq. (4-56)). Hence Eq. (4-123) can + f0 ∑ X ( q ) X ( q ′ ) X ( q ′′ ) X ( q ′′′ )
4! q, q ′, q ′′, q ′′′
obviously be generalized as follows (see
also De Fontaine, 1975): × d ( q + q ′ + q ′′ + q ′′′ − G ) (4-131)
Ᏺ MFA = − 2 ∑ J ( xi − x j ) x i x j (4-126) where G is a reciprocal lattice vector and

i≠ j f̃ ¢¢(q) can be written as
+ kB T ∑ [ci ln ci + (1 − ci ) ln (1 − ci )] f̃ ¢¢(q) = – 4 J˜ (q) + kB T/[c (1 – c)] (4-132)
i
where xi = ci – c = (M – Si)/2, and constant Comparing Eqs. (4-121) and (4-132), we

terms have been omitted. Expanding Eq. realize that the inverse structure factor (or
(4-126) in terms of xi yields, again omitting inverse “susceptibility”, c – 1 (q)) is simply
a constant term, proportional to f̃ ¢¢(q), as it should be, since
1 X (q) and H (q) (apart from constants) are
DᏲ MFA = ∑ f0′′( xi − x j ) x i x j (4-127) canonically conjugate thermodynamic var-
2 i, j
iables. Of special importance now are the
1 1 IV points qc where J˜ (q) has its maximum and
+ f0′′′∑ x i3 + f0 ∑ x i4
3! i 4! i correspondingly f˜ ¢¢(q) has a minimum, be-
with the coefficients cause for these wavevectors a sign change
of f˜ ¢¢(q) in Eq. (4-131) occurs first (at the
f0¢¢(x ≠ 0) = – 4 J (xi – xj) highest temperature). Hence a stability
f0¢¢(x = 0) = kBT /[c (1 – c)] (4-128) limit of the disordered phase is predicted as
and kBTc = 4 J˜ (qc) c (1 – c) (4-133)
f0¢¢¢ = – kBT (2 c – 1)/[c (1 – c)]2 (4-129) If at this temperature a second-order transi-

tion occurs, it should actually be described
f0IV = 2
2 kBT [3 c – 3 c + 1)]/[c (1 – c)] 3
by a “concentration wave” X (qc) as an or-
If we group the x i s properly into the sub- der parameter (see also the discussion
lattices reflecting the (known or assumed) following Eq. (4-31)). In many cases the
state, Eq. (4-127) essentially yields the maxima of J˜ (q) occur at special symme-
Landau expansion in terms of the order-pa- try points of the Brillouin zone, and sym-
rameter components, as discussed in Sec. metry considerations in reciprocal space
4.2.1. Rather than defining sublattices a are useful for discussing the resulting or-
dered structures (De Fontaine, 1975, 1979; of an order-parameter component for the
Khachaturyan, 1973, 1983). transition: in mean-field theory, the asso-
This formulation of order – disorder tran- ciate eigenfrequency vanishes at a temper-
sitions in alloys is analogous to the treat- ature Tc (“soft phonon”). If this happens for
ment of structural transitions, for which a a phonon with wavevector k0 at the Bril-
model Hamiltonian similar to Eq. (4-127) louin zone edge, we have an antiferro-
can be written: rather than concentration electric order, provided that the phonon is
deviations xi we now have displacement polar, i.e., it produces a local dipole mo-
vectors ul (x) associated with a lattice vec- ment. For non-polar phonons, such as for
tor x for the l th atomic species in the unit the transition in SrTiO3 at Tc = 106 K where
cell. Just as it is useful to relate xi to the k0 = p (–12 , –12 , –12 )/a (the soft phonon there
Fourier transform X (q) of the concentra- physically corresponds to an antiphase ro-
tion deviation, it is useful to relate ul (x) to tation of neighboring TiO6 octahedra, as
the phonon normal coordinate Qk, l , de- indicated schematically in Fig. 4-12), the
fined as transition leads to “antiferrodistortive” or-
(4-134) der. Long-wavelength distortions corre-
1
ul ( x ) = ∑ exp( i k ⋅ x ) el ( k , l ) Qk, l sponding to optical phonons give rise
N Ml k, l
to ferroelectric ordering (an example is
where Ml is the mass of the atom of type l Pb5Ge3O11, where k0 = 0 and Tc = 450 K
at site Rli in the i th unit cell, el (k, x) is a (Gebhardt and Krey, 1979)). There are also
phonon polarization vector, l labels the paraelectric – ferroelectric transitions of
phonon branch and k its wavevector (Fig. first order, e.g., the cubic – tetragonal tran-
4-16). In this case, ·Qk0 , l0 ÒT plays the role sition of BaTiO3, such that no soft mode
occurs.
Just as the “macroscopic” ferroelectric
ordering can be associated with the normal
coordinate Qk0 = 0, l of the associate optical
phonon as a microscopic order parameter
characterizing the displacements on the
atomic scale, the “macroscopic” ferroelas-
tic orderings, where in the phenomenologi-
cal theory, Eq. (4-27), a component of the
strain tensor emn is used as an order pa-
rameter, can be related to acoustical pho-
nons. Examples are the martensitic mate-
rial In-25 at.% Tl where the combination
c11 – c12 of the elastic constants nearly sof-
tens at Tc = 195 K, and LaP5O14 where c55
Figure 4-16. (a) Schematic temperature variation of softens at Tc = 400 K and the structure
order parameter F = ·Q k0 , l 0 Ò T and square of the changes from orthorhombic to monoclinic
“soft-mode” frequency w 2 (k0 , l 0 ) at a displacive (Gebhardt and Krey, 1979). Of course, this
structural transition. (b) Schematic phonon spectrum
correspondence between the microscopic
of the solid at T > Tc . Note that either optical or
acoustic phonons may go soft, and for optical pho- description of displacements in crystals in
nons a softening may often occur at the boundary of terms of phonons and the phenomenologi-
the first Brillouin zone rather than at its center. cal macroscopic description in terms of po-
larization fields, strain fields, etc., is a component structural transition as follows

well-known feature of condensed-matter (Bruce and Cowley, 1981):
theory (Kittel, 1967).
ᏴF 4 = ∑ ⎡⎢ r Fi2 + uFi4 ⎤⎥
1 1
As is well known, the Qk, l are defined
i ⎣ 2 4 ⎦
such that the Hamiltonian of the crystal in
1
the quasi-harmonic approximation is diag- + ∑ C (Fi − F j )2 (4-136)
onalized (Born and Huang, 1954): 〈 i, j 〉 2
where r, u and C are phenomenological co-

1
Ᏼ = U0 + ∑ [ ∂ 2U /( ∂xil )a ( ∂x lj′ )b ] efficients. The last term corresponds to
2 i, j , l , l ′, a , b the harmonic interaction between displace-
× ula ( xi ) ulb′ (x j ) ments in neighboring lattice cells, while
1 the anharmonicity has been restricted to the
= U0 + ∑ w 2 ( k , l ) | Qk, l |2 (4-135) “single-site Hamiltonian” [r Fi2/2 + u Fi4/4].
2 k, l
Note that Eq. (4-136) is fully equivalent to
Eq. (4-135) corresponds precisely to Eqs. the “Hamiltonian” DᏲ MFA in Eq. (4-127) if
(4-127) and (4-131) if only the “harmonic f0¢¢¢= 0 is chosen, since (Fi – Fj)2 = Fi2 + Fj2
terms” (i.e., quadratic terms in the expan- – 2 Fi Fj, and the coefficient C is thus
sion with respect to the xi s) are retained. equivalent to f0¢¢(xi – xj) where xi – xj is a
The vanishing of a soft mode, w2 (k0 , l0) ~ nearest-neighbor distance, a. On the other
(T – Tc ), is again equivalent to the vanish- hand, the “F 4-model” in Eq. (4-136) can
ing of a coefficient r ~ (T – Tc ) of a quad- be thought of as a lattice analog of the
ratic term in a Landau expansion. Of Helmholtz energy functional Eq. (4-10),
course, as in the Landau theory, higher-or- putting (Fi – Fj) ≈ a · —Fi (see Milchev
der “anharmonic” terms in Q are crucial for et al. (1986) for a discussion).
the description of the ordered phase in An interesting distinction concerns this
terms of a stable order parameter ·Qk0 , l0 Ò effective single-site Hamiltonian felt by the
for T < Tc . atoms undergoing the distortion. We have
These anharmonic terms thoroughly assumed that the ordered structure is
modify the picture of the transition as ob- doubly degenerate; the atoms below Tc can
tained from MFA, as they lead to a cou- sit in the right or the left minimum of a
pling of the soft mode Qk0, l0 with other double-well potential. If the single-site po-
noncritical modes. This coupling among tential above Tc is essentially of the same
modes gives rise to a damping of the soft type, and only the distribution of the atoms
mode. In fact, under certain circumstances over the minima is more or less random,
even an overdamped soft mode and the ap- the transition is called “order – disorder
pearance of a central peak are expected type”. This occurs, for example, for hydro-
(Gebhardt and Krey, 1979; Bruce and gen-bonded ferroelectrics and is analogous
Cowley, 1981). to the sublattice ordering described above
A treatment of the anharmonic higher- for alloys. On the other hand, if the single-
order terms of a Landau-like expansion site potential itself changes above Tc to a
in reciprocal space is cumbersome, as Eq. single-well form, the transition is called
(4-131) demonstrates. Denoting the ampli- “displacive”. Whereas it was often thought
tude of the displacement vector produced that displacive structural transitions exhibit
by the soft mode Qk0 , l0 in the unit cell i as well-defined soft phonons right up to their
Fi , we may formulate a model for a one- transition temperature Tc, it has now be-
come clear that all these structural transi- different structures can be handled by per-
tions with a one-component order parame- forming this calculation for both phases
ter acquire characteristics of order – disor- and identifying the temperature Tc where
der transitions close to Tc, as expected from the free energy branches of the two phases
the “universality principle”, and therefore cross. Since the quasi-harmonic theory is a
the distinction between the character of a calculation of the mean-field type, as
structural transition as being “order – disor- pointed out above, first order transitions
der” or “displacive” is not a sharp one also show up via stability limits of the
(Bruce and Cowley, 1981). phases, where the soft modes vanish; thus
It should also be noted that the softening we are not locating Tc but rather tempera-
of w (k0, l0) near Tc does not mean that dis- tures T0 or T1 (cf. Fig. 4-6 b), which are
placements ul (x) become very large. In often not very far from the actual transi-
fact, the mean square displacement of an tion temperature. This quasi-harmonic ap-
atom at a structural transition is only ex- proach to structural phase transitions has
pected to have an energy-like singularity been tried for many materials. Typical ex-
(Meißner and Binder, 1975) amples include RbCaF3 (Boyer and Hardy,
1981) and the systems CaF2 and SrF2
·u2l ÒT – ·u2l ÒTc ~ (T /Tc – 1)1 – a (4-137)
(Boyer, 1980, 1981 a, b), which show phase
where a is the specific heat exponent. This transitions to a superionic conducting state.
is important because ·u2l ÒT is easily de-
duced experimentally from the Mössbauer
effect, from the Debye – Waller factor de- 4.3.3 Computer Simulation Techniques
scribing the temperature variation of Bragg In a computer simulation, we consider a
peaks in X-ray or neutron scattering, etc. finite system (e.g., a cubic box of size L3
We end this subsection with a comment with periodic boundary conditions to avoid
on the theory of first-order structural tran- surface effects) and obtain information on
sitions. The common approach is to restrict the thermodynamic properties, correlation
the analysis entirely to the framework of functions, etc. of the system (as specified
the quasi-harmonic approximation, in by its model Hamiltonian) which is exact,
which the Helmholtz energy at volume V apart from statistical errors. However, this
and temperature T is written as approach is restricted to classical statistical
mechanics (including quasi-classical mod-
F (T , V ) = U − T S
els such as Ising or Potts models), although
1
= U0 ( V ) + ∑ w V ( k , l ) (4-138) remarkable progress on application to
2 k, l quantum problems has been made (Kalos,
+ kB T ∑ ln [1 − exp( − w V ( k , l )/ kB T )] 1985; Kalos and Schmidt, 1984; De Raedt
k, l and Lagendijk, 1985; Suzuki, 1992; Ceper-
Therefore, if the effective potentials speci- ley, 1995). The principal approaches of this
fying the dynamical matrix ∂2 U/[(∂xil )a type are the molecular dynamics (MD)
∂xjl )b ] in Eq. (4-135) are known, the pho- technique and the Monte Carlo (MC) tech-
non frequencies wV (k, l) for a given vol- nique (for reviews of MD, see Ciccotti
ume and the free energy F (T, V) are ob- et al., 1987; Hoover, 1987; Hockney and
tained. Of course in this approach, knowl- Eastwood, 1988; Binder and Ciccotti,
edge of the structure of the material is 1996; of MC, see Binder, 1979, 1984 a;
assumed. First-order transitions between Mouritsen, 1984; Binder and Heermann,
1988; Binder and Ciccotti 1996). In the principle, this problem is well understood
MD method, we numerically integrate New- (Fisher 1971; Challa et al., 1986; Binder,
ton’s equation of motion which follows 1987 b; Privman, 1990). In practice, this
from the chosen Hamiltonian, assuming er- makes it difficult to distinguish between
godic behavior, and the quantities of inter- second-order and weakly first-order transi-
est are obtained as time averages from the tions. For example, extensive MD work
simulation. This method requires that the was necessary to obtain evidence that the
relevant physical time scales involved in melting transition of two-dimensional sol-
the problem are not too different from each ids with pure Lennard – Jones interaction is
other. The MD approach has often been ap- first order (Abraham, 1983, 1984; Bakker
plied successfully to liquid – solid transi- et al., 1984), and that the suggested two
tions. It would not be suitable to study or- continuous transitions involving the hexa-
der – disorder phenomena in solid alloys, tic phase do not occur in these systems
since the time step in the MD method for a (Nelson and Halperin, 1979). However,
solid must be much less than a phonon fre- this conclusion has been called into ques-
quency, and this time scale is orders of tion by recent simulations for hard-disk
magnitude smaller than the time between fluids (Jaster, 1998) providing evidence for
diffusive hops of atoms to neighboring va- continuous two-dimensional melting.
cant sites, which is the process relevant to Another difficulty is that the periodic
equilibration of configurational degrees of boundary condition (for a chosen shape of
freedom (Fig. 4-12). the box) prefers certain structures of a solid
For problems of the latter type, the MC and suppresses others which do not “fit”:
method clearly is to be preferred. In the this is particularly cumbersome for incom-
MC method, random numbers are used to mensurate modulated structures (Selke, 1988,
construct a random walk through the con- 1989; 1992) and for off-lattice systems,
figuration space of the model system. Us- such as studies of the fluid – solid transition
ing the Hamiltonian, transition probabilities or phase transitions between different lat-
between configurations are adjusted such tice symmetries. For example, particles in-
that configurations are visited according to teracting with a screened Coulomb potential
their proper statistical weight (Binder, (this is a model for colloidal suspensions or
1979, 1984 a; Binder and Ciccotti, 1996). colloidal crystals (see Alexander et al.,
Again averages are obtained as “pseudo- 1984)) may exhibit a fluid phase in addi-
time averages” along the trajectory of the tion to an f.c.c. and a b.c.c. crystal, and the
system in phase space, the only difference determination of a complete phase diagram
from the MD method being that the tra- is correspondingly difficult (Kremer et al.,
jectory is now stochastic rather than deter- 1986, 1987; Robbins et al., 1988). The tra-
ministic. Both methods have been exten- ditional approach to dealing with such prob-
sively reviewed (see the references quoted lems is to repeat the calculation for differ-
above); therefore, we do not give any de- ent box shapes and compare the free ener-
tails here, but only briefly mention the dif- gies of the different phases. An interesting
ficulties encountered when phase transi- alternative method has been proposed by
tions are studied, and discuss a few typical Parrinello and Rahman (1980) and Parri-
examples of their application. nello et al. (1983), who generalized the MD
One principal difficulty is the finite-size method by including the linear dimensions
rounding and shifting of the transition. In of the box as separate dynamic variables.
Another severe problem is the occur-

rence of metastability and hysteresis; the
system may become trapped in a meta-
stable state, the lifetime of which is longer
than the observation time of the simulation.
The distinction of such long-lived meta-
stable states from true equilibrium states is
difficult and may require computation of
the Helmholtz energies of the phases in
question.
These problems must be considered
when first-order transitions are studied by
computer simulation so that suitable box
sizes and observation times are chosen and
the initial state is prepared accordingly.
Then, employing sufficient effort in com-
puting time, and performing a careful anal-
ysis of all the possible pitfalls mentioned
above, very reliable and useful results can
be obtained which are superior in most
cases to any of the other methods that have
been discussed so far (see Fig. 4-15 for a
comparison in the case of f.c.c. alloys).
Misjudgements of the problems mentioned
above have also led to erroneous con- Figure 4-17. Temperature dependence of the long-
clusions: in the f.c.c. Ising antiferromagnet range order parameter Y of the Cu3Au structure and
of the nearest- and next-nearest-neighbor short-range
with nearest interaction J, the triple point order parameter a 1 and a 2 . Curves are Monte Carlo
between the AB and A3B structures results of Binder (1986). Points show experimental
(Fig. 4-8 a, c) was suggested to occur at data of Cowley (1950), Schwartz and Cohen (1965),
T = 0 (Binder, 1980) while later a nonzero Moss (1964), Bardham and Cohen (1976), and Keat-
ut low temperature was found, kBT /| J | ~ 1.0 ing and Warren (1951). From Binder (1986).
(Gahn, 1986; Diep et al., 1986; Kämmerer
et al., 1996). Unlike in the other methods,
such problems can always be clarified by sive (Jnn) and next-nearest-neighbor attrac-
substantially increasing the computational tive interaction Jnnn (Eq. (4-110)), for vari-
effort and carrying out more detailed anal- ous choices of their ratio R = Jnnn/Jnn. The
yses of the simulation “data”. comparison with experimental data shows
A significant advantage of such a “com- that a reasonable description of Cu3Au is
puter experiment” is that interaction pa- obtained for R = – 0.2, for T < Tc , whereas
rameters of a model can be varied system- better experimental data for T > Tc are
atically. As an example, Fig. 4-17 shows needed before a more refined fit of effec-
the temperature dependence of the long- tive interaction parameters can be per-
range order parameters and some short- formed.
range order parameters of an f.c.c model of As an example showing that the MC
an A3B alloy with nearest-neighbor repul- method can deal with very complex phase
diagrams, we consider the model in Eq.

(4-115) for alloys with one magnetic com-
ponent. However, in order to facilitate
comparison with the MFA and CV meth-
ods, the magnetic interaction is chosen to be
of Ising rather than Heisenberg type (Dün-
weg and Binder, 1987). Fig. 4-18 shows
the resulting phase diagrams for the choice
of interaction parameters Jnnn/Jnn = 0.5 and
Jm /| Jnn | = 0.7. In this case, the MFA pre-
dicts a wrong phase diagram “topology”
(e.g., a direct transition between paramag-
netic D03 and A2 phases never occurs,
there is always a B2 phase in between), and
it also grossly overestimates the transition
temperatures. In contrast, the CV method
yields the phase diagram “topology” cor-
rectly, and it overestimates the transition
temperature by only a few per cent. Al-
though the CV method does not always
perform so well (see Fig. 4-15, and Binder
(1980) and Diep et al. (1986) for a discus-
sion in the case of f.c.c. alloys), it is always
much superior to the simple MFA, which in
many cases fails dramatically.
One great advantage of the MC method
over the CV method is that it can also be
applied straightforwardly to models with
continuous degrees of freedom, for which
the CV method would be cumbersome to
work out. As examples, Figs. 4-19 and
Figure 4-18. Phase diagram of the body-centered
4-20 show the phase diagrams of the model
cubic binary alloy model with nearest and next near-
in Eqs. (4-93) and (4-136), namely the an- est neighbor crystallographic interactions Jnn and
isotropic classical Heisenberg antiferro- Jnnn (both being “antiferromagnetic” if Ising-model
magnet in a uniform field H|| along the easy terminology is used) and a ferromagnetic nearest-
axis and the F 4 model on the square lat- neighbor interaction Jm between one species, for
tice. Note that in the latter case the result Jnnn /Jnn = 0.5, Jm / | Jnn | = 0.7; c denotes the concentra-
tion of the magnetic species. (a) Result of the MFA
of the MFA would be off-scale in this (Bragg – Williams approximation); (b) result of the
figure (e.g., Kc– 1 = 4 in the Ising limit). Fi- CV method in the tetrahedron approximation, and
nally, Fig. 4-21 shows the phase diagram of part (c) the MC result. The A2 phase is the crystallo-
a model for EuxSr1 – xS (Binder et al., graphically disordered phase, the orderings of the B2
1979): and D03 phases are shown in Fig. 4-3. Two-phase co-
existence regions are shaded. The magnetic ordering
of the phases is indicated as para(-magnetic) and
1
Ᏼ=− ∑ Jij xi x j Si ⋅ S j (4-139) ferro(-magnetic), respectively. From Dünweg and
2 i≠ j Binder (1987).
Figure 4-19. Phase diagram of a uniaxial classical Figure 4-20. Critical line Kc–1 of the F 4 model
Heisenberg antiferromagnet on the simple cubic lat- (Eq. (4-136)) on the square lattice, shown in the
tice, as a function of temperature T and field H|| ap- space of couplings K –1, L–1 defined as K = – r C/u,
plied in the direction of the easy axis. The anisotropy L = r (r + 4 C)/(4 u). Note that for L–1 = 0, and K finite,
parameter D in the Hamiltonian Eq. (4-93) is chosen the square Ising model results, the transition temper-
as D = 0.2. Both Monte Carlo results (crosses, circles, ature of which is exactly known (Onsager, 1944; ar-
full curves) and the Landau theory fitted to the phase row). (쐌) Monte Carlo simulation of Milchev et al.
diagram in the region off the true bicritical point is (1986); (+) MD results of Schneider and Stoll
shown (dash-dotted straight line). The Landau theory (1976); (왌, ¥) from a real space renormalization
would overestimate the location of the bicritical tem- group calculation of Burkhardt and Kinzel (1979).
perature Tb (Tb* > Tb ), and fails to yield the singular The disordered phase occurs above the critical line.
umbilicus shape of the phase diagram lines Tc^, Tc|| From Milchev et al. (1986).
near the bicritical point. The broken straight lines de-
note the appropriate choices of “scaling axes” for a
crossover scaling analysis near Tb . The triangles nearest neighbors only, further-neighbor
show another phase diagram, namely when the field interactions being negligibly small, with
is oriented in the perpendicular direction to the easy
axis. The nature of the phases is denoted as AF (anti-
Jnnn/ Jnn = – 1/2. Thus Fig. 4-21 exhibits a
ferromagnetic), SF (spin-flop) and P (paramagnetic). remarkable agreement between computer
See Fig. 4-9 a for a schematic explanation of this simulation and experiment (Maletta and
phase diagram. From Landau and Binder (1978). Felsch, 1979); there are no adjustable pa-
rameters involved whatsoever. Note that
where Si are unit vectors in the direction of analytical methods for such problems with
the Eu magnetic moment, and from spin quenched disorder would be notoriously
wave measurements (Bohn et al., 1980) it difficult to apply.
is known that EuS essentially exhibits To conclude this subsection, we state
superexchange between nearest and next- that computer simulation techniques are
4.4 Concepts About Metastability

Metastable phases are very common in
nature, and, for many practical purposes,
not at all distinct from stable phases (re-
member that diamond is only a metastable
modification of graphite!). Also, approxi-
mate theories of first-order phase transi-
tions easily yield Helmholtz energy
branches that do not correspond to the ther-
mal equilibrium states of minimum
Helmholtz energy, and hence are com-
Figure 4-21. Ferromagnetic critical temperature of monly interpreted as metastable or unstable
the model Eq. (4-139) of Eu x Sr1– x S vs. concentration states (cf. Figs. 4-6 and 4-22). From the an-
x of magnetic atoms. Circles denote experimental alog of Eq. (4-12) for H (0,
data due to Maletta and Felsch (1979), triangles the (4-140)
MC results for the diluted classical Heisenberg fcc 1 ⎛ ∂F ⎞ H
⎜ ⎟ = r F + uF 3 − =0
kB T V ⎝ ∂F ⎠ T
ferromagnet with nearest (Jnn ) and next-nearest
(Jnnn ) exchange, Jnnn = – –12 Jnn . From Binder et al.
kB T
(1979).
the limit of metastability where c T =
(∂F /∂H )T diverges as follows:
well suited for studying the phase diagrams 1 1
r + 3 uFs2 = =0
of various systems, magnetic systems, me- kB T c T
tallic alloys, structural transitions, ad-
Fs = − r / 3 u = F0 / 3
sorbed layers at surfaces, etc. provided that
a suitable model Hamiltonian is known. 2r −r
Hc = − (4-141)
Apart from the phase diagram, detailed in- 3 3u
formation on long- and short-range order is
also accessible. Also, for certain problems, Since in the metastable states
the kinetics of phase transitions can be
cT = (3 kB Tu)– 1 (F 2 – F s2)– 1
studied, in particular for alloys where the
Monte Carlo process can directly model we see that cT Æ • as F Æ Fs; in the
the atomic jump processes on the crystal (T, F ) plane, the spinodal curve F = Fs (T )
lattice (Binder, 1979, 1984 a; Kehr et al., plays the part of a line of critical points.
1989). Of course, there is also interest in Similar behavior also occurs in many
clarifying the phase diagrams of materials other theories: for example, the van der
in their fluid states. Computer simulations Waals equation of state describing gas – liq-
can now obtain accurate gas – liquid phase uid condensation exhibits an analogous
diagrams (Wilding, 1997), liquid binary loop of one-phase states in the two-phase
mixtures (Wilding et al., 1998) and surfac- coexistence region.
tants (Schmid, 1999). Unfortunately, although the description
of metastability in the framework of the
MFA seems straightforward, this is not so
if more accurate methods of statistical ther-
modynamics are used. A heuristic compu-
4.4 Concepts About Metastability 299
tational approach, discussed in Sec. 4.3.2,

is the CV method where we systematically
improve upon the MFA by taking more and
more short-range correlations into account,
the larger the cluster is chosen. Computing
the equation of state for a nearest-neighbor
Ising ferromagnet in this way (Kikuchi,
1967), it is found that the stable branch is
nicely convergent, whereas the metastable
loop becomes flatter and flatter the larger
the cluster, i.e., the critical field Hc (Eq. Figure 4-22. Order parameter F vs. conjugate field
(4-141)) converges towards zero. An exact H according to the phenomenological Landau theory
calculation of the equation of state yields for system at a temperature T less than the critical
the magnetization jump from F0 to – F0 temperature Tc of a second-order phase transition
as H changes from 0+ to 0– in Fig. 4-22, but (schematic). At H = 0, a first-order transition from F0
to – F0 occurs (thick straight line). The metastable
does not yield any metastable state! This is branches (dash-dotted) end at the “limit of meta-
not surprising, however, because statistical stability” or “spinodal point” (Fs , – Hs ), respectively,
mechanics is constructed to yield informa- and are characterized by a positive-order parameter
tion only on thermal equilibrium states. susceptibility c T > 0, whereas for the unstable branch
The partition function is dominated by the (broken curve) c T < 0.
system configurations in the vicinity of the
minimum of the Helmholtz energy. In a
magnet with H < 0, states with negative magnetization in equilibrium is negative,
magnetization have lower Helmholtz en- but in the course of the nonequilibrium re-
ergy than those with positive magnetiza- laxation process from the initial state at F 0
tion. Hence the latter do not result from the a metastable state may occur with Fms > 0,
partition function in the thermodynamic although H < 0, which exists only over a
limit. Similar conclusions emerge from a finite lifetime tms . In order that Fms can
rigorous treatment of the gas – fluid transi- clearly be identified in this dynamic pro-
tion in systems with long-range interac- cess where the time-dependent order pa-
tions (Lebowitz and Penrose, 1966). rameter F (t) relaxes from F (t = 0) = F 0 to
Two concepts emerged for the descrip- the negative equilibrium value F (t Æ •),
tion of metastable states (for a more de- it is necessary that tms is much larger than
tailed discussion see Binder (1987 a)). One any intrinsic relaxation time. Then the time
concept (Binder, 1973) suggests the defini- tms for the decay of the metastable state far
tion of metastability be based on the con- exceeds the time needed for the system to
sideration of kinetics: starting with the sys- relax from the initial state towards the met-
tem in a state of stable equilibrium, the sys- astable state, and, in the “nonequilibrium
tem is brought out of equilibrium by a sud- relaxation function” F (t) the metastable
den change of external parameters (temper- state shows up as a long-lived flat part
ature, pressure, fields, etc.). For example, where F (t) ≈ Fms. The second concept
in the Ising-type ferromagnet of Fig. 4-22, tries to define a metastable state in the
at times t < 0 we may assume we have a framework of equilibrium statistical me-
state at H = 0+, F = F 0 > 0: at t = 0, the field chanics by constraining the phase space so
is switched to a negative value: then the as to forbid the two-phase configurations
that otherwise dominate the partition func- F = Fs (T ) acts like a line of critical points,
tion. Langer (1974) suggested that phase as discussed above, we require (cf. Eq.
separation into two phases with order pa- (4-58))
rameters – F 0 and + F 0 coexisting at the
·[F (x) – F]2 ÒT, L O [F – Fs (T )]2 (4-142)
first-order transition at H = 0 in Fig. 4-1
and Fig. 4-22 is suppressed if we consider i.e., the mean-square fluctuations in a vol-
a system (a) at fixed F between – F 0 and ume Ld must be much smaller than the
+ F 0 but (b) constrain the system and di- squared distance of the order parameter
vide it into cells of size Ld, and (c) require from its value at criticality, which in this
that the order parameter is not only glo- case is the value at the spinodal. Now, the
bally fixed at F but also inside each cell. maximum permissible value for L is the
If L is small enough, namely a O L O xcoex correlation length x in the metastable state,
(this is the same condition as noted for which becomes (4-143)
the construction of a continuum model
from the microscopic Hamiltonian, see x = {R/[6 d u Fs (T )]1/2} [F – Fs (T )]– 1/2
Sec. 4.2.2, and Eqs. (4-9) and (4-57)), Eq. (4-143) exhibits the critical singularity
phase separation inside a cell cannot occur, of x at Fs (T ), corresponding to that of cT
and hence a coarse-grained Helmholtz en- noted above. With a little algebra we find
ergy density fcg (F ) of states with uniform from Eqs. (4-142) and (4-143) (Binder,
order parameter F is obtained, which has 1984 c)
exactly the double-well shape of the Lan-
dau theory (Fig. 4-6 a). Since fcg (F ) then 1 O R2 x d − 2 [F − Fs ( T )]2
necessarily depends on this coarse-grained ~ R d [Fs ( T )]( 2 − d )/ 2 [F − Fs ( T )]( 6 − d )/ 2
length scale L, it is clear that the metastable
~ R d ( Hc − H )( 6 − d )/ 4 (4-144)
states of fcg (F ) are not precisely those
observed in experiments, unlike the dy- This condition can only be fulfilled if the
namic definition. While fcg (F ) defines a interaction range is very large. Even then
“limit of metastability” or “spinodal curve” for d = 3 this inequality is not fulfilled very
Fs where (∂2 fcg (F )/∂F 2 )T = c T– 1 vanishes close to the spinodal curve.
and changes sign, this limit of metastability In order to elucidate the physical signifi-
also depends on L and hence is not related cance of this condition further and to dis-
to any physical limit of metastability. In the cuss the problem of the lifetime of meta-
context of spinodal decomposition of mix- stable states, we briefly discuss the mean
tures, the problem of the extent to which a field theory of nucleation phenomena. In
spinodal curve is relevant also arises and is practice, for solid materials, heterogeneous
discussed in detail in the chapter by Binder nucleation (nucleation at surfaces, grain
and Fratzl (2001). boundaries, dislocations, etc.) will domi-
In some systems where a mean-field de- nate; and this may restrict the existence of
scription is appropriate, the lifetime of metastable states much more than expected
metastable states can be extremely long, from Eq. (4-144). However, we consider
and then the mean-field concept of a limit here only the idealized case that homoge-
of metastability may be very useful. This neous nucleation (formation of “droplets”
fact can again be understood in terms of of the stable phase due to statistical fluctu-
the Ginzburg criterion concept as outlined ations) is the mechanism by which the met-
in Sec. 4.2.2. Since the spinodal curve astable state decays. Since there is no
4.4 Concepts About Metastability 301
mechanism by which this process can be

suppressed, metastable states cannot exist
in a region where strong homogeneous nu-
cleation occurs. Therefore we estimate the
intrinsic and ultimate limit of metastability
by the condition that the Helmholtz energy
barrier against homogeneous nucleation
decreases to the order of the thermal en-
ergy kBT (Binder and Stauffer, 1976 b).
As is well known, the nucleation barrier
arises from competition between the favor-
able) volume energy of a droplet of the new
phase and the (unfavorable) surface Helm-
holtz energy between the droplet and the
surrounding metastable background phase.
In order to calculate this Helmholtz energy
barrier in the framework of the Landau the-
ory, we can still use Eqs. (4-10) and (4-33).
Putting H (x) = 0 in Eq. (4-33) and consid-
ering the situation that F (z Æ •) = F 0 ,
F (z Æ – •) = – F 0 and solving for the con-
centration profile F (z) (Cahn and Hilliard,
Figure 4-23. Order-parameter profile F (z) across
1958), we can obtain the interface Helm-
an interface between two coexisting phases ± Fcoex ,
holtz energy associated with a planar (infi- the interface being oriented perpendicular to the z-di-
nitely extended) interface perpendicular to rection (a) and the radial order-parameter profile for
the z-direction (Fig. 4-23 a). In the “classi- a marginally stable droplet in a metastable state
cal theory of nucleation” (Zettlemoyer, which is close to the coexistence curve (b) or close to
the spinodal curve (c). In (a) and (b) the intrinsic
1969), this Helmholtz energy is then used
“thickness” of the interface is of the order of the cor-
to estimate the Helmholtz energy barrier. relation length at coexistence x coex , whereas in (c) it
However, this “classical theory of nuclea- is of the same order as the critical radius R*. From
tion” is expected to be reliable only for Binder (1984 c).
metastable states near the coexistence
curve, where the radius R* of a critical permitted, and a boundary condition
droplet (corresponding to a droplet Helm- F (r Æ •) = F ms is imposed (Fig. 4-
holtz energy exactly at the Helmholtz en- 23 b, c). Whereas for Fms near Fcoex = F 0
ergy barrier DF*) is much larger than the this treatment agrees with the “classical
width of the interfacial profile (which then theory of nucleation”, it differs signifi-
is also of the same order as the correlation cantly from it for F near Fs (T ): then the
length xcoex at the coexistence curve, cf. critical droplet radius R* is of the same or-
Figs. 4-23 a, b). Therefore it cannot be used der as the (nearly divergent!) correlation
close to the limit of metastability. Cahn and length x (Eq. (4-143)), and the profile is
Hilliard (1959) have extended the Landau extremely flat, F (r) reaches in the droplet
theory to this problem, solving Eq. (4-33) center only a value slightly below Fs rather
for a spherical geometry, where only a than the other branch of the coexistence
radial variation of F (r) with radius r is curve. Calculating the Helmholtz energy
barrier DF*, we obtain for T near Tc

(Binder, 1984 c; Klein and Unger, 1983)
( 4 − d )/ 2 ( 6 − d )/ 2
DF* ⎛ T⎞ ⎛ F − Fs ⎞
~ Rd ⎜1 − ⎟ ⎜ ⎟
kB Tc ⎝ Tc ⎠ ⎝ Fcoex ⎠
(4-145)
whereas near the coexistence curve the re-
sult is
( 4 − d )/ 2 − ( d − 1)
DF* ⎛ T⎞ ⎛ Fcoex − F ⎞
~ Rd ⎜1 − ⎟ ⎜ ⎟
kB Tc ⎝ Tc ⎠ ⎝ Fcoex ⎠
(4-146)
where all prefactors of order unity are omit-
ted. In a system with a large range R of in-
teraction, the nucleation barrier is very high
in the mean-field critical region, in which
R d (1 – T /Tc)(4 – d )/2 o 1 (cf. Eq. (4-62)); this
factor, which controls the Ginzburg criter-
ion, also controls the scale of the nuclea-
tion barrier as a prefactor (see Fig. 4-24).
Figure 4-24. Schematic plots of the Helmholtz en-
In this region the condition for the limit of ergy barrier for (a) the mean field critical region, i.e.,
metastability, DF*/kBT ≈ 1, is located very R d (1 – T/Tc ) (4 – d )/2 o 1, and (b) the non-mean field
close to the mean-field spinodal. Then the critical region i.e., R d (1 – T/Tc ) (4 – d )/2 O 1. Note that
description of nucleation phenomena in owing to large prefactors to the nucleation rate, the
terms of the diffuse droplets described by constant of order unity where the gradual transition
from nucleation to spinodal decomposition occurs is
Fig. 4-23 c near the spinodal curve is mean- about 101 rather than 100. From Binder (1984 c).
ingful (“spinodal nucleation”). On the
other hand, for a system with short-range
interactions where R (measured in units
of the lattice spacing in Eqs. (4-145) and new phase) are also extremely rare, and the
(4-146)) is unity, the Helmholtz energy Helmholtz energy cost to form them should
barrier becomes of order unity long before be very high, as implied by Eqs. (4-145)
the spinodal curve is reached. The singu- and (4-146).
larity at the spinodal then completely lacks These concepts have been tested by
any physical significance, as the meta- Monte Carlo simulations on simple-cubic
stable state decays to the stable phase long Ising models with various choices of the
before the spinodal is reached. range of the interaction (Fig. 4-25, Heer-
It is instructive to compare the condition mann et al., 1982). For the case of nearest-
DF*/kBT O 1 with Eq. (4-144): this shows neighbor interaction (each spin interacts
that these conditions are essentially iden- with q = 6 neighbors) it is seen that cT– 1 dif-
tical! This is not surprising: the MFA is fers considerably from the MFA result, and
essentially correct as long as effects of the range of fields over which metastable
statistical fluctuations are very small: the states can be observed is not very large. With
“heterophase fluctuations” (droplets of the an increase in the number of neighbors q,
4.5 Discussion 303
have a strong influence on certain physical

properties. The approach of statistical me-
chanics tries to provide general concepts
for dealing with such phenomena: classifi-
cation methods are developed which also
try to clarify which aspects of a phase
transition are specific for a particular mate-
rial and which are general (“universal”). At
the same time, theoretical descriptions are
available both on a phenomenological
level, where thermodynamic potentials are
expanded in terms of suitable order param-
eters and the expansion coefficients are un-
determined, and can only be adjusted to ex-
perimental data, and on a microscopic
level, where we start from a model Hamil-
tonian which is treated either by molecular
field approximations or variants thereof or
by computer simulation techniques.
This chapter has not given full details of
all these approaches, but rather tried to
Figure 4-25. Inverse susceptibility of Ising ferro-
magnets plotted against h = – H/kB T at T/TcMFA = 4/9 give a discussion which shows what these
for various ranges of the exchange interaction; each methods can achieve, and to give the reader
spin interacts with equal strength with q neighbors. a guide to more detailed literature on the
The full curve is the MFA, the broken curve a fit to a subject. An attempt has been made to sum-
droplet model description. From Heermann et al.
marize the main ideas and concepts in the
(1982).
field and to describe the general facts that
have been established, while actual materi-
als and their phase transitions are men-
this range increases, the value of cT– 1 at tioned as illustrative examples only, with
the field where nucleation becomes appre- no attempt at completenesss being made.
ciable falls also, and the cT– 1 vs. H curve Although the statistical thermodynamics of
quickly converges towards the MFA pre- phase transitions has provided much physi-
diction, except in the immediate neighbor- cal insight and the merits and limitations of
hood of the limit of metastability Hc. the various theoretical approaches are now
It is expected that the ideas sketched well understood generally, the detailed
here will carry over to more realistic sys- understanding of many materials is still ru-
tems. dimentary in many cases; often there are
not enough experimental data on the order-
ing phenomenon in question, or the data
4.5 Discussion are not precise enough; a microscopic
model Hamiltonian describing the interac-
There are many different kinds of phase tion relevant for the considered ordering is
transitions in materials, and the fact that a often not explicitly known, or is so com-
material can exist in several phases may plicated that a detailed theory based on
such a model Hamiltonian does not exist. Baxter, R. J. (1982), Exactly Solved Models in Statis-
tical Mechanics. London: Academic.
Many of the more sophisticated methods Bean, C. P., Rodbell, D. S. (1962), Phys. Rev. 126,
(CV method, computer simulation, etc.) 104.
are restricted in practice to relatively sim- Belanger, D. P. (1998), in: Spin Glasses and Random
Fields: Young, A. P. (Ed.). Singapore: World Sci-
ple models. Hence, there still remains a lot entific, p. 251.
to be done to improve our understanding of Berlin, T. H., Kac, M. (1952), Phys. Rev. 86, 821.
phase transitions. Bethe, H. (1935), Proc. Roy. Soc. A 150, 552.
Bieber, A. Gautier, F. (1984 a), J. Phys. Soc. Japan
53, 2061.
Bieber, A., Gautier, F. (1984 b), Z. Phys. B 57, 335.
4.6 References Binder, K. (1973), Phys. Rev. B 8, 3423.
Binder, K. (Ed.) (1979), Monte Carlo Methods in
Abraham, F. F. (1974), Homogeneous Nucleation Statistical Physics. Berlin: Springer.
Theory. New York: Academic. Binder, K. (1980), Phys. Rev. Lett. 45, 811.
Abraham, F. F. (1983), Phys. Rev. Lett. 50, 978. Binder, K. (1983 a), Z. Physik B 50, 343.
Abraham, F. F. (1984), Phys. Rev. B 29, 2606. Binder, K. (1983 b), in: Phase Transitions and Criti-
Aharony, A. (1973), Phys. Rev. B 8, 1973. cal Phenomena, Vol. 8: Domb, C., Green, M. S.
Aharony, A. (1976), in: Phase Transitions and Criti- (Eds.). London: Academic Press, p. 1.
cal Phenomena, Vol. 6: Domb, C., Green, M. S. Binder, K. (1984 a), Applications of the Monte Carlo
(Eds.). London: Academic, p. 358. Method in Statistical Physics. Berlin: Springer
Alefeld, G. (1969), Phys. Stat. Sol. 32, 67. Verlag.
Alefeld, G., Völkl, J. (Eds.) (1978), Hydrogen in Binder, K. (1984 b), in: Condensed Matter Research
Metals I, II. Berlin: Springer Verlag (Topics in Using Neutrons: Lovesey, S. W., Scherm, R.
Appl. Physics, Vols. 28, 29). (Eds.). New York: Plenum Press, p. 1.
Alexander, S. (1975), Phys. Lett. 54 A, 353. Binder, K. (1984 c), Phys. Rev. A 29, 341.
Alexander, S., Chaikin, P. M., Grant, P., Morales, G. Binder, K. (1986), in: Atomic Transport and Defects
J., Pincus, P., Hone, D. (1984), J. Chem. Phys. 80, in Metals by Neutron Scattering: Janot, C., Petry,
5776. T., Richter, D., Springer, T. (Eds.). Berlin: Sprin-
Als-Nielsen, J. (1976), in: Phase Transitions and ger, p. 12.
Critical Phenomena, Vol. 5 A: Domb, C., Green, Binder, K. (1987 a), Rep. Progr. Phys. 50, 783.
M. S. (Eds.). New York: Academic, p. 88. Binder, K. (1987 b), Ferroelectrics 73, 43.
Als-Nielsen, J., Laursen, I. (1980), in: Ordering of Binder, K. (1989), in: Alloy Phase Stability: Gonis,
Strongly Fluctuating Condensed Matter Systems: A., Stocks, L. M. (Eds.). Dordrecht: Kluwer Acad.
Riste, T. (Ed.). New York: Plenum Press, p. 39. Publ., p. 232.
Amit, D. J. (1984), Field Theory, the Renormaliza- Binder, K. (1994), Adv. Polymer Sci. 112, 181.
tion Group and Critical Phenomena. Singapore: Binder, K. (1998), in: Spin Glasses and Random
World Scientific. Fields: Young, A. P. (Ed.). Singapore: World Sci-
Angell, C. A., Goldstein, M. (Eds.) (1986), Dynamic entific, p. 99.
Aspects of Structure Change in Liquids and Binder, K., Fratzl, P. (2001), in: Phase Transforma-
Glasses. New York: N.Y. Acad. of Sciences. tions in Materials: Kostorz, G. (Ed.). Weinheim:
Ausloos, M., Elliott, R. J. (1983), Magnetic Phase Wiley-VCH, Chap. 6.
Transitions. Berlin: Springer Verlag. Binder, K., Heermann, D. W. (1988), Monte Carlo
Bak, P., Domany, E. (1979), Phys. Rev. B 20, 2818. Simulation in Statistical Physics: An Introduction.
Bak, P., Mukamel, D. (1979), Phys. Rev. B 19, 1604. Berlin: Springer.
Bak, P., Mukamel, D., Villain, J., Wentowska, K. Binder, K. Landau, D. P. (1981), Surface Sci. 108, 503.
(1979), Phys. Rev. B 19, 1610. Binder, K., Landau, D. P. (1989), in: Molecule – Sur-
Bakker, A. F., Bruin, C., Hilhorst, H. J. (1984), Phys. face Interactions: Lawley, K. P. (Ed.). New York:
Rev. Let. 52, 449. Wiley & Sons, p. 91.
Bardham, P., Cohen, J. B. (1976), Acta Cryst. A 32, Binder, K., Stauffer, D. (1976 a), Z. Phys. B 24,
597. 407.
Bates, S. F., Rosedale, J. H., Fredrickson, G. H., Binder, K., Stauffer, D. (1976 b), Adv. Phys. 25, 343.
Glinka, C. J. (1988), Phys. Rev. Lett. 61, 229. Binder, K., Young, A. P. (1986), Rev. Mod. Phys. 58,
Baus, M. (1987), J. Stat. Phys. 48, 1129. 801.
Baus, M., Rull, L. F., Ryckaert, J. P. (Eds.) (1995), Binder, K., Meißner, G., Mais, H. (1976), Phys. Rev.
Observation, Prediction and Simulation of Phase B 15, 267.
Transitions in Complex Fluids. Dordrecht: Kluwer Binder, K., Kinzel, W., Stauffer, D. (1979), Z. Phys.
Acad. Publ. B 36, 162.
Baxter, R. J. (1973), J. Phys. C 6, L445. Binder, K. Reger, J. D. (1992), Adv. Phys. 41, 547.
4.6 References 305
Binder, K., Ciccotti, G. (Eds.) (1996), Monte Carlo Cusack, N. E. (1987), The Physics of Structurally
and Molecular Dynamics of Condensed Matter. Disordered Matter: An Introduction. Bristol:
Bologna: Società Italiana di Fisica. Adam Hilger.
Blinc, R., Levanyuk, A. P. (1986), Incommensurate De Fontaine, D. (1975), Acta Met. 23, 553.
Phases in Dielectrics, Vols. 1, 2. Amsterdam: De Fontaine, D. (1979), in: Solid State Physics, Vol.
North-Holland. 34: Ehrenreich, H., Seitz, F., Turnbull, D. (Eds.).
Blinc, R., Zeks, B. (1974), Soft Modes in Ferro- London: Academic, p. 73.
electrics and Antiferroelectrics. Amsterdam: De Fontaine, D., Kulik, J. (1985), Acta Metall. 33, 145.
North-Holland. De Gennes, P. G. (1974), The Physics of Liquid Crys-
Bloch, D., Mauri, R. (1973), Phys. Rev. B 7, 4883. tals. Oxford: Clarendon Press.
Blöte, H. W., Swendsen, R. H. (1979), Phys. Rev. De Gennes, P. G. (1979), Scaling Concepts in Poly-
Lett. 43, 799. mer Physics. Ithaca: Cornell University Press.
Blume, M. (1966), Phys. Rev. 141, 517. De Jongh, J. L., Miedema, A. R. (1974), Adv. Phys.
Blume, M., Emery, V., Griffiths, R. B (1971), Phys. 23, 1.
Rev. A 4, 1971. Delaey, L. (2001), in: Phase Transformations in Ma-
Bohn, H. G., Zinn, W., Dorner, B., Kollmar, A. terials: Kostorz, G. (Ed.). Weinheim: Wiley-VCH,
(1980), Phys. Rev. B 22, 5447. Chap. 9.
Born, M., Huang, K. (1954), Dynamical Theory of Den Nijs, M. P. M. (1979), J. Phys. A 12, 1857.
Crystal Lattices. Oxford: Oxford Univ. Press. De Raedt, H., Lagendijk, A. (1985), Phys. Repts.
Boyer, L. L. (1980), Phys. Rev. Lett. 45, 1858. 127, 235.
Boyer, L. L. (1981 a), Phys. Rev. Lett. 56, 1172. De Raedt, B., Binder, K., Michel, K. H. (1981), J.
Boyer, L. L. (1981 b), Ferroelectrics 35, 83. Chem. Phys. 75, 2977.
Boyer, L. L., Hardy, J. R. (1981), Phys. Rev. B 24, Diep, H. T., Ghazali, A., Berge, B., Lallemand, P.
2577. (1986), Europhys. Lett. 2, 603.
Bragg, W. L., Williams, E. J (1934), Proc. Roy. Soc. Dietrich, S. (1988), in: Phase Transitions and Criti-
A 145, 699. cal Phenomena, Vol. 12: Domb, C., Lebowitz, J. L.
Brazovskii, S. A. (1975), Soviet Phys. JETP 41, 85. (Eds.). New York: Academic, p. 1.
Bretz, M. (1977), Phys. Rev. Lett. 38, 501. Dillon, J. F., Chen, E. Y., Guggenheim, H. J. (1978),
Broddin, D., Van Tendeloo, G., Van Landuyt, J., Phys. Rev. B 18, 377.
Amelinckx, S., Portier, R., Guymont, M., Loiseau, Di Tolla, D. F., Tosatti, E., Ercolessi, F. (1996), in:
A. (1986), Phil. Mag. A 54, 395. Monte Carlo and Molecular Dynamics of Con-
Brout, R. (1959), Phys. Rev. 115, 824. densed Matter Systems: Binder, K., Ciccotti, G.
Brout, R. (1965), Phase Transitions. New York: Ben- (Eds.). Bologna: Società Italiana di Fisica, p. 345.
jamin Press. Domany, E., Schick, M., Walker, J. S., Griffiths, R.
Bruce, A. D., Aharony, A. (1975), Phys. Rev. B 11, B. (1978), Phys. Rev. B 18, 2209.
478. Domany, E., Shnidman, Y., Mukamel, D. (1982), J.
Bruce, A. D., Cowley, R. A. (1981), Structural Phase Phys. C 15, L495.
Transitions. London: Taylor and Francis. Domb, C., Green, M. S. (Eds.) (1972 – 1976), Phase
Burkhardt, T. W., Kinzel, W. (1979), Phys. Rev. B 20, Transitions and Critical Phenomena, Vols. 1 – 6.
4730. New York: Academic.
Burkhardt, T. W., van Leeuwen, J. M. J. (Eds.) (1982), Domb, C, Green, M. S. (Eds.) (1983 – 1997), Phase
Real-Space Renormalization. Berlin: Springer. Transitions and Critical Phenomena, Vols. 6 – 17.
Buzaré, J. Y., Fayet, J. C., Berlinger, W., Müller, K. New York: Academic.
A. (1979), Phys. Rev. Lett. 42, 465. Dünweg, B., Binder, K. (1987), Phys. Rev. B 36, 6935.
Cahn, J. W., Hilliard, J. E. (1958), J. Chem. Phys. 28, Edwards, S. F., Anderson, P. W. (1975), J. Phys. F 5,
258. 965.
Cahn, J. W., Hilliard, J. E. (1959), J. Chem. Phys. 31, Enz, C. P. (Ed.) (1979), Dynamic Critical Phenom-
688. ena and Related Topics. Proceedings. Berlin:
Ceperley, D. M. (1995), Rev. Mod. Phys. 67, 279. Springer Verlag (Springer Lecture Notes in Phys-
Challa, M. S. S., Binder, K., Landau, D. P. (1986), ics, Vol. 104).
Phys. Rev. B 34, 1841. Ferrell, R. A., Menyhard, N., Schmidt, H., Schwabl,
Chandrasekhar, S. (1992), Liquid Crystals. Cam- F., Szepfalusy, P. (1967), Phys. Rev. Lett. 18, 891.
bridge: Cambridge University Press. Finel, A. (1994), in: Static and Dynamics of Alloy
Ciccotti, G., Frenkel, D., McDonald, I. R. (Eds.) Phase Transformations: Turchi, P. E. A., Gonis, A.
(1987), Simulations of Liquids and Solids. Amster- (Eds.). New York: Plenum Press, p. 495.
dam: North-Holland. Fisher, M. E. (1968), Phys. Rev. 176, 257.
Clapp, P. C., Moss, S. C. (1966), Phys. Rev. 142, 418. Fisher, M. E. (1971), in: Critical Phenomena: Green
Clapp, P. C., Moss, S. C. (1968), Phys. Rev. 171, 754. M. S. (Ed.). London: Academic, p. 1.
Cowley, J. M. (1950), Phys. Rev. 77, 669. Fisher, M. E. (1974), Rev. Mod. Phys. 46, 587.
Cowley, J. M. (1976), Phys. Rev. B 13, 4877. Fisher, M. E. (1975 a), AIP Conf. Proc. 24, 273.
Fisher, M. E. (1975 b), Phys. Rev. Lett. 34, 1634. Halperin, B. I., Hohenberg, P. C. (1967), Phys. Rev.
Fisher, M. E., Langer, J. S. (1968), Phys. Rev. Lett. Lett. 19, 700.
20, 665. Halperin, B. I., Varma, C. (1976), Phys. Rev. B14, 4030.
Fisher, M. E., Nelson, D. R. (1974), Phys. Rev. Lett. Halperin, B. I., Hohenberg, P. C., Ma, S. K. (1974),
32, 1350. Phys. Rev. B 10, 139.
Fisher, M. E., Racz, Z. (1976), Phys. Rev. B 13, 5039. Halperin, B. I., Hohenberg, P. C., Siggia, E. (1976),
Fishman, S., Aharony, A. (1979), J. Phys. C 12, L 729. Phys. Rev. B 13, 4119.
Flory, P. J. (1953), Principles of Polymer Chemistry. Hansen, M. (1958), Constitution of Binary Alloys.
Ithaca: Cornell University Press. New York: McGraw Hill.
Folk, R., Iro, H., Schwabl, F. (1976), Z. Phys. B 25, 69. Harris, A. B. (1974), J. Phys. C 7, 1671.
Franz, S., Parisi, G. (1998), Physica A 261, 317. Harris, A. B., Meyer, H. (1985), Can. J. Phys. 63, 3.
Frazer, B. C., Shirane, G., Cox, D. E. (1965), Phys. Harris, R., Plischke, M., Zuckermann, M. J. (1973),
Rev. 140 A, 1448. Phys. Rev. Lett. 31, 160.
Fredrickson, G. H., Binder, K. (1989), J. Chem. Heermann, D. W., Klein, W., Stauffer, D. (1982),
Phys. 91, 7265. Phys. Rev. Lett. 49, 1262.
Fredrickson, G. H., Helfand, E. (1987), J. Chem. Herrmann, H. J., Janke, W., Karsch, F. (Eds.) (1992),
Phys. 87, 697. Dynamic of First Order Phase Transitions. Singa-
Friedmann, L. F., Tunstall, D. P., (Eds.) (1978), The pore: World Scientific.
Metal – Non Metal Transition in Disordered Hockney, R. W., Eastwood, J. W. (1988), Computer
Systems. Edinburgh: SUSSP. Simulation Using Particles. Bristol: Adam Hilger.
Fujimoto, M. (1997), The Physics of Structural Höchli, U. T., Knorr, K., Loidl, A. (1990), Adv. Phys.
Phase Transitions. Berlin: Springer. 39, 405.
Gahn, U. (1986), J. Phys. Chem. Solids 47, 1153. Hohenberg, P. C., Halperin, B. I. (1977), Rev. Mod.
Gebhardt, W., Krey, U. (1979), Phasenübergänge Phys. 49, 435.
und kritische Phänomene – Eine Einführung. Hoover, W. G. (1987), Molecular Dynamics. Berlin:
Braunschweig: Vieweg. Springer.
Gehring, G. A. (1977), J. Phys. C 10, 531. Hornreich, R. M., Luban, M., Shtrikman, S. (1975),
Gehring, G. A., Gehring, K. A. (1975), Rep. Progr. Phys. Rev. Lett. 35, 1678.
Phys. 38, 1. Hulliger, F., Siegrist, T. (1979), Z. Phys. B 35, 81.
Gerold, V., Kern, J. (1986), in: Atomic Transport and Imbrie, J. Z. (1984), Phys. Rev. Lett. 53, 1747.
Defects in Metals by Neutron Scattering: Janot, C., Imry, Y. (1980), Phys. Rev. B 21, 2042.
Petry, T., Richter, D., Springer, T. (Eds.). Berlin: Imry, Y., Ma, S. K. (1975), Phys. Rev. Lett. 35, 1399.
Springer Verlag, p. 17. Imry, Y., Wortis, M. (1979), Phys. Rev. B 19, 3580.
Gerold, V., Kern, J. (1987), Acta Met. 35, 393. Inden, G. (2001), in: Phase Transformations in Mate-
Ginzburg, V. L. (1960), Sov. Phys.-Solid State 2, rials: Kostorz, G. (Ed.). Weinheim: Wiley-VCH,
1824. Chap. 8.
Giordano, N., Wolf, W. P. (1975), Phys. Rev. Lett. 35, Jäckle, J. (1986), Rep. Progr. Phys. 49, 171.
799. Jaster, A. (1998), Europhys. Lett. 42, 277.
Golosov, N. S., Popov, L. E., Rudan, L. Y. (1973), Jona, F., Shirane, G. (1962), Ferroelectric Crystals.
J. Phys. Chem. Solids 34, 1149 – 1159. New York: Pergamon Press.
Gonis, A., Stocks, L. M. (Eds.) (1989), Alloy Phase Joyce, J. S. (1972), in: Phase Transitions and Criti-
Stability. Dordrecht: Kluwer Acad. Publ. cal Phenomena, Vol. 2: Domb, C., Green, M. S.
Grazhdankina, N. P. (1969), Sov. Phys.-Usp. 11, 727; (Eds.). New York: Academic, Chapter X.
Usp. Fiz. Nauk. 96, 291. Kämmerer, S., Dünweg, B., Binder, K., D’Onorio De
Grinstein, G. (1985), in: Fundamental Problems in Meo, M. (1996), Phys. Rev. B 53, 2345.
Statistical Mechanics, VI: Cohen, E. G. D. (Ed.). Kalos, M. H. (Ed.) (1985), Monte Carlo Methods in
Amsterdam, North-Holland, p. 147. Quantum Problems. Dordrecht: Kluwer Acad. Publ.
Guerard, D., Herold, A. (1975), Carbon 13, 337. Kalos, M. H., Schmidt, K. E. (1984), in: Applications
Gunton, J. D., Droz, M. (1983), Introduction to the of the Monte Carlo Method in Statistical Physics:
Theory of Metastable and Unstable States. Berlin: Binder, K. (Ed.). Berlin: Springer Verlag, p. 125.
Springer. Kaplan, T. A. (1959), Phys. Rev. 116, 888.
Gunton, J. D., San Miguel, M., Sahni, P. S. (1983), Keating, D. T., Warren, B. E. (1951), J. Appl. Phys.
in: Phase Transitions and Critical Phenomena, 22, 286.
Vol. 8: Domb, C., Lebowitz, J. L. (Eds.). London: Kehr, K. W., Binder, K., Reulein, S. M. (1989), Phys.
Academic, p. 267. Rev. B 39, 4891.
Guymont, M. (1978), Phys. Rev. B 18, 5385. Khachaturyan, A. G. (1962), Fiz. Metallov i Metallo-
Guymont, M. (1981), Phys. Rev. B 24, 2647. vedenie 13, 493.
Haasen, P., Gerold, V., Wagner, R., Ashby, M. F. Khachaturyan, A. G. (1963), Fiz. tverd. Tela 5, 26;
(Eds.) (1984), Decomposition of Alloys: The Early ibid 5, 750.
Stages. New York: Pergamon Press. Khachaturyan, A. G. (1973), Phys. stat. sol. (b) 60, 9.
4.6 References 307
Khachaturyan, A. G. (1983), Theory of Structural Michel, K. H., Naudts, J. (1977), J. Chem. Phys. 67,
Transformation in Solids. New York: J. Wiley & 547.
Sons. Michel, K. H., Naudts, J. (1978), J. Chem. Phys. 68,
Kikuchi, R. (1951), Phys. Rev. 81, 998. 216.
Kikuchi, R. (1967), J. Chem. Phys. 47, 1664. Milchev, A., Heermann, D. W., Binder, K. (1986),
Kittel, C. (1967), Quantum Theory of Solids. New J. Stat. Phys. 44, 749.
York: J. Wiley & Sons. Moncton, D. E., Axe, J. D., Di Salvo, F. J. (1977),
Kittler, R. C., Falicov, L. M. (1978), J. Phys. B18, 2506. Phys. Rev. B 16, 801.
Kittler, R. C., Falicov, L. M. (1979), Phys. Rev. B 19, Morita, T. (1972), J. Math. Phys. 13, 115.
291. Moss, S. C. (1964), J. Appl. Phys. 35, 3547.
Kleemann, W., Schäfer, F. J., Tannhäuser, D. S. Moss, S. C., Clapp, P. C. (1968), Phys. Rev. 171, 764.
(1980), J. Magn. Magnet. Mater. 15 – 18, 415. Mott, N. F. (1974), Metal-Insulator Transitions. Lon-
Klein, W., Unger, C. (1983), Phys. Rev. B 28, 445. don: Taylor and Francis.
Koch, S. W. (1984), Dynamics of First-Ordered Mouritsen, O. G. (1984), Computer Studies of Phase
Phase Transitions in Equilibrium and Nonequilib- Transitions and Critical Phenomena. Berlin:
rium. Berlin: Springer. Springer.
Kötzler, J. (1984), Z. Phys. B 55, 119. Müller, K. A., Berlinger, W. (1975), Phys. Rev. Lett.
Kötzler, J., Raffius, G., Loidl, A., Zeyen, C. M. E. 22, 1547.
(1979). Z. Phys. B 35, 125. Müller-Krumbhaar, H. (1977), in: Crystal Growth
Kosterlitz, J. M., Thouless, D. J. (1973), J. Phys. C 6, and Materials: Kaldis, E. (Ed.). Amsterdam:
1181. North-Holland, p. 79.
Kremer, K., Robbins, M. O., Grest, G. S. (1986), Mukamel, D., Krinsky, S. (1976), Phys. Rev. B 13,
Phys. Rev. Lett. 57, 2694. 5065.
Kremer, K., Grest, G. S., Robbins, O. (1987), J. Mukamel, D., Fisher, M. E., Domany, E. (1976 a),
Phys. A 20, L181. Phys. Rev. Lett. 37, 565.
Krinsky, S., Mukamel, D. (1977), Phys. Rev. B 16, Mukamel, D., Krinsky, S., Bak, P. (1976 b), AIP
2313. Conf. Proc. 29, 474.
Krivoglaz, M. A. (1969), Theory of X-Ray and Ther- Nattermann, T. (1998), in: Spin Classes and Random
mal Neutron Scattering by Real Crystals. New Fields: Young, A. P. (Ed.). Singapore: World Sci-
York: Plenum Press. entific, p. 277.
Landau, D. P., Binder, K. (1978), Phys. Rev. B17, 2328. Nelson, D. R., Halperin, B. I. (1979), Phys. Rev. B19,
Landau, D. P., Binder, K. (1985), Phys. Rev. B31, 5946. 2457.
Landau, L. D., Lifshitz, E. M. (1958), Statistical Nienhuis, B., Riedel, E. K., Schick, M. (1980), J.
Physics. Oxford: Pergamon Press. Phys. A 13, L189.
Langer, J. S. (1974). Physica 73, 61. Nishiyama, Z. (1979), Martensitic Transformation.
Larkin, A. I., Khmelnitskii, D. E. (1969), Zh. Eksp. New York: Academic.
Fiz. 56, 2087; Sov. Phys. JETP 29, 1123. Onsager, L. (1944), Phys. Rev. 65, 117.
Lebowitz, J. L., Penrose, O. (1966), J. Math. Phys. 7, 98. Orr, R. L. (1960), Acta Met. 8, 489.
LeGuillou, J. C., Zinn-Justin, J. (1980), Phys. Rev. Orr, R. L., Luciat-Labry, J., Hultgren, R. (1960), Acta
B 21, 3976. Met. 8, 431.
Leibler, L. (1980), Macromolecules 13, 1602. Parrinello, M., Rahman, A. (1980), Phys. Rev. Lett.
Lifshitz, E. M. (1942), J. Physique 6, 61. 45, 1196.
Lipowsky, R. (1984), J. Appl. Phys. 55, 213. Parrinello, M., Rahman, A., Vashishta, P. (1983),
Loidl, A. (1989), Ann. Rev. Phys. Chem. 40, 29. Phys. Rev. Lett. 50, 1073.
Loiseau, A., Van Tendeloo, G., Portier, R., Ducas- Patashinskii, A. Z., Pokrovskii, V. I. (1979), Fluctua-
telle, F. (1985), J. Physique 46, 595. tion Theory of Phase Transitions. Oxford: Per-
Lynden-Bell, R. M., Michel, K. H. (1994), Rev. Mod. gamon Press.
Phys. 66, 721. Pershan, P. S. (1988), Structure of Liquid Crystal
Ma, S.-K. (1976), Modern Theory of Phase Transi- Phases. Singapore: World Scientific.
tions. Reading: Benjamin Press. Piercy, P., Pfnür, H. (1987), Phys. Rev. Lett. 59, 1124.
Maletta, H., Felsch, W. (1979), Phys. Rev. B 20, 1079. Pimenta, M. A., Echegut, P., Luspin, Y., Hauset, G.,
Mandelbrot, B. B. (1982), The Fractal Geometry of Gervais, F., Abélard, P. (1989), Phys. Rev. B 39,
Nature. San Francisco: Freeman. 3361.
Marx, R. (1989), Phys. Rev. B 40, 2585. Potts, R. B. (1952), Proc. Camb. Phil. Soc. 48, 106.
McCoy, B. M., Wu, T. T. (1973), The Two-Dimen- Privman, V. (Ed.) (1990), Finite Size Scaling and Nu-
sional Ising Model. Cambridge, Mass: Harvard merical Simulation. Singapore: World Scientific.
University Press. Privman, V., Hohenberg, P. C., Aharony, A. (1991),
McGuire, T. R., Gambino, R. J., Pickart, S. J., Alpe- in: Phase Transitions and Critical Phenomena,
rin, H. A. (1969), J. Appl. Phys. 40, 1009. Vol. 14: Domb, C., Lebowitz, J. L. (Eds.). London:
Meißner, G., Binder, K. (1975), Phys. Rev. B 12, 3948. Academic Press, Chapter 1.
Pynn, T., Skjeltorp, A. (Eds.) (1983), Multicritical Suzuki, M. (Ed.) (1992), Quantum Monte Carlo
Phenomena. New York: Plenum Press. Methods in Condensed Matter Physics. Singapore:
Riste, T., Sherrington, D. (Eds.) (1989), Phase Tran- World Scientific.
sitions in Soft Condensed Matter. New York: Ple- Swann, P. S., Duff, W. R., Fisher, R. M. (1972),
num Press. Metallurg. Trans. 3, 409.
Robbins, M. O., Kremer, K., Grest, G. S. (1988), J. Swendsen, R. H., Krinsky, S. (1979), Phys. Rev. Lett.
Chem. Phys. 88, 3286. 43, 177.
Rohrer, H. (1975), Phys. Rev. Lett. 34, 1638. Tanisaki, S. (1961), Phys. Soc. Japan 16, 579.
Rohrer, H., Gerber, Ch. (1977), Phys. Rev. Lett. 38, 909. Tolédano, J. C. (1981), Ferroelectrics 35, 31.
Roth, W. L. (1958), Phys. Rev. 110, 1333. Tolédano, P., Pascoli, G. (1980), Ferroelectrics 25,
Sadiq, A., Binder, K. (1984), J. Stat. Phys. 35, 517. 427.
Salanon, B., Fabre, F., Lapoujoulade, J., Selke, W. Tolédano, J. C., Tolédano, P. (1987), The Landau
(1988), Phys. Rev. B 38, 7385. Theory of Phase Transitions. Singapore: World
Salje, E. (1990), Phase Transitions in Ferroelastic Scientific.
and Co-elastic Crystals. Cambridge: Cambridge Tsakalakos, T. (1984), Phase Transformations in Sol-
Univ. Press. ids. Amsterdam: North-Holland.
Sanchez, J. M., De Fontaine, D. (1980), Phys. Rev. Turchi, P. E. A., Gonis, A. (Eds.) (1994), Statics and
B 21, 216. Dynamics of Alloy Phase Transformations. New
Sanchez, J. M., De Fontaine, D. (1982), Phys. Rev. York: Plenum Press.
B 25, 1759. van Hove, L. (1954), Phys. Rev. 93, 1374.
Schick, M. (1981), Progr. Surf. Sci. 11, 245. Villain, J. (1959), J. Phys. Chem. Solids 11, 303.
Schmid, F. (1999), in: Computational Methods in Villain, J. (1985), in: Scaling Phenomena in Disor-
Colloid and Interface Science: Borowko, M. (Ed.). dered Systems: Pynn, R., Skjeltorp, A. (Eds.). New
New York: M. Dekker, Chapter 10. York: Plenum Press, p. 423.
Schmidt, V. H. (1978), in: High Pressure and Low Wagner, H., Horner, H. (1974), Adv. Phys. 23, 587.
Temperature Physics: Chu, C. W., Wollam, J. A. Wagner, R., Kampmann, R., Vorhees, P. W. (2001),
(Eds.). New York: Plenum Press, p. 237. in: Phase Transformations in Materials: Kostorz,
Schneider, T., Stoll, E. (1976), Phys. Rev. B 13, 1216. G. (Ed.). Weinheim: Wiley-VCH, Chap. 5.
Schwartz, L. H., Cohen, J. B. (1965), J. Appl. Phys. Wansleben, S., Landau, D. P. (1987), J. Appl. Phys.
36, 598. 61, 3968.
Schweika, W. (1989), in: Alloy Phase Stability: Stocks, Weeks, J. D. (1980), in: Ordering in Strongly Fluctu-
G. M., Gonis, A. (Eds.). Dordrecht: Kluwer, p. 137. ating Condensed Matter Systems: Riste, T. (Ed.).
Schweika, W. (1994), in: Statics and Dynamics of New York: Plenum Press, p. 293.
Alloy Phase Transformations: Turchi, P. E. A., Wiechert, H., Arlt, S. A. (1993), Phys. Rev. Lett. 71,
Gonis, A. (Eds.). New York: Plenum Press, p. 103. 2090.
Schweika, W., Hauboldt, H.-G. (1986), in: Atomic Wilding, N. B. (1997), J. Phys.: Condens. Matter 9,
Transport and Defects in Metals by Neutron Scat- 585.
tering: Janot, C., Petry, T., Richter, D., Springer, T. Wilding, N. B., Schmid, F., Nielaba, P. (1998), Phys.
(Eds.). Berlin: Springer Verlag, p. 22. Rev. E 58, 2201.
Scott, J. F. (1987), in: Phase Transitions and Critical Wilson, K. G., Kogut, J. (1974), Phys. Repts. 12 C, 75.
Phenomena: Domb, C., Lebowitz, J. L. (Eds.). Wu, F. Y. (1982), Rev. Mod. Phys. 54, 235.
London: Academic Press, Chapter 2. Yelon, W. B., Cox, D. E., Kortmann, P. J. (1974),
Selke, W. (1988), Phys. Repts. 170, 213. Phys. Rev. B 9, 4843.
Selke, W. (1989), in: Alloy Phase Stability: Gonis, A., Yeomans, J. (1992), Statistical Mechanics of Phase
Stocks, L. M. (Eds.). Dordrecht: Kluwer Akad. Publ. Transitions. Oxford: Clarendon Press.
Selke, W. (1992), in: Phase Transitions and Critical Young, A. F. (Ed.) (1998), Spin Glasses and Random
Phenomena, Vol. 15: Domb, C., Lebowitz, J. L. Fields. Singapore: World Scientific.
(Eds.). London: Academic Press, p. 1. Zallen, T. (1983), The Physics of Amorphous Solids.
Semenovskaya, S. V. (1974), Phys. Stat. Sol. (b) 64, New York: J. Wiley and Sons.
291. Zarzycki, J. (Ed.) (1991), Glasses and Amorphous
Smart, J. S. (1966), Effective Field Theories of Mag- Materials (Materials Science and Technology,
netism. New York: W. B. Saunders. Vol. 9). Weinheim: VCH.
Stanley, H. E. (1971), Introduction to Phase Transi- Zettlemoyer, A. C. (Ed.) (1969), Nucleation. New
tions and Critical Phenomena. Oxford: Oxford York: Marcel Dekker.
University Press. Zunger, A. (1994), in: Statics and Dynamics of Alloy
Stauffer, D., Aharony, A. (1992), An Introduction to Phase Transformations: Turchi, P. E. A., Gonis, A.
Percolation Theory. London: Taylor and Francis. (Eds.). New York: Plenum Press, p. 361.
Sullivan, D. E., Telo da Gama, M. M. (1985), in:
Fluid Interfacial Phenomena: Croxton, C. A.
(Ed.). New York: J. Wiley and Sons, p. 45.
5 Homogeneous Second-Phase Precipitation
Richard Wagner
Forschungszentrum Jülich GmbH, Jülich, Germany
Reinhard Kampmann
Institut für Werkstofforschung, GKSS-Forschungszentrum GmbH, Geesthacht, Germany
Peter W. Voorhees
Department of Materials Science and Engineering, Northwestern University, Evanston,

Ill., USA

5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 314
5.2 General Considerations . . . . . . . . . . . . . . . . . . . . . . . . . . 315
5.2.1 General Course of an Isothermal Precipitation Reaction . . . . . . . . . . 315
5.2.2 Thermodynamic Considerations – Metastability and Instability . . . . . . 317
5.2.3 Decomposition Mechanisms: Nucleation and Growth versus Spinodal
Decomposition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 319
5.2.4 Thermodynamic Driving Forces for Phase Separation . . . . . . . . . . . 322
5.3 Experimental Techniques for Studying Decomposition Kinetics . . . . 326
5.3.1 Microanalytical Tools . . . . . . . . . . . . . . . . . . . . . . . . . . . . 326
5.3.1.1 Direct Imaging Techniques . . . . . . . . . . . . . . . . . . . . . . . . . 326
5.3.1.2 Scattering Techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . 328
5.3.2 Experimental Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . 330
5.3.2.1 Influence of Quenching Rate on Kinetics . . . . . . . . . . . . . . . . . . 330
5.3.2.2 Distinction of the Mode of Decomposition . . . . . . . . . . . . . . . . . 332
5.4 Precipitate Morphologies . . . . . . . . . . . . . . . . . . . . . . . . . . 334
5.4.1 Experimental Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . 334
5.4.2 Factors Controlling the Shapes and Morphologies of Precipitates . . . . . 336
5.5 Early Stage Decomposition Kinetics . . . . . . . . . . . . . . . . . . . . 339
5.5.1 Cluster-Kinetics Approach . . . . . . . . . . . . . . . . . . . . . . . . . . 340
5.5.1.1 Classical Nucleation – Sharp Interface Model . . . . . . . . . . . . . . . . 340
5.5.1.2 Time-Dependent Nucleation Rate . . . . . . . . . . . . . . . . . . . . . . 343
5.5.1.3 Experimental Assessment of Classical Nucleation Theory . . . . . . . . . 345
5.5.1.4 Non-Classical Nucleation – Diffuse Interface Model . . . . . . . . . . . . 347
5.5.1.5 Distinction Between Classical and Non-Classical Nucleation . . . . . . . . 349
5.5.2 Diffusion-Controlled Growth of Nuclei from the Supersaturated Matrix . . 350

ISBN: 3-527-30256-5
310 5 Homogeneous Second-Phase Precipitation
5.5.3 The Cluster-Dynamics Approach to Generalized Nucleation Theory . . . . 352

5.5.4 Spinodal Theories . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 356
5.5.5 The Philosophy of Defining of ‘Spinodol Alloy’ – Morphologies of
‘Spinodal Alloys’ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 363
5.5.6 Monte Carlo Studies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 367
5.6 Coarsening of Precipitates . . . . . . . . . . . . . . . . . . . . . . . . . 370
5.6.1 General Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 370
5.6.2 The LSW Theory of Coarsening . . . . . . . . . . . . . . . . . . . . . . . 370
5.6.3 Extensions of the Coarsening Theory to Finite Precipitate Volume Fractions 373
5.6.4 Other Approaches Towards Coarsening . . . . . . . . . . . . . . . . . . . 377
5.6.5 Influence of Coherency Strains on the Mechanism and Kinetics of
Coarsening – Particle Splitting . . . . . . . . . . . . . . . . . . . . . . . 377
5.7 Numerical Approaches Treating Nucleation, Growth and Coarsening as
Concomitant Processes . . . . . . . . . . . . . . . . . . . . . . . . . . . 381
5.7.1 General Remarks on the Interpretation of Experimental Kinetic Data
of Early Decomposition Stages . . . . . . . . . . . . . . . . . . . . . . . 381
5.7.2 The Langer and Schwartz Theory (LS Model) and its Modification by
Kampmann and Wagner (MLS Model) . . . . . . . . . . . . . . . . . . . 383
5.7.3 The Numerical Modell (N Model) of Kampmann and Wagner (KW) . . . . 385
5.7.4 Decomposition of a Homogeneous Solid Solution . . . . . . . . . . . . . 385
5.7.4.1 General Course of Decomposition . . . . . . . . . . . . . . . . . . . . . . 385
5.7.4.2 Comparison Between the MLS Model and the N Model . . . . . . . . . . 387
5.7.4.3 The Appearance and Experimental Identification of the Growth
and Coarsening Stages . . . . . . . . . . . . . . . . . . . . . . . . . . . . 387
5.7.4.4 Extraction of the Interfacial Energy and the Diffusion Constant from
Experimental Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 389
5.7.5 Decomposition Kinetics in Alloys Pre-Decomposed During Quenching . . 391
5.7.6 Influence of the Loss of Particle Coherency on the Precipitation Kinetics . 392
5.7.7 Combined Cluster-Dynamic and Deterministic Description
of Decomposition Kinetics . . . . . . . . . . . . . . . . . . . . . . . . . . 394
5.8 Self-Similarity, Dynamical Scaling and Power-Law Approximations . . 395
5.8.1 Dynamical Scaling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 395
5.8.2 Power-Law Approximations . . . . . . . . . . . . . . . . . . . . . . . . . 398
5.9 Non-Isothermal Precipitation Reactions . . . . . . . . . . . . . . . . . 401
5.11 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 402

a lattice parameter
A solvent atoms
aij(c) activity of atomic species i in phase j at composition c
B solute atoms
c composition
cA atomic fraction of A
c0 initial composition
r
c mean composition
Dc supersaturation or composition difference
cae equilibrium composition of matrix phase a
c¢ae composition of matrix a in metastable equilibrium with coherent precipi-
tates
cbe equilibrium composition of precipitate phase b
cp composition of incoherent particles
c¢p composition of coherent particles
cij elastic constant
cR composition of matrix at the matrix/particle interface
C (i) equilibrium cluster distribution
–
D mean diameter of precipitated particles
E Young’s modulus
f (R) size distribution function of precipitates with radii R
F (c), f (c) Helmholtz energy or energy density
F¢(c), f ¢(c) Helmholtz constraint energy or constraint energy density
DF * nucleation barrier
DFel elastic free energy
DFa/b interfacial free energy
DFch (c) chemical driving force
fp precipitated volume fraction
F̃ (x) time-independent scaling function
G Gibbs energy
G(| r – r0 |) two-point correlation function at spatial positions r, r0
H enthalpy
i, i* number of atoms in a cluster or in a cluster of critical size
J s, J* steady state and time-dependent nucleation rate
K aspect ratio of an ellipsoid of revolution
K* gradient energy coefficient in the CH spinodal theory
KRLSW coarsening rate according to the LSW theory
k Boltzmann constant
L ratio of elastic to interfacial energy
M atomic mobility
Nv number density of precipitates
nv number of atoms per unit volume
p pressure
R radius of precipitates
–
R mean radius of precipitates
R* radius of a critical nucleus or a particle being in unstable equilibrium
with the matrix
Rg molar gas constant
R (k); R (l) amplification factor in the CH spinodal theory
S entropy
S(k) structure function
Sm = S (km) maximum of structure function
T temperature
TA annealing temperature
TH homogenization temperature
t aging time
U internal energy
V volume
Va , Vb molar volume of a or b phase
Z Zeldovich factor
a, a ¢ matrix phase
b equilibrium precipitated phase
b¢ metastable precipitated phase; transition phase
g shear strain
d misfit parameter
h = (1/a0) (∂a/∂c) atomic size factor
k, k scattering vector and its magnitude
km wavenumber of the maximum of the structure function S (k)
l wavelength
m shear modulus
m ij chemical potential of component i in phase j
n Poisson’s ratio
W atomic volume
s ab specific interfacial energy
t incubation period or scaled time in the MLS and KW models
AEM analytical transmission electron microscopy

AFIM analytical field ion microscopy (atom probe field ion microscope)
CH Cahn–Hilliard spinodal theory
CHC Cahn–Hilliard–Cook spinodal theory
CTEM conventional transmission electron microscopy
EDX energy-dispersive X-ray analysis
EELS electron energy loss spectroscopy
FIM field ion microscopy
HREM high resolution electron microscopy
KW Kampmann–Wagner model
LBM Langer–Bar-On–Miller non-linear spinodal theory

LS Langer–Schwartz theory
LSW Lifshitz–Slyozov–Wagner theory
MCS Monte Carlo simulation
MLS modified Langer–Schwartz model
n.g. nucleation and growth
SAXS small angle X-ray scattering
SANS small angle neutron scattering
s. d. spinodal decomposition
TAP tomographic atom probe
5.1 Introduction In order to initiate a precipitation reac-

tion, the alloy is first homogenized in the
Many technologically important proper- single-phase region at TH and then either
ties of alloys, such as their mechanical a) cooled down slowly into the two-phase
strength and toughness, creep and corro- region a¢ + b, or
sion resistance, and magnetic and super- b) quenched into brine prior to isothermal
conducting properties, are essentially con- aging at a temperature TA within the
trolled by the presence of precipitated par- two-phase region (Fig. 5-1).
ticles of a second phase. This commonly
results from the decomposition of a solid In both cases, thermodynamic equilib-
solution during cooling. A fundamental rium is reached if the supersaturation Dc,
understanding of the thermodynamics, the defined as
mechanism and the kinetics of precipita- Dc (t) = c̄ – cae (5-1)
tion reactions in metallic solids, leading to
a well-defined microstructure, is therefore becomes zero. (Here c̄ (t) is the mean matrix
of great interest in materials science. composition at time t with c̄ (t = 0) ∫ c0 .)
As is reflected by the schematic binary For case a), which frequently prevails
phase diagram of Fig. 5-1, for reasons of during industrial processing, the aging
entropy the single-phase state a of a solid temperature and the associated equilibrium
solution with composition c0 is thermody- solubility limit cae (T ) decrease continu-
namically stable only at elevated tempera- ously. Equilibrium can only be reached if
tures. At lower temperatures the free en- the cooling rate is sufficiently low within a
ergy of the system is lowered through un- temperature range where the diffusion of
mixing (‘decomposition’ or ‘phase separa- the solute atoms is still adequately high.
tion’) of a into two phases, a¢ and b. The precipitated volume fraction ( fp) and
the dispersion of the particles of the second
phase can thus be controlled via the cool-
ing rate.
Procedure b) is frequently used for stud-
ies of decomposition kinetics under condi-
tions which are easier to control and de-
scribe theoretically (T = TA = const.; D =
const.) than for case a). This leads to pre-
cipitate microstructures whose volume
fraction and particle dispersion depend on
Dc (T) and the aging time t.
Decomposition reactions involve diffu-
sion of the atomic species via the vacancy
and/or the interstitial mechanism. Hence,
the precipitate microstructure proceed-
ing towards thermodynamic equilibrium
Figure 5-1. Schematic phase diagram of a binary al-
evolves as a function of both time and tem-
loy displaying a miscibility gap. Dashed lines show
the metastable coherent solvus line and a possible perature. In practice, a metallurgist is often
metastable intermetallic phase b¢. The long arrow in- requested to tailor an alloy with a specific
dicates the quenching process. precipitation microstructure. For this pur-
5.2 General Considerations 315
pose it would be desirable to have a theo- 5.2 General Considerations

retical concept, preferentially available as
a computer algorithm, that allows suitable 5.2.1 General Course of an Isothermal
processing parameters to be derived for es- Precipitation Reaction
tablishing the specific microstructure on As illustrated for a Ni–37 at.% Cu–8
grounds of the given thermodynamics of at.% Al alloy (Fig. 5-2 and 5-3), an isother-
the alloy. This less empirical approach to mal precipitation reaction (the kinetics of
alloy design, however, would require a which will be dealt with in more detail in
comprehensive theory of precipitation ki- Secs. 5.5, 5.6 and 5.7) is qualitatively char-
netics, which has not yet been developed to acterized by an early stage during which an
a satisfactory level, despite recent efforts increasing number density
and progress made in elucidating the kinet-
ics of first-order phase transitions in solids. Nv (t) = Ú f (R) dR (5-2)
Furthermore, in materials science it is fre- of more or less spherical solute-rich clus-
quently desirable to predict the kinetic evo- ters (‘particles’) with a size distribution
lution of an originally optimized precipitate f [R (t)] and a mean radius
microstructure under service conditions,
e.g., for high-temperature applications in R=
∫ R f ( R) dR (5-3)
two-phase materials, where the precipitate ∫ f ( R ) dR
distribution might undergo changes be- are formed. In common with all homogene-
cause of coarsening. Even though the ki- ous precipitation reactions studied so far, in
netics of coarsening are of great practi- the earliest stages of the reaction the parent
cal importance, a completely satisfactory phase a and the precipitate phase b share a
coarsening theory has so far only been de- common crystal lattice, i.e., the two phases
veloped in the limit of zero precipitated are coherent. As inferred from the field ion
volume fraction (see Sec. 5.6). This limit, micrographs of Fig. 5.2 a – c and from the
however, is never realized in technical al- quantitative data of Fig. 5-3, during the
–
loys, where the volume fraction of the mi- early stages R increases somewhat and the
nor phase frequently exceeds 30%. supersaturation, Dc, decreases slowly. This
In the present chapter recent theoretical small reduction in Dc, however, is sufficient
and experimental studies on the kinetics of to terminate the nucleation of new particles,
phase separation in solids are reviewed from as indicated by the maximum of Nv (t).
the point of view of the experimentalist. Beyond this maximum the precipitate
Special emphasis is placed on the questions number density decreases (Fig. 5-3) due to
to what extent the theoretical results can be the onset of the coarsening reaction, during
verified experimentally and to what extent which the smaller particles redissolve, thus
they might be of practical use to the physical enabling the larger ones to grow (see Sec.
metallurgist. We confine ourselves to deal- 5.6). During this coarsening process the
ing only with homogeneous and continuous supersaturation Dc decreases asymptoti-
phase separation mechanisms. Heterogene- cally towards zero.
ous nucleation at crystal defects, and dis- As mentioned previously, within the mis-
continuous precipitation reactions at mov- cibility gap, the solid solution a becomes un-
ing interfaces, as well as unmixing in solids stable and decomposes into the stable solid
under irradiation are treated separately, in solution a¢ and the precipitated phase b:
the chapter by Purdy and Bréchet (2001). a Æ a¢ + b
– –
a) 2 min.; 2 R = 2,4 nm b) 180 min.; 2 R = 8 nm
–
c) 420 min.; 2 R = 11 nm d)
Figure 5-2. a) – c) Neon field ion image of g ¢-precipitates (bright images in dark matrix) in Ni–36.8 at.% Cu–8
at.% Al aged for the given times at 580 °C (Liu and Wagner, 1984). d) Three-dimensional distribution of Mg and
Zn atoms in a commercial Al–5.5 at.% Mg–1.35 at.% Zn alloy analyzed with the tomographic atom probe. Each
point represents an atom. The T ¢ precipitates are enriched in Mg and Zn (Bigot et al., 1997).
The terminal solute concentration of the During the first step of the precipitation
solvent (A)-rich matrix a¢ is given by the sequence a metastable precipitate (transi-
equilibrium solubility limit cae (T ); the ter- tion phase b¢) is formed, frequently with
minal composition cp of the precipitated large associated coherency strains and a
phase is given by the solubility at TA at the small interfacial energy s ab. Often the
B-rich side of the phase diagram (Fig. 5-1). metastable phase b¢ is an intermetallic
Frequently, the interfacial energy s ab compound with a lower solute concentra-
(J/m2) between the matrix and the equilib- tion (c¢be ∫ c¢p) than the equilibrium precipi-
rium precipitate is rather high, particularly tate (cp). Because of the large coherency
if the width of the miscibility gap is large strains, the metastable (‘coherent’) solvus
(see Sec. 5.7.4.4). In this case, the decom- line (dashed line in Fig. 5-1) is shifted to-
position of a proceeds via the sequence wards higher solute concentrations. This
a Æ a≤ + b¢ Æ a¢ + b leads to a reduction of the supersaturation
Figure 5-3. Time evolution

–
of the mean radius R , the
number density Nv, and
the supersaturation Dc of
g ¢-precipitates in Ni–36
at.% Cu–9 at.% Al during
aging at 500 °C (Liu and
Wagner, 1984).
with respect to the (incoherent) equilib- 2nd step extended

rium solvus line (Fig. 5-1), and, hence, to a aging
reduction of the driving force for precipita- at T > 350 °C
tion (see Sec. 5.2.4). Often only after ex-
tended aging do the metastable phases a≤ a¢-Cu–Ti (f.c.c. solid solution) +
and b¢ transform further to the final equi- b-Cu3Ti (stable, hexagonal structure,
librium phases a¢ and b. The crystal struc- incoherent, small strain, large
ture of the equilibrium phase b finally pre- interfacial energy,
cipitated is different from that of the parent sab > 0.6 J/m2)
phase, leading to either a coherent, a semi-
coherent or a fully incoherent a¢/b interface
boundary (Gleiter, 1983). The atomic struc- 5.2.2 Thermodynamic Considerations –
ture of the latter resembles that of a high- Metastability and Instability
energy, high-angle grain boundary and, thus,
is associated with a rather large interfacial Let us consider a binary alloy consisting
energy and small elastic strain energies. of NA solvent atoms A and NB solute atoms
The decomposition of Cu–Ti alloys with B with NA + NB = N, or, in terms of atomic
Ti contents between ≈ 1 at.% and 5 at.% fractions, cA = NA/N and cB = NB/N, with
serves as an example of such a complex pre- cA + cB = 1. (The concentrations of atoms
cipitation sequence (Wagner et al., 1988): A and B are then given as cA nv and cB nv,
where nv is the number of atoms per unit
a-Cu–(1 . . . 5 at.%) Ti volume.) As only one independent variable
remains, we can refer to the composition of
1st step T ≈ 350 °C the alloy as c ∫ cB (0 ≤ c ≤ 1).
Decomposition of a supersaturated sin-
gle-phase alloy into a two-phase state com-
a≤-Cu–Ti (f.c.c. solid solution) + monly occurs at constant temperature T
b¢-Cu4Ti (metastable, coherent, body- and pressure p, and is thus prompted by a
centered-tetragonal structure, possible reduction in Gibbs energy (Gas-
large coherency strains, small kell, 1983)
interfacial energy,
sab = 0.067 J/m2) G = H – TS (5-4)
with the enthalpy

H = U + pV (5-5)
Thermodynamic equilibrium is attained
when G has reached a minimum, i.e.,
d GT, p = dU + p dV – T dS = 0 (5-6)
For phase separation in solids the term p dV
can usually be neglected with respect to the
others in Eq. (5-6). Thus a good approxi-
mation for G is given by the Helmholtz en-
ergy
F = U – TS (5-7)
which will be used in the following sec-
tions as the relevant thermodynamic func-
tion. Equilibrium is achieved if F or the Figure 5-4. Helmholtz energy F as a function of
corresponding Helmholtz energy density composition for a binary alloy with a miscibility gap.
(F per unit volume or per mole) are mini- The changes in F and the resulting driving forces for
unmixing are illustrated.
mized. Unmixing only takes place if the
transition from the single-phase state low-
ers the Helmholtz energy, i.e., by conven- tangent to the Helmholtz energy curves of
tion if D F < 0. the a- and b-phases; thus cbe – cae represents
For a given temperature, volume and so- the width of the miscibility gap at a given
lute concentration of a heterogeneous bi- temperature (cf. Fig. 5.4).
nary alloy, equilibrium between the two Fig. 5-5 a shows schematically the phase
phases a and b can only be achieved if the diagram of a binary alloy with a two-phase
concentrations of A and B in the two region at lower temperatures for T = T1; the
phases have been established such that associated Helmholtz energy versus com-
position curve, F (c), is shown in Fig. 5-5 b.
⎛ ∂F α ⎞ ⎛ ∂F β ⎞
⎜ ∂n ⎟ = ⎜ ⎟ (5-8) In the single-phase field a, F initially de-
⎝ i ⎠ T , V, n ⎝ ∂ni ⎠ T , V, n creases with increasing solute concentra-
j j
tion due to the growing entropy of mixing
In other words, at equilibrium the chemical (Eq. (5-7)). In the thermodynamically
∂F α ∂F β equilibrated two-phase region, F varies lin-
potentials m iα = and m iβ = of the
∂ni ∂ni early with c (bold straight line satisfying
component i (either A or B) in the two the equilibrium condition m ai = m bi ).
phases are identical and the two phases The mean field theories (see Secs. 5.5.1
have a common tangent to the associated and 5.5.4) dealing with the unmixing kinet-
free energy curves. This fact is illustrated ics of solid solutions quenched into the
in Fig. 5-4 for a supersaturated solid solu- miscibility gap are now based on the (ques-
tion of composition c0, which decomposes tionable) assumption that the quenched-in
into a B-depleted phases a¢ and a B-rich single-phase states within the two-phase
phase b of composition cae and cbe , respec- region can be described by a ‘constraint’
tively. cae and cbe are fixed by the common Helmholtz energy F ′(c) > F(c), e.g., as is
inflection points (∂ 2 F ′/ ∂c2 )T ∫ 0, separ-

ates the two regions (Fig. 5-5 a). The es-
sence of the distinction between meta-
stability and instability will be discussed in
the following section.
5.2.3 Decomposition Mechanisms:

Nucleation and Growth versus Spinodal
Decomposition
Experimentalists are often inclined to
distinguish between two different kinds of
decomposition reactions, depending on
whether i) the solid solution experiences a
shallow quench (e.g., from points 0 to 1,
Fig. 5-5 a) into the metastable region or ii)
whether it is quenched deeply into the un-
stable region of the miscibility gap (e.g.,
from points 0 to 2).
For case i, unmixing is initiated via the
formation of energetically stable solute-
rich clusters (‘nuclei’). As inferred from
Fig.5-5 b) only thermal composition fluctu-
ations with sufficiently large compositional
amplitudes ·c – c0Ò lower (DF < 0) the
Helmholtz energy of the system and,
hence, can lead to the formation of stable
nuclei. According to the tangent construc-
Figure 5-5. a) Phase diagram of a binary model al- tion for the given composition c0 of the in-
loy with constituents A and B. The two-phase region itial solid solution, the largest decrease in
is subdivided by the mean-field spinodal curve into
DF is obtained for a nucleus with composi-
metastable (hatched) and unstable regions (cross-
hatched). b) Schematic Helmholtz energy versus tion cpnuc (which depends on c0) rather than
composition curves at temperature T1. The bold with cp = cbe . The latter is the composition
dashed curve in the two-phase region shows the of the terminal second phase b coexisting
‘constraint’ Helmholtz energy F ¢ (c) of the unstable in equilibrium with the a phase of compo-
solid solution. sition cae . The formation of stable nuclei via
localized ‘heterophase’ thermal composi-
tion fluctuations (Fig. 5-6 a) requires a nu-
shown in Fig. 5-5 b (bold dashed line). cleation barrier (typically larger than 5 kT )
Based on this concept, the miscibility gap to be overcome (see Sec. 5.5.1) and is char-
can be subdivided into a metastable region, acterized by an incubation period. This de-
where F ′ (c) > F (c) and ∂ 2 F ′/ ∂c2 > 0, and fines the homogeneous solid solution at
an unstable region, for which F ′(c) > F (c) point 1 (Fig. 5-5 a) as being metastable,
but ∂ 2 F ′/ ∂c2 < 0. The unique spinodal and the transformation as a nucleation and
curve, which is defined as the locus of the growth reaction.
The time evolution of the cluster size dis-

tribution and, hence, the dynamics of the
decay of the metastable alloy, are assumed
to proceed via the condensation or evapo-
ration of single solute atoms at each clus-
ter.
In case ii, the non-equilibrium solid so-
lution with an initial composition c0 > cas ,
(Fig. 5-5 b) is unstable with respect to the
formation of non-localized, spatially ex-
tended thermal composition fluctuations
with small amplitudes. Hence, the unmix-
ing reaction of an unstable solid solution,
which is termed spinodal decomposition, is
initiated via the spontaneous formation
and subsequent growth of coherent
(‘homophase’) composition fluctuations
(Fig. 5-6 b).
The dynamic behavior of an unstable al-
loy proceeding towards equilibrium has of-
ten been theoretically approached in terms
Figure 5-6. Spatial variation in solute distribution of spinodal theories (see Sec. 5.5.4 and
c (r = (x, y, z)) during a) a nucleation and growth re-
action, and b) a continuous spinodal reaction at the
Chapter 6), amongst which the most well-
beginning (time t1) and towards the end (t3) of the known linear theory is due to Cahn (1966)
unmixing reaction. The notation of the compositions and the most elaborate non-linear one is
refers to Fig. 5-5; c0ms and c0us are the nominal com- due to Langer et al. (1975). As discussed in
positions of the quenched-in metastable and unstable Sec. 5.2.2, the spinodal theories are based
solid solutions, respectively; R* is the critical radius
of the nuclei and l the wavelength of the composition
on the assumption that in each stage of the
fluctuations. The direction of the solute flux is indi- decomposition reaction the Helmholtz en-
cated by the arrows. After an extended reaction time ery of the non-equilibrium solid solution,
(e.g., after t3), the transformation products are simi- which contains compositional fluctuations,
lar and do not allow any conclusions to be drawn can be defined. As in the cluster kinetics
with respect to the early decomposition mode.
models, the driving force for phase separa-
tion is again provided by lowering the
As will be outlined in Sec. 5.5.1, the de- Helmholtz energy of the alloy. If the form
cay of a metastable solid solution via nu- of the ‘constraint’ Helmholtz energy F¢ (c)
cleation and growth has frequently been of the non-uniform system is properly cho-
described in terms of cluster kinetics mod- sen, the composition profile with asso-
els. The cluster kinetics approaches are es- ciated minimum Helmholtz energy can be
sentially based on the Becker–Döring the- determined (in principle!) at any instant of
ory (Becker and Döring, 1935) of the dy- the phase transformation.
namics of solute cluster formation. There it The existence of a unique spinodal curve
is assumed that the non-equilibrium system within the framework of the mean field
consists of non-interacting solute-rich clus- theories (Cahn, 1966; Cook, 1970; Skripov
ters of various size embedded in the matrix. and Skripov, 1979) has led to the idea (still
widespread in the community of metallur- qualitative insight into early-stage unmix-

gists) that there is a discontinuity of the ing behavior of a binary alloy, they have re-
mechanism and, in particular, of the de- vealed that there is no discontinuity in the
composition kinetics at the boundary decomposition kinetics to be expected dur-
between the metastable and unstable re- ing crossing of the mean-field spinodal
gions. Therefore, many experiments have curve by either increasing the concentra-
been carried out in order to determine the tion of the alloy and keeping the reaction
spinodal curve and to search for a kinetic temperature (TA ) constant or by lowering
distinction between metastable and un- the reaction temperature and keeping the
stable states (see Sec. 5.5.5). composition (c0) constant. On the other
In reality, there is no need to develop dy- hand, the mean-field description is strictly
namical concepts which are confined to ei- valid only for systems with infinitely long-
ther the metastable nucleation and growth range interaction forces (Gunton, 1984)
regime (case i) or to the unstable spinodal and, hence, in general does not apply to
regime (case ii). In fact, the cluster-kinetics metallic alloys (polymer mixtures might be
models and the spinodal theories can be close to the mean-field limit: Binder, 1983,
seen as two different approaches used to 1984; Izumitani and Hashimoto, 1985; see
describe phase separation, the dynamics of Sec. 5.5.4). Therefore, numerous experi-
which are controlled by the same mecha- ments that have been designed by metallur-
nism. i.e. diffusion of solvent and solute at- gists in order to determine a unique spino-
oms driven by the gradient of the chemical dal curve simply by searching for drastic
potential (Martin, 1978). changes in the dynamic behavior of an al-
This fact is reflected in the attempts to loy quenched into the vicinity of the mean-
develop ‘unified theories’ which comprise field spinodal must be considered with
spinodal decomposition as well as nuclea- some reservations.
tion and growth. Langer et al. (1975) tried In principle, the above-mentioned theo-
to develop such a theory on the basis of a ries and, in particular, the Monte Carlo
non-linear spinodal theory (see Sec. 5.5.4), simulations deal mainly with the dynamic
whereas Binder and coworkers (Binder et evolution of a two-phase mixture in its
al., 1978; Mirold and Binder, 1977) chose early stages. They frequently do not ac-
the cluster kinetics approach by treating spi- count for a further evolution of the precipi-
nodal decomposition in the form of a gener- tate or cluster size distribution with aging
alized nucleation theory (see Sec. 5.5.3). time (i.e., coarsening; see Sec. 5.6) once
These theories involve several assumptions the precipitated volume fraction is close to
whose validity is difficult to assess a priori. its equilibrium value. On the other hand,
The quality of all spinodal or cluster kinetics the time evolution of the precipitate micro-
concepts, however, can be scrutinized by structure beyond its initial clustering stages
Monte Carlo simulations of the unmixing has been the subject of many experimental
kinetics of binary ‘model’ alloys. These are studies and is of major interest in practical
quenched into either a metastable or an un- metallurgy. For this purpose, numerical ap-
stable state (see Sec. 5.5.6) and can be de- proaches have been devised (Langer and
scribed in terms of an Ising model (Kalos et Schwartz, 1980; Kampmann and Wagner,
al., 1978; Penrose et al., 1978). 1984) which treat nucleation, growth and
Although both the ‘unified theories’ and coarsening as concomitant processes and
the Monte Carlo simulations only provide thus allow the dynamic evolution of the
two-phase microstructure to be computed 5.2.4 Thermodynamic Driving Forces

during the entire course of a precipitation for Phase Separation
reaction (see Sec. 5.7).
Even in the single-phase equilibrium
As was pointed out in Sec 5.2.1, the re-
state, the mobility of the solvent and solute
action path of a supersaturated solid solu-
atoms at elevated temperatures permits the
tion can be rather complex, sometimes in-
formation of composition fluctuations
volving the formation of one or more inter-
which grow and decay again reversibly
mediate non-equilibrium phases prior to
with time. If the solid solution is quenched
reaching the equilibrium two-phase micro-
into the miscibility gap the two-phase mix-
structure. Unlike in the ‘early stage
ture is the more stable state and, thus, some
theories’ mentioned above, these compli-
of these fluctuations grow irreversibly ow-
cations, which are of practical relevance,
ing to the associated reduction in Helm-
can be taken into consideration in numeri-
holtz energy. The reduction in Helmholtz
cal approaches. Even though they still con-
energy during the transformation from the
tain a few shortcomings, numerical ap-
initial to the final state provides the driving
proaches lead to a practical description of
force DF. As we shall see in Sec. 5.5.1, it is
the kinetic course of a precipitation reac-
possible to calculate the formation rate and
tion which lies closest to reality.
the size of stable composition fluctuations
There are several comprehensive review
(‘nuclei’ of the second phase) by means of
articles and books dealing in a more gen-
the cluster kinetics approach once DF is
eral manner with the kinetics of first-order
known. It should, however, be emphasized
phase transitions (Gunton and Droz, 1984;
at this point that for most alloys it is rather
Gunton et al., 1983; Binder, 1987; Gunton,
difficult to calculate DF with sufficient ac-
1984; Penrose and Lebowitz, 1979). Phase
curacy. This is seen as one of the major
separation in solids (crystalline and amor-
hindrances to performing a quantitative
phous alloys, polymer blends, oxides and
comparison between theory and experi-
oxide glasses) via homogeneous nucleation
ment.
and growth or via spinodal decomposition
The driving force for precipitation is
represents only one aspect among many
made up of two different contributions:
others (Gunton and Droz, 1984). Apart
i) the gain in chemical Helmholtz en-
from the above-mentioned numerical ap-
ergy, DFch < 0, associated with the forma-
proaches, the comprehensive articles by
tion of a unit volume of the precipitating
Martin (1978) and by de Fontaine (1982)
phase b, and
cover most of the theoretical developments
ii) the expenditure of distortion Helm-
relevant to the kinetics of (homogeneous)
holtz energy, DFel > 0, accounting for the
phase separation in metallic systems. A
coherency strains which result from a
general overview of the broad field of dif-
likely variation of the lattice parameter
fusive phase transformations in materials
with the spatial composition fluctuations.
science including heterogeneous nuclea-
tion and discontinuous precipitation not i) Chemical contribution, D Fch. Accord-
covered in the present chapter, can be ing to the tangent rule and referring to Fig.
found in the article by Doherty (1983) or, 5-4, the chemical driving force for precipi-
as an introduction to this field, in the book tation of the equilibrium b phase out of a
by Christian (1975), see also chapter by solid solution with composition c0 is given
Purdy and Bréchet (2001). ––. Assuming the
by the numerical value of xy
precipitating phase b already has the final Hence, if, for example, F (c, T ) is known
bulk composition cbe ∫ cp rather than cpnuc for the stable solid solution, this value can
(which is only a reasonable assumption if readily be extrapolated into the adjacent
the supersaturation is not too large (Cahn two-phase region to yield F¢(c) (see Fig.
and Hilliard, 1959 a, b)), DFch has been de- 5-5) of the supersaturated homogeneous
rived for a unit volume of b phase with mo- solid solution or the related chemical po-
lar volume Vb as (Aaronson et al., 1970 a, tentials and activities. Of course, physi-
b): cally this is only meaningful if we assume
Rg T (as, in fact is done for the derivation of Eq.
DFch = – (5-9)
Vβ (5-9)) that the free energy of the non-equi-
librium solid solution can be properly de-
⎡ a (c ) a (c ) ⎤ fined within the miscibility gap.
× ⎢cβe ln B 0e + (1 – cβe )ln A 0e ⎥
⎣ aB ( cα ) aA ( cα ) ⎦ For many binary alloys and for some ter-
nary alloys of technological significance,
where ai (c) is the activity of the solvent the thermodynamic functions have been
(i = A) or solute component (i = B) in the evaluated by means of the CALPHAD
parent phase a for the given composition. method and are compiled in volumes of the
For most alloy systems the activity data CALPHAD series (Kaufman, 1977).
required for a computation of DFch by Hitherto, due to the lack of available ac-
means of Eq. (5-9) are not available. In tivity data, the activities entering Eq. (5-9)
principle, they can be derived from a com- were frequently replaced by concentra-
putation of the thermodynamic functions tions, i.e.,
by means of the CALPHAD method. Orig-
Rg T
inally this method was developed for the DFch = – (5-10)
calculation of phase diagrams by Kaufman Vβ
and coworkers (Kaufman and Bernstein, ⎡ c ⎛ 1 – c0 ⎞ ⎤
1970) on the basis of a few accurately × ⎢cβe ln 0e + (1 – cβe ) ln ⎜ ⎟⎥
measured thermodynamic data to which ⎣ cα ⎝ 1 – cαe ⎠ ⎦
suitable expressions for the thermodynam- If b is almost pure B, then cbe ≈ 1 and the
ic functions had been fitted. For the sake of chemical driving force is approximated as
a simplified mathematical description, the
stable solid solutions of the particular alloy Rg T c
DFch ≈ – ln 0e (5-11)
system are frequently described in terms of Vβ cα
the regular solution model, whereas the For many alloys the condition cbe ≈ 1 is
phase fields of the intermetallic com- not met. Nevertheless, it has been used for
pounds are approximated by line com- computation of DF. As discussed by Aa-
pounds. The thermodynamic functions ob- ronson et al. (1970 a), the resulting error
tained thereby are then used to reconstruct can be rather large. If the nucleating phase
the phase diagram (or part of it). The de- is a solvent-rich intermetallic compound
gree of self-consistency between the recon- (e.g., Ni3Al or Cu4Ti), the computation of
structed and the experimentally determined DFch is particularly difficult, since up to
phase diagram (or the agreement between now the corresponding activities are nei-
the measured thermodynamic data and the ther known nor furnished by the CAL-
derived data) serves as a measure of the ac- PHAD method, and Eqs. (5-10) and (5-11)
curacy of the thermodynamic functions. are no longer valid.
ii) Reduction of the driving force by the The problem of calculating DFel for co-
elastic strain energy, D Fel. The formation herent inclusions is rather complex. It has
of a composition fluctuation is associated generally been treated within the frame-
with the expenditure of elastic strain Helm- work of isotropic elastic theory for ellip-
holtz energy, DFel , if the solvent and the so- soidal precipitates with varying axial ra-
lute atoms have different atomic radii. Ac- tios, e.g. Eshelby (1957), and has been ex-
cording to Cahn (1962), the Helmholtz en- tended to the anisotropic case (Lee and
ergy of a system containing homophase Johnson, 1982). Basically, these treatments
fluctuations is raised by show that the strain energy depends on the
particular shape of the new phase. Only if
DFel = h2 Dc2 Y (5-12)
the transformation strains are purely dilata-
where h = (1/a0) (∂a/∂c) denotes the change tional and if a and b have about the same
in lattice parameter (a0 for the homogene- elastic constants will DFel become inde-
ous solid solution of composition c0) with pendent of the shape. It is then given as
composition, and Dc the composition am- (Eshelby, 1957):
plitude. Y is a combination of elastic con-
1+n ⎞ 2
stants and depends on the crystallographic DFel = 2 g ⎛ d (5-14)
direction of the composition modulation. It ⎝1–n ⎠
reduces to E/(1 – v), E and v being Young’s with g being the shear modulus.
modulus and Poisson’s ratio, respectively, As is shown schematically in Fig. 5-7,
if the elastic anisotropy A ∫ 2 c44 + c12 – with respect to the Helmholtz energy of the
c11 is zero; otherwise, in order to minimize incoherent equilibrium phase b, the Helm-
DFel , the composition fluctuations are ex- holtz energy of the coherent phase b¢ is
pected to grow along the elastically soft di- raised by the elastic energy to match the
rections, which for cubic crystals and A > 0 two lattices. The driving force DF inc ∫ ––
xy
are the ·100Ò directions.
In the context of heterophase fluctuations,
the barrier against nucleation of the new
phase is dominated by the matrix/ nucleus
interfacial energy (see Sec. 5.5.1). Since co-
herent interfaces have a lower energy than
incoherent ones, a precipitate is usually co-
herent (or at least semi-coherent) during the
early stages of nucleation and growth. Often
its lattice parameter, ab , is slightly different
from that of the parent phase, aa. The result-
ing misfit between the (unstrained) matrix
and the (unstrained) precipitate:
aα – aβ
d=2 (5-13)
aα + aβ
can be accommodated by an elastic strain
if both d and the particles are sufficiently Figure 5-7. Helmholtz energy curves and associated
small; this is commonly the case during the driving forces for precipitates of the coherent meta-
early stages of nucleation and growth. stable (Fb¢) and incoherent equilibrium (Fb) phases.
is reduced to the value of DF coh ∫ uv–– , and

the solubility limit in the a phase increases
from cae to c¢ae.
Hence, if we refer to the resulting coher-
ent phase diagram (cf. Fig. 5-1), the total
driving force, DF , for coherent unmixing,
implicitly already containing the elastic
contribution, is obtained by replacing, e.g.,
in Eq. (5-11), the composition cae of the in-
coherent phase by the composition c¢ae of
the coherent one:
Rg T c
DF = – ln 0e (5-15)
Vβ′ ca′
As an example, for metastable iron-rich
f.c.c. precipitates (cbe ≈ 99.9 at.%) with
d = – 8 ¥ 10–3 formed at 500 °C in a super- Figure 5-8. Solubility limits of Cu in Al in the pres-
saturated Cu–1.15 at.% Fe alloy (c0 = 1.15 ence of the metastable q≤ and q¢ phases (dashed) and
at.%, (cae ≈ 0,05 at.%; Kampmann and Wag- the stable q phase, as a function of temperature (after
Hornbogen, 1967).
ner, 1986), Eq. (5-12) yields DFel≈0.13
kJ/mol. This value is negligibly small with
respect to the rather large chemical driving
force (≈20 kJ/mol). Since, however, DFch loys, amongst which Al–Cu (c0 ≤ 2.5 at.%
decreases markedly with increasing tem- Cu) is probably the best known. There the
perature (or decreasing supersaturation), stable phase (q-CuAl2) is incoherent, with
whereas DFel does not, coherent nucleation a high associated sab (R 1 J/m2) (Hornbo-
commonly occurs at larger undercoolings gen, 1967). This inhibits homogeneous nu-
or supersaturations, whereas nucleation of cleation and q is found to form only at
incoherent precipitates takes place at small undercoolings, preferentially at grain
smaller ones. This is in fact observed in the boundaries (see Fig. 5-8). At larger under-
Cu–Ti (see Sec. 5.2.1) and Al–Cu (Hornbo- coolings, a series of metastable copper-rich
gen, 1967) systems. precipitates is formed in the order: GPI
As will be shown in Sec. 5.5.1.1, the bar- zones Æ GPII zones (q≤) Æ q¢ with differ-
rier DF* against formation of the new ent crystal structures (see Fig. 5-8). Gui-
phase, and hence the nucleation rate, is not nier–Preston zones of type I (GPI) and type
only a function of the driving force but is II (GPII or q≤) are coherent and nucleate
rather sensitive to the nucleus/matrix inter- homogeneously, whereas q¢ is semi-coher-
face energy sab . Therefore, the phase nu- ent and nucleates preferentially at disloca-
cleating first will not necessarily be the tions.
equilibrium phase with the lowest Helm- Al–Mg–Si alloys represent an important
holtz energy but that with the lowest DF*, group of age-hardenable structural materi-
e.g. a coherent metastable phase with a als heavily used by industry in both cast
low value of sab . This explains the likely and wrought form. Hence, control and opti-
formation of a series of metastable phases mization of the precipitate microstructure
(see Sec. 5.2.1) in various decomposing al- during cooling and heat treatment is of par-
amount interest for optimum mechanical with respect to both the detection limit and
properties (Bratland et al., 1997). How- the spatial resolution of microanalysis.
ever, due to its inherent complexity, the
precipitation sequence of these commercial
5.3.1.1 Direct Imaging Techniques
alloys is still a matter of controversy (Dutta
and Allen, 1991; Gupta and Lloyd, 1992; Field ion microscopy (FIM) (Wagner,
Edwards et al., 1998). 1982) as well as conventional (CTEM)
(Hobbs et al., 1986) and high resolution
(HREM) transmission electron microscopy
5.3 Experimental Techniques for (Smith, 1983) allow for direct imaging of
Studying Decomposition Kinetics the second-phase particles, provided the
contrast between precipitate and matrix is
sufficient.
5.3.1 Microanalytical Tools
In CTEM both the bright field and the
In general, the course of a decomposi- dark field contrast of particles less than
tion reaction, including the early stages ≈ 5 nm in diameter are often either too
(during which composition fluctuations weak or too blurred for an accurate quanti-
and second-phase nuclei are formed, see tative determination of the relevant structu-
Sec. 5.2.1) and the coarsening stages, can- ral precipitate parameters. Hence, CTEM
not be followed continuously by any one does not provide access to an experimental
microanalytical technique. The progress of investigation of the early stages of decom-
the reaction is usually reconstructed from position but remains a technique for study-
the microstructure that develops at various ing the later stages. In contrast, HREM al-
stages of the phase transformation. Thus it lows solute clusters of less than 1 nm diam-
is necessary to analyze the spatial exten- eter to be imaged, as was demonstrated for
sion and the amplitude of composition Ni3Al precipitates in Ni–12 at.% (Si-Qun
fluctuations of incipient second-phase par- Xiao, 1989) and for silver-rich particles in
ticles, as well as the morphology, number Al–1 at.% Ag (Ernst and Haasen, 1988;
density, size and chemical composition of Ernst et al., 1987). As is shown in Fig. 5-9
individual precipitates at various stages of for the latter alloy, the particles are not
the phase transformation. For this purpose ideally spherical but show some irregular-
microanalytical tools are required that are ities which, however, are small in compari-
capable of resolving very small (typically a son with their overall dimensions. Hence,
few nm) solute clusters, and which allow describing their shapes by spheres of radius
(frequently simultaneously) an analysis of R is still a rather good approximation.
their chemical composition to be made. Prior to measuring the particle sizes di-
The tools that, in principle, meet these rectly from the HREM micrographs, it
requirements can be subdivided into two must be established via computer image
groups: direct imaging techniques and simulations that there exists a one-to-one
scattering techniques. correspondence between the width of the
It is beyond the scope of this chapter to precipitate contrast (dark area in Fig. 5-9)
discuss any one of the techniques belong- and its true size. This was verified for the
ing to either group in any detail. We shall HREM imaging conditions used in Fig.
only briefly summarize the merits and the 5.9. The evolution of the size distribution
shortcomings of the various techniques in Al–1 at.% Ag with aging time at 413 K
5.3 Experimental Techniques for Studying Decomposition Kinetics 327
as derived from HREM micrographs is

shown in Fig. 5-10.
The HREM lattice imaging technique
has also been employed to determine the
spacings of adjacent lattice planes in vari-
ous alloys undergoing phase separation
(Sinclair and Thomas, 1974; Gronsky et
al., 1975). The smooth variations observed
were attributed to variations in lattice pa-
rameter caused by composition modula-
tions such as we would expect from spino-
dal decomposition (e.g. Fig. 5-6 b at t1 or Figure 5-9. HREM lattice fringe image of Al-1 at.%
t2). Subsequent model calculations of high- Ag aged for 92 h at 413 K with silver-rich precipi-
resolution images, however, showed that tates (dark areas). For imaging the Al matrix a rather
large defocus (⬃ 280 nm) had to be chosen, which
the spatial modulation of the lattice fringes gives rise to the bright Fresnel fringes surrounding
observed on the HREM image can be sig- the particles (reproduced by courtesy of F. Ernst
nificantly different from the spatial mod- (Ernst and Haasen, 1988)).
ulation of the lattice plane spacings in the
specimen (Cockayne and Gronsky, 1981).
In practice, it does not appear possible to
derive reliable information on composition
modulations from modulations of lattice
fringe spacings in HREM images.
The composition of the imaged particles
can, in principle, be obtained from analyti-
cal electron microscopy (AEM), which is
routinely based on energy dispersive X-ray
analysis (EDX) or, less frequently, on elec-
tron energy loss spectroscopy (EELS)
(Williams and Carter, 1996). Due to beam
spreading by electron beam/specimen
interaction, the spatial resolution for rou-
tine EDX microanalysis is limited to
R 10 nm. This value is rather large and con-
fines EDX microanalysis of unmixing al-
loys to the later stages of precipitation. The
more interesting early stages of decom-
position, however, where composition
changes of either extended or localized so-
lute fluctuations are expected to occur, are
not amenable to EDX microanalysis.
The spatial resolution of EELS may,
with difficulty, reach 1 nm (Williams and Figure 5-10. Time evolution of the particle size dis-
Carter, 1996), but so far, EELS has not tribution in Al–1 at.% Ag during aging at 413 K (af-
been employed for a systematic microanal- ter Ernst and Haasen, 1988).
ysis of composition fluctuations during scale (Fig. 5-2 d). The spatial distribution
early-stage decomposition. of each chemical species can be directly
Analytical field ion microscopy (AFIM) observed with a spatial resolution better
(Wagner, 1982; Miller et al., 1996) can fa- than 0.5 nm (Auger et al., 1995; Pareige et
vorably be applied for an analysis of ultra- al., 1999; Al-Kassab et al., 1997). The vol-
fine solute clusters in the early stages of ume typically sampled is about 15 × 15
decomposition, as both the imaging resolu- × 100 nm3 and, thus, about a hundred times
tion of the FIM (see Fig. 5-2) and the spatial larger than the volume analyzed with the
resolution of microanalysis of the integrated AFIM.
time-of-flight spectrometer (‘atom probe’)
are sufficient, i.e., <1 nm and ª 2 nm, re-
5.3.1.2 Scattering Techniques
spectively. (Although the atom probe de-
tects single atoms from the probed volume, The time evolution of the structure of
the requirement for the analysis to be sta- supersaturated alloys, as well as of oxides
tistically significant confines the microana- and polymer blends undergoing phase sep-
lytical spatial resolution to ≥ 2 nm.) As is aration, can be analyzed by means of small
shown in Figs. 5-2 a – c and 5-19 c, both the angle scattering of X-rays (SAXS), neu-
morphologies of small particles and their trons (SANS), and light. Light scattering is
three-dimensional arrangements can also of course confined to transparent speci-
be determined in the FIM, at least for pre- mens, in which the domain sizes of the
cipitates with sufficient contrast. evolving second phase must be of the order
The volume sampled during an atom of ≈ 1 µm. It has been successfully applied
probe FIM analysis is rather small (typi- in studies of decomposition, e.g., of a poly-
cally about 200 nm3). Hence, in order to mer mixture of polyvinyl methyl ether
obtain statistically significant data con- (PVME) and polystyrene (PS) (Snyder and
cerning the average size (R̄) and the num- Meakin 1983 a, b, 1985), and of various
ber density (Nv) of the precipitates, the lat- glass-forming oxide systems (Goldstein,
ter ought to exceed 1023 m–3. Since after 1965; Rindtone, 1975).
the early stages of precipitation Nv de- In principle, SAXS (Glatter, 1982) and
crease with time (see Sec. 5.2.1), AFIM SANS (Kostorz, 1979; Sequeira et al.,
shows its full potential as a microanalytical 1995) provide access to a structural analy-
tool in studies of the early stages of precip- sis of unmixing alloys, in both the early
itation, during which R̄ is commonly small stages where the composition fluctuations
and Nv sufficiently large (Haasen and Wag- can be small in spatial extension and in am-
ner, 1985). plitude, as well as in the later stages of de-
With respect to analyses of the spatial ar- composition where R̄ and Nv may have at-
rangement and the composition of precipi- tained values which are quite unfavorable
tated phases, the versatility of the AFIM for a quantitative analysis by TEM or FIM.
was considerably improved by the devel- In addition, evaluation of the Laue scatter-
opment of the tomographic atom probe ing allows the remaining supersaturation of
(TAP) (Blavette et al., 1993, 1998; Miller solute atoms in the matrix to be deter-
et al., 1996). This instrument allows for mined.
the three-dimensional reconstruction of a The metallurgist is left with the problem
small volume of the microscopically heter- of extracting in a quantitative manner all
ogeneous material on a sub-nanometric the structural data contained in scattering
gle. For binary alloys S ( k, t) is the Fourier

transform of the two-point correlation
function at time t (Langer, 1975):
(5-16)
G (| r – r0|, t) = ·[c (r, t) – c0] [c (r0, t) – c0]Ò
The right-hand side in Eq. (5-16) denotes

the non-equilibrium average of the product
of the composition amplitudes at two dif-
ferent spatial positions, r and r0 , in the al-
loy with average composition c0.
Figure 5-11. SANS curves of Cu-2.9 at.% Ti single
S (k, t) contains all the structural infor-
crystals aged for the given times at 350 °C. Note that mation on the phase-separating system.
even the homogenized and quenched (‘hom’) speci- Many theories and computer simulations
men yields SANS intensity, indicating that phase dealing with phase separation yield the ev-
separation has occurred during quenching (Eckerlebe olution of S (k, t) and predict the shift of
et al., 1986).
km (which is a measure of the average
spacing of solute clusters) and the growth
of Sm (km) with time (Langer et al., 1975;
curves such as are shown in Fig. 5-11. This Marro et al., 1975, 1977; Binder et al.,
is often not trivial (Glatter, 1982), in partic- 1978).
ular for concentrated alloys (i.e., for most Hence, SAXS and SANS curves meas-
technical two-phase alloys). Here the scat- ured after different aging times can be di-
tering curve reveals a maximum (at posi- rectly compared with the predictions from
tion km and with height Sm (km) in Fig. various theoretical kinetic concepts (see
5-11) which results from an interparticle Secs. 5.5.4, 5.5.6 and 5.8.1).
interference of the scattered waves (Kos- In ternary systems the situation is much
torz, 1979). In general, the interparticle more complicated. For a substitutional ter-
interference function is not known, thus nary alloy, there are three linearly indepen-
impeding a straightforward quantitative dent pair correlation functions with three
analysis. Moreover, the composition and related partial structure functions (de Fon-
morphology of the fluctuations are re- taine, 1971, 1973), which linearly combine
quired in order to perform such an analysis. to the measured SAS intensity. An unam-
For many alloys this information can only biguous characterization of the kinetics of
be obtained from AFIM or AEM. These phase separation in ternary substitutional
complicating factors often demand the alloys requires the three partial structure
above-mentioned microanalytical tools to functions to be determined separately. This
be employed jointly, rendering experimen- has been attempted by employing the
tal studies of decomposition rather difficult ‘anomalous small angle X-ray scattering’
and tedious. technique for investigations of phase separ-
The small angle scattering intensity is ation in Al–Zn–Ag (Hoyt et al., 1987),
proportional to the structure function S ( k, Cu–Ni–Fe (Lyon and Simon, 1987, 1988)
t); k is the scattering vector with k ∫ | k | = and Fe–Cr–Co (Simon and Lyon, 1989).
4 p sin (q /2)/l; l is the wavelength of the Unlike for X-rays, for thermal neutrons
X-rays or neutrons and q the scattering an- the atomic nuclear scattering length is not
Due to relatively simple calibration pro-

cedures in SANS experiments, the scat-
tered intensity which is expressed in terms
of the coherent (nuclear or magnetic) scat-
tering cross-section per unit volume,
dS ( k, t)/dW,can be measured in absolute
units. This is directly related to the struc-
ture function S ( k, t) via (Hennion et al.,
1982):
dS 1
(k , t ) = 2 ( bM – bp )2 S (k , t ) (5-17)
Figure 5-12. Nuclear and magnetic SANS curves of dW W
Fe–1.4 at.% Cu aged for 200 h at 400 °C. From the – –
where bM and bp are the locally averaged
ratio of magnetic and nuclear scattering intensities,
the composition of the copper-rich clusters can be de- (nuclear or magnetic) scattering lengths of
rived. Theoretical curves are shown as full lines the matrix and of the solute-rich clusters,
–
(Kampmann and Wagner, 1986). respectively, and W is the mean atomic vol-
ume.
monotonically dependent on the atomic

number (Kostorz, 1979). For this reason 5.3.2 Experimental Problems
use of SANS is more universal and is fre-
5.3.2.1 Influence of Quenching Rate
quently superior ro SAXS for decomposi-
on Kinetics
tion studies of binary alloys with mainly
transition metal constituents having similar For studies on decomposition kinetics
atomic numbers. Consequently, SAXS has the alloy is commonly homogenized in
only been used extensively for studies of the single-phase region at TH (Fig. 5-1),
the unmixing kinetics in Al–Zn where the quenched into brine and subsequently iso-
difference (this, in fact, controls the con- thermally aged at TA. In order to capture
trast in SAS) in atomic scattering lengths the initial stages of the decay of the super-
for X-rays is sufficiently large (Rundman saturated solid solution, both the quench-
and Hilliard, 1967; Hennion et al., 1982; ing and the heating rate to TA are required
Forouhi and de Fontaine, 1987). to be sufficiently high in order to avoid
Furthermore, if the two phases differ not phase separation prior to isothermal aging.
only in composition but also in magnetiza- This can be comfortably achieved in alloys
tion, both nuclear and magnetic SANS with small supersaturations. It is, however,
curves can be recorded (see Fig. 5-12). often a problem (or even impossible) for
These two independently measurable alloys with large supersaturations and/or
SANS curves sometimes even allow the small interfacial energies, where the nucle-
composition of the scattering centers to be ation barriers are small (see Secs. 5.2.3 and
determined, e.g., for Fe–Cu, where dia- 5.5.1) and, hence, the nucleation rates
magnetic copper-rich particles precipitate large.
in the ferromagnetic a-Fe matrix (Kamp- The driving force for unmixing in Al-1
mann and Wagner, 1986), or for phase-sep- at.% Ag at 140 °C is rather small
arated amorphous Fe40Ni40P20 (Gerling et (≈0.5 kJ/mol). Nevertheless, because of the
at., 1988). extremely small interfacial energy of only
≈0.01 J/m2 (Le Goues et al., 1984 c), the

barrier to nucleation is so low that precipi-
tation commences instantaneously almost
as soon as the solvus line is crossed. Thus,
except for extremely high quenching rates,
the non-equilibrium single-phase state can-
not be frozen in and the as-quenched mi-
crostructure already contains a large num-
ber of small GP zones (Fig. 5-13).
The accessible quenching rates have
been found to be insufficient for a suppres-
sion of phase separation during quenching Figure 5-13. HREM micrograph of GP zones
formed during quenching of Al–1 at.% Ag. The
of homogenized Cu–Ti alloys with Ti con-
smallest GP zones have diameters of only ⬃ 1 nm
centrations exceeding ≈2.5 at.%. This is (reproduced by courtesy of P. Wilbrandt (Ernst et al.,
discernible from the SANS curve of the as- 1987)).
quenched specimen (Fig. 5-11) (see Sec.
5.7.5).
Further problems result from the fact aged for 10 min at 350 °C. The maximum
that the decomposition kinetics at TA are in the SANS curves is found to be higher,
strongly influenced by the concentration of and its position lower, the higher the ho-
quenched-in excess vacancies. This de- mogenization temperature. This indicates
pends strongly on the chosen homogeniza- that phase separation in the early stages
tion temperature and quenching rate. progresses the faster the higher the homog-
Fig. 5-14 shows SANS curves from enization temperature is chosen. Obviously,
specimens that were solution-treated at the in Cu–Ti, the concentration of quenched-in
given homogenization temperature TH . excess vacancies is correlated to TH and
Following quenching, each specimen was significantly influences the decomposition
Figure 5-14. SANS curves

of Cu–2.9 at.% Ti quenched
from the given homogeniza-
tion temperatures TH, and
(apart from the bottom
curve) subsequently aged
for 10 min at 350 °C (Ecker-
lebe et al. 1986).
kinetics. This effect may be minimized if

the alloy is first solution-treated at high TH
and subsequently equilibrated at a homog-
enization temperature slightly above the
solvus temperature prior to quenching.
The inconsistency in the exact mode and
kinetics of decomposition in Al–22 at.%
Zn, investigated by different authors
(Rundman and Hilliard, 1967; Gerold and
Merz, 1967), has been explained by the
presence of different quenching rates
which unavoidably lead to different states
of solute clustering in the as-quenched
specimens (Agarwal and Herman, 1973;
Bartel and Rundman, 1975).
The problems stated above make experi-
ments devised for examination of the vari-
ous spinodal theories particularly difficult.
5.3.2.2 Distinction of the Mode

of Decomposition
The criterion that must be satisfied in or-
der to distinguish spinodal decomposition
(s.d.) from a nucleation and growth (n.g.)
reaction is to prove by any microanalytical
technique that the amplitude of the compo- –
Figure 5-15. a) Composition and diameter D, of g ¢-
sition modulations of an alloy deeply precipitates in Ni–36 at.% Cu–9 at.% Al as a func-
quenched into the miscibility gap increases tion of aging time at 580 °C. The corresponding pre-
with time during the initial stages of phase cipitate microstructure is shown in Fig. 5-2 a – c (Liu
–
separation (see Sec. 5.2.2 and Fig. 5-6). and Wagner, 1984). b) Composition and diameterr D
Even apart from the problem outlined in of the a1 (matrix) and a2 (precipitate ) phases in
Fe–29 at.% Cr–24 at.% Co as a function of aging
the preceding section, this is a rather diffi- time at 640 °C (Zhu et al., 1986).
cult task, since the cluster diameters or the
modulation wavelengths in most of the
more concentrated alloys investigated so
far have been found to range below the the clusters have attained sizes that are ac-
resolution limit of composition analysis, cessible to chemical analysis by AFIM
e.g. ≈2 nm for AFIM. Figure 5-15 reveals (3 nm and 1.8 nm, respectively). (In fact
that the compositions of the solute clusters both ternary alloys may be regarded as
in Ni–36 at.% Cu–9 at.% Al (Liu and pseudobinary. Thus the thermodynamic
Wagner, 1984) and in a hard-magnetic considerations of Sec. 5.2.2 still apply.)
Fe–29 at.% Cr–24 at.% Co (Zhu et al., The observed features suggest the reaction
1986) alloy have reached their equilibrium in both alloys to be of the n.g. type. How-
values and, hence, remain constant once ever, it cannot be ruled out that these clus-
Figure 5-16. Field ion image of Fe–29 Cr–14 Co–2

Al–0.15 Zr (at.%) aged into its optimum magnetic
state (Zhu et al., 1986).
ters result from a s.d. reaction which was

terminated after even shorter aging times
than could be covered in these studies.
Fig. 5-16 shows the interconnected pre-
cipitate microstructure of another hard-
magnetic alloy, Fe–29 at.% Cr–14 at.%
Co–21 at.% Al–0.15 at.% Zr, aged to yield
its optimum magnetic properties. In order
to characterize the decomposition process
with respect to the distribution of Cr Figure 5-17. a) Evolution of the Cr concentration at
between the two phases, atom probe micro- 525 °C and 600 °C in the Cr-rich a2-phase as a func-
analysis was performed. tion of aging time. b) Composition of the Fe-rich (a1)
The measured Cr concentration of the and Cr-rich (a2) phase after aging at 525 °C for the
darkly imaging precipitating a2-phase as a given times (in minutes). The dashed line corre-
sponds to the tie-line with a Fe/Co ratio of 3/1 (Zhu
function of aging time at T = 525 °C and
et al., 1986).
T = 600 °C is shown in Fig. 5-17 a. During
aging at 600 °C for less than 5 min it in-
creases continuously before it attains its
constant equilibrium value. This behavior centration in the a2-phase with decreasing
indicates unequivocally a spinodal decom- aging temperature is due to the widening of
position mechanism. At the lower aging the miscibility gap. As the Cr concentra-
temperature of TA = 525°C, the Cr ampli- tion in the a2-phase increases, the propor-
tude grows more slowly deeper within the tions of Co and Fe are reduced in this phase
miscibility gap due to the slower diffusion, and are correspondingly increased in the
but ultimately reaches higher equilibrium a1-phase. Atom probe analysis yields a
concentrations in the a2-precipitates, thus Fe/Co ratio of 3 : 1 for both the a1- and a2-
showing the spinodal behavior even more phases. This ratio corresponds to the tie-
clearly. The observed increase in Cr con- line in the ternary system along which
spinodal decomposition proceeds. Fig.

5-17b shows the compositions of the iron-
rich and chromium-rich phases for various
aging times at 525 °C and the tie-line corre-
sponding to an Fe/Co ratio of 3 : 1.
This study is one of the very rare ones on
metallic alloys to have identified a decom-
position reaction unequivocally to be of the
spinodal type.
Frequently, the occurrence of either
quasi-periodically aligned precipitates
which give rise to side-bands in X-ray or
TEM diffraction patterns (Fig. 5-18) or of a
precipitated phase with a high degree of
interconnectivity (e.g., Fig. 5-16) has been
employed as a unique criterion by which to
define an alloy as spinodal. However, as
will be pointed out in Secs. 5.4.1 and 5.5.5,
morphology alone cannot be used to unam-
biguously distinguish spinodal decomposi-
tion from a nucleation and growth reaction
(Cahn and Charles, 1965).
5.4 Precipitate Morphologies

There is a wide variety of different
shapes in individual precipitates and of
morphologies in precipitate microstruc-
tures. This is illustrated in Fig. 5-19 for
three different Cu–Ni-based alloys with
ternary additions of Al, Cr or Fe, the pre-
cipitated volume fraction ( fp) of which is
rather large, e.g., fp (CuNiAl) ≈0.20, fp
(CuNiCr) ≈0.19, and fp (CuNiFe) ≈0.37.
Figure 5-18. TEM micrographs of Cu–48 at.% Ni–8 Aging of Ni–37 at.% Cu–8 at.% Al for
at.% Fe aged for a) 8 h, b) 23 h, and c) 65 h at 500 °C. 167 h at 580 °C yields randomly distrib-
–
Each insert shows two satellites around the bright uted spherical particles with R ≈11 nm and
(002) matrix reflections. The distance between the a number density of ≈3.6 ¥ 1022 m–3 (Fig.
satellites and the fundamental reflection is inversely
5-19 a). In contrast, TEM of Cu–36 at.%
proportional to the wavelength of composition mod-
ulations or to the precipitate spacing in modulated Ni–4 at.% Cr aged for 240 h at 650 °C re-
structures (reproduced by courtesy of R. P. Wahi veals a modulated precipitate structure
(Wahi and Stajer, 1984)). with cuboidal particles aligned along the
5.4 Precipitate Morphologies 335
three ·100Ò directions (Fig. 5-19 b). Re-

placing Al or Cr by Fe leads to a mottled
(‘sponge-like’) precipitate microstructure
in Cu–48 at.% Ni–8 at.% Fe aged for 8 h at
500 °C. The three-dimensional intercon-
nectivity of the mottled structure becomes
discernible by a reconstruction of a se-
quence of FIM images which were taken at
various distances underneath the original
surface of the FIM specimen (Fig. 5-19 c).
Modulated structures have been pre-
dicted (Cahn, 1965; Hilliard, 1970) to
evolve from spinodal decomposition in
elastically anisotropic cubic matrices, ow-
ing to the tendency to minimize the cohe-
rency strain energy (see Sec. 5.2.4, Eq.
(5-12)), Furthermore, in material which is
either isotropic or for which the elastic en-
ergy (see Eq. (5-12)) is negligibly small,
spinodal decomposition is predicted to
generate an interconnected mottled precip-
itate morphology. Computer simulations of
the latter case, which were based on a
superposition of randomly oriented sinu-
soidal composition fluctuations of fixed
wavelength, random phase shifts and a
Gaussian distribution of amplitudes,
Figure 5-19. a) TEM dark field image of g ¢-precipi-

tates in Ni–37 at.% Cu–8 at.% Al aged for 167 h at
580 °C (Wagner et al. 1988). b) TEM bright field im-
age of Cu–36 at.% Ni–4 at.% Cr aged for 240 h at
650 °C displaying particle alignment along the ·100Ò
matrix directions (Wagner et al., 1988). c) FIM mi-
crographs of Cu–48 at.% Ni–8 at.% Fe aged for 8 h
at 500 °C. Between each FIM micrograph 2 nm of the
specimen surface were removed (‘field evaporated’
(Wagner, 1982)) in order to reveal the three-dimen-
sional arrangement of the brightly imaged (Ni, Fe)-
rich precipitated phase. In a the precipitates C, D, E,
F, G are apparently isolated; after having removed
2 nm, they have merged into one large extended par-
ticle (b). At still greater depth (d; 6 nm) ‘particle’ C
appears again to be isolated from the others by the
darkly imaged matrix (reproduced by courtesy of W.
Wagner (Piller et al., 1984)).
complete study of the evolution of the new

phase with aging time (see Sec. 5.5.5).
5.4.2 Factors Controlling the Shapes

and Morphologies of Precipitates
In the case of homogeneous nucleation,
the precipitating particles are commonly
coherent with the matrix. Their shape is
controlled by the rather complex interplay
of various factors, such as the magnitude
and anisotropy of the interfacial energy, the
difference in the elastic constants between
matrix and precipitate, and the crystal
Figure 5-20. Computer-simulated cross-section of a structure of the latter (Khachaturyan,
spinodal structure in an isotropic solid displaying the 1983).
interconnectivity of the two incipient phases 1 and 2 The complexity of the situation is illus-
with equal volume fraction. (After Cahn, 1965).
trated in Fig. 5-21 a (Lee et al., 1977).
There the anisotropic elastic strain energy
yielded an interconnectivity of the two DFel/d 2 of an ellipsoidal Ag-rich particle
conjugate phases for volume fractions with different orientational relationships
ranging from ≈0.15 to ≈0.85 (Cahn, 1965). with respect to a Cu or Al matrix is plotted
The simulated morphological pattern (Fig. as a function of the aspect ratio K ∫ T/R,
5-20) resembles the FIM images shown in where T and R denote the two axes of an el-
Figs. 5-16 and 5-19 c remarkably well. lipsoid of revolution.
This has made it quite tempting to define Assuming the interfacial energy to be
these alloys and, more generally, all alloys isotropic, Fig. 5-21 a reveals that the mini-
displaying interconnected or quasi-periodi- mum strain energy is obtained for platelet-
cal morphologies, as spinodal alloys. In shaped particles (K Ⰶ1), the cubic direc-
general, however, in more concentrated al- tions of which lie parallel to those of the Al
loys where the number density of clusters or Cu matrices.
of the new phase is large and, hence, the More generally, theory predicts (Lee et
intercluster spacing small, interconnectiv- al., 1977) platelets to have the minimum
ity and quasi-periodicity in the late-stage and spheres to have the maximum strain
microstructure may result from other energy if the precipitated phase is elasti-
mechanisms. Examples of such mecha- cally softer than the matrix, regardless of
nisms are coalescence of neighboring parti- the orientation relationship and elastic an-
cles (e.g., in the BaO–SiO2 system) or se- isotropy. For the reverse situation of a hard
lective coarsening of elastically favorably particle embedded in a softer matrix, the
oriented particles (e.g., in Ni–Al (Ardell sphere represents the minimum strain en-
and Nicholson, 1966; Doi et al., 1988)). ergy shape.
Hence, in essence, the distinction between However, the elastic constants of the
a spinodal reaction and nucleation and precipitating phase are often not available,
growth cannot be based solely on any spe- rendering the computation of the shape
cific morphological features but requires a with minimum strain energy difficult. In
5.4 Precipitate Morphologies 337
a) b)
Figure 5-21. a) Strain energy vs. aspect ratio of an Ag precipitate in Cu and Al matrices. The numbers 1, 4, 5
and 7 designate different orientation relationships between matrix and precipitate. The lowest strain energy is
obtained if the cubic directions of both particle and matrix are parallel to each other (curves 1), d is the linear
misfit. After Lee et al. (1977). b) The variation in a parameter, aR2, as a function of L. aR2 is used to characterize
the symmetry of a given family of equilibrium shapes: it is zero if the shape is four-fold symmetric and non-zero
if it is two-fold symmetric (Thompson et al., 1994).
particular, this holds true if the nucleating sions they found that at a critical size the
phase is a metastable transition phase. shape of a circular precipitate will change
In essence, however, the adopted shape to an ellipse in a smooth manner as the size
is determined by the balance between the of the precipitate is increased; the shape bi-
interfacial Helmholtz energy D Fa/b and the furcation is supercritical. A supercritical
elastic Helmholtz energy DFel. D Fa/b is bifurcation is analogous to the change in
minimized for particles with spherical or the order parameter on passing through a
faceted shape (e.g., Figs. 5-9 and 5-19) second-order phase transition. In contrast,
whereas DFel , which is related to the parti- in three dimensions, transcritical bifurca-
cle volume, commonly prefers a platelet- tions (first-order transitions) are possible.
like morphology (e.g., Fig. 5-21 a). Thus, Such a bifurcation implies that discontinu-
during the early stages of precipitation, ous changes in particle shape will occur as
D Fa/b is the dominating term, whereas the particle increases its size.
DFel prevails in the limit of large particles. The work of Johnson and Cahn clearly
The nature of this elastic stress-induced showed the importance of particle shape
precipitate–shape transition has been in- bifurcations in understanding microstruc-
vestigated by Johnson and Cahn (1984) us- tural evolution in systems with misfitting
ing bifurcation theory. Assuming an elasti- particles. Their study, however, was re-
cally isotropic inhomogeneous system and stricted to ellipsoidal particle shapes. This
by restraining the morphology of the parti- was remedied in the work of Thompson and
cle to be ellipsoidal it was possible to ad- Voorhees (1999) where the equilibrium
dress analytically the structure of the shape shape of a particle in an elastically aniso-
bifurcation. For example, in two dimen- tropic homogeneous system with isotropic
interfacial energy was determined numerically anisotropic system with a tetragonal
cally, see Fig. 5-21 b. Here the equilibrium misfit, and Schmidt and Gross (1997)
shapes were found as a function of the di- found strongly non-convex global energy
mensionless parameter L = d 2 C44 l/sab minimizing, pincushion-like shapes when
where C44 is an elastic constant in a system the particle is sufficiently soft in an elasti-
with cubic elastic anisotropy, and l is the cally anisotropic system. Similar pincush-
size of a particle, e.g., l = 3 V1/3/4 p where ion-like precipitates have been observed
V is the volume of a precipitate, L is the ra- experimentally, but in a system with a
tio of the characteristic elastic to interfacial much smaller difference in elastic con-
energy. When the particle is small, interfa- stants (Maheshwari and Ardell, 1992). Em-
cial energy is the dominant factor in setting phasizing the difficulty in obtaining true
the particle shape and as L increases, the equilibrium shapes experimentally, Wang
effects of elastic stress become important. and Khachaturyan (1995) have suggested,
When L is small and the equilibrium shape however, that the shapes observed by Ma-
has a fourfold symmetry, a cube-like mor- heshwari and Ardell are kinetic in origin.
phology with rounded corners and rounded In an internally nitrided Fe 3 at.% Mo al-
sides will result. As L increases, at a criti- loy, (Fe, Mo)16N2-type precipitates were
cal size there is a supercritical bifurcation found to nucleate as thin platelets with
to one of two plate-like shapes oriented K ≈ 0.1 and with an undistored {100}a-Fe
along the elastically soft directions. How- plane common to both the a-matrix and the
ever, due to the presence of interfacial en- nitrides. The observed morphology, as well
ergy, the equilibrium shapes above the bias the {100}a-Fe habit planes, have been
furcation point are not plates for any finite accurately predicted by applying macro-
L. In the limit of infinite L the equilibrium scopic linear elastic theory to tetragonal
shape is an infinitely long plate of zero Fe16N2 precipitates in ferritic iron, and as-
thickness (Khachaturyan, 1983). The loca- suming the interfacial energy to be iso-
tion of the bifurcation point is a function of tropic (Khachaturyan, 1983). The latter as-
both the elastic anisotropy and the differ- sumption, however, is debatable, since dur-
ence in elastic constants between the ing aging at 600 °C (Fe, Mo)16N2 platelets
phases (Schmidt and Gross, 1997). Three- only grow markedly in diameter; their
dimensional calculations have also been thickness remains almost constant (Fig.
performed where the particles are not con- 5-22). This has been interpreted in terms of
strained to be of a simple geometric shape the large difference between the interfacial
(Mueller and Gross, 1998, Lee, 1997; energy of the habit plane (≈0.05 J/m2) and
Thompson and Voorhees, 1999). that of the peripheral plane (≈0.3 J/m2)
Another qualitative difference between (Wagner and Brenner, 1978).
the possible equilibrium shapes of a parti- Under certain conditions, the elastic or
cle in the presence of stress and that in the magnetic interaction of precipitates with
absence of stress is that the morphology of external stress fields or magnetic fields al-
misfitting particles can be non-convex. In lows for the generation of anisotropic,
the absence of stress, regardless of the type highly oriented precipitate microstructures
of interfacial energy anisotropy, the equi- which are sometimes of technological im-
librium shape must be convex. In contrast, portance. If the transformation strain is
Thompson and Voorhees (1999) found non-spherical, such as for Fe16N2 in ferritic
metastable non-convex shapes in an elasti- Fe–N, the particles may interact with an
5.5 Early Stage Decomposition Kinetics 339
externally applied stress field. The result-

ing elastic interaction energy depends on
the particular orientation of the particle
with respect to this field. During aging, this
causes a selective coarsening of favorably
oriented particles at the expense of the en-
ergetically less favorably oriented ones.
For instance, if the originally isotropically
distributed Fe16N2 platelets undergo coars-
ening in the presence of an external tensile
stress applied along [001], a highly
oriented (anisotropic) precipitate structure
will result (Ferguson and Jack, 1984,
1985), with the habit planes of all particles
being parallel to each other and perpendic-
ular to [001]. In addition, these platelets
show a strong uniaxial magnetic anisotropy
with the direction of easy magnetization
parallel to the plate normal. This magnetic
property can also be used for completely
orienting the platelets during aging in an
external magnetic field (Sauthoff and
Pitsch, 1987).
5.5 Early Stage Decomposition

Kinetics
As outlined in Sec. 5.2.3, the early
stages of unmixing of a solid solution
quenched into the miscibility gap are trig-
gered by the growth and the decay of con-
centration fluctuations. Basically, the ob-
jective of any theory dealing with the ki-
netics of early stage decomposition is the
prediction of the particular shape, ampli-
Figure 5-22. FIM micrographs of (Fe, Mo)16N2

platelets (bright) in an a-iron matrix (dark). During
coarsening at 600 °C for the given times, the aspect
ratio decreases from ⬃0.1 (t = 0) to ⬃0.04 (623 h),
whereas the platelet thickness increases only from
⬃0.7 nm to ⬃1.0 nm. The platelets intersect the sur-
face of the semi-spherical field ion tip and thus ap-
pear to be curved (Wagner and Brenner, 1978).
tude and spatial extension (or number of at- scientist. For more recent developments,
oms) of a solute fluctuation which becomes see the chapter by Binder and Fratzl (2001).
critical and, hence, stable against decay.
Once formed, the critical fluctuations ren- 5.5.1 Cluster-Kinetics Approach
der the supersaturated alloy unstable with
5.5.1.1 Classical Nucleation –
respect to further unmixing.
Sharp Interface Model
According to Sec. 5.2.4 and referring to
Fig. 5-6 a, nucleation theories consider the Suppose the homogenized solid solution
formation rate of stable nuclei, the latter is quenched not too deeply into the meta-
representing spatially localized solute-rich stable regime of the miscibility gap (e.g., to
clusters (‘particles’ or ‘droplets’) with point 1 in Fig. 5-5 a). There it is isother-
large concentration amplitudes. In this con- mally aged at a temperature sufficiently
text, a distinction is made between classi- high for solute diffusion. After a certain
cal (Sec. 5.5.1.1) and non-classical nuclea- time, a distribution of microclusters con-
tion theory (Sec. 5.5.1.4) depending on taining i atoms (i-mers) will form in the
how the Helmholtz energy of the non-uni- matrix.
form solid solution containing the cluster Generally, classical nucleation is now
distribution is evaluated. Once the Helm- based on both a static and a dynamic part.
holtz energy has been specified, the equi- In the static part the changes of Helmholtz
librium distribution of heterophase fluctua- energy associated with the formation of an
tions and the nucleation barrier (Sec. 5.2.3) i-mer and the cluster distribution f (i ) must
can be calculated. The nucleation rate is be evaluated. In the dynamic part the kinet-
then obtained as the transfer rate at which ics of the decay of the solid solution which
smaller clusters attain the critical size. now is described by the given distribution
A ‘composition wave’ picture rather of non-interacting microclusters, are calcu-
than a ‘discrete droplet’ formalism is em- lated in terms of the time evolution of f (i );
ployed in the spinodal theories (Sec. 5.5.4) ultimately this will furnish the formation
which, as another approach, describe the rate of stable clusters, i.e. the nucleation
early-stage decomposition kinetics in rate.
terms of the time evolution of the ampli- Classical nucleation theory treats the so-
tude and the wavelength of certain stable lute fluctuations as droplets which were cut
‘homophase’ fluctuations (Fig. 5-6 b). from the equilibrium precipitate phase b
There are several papers dealing with the and embedded into the a matrix; in this
early-stage unmixing kinetics. Russel capillarity or droplet model the interface
(1980) and Aaronson and Russel (1982) between a and b is assumed to be sharp,
consider both classical and non-classical e.g., Fig. 5-6 a. In essence, the approxima-
nucleation phenomena from a more metall- tion reduces the number of independent
urgical point of view; the formulation of a variables which characterize a cluster, and
microscopic cluster theory of nucleation which may vary during the nucleation pro-
can be found in the article by Binder and cess (e.g., the solute concentration, the
Stauffer (1976). The comprehensive arti- atomic distribution within the cluster, its
cles by Martin (1978) and by Gunton and shape, the composition profile across the
Droz (1984) disclose developments of both interface, etc.) and, hence, determine the
nucleation and spinodal concepts not only Helmholtz energy of the system, to virtu-
to the theoretican but also to the materials ally a single one. That is the number i of
atoms contained in the cluster or, in terms

of a more macroscopic picture, the radius R
of the droplet; (4 p /3) R3 = i Wb if Wb de-
notes the atomic volume in the droplet.
Once the small nucleus containing only a
few atoms is treated as a droplet of the new
phase b having bulk properties (which, in
fact, is a rather debatable approximation),
the interfacial Helmholtz energy and the
Helmholtz energy are also considered to be
entirely macroscopic in nature.
The formation of a coherent droplet with
radius R which gives rise to some elastic
coherency strains, leads to a change of Figure 5-23. Schematic representation of Helmholtz
Helmholtz energy, energy changes associated with cluster formation as
(5-18)
a function of cluster radius (R) or number i of atoms
4p 3
DF ( R) = ( D fch + D fel ) R + 4 p R2 s ab in the cluster. DF 0 is the Helmholtz energy of the
3 homogeneous solid solution, Z the Zeldovich factor.
where, according to Sec. 5.2.4, (D fch +
D fel) is the driving force per unit volume
and sab the specific interfacial energy. dius R* of the critical nucleus are predicted
The first term in Eq. (5-18) which scales to grow continuously. This requires a fluc-
with R3 accounts for the gain of Helmholtz tuation to become a critical nucleus, first to
energy on forming the droplet (i.e., it is overcome the activation barrier for nuclea-
negative). The second term which scales tion, or the nucleation energy,
with R2 has to be expanded on forming the 3
(5-20)
16 p s ab
interphase boundary and hence is a positive DF ( R*) ≡ DF* =
contribution to DF (R). 3 ( D fchem + D fel )2
Fig. 5-23 shows the dependence of the As outlined in Sec. 5.2.4, D fchem de-
two contributions in Eq. (5-18) on the creases strongly with decreasing supersatu-
droplet radius. The resulting droplet forma- ration, or, for a given concentration, with
tion energy DF (R) passes through a maxi- increasing aging temperature (Fig. 5-1).
mum at R ∫ R* or i ∫ i* with Correspondingly, since s ab and D fel are
2 s ab only weakly dependent on temperature,
R* = or both R* and D F * increase strongly with
– ( D fchem + D fel )
decreasing supersaturation. Hence, the
3
4p ⎡ 2 s ab ⎤ droplet model ought to approximate the en-
i* = (5-19)
3Wb ⎣ – ( D fchem + D fel ) ⎥⎦
⎢ ergetics of an unmixing alloy better the
smaller the supersaturation.
Accordingly, clusters with R = R* are in Once we have specified the free energy
unstable equilibrium with the solid solu- of formation of a droplet in terms of its size
tion, i.e., the Helmholtz energy of the R or i (Eq. 5-18)), the Helmholtz energy of
system is lower if it contains clusters with the system containing a number of f (i, t) of
sizes below or beyond R* or i*. Therefore, non-interacting clusters of size i per unit
only clusters with radii that exceed the ra- volume at time t is given as (Frenkel, 1939;
Russell, 1980): a (i + 1) derived for the equilibrium situa-

n tion (‘principle of detailed balance’) is still
F = ∑ DF (i ) f (i, t ) – T Smix (5-21) valid for the non-equilibrated system
1
where f (i, t) ≠ C (i ) and Ji ≠ 0. Under such
The entropy of mixing Smix arises from an assumption, which becomes reasonable
distributing the clusters on the available N0 when i-mers are able to relax internally
lattice sites per unit volume of the crystal. between atomic condensation or evapora-
The kinetics of early-stage unmixing are tion, the flux of clusters between size
governed by the change in the cluster size classes i and i + 1 is obtained from Eqs.
distribution function f (i, t) with aging (5-22) and (5-23) (Russell, 1980):
time. Microscopically, this may occur via
⎡ f (i + 1, t ) f (i, t ) ⎤
different processes. Ji = b ( i ) C ( i ) ⎢ –
Volmer and Weber (1926), Becker and ⎣ C (i + 1) C (i ) ⎥⎦
Döring (1935) and Zeldovich (1943), the ⎡ ∂ f ( i, t ) / C ( i ) ⎤
works which most nucleation theories are ≈ – b (i ) C (i ) ⎢ ⎥⎦ (5-25)
⎣ ∂i
based on, assume the transitions between
size classes in an assembly of non-interact- In the earliest theory on nucleation
ing droplets to occur via the condensation (Volmer and Weber; VW, 1926) it is as-
or evaporation of single solute atoms. sumed that f (i, t) ∫ 0 for clusters with
Hence, since only a transition between size i > i* (or R > R*), and that clusters with R
classes i and i + 1 is allowed, its flux, Ji is > R* decay artificially into monomers,
given as (Russell, 1980): thus keeping the matrix supersaturation
about constant. The resulting quasi-steady-
Ji = b (i ) f (i, t) – a (i + 1) f (i + 1, t) (5-22)
state distribution of cluster sizes is ob-
where b (i) is the condensation rate and tained from Eq. (5-23) and shown in Fig.
a (i + 1) the evoporation rate of a single 5-24). In this theory the steady-state nucle-
s
atom to a cluster of size i or from a cluster ation rate JV–W is obtained as the product
of size (i + 1), respectively. of the number C* of critical nuclei and the
In equilibrium, the fluxes Ji must vanish rate b at which a single solute atom im-
and f (i, t) becomes identical with the equi- pinges on the critical nucleus rendering it
librium cluster size distribution C (i ) for supercritical, i.e.:
which F in Eq. (5-21) attains a minimum: s
JVW = b C* = b N0 exp (–D F */kT ) (5-26)
C (i) = N0 exp [–DF (i )/kT] (5-23)
One of the shortcomings of the VW the-
It is worth noting that in this case the ory is that supercritical droplets with
number of critical nuclei C* ∫ C (i ∫ i*) is i > i* are assumed not to belong to the
proportional to exp (–DF */kT ). Then, ac- cluster size distribution. This was cleared
cording to Eq. (5-22), the condensation up by the theory of Becker and Döring
rate and evaporation rate are related to each (1935). In their theory (BD) the non-equi-
other via librium steady-state distribution of small
clusters with i Ⰶ i* is identical with that of
C (i )
a (i + 1) = b (i ) (5-24) VW but unlike in the latter theory, clusters
C (i + 1) with i* ⬉i ⬉ic (ic is a somewhat arbitrarily
In the conventional nucleation theory it chosen cut-off size) are considered to be-
is now assumed that the evaporation rate long to the size distribution (Fig. 5-24).
The Zeldovich factor Z is of the order of

1/20 to 1/40; graphically, its reciprocal
value corresponds approximately to the
width of the potential barrier, DF (R) or
D F (i ), at a distance kT below the maxi-
mum (Fig. 5-23).
5.5.1.2 Time-Dependent Nucleation

Rate
It must be pointed out that the above
given steady-state nucleation theories do
not provide any information on the mo-
Figure 5-24. Quasi-stationary cluster size distribu- mentary cluster size distribution or on the
tions of the Volmer–Weber (V–W) and Becker– nucleation rate prior to reaching steady-
Döring (B–D) theory. state conditions, i.e., the time-dependent
nucleation rate J* (t). Commonly the latter
is considered in terms of the steady-state
Hence, according to Eq. (5-22) a decay
nucleation rate J s and an incubation period
or dissolution of supercritical droplets with
t via:
i > i* becomes a likely process in the BD
theory and is accounted for by the Zeldo- J* (t) = J s exp (– t /t) (5-29)
vich factor Z. With the assumption that the
Steady-state will be achieved once the
rate b ∫ b* at which a solute atom impinges
clusters have attained sizes for which the
on a critical droplet is proportional to its
probability of redissolution is negligibly
surface area, integration of the cluster flux
small. Referring to Fig. 5-23 and recalling
equation (Eq. (5-25)) yields the steady-
S the physical meaning of the Zeldovich fac-
state nucleation rate JBD of the BD theory
tor Z, this will be the case for clusters with
as (Russell, 1980):
i > i* +1/2 Z (Feder et al., 1966). As the
s
JBD = Z b * N0 exp (–D F */kT ) (5-27) gradient of D F (i ) within the region 1/Z is
rather small, the clusters will move across
with
1/ 2
this region predominantly by random walk
⎡ – 1 ∂ 2 ∆F ⎤ with the jump frequency b *. The time
Z=⎢ ⎥ (5-28)
⎢⎣ 2 p kT ∂i i = i* ⎥⎦
2 1
t= (5-30)
2 b* Z
According to Eq. (5-27), nucleation is a
thermally activated process with an activa- to cover the distance 1/Z by random walk is
tion energy identical to that (DF *) of identified with the incubation period.
forming a critical nucleus of size i* or R*. So far it has been assumed that during
Furthermore, like the thermodynamic nucleation the supersaturation and, hence,
model that yields the number of critical the driving force remains unchanged. At its
nuclei proportional to exp (– DF */kT ) (Eq. best this may be valid for extremely low
(5-23)), the kinetic treatment predicts the nucleation rates: in this case D F *, Z and t
steady-state nucleation rate also to be pro- may be considered time-independent quan-
portional to the same exponential factor. tities. Russell (1980) proposed the nuclea-
tion process to terminate itself under the expressions for D F *, Z, b* and t given
steady-state conditions if the incubation above for spherical clusters, have merely to
time t is shorter than some critical time tc. be multiplied with numerical factors but
If, however, t > tc , phase separation will be otherwise remain unchanged. The magni-
completed without steady-state ever having tude of these numerical factors depends on
been achieved. This situation, which in fact the particular equilibrium shape of the nu-
is met in most decomposing alloys studied cleus. Ignoring volume strain energy, the
up to now (see Sec. 5.7.4.1), is referred to equilibrium shape of the nucleus can be de-
as catastrophic nucleation; under such termined by means of the Wulff construc-
conditions D F *, Z and t become time de- tion on a three-dimensional sab-plot (Her-
pendent. Assuming that nucleation ceases ring, 1953). The latter is obtained from a
once the diffusion fields and around pre- radial plot of different vectors in every di-
cipitates with radius R and composition cbe rection e.g., v 1, v 2, v 3 in Fig. 5-25), the
embedded in a matrix with initial composi- length of which is proportional to the en-
tion c0 begin to overlap (Fig. 5-6 a), and ergy of the interface perpendicular to the
furthermore that nucleation becomes un- particular vector. The surface connecting
likely within a region R/e (e ≈1/10) around the tips of all vectors represents the polar
the particle where the solute content has sab-plot, wherein the cusps indicate inter-
decreased below (1 – e) c0, the critical time faces with good atomic matching , i.e., low
tc is estimated to be: interfacial energies. Subsequently planes
1/ 5
⎡ ⎛ ce ⎞ 3 ⎤
β 1 1 ⎥
tc = ⎢e 6 ⎜ ⎟ (5-31)
⎢ ⎝ c0 ⎠ ( 2 D)3 J s ⎥
2
⎣ ⎦
where D is the solute diffusion coefficient.
For a quantitative assessment of classi-
cal nucleation theory, the atomic impinge-
ment rate b* in Eq. (5-27) must be known.
For spherical nuclei it was evaluated to be
(Russell, 1970):
4 p R *2 Dc0
b* = (5-32)
a4
where a is the lattice parameter. b* is pro-
portional to the nucleus surface, as was as-
sumed in the original Becker–Döring the-
ory.
According to Sec. 5.4.2 the nucleus
shape with minimum energy may deviate
from the spherical one due to different en- Figure 5-25. Polar sab-plot for particles with aniso-
ergies of the interfaces (some may be co- tropic interfacial energies sab. The normals drawn at
herent, some semi- or incoherent, Sec. the tip of each vector v are the Wulff planes, the inner
envelope of which gives the equilibrium shape of the
5.2.1) bounding the nucleus in different precipitate. If the sab-plot displays deep cusps (e.g.,
crystallographic directions. It has been for v 1, v 3, and v 5) the Wulff construction yields a fac-
shown (Chan et al., 1978) that in this case eted polyhedron.
are drawn (the so-called Wulff planes) per- face between Al-rich matrix and Ag-rich
pendicular to the vectors v 1, v 2, v 3, etc. GP zones are found to be equal within 5%,
which intersect the sab-plot (Fig. 5-25). and to increase from zero to about
The inner envelope of the Wulff planes 30 mJ/m2 if the temperature is lowered
yields the equilibrium shape of the nucleus from Tc to ~ 0.5 Tc. This result is compat-
with facets at the cusps of the sab-plot ible with the spherical particle morphology
(Martin and Doherty, 1976). resolved in the HREM micrographs (Figs.
LeGoues et al. (1982) employed a near- 5-9 and 5-13).
est-neighbor interaction, regular solution
model for a computation of the sab-plot as
5.5.1.3 Experimental Assessment of
a function of T/Tc (Tc is the critical temper-
Classical Nucleation Theory
ature) within the miscibility gap of an f.c.c.
solid solution. At low temperatures they A quantitative assessment of classical
found the nucleus shape to be fully faceted nucleation theory in solids is inherently
by {100} and {111} planes whereas at 0.5 difficult and thus has prompted only a few
Tc and near-zero supersaturation, the nu- studies. First of all, the range of supercool-
cleus shape can be rather well approxi- ing (or supersaturation) has to be chosen
mated by a sphere. They furthermore such that nucleation is homogeneous and
showed that beyond T ≈0.4 Tc the parame- the nucleation rates are neither unmeasure-
ters DF *, Z, t, and b* which enter the time- ably slow nor beyond the limits of catas-
dependent (classical) nucleation rate (Eqs. trophic nucleation. Secondly, the quench-
(5-27) and (5-29)) need no longer be cor- ing rate must be sufficiently high in order
rected for deviations from spherical shape. to avoid phase separation during the
Recently the question of to what extent quench but also sufficiently low in order to
the energy of the coherent interphase boun- avoid excess vacancies to be quenched in.
dary depends on its crystallographic orien- Servi and Turnbull (1966) studied in a
tation and on temperature has been read- well-designed experiment homogeneous
dressed by employing the cluster variation nucleation kinetics of coherent Co-rich
method (CVM) with effective cluster inter- precipitates in Cu–1 … 2.7 wt.% Co al-
action parameters (Sluiter and Kawazoe, loys. By using electrical resistivity meas-
1996; Asta, 1996). The latter are specific urements, they could determine rather ac-
for the particular two-phase alloy system curately the precipitated volume fraction.
under consideration. They can be derived Assuming that the growth is diffusion-con-
from the results of ab initio total energy trolled (Sec. 5.5.2), from the latter the par-
calculations as was demonstrated by com- ticle density at the end of the precipitation
puting the thermodynamic properties of process could be derived as a function of
interphase boundaries between disordered aging temperature and composition. The
f.c.c. matrices and ordered (L12 type) pre- thus indirectly obtained number density of
cipitates in Al–Li (Sluiter and Kawazoe, Co-rich particles, which was later corrobo-
1996) and Al–Sc (Asta et al., 1998) as well rated by CTEM studies, agreed within one
as disordered GP zones (cf. Fig. 5-13) in order of magnitude with the value pre-
Al–Ag alloys (Asta and Hoyt, 1999). dicted by classical nucleation theory if the
Between Tc and T = 0.5 Tc the CVM-calcu- interfacial energy was taken as ≈0.19 J/m2;
lated energies for the cube {100} and octa- this value was derived from discrete lattice
hedral {111} orientations of the flat inter- calculations (Shiflet et al., 1981).
The validity of classical nucleation the-

ory, as proven by the Servi–Turnbull study
on Cu–Co alloys, was later challenged by
LeGoues and Aaronson (1984). They
argued that the supersaturation employed
in the Servi–Turnbull investigations was
probably too high to avoid phase separa-
tion during quenching and also probably
too high to avoid concomitant coarsening
during the precipitation reaction. Employ-
ing a discrete lattice point model, which in-
corporates coherency strain energy, Le-
Goues and Aaronson first evaluated the
‘window’ of temperatures (DT ≈50 °C) and
compositions (0.5 to 1.0 at.% Co) at which
homogeneous nucleation kinetics would be
neither too sluggish nor too fast, and at
which no interference with coarsening
would be expected. Prior to CTEM analy-
ses, the isothermally nucleated particles
had to be subjected to diffusion-controlled
growth in order to increase their radius be-
yond a certain size (R ≈5 nm) at which the
Co-rich particles became easily discernible
in the CTEM. The experimental results
were interpreted in terms of classical nu-
cleation theory (Eq. (5-29)). The agree-
ment between the experimentally obtained
nucleation rates and the theoretically pre- Figure 5-26. Nucleation barrier DF * (a) and critical
radius R* (b) as a function of supersaturation at T =
dicted ones was again rather good, thus 0.25 Tc according to classical theory (cl. th.), non-
providing further support for the validity of classical Cahn–Hilliard continuum model (C–H) and
classical nucleation theory. Furthermore, discrete lattice model (DLM). R* is determined such
as is shown in Fig. 5-26, for smaller super- that it corresponds to (c0 + cc)/2 where cc is the com-
saturations the nucleation energies DF * position of the nucleus at its center. (After Le Goues
et al., 1984 a).
and the critical radii R* as evaluated from
classical nucleation theory are almost iden-
tical to the corresponding quantities calcu-
lated from either the non-classical nuclea- (Kirkwood, 1970; West and Kirkwood,
tion theory (cf. Sec. 5.5.1.4, Eq. (5-37)) or a 1976; Hirata and Kirkwood, 1977) studied
discrete lattice point theory. Hence, at least early-stage precipitation of g ¢-Ni3Al in
for smaller supersaturation, the classical Ni–Al alloys, also using CTEM. They
theory predicts the nucleation rates about as found the g ¢-Ni3Al precipitate number den-
well as the two more sophisticated theories. sity to decrease instantaneously upon ag-
In another attempt to assess classical nu- ing, which is indicative of an extremely
cleation theory, Kirkwood and coworkers fast nucleation process and the observation
of a coarsening process rather than a nucle- as:

ation event. By making certain assumptions DF = ( F – F0 ) (5-33 c)
they attempted to infer the steady-state nu-
cleation rate from the particle density ob- = ∫ [ f ( c ) – f ( c0 ) + K * (∇ c )2 ] dV
V
served during coarsening. The observed
lack of agreement between the experimen- The basic idea behind these expressions is
tally derived nucleation rates and the theo- to subdivide the solid with volume V into
retically predicted ones may be seen to re- many small volume elements dV, a proce-
sult from these experimental difficulties. dure which is often referred to as ‘coarse
graining’ (Langer, 1971). The free energy
of each volume element is taken as f (c) dV,
5.5.1.4 Non-Classical Nucleation – f (c) is considered to be the Helmholtz en-
Diffuse Interface Model ergy per unit volume of the bulk material
In the droplet model that classical nucle- with composition c which is equal to the
ation is based on, it is assumed that the mean composition c (r) within dV located
composition of the nucleus is more or less at position r. In essence, coarse graining re-
constant throughout its volume and that its quires the number of atoms within each
interface is sharp. This made it possible to volume element to be sufficiently large
take the change of volume Helmholtz en- such that c (r), as well as f (c), can be spec-
ergy and the interfacial Helmholtz energy ified as continuous functions. On the other
separately into account (Eq. (5-18)). In hand, the number of atoms within dV must
non-classical nucleation theory, developed be small enough in order to avoid phase
by Cahn and Hilliard (1959 a, b), the inho- separation within the individual volume
mogeneous solid solution in its metastable element. For the non-uniform system with
state is considered to contain homophase composition fluctuations, Cahn and Hilli-
fluctuations with diffuse interfaces and a ard (1958, 1959 a, b) assumed that the local
composition which varies with position Helmholtz energy contains further terms
throughout the cluster (Fig. 5-6 b). Hence, depending on the composition gradient
unlike in the droplet model a critical fluctu- (—c). They finally showed that the non-uni-
ation has now to be characterized by at form environment of atoms in a composi-
least two parameters, its spatial extension tion gradient may be accounted for by add-
or wavelength l and its spatial composition ing to the local Helmholtz energy a single
variation. The necessary Helmholtz energy gradient energy term which is proportional
formalism, which no longer treats volume to (—c)2. Hence, the resulting Helmholtz
energy and surface energy separately, was energy of a volume element dV is ex-
elaborated by Cahn and Hilliard (1958). pressed as f (c) + K* (—c)2 where the con-
They wrote the Helmholtz energy change stant K* denotes the gradient energy coeffi-
associated with the transfer from the homo- cient. Summing up all contributions from
geneous system with composition c0, the various volume elements yields Eq.
(5-33 c).
F0 = ∫ f ( c0 ) dV (5-33 a) The continuum model is base on the as-
V
sumption that the Helmholtz energy f (c)
to that of the inhomogeneous system with varies smoothly with composition, i.e. the
F = ∫ [ f ( c ) + K * ( ∇ c ) 2 ] dV (5-33 b) wavelength of the fluctuations must be
V large compared with the interatomic spac-
ing. Furthermore, for a continuum model 1982),

its is required that the two evolving phases
are fully coherent with each other and thus ∆F [c *( r )] ≡ ∆F * (5-37)
∞
have the same crystal structure with similar ⎡ ⎛ dc ⎞ ⎤ 2
2
lattice parameters. If the lattice parameter = 4p ∫ ⎢ ∆ f ( c*) + K * ⎝ ⎠ ⎥ r dr

0 ⎣ dr ⎦
changes with composition, which is com-
monly the case in crystalline solids, the re- attains a minimum. Like in classical nucle-
quirement for coherency leads to cohe- ation, D f * is the vertical distance from the
rency strains which according to Eq. (5-12) tangent at c0 to the Helmholtz energy curve
are accounted for by an elastic energy term at c* (cf. Fig. 5-5 b). Suppose D F* is
known; the nucleation rate is then obtained
fel (c) = h2 Y (c – c0)2 (5-34) from Eq. (5-27) or (5-29). Provided the
Combining Eqs. (5-34) and (5-33c) yields constraint free energy is known, the com-
position profile of a critical nucleus in a
DF = ∫ [ f ( c ) – f ( c0 ) + K * (∇ c )2 solid solution with composition c0 and gra-
V dient energy coefficient K* (K* is of the or-
+ h 2 Y ( c – c0 )2 ] dV (5-35) der 10 to 100 J m–1 (at. fract.)–2) can be ob-
This expression may be interpreted in sim- tained from integrating Eq. (5-36). Fig.
ilar terms as for classical nucleation [Eq. 5-27 shows the composition profiles of nu-
(5-18)]: neglecting the elastic energy term, clei in supersaturated solid solutions with
the positive contribution of the gradient en- different solute concentrations c0 quenched
ergy as a barrier to nucleation acts like the to T/Tc = 0.25, where Tc is the critical tem-
surface Helmholtz energy in the droplet perature (LeGoues et al., 1984 b); these cal-
model and is finally overcome by the gain culations are based on the regular solution
in chemical Helmholtz energy once the model for f (c) (cf. Sec. 5.2.4). For small
composition difference between the fluctu- supersaturation (c0 = 2 ¥ 10–3) the compo-
ation and the homogeneous matrix has be- sition within the nucleus is constant and
come sufficiently large. corresponds to cbe of the equilibrium b-
Assuming isotropy, the composition pro- phase. The interface is almost sharp, simi-
file c (r) of a spherical fluctuation (r is the lar to that assumed in the classical droplet
radial distance from the fluctuation center) model. Correspondingly, the free energy of
is obtained from a numerical integration of formation of the nucleus (DF *) and its size
(Cahn and Hilliard, 1959 a, b): (R*) as evaluated from either classical or
non-classical theory are almost identical
d 2c K * dc ∂f ∂f (Fig. 5-26) and tend to infinity as c0 ap-
2K * 2 +4 = – (5-36) proaches cae . With increasing supersatura-
dr r dr ∂c c ∂c c0
tion the composition profile becomes in-
with the boundary conditions dc/dr = 0 at creasingly diffuse and the solute concentra-
the nucleus center (r ≈ 0) and far away from tion at the nucleus center decreases (Fig.
it (r Æ •) where c ∫ c0. The critical nu- 5-27); this situation is no longer accounted
cleus is then determined as the fluctuation for by the droplet model. Furthermore, as
which, like a critical droplet, is in unstable the initial composition (c0) approaches the
equilibrium (Eq. (5-35)) with the matrix. spinodal one (csa ), unlike the droplet
Its composition is established such that model, non-classical theory predicts DF *
the nucleation barrier (e.g., de Fontaine, to go to zero (Fig. 5-26 a), and the spatial
less, their formation energies are only

slightly higher than DF *. In this case, the
probability of the alloy decaying via the
formation of those ‘short-wavelength’
fluctuations requiring a slightly higher ac-
tivation energy is much higher than for a
decay via the formation of critical ‘long-
wavelength’ fluctuations.
5.5.1.5 Distinction Between Classical

and Non-Classical Nucleation
For a practical application of non-classi-
cal nucleation theory to experimental stud-
ies of unmixing, the total constraint free
energy f¢ (c) rather far away from equilib-
Figure 5-27. Composition profile of a spherical nu- rium, as well as K*, have to be known in
cleus (center at r = 0) in four different metastable order to calculate the composition profile
solid solutions with composition c0 quenched to T = [Eq. (5-36)] and the nucleation energy [Eq.
0.25 Tc into the miscibility gap. The continuous line
was calculated from the Cahn–Hilliard continuum
(5-37)]. This is the case only for simple al-
model (Eq. (5-36)), the crosses represent calcula- loy systems and, hitherto, has inhibited the
tions based on the discrete lattice point model (after application of the non-classical theory to
LeGoues et al., 1984 a). The computations are based either scientifically or industrially impor-
on a regular solution model with the solubility limit tant alloy systems. Furthermore, so far it
cae = 0.37 ¥ 10–3 and the spinodal composition cas =
6.41 ¥ 10–2, a is the lattice parameter of an assumed
has been assumed that a nucleus is in equi-
f.c.c. lattice and l the interfacial width. librium with the infinite matrix and that the
nuclei do not interact during the nucleation
stage. This may be a valid approximation
as long as the supersaturation is small, i.e.,
extent of the critical fluctuations (or R*) to where according to Figs. 5-26 and 5-27
increase again; at c0 ∫ csa , R* finally be- classical nucleation theory applies. It ought
comes infinite (Fig. 5-26 b) suggesting a to become, however, a poor approximation
discontinuity of the decomposition mecha- for more concentrated alloys where non-
nism at the spinodal line (cf. Sec. 5.2.3). classical nucleation applies. In this case the
This apparent discontinuity at the spino- nucleation rate is high (then the steady-
dal results from the assumption that the de- state nucleation regime will not be ob-
composition path is controlled by the mini- served) and, hence, the nucleus density
mum height of the nucleation barrier, i.e., large. These complications in non-classical
by DF *, which close to the spinodal is in nucleation make it desirable to provide the
the order of a few k T (Fig. 5-26 a) and, experimentalist with a criterion for the ap-
hence, rather low. As pointed out by de plicability of classical nucleation theory
Fontaine (1969), in a system close to the which is much easier to handle. Cahn and
spinodal many fluctuations exist, the spa- Hilliard (1959 a, b) suggest classical theory
tial extent of which is much smaller than will apply if the width l of the (diffuse)
that of the critical fluctuation; neverthe- interface is considerably smaller than the
core of the nucleus where the composition Although some caution is advisable in
is about constant (Fig. 5-27). treating these tiny particles (critical nuclei
If we assume that a perceptible nuclea- in solid-state transformation are tens to
tion rate is obtained for D F * ≤ 25 kT (in hundreds of atoms in size) in terms of a
the literature, DF * ≈ 60 kT is frequently continuum theory1, and assigning them a
stated for the maximum barrier at which macroscopic surface and thermodynamic
nucleation becomes measurable; this value, bulk properties, classical nucleation theory
however, is much too high for homogene- seems to be appropriate for an estimation
ous nucleation to occur in solids) and em- of the nucleation rate. This conclusion will
ploys Eqs. (5-19) and (5-20), classical nu- be corroborated in Sec. 5.7.4.
cleation theory may be applied if (Cahn
and Hilliard, 1959 a, b): 5.5.2 Diffusion-Controlled Growth
of Nuclei from the Supersaturated Matrix
4 p s ab
l2 Ⰶ1 (5-38) Suppose that after nucleation the stable
75 kT nucleus is embedded in a still supersatu-
rated matrix. As is illustrated in Fig. 5-28,
For Cu–Ti (s ab = 0.067 J/m2; T = 623 K), the particle will then be surrounded by a
Fe–Cu (s ab = 0.25 J/m2; T = 773 K), and concentration gradient which provides the
Cu–Co (s ab = 0.17 J/m2; T = 893 K) this driving force for solute diffusion, and thus
means that the interfacial width has to be gives rise to its growth.
smaller than ≈ 0.56, 0.33, and 0.42 nm, re- The growth rate can be controlled either
spectively, and, hence, has to be rather by the rate at which atoms are supplied to
sharp. So far it has not been possible to the particle/matrix interface by diffusion or
measure the composition profile across a by the rate at which they cross the interface
nucleus/matrix interface. However, analyt- (Tien et al., 1973). It may be rationalized
ical field ion microscopy of Cu–1.9 at.% Ti that for small particles the interface reac-
aged at 350 °C for 150 s revealed the Ti tion is likely to be the rate-controlling step
concentration of particles with radii of only since the diffusion distances are rather
≈1 nm to decrease from 20 at.% (corre- short; once the particles have grown to a
sponding to Cu4Ti, cf. Sec. 5.2.1) to that of certain size, the matrix will be depleted
the matrix within one to two atomic (111)- from solute atoms and the associated re-
layers (von Alvensleben und Wagner, duction of the driving force makes diffu-
1984). For the chosen aging conditions nu- sion likely to be the slower and, thus, rate-
cleation was evaluated to terminate within controlling step (Shewmon, 1965). The
about 60 s (Kampmann and Wagner, 1984), transition from one step to the other de-
thus after aging for 150 s the analyzed par- pends upon the relative magnitudes of so-
ticles must have already experienced some lute diffusion and interface mobility.
growth beyond the original size of the nu- After termination of nucleation it is com-
cleus. Nevertheless, these results provide monly assumed that the mobility of the
rather good evidence that during nucleation interface is sufficiently high in order to al-
the Cu4Ti clusters may be considered drop-
lets with sharp interfaces rather than spa-
1
tially extended (long-wavelength) fluctua- Many FIM studies of two-phase alloys (e.g., Cu–1
at.% Fe) revealed the shapes of even tiny clusters
tions with diffuse interfaces, even though with as little as about twenty atoms to be already
DF * is only about 10 kT. compact rather than ramified (Wagner, 1982).
in a manner which, in particular for larger

values of k*, depends upon the approxima-
tion assumed (Aaron et al., 1970). For pre-
cipitating intermetallic compounds, k* is
one or two orders of magnitude larger (e.g.,
≈ 0.3 for g ¢-Ni3Al in Ni–Al and ≈ 0.2 for
b¢-Cu4Ti in Cu–Ti) than for precipitating
phases with cp ≈ 1 such as Fe–Cu, Cu–Fe
and Cu–Co.
For small supersaturations, c (r, t) in Eq.
(5-39) may be approximated as being time-
Figure 5-28. Schematic concentration field in the independent. In this case, an isolated spher-
matrix surrounding a nucleus with radius R and com-
position cp.
ical particle with radius R surrounded by
the concentration field illustrated in Fig.
5-28 will grow at a rate
low the solute concentration cR at the dR c ( t ) – cR D
curved interface to achieve local equilib- ≡ v ( R) = (5-43)
dt c p – cR R
rium (Doherty, 1983). In this case diffusion
is the rate-controlling step for the growth where D (assumed to be independent of
of stable coherent nuclei in a homogeneous composition) is the volume diffusion coef-
precipitation reaction. An analytical solu- ficient in the matrix. According to the
tion of the adequate field equation, Gibbs–Thomson equation, the composition
cR of the matrix phase at a curved interface
∂c ( r , t ) is different from that of a flat interface, the
D∇ 2 c ( r , t ) = (5-39)
∂t latter being in equilibrium with the equilib-
requires certain approximations (Zener, rium solute concentration cae of the a-ma-
1949). These were critically examined and trix phase, and varies with the precipitate
compared by Aaron et al. (1970) for the radius as (Martin and Doherty, 1976)
diffusion-controlled growth of spherical (5-44)
and platelet-shaped particles. Regardless of cR ( R) 1 – ca ⎛ 2 s ab Vb ⎞
e
g (c )
ln e = e ⎜ ⎟ – ln a Re
the particular approximation used, both the ca cp – ca ⎝ Rg T R ⎠ g a (ca )
radius R of a spherical precipitate and the
half-thickness T of a platelet grow with time with ga (cR) and ga (cae ) being the activity
according to a parabolic growth law as: coefficients of the solute atoms in the a-
phase at the concentration cR and cae , re-
R = l i (D t)1/2 (5-40) spectively. If the solid solution shows reg-
and ular solution behavior (ga then becomes in-
dependent of c), Eq. (5-44) yields the well-
T = l j (D t)1/2 (5-41) known Gibbs–Thomson equation:
respectively. The rate constants li and lj in- ⎛ 2 s ab Vb 1 ⎞
crease with increasing supersaturation, or, cR ( R) = cae exp⎜ ⎟ (5-45)
⎝ Rg T R ⎠
more specifically, with an increase in the
factor or for larger radii in its linearized version,
c ( t ) – cR ⎛ 2 s ab Vb 1 ⎞
k* = 2 (5-42) cR ( R) = cae ⎜ 1 + (5-46)
c p – cR ⎝ Rg T R ⎟⎠
By assuming a monodispersive particle into a bulk and a surface term appears rather
distribution and cR ≈ c̄ (t Æ •) = cae (or c¢ae), debatable. To avoid some of these deficits,
and c̄ (t) ≈ c0 , integration of Eq. (5-43) yields there have been several attempts to develop
1/ 2 more accurate descriptions of clusters, tak-
⎛ c – ce ⎞
R (t ) = ⎜ 2 0 ae ⎟ ( D t )1 / 2 (5-47) ing into account their different sizes and
⎝ cp – ca ⎠ shapes (e.g., Binder and Stauffer, 1976;
It should, however, be pointed out that, so Stauffer, 1979), and to derive an equilibrium
far, there is little experimental evidence for distribution of clusters which is more realis-
the existence of a precipitation regime dur- tic than the one given by Eq. (5-23). As in
ing which particle growth strictly follows the classical nucleation theory, the latter is
Eq. (5-47) (Kampmann and Wagner, 1984). an important quantity for a computation of
In Sec. 5.7.4.3 the reason for this lack of the cluster formation rate in terms of the
experimental evidence will be provided, as more recently elaborated theories of cluster
well as a guideline to the design of experi- dynamics which, in essence, are extensions
ments that allow a verification of the exis- of the Becker–Döring theory (Binder and
tence of diffusion-controlled particle Stauffer, 1976; Penrose and Lebowitz,
growth with the predicted kinetics to be 1979; cf. Gunton et al., 1983 and the chapter
made. Most quantitative experimental ob- by Binder and Fratzl (2001) for a general
servations of growth rates which yielded discussion of the various developments).
good agreement with the diffusion-con- Unlike Becker and Döring, however,
trolled models outlined above, were con- who confined the microscopic mechanism
fined to large particles, frequently with of cluster growth or shrinkage to the con-
sizes in the micrometer range (Aaronson et densation or evaporation of monomers
al., 1970 a, b; Doherty, 1982). (Sec. 5.5.1.1), in their generalized nuclea-
tion theory Binder and coworkers consider
the time evolution of the cluster size distri-
5.5.3 The Cluster-Dynamics Approach
bution function f (i, t) more generally in
to Generalized Nucleation Theory
terms of a cluster coagulation or cluster
As has been pointed out in Sec. 5.5.1.1, splitting mechanism, i.e., in a single cluster
the cluster-dynamics approach chosen by reaction step i-mers (i = 2, 3, …) are also
Becker and Döring (1935) is based on the allowed to be added or separated from an
assumption that the equilibrium cluster dis- existing cluster (e.g., Binder, 1977; Mirold
tribution function can be specified by Eqs. and Binder, 1977). Thus, the related kinetic
(5-23) and (5-18). Even though the result- equation for the time evolution of f (i, t)
ing predictions of classical nucleation the- (Binder, 1977),
ory ought to become valid asymptotically ∞
d
for large droplets at small supersaturations, f (i, t ) = ∑ ai + i ′, i ′ f (i + i ′, t ) (5-48)
dt i′ = 1
there remain some inherent deficits
i –1 i –1
(Binder, 1980; Gunton et al., 1983). 1 1
– ∑ ai, i ′ f (i, t ) + ∑ bi – i ′, i ′
For instance, for small clusters the sur- 2 i ′= 1 2 i ′= 1
face energy entering Eq. (5-18) should ∞
contain size-dependent corrections to the × f ( i ′, t ) f ( i – i ′, t ) – ∑ bi, i ′ f (i, t ) f (i ′, t )
i ′= 1
macroscopic equilibrium energy of a flat
interface. Furthermore, the separation of contains four different terms. The first ac-
the droplet formation Helmholtz energy counts for an increase of i-mers due to
splitting reactions (i + i¢) Æ (i, i¢) which

are assumed to be proportional to the mo-
mentary number f (i + i¢, t) of clusters of
size (i + i¢), where the proportionality co-
efficient ai + i¢, i¢ is the rate constant. The
second term describes the decrease of
i-mers because of splitting reactions i Æ
(i – i¢, i¢); as a reverse process, coagulation
of clusters with sizes (i – i¢) and i¢ contrib-
utes to a further increase of i-mers (third
term), some of which are lost again by coag-
ulation reactions between i-mers and i¢-
mers to yield (i + i¢)-mers (fourth term).
Evidently, for i¢ = 1 the monomer evapora-
tion and condensation process, assumed to
be the rate controlling step in the Becker–
Döring theory, is contained in Eq. (5-48).
If, again as in classical nucleation (Sec.
5.5.1.1, Eq. (5-24)), detailed balance con- Figure 5-29. Cluster size distribution at different
aging times (arbitrary units) as obtained from a nu-
ditions are assumed to apply between split- merical solution of Eqs. (5-48) and (5-49). (After
ting and coagulation, then the rates a and b Binder and Stauffer, 1976.)
in Eq. (5-48) can be replaced by a single
reaction rate W, e.g.,
(5-49)
though basically qualitative, insight into
W (i, i¢) ∫ ai + i¢, i¢ C (i + i¢) = bi, i¢ C (i ) C (i ¢)
the dynamics of cluster formation and
Here, C (i ) again denotes the cluster con- growth. As is shown in Fig. 5-29, the in-
centration which is in thermal equilibrium itially (t = 0) monotonically decreasing
with the metastable matrix. cluster size distribution attains for interme-
Numerical integration of Eq. (5-48) with diate times a minimum at a size that corre-
Eq. (5-49) provides the time evolution of sponds to about that of the critical nucleus
the cluster concentration. This requires, in classical nucleation theory.
however, a knowledge of the initial cluster The broad maximum of f (i, t) (which,
distribution f (i, t = 0), the reaction rates in fact, does not appear in the correspond-
such as W (i, i¢), and of C (i). These quan- ing curve of the Becker–Döring nucleation
tities are not commonly available for alloys theory (cf. Fig. 5-24), and its shift to larger
undergoing phase separation; this has so sizes with increasing time is due to those
far impeded a quantitative comparison particles which have nucleated at earlier
between the predictions of the generalized times and, hence, have already experienced
nucleation theory and experimental results growth. At t Æ • equilibrium is reached.
on nucleation kinetics. In this case large clusters (iÆ •) of the
However, the solution of the kinetic precipitated equilibrium phase with com-
equations with plausible assumptions for position cbe are embedded in a matrix
the missing quantities (Mirold and Binder, (with composition cae ) which then still con-
1977, see also the chapter by Binder and tains small clusters with the equilibrium
Fratzl (2001) has provided a profound, size distribution shown in Fig. 5-29.
The qualitative features of the f (i, t)-

curves emerging from the generalized nu-
cleation theory and displayed in Fig. 5-29
are corroborated by computer simulations
(cf. Sec. 5.5.6). They are indicative of the
occurrence of nucleation and growth as
concomitant processes during the early
stages of unmixing. For most solids under-
going phase separation, therefore, it is not
possible to investigate experimentally nu-
cleation and growth as individual processes
proceeding subsequently on the time scale.
Apart from well designed experiments, it is
thus usually impossible to verify experi-
mentally the kinetics predicted by classical
or non-classical nucleation theory (Secs.
5.5.1.1 to 5.5.1.3) or predicted by growth
theories (Sec. 5.5.2). This will be substan-
tiated in more detail in Sec. 5.7.4.
Figure 5-30. a) Schematic phase diagram of a bi-
The generalizing character of the nucle- nary alloy (components A and B) with a symmetrical
ation theory of Binder and coworkers is miscibility gap.
founded on the fact that, in the sense of b) Variation of the critical radius R* of a nucleus and
Sec. 5.2.3, it comprises nucleation in the of the wavelength lc of a critical fluctuation with
composition of the alloy at T0 according to the non-
metastable regime as well as the transition
classical Cahn–Hilliard nucleation theory (C–H non-
to spinodal decomposition in the unstable cl.) and the Cahn–Hilliard spinodal theory (C–H
regime of the two-phase region. Conse- spin.). Approaching the spinodal composition cas
quently, the artificial divergency of both from either the metastable or the instable region
the critical radius R* of the nucleus and the causes R* and, respectively, to diverge. In contrast,
the generalized nucleation theory of Binder and co-
wavelength l* of a critical fluctuation, in-
workers yields the size of the critical cluster to de-
herent to the Cahn–Hilliard theories of non- crease steadily until it becomes comparable with the
classical nucleation (Sec. 5.5.1.4) and spi- correlation length of typical thermal fluctuations.
nodal decomposition (Sec. 5.5.4), is no There is no peculiarity on crossing the spinodal
longer discernible on approaching the spi- curve.
nodal line either from the metastable
(‘nucleation’) or from the unstable From the experimental point of view it is
(‘spinodal’) regime of the two-phase re- rather difficult and tedious to obtain statis-
gion (Fig. 5-30). On crossing the spinodal tically significant information from direct
curves the size of a critical cluster de- imaging techniques (FIM, HREM) on the
creases steadily until it becomes compar- time evolution of the cluster size distribu-
able to the correlation length of typical tion during the earliest stages of unmixing.
thermal fluctuations (Binder et al., 1978). Attempts were made by Si-Qun Xiao and
Thus, no discontinuity of the mechanism or Haasen (1991) who employed HREM to
of the decomposition kinetics is expected binary Ni–Al alloys. This information,
to occur on crossing the border between the however, is implicitly contained in the
metastable and unstable regions. structure function S (k, t) from small angle
scattering techniques, e.g., SAXS or SANS

(Sec. 5.3.1.2). Binder et al. (1978) com-
puted the time evolution of S (k, t) on
grounds of their cluster dynamics model.
For this purpose, rather crude approxima-
tions and simple assumptions had to be
made concerning the originally complex
shape of the size distribution function of
the clusters, their shapes and their compo-
sition profiles. Hence, the resulting S (k, t)
is only qualitative in nature and cannot be
used for a quantitative comparison with ex-
perimentally obtained structure functions.
As is shown in Fig. 5-31 a, S (k, t) is pre-
dicted to develop a peak the height Sm ∫
S (km) of which increases with time due to
the increasing cluster volume fraction,
while the peak position is shifted towards
smaller values of k, reflecting the coarsen-
ing of the cluster size distribution with
time. Fig. 5-31 b shows the variation of
both the peak height Sm (k ∫ km) and the
peak position km with time. On a double-
logarithmic plot both curves display som
curvature, though the shift of km may be
reasonably well represented as a power law,
km (t) µ t–1/6 (5-50)
during the ‘aging period’ covered in Fig.
5-31 a and 5-31 b.
Even though it is not feasible to con-
vert the scaled times shown in Fig. 5-31 a
and 5-31 b to real times, experimental
SANS curves (Fig. 5-32 a) and the related Figure 5-31. a) Time evolution of the structure func-
tion S (k, t), and b) of its peak height Sm and peak po-
Sm (t) µ dS m (t)/dW (cf. Eq. 5-17)) and sition km as predicted by the generalized nucleation
km (t)-curves (Fig. 5-32 b) from Cu–2.9 theory of Binder and coworkers for a 3-dimensional
at.% Ti homogenized at TH, quenched and Ising model (c0 = 0.1, T/Tc = 0.6). After Binder et al.
isothermally aged at 350 °C (Eckerlebe et (1978), see also the chapter by Binder and Fratzl
al., 1986) display the characteristic fea- (2001).
tures qualitatively predicted by the gener-
alized nucleation theory. Unlike previous models, in this model a
Kampmann et al. (1992) developed a ge- distinction can be made between the clus-
neric cluster-dynamics model that can be tering phase being a disordered solid solu-
used for the quantitative interpretation of tion, e.g., Cu–Co with a positive pair ex-
experimental kinetic data. change energy, or an ordered phase, e.g.,
Ni3Al in the Ni–Al system with a negative 5.5.4 Spinodal Theories

pair exchange energy (Staron and Kamp-
As has been pointed out in Sec. 5.2.3 the
mann, 2000 a, b). In the latter case the con-
evolution of non-localized, spatially ex-
tinuation of the atomic order must be ac-
tended solute-enriched fluctuations into
counted for when computing the growth
stable second phase particles is treated in
probability of the long-range ordered
terms of spinodal theories. The general
Ni3Al cluster, the growth of which is as-
concepts of the spinodal theories, including
sumed to proceed via diffusional incorpo-
more recent extensions, are comprehen-
ration of a solute monomer (cf. Sec. 5.7.7).
sively discussed by Binder and Fratzl
(2001) in Chapter 6 of this volume. We will
therefore restrict ourselves to a summary of
the essential predictions of those theories
which may be examined by experimental
studies. Although often not justified, in ma-
terials science some of these predictions are
frequently employed as sufficient criteria to
distinguish a spinodal reaction from an un-
mixing reaction via nucleation and growth.
In their continuum model for spinodal
decomposition, which is based on the free
energy formalism of a non-uniform binary
alloy outlined in Sec. 5.5.1.4 (Eq. (5-35)),
Cahn and Hilliard (1959 a, b) derived the
following (linearized) diffusion equation:
⎡
∂c ( r , t ) M 2 ⎢⎛ ∂ 2 f ⎞
= ∇ ⎜ 2 + 2h 2 Y ⎟ (5-51)
∂t nv ⎢⎝ ∂c ⎠
⎣ c0
⎤
× c ( r , t ) – 2 K * ∇ 4 c ( r , t )⎥
⎦
Figure 5-32. a) Time evolu-

tion of the structure function
(SANS experiments) of a
Cu–2.9 at.% Ti single crystal
aged at 350 °C for the given
times. b) Time evolution of the
peak height Sm µ dS m/dW and
the peak position km. (After
Eckerlebe et al., 1986.)
for the time dependence of the composition is defined as: (5-56)

c (r, t) at position r. As the number of at- ⎤
M ⎡∂ f 2
oms per unit volume, nv, accounts for the R (k ) = – ⎢ + 2h 2 Y + 2 K * k 2 ⎥ k 2
fact that the derivative has to be taken with nv ⎣ ∂c c0
2
⎥⎦
respect to the concentration of component With reference to Eqs. (5-16), (5-54) and
B, M is the atomic mobility and is related to (5-55), the structure function S ( k, t) or the
the interdiffusion coefficient D̃ via related small angle scattering intensity (Eq.
(5-17)) at time t is obtained as
˜ nv ∂2 f
M≡D (5-52)
∂c 2 S (k , t ) = S (k , 0 ) exp [2 R (k ) t ] (5-57)
c0
The other symbols in Eq. (5-51) were de- where S ( k, 0) denotes the equal-time
fined in Sec. 5.5.1.4. structure function of the system equili-
As M is always positive, D̃ takes the sign brated at the homogenization temperature.
of ∂2 f /∂c2 and so is negative inside the spi- Inside the spinodal region (∂2 f /∂ c2)|c0 is
nodal regime (Sec. 5.2.2), giving rise to an negative. Thus for a given negative value
‘uphill diffusion’ flux of solute atoms (cf. of the latter, R ( k) becomes positive for all
Fig. 5-6 b). wavenumbers k satisfying
The linearized version of the more gen- ⎞
⎛ ∂2 f
eral nonlinear diffusion equation holds if k 2 < k c2 = ⎜ 2 – 2h 2 Y ⎟ 2 K * (5-58 a)
the amplitude c (r) – c0 of the composition ⎝ ∂c c0 ⎠
fluctuation is rather small and both M and or, correspondingly, for all fluctuation
(∂2 f /∂ c2)|c0 are independent of composi- wavelengths l larger than lc with
tion. These approximations inherently con-
fine the Cahn–Hilliard theory (CH theory) ⎛ ∂2 f ⎞
to the earliest stages of phase separation. l2c = 4 p K * ⎜ 2 – 2h 2 Y ⎟ (5-58 b)
⎝ ∂c c0 ⎠
By expanding the atomic distribution
c (r, t) into a Fourier series, Eq. (5-51) can Long wavelength fluctuations with l > lc
be written as thus will be amplified exponentially
∂A (k ) whereas short wavelength fluctuations will
= R (k ) A (k ) (5-53) decay with aging time. For lm ∫ 앀옽2 lc , R (l)
∂t
attains a maximum giving rise to fastest
with the amplitude growth of fluctuations with wavelength lm.
On approaching the coherent spinodal
A (k ) = ∫ [c ( r ) – c0 ] exp (i k r ) dr (5-54) curve where the denominator in Eq.
(5-58 b) becomes zero (cf. Sec. 5.5.3), the
of a Fourier component with wavenumber
critical wavelength lc diverges in a similar
k = 2p /l if l denotes the wavelength of the
manner as does the critical radius R* (Fig.
composition fluctuation. Eq. (5-51) has a
5-30). Thus both the linear theory on spino-
solution for every A ( k, t) with
dal decomposition and the non-classical
A (k , t ) = A (k , 0 ) exp [ R (k ) t ] (5-55) nucleation theory (Sec. 5.5.1.4) predict
abrupt changes in the decomposition kinet-
where A ( k, 0) is the initial amplitude of ics on crossing the spinodal line from ei-
the Fourier component with wavenumber ther side. This has prompted several ex-
k . The so-called amplification factor R ( k) perimental studies to determine the spino-
curves taken at various times should cross at

a common cross-over point, at kc .
Most SAXS or SANS studies of binary
alloys, oxides or glasses that were deeply
quenched into the miscibility gap failed to
corroborate the predictions of the linear-
ized CH theory. This is exemplified by
Cu–2.9 at.% Ti isothermally aged at
350 °C. As illustrated in Fig. 5-32 b, the
peak position of the structure function,
which is frequently identified with the
– –
mean cluster spacing l (l ≈ 2 p /km = lm ), is
not found to be time-independent but is
rather shifted towards smaller values of k
owing to spontaneous coarsening of the
Figure 5-33. Time evolution of dS /dW for Cu–2.9 clustering system. Furthermore, no com-
at.% Ti aged at 350 °C for various constant wave- mon cross-over point at any kc (Fig. 5-32 a)
numbers k (Eckerlebe et al., 1986). exists nor does S (k, t) grow exponentially
for any value of k (Fig. 5-33).
According to Eqs. (5-56) and (5-57), a
plot of (5-59)
2 ∂
R (k ) / k = (1 / 2 k ) ln [ S (k , t ) / S (k , 0 )]
2
∂t
versus k2 should yield a straight line with

R (k) ∫ 0 at kc , whereas a pronounced cur-
vature at larger values of k has been ob-
served experimentally in alloys (e.g., Fig.
5-34) and glass systems (Neilson, 1969).
Cook (1970) attributed this curvature to
random thermal composition fluctuations
which were not accounted for in the origi-
Figure 5-34. Variation of R (k, t)/k 2 with k 2 as de-
nal CH theory. Even with the incorporation
termined for two different time intervals (Eckerlebe of the thermal fluctuations into the linear-
et al., 1986). ized theory, at its best the resulting
Cahn–Hilliard–Cook (CHC) spinodal the-
ory is seen (Langer, 1973; Gunton, 1984)
dal curve on grounds of a kinetic distinction to be valid only for the earliest stages of
between metastable and unstable states. decomposition in systems for which the
In terms of the time evolution of the struc- range of the interaction force is consider-
ture function [Eq. (5-57); now assumed to be ably larger than the nearest-neighbor dis-
isotropic] or the small angle scattering inten- tance, i.e., in systems which are almost
sity, S (k, t) should increase exponentially for mean-field-like in nature. This is not the
k < kc with a peak at the time-independent case for metallic alloys, oxides and glasses,
position km = kc /앀옽 2. Furthermore, all S (k) but is for polymer blends. In fact, results
from kinetic studies of decomposition in curve exists for these solid mixtures (cf.
various polymer mixtures, some of which Binder and Fratzl (2001), Sec. 6.2.5, this
are cited in Table 5-1 (for reviews, cf. Gun- volume). Thus in general, it is currently not
ton et al., 1983; Binder and Fratzl (2001), feasible to predict on theoretical grounds
Chapter 6 of this volume), were found to be whether a metallic, glass or oxide solid
consistent with the predictions of the line- mixture has been quenched into the meta-
arized spinodal CHC theory (e.g., Okada stable or unstable region of the miscibility
and Han, 1986). In contrast, apart from a gap. In this context, it remains debatable
SANS study on the kinetics of phase separ- whether each of the metallic, glass or oxide
ation in Mn–33 at.% Cu (Gaulin et al., mixtures that were studied by means of
1987), none of the scattering experiments SAXS or SANS, and which have been re-
on metallic alloys or oxides which are listed ported to undergo spinodal decomposition
in Table 5-1 and which probably had been (Table 5-1), were truly quenched into and
quenched into the unstable region of the aged within the spinodal region of the
miscibility gap, corroborated the predic- phase diagram. Nevertheless, extensive ex-
tions of the CHC theory. SANS curves perimental SAXS or SANS studies, in par-
taken from Mn–33 at.% Cu during short- ticular on Al–Zn alloys and Fe–Cr (Table
term aging (between 65 s and 210 s at 5-1), did not reveal any evidence for a dras-
450°C) could be fitted to the CHC structure tic change in the time evolution of the S (k, t)
function with km being time-independent curves with a change in the initial supersat-
during this period. Thus, it was concluded uration. For the Al–Zn systems in particu-
that phase separation in Mn-33 at.% Cu is of lar, it was shown (Simon et al., 1984) that
the spinodal type and the kinetics follow the the clustering rate increases rapidly with
CHC predictions (Gaulin et al., 1987). The increasing supersaturation, i.e. both km and
S (k) curve of the as-quenched state already Sm are larger the deeper the quench.
revealed a peak, in fact, at a smaller value of Extensions of the Cahn–Hilliard theory
km than for the aged samples. This is indic- of spinodal decomposition to ternary or
ative of the fact that phase separation oc- multicomponent systems have been elab-
curred during the quench owing to a finite orated by de Fontaine (1972, 1973) and
quench rate. In this case, as has been exten- Moral and Calm (1971). However, to our
sively discussed by Hoyt et al. (1989), km knowledge the kinetic predictions from
may indeed become initially time-indepen- these extensions have not yet been com-
dent, thus pretending an experimental con- pared with experimental results, probably
firmation of the CHC linear approximation. because of difficulties in determining the
In the framework of the mean-field theo- partial structure functions of ternary alloys
ries the spinodal curve is well defined as (cf. Sec. 5.3.1.2).
the locus where ∂2 f /∂ c2 vanishes (Sec. As has been pointed out in Sec. 5.2.3,
5.2.2). It can then be easily evaluated, e.g., the formulation of a ‘unified theory’ com-
from an extrapolation of the thermody- prising both nucleation and growth as well
namic data which are known for the single- as spinodal decomposition can also be at-
phase solid solution, into the (‘mean-field’) tacked on grounds of a spinodal theory.
spinodal regime (Hilliard, 1970). Since, This has been attempted in the statistical
however, neither metallic alloys nor glass model of Langer, Bar-On and Miller
or oxide mixtures behave as mean-field- (1975), which takes into account thermal
like systems, no uniquely defined spinodal fluctuations and nonlinear terms which are
Table 5-1. Some small angle scattering experiments – sometimes jointly employed with other microanalytical
tools for studying the morphology – primarily designed for an investigation of spinodal decomposition. If not
otherwise stated the concentrations are in at.% (N.d.: not determined.)
Decomposing solid Experimental Morphology of two- Authors

technique phase microstructure
Metallic alloys
Al–4 wt.% Cu SAXS Modulated structure Naudon et al. (1976)
Al–22% Zn SAXS Gerold and Merz (1967)
Al–22% Zn SAXS, TEM Modulated structure Agarwal and Herman (1973)
Al–20.7 … 49.1% Zn SAXS Bonfiglioli and Guinier (1966)
Al–5.3 … 6.8% Zn SANS GP-zones Hennion et al. (1982)
Al–5.3 … 12.1% Zn SANS, TEM Guyot and Simon (1981)
Simon et al. (1984)
Al–22% Zn–0.1 Mg SAXS, TEM Interconnected spherical Forouhi and de Fontaine
clusters at early times; (1987)
regularly spaced platelets
at later stages
Al–12 … 32 Zn SAXS (synchro- Hoyt et al. (1987, 1989)
tron radiation)
Al–2.4% Zn–1.3 Mg SANS Blaschko et al. (1982, 1983)
Al–3.8 … 9% Li with SANS N.d. Pike et al. (1989)
additions of Cu, Mn
Au-60% Pt SANS Singhal et al. (1978)
Fe-28% Cr–10% Co SANS, TEM, AFIM ‘Sponge-like’ structure Miller et al. (1984)
Fe–34% Cr SANS N.d. Katano and Iizumi (1984)
Fe–30 … 50% Cr SANS N.d. Ujihara and Osamura (2000)
Fe–29.5% Cr–12.5% Co Anomalous SAXS N.d. Simon and Lyon (1989)
(synchrotron source)
Fe–52% Cr SANS N.d. La Salle and Schwartz (1984)
Fe–20 … 60% Cr SANS N.d. Furusaka et al. (1986)
Cu–2.9% Ti SANS, TEM Modulated structure Eckerlebe et al. (1986)
CuNiFe SANS Modulated structure Aalders et al. (1984)
CuNiFe SANS, TEM, AFIM Mottled structure Wagner et al. (1984)
CuNiFe Anomalous SAXS N.d. Lyon and Simon (1987)
(synchrotron source)
Cu–2% Co SANS N.d. Steiner et al. (1983)
Mn–25 … 52% Cu SANS (analysis of N.d. Vintaikin et al. (1979)
integrated intensity)
Mn–33% Cu SANS N.d. Gaulin et al. (1987)
Ni–13% Al SANS Anisotropic clustering Beddoe et al. (1984)
Ni–12.5% Si SANS, SAXS, TEM Modulated structure Polat et al. (1986, 1989)
(‘side bands’)
Ni–11.5% Ti SANS, TEM Modulated structures Cerri et al. (1987, 1990)
Table 5-1. Cont.

Glasses, oxides a
B2O3–(15 wt.% PbO SAXS Craievich (1975)
– 5 wt.% Al2O3) Acuña and Craievich (1979)
(quasi-binary system) Craievich and Olivieri (1981)
B2O3 – (27 wt.% PbO SAXS Craievich et al. (1986)
– 9 wt.% Al2O3) (synchrotron source)
Vycor glass SANS ‘Sponge-like’ structure Wiltzius et al. (1987)
consisting of a SiO2-rich
and a B2O3 -alkali
oxide-rich phase
SiO2–13 mole% Na2O SAXS, TEM ‘Sponge-like’ structure Neilson (1969)
TiO2–(20–80 mole%) SnO2 SAXS, TEM, X-ray Lamellar modulations Park et al. (1976)
diffraction (‘side along [001] (Fig. 5-38)
band’ analysis)
Polymer mixtures
Critical mixture of per- Light scattering Wiltzius et al. (1988)
deuterated and protonated Bates and Wiltzius (1989)
1,4-polybutadiene
polybutadiene and Izumitani and Hashimoto (1985)
styrene–butadiene
copolymer mixtures
polystyrene–polyvinyl- Light scattering Sato and Han (1988)
methylether (PS–PVME) Sato et al. (1989)
Snyder et al. (1983 a, b)
Hashimoto et al. (1986 a, b)
Okada and Han (1986)
a
For a comprehensive survey on oxides and glasses up to 1978, see Jantzen and Herman (1978).
neglected in the CH theory. The LBM non- With certain approximations, nonlinear ef-
linear theory of spinodal decomposition fects and thermal fluctuations are con-
also accounts for the later stages of phase tained in the term A* (t) and in the last
separation, though not for the coarsening term, respectively. Setting A* (t) ∫ 0 repro-
regime. If coherency strains are neglected duces the corresponding equation of mo-
it approximates the kinetic evolution of the tion of the CHC theory, and, if the fluctua-
structure function as tion term is also omitted, Eq. (5-57) of the
original linear CH theory is regained.
dS (k , t ) M 2
=–2 k Langer et al. (1975) proposed a computa-
dt nv tional technique for solving Eq. (5-60).
⎡⎛ ∂2 f ⎞ ⎤ This required several approximations to be
× ⎢⎜ K * k 2 + 2 ⎟ S (k ) + A * (t )⎥ made, e.g., on f (c) and on M, which inher-
⎢⎣⎝ ∂c c0⎠ ⎥⎦
ently leaves the LBM approach with some
M
+2 k T k2 (5-60) fundamental shortcomings. Nevertheless,
nv with respect to the features displayed in the
Figure 5-35. a) Phase diagram of a binary alloy with

a symmetric miscibility gap centered at csym.
b) Time evolution of the structure function of an al-
loy with c0 = csym quenched and aged at TA. The in-
sert shows the distribution function of composition
configurations at two different times. At t = 80 the
evolving two-phase structure with compositions c1
and c2 or y1 and y2, respectively, with
c1 – csym c2 – csym
y1 = and y2 =
cae – csym cbe – csym
already becomes discernible. All units are dimen-
sionless.
c) LBM predictions on the variation of R (k, t)/k 2
with k 2 as determined from b) according to Eq.
(5-59). Qualitatively there is good agreement with
the LBM predictions and experimental results, e.g.,
displayed in Fig. 5-34. (After Langer et al., 1975.)
Figure 5-36. Time evolution

of the structure function
(SANS intensity) of Fe–40
at.% Cr aged at 515 °C for the
given times. Full lines are
calculated from the LBM the-
ory with three free fitting pa-
rameters. (After Furusaka et
al. 1986.)
time evolution of S (k, t), the predictions of region of the miscibility gap. Furthermore,
the LBM theory (Fig. 5-35) are in rather the various spinodal theories are difficult to
good agreement with both experimental handle and many parameters in the result-
studies (e.g., Fig. 5-32 a and 5-34) and ing kinetic equations are often not avail-
Monte Carlo studies (Sec. 5.5.6, Fig. 5- able for most solid mixtures. Hence, like
41). the non-classical nucleation theory, the
In their study of phase separation in various elegant theories that describe the
Fe–(20, 30, 40 and 60) at.% Cr alloys, Fu- kinetics of phase separation of a ‘spinodal
rusaka et al. (1986) compared the experi- alloy’ are of little use for the practical met-
mental S (k, t) curves with Eq. (5-60) of allurgist.
the LBM theory. For shorter aging times at Because of these problems, materials
515 °C, the experimental data points taken scientists have up to now employed mor-
from Fe–40 at.% Cr could well be fitted by phological criteria for the definition of a
Eq. (5-60) (Fig. 5-36) if k T/K*, M K*, and ‘spinodal alloy’. These are simply related
A*/K* were employed as three independent to the predictions of the linear spinodal CH
fitting parameters. On the grounds of the theory concerning the morphological evo-
agreement with the spinodal theory of lution of a solid mixture undergoing spino-
LBM, and additionally, as this particular dal decomposition.
alloy was evaluated to lie within the Most crystalline solid solutions show a
(mean-field) spinodal region, it has been variation of lattice parameter with compo-
concluded that Fe–40 at.% Cr is a ‘spino- sition leading to coherency strains. The as-
dal alloy’. sociated strain energy fel = h 2 Y (c – c0)2
Based on a similar LBM analysis of (Eq. (5-34)), which is accounted for in Eq.
SANS data from Fe–30 … 50 at.% Cr, Uji- (5-51), reduces the driving force for phase
hara and Osamura (2000) corroborated this separation. This effect shifts the locus of
conclusion recently. the original chemical (mean field) spinodal
The nonlinear theory of spinodal decom- to lower temperatures, yielding the coher-
position developed by Langer (1971) for ent spinodal curve (Eq. (5-58)). If the pa-
binary alloys has also been extended to ter- rameter Y (cf. Sec. 5.2.4), which is a com-
nary substitutional systems (Hoyt, 1989) bination of various elastic constants, de-
by deriving the time-dependent behavior of pends on the crystallographic direction, fel
the three linearly independent partial struc- also becomes anisotropic. Therefore, the
ture functions (cf. Sec. 5.3.1.2). An experi- locus of the coherent spinodal may also
mental examination of the kinetic predic- vary with the crystallographic direction.
tions, however, is still lacking. This becomes particularly discernible for
the tetragonal TiO2–SnO2 oxide system
5.5.5 The Philosophy of Defining (Park et al., 1975, 1976), where fel attains a
a ‘Spinodal Alloy’ – minimum for composition waves along
Morphologies of ‘Spinodal Alloys’ [001] and is larger for waves along [100]
and [010]. Hence, the coherent spinodal
Referring to the previous section it is splits up into a [001], a ·101Ò, and a ·100Ò
rather difficult or even impossible to assess branch (Fig. 5-37). As a consequence the
on a thermodynamic basis whether an al- SnO2-rich modulations form preferentially
loy, glass or oxide system was truly along [001], giving rise to a lamellar struc-
quenched into and aged within the spinodal ture at later aging stages (Fig. 5-38). For
Figure 5-37. The phase

diagram and the spinodal
curves for composition
waves along [001], ·101Ò,
and ·100Ò directions for
elastically anisotropic te-
tragonal TiO2–SnO2. The
spinodals were calculated
on the basis of the regular
solution model. (From
Park et al., 1976.)
most cubic metallic systems the elastic an- less steel that undergoes phase separation
isotropy parameter A (Sec. 5.2.4 ii) is posi- during tempering between 350 °C and
tive, rendering the minimum fel along the 450 °C with an associated embrittlement
elastically soft ·100Ò directions. According (Auger et al., 1989). Increasing the Cr con-
to the CH theory the growth rate will thus tent of the ferritic Fe–Ni–Cr solid solution
be highest along the three ·100Ò directions, leads to the formation of Cr-rich a¢-precip-
giving rise to the frequently observed mod- itates during tempering. As the a¢-phase is
ulated precipitate microstructures (Table discernible as individual particles by
5-2, Figs. 5-18, 5-19 b). In isotropic materi- CTEM, it is concluded that phase separa-
als such as polymers, glasses, or, for in- tion occurred via nucleation and growth. A
stance, Fe–Cr and Fe–Cr–Co alloys, the further increase in the Cr content, to about
modulations do not grow along any prefe- 25 wt.%, yielded a ‘sponge-like’ micro-
rential directions. The resulting two-phase structure after tempering below 400 °C.
microstructure is of the ‘sponge-like’ type This was attributed to a spinodal mecha-
and sometimes referred to as a ‘mottled’ or nism. Thus the spinodal is drawn as the line
‘interconnected’ precipitate microstructure that separates the two morphologies (Fig.
(e.g., Fig. 5-16, 5-19 c, 5-20). Based on 5-39).
these considerations, two-phase alloys dis- It must be pointed out, however, that
playing either lamellar, modulated or interconnected or modulated structures
interconnected precipitate microstructures represent two-phase microstructures in the
are commonly termed ‘spinodal alloys’ by later stages of the reaction (e.g., Fig. 5-18).
metallurgists. Furthermore, the morphol- Even though they are widely believed to
ogy of a two-phase material is sometimes result from spinodal decomposition in the
employed to derive the locus of the spino- sense of the CH theory, their formation
dal. This is illustrated in Fig. 5-39 for the might be of a rather different origin. For in-
Cr-rich ferrite phase of a cast duplex stain- stance, a strong elastic interaction of a high
Figure 5-39. Variation of the phase boundaries in

the Fe–Ni–Cr ferrite phase with chromium content as
derived from microstructural observations. (After
Auger et al., 1989.)
Figure 5-38. CTEM micrograph of equimolar

TiO2– SnO2 displaying a lamellar structure consist-
only shown for decomposing Fe–Co–Cr
ing of alternating TiO2- and SnO2-rich layers formed
after aging at 900 °C for 60 min. From Park et al. (Figs. 5-16 and 5-17), AlNiCo permanent
(1976). magnetic materials (Hütten and Haasen,
1986) and Fe–Cr (Brenner et al., 1984) by
means of AFIM, and for a phase separating
number density of individual nuclei, each polystyrene–polyvinylmethylether poly-
of which is surrounded by a solute-de- mer mixture employing nuclear magnetic
pleted zone in which no further nucleation resonance methods (Nishi et al., 1975).
can occur, may also lead to regularly ar- As has ben outlined in Sec. 5.2.3, re-
ranged precipitates, i.e., modulated struc- gardless of whether phase separation oc-
tures (Ardell et al., 1966; Doi et al., 1984, cours via nucleation and growth or via spi-
1988; Doi and Miyazaki, 1986). On the nodal decomposition, the underlying mi-
other hand, initially interconnected micro- croscopic mechanism is diffusion of the
structures in Cu–Ni–Fe alloys were found solvent and solute atoms. In this sense
to break up into isolated plates (Piller et al., there is thus no need to distinguish between
1984). The later-stage microstructure is the two different decomposition modes and
thus neither sufficient to draw any conclu- the term ‘spinodal alloy’ is simply seman-
sions on the early-stage decomposition tic in nature. This is also reflected in the
mode nor for a definition of a ‘spinodal various attempts to develop ‘unified
alloy’. We are therefore still left with the theories’ comprising either mode.
question as to what really is a spinodal al- From the practical point of view the mi-
loy? To answer this question unequiv- crostructure of virtually all technical two-
ocally, we must verify by any microanalyt- phase alloys corresponds to that of the later
ical technique that the amplitude of the stages. Hence, the practical metallurgist
composition waves increases gradually worries little about the initial stages of un-
with time until the evolving second phase mixing but is instead interested in predicting
has finally reached its equilibrium compo- the growth and coarsening behavior of pre-
sition. Such an experimental verification is cipitate microstructures in the later stages.
a difficult task and, to our knowledge, was This will be the subject of Secs. 5-6 and 5-7.
Table 5-2. TEM and AFIM studies on phase separating solids that revealed ‘spinodal precipitate
microstructures’. If not otherwise stated the concentrations are in at.%.

Metallic alloys
Al–4 wt.% Cu TEM (‘side-bands’) Modulated structure Rioja and Laughlin (1977)
Al–(2.4 –3%) Li HREM, Radmilovic et al. (1989)
X-ray diffraction
Cu3–4MnxAl TEM (‘side-bands’) Modulated structure; Bouchard and Thomas (1975)
at later stages
Cu2MnAl + Cu3Al-plates
Cu–(1.5–5.2 wt.%) Ti TEM (‘satellite Modulated structure Laughlin and Cahn (1975)
analysis’)
Cu–2.7% Ti AFIM Modulated structure Biehl and Wagner (1982)
Ni–12% Ti AFIM Modulated structure Grüne (1988)
Ni-29% Cu–21% Pd TEM Modulated structure Murata and Iwama (1981)
Ni-base superalloys AFIM Short wavelength Bouchon et al. (1990)
(≈ 2.5 nm) Cr fluctuation
in g-phase
Nimonic 80 A TEM (‘side-bands’) Modulated structure Wood et al. (1979)
Fe–25% Be AFIM Modulated structure Miller et al. (1984, 1986)
CuNiFe AFIM Interconnected percolated Piller et al. (1984)
structure (cf. Fig. 5-19 c)
CuNiFe TEM (‘side-bands’) Modulated structure Wahi and Stager (1984)
(cf. Fig. 5-18) Livak and Thomas (1974)
Co–10% Ti FIM Modulated structure Davies and Ralph (1972)
Co–3 wt.% Ti TEM (‘side-bands’) Modulated structure Singh et al. (1980)
– 1 … 2 wt.% Fe
Mn–30 wt.% Cu TEM, magnetic Interconnected Yin et al. (2000)
susceptibility structure
Steels
Cast Duplex stainless TEM, AFIM After decomposition Auger et al. (1989)
original ferritic
Fe–Cr–Ni shows
‘sponge-like’ structure
Ferrous martensite
Fe–15 wt.% Ni–1 wt.% C TEM Tweed structure Taylor (1985)
Fe-25 wt.% Ni TEM Tweed structure Taylor (1985)
– 0.4 wt.% C
Fe–13 … 20 wt.% Mn X-ray diffraction Modulated structure at Miyazaki et al. (1982)
(‘side-bands’), TEM early times; isolated
Fe2Mo particles at later
stages
Amorphous alloys
Ti50Be40Zr10 AFIM Wavy composition profile Grüne et al. (1985)
Table 5-2. Cont.

Glasses, ocides
TiO2–60 mole% SnO2 TEM Tetragonal system with Stubican and Schultz (1970)
lamellar modulations
along [001]
TiO2–50 mole% SnO2 HREM Tetragonal system with Horiuchi et al. (1984)
lamellar modulations
along [001]
SiC–(50–75 mole%) AlN TEM (‘satellites’) Modulated structure Kuo and Virkar (1987)
5.5.6 Monte Carlo Studies Carlo techniques are then employed to

carry out this stochastic process.
The basic features of early stage decom-
The Kawasaki dynamics employed for
position as a stochastic process, without ar-
MC simulations are far from being repre-
tificial distinction between nucleation,
sentative of a real binary alloy system as
growth and coarsening regimes, have been
extensively studied in model alloys by
means of Monte Carlo (MC) computer sim-
ulations, mainly by Lebowitz, Kalos and
coworkers (Lebowitz and Kalos, 1976;
Binder et al., 1979; Marro et al., 1975,
1977; Penrose et al., 1978; Lebowitz et al.,
1982; see also Binder and Fratzl, 2001;
Chapter 6 of this volume). The binary model
alloys are usually described in terms of a
three-dimensional Ising model with pairwise
nearest-neighbor interactions on a simple
rigid cubic lattice the sites of which are oc-
cupied by either A or B atoms, and a phase
diagram which displays a symmetrical mis-
cibility gap centered at 50 at.% solute con-
centration (Fig. 5-40). The microscopic dy-
namics of this system have commonly been
described by the Kawasaki model (Kawa-
saki, 1972). There a nearest-neighbor pair
of lattice sites is chosen at random, then the Figure 5-40. Phase diagram of the 3-dimensional Is-
atoms on those sites may be interchanged ing model approximating a binary model alloy. In
with a probability that depends on the ener- terms of the mean field theory (Sec. 5.2.2) ‘alloys’
# 1 to # 4 are quenched into the metastable regime,
gies of the configuration before and after ‘alloys’ # 5 to # 7 beyond the classical spinodal line
the exchange in such a way that detailed are quenched into the unstable region of the phase di-
balancing holds (Penrose, 1978). Monte agram. (After Lebowitz et al., 1982.)
the atomic exchange is assumed to occur close to the spinodal line within the meta-
directly rather than indirectly via the va- stable regime, evolves similarly to that of
cancy mechanism. Nevertheless, MC simu- ‘alloy’ # 5 quenched into the center of the
lations based on Kawasaki dynamics on an spinodal region (Fig. 5-41c). Thus, in
atomic scale yielded the time evolution of agreement with the generalized nucleation
the cluster configuration and the structure theory and the LBM spinodal approach, but
function S (k, t) of model alloys quenched unlike the predictions of linearized CH the-
into any region below the solubility line ory of spinodal decomposition, MC simu-
(Fig. 5-40) without worrying about com- lations again reveal i) no evidence for any
plicating factors, such as an insufficient abrupt change of the decomposition kinet-
quenching rate, excess vacancies, or lattice ics on crossing the spinodal curve, ii) no
defects, which we commonly face when in- common cross-over point of the S (k)
vestigating real alloys (Sec. 5.3.2). Fur- curves taken after different aging times, iii)
thermore, MC simulations have allowed a no exponential growth of the scattering in-
critical examination of the various theoret- tensity for a certain time-independent wave
ical approaches in terms of cluster dynam- vector in any time regime, and, further-
ics models or spinodal models to be made. more, iv) the peak position of S (k, t) at km
For practical computational limitations, is not found to be time-independent but is
however, the maximum size of the model shifted towards smaller values of k indicat-
alloy has commonly been restricted to ing the immediate growth of clusters.
about 50 ¥ 50 ¥ 50 lattice sites. Because of Frequently, the MC data for the time ev-
the size limitation, in general MC kinetic olution of the peak position (km) and the
experiments can only cover the earlier peak height (Sm) of the structure function
stages of a precipitation reaction in alloys have been fitted to simple power laws (cf.
where the supersaturation is sufficiently Sec. 5.8.2), such as km (t) µ t –a and Sm (t)
high for the formation of a large number d t b ; a and b were estimated to range from
density of clusters or nuclei, and where the 0.16 to 0.25 and 0.41 to 0.74, respectively,
cluster sizes are still considerably smaller depending on the initial supersaturation of
than the linear dimension of the model the ‘alloy’ (Marro et al., 1975, 1977; Sur et
system. This is frequently not the case in al., 1977). Lebowitz et al. (1982) pointed
real alloys. out, however, that due to the finite (small)
In essence, the results from MC simula- size of the system it is difficult to extract
tions based on Kawasaki dynamics have re- from computer simulations precise and re-
covered the predictions both from the gen- liable information about the analytical
eralized nucleation theory of Binder and form of km (t), for example and that it is
coworkers (Sec. 5.5.3) and from the non- possible to fit the same MC data with other
linear spinodal LBM approach. In particu- functional forms than the power laws given
lar, the time evolution of the structure func- above. (In Sec. 5.8.2 we will show, in fact,
tion in the kinetic Ising model displays that apart from the late stages of coarsen-
qualitatively the same features as the corre- ing, it is usually not feasible to interpret ex-
sponding ones obtained from the latter the- perimental kinetic data over an extended
ories (cf. Figs. 5-31 and 5-35 b) or from ex- aging period in terms of a power-law be-
periment (e.g., Fig. 5.32 a). havior with a time-independent exponent.)
For instance, S (k, t) of ‘alloy’ # 4 (Fig. The Kawasaki dynamics are based on an
5-41 b), which according to Fig. 5-40 lies unrealistic exchange mechanism between
Figure 5-41. Time evolution of the structure func-

tion at T/Tc = 0.59 as obtained from MC computer
simulations. With reference to Fig. 5-40:
a) for ‘alloy’ # 1 with c0 = 0.05 (after Lebowitz et al.,
1982),
b) for ‘alloy’ # 4 with c0 = 0.2 (after Sur et al., 1977)
and
c) for ‘alloy’ # 5 with c0 = 0.5 (after Marro et al.,
1975). The given times are in units of a Monte Carlo
step, i.e., the average time interval between two at-
tempts at exchanging the occupancy of a specific
site. The numerical results for S (k, t) at the discrete
values of k were connected by straight lines.
neighboring atoms. In order to account for and Penrose, 1994). Comparing the growth
the more realistic case of atomic diffusion rates of clusters in the Ising model of a
being based on the vacancy mechanism, two-dimensional model alloy showed that
the Ising model was more recently ex- the asymptotic regime of cluster coarsen-
–
tended to binary alloys comprising vacan- ing (R ~ t1/3; cf. Sec. 5.6.2) is approached
cies (Yaldram and Binder, 1991 a,b; Fratzl much faster with vacancy dynamics than
with Kawasaki dynamics, in particular at bility cR (R) in the presence of small parti-
larger supersaturations; the cluster shapes cles with a large ratio of surface area to
are found not to depend on the particular volume is larger than that for larger ones.
dynamics model (Fratzl and Penrose, With reference to Eq. (5-43), this leads to a
1994). size-dependent growth rate, which is posi-
In a first attempt to interpret experimen- tive for larger particles with c̄ > cR and neg-
tal data with kinetic data obtained from ative for smaller ones with c̄ < cR. The
MC simulations, Soisson et al. (1996) sim- growth rate becomes zero for particles with
ulated phase separation in a Fe–1.34 at.% c̄ = cR which are in unstable equilibrium
Cu alloy. The alloy was modeled as a rigid with the matrix. Their radius R* is derived
b.c.c. crystal with 2 ¥ L3 lattice sites (L var- from Eq. (5-46) as (5-61)
ying from 32 to 64) containing one single
2 s ab Vb 1 1
vacancy; the energetics and the parameters R* ≈ ≡ K ′ s ab
of the atomistic kinetic model were ad- Rg T ln ( c /cae ) ln ( c /cae )
justed to the thermodynamic and diffusion Hence, driven by the release of excess
data of the Fe–Cu system. The MC results interfacial energy, larger precipitates will
confirmed earlier results from magnetic grow at the expense of smaller ones which
SANS studies (Beaven et al., 1986) that the dissolve again given rise to a change in the
precipitating clusters are pure copper; fur- precipitate size distribution. This process,
thermore the time evolution of the precipi- which is commonly referred to as coarsen-
tated volume fraction agreed quite well ing or Ostwald ripening2, frequently re-
with the one experimentally determined duces the precipitate number density form
from resistivity measurements (Lê et al., ≈ 1025 m–3 to less than 1019 m–3 in typical
1992). In the later stages dynamical scaling two-phase alloys during aging (cf. Fig.
behavior was shown to hold, indicating that 5-3). Usually, the coarsening process is
the cluster pattern remains similar as aging considered to be confined to the latest
time increases (cf. Sec. 5.8.1). stages of a precipitation reaction. However,
as will be shown in Sec. 5-7, coarsening
may accompany the growth process out-
5.6 Coarsening of Precipitates lined in Sec. 5.5.2, or may even start while
the system is still in its nucleation period,
5.6.1 General Remarks depending on the initial supersaturation of
the solid solution.
For most two-phase alloys, the simple
model of diffusional growth of isolated 5.6.2 The LSW Theory of Coarsening
non-interacting particles with uniform size,
on which Eq. (5-47) is based, frequently In essence, the coarsening of randomly
does not give a realistic description of the dispersed second-phase particles is a multi-
further dynamic evolution of the precipi- particle diffusion problem which is diffi-
tate microstructure beyond its nucleation cult to handle theoretically. In their classic
stage. In reality, towards the end of the nu- LSW coarsening theory, Lifshitz and Slyo-
cleation period a more or less broad parti-
2
cle size distribution f (R) is established In its original meaning, Ostwald ripening is con-
fined to a coarsening reaction where the second-
(Fig. 5-10 and 5-29). According to the phase particles act as the only sinks or sources of so-
Gibbs–Thomson equation (5-45), the solu- lute atoms.
5.6 Coarsening of Precipitates 371
zov (1961), and Wagner (1961) calculated

the time evolution of f (R, t) which satis-
fies the continuity equation:
∂f ∂ ⎡ dR ⎤
+ f =0 (5-62)
∂t ∂R ⎢⎣ dt ⎦⎥
On the basis of the continuity equation and
of Eq. (5-43), the time evolution of the
–
mean particle radius R (t) and the precipi-
tate number density Nv (t) are derived. Cer-
tain approximations, however, had to be
made in order to solve the equations of mo-
tion analytically:
a) Both terminal phases a and b are dilute
solutions; their thermodynamics can be Figure 5-42. CTEM analyses of the normalized g ¢-
described by a dilute solution and the particle size distribution in Ni–8.74 wt.% Ti after ag-
linearized version (Eq. 5-46) of the ing at 692 °C for the given times. For comparison, the
Gibbs–Thomson equation may be used. shape-invariant distribution function fLSW of the
b) The precipitated volume fraction fp = LSW theory is included. (After Ardell, 1970.)
–
(4 p /3) R 3 Nv is close to zero. In such a
dilute system interparticle diffusional
must be emphasized that these predictions
interactions such as those occurring in
hold only in the limit t approaching infin-
more concentrated alloys (cf. Sec.
ity, since the particle size distribution that
5.6.3) can be neglected, and a particle
is present at the beginning of coarsening
only interacts with the infinite matrix.
can be quite different from the time-invari-
c) fp ≈ const., i.e., the decomposition is
ant form.
close to completion with the supersatu-
Extensions of the LSW theory to binary
ration Dc ≈ 0. This inherently confines
systems with non-zero solubilities and non-
the LSW theory to the late stages of a
ideal solution thermodynamics, and ade-
precipitation reaction.
quate modification of the Gibbs–Thomson
With these assumptions the LSW theory equation, have reproduced the temporal
yields in the asymptotic limit Dc (t) Æ 0 power laws (Eq. 5-63). The corresponding
temporal power laws rate constants for this more realistic case
– were derived as (Schmitz and Haasen,
R 3 (t) = KRLSW t (5-63 a)
1992; Calderon et al., 1994):
Nv (t) = KNLSW t –1 (5-63 b)
8 D Vb s ab
Dc (t) = KCLSW t –1/3 (5-63 c) KR = (5-64 a)
9 ( cbe – cae )2 Ga′′
for the time evolution of the average parti-
cle radius, the particle density and the 3 fp 1
KN = (5-64 b)
supersaturation, respectively, and a particle 4 p K R f3
–
size distribution fLSW (R/R ), the shape of
which is time invariant under the scaling of 9 ( cbe – cae )
– KC = K R2 / 3 (5-64 c)
the average particle size R (Fig. 5-42). It 4D
Ga≤ is the second derivative of the molar longer feasible to follow minor changes of
Helmholtz energy of the a phase; f3 is the Dc with time quantitatively by any of the
third moment of the time-independent experimental techniques frequently em-
scaled particle radius distribution function ployed, such as CTEM, AFIM, SAXS or
–
f (R/R ) which in the zero volume fraction SANS (Sec. 5.3.1); in the asymptotic limit
limit (i.e. f = fLSW; see Sec. 5.6.3) is 1.129. Dc Æ 0, it even appears difficult to measure
For an ideal solution and the dilute solu- Dc (t) in alloys containing a ferromagnetic
tion limit (i.e. cae Ⰶ1, cbe ≈ 1) the rate con- phase with magnetic techniques (Ardell,
stant KR (Eq. 5-64 a) adopts the form 1967, 1968), though these direct methods
are certainly more sensitive than CTEM,
8 D Vb cae s ab AFIM or SAS. On the other hand, in those
K RLSW = (5-65)
earlier decomposition stages, where Dc (t)
9 Rg T
is experimentally accessible, an LSW anal-
of the original LSW theory (Calderon et ysis cannot be performed, and, in particu-
al., 1994). lar, the rate constant KRLSW (Eq. (5-65)) is
Numerous experimental studies on a no longer valid (see Sec. 5.7.4.3). Thus, in
wide variety of two-phase alloys attempted practice, an LSW analysis based on Eqs.
to examine the coarsening kinetics pre- (5-63) and (5-65), at its best, can be per-
dicted by the LSW theory. Regardless of formed only in dilute systems and yields
the particular alloy system and the micro- only a value for the product Ds ab . This fact
analytical technique employed (e.g., was ignored in most LSW analyses based
CTEM on Ni–Al, Ni–Ti: Ardell, 1967, on the independent measurement of both
–
1968, 1970; and on Ni–Si: Cho and Ardell, R (t1/3) and Dc (t –1/3); hence, the values of
1997; AFIM on Ni–Al: Wendt and Haasen, D and s ab derived from this type of analy-
1983; SANS on Fe–Cu: Kampmann and sis must be regarded with some reserva-
Wagner, 1986), these studies frequently re- tion. However, frequently the interfacial
vealed the experimental size distribution energy was derived exclusively from Eq.
function to be considerably broader than (5-65) under the assumption that the effec-
– –
fLSW (R/R ) (Fig. 5-42), whereas a plot of R 3 tive diffusivity D is identical with that ob-
or Nv–1 vs. t yielded more or less straight tained from an extrapolation of available
lines (due to limited statistics, the error high-temperature data to the aging temper-
bars are usually rather large). From the ature, thus neglecting the influence of
slopes of these LWS plots the product s ab D quenched-in vacancies (Sec. 5.3.2.1). (In
can be derived with cae usually taken from Sec. 5.7.4.4 two different methods will be
the known phase diagram. Frequently Dc discussed which allow a separate deriva-
was also measured and plotted versus t–1/3 tion of s ab and D to be made from experi-
2
(Eq. (5-63 c)) in order to determine D/s ab mental data, one method without even
(e.g., Ardell, 1967, 1968, 1995; Wendt knowing Dc (t).)
and Haasen, 1983). Thus apparently abso- For non-ideal solid solutions, s ab must
lute values for both s ab and D have been be derived from Eq. (5-64 a). Apart from
determined from values of s ab D and knowing D, this requires detailed informa-
2
D/s ab . However, the rate constants KRLSW tion on the thermodynamics of the a matrix
and KCLSW in Eq. (5-63) were derived phase in order to derive Ga≤. For many bi-
with the assumption c): Dc ≈ 0 or fp = const. nary alloys the necessary thermodynamic
Once this condition is fulfilled, it is no functions have become available via the
CALPHAD method (Sec. 5.2.4). If not

available, they must be derived from more
or less adequate solution models, with the
regular solution model being the less so-
phisticated one. For the case of a binary
model alloy with a miscibility gap and reg-
ular solution behavior, Calderon et al.
(1994) demonstrated that depending on ag-
ing temperature, s ab may be significantly
different if assessed from the correct form
of the rate constant for a regular solution
or from the original LSW form, i.e. Eq.
(5-65).
Figure 5-43. The rate constant K, relative to that of
LSW, KLSW, as a function of the volume fraction fP.
5.6.3 Extensions of the Coarsening Shown on the figure are the predictions of Brailsford
Theory to Finite Precipitate Volume and Wynblatt (BW), Marsh and Glicksman (MG),
Fractions Marqusee and Ross (MR), Tokoyama and Kawasaki
(TK) and the simulations of Akaiwa and Voorhees
There have been many efforts to extend (AV) where 쑗 denote the use of the monopole ap-
the LSW theory to the more realistic case proximation of the diffusion field and 쎲 denote
monopole and dipole approximations to the diffusion
of a finite precipitated volume fraction and
field.
to investigate its influence on the shape of
the distribution function and the coarsen-
ing kinetics (Ardell, 1972). The central
challenge is to determine the effects of more symmetric as the volume fraction in-
interparticle diffusional interactions on the creases (Fig. 5-44). The reason for the in-
growth rate of a particle of a given size. crease in the rate constant is clear: as the
Theories accounting for these diffusional volume fraction increases, shrinking parti-
interactions fall into two broad categories: cles move closer to growing particles and
effective medium theories, and statistical thus the concentration gradients are larger,
theories that are based upon a solution to and the rate of growth and shrinkage in-
the diffusion field in this multiparticle creases.
system. All theories are in agreement that The effective medium theories deter-
the presence of a non-zero volume fraction mine the growth or shrinkage rate of a par-
does not change the exponents of the tem- ticle of a given size using a medium that is
poral power laws given by LSW (Eq. constructed, presumably, to give the statis-
(5-63)) or the existence of a time-indepen- tically averaged growth rate of particles of
dent scaled particle size distribution. The a certain size. Examples of such theories
presence of a non-zero volume fraction, are due to Tsumuraya and Miyata (1983),
however, does change the amplitudes of the Brailsford and Wynblatt (1979) and, more
temporal power laws and the shape of the recently, Marsh and Glicksman (1996).
scaled distribution functions. The rate con- While the effective medium employed by
stant KRLSW increases with the volume frac- Brailsford and Wynblatt has been shown
tion (Fig. 5-43) and the time-independent analytically to be consistent with a solution
distribution function becomes broader and to the multiparticle diffusion equation, this
terms must be included (Akaiwa and Voor-

hees, 1994). Given, then, a spatial distribu-
tion of particles and a particle size distribu-
tion, it is possible to determine the coarsen-
ing rate of each particle. The more chal-
lenging step is to determine from this infor-
mation the statistically averaged growth
rate of a particle of a given size. This can
be done analytically (Marqusee and Ross,
1984; Tokuyama and Kawasaki, 1984; Yao
et al., 1993), or numerically by placing a
large number of particles in a box and de-
termining their coarsening rate (Voorhees
Figure 5-44. Steady-state (time-invariant) precipi-
and Glicksman, 1984; Akaiwa and Voo-
tate size distributions at various volume fractions. rhees, 1994; Mandyam et al., 1998). Al-
For comparison the corresponding LSW distribution though all of these theories begin with the
for zero volume fraction is also shown. (After To- same solution to the diffusion equation,
kuyama and Kawasaki, 1984.) none of the predictions for the rate con-
stants and particle size distributions are in
agreement, illustrating the difficulty in per-
has not been done with the other theories. forming the statistical averaging.
This, then, is the major disadvantage of Nevertheless, a number of qualitative ef-
these theories. A significant advantage, fects of a finite volume fraction become
however, is that they are simple to apply clear after examining the predictions of
and make predictions over a wide range of these theories. (a) The growth rate of a par-
volume fractions, even as high as 0.95. In ticle of a given size is a function of its sur-
addition, both the Brailsford and Wynblatt rounding particles. For example, if a parti-
and Marsh and Glicksman theories corre- cle of size R1 is surrounded by particles of
spond reasonably well with the first-princi- radii R < R1 , this particle will grow, but if
ples statistical theories for volume frac- this particle is surrounded by particles with
tions below 0.3. Above 0.3, these are the radii R > R1 , it will shrink. Thus, unlike in
only theories that make predictions on the the LSW theory the growth rate of a parti-
rate constants and particle size distribu- cle is not solely a function of its size. (b)
tions. The local diffusional interactions give rise
The statistical theories are based upon a to spatial correlations between particles
solution to the multiparticle diffusion prob- that are not random. For example, the aver-
lem. The particles are assumed to be spher- age interparticle separation for a system
ical and the concentration in the matrix at undergoing coarsening is larger than that
the interface of each particle is given by the for a random spatial distribution. This is
Gibbs–Thompson equation, (5-45). For because the probability of finding a small
volume fractions below 0.1 the solution to and large particle almost touching, which
the diffusion equation is represented as a can occur in a system with a random spatial
monopole source or sink of solute in the distribution, is low, as the strong diffu-
center of each particle (Weins and Cahn, sional interactions that occur when a large
1973). At higher volume fractions dipolar and a small particle are located close to
each other causes this small particle to dis- formed in a system wherein the materials
appear. At larger interparticle separation parameters, such as the interfacial energy
distances, however, large particles tend to and diffusion coefficient, are known (inde-
be surrounded by small particles as they pendent of a coarsening experiment) and
are feeding solute to the growing large par- where this system satisfies all the assump-
ticles (Akaiwa and Voorhees, 1994). Thus, tions of theory. Therefore, it is unclear if
there are both spatial and particle size cor- the disagreement between theory and ex-
relations (Marder, 1987). Information on periments is due to an artifact of the system
the spatial correlations between particles is employed in the experiments or a defect in
needed to determine the structure function the theories. Recent experiments that em-
that is measured using small angle scatter- ploy a system in which the materials pa-
ing. The agreement between the predic- rameters are known and satisfy all the as-
tions of Akaiwa and Voorhees (1994) for sumptions of theory have been performed
the structure function and the experimen- using solid Sn-rich particles in a Pb-rich
tally measured structure function in Al–Li eutectic liquid (Alkemper et al., 1999).
alloys is reasonable (Che et al., 1997). (c) There was no convective motion of the par-
Another effect of interparticle diffusional ticles because the experiments were per-
interactions is that the center of mass of the formed in the microgravity environment of
particles is not fixed, but moves in a man- the Space Shuttle. These experiments show
ner consistent with the concentration gra- that there can be very long transients asso-
dients in the system. This particle motion ciated with the coarsening process. In par-
has been determined theoretically (Mar- ticular, during the course of the experiment
der,1987; Akaiwa and Voorhees, 1994), and the initially broad particle size distribution
has been observed in experiments on coars- evolved slowly towards that predicted by
ening in transparent solid–liquid mixtures theory. Even with a factor of four change in
(Voorhees and Schaefer, 1987). the average particle size the distribution
Experiments in which the coarsening ki- never reached the steady-state distribution
netics of solid particles in a liquid (Fig. predicted by theory. This may be the reason
5-45) have been measured show clearly why the experimentally measured particle
that the coarsening rate increases with vol- size distributions (measured using other
ume fraction (Hardy and Voorhees, 1988, systems) rarely agree with theory. Never-
and references therein). In two-phase solid theless, the evolution of the distribution
systems, however, there are reports that the was found to correspond quite well with
rate constant is independent of the volume the predictions of two theories for transient
fraction (Cho and Ardell, 1997). The rea- Ostwald ripening.
son for these contradictory results may be Coarsening of two-phase alloys with
the elastic stress that is present in the two- modulated microstructures consisting of
phase solid systems. Although the pre- large volume fractions ( fp ≥ 30%) of iso-
dicted particle size distributions of the non- lated coherently misfitting particles, e.g.,
zero volume fraction theories are broader Cu–Ni–Fe and Cu–Ni–Cr (Fig. 5.19),
than those of LSW, nearly all experimen- Cu–Ni–Si (Yoshida et al., 1987), or mod-
tally measured particle size distributions ulated Co–Cu (Miyazaki et al., 1986), was
are broader than the predictions of the non- reported to be rather sluggish. The time ex-
zero volume fraction theories. Until re- ponent a of the coarsening rate, defined by
–
cently there have been no experiments per- R µ t a, of these alloys was consistently
Figure 5-45. Evolution of Sn-rich particles with fp = 0.64 in a liquid Pb–Sn matrix during coarsening at 185 °C.
The top row is at constant magnification and illustrates a typical coarsening process. For scaling the absolute
size of the microstructure, in the bottom row the magnification has been multiplied by the ratio of the average
– –
intercept length L at time t to L at t = 75 min. (From Hardy and Voorhees, 1988.)
found to be less than 1/3, the value ex- ing the intermediate stages of coarsening.
pected from conventional coarsening the- Depending on the specific (local) micro-
ory on the basis of Ostwald ripening. Fur- scopic diffusional mechanism which is as-
thermore, it was shown that coarsening of a sumed to contribute to the shift of the cen-
·100Ò modulated structure in less concen- ter of gravity of the particles and their
trated Co–Cu alloys (≤ 20 at.% Cu; fp ≈ likely coagulation, the time exponents a for
20%) still follows the t1/3 kinetics, whereas the related coarsening rate were evaluated
the coarsening rate becomes extremely between a = 1/6 and a = 1/4 and hence are
small in a Co–50 at.% Cu alloy ( fp ≈50%), smaller than predicted by the LSW theory
as indicated by the small time exponent (a = 1/3).
a < 1/50 (Miyazaki et al., 1986). According to Fig. 5-32 b the position of
As was pointed out in Sec. 5.5.5 “mot- the SANS peak intensity at k m varies in
tled structures” or “interconnected micro- proportion to t –0.25. This might, in fact, be
structures” can result from spinodal de- interpreted in terms of a cluster-diffusion-
composition of alloys with vanishing cohe- coagulation mechanism being dominant
rency strains. Unlike for modulated struc- prior to LSW-type coarsening. In Sec.
tures, elastic interactions can therefore be 5.8.2 it will be shown, however, that even
ruled out as a likely reason for the stability if single atom evaporation or condensation
of interconnected microstructures against in the LSW sense is assumed to be the only
coarsening. Nevertheless, the coarsening microscopic mechanism contributing to
rate of interconnected phases is frequently particle growth, the time exponent a al-
also found not to follow the t1/3 kinetics. ready displays a strong time dependence.
Experimental studies on the coarsening ki- Depending on the initial supersaturation of
netics of interconnected phases in hard the alloy, a (t) may then vary between al-
magnet materials such as Fe–Cr–Co (Zhu most zero and 0.5 during the course of a pre-
et al., 1986), Al–Ni–Co (Hütten and Haa- cipitation reaction. The question of whether
sen, 1986; Katano and Iizumi, 1982) in- a even reaches its asymptotic value 1/3 de-
stead yielded time exponents a between 1/4 pends on whether the experiment spans a
and 1/10. Evidently, an adequate theoreti- sufficiently long range of aging times.
cal description of the coarsening kinetics of From the experimental point of view it
both modulated and mottled microstruc- thus does not appear feasible to decide
tures has to go beyond the mere modeling merely on grounds of the measured time
of diffusional interaction between an en- exponent whether the cluster-diffusion-
semble of isolated spherical particles. coagulation mechanism influences the
growth or coarsening kinetics of a precipi-
tate microstructure.
5.6.4 Other Approaches Towards
Coarsening
5.6.5 Influence of Coherency Strains
In addition to the LSW-type coarsening
on the Mechanism and Kinetics
mechanism, which is based on the evopora-
of Coarsening – Particle Splitting
tion and condensation of single atoms from
dissolving and growing precipitates, The LSW theory, as well as the exten-
Binder and Heermann (1985) also con- sions to finite volume fractions, assume
siderred a cluster-diffusion-coagulation that the coarsening process is driven en-
mechanism likely to become operative dur- tirely by the associated release of interfa-
Figure 5-46. High-magnification scanning

electron micrograph of concave g ¢-precipi-
tates in an aged Ni–23.4 Co–4.7Cr–4 Al
– 4.3Ti (wt.%) superalloy prior to splitting.
(Reproduced by courtesy of D. Y. Yoon
(Yoo et al., 1995).)
Figure 5-47. As in Fig. 5-46, showing a

split g ¢-precipitate.
cial energy. In the presence of elastic tion energy is also a function of the shape
stresses, however, such as those induced by of the particles, the importance of which
coherency strains, the coarsening process has been emphasized by Onuki and Nishi-
can be driven by the release of both the mori (1991).
interfacial and elastic energy. The elastic In the presence of misfit strains, inter-
energy can be decomposed into an elastic particle elastic interactions can give rise to
self-energy, the elastic energy of an iso- pronounced spatial correlations between
lated particle in an infinite medium, and an the precipitates. Ardell et al. (1966) pro-
interaction energy that is due to the pres- posed that the elastic interactions between
ence of other particles in the system. The g¢–Ni3Al particles are responsible for the
elastic self-energy is a strong function of alignment of the particles along the elasti-
the shape of a particle and is responsible cally soft directions of the crystal during
for the shape bifurcations mentioned in coarsening. The resulting structure of
Sec. 5.4.2. Moreover, the elastic interac- many nickel-based alloys reveals that the
g¢-type precipitates are rather uniformly on the morphology of the particles and
distributed (Fig. 5-19 b), which is some- hence on the magnitude of the elastic inter-
times reminiscent of modulated structures. action energy itself. In fact, two-dimen-
These strong spatial correlations develop sional calculations (Thompson et al., 1994)
by two basic mechanisms. The first is that wherein the morphology of a particle was
particles aligned as stringers along the not constrained found that a misfitting
elastically soft directions will grow at the fourfold symmetric particle in an elasti-
expense of particles not so aligned (Ardell cally anisotropic homogeneous crystal was
et al., 1966). The second is that alignment stable with respect to interfacial perturba-
will occur by particles migrating through tions at least up to L = 26 (cf. Sec. 5.4.2).
the matrix (Voorhees and Johnson, 1988). Thus the conclusion reached on the basis of
This migration is a result of the elastic elastic energy considerations that splitting
interactions inducing a non-uniform inter- is possible appears to be due to the assump-
facial concentration and, hence, non-uni- tion that the morphology of the precipitate
form concentration gradients along the paris invariant upon splitting. Miyazaki et al.
ticle interface. (1982) proposed that the splitting process
It has been observed (Miyazaki et al., begins with the formation of the matrix
1982; Doi et al., 1984; Kaufmann et al., phase in the middle of the precipitate. This
1989) that during coarsening, individual idea was verified through the diffuse inter-
g¢-type precipitates can split into groups of face calculations of Wang et al. (1991).
two or eight smaller cuboidal particles. Us- However, the experimental results of Yoo
ing a deep etching technique and a scan- et al. (1995) confirmed the hypothesis of
ning electron microscope, the recent experi- Kaufmann et al. (1989) that the splitting
ments of Yoo et al.(1995) have illustrated process is initiated via the amplification of
clearly the three-dimensional morphology perturbations along the precipitate–matrix
of the particles undergoing the splitting interface and not by the formation of the
process (Figs. 5-46 and 5-47). Such a split- matrix phase in the center of the particle.
ting process clearly cannot be driven by a Thus, the cause of the splitting remains, at
reduction in interfacial energy and thus the this point, unexplained. However, the
cause of the splitting has been ascribed to a stability of a cuboidal shaped particle in
decrease in the elastic energy. Although the three dimensions has not been examined,
elastic self-energy does not change on and the effects of elastic inhomogeneity are
splitting, assuming that the particle mor- still to be explored fully. Recent work by
phologies are the same before and after Lee (2000) has shown that splitting is pos-
splitting, if the elastic interaction energy is sible for certain differences in elastic con-
negative, e.g., as is the case for particles stants between particle and matrix.
aligned along the elastically soft directions In addition to inducing particle migra-
of an elastically anisotropic crystal, the to- tion and selective coarsening, elastic inter-
tal elastic energy will decrease upon split- actions may alter the kinetics of the coars-
ting (Khachaturyan and Airapetyan, 1974; ening process. Calculations employing
Khachaturyan et al., 1988; Johnson and fixed (spherical or circular) particle mor-
Cahn, 1984; Miyazaki et al., 1982). How- phologies have shown that in a two-particle
ever, constraining the morphology of the system inverse coarsening is possible,
particles to be invariant upon splitting ne- wherein a small particle will grow at the
glects the effects of the elastic interaction expense of a large particle (Johnson,1984;
Miyazaki et al.,1986; Johnson et al., 1990).

However, similar calculations of the evolu-
tion of two precipitates in an elastically an-
isotropic homogeneous system wherein the
morphology is not fixed have shown that
inverse coarsening does not occur (Su and
Voorhees, 1996). The difference between
these two results is due to the strong shape
dependence of the elastic interaction en-
ergy; the particle morphology must be al-
–
lowed to change in a manner that is consis- Figure 5-48. Variation of the mean radius R of g ¢-
tent with the elastic and diffusion fields in particles with aging time in two different Ni–Cu–Si
alloys with fp = 0.18 (Ni–47.4 Cu–5Si) and fp = 0.5
the system. Inverse coarsening can still oc- (Ni–35.1 Cu–9.8 Si), numbers indicate at.%. (After
cur, however, when more than two particles Miyazaki and Doi, 1989.)
interact elastically (Su and Voorhees, 1996;
Wang et al., 1992). If, however, L is suffi-
ciently large and the system is both elasti-
cally anisotropic and inhomogeneous,
Schmidt et al. (1998) have shown that two
arbitrarily shaped particles can indeed be
stable with respect to coarsening. These
two results imply that it is essential not to
constrain the morphology of the precipi-
tates when computing the evolution of the
microstructure and that elastic inhomoge-
neity can be an important factor in deter- Figure 5-49. Variation of the standard deviation s of
mining the stability of the system with re- the g ¢-particle size distribution with aging time for
spect to coarsening. Paris et al. (1995) have the same alloys as in Fig. 5-48. (After Miyazaki and
also emphasized the importance of elastic Doi, 1989.)
inhomogeneity in determining the stability
of a system with respect to coarsening. particle size is time dependent and, at the
This implies that determining the evolution very least, much less than the classical
of a system during coarsening with a large value of 1/3 predicted by the LSW theory.
number, sufficient to provide accurate sta- In contrast, Ising model simulations (Fratzl
tistical information, of arbitrarily shaped and Penrose, 1996) and diffuse interface
particles is a very challenging problem. calculations (Nishimori and Onuki, 1990)
Nevertheless, some attempts have been in an elastically anisotropic homogeneous
made. For example, Onuki and Nishimori system do not show stabilization with re-
(1991) have found through diffuse inter- spect to coarsening, again indicating the
face calculations that when the precipitates importance of elastic inhomogeneity. By
are softer than the matrix, stabilization of assuming that the systems are statistically
the system with respect to coarsening may invariant, Leo et al. (1990) have shown that
be possible. Although the size of the the exponent for the average size scale of
system examined is small, the exponent of the precipitates can be altered by the pres-
the temporal power law for the average ence of elastic stress, in this case attaining
5.7 Numerical Approaches to Concomitant Processes 381
a value of 1/2. This result does not contra- tion, growth and coarsening. In contrast to
dict Onuki’s, as the onset of stabilization the considerations of the preceding sec-
clearly violates the statistically time invar- tions we now ask how experimental kinetic
iant assumption of Leo et al. data can be interpreted if they refer to very
There have been many investigations of early decomposition stages which include
the kinetics of coarsening in elastically nucleation, growth and coarsening as con-
stressed solids. Experiments performed in comitant rather than consecutive processes
Ni–Cu–Si alloys by Yoshida et al. (1987), on the time scale. In attempting such inter-
and Miyazaki and Doi (1989) show a clear pretations we have to recall that these early
departure from the classical t1/3 predictions decomposition stages especially are char-
for the average particle radius with the acterized by dramatic changes in supersat-
growth rate of the average particle radius uration and, thus, in the driving forces for
and the width of the particle size distribution nucleation and growth processes. Further-
decreasing in time (Figs. 5-48 and 5-49). A more, during these early stages the size dis-
similar departure from the t1/3 function has tribution function evolves and is subjected
also been observed in the Ti–Mo system by to drastic alterations with rather short ag-
Langmayr et al. (1994). In contrast, in ing periods where nucleation, growth and
Ni–Al (Ardell, 1990), Ni–Mo–Al (Fähr- coarsening must be seen as competing and
mann et al., 1995) and Ni–Si (Cho and Ar- overlapping processes. Evidently these
dell, 1997) alloys the exponent for the aver- rather complicated phenomena are not
age particle size appears to be 1/3. The rate properly accounted for by splitting the
constant, however, appears to be indepen- course of decomposition into a nucleation
dent of the volume fraction of precipitate. regime, a growth regime, and a coarsening
Although it has been speculated that elastic regime. Moreover, the kinetic theories de-
interactions are responsible for this result, veloped for either regime (Secs. 5.5.1, 5.5.2
this has not been confirmed due to the diffi- and 5.6, respectively) are based on ideal-
culty of performing calculations in systems ized assumptions which are frequently not
that are sufficiently large to yield accurate expected to be close to reality. This holds
statistical information. true for the classical nucleation theories of
Volmer and Weber and Becker and Döring
(Sec. 5.5.1). In these theories the supersat-
uration is assumed to be constant. This may
5.7 Numerical Approaches only be fulfilled – if at all – during the ear-
Treating Nucleation, Growth liest nucleation stage. Furthermore, these
and Coarsening as Concomitant theories are based on artificial assumptions
Processes of the cluster size distribution in the vicin-
ity of the critical radius (Fig. 5-24), which
are not consistent with the fact that during
5.7.1 General Remarks on the
the nucleation process, many growing pre-
Interpretation of Experimental Kinetic
cipitates slightly larger than the critical
Data of Early Decomposition Stages
size are formed. The theory of diffusional
In this section we turn our attention growth by Zener and Ham (Sec. 5.5.2) de-
to decomposition studies of alloys with scribes only the time evolution of precipi-
sufficiently high nucleation barriers tates with uniform size. However, towards
(DF */k T t 7) decomposing via nuclea- the end of nucleation as well as at the be-
ginning of the LSW coarsening regime we mann and Wagner (MLS model; 1984) in
certainly have to deal with a polydispersed such a way that it could be applied for the
precipitate microstructure. Thus, even if description of the kinetics of precipitate
nucleation and LSW coarsening were to formation and growth in metastable alloys
be well separated on the time scale, in of rather high degrees of supersaturation.
between these two regimes the Zener–Ham The MLS model is still based on the same
theory cannot be expected to correctly pre- assumptions as the original LS theory. In
dict the measured growth kinetics quantita- particular, the explicit form of the size dis-
tively. Finally, the coarsening theories of tribution is not accounted for and the long-
Lifshitz, Slyozov and Wagner are based on time coarsening behavior is assumed to
the linearized version of the Gibbs–Thom- match the LSW results, i.e., is described by
son equation (Eq. (5-46)) and on the as- Eqs. (5-63 a, 5-65). A priori, it is not pos-
sumption that the supersaturation is close sible to foresee the influence of these as-
to zero. These restrictions also hold for the sumptions on the precipitation kinetics.
more recent theories (Sec. 5.6.3) which Therefore, Kampmann and Wagner
take into account finite precipitated vol- (KW; 1984) have devised an algorithm that
ume fractions and overlapping concentra- accurately describes the entire course of
tion profiles between precipitates. precipitation within the framework of clas-
Thus, any theory for the kinetics of pre- sical nucleation and growth theories. This
cipitation that can be employed either for a algorithm is accurate in the sense that, un-
more realistic interpretation of experimen- like the LS and the MLS theories, it con-
tal data or for a prediction of the dynamic tains no simplifying assumption; in partic-
evolution of a second-phase microstructure ular, in this algorithm, termed the Numeri-
under elevated temperature service condi- cal model (N model), the time evolution of
tions, has to treat nucleation, growth and the size distribution is computed without
coarsening as concomitant processes. This any approximations. From a comparison of
was accounted for in the cluster dynamics the N model with experimental data it is
theories of Binder and coworkers (Secs. possible to determine some crucial precipi-
5.5.3, 5.6.4 and Binder and Fratzl (2001), tation parameters of the particular alloy
Chapter 6 of this volume). However, as system as well as to scrutinize the existing
was pointed out by these authors (Binder et nucleation and growth theories with re-
al., 1978; Mirold and Binder, 1977), the spect to their applicability to decomposi-
theory developed yields a reasonable qual- tion reactions in real materials. Further-
itative prediction of the features of experi- more, the N model allows an evaluation to
mentally observable quantities but does not be made of how realistic the various ap-
attempt their quantitative interpretation. proximations are which occur in both the
A further decomposition theory treating LWS theory and the MLS model.
nucleation, growth, and coarsening as con- In this section we briefly introduce the
comitant processes was developed by MLS and the N models. An attempt is
Langer and Schwartz (LS model; Langer made to demonstrate their ability to de-
and Schwartz, 1980). This theory was for- scribe the entire course of the decomposi-
mulated for describing droplet formation tion reaction; we will also show how some
and growth in near-critical fluids. It was essential parameters of the decomposing
later modified by Wendt and Haasen alloy, such as the diffusion constant and the
(1983) and further improved by Kamp- specific interfacial energy, can be evalu-
ated by fitting the theoretically predicted

–
curves R (t) and Nv (t) to the corresponding
experimental data. Furthermore, on the
grounds of the N model, it is possible to ex-
amine whether certain time intervals exist
during the course of a precipitation reac-
tion in which the kinetic evolution is pre-
dicted by either the growth or the coarsen-
ing theory with sufficient accuracy.
5.7.2 The Langer and Schwartz Theory

(LS Model) and its Modification
by Kampmann and Wagner (MLS
Model)
In the LS theory it is assumed that the Figure 5-50. Relationship between the ‘true’ contin-
system contains NLS droplets per unit vol- uous size distribution function f (R, t) yielding the
– –
mean radius R and the related parameters of the LS
ume of uniform size R LS . In order to ac-
count for coarsening, the continuous distri- model. LS assumed a monodispersive distribution,
–
fLS, of particles with radius R LS . fa (R*, t) dR* is the
bution function f (R, t) and the number of apparent number density of particles with radii
particles of critical size, i.e., f (R*, t) dR*, between R* and R* + dR*. In the LS theory, only par-
must be known. However, in the LS theory ticles in the hatched region belong to the precipitated
this is not the case. LS, therefore, intro- phase.
duced an apparent density fa (R*, t) (Fig. 5-
50) which is given as:
–
b The growth of particles of mean size R LS is
fa ( R*, t ) = N LS (5-66) given as:
RLS – R *
fa (R*, t) is thus proportional to NLS and in- dRLS (5-69)
versely proportional to the width of f (R, t). = v ( RLS )
dt
The constant b = 0.317 is chosen in such a fa ( R*, t ) dR *
way that for t Æ • the coarsening rate + ( RLS – R*)
–3 N LS dt
dR LS /dt is identical to KRLSW (Eq. (5-65)).
– 1
Unlike the LSW theory, where R = R*, in + J [ R * (t )] ( R * + d R * – RLS )
the LS theory only particles with R > R*are N LS
counted as belonging to f (R, t), i.e. –
∞
The term v (R LS ) is given by Eq. (5-43) and
1 accounts for the growth rate of the particles
RLS = ∫ f ( R, t ) R dR (5-67)
N LS R* in the supersaturated matrix. The second
– term accounts for the change in the true
Thus, R LS > R* at all stages of decomposi- distribution function caused by the dissolu-
tion, keeping fa (R*, t) in Eq. (5-66) finite. tion of fa (R*, t) dR* particles with radii
Due to nucleation at a rate J and dissolu- between R* and R* + dR*. The third term
tion, NLS changes with time according to: describes the change in f (R, t) caused by
dN LS dR * the nucleation of particles which must be
= J – fa ( R*, t ) (5-68)
dt dt slightly larger than those of critical size,
i.e., R = R* + dR* and dR* O R*. Together 2 Vm s 0

with the continuity equation3, RN = ; ks = s ab / s 0 ; x 0 = c0 / cae ;
Rg T
4p 3 x = c / cae ; x p = cp / cae ; r = RLS / RN ;
( cp – c ) RLS N LS = ( c0 – c ) (5-70)
3
r * = R * / RN ; dr * = dR * / RN ;
and Eq. (5-66), Eqs. (5-68) and (5-69) are 4p 3 D dn dt
the rate equations which describe the entire n = N LS RN ; t = 2 t ; J˜ = J
3 RN d N dt
course of precipitation in the LS model.
(5-72)
After proper scaling, these equations were
numerically integrated in conjunction with Apart from RN, all parameters are dimen-
steady state nucleation theory (Eq. (5-27)). sionless; unlike in both the original LS
The LS model is based on the assumption model and the study of Wendt and Haasen
that the equilibrium solubility of small (1983), in the MLS model RN rather than
clusters can be determined from the linear- the correlation length is used as the scaling
ized version of the Gibbs–Thomson equa- length. By straigthforward scaling of Eqs.
tion, i.e. Eq. (5-46). This linearization, (5-27), (5-61) and (5-71) we obtain the
however, generally does not hold for small scaled version of the equations of motion
clusters in metallic alloys (cf. Sec. 5.6.2). (5-68) and (5-69). From these, the number
This becomes immediately evident for the density n of particles is eliminated by vir-
Cu–1.9 at.% Ti system isothermally aged tue of the scaled continuity equation
at 350 °C. At t = 0, R* ist 0.48 nm or 0.13 1 x0 – x
nm, depending on whether R* is computed n= (5-73)
r3 xp – 1
(s ab = 0.067 J/m2, see Sec. 5.7.4.4) from
the non-linearized or from the linearized We finally obtain
version of Eq. (5-46). This example clearly
dr r ⎛ 1 b ks 1 ⎞ dx
demonstrates that Eq. (5-46) must be used + ⎜ + ⎟
in its non-linearized version, particulary dt 3 ⎝ x 0 – x r ln x – ks x ln x ⎠ dt
for systems with large values of s ab and/or r4 xp – x
large supersaturations. = – J˜ (5-74)
3 x0 – x
The MLS model is based on the non-line-
arized Gibbs–Thomson equation (Eq. (5- dr b ks dx
+
45)). Thus, the growth rate in the MLS dt x (ln x )2 dt
model is: 1 1
= [x – exp ( ks /r )] (5-75)
dRLS 1 D (5-71) xp – 1 r
= v ( RLS ) = e
dt cp – ca RLS x p – 1 ˜ ⎛ ks ⎞
+ r3 J⎜ + dr * –r ⎟
⎡ ⎛ 2 s ab Vm 1 ⎞ ⎤ x 0 – x ⎝ ln x ⎠
× ⎢c – cae exp ⎜ ⎟⎥
⎢⎣ ⎝ Rg T RLS ⎠ ⎥⎦ Eqs. (5-74) and (5-75) are the basic equa-
In order to write the rate equations (5-68) tions of the MLS model and are numeri-
to (5-70) in a properly scaled version, we cally integrated with s0 = 0.1 J/m2; values
introduce the following parameters: of dr*/r* (x 0) used by Kampmann and
Wagner (1984) ranged from 0.05 to 0.2.
3
Note that particles with R = R* are not contained in In the LS theory only the steady-state
Eq. (5-70); we therefore call these particles ‘apparent’. nucleation rate has been used; in contrast,
in Eqs. (5-74) and (5-75) of the MLS 1

with r j = (r j + r j + 1 ) (5-77)
model, KW employed the time-dependent 2
nucleation rate J* (Eq. (5-29)), which in- The continuity equation in the N model
volves the incubation time. In the scaled then reads:
version the latter (tw) was evaluated from: j0
1 xp – 1 2 2 (x p – 1) ∑ r j3 n j = x 0 – x (5-78)
t w (x ) = r * cw ≅ t w, min cw2 (5-76) j =1
2 x –1
Thus the continuous time evolution of
When deriving this equation, it was as- f (r, t) is split into a sequence of individual
sumed that the minimum time (tw, min) for a decomposition steps; these step are chosen
particle to reach the critical size r* is given in such a way that within each correspond-
by Eq. (5-47). However, for subcritical nu- ing time interval Dti the changes of all ra-
clei cR is considerably larger than cea, and dii r j, (ti) and of the supersaturation x (ti)
the probability of their redissolution is remain sufficiently small. Then, both the
rather large; hence, Eq. (5-47) overesti- nucleation and the growth rates can be con-
mates the growth rate significantly. This sidered as being constant during Dti and
fact is counterbalanced by the introduction the change of f (r, t) and of x can be reli-
of the parameter cw. For each particular al- ably computed. The N model of KW con-
loy system, cw is determined in such a way tains some algorithms which ensure a
that after a period tw the first particles be- rather high accuracy of the numerical cal-
come ‘observable’ with a reasonable number culation; it amounts to ≈ 0.5% in the case
density. For all alloy systems investigated so of the growth rate of r.
far, cw ranges from 1.4 to 3.5. Hence, be-
cause of the stochastic nature of cluster 5.7.4 Decomposition of a Homogeneous
growth and dissolution, on average it takes Solid Solution
about two to twelve times longer for a clus-
ter to attain a size R > R* beyond which its 5.7.4.1 General Course of Decomposition
further growth may be evaluated in a deter-
ministic manner on the basis of Eq. (5-47). KW discussed the decomposition reac-
tion of an ideally quenched alloy with cae =
0.22 at.%, cp = 0.20 at.%, c0 = 1.9 at.%
5.7.3 The Numerical Model (N Model) and sab = 0.067 J/m2. Just on the basis of
of Kampmann and Wagner (KW) these input data, the decomposition reac-
Unlike the MLS model, in the N model tion can be calculated within the frame-
the time evolution of f (R, t) – or f (r, t) if work of the MLS and N models. Fig. 5-51
we stick to the same nomenclature – is shows the predictions from both the MLS
computed. For this purpose f (r, t) is sub- (full lines) and the N models (discrete sym-
divided into intervals [r j + 1, r j ] with bols) for the time evolution of the radii
| r j – r j + 1|/r j Ⰶ1 and nj particles in the j-th r– (t) and r * (t), the number density n (t),
interval. In contrast to the MLS model, dis- the supersaturation x (t), and the nuclea-
solving particles with r < r * are assumed tion rate J̃ (t), The chosen values of x0 =
to belong to the precipitating phase, i.e. 8.7, cae , s and D correspond to those for
Cu–1.9 at.% Ti aged at 350 °C (see Secs.
j0 j0
1 5.7.4.3 and 5.7.4.4). Surprisingly, both
n= ∑ n j and r = ∑ nj r j
j =1 n j =1 models yield qualitatively similar results.
Figure 5-51. Evolution of various scaled precipitation parameters with scaled aging time t according to both the
N model and the MLS model. The chosen values of x 0, sab correspond to those for Cu–1.9 at.% Ti aged at 350 °C.
Figure 5-52. Evolution of the scaled size distribution function in Cu-1.9 at.% Ti with aging times as computed
with the N model. For comparison, the distribution functions fLSW (r, t) of the LSW theory with the known
–
values of Nv (t) and R (t) are also shown. D = 2.5 × 10–15 cm2/sec.
During the early nucleation period (t Ù 1) growth rate of nucleated precipitates is also
both J̃ and n increase, whereas x and, close to zero, i.e., r– remains about con-
hence, r* remain roughly constant. At this stant. After (t Û 1, those precipitates nucle-
stage, the N model gives a rather narrow ated first become considerably larger than
size distribution function f (r, t) (Fig. r* and f (r, t) becomes much broader. This
5-52). Since cR is still close to c– ≈ c0, the is the beginning of the growth period
which is characterized by (i) the highest served in Fig. 5-51. During the later
growth rate ever observed during the growth period, dissolution of particles with
course of precipitation; (ii) a ratio r–/r* r < r* commences. At this stage, the MLS
which becomes significantly larger than 1; model only counts particles with r > r* as
(iii) the maximum number density (nmax, belonging to the second phase (Eq. (5-67)),
Nv, max) of particles which remains about and, furthermore, assumes fa (r = r*) to be
constant; (iv) a stronger decrease of x; and, proportional to 1/(r– – r*) (Eq. (5-66)). At
consequently, (v) a decrease of J̃ from its this stage, this is a rather poor approxima-
maximum value; in this particular case, J̃ tion because the N model yields a steep
never reaches its steady-state value. At the slope for f (r Æ r*, t) with a rather small
end of the growth regime (t ≈ 15) the density n (r = r*). These facts make the
supersaturation has dropped significantly MLS model predict considerably larger rate
and the growth rate becomes small. This constants for the decrease of the particle
effect causes r* to converge towards r– and number density and, hence, for the growth
makes dn/dt < 0 (Eq. (5-68)). During the of r– at the end of the growth period (t ≈ 15).
subsequent transition period (t ≈ 15), the Depending on the particular choice of b
growth rate of r– is primarily controlled by (Eq. (5-66)), the coarsening rate of the
the dissolution of particles with r < r* and MLS model approaches asymptotically the
only to a lesser extent by the uptake of so- value KRLSW (Eq. (5-65)) from the LSW the-
lute atoms from the matrix, the supersatu- ory. In this asymptotic limit the mean radii
ration of which is still about 20%. During from both models and, hence, their coars-
the transition period, the true distribution ening rates become identical. At this stage
function f (r, t) continuously approaches the N model yields r– Ó r* as predicted by
the one predicted by LSW (Fig. 5-52). At LSW, whereas the MLS model yields r– –
–
this stage, however, fLSW is still a rather r* = const., i.e., R LS > R*, as required by
poor approximation for f (r, t). This simply Eq. (5-67).
reflects the influence of the linearization of We can conclude that the MLS model,
the Gibbs–Thomson equation (Eq. (5-46)) which requires much less computing time
on which the LSW theory is based. For than the N model, provides a good survey
3
t Æ •, dr– /dt approaches asymptotically of the general course of precipitation.
a constant value, i.e., the reaction is within However, due to the simplifying assump-
the asymptotic limit of coarsening where tions made, it does not predict the precipi-
the supersaturation is almost zero. At this tation kinetics with the same accuracy as
stage f (r, t) is well approximated by fLSW the N model. This is particularly evident
with only minor deviations for small parti- for those precipitation stages where the
cle radii. shape of f (r, t) is extreme, as, for in-
stance, during the later growth stages in the
example discussed above.
5.7.4.2 Comparison Between the MLS
Model and the N Model
5.7.4.3 The Appearance and
The precipitation reaction starts with an
Experimental Identification
identical nucleation rate and identical par-
of the Growth and Coarsening Stages
ticle sizes in both the MLS and the N mod-
els. Therefore, both models are expected to In the following, the results from the
yield identical results which, in fact, is ob- MLS and the N models are compared with
–
Figure 5-53. Variation of R ,
–
R*, RLS and of the relative
supersaturation with aging
time for Cu–1.9 at.% Ti as
computed with the N model
and the MLS model for the
given set of thermodynamics
data; also shown are the ex-
–
perimental data for R (t) from
von Alvensleben and Wagner
(1984).
Figure 5-54. Variation of R2 and R*2 with time during the growth regime for Cu–1.9 at.% Ti. During the period
–
marked by the two arrows the kinetics follow the power-law R ⬃ t1/2.
–3
b) Variation of the coarsening rate dR /dt with aging time; KRLSW is the value predicted by the LSW theory,
Eq. (5-65).
experimental data obtained from a FIM theory (Eq. (5-65)) is only reached for ag-
–
atom probe study of early-stage precipita- ing times beyond ≈ 104min! At this stage R
tion in Cu–1.9 at.% Ti at 350 °C (von Al- has already grown to ≈ 6.4 nm. From this
vensleben and Wagner, 1984). result, which reflects the influence of the
Fig. 5-53 shows the time evolution of R* linearization of Eq. (5-46), KW inferred
–
and or R in physical units. Again these ex- that the LSW theory predicts the correct
perimental data points are well described coarsening rate once
by the N model. Fig. 5-53 also reveals that Vm s ab
the experimental data cover neither the R ⭌13 (5-79)
early nucleation period nor the growth re- Rg T
gime, but rather start (t = 2.5 min) at the If this relation holds, f (R, t) is almost iden-
end of the latter region. From Fig. 5-54 a,
– tical to fLSW (Fig. 5-52) and c̄/cae ≈ 1.
where R 2 is plotted as a function of t, it is According to the results from the N
recognized that during the period t ≈ model, during the early coarsening stages
–
0.7 min to 1.2 min, R 2 varies linearly with t the precipitation kinetics deviate signifi-
as suggested by Eq. (5-40). Since this time cantly from those predicted by the LSW
period extends only over 0.5 min, KW con- theory (e.g., Fig. 5-54 b). Thus the widely
cluded that for Cu–1.9 at.% Ti the time used LSW analyses resulting in a determi-
window is too short for revealing the nation of D and s ab on the basis of Eq.
–
R ⬃t1/2 kinetics experimentally; in fact, the (5-63 a) and Eq. (5-63 c) should not be ap-
same holds true for many other alloy plied to early coarsening stages during
systems analyzed by KW using the N which the relation does not hold.
model. Furthermore, from Fig. 5-54 a the
slope of the straight line has been evaluated
5.7.4.4 Extraction of the Interfacial
to be 0.88 nm2/min, whereas the corre-
Energy and the Diffusion Constant
sponding growth rate from Eq. (5-47) is
from Experimental Data
computed (cae = 0.22 at.%, cp = 20 at.%,
D = 2.5 ¥ 10–15 cm2/sec) to be 2.5 nm2/ KW determined s ab and D separately by
–
min. This result clearly demonstrates that fitting R (t) and Nv (t) as obtained from the
no growth regime exists which is ade- N model to the corresponding experimental
quately accounted for by Eq. (5-47). In data. Fig. 5-55 shows the variation of Nv
other words, the idealizations made in the and J* with aging time as computed with
derivation of Eq. (5-47) do not approxi- the N and the MLS models, together with
mate the true situation in Cu–1.9 at.% Ti. the corresponding experimental data. The
However, if c̄ (t) and cR in Eq. (5-43) are peak number density Nv, max of particles in
replaced by their mean values respectively, a precipitation reaction is essentially gov-
–
during the period for which R ⬃t1/2 holds, erned by the value of DF* via the nuclea-
then Eqs. (5-40 and 5-42) with li = 앀앙 k* tion rate equation (5-27). Since D F* ⬃ s ab
3
yield a value for the growth rate (0.104 nm2/ (Eq. (5-20)), Nv, max depends sensitively on
min) which is only 16% higher than the the value of the interfacial energy s ab. The
true value. very strong dependence of Nv, max on s ab is
–
In Fig. 5-54 b the rate constant dR 3/dt for clearly revealed by Fig. 5-55, showing a
coarsening is plotted versus t. It is evident good fit of Nv (t) for s ab = 0.067 J/m2 and
that for Cu–1.9 at.% Ti the rate constant only poor agreement for s ab = 0.071 J/m2.
KRLSW= 1.2 ¥ 10–24 cm3/sec from the LSW On the other hand, a variation in D mani-
Figure 5-55. Variation of Nv and J* with aging time for Cu–1.9 at.% Ti as computed with the N model and the
MLS model for sab = 0.067 J/m2; for this value of sab the computed Nv (t) curve agrees well with the experi-
mental data; poor agreement is obtained for sab = 0.071 J/m2.
Table 5-3. Correlation between the width of the miscibility gap and the coherent interfacial energies sab for
various two-phase alloys as determined from a fit of the N model to experimental kinetic data.
Alloy Aging Composition/type Width of coherent Coherent interfacial

temperature of precipitates miscibility gap energy
at.% °C at.% sab J/m2
Ni–14 Al 1
550, 500 g ¢-Ni3Al ≈ 15 ≈ 0.016
Ni–26 Cu–9 Al 2
550, 500 g ¢-(Cu, Ni)3Al ≈ 20 ≈ 0.052
540, 500 Not determined ≈ 20 ≈ 0.050
580, 500 Not determined ≈ 20 ≈ 0.052
Cu–1.9 Ti 3
350, 500 b¢-Cu4Ti ≈ 20 ≈ 0.067
Cu–2.7 Ti 4
350, 500 b¢-Cu4Ti ≈ 20 ≈ 0.067
Cu–1.5 Co 5
500, 500 > 95 at.% Co ≈ 95 ≈ 0.171
Fe–1.4 Cu 6
400, 500 > 98 at.% Cu ≈ 100 ≈ 0.250
Fe–0.64 Cu 6
400, 500 > 98 at.% Cu ≈ 100 ≈ 0.250
1 Wendt and Haasen (1983) (AFIM)

2 Liu and Wagner (1984) (AFIM)
3 Alvensleben and Wagner (1984) (AFIM, CTEM)
4 Eckerlebe et al. (1986) (SANS)
5 Gust (1986), unpublished (magnetic measurements)
6 Kampmann and Wagner (1986) (SANS)
fests itself in a parallel shift of the entire From a variety of different two-phase al-
–
curve on the time scale. Thus, a fit of R (t) loys, the available kinetic data have been
and of Nv (t) as obtained from the N model interpreted in terms of the N model. Table
to the experimental curves allows both s ab 5-3 presents the interfacial energies s ab to-
and D to be determined quite accurately. gether with the width of the (coherent) mis-
cibility gap for each given alloy. It is evi- ones. After aging for ≈ 100 min at 350 °C
dent that there is a pronounced correlation the metastable b¢-solvus line is nearly
between s ab and the compositional width, reached at c¢ae≈0.22 at.% Ti (cf. Sec. 5.2.1).
i.e., the broader the gap, the larger is s ab. The experimentally determined kinetic
This is consistent with various theoretical behavior of the precipitate number density,
predictions on the interfacial energy (see of their mean radius, and of the supersatu-
Lee and Aaronson, 1980, for a comprehen- ration are displayed in Fig. 5-57 a – c and
sive discussion of this aspect). compared with the predictions of the N
model. For the computations, the result
from the SANS evaluation was taken into
5.7.5 Decomposition Kinetics in Alloys
account, which yielded the homogenized
Pre-Decomposed During Quenching
sample to already contain ≈2 ¥ 1025 clus-
–
The versatility of the N model is further ters/m3 with R ≈ 0.7 nm. These could im-
illustrated by its ability to predict the pre- mediately grow by further depleting the
cipitation kinetics in alloys which have ex- matrix from solute atoms. Moreover, at
perienced some phase separation during t = 0 the supersaturation was still large
quenching. This is exemplified for Cu–2.9 enough for nucleating new clusters with
at.% Ti, the decomposition reaction of smaller radii at a nucleation rate J* (Fig.
which was studied by Kampmann et al. 5-57 a). Thus, during the first minutes of
(1987) by means of SANS. They found that aging, the cluster number density in-
the cooling rate of their specimen was not creased. At this stage, the alloy contained a
sufficient to suppress the formation of sort of bimodal cluster distribution: the
Cu4Ti precipitates during the quench; in larger ones formed at a smaller supersatu-
fact, as is shown in Fig. 5-56, the solute ration during quenching, and the smaller
concentration decreased from c0 = 2.9 at.% ones resulted from nucleation at 350 °C.
to 2.2 at.% Ti during the quench. With fur- Due to both nucleation of new clusters and
ther aging, the supersaturation decreased growth of pre-existing ones, the supersatu-
continuously through the formation of ad- ration and, hence, the nucleation rate de-
ditional clusters and the growth of existing creased rapidly; after aging for ≈ 3 min, nu-
Figure 5-56. Decrease in the solute con-

centration in the matrix with aging time, as
determined from Laue scattering (쑗 · ), from
the integrated intensity (¥), and the Guinier
approximation (쎲) of SANS curves.
cleation is virtually terminated. After about

10 min the critical radius R*, which is cor-
related with the momentary supersatura-
–
tion, reaches the mean value R of the glo-
bal size distribution. At this instant, R* has
grown beyond the mean radius of the
smaller, freshly nucleated clusters; these
now redissolve leading to a further de-
crease in Nv. Now the size distribution is
again governed by the larger precipitates
formed during quenching. After about
500 min, Nv decreases with t as expected
from the LSW theory.
With regard to the accuracy of both the
SANS experiment and, in particular, the
SANS data evaluation, the agreement
between the experimental kinetic data and
those from the N model is rather good for
an interfacial energy s ab = 0.067 J/m2.
This value is identical to that determined
for the less concentrated Cu–1.9 at.% Ti al-
loy (Sec. 5.7.4.4). In the early stages (t ⬉
–
10 min) the experimental R is considerably
larger than the theoretical one. This simply
reflects the fact that the scattering power of
a particle is weighted with R6; hence, for
t ⬉ 10 min, essentially the radius of only
the larger particles within the bimodal dis-
tribution was determined. The diffusion
coefficient (D = 3 ¥ 10–16 cm2/sec) was
found to be about one order of magnitude
smaller than in the Cu-1.9 at.% Ti alloy
(Sec. 5.7.4.4). It was not possible to decide
whether this difference in the D values re-
flects the concentration dependence of D,
or whether it was caused by differences in
the homogenization temperatures of the
two alloys (Cu–2.9 at.% Ti: TH = 780°C;
Cu–1.9 at.% Ti: TH = 910 °C).
5.7.6 Influence of the Loss of Particle

Figure 5-57. Time evolution of (a) the cluster num- Coherency on the Precipitation Kinetics
ber density Nv and the nucleation rate J*, (b) of their
– In many two-phase systems the particles
mean radius R and of the critical radius R*, and (c) of
the supersaturations. Experimental SANS results: lose coherency once they have grown be-
discrete symbols; computational results: full lines.
yond a certain size RT . The associated in-

crease in the interfacial energy and de-
crease in the solubility limit (cf. Fig. 5-1)
leads to an enhanced driving force and,
hence, to accelerated kinetics for further
coarsening of the incoherent microstruc-
ture with respect to the coherent one. This
effect can also be accounted for by the N
model. This is shown in Fig. 5-58 for
Fe–1.38 at.% Cu aged at 500 °C. As in-
ferred from CTEM, the Cu-rich particles
transform at RT ≈ 2.8 nm from the meta-
stable b.c.c. structure into the f.c.c. equilib-
rium structure. The associated loss of cohe-
rency occurs at particle number densities
well beyond the maximum number density
Nv, max (Fig. 5-58). Thus, following the
procedure outlined in Sec. 5.7.4.4, the co-
herent interfacial energy could be deter-
mined by fitting the N model to the experi-
mental (SANS) kinetic data of the still co-
herent system. The value s ab ≈ 0.27 J/m2
obtained is considerably smaller than the
corresponding value s ab inc
≈ 0.50 J/m2
which was derived from a thermodynami-
cal analysis of the Fe–Cu system (Kamp-
mann and Wagner, 1986). As is shown in
Fig. 5-58, the loss of coherency in fact
leads to a momentary acceleration of the
growth kinetics. It is, however, not suffi-
cient to bridge the discrepancy between the
experimental kinetic data and the theoreti-
cally predicted ones displayed in Fig. 5-58.
As the predicted kinetics are much more
sluggish than the experimentally deter-
mined ones, we might speculate that ne-
glecting particle interaction accounts for
the observed discrepancy. Inspection of
Fig. 5-43, however, reveals that a consider-
ation of finite volume effects in Fe–1.38 Figure 5-58. Kinetic evolution of the precipitated
at.% Cu with fp ≈ 1% would increase the number density (top) and of the mean radius (bottom)
as predicted by the N model for Fe–1.38 at.% Cu.
coarsening rate only by less than a factor
The solid points refer to experimental data derived
1.3 whereas a factor of ≈10 is required to from nuclear and magnetic SANS experiments. The
match the results from the N model and the dashed lines show the kinetic evolution without con-
SANS experiments at the later stages of sideration of the b.c.c. Æ f.c.c. transformation of the
copper-rich particles.
precipitation. At present, it is unclear adequate for describing the growth (R > R*)
whether the various theories dealing with or dissolution (R < R*) of clusters.
finite volume effects (Sec. 5.6.3) are still To overcome these shortcomings, Kamp-
insufficient or whether some heterogenous mann et al. (Kampmann et al., 1992; Staron
precipitation at lattice defects accounts for and Kampmann, 2000 a, b) extended the N
the observed discrepancy. model by a cluster-dynamics (C–D) simu-
At first glance the experimental kinetic lation of the kinetics during the nucleation
–
data for R (t) and Nv (t) in Fig. 5-58 might and growth stages (cf. Sec. 5.5.3). To save
be seen as being amenable to an LSW anal- computation time, in the later growth and
ysis in terms of Eq. (5-63). Analyses of the coarsening stages where the stochastic pro-
SANS data for t Ⰷ103 min, however, cess is no longer relevant, the cluster-dy-
showed the width of the particle size distri- namics approach is linked to the determin-
bution to be much broader (standard devia- istic description of the original N model.
tion: s ≈ 0.31) than expected from the As in the latter, the input parameters enter-
LSW theory or its modification (s ≈ 0.23). ing the extended C–D model are c0, cae , s
Furthermore, the measured supersaturation and D with variations in D leading only to
–
was still far from being close to zero. shifts of the computed curves (e.g., R (t) or
Hence, the conditions for an LSW analysis J (t)) parallel to the time scale. Therefore,
of the experimental data are not at all ful- this C–D simulation can be used in as ver-
filled. satile a manner as the original N model
for the interpretation of experimental data
(Staron and Kampmann, 2000 b).
5.7.7 Combined Cluster-Dynamic
The cluster-dynamics algorithm was de-
and Deterministic Description
vised such that its corresponding Helm-
of Decomposition Kinetics
holtz energy functional approximates
There remain two main shortcomings of closely that of the regular solution model.
the N model outlined in Sec. 5.7.3. Firstly, As proved by CALPHAD studies, this is
the stochastic nature of the nucleation pro- a rather good approximation for Cu–Co.
cess is solely accounted for by using the Fig. 5-59 shows the computed time evolu-
time- and concentration-dependent nucleation of the cluster size distribution during
tion rate (i.e., Eq. (5-29)), whereas the the nucleation period and the nucleation
growth (and shrinkage!) is computed in a rate with input parameters corresponding
deterministic manner on the basis of Eq. to a Cu–0.8 at.% Co model alloy aged at
(5-43). Thus, the stochastic nature of early- 500 °C. At t = 5 sec all clusters are still be-
stage growth which, in particular, becomes low the critical size4 and, thus, Fig. 5-59 a
effective in systems with small nucleation provides insight into the incubation process.
S
energies, is not adequately accounted for in The nucleation rate JB–D calculated ac-
the N model. cording to Eq. (5-27) with c̄ ∫ c0 exceeds
Secondly, as in all previous theories, the the maximum computed one (Fig. 5-59 b)
matrix concentration cR at the cluster–ma- by more than a factor of 1000. This dis-
trix interface was also calculated on the ba- crepancy cannot be attributed to incubation
sis of the Gibbs–Thomson equation, Eq.
(5-46). This relation, which describes an 4
In the C –D approach the critical radius is defined
equilibrium between the cluster size R and as that size for which the probabilities of cluster
the matrix concentration c̄, is generally not growth and shrinkage are identical.
5.8 Self-Similarity, Dynamical Scaling and Power-Law Approximations 395
effects; it can only be resolved if – instead

of Eq. (5-27) – a more adequate equation is
used for calculating the Becker–Döring nu-
cleation rate:
1
JˆB–D
S
= Z b * n1 (t ) (5-80)
2
× exp {– [ DF *( c1 ) – DF *(1)]/ kT}
The modified exponent takes into account
that no Helmholtz energy D F* (1) is
needed for forming monomers (which are
already dissolved in the matrix during the
solution treatment).
Accordingly the number of lattice points
(N0) in Eq. (5-28) must be replaced by the
number of monomers n1 (t), and for a calcu-
lation of D F* the concentration of
monomers c1 (according to Fig. 5-60, c1 =
0.7 at.% rather than the nominal concentra-
tion of c0 = 0.8 at.%) must be used. The
factor 1⁄2 in Eq. (5-80) takes into account
that monomers serve both as initial clusters
(containing i = 1 solute atom) capable of
growing, and also as reaction species being
able to initiate the growth of “monomer
clusters”. As shown in Fig. 5-59 b, with
these physically justified modifications,
excellent agreement is obtained between
the maximum nucleation rate (Jmax) as ob-
tained from the cluster-dynamics model,
and JˆB–D
S
(Eq. (5-80)).
5.8 Self-Similarity,
Figure 5-59. a) Evolution of the cluster size distri- Dynamical Scaling
bution during the nucleation stage for a Cu–0.8 at.% and Power-Law Approximations
Co model alloy aged at T = 500 °C, with pair ex-
change parameter W = 6.24, cae = 0.2 at.%, DCuCo =
1¥ 10–19 m2/s. 5.8.1 Dynamical Scaling
b) Time evolution of the nucleation rate J (t) (solid
According to the LSW theory and its ex-
line) according to the cluster-dynamics calculation.
S
JB–D S
is based on Eq. (5-27), Ĵ B–D is the modified tensions to finite volume fractions (Sec.
Becker–Döring nucleation rate (Eq. (5-80)). (After 5.6), the distribution of relative particle
–
Kampmann et al., 1992.) sizes R /R evolves during extended aging
(t Æ •) towards an asymptotic, time-invar-
iant form, the particular shape of which de-
Figure 5-60. Time evolution of the

number density (nv), the mean (dy-
namical) critical radius (R*), the
mean radius of supercritical clus-
–
ters (R ); the conventionally defined
supersaturation x – 1 = c̄/cae – 1 is
compared with the supersaturation
x mono – 1 = c1/cae – 1 of monomers.
For t ≤ 100 min the computation
was performed cluster-dynamically
and afterwards continued determin-
istically on the basis of the N
model of Sec. 5.7.3. (After Kamp-
mann et al., 1992.)
pends on the precipitated volume fraction was first provided by Monte Carlo simula-
(Fig. 5-44). The final time invariance of tions of the time evolution of binary model
–
f (R /R ) reflects the fact that once the pre- alloys (Sec. 5.5.6). From these studies it
cipitated volume fraction has reached its was concluded that there is a small though
equilibrium value, consecutive configura- systematic dependence of F̃ (x) on the in-
tions of the precipitate microstructure are itial supersaturation, at least for small x
geometrically similar in a statistical sense, (Lebowitz et al., 1982); for large values of
i.e., all consecutive configurations are sta- x, the scaling function appears to be uni-
tistically uniform on a scale that is consid- versal in that it becomes independent of
erably larger than some characteristic temperature and precipitated volume frac-
–
length such as the mean particle size R or tion and even of the investigated material
–
the mean center-to-center distance L = (Fratzl et al., 1983). By analogy to the
Nv–1/3. The self-similarity of the microstruc- Porod law of small-angle scattering, in this
tural evolution has found its expression in regime F̃ (x) decays in proportion to x–4.
the dynamical scaling of the structure func- In experiments designed to test the valid-
tion S (k, t) (Binder and Stauffer, 1974; ity of the scaling behavior, l (t) is com-
Binder et al., 1978). Furukawa (1981) pro- monly related to either the radius of gyra-
posed S (k, t) to satisfy (after some tran- tion RG , the mean particle radius, or to ei-
sient time t0) a scaling law of the form: ther km–1 (t) or k1–1 (t) if km and k1 denote
the maximum and the first moment of S (k,
S (k, t) = l 3 (t) F̃ [k l (t)]; t ⭌ t0 (5-81)
t), respectively. The scaling function F̃ (x)
where F̃ [k l (t)] ∫ F̃ (x) is the time-inde- is then simply obtained, for instance, by
pendent scaling function. As the scaling plotting km3 S (k, t) versus k /km. If scaling
parameter, l (t) denotes some characteristic holds, F̃ (k /km) is time independent 5. After
length and contains exclusively the time
5
dependence of S (k, t). In order to test whether experimental data satisfy
Strong theoretical support for the valid- the scaling law, Fratzl et al. (1983) have proposed
– a
direct method by which the evaluation of RG, R, km
ity of the scaling hypothesis, Eq. (5-81), or k1 can be avoided and by means of which F̃ (x)
during the later stages of decomposition can be determined graphically.
some initial transient time this, in fact, was

observed in the glass systems B2O3–PbO–
Al2O3 (Craievich et al., 1981, 1986) and in
several binary alloys such as Mn–Cu (Fig.
5-61), Al–Zn (e.g., Simon et al., 1984;
Hoyt and de Fontaine, 1989) or Ni–Si (Po-
lat et al., 1989). Dynamical scaling behav-
ior was also found in some ternary alloys
such as Al–Zn–Mg (Blaschko and Fratzl,
1983). For Cu–Ni–Fe, which was studied
by means of anomalous SAXS (Lyon and
Simon, 1987), the scaling behavior is
found to be obeyed only by the partial
structure functions, indicating that this
system does not behave like a pseudo-bi-
nary system. For Fe–Cr the results are con-
troversial. In contrast to Katano and Iizumi
(1984) and Furusaka et al. (1986), La Salle
and Schwartz (1984) reported that dynami-
cal scaling does not hold. The decomposi-
tion kinetics in Fe–Cr at about 500 °C are
fairly sluggish and it may well be that even
after the longest chosen aging time (100 h)
the system had not yet reached the scaling
region where the microstructure displays
self-similarity.
In principle, the structure function S (k,
t) contains all the information on the vari-
ous structural parameters of a decomposing
–
solid such as f (R, t), Nv (t), R (t), morphol- Figure 5-61. Time dependence of the scaling func-
ogy, etc. which, for instance, may control tion F̃ (k /km) = k 3m S (k, t) for Mn–33 at.% Cu at
its mechanical properties. In practice, how- 450 °C. For later times (t > 5115 s) F̃(k /km) becomes
– time independent and, hence, dynamic scaling holds
ever, commonly only R and Nv and neither (top). The structure functions taken at earlier times
f (R, t) nor the morphology can be extracted (965, 1602, 2239, 2886 and 3532 s) do not yet dis-
from experimental data. This stems mainly play scaling behavior (bottom). (After Gaulin et al.,
from a lack of knowledge of the interparti- 1987.)
cle interference function, which contains
the spatial correlations of the precipitate the quantitative extraction of information
microstructure and which manifests itself on the precipitate microstructure rather dif-
in the appearance of a maximum in the S ficult.
(k, t) curves of less dilute systems. Further- From a practical point of view, scaling
more, both the limited range of k over analyses are sometimes seen to allow all
which S (k, t) can be measured and the information contained in S (k, t) to be deci-
large background in conventional small- phered. Provided the explicit form of F̃ (x)
angle scattering experiments often render could be predicted on grounds of a first-
principles theory, it could be compared In contrast to Al–Zn, the small misfit

with experimental data and employed for a strains between the Al-rich matrix and the
comparison of S (k, t) curves taken from spherical d¢-Al3Li precipitates renders the
different materials. Up to now, however, Al–Li system ideal for an accurate test of
this is not feasible. For this reason various scaling in the later stages of coarsening.
phenomenological theories for F̃ (x) have Using SAXS, Che et al. (1997) found scal-
been conceived (e.g., Furukawa, 1981; ing behavior to be obeyed in the coarsening
Hennion et al., 1982) amongst which the regime of each of the Al–Li alloys with dif-
model of Rikvold and Gunton (1982) may ferent volume fractions ranging from 0.18
be regarded as the one that is most conven- to 0.23. The breadth of the scaled structure
ient for a comparison with experimental function, measured by the full width at half
data as it contains the precipitated volume maximum, versus the equilibrium volume
fraction as the only parameter. This model fraction, agreed well with the boundary in-
assumes the two-phase microstructure to tegral-based computer simulation of
consist of a ‘gas’ of spherical second phase Akaiwa and Voorhees (1994) (cf. Sec.
particles (with an identical scattering form 5.6.3).
factor) each of which is surrounded by a As a concluding remark to this section it
zone depleted of solute atoms. With simple is probably fair to state that scaling analy-
approximations on the probability distribu- ses currently can neither furnish the practi-
tion for pairs of particles with certain inter- cal metallurgist with more information on
particle spacings, the explicit analytical the precipitate microstructure nor on its dy-
form for F̃ (x) was derived. However, due namic evolution than has been possible by
to the various assumptions invoked, the conventional analyses of S (k, t) curves
Rikvold–Gunton model is restricted to prior to the emergence of the scaling hy-
smaller precipitated volume fractions. In potheses. It is felt that precise information
spite of its simplicity, fair agreement was on the size distribution, the morphology,
reported between the theoretical F̃ (x) and and the spatial arrangement of precipitates
the scaling functions obtained from com- in a two-phase microstructure becomes
puter simulations and from scaling analy- more readily available from studies em-
ses of S (k, t) curves taken from decom- ploying direct imaging techniques, e.g.,
posed Al–Zn and Al–Ag–Zn alloys (Simon CTEM or AFIM, in particular, as dynami-
et al., 1984). In constrast, in studies of cal scaling only holds in the later stages of
Al–Zn (Forouhi and de Fontaine, 1987) aging where the precipitate microstructure,
and of Ni–Si (Chen et al., 1988) the pre- in general, can be easily imaged and re-
dicted F̃ (x) was found to be much broader solved by these techniques. Furthermore,
than the experimental ones; scaling analy- once dynamical scaling is satisfied, the
ses for borate glasses also could not verify system is close to the asymptotic limit
the theoretically predicted form of F̃ (x) where the LSW theory or its extensions
(Craievich et al., 1986). This may stem may be applied for a prediction of its dy-
from the inadequate assumptions on the namic evolution during further aging.
chosen probability distribution in which
long-range correlations are neglected
5.8.2 Power-Law Approximations
and/or from precipitate morphologies de-
viating from spheres, e.g., platelets in So far not explicit assumption has been
Al–Zn. made about the time dependence of the
chosen characteristic length, e.g., km–1 (t), corresponding data from computer simula-
entering the scaling law, Eq. (5-81). As the tions.
self-similarity of a precipitate microstruc- Inspired by the theoretical predictions,
ture and, hence, the scaling law is impli- many scattering experiments on alloys
citly contained in the LSW theory of coars- were interpreted in terms of power-law ap-
ening, the region of validity of dynamical proximations. Frequently the existence of
scaling coincides with the LSW regime of two well-defined kinetic regimes with dis-
coarsening. km–1 (t) is thus expected to show tinct values of a has been reported (e.g.,
the simple power-law behavior, Fig. 5-32 b). At earlier times, a ranges
between Ù0.1 and ≈ 0.2, whereas at later
km–1 (t) ⬃ t a (5-82)
stages a and b are found close to the values
with a = 1/3. Accordingly, if scaling holds, 1/3 and 1 predicted by the LSW theory.
the maximum of the structure function Sometimes this has been taken as evidence
must evolve in time as (e.g., Katano and Iizumi, 1984) for the first
regime to be dominated by the cluster–
Sm (t) ⬃ t b (5-83)
diffusion–coagulation mechanism (Sec.
with b = 3 a (Eq. (5-81)). Such a power-law 5.6.4), whereas in the second regime, the
behavior was frequently corroborated by evolution proceeds according to the LSW
scattering experiments on materials that mechanism via the evaporation and con-
were aged in the scaling region, and also by densation of single solute atoms.
computer simulations (e.g., Lebowitz et al., However, the interpretation of and Sm (t)
1982), which indicated that scaling would and km (t) in terms of two distinct kinetic
hold. regimes, each of which is well described by
The more recent theoretical develop- a power law, seems rather debatable. A
ments on the kinetics of phase separation closer examination of the Sm (t) and km
have predicted various other values for the curves (e.g., Fig. 5-32 b) always reveals
exponent a. On the basis of their cluster– some curvature prior to reaching the scal-
diffusion–coagulation model (Sec. 5.6.4), ing region. This clearly shows that the ex-
for intermediate times Binder and cowork- ponents a and b are time dependent; thus,
ers predict a = 1/6 and a = 1/5 or 1/4 for apart from the LSW regime, a power-law
low and intermediate temperatures, respec- approximation must be seen as a rather
tively (Binder and Stauffer, 1974; Binder poor description of the dynamic evolution
1977; Binder et al., 1978). As is shown in of a decomposing solid and commonly
Fig. 5-31 b, approximation of Sm (t), which does not disclose the specific growth
displays some curvature, by a power-law mechanism dominating at a certain aging
(Eq. (5-83)) yields b = 0.7 rather than 0.48 regime. This becomes particularly evident
as implied by scaling. The LBM theory of by employing the Numerical Model (Sec.
spinodal decomposition, which accounts 5.7.3) for a derivation of the exponent a (t)
–
for some coarsening at earlier stages (Sec. = ∂ log R /∂ log t. As the N model com-
5.5.4), yields a = 0.21. As outlined in Sec. prises nucleation, growth and coarsening
5.5.6, these values agree quite well with as concomitant processes on the basis of
the corresponding values (a = 0.16 to 0.25 just one growth mechanism – single-atom
and b = 0.41 to 0.71, depending on the evaporation or condensation in the LSW
supersaturation and ‘aging temperature’) sense – a plot of a (t) versus t allows a
obtained from fitting power laws to the closer examination to be made of the valid-
Figure 5-62. Variation of

the time exponent a with ag-
ing time as evaluated for
Cu–1.9 at.% Ti by means of
the N model.
Figure 5-63. Dynamical scaling of the structure function of Cu-2.9 at.% Ti beyond t = 250 min; SANS results
ity of power-law approximations at any in- sharply from ≈ 0.15 to its maximum value a
stant. This is shown in Fig. 5-62 for Cu–1.9 = 0.56 which is indicative of a diffusion-
at.% Ti, the experimental kinetic data of controlled growth of the particles. The du-
which are well described by the N model 6
Accounting for modified boundary conditions, the
(cf. Figs. 5-53 and 5-55). Towards the end cluster-dynamics approach even yields a maximum
of the nucleation regime, a (t) increases value a = 0.7 (Kampmann et al., 1992).
5.9 Non-Isothermal Precipitation Reactions 401
–
ration of the growth regime where R 5.9 Non-Isothermal Precipitation
evolves according to the parabolic power Reactions
–
law R ⬃t1/2 is rather short for this alloy
(cf. Sec. 5.7.4.3). At the end of the growth For many age-hardening commercial al-
regime, where the particle number density loys, the microstructure is established dur-
has reached its maximum value (Fig. ing continuous cooling from the processing
5-55), a (t) drops within about 250 s to a temperature, i.e., the solidification or hot-
value of less than 0.1. In the subsequent working temperature in cast or wrought al-
transition regime at intermediate times, loys, respectively, or during cooling after
a (t) increases continuously and approaches laser surface treatments or welding (Brat-
only slowly the LSW-coarsening regime land et al., 1997, and references therein).
where dynamical scaling holds. Evidently, The resulting non-isothermal transforma-
there is no time regime between the appear- tion usually involves nucleation and
ance of the growth regime and the LSW re- growth of the precipitates as concomitant
gion where power-law behavior is ob- processes. Modeling of non-isothermal
served. On the other hand, it may be in- transformation rates thus ought to account
ferred from Fig. 5-62 that the kinetic evo- for the independent variations of the nucle-
–
lution of R during intermediate aging ation and the growth rates with temperature
stages might be artificially interpreted in and with the temperature-dependent solu-
terms of power laws if the time window bilities in the matrix and the precipitated
covered by the experiment was too short; phase. Furthermore, heterogeneous nuclea-
in this case any exponent between 0.15 and tion at lattice defects often must also be
0.33 may be derived. Thus, if the time win- taken into account, as usually the driving
dow for the kinetic experiment is not prop- force for homogeneous nucleation be-
erly chosen, a ≈ 0.2 may be obtained, but comes sufficient only once the alloy has
without the cluster–diffusion–coagulation been cooled sufficiently deep into the mis-
mechanism being operative. As has been cibility gap.
pointed out in Sec. 5.6.4.3, with increasing Whereas previous models, which have
supersaturation, the growth regime with been based on isokinetic behavior and the
–
R µ t 1/2 disappears completely and a (t) additivity concept, are limited to a descrip-
takes values only between zero and 1/3. tion of only diffusional growth (or dissolu-
Furthermore, the transition period (where tion during reheating) (e.g., Shercliff
a < 1/3) becomes shorter. In this case, LSW and Ashby, 1990a, b; Myhr and Grong,
coarsening and dynamical scaling are ob- 1991a, b; Onsoien et al., 1999), Grong and
served after rather short aging times. This is Myhr (2000) considered the non-isother-
illustrated in Fig. 5-63 for Cu–2.9 at.% Ti mal precipitation reaction as a coupled nu-
which satisfies dynamical scaling after ag- cleation and growth process in terms of a
ing for ≈ 250 min at 350 °C, whereas for the numerical solution and two analytical mod-
less concentrated Cu–1.9 at.% Ti alloy scal- els. As validated through a comparison
ing only holds after about 5 ¥ 104 min. with the numerical results, both analytical
models, i.e., a simplified state variable so-
lution and a solution based on the Avrami
equation, yielded an adequate description
of the overall non-isothermal transforma-
tion behavior comprising the variation of
the nucleation and growth rates as well as Ardell, A. J. (1990), Scripta Metall. Mater. 24, 343.
Ardell, A. J. (1995), Interface Sci. 3, 119.
the precipitated volume fraction with tem- Ardell, A. J., Nicholson, R. B. (1966), Acta Metall.
perature and cooling rate. 14, 1793.
Ardell, A. J., Nicholson, R. B., Eshelby, J. D. (1966),
Acta Metall. 14, 1295.
Asta, M. (1996), Acta Mater. 44, 4131.
5.10 Acknowledgements Asta, M., Hoyt, J. J. (1999), Acta Mater. 48, 1089.
Asta, M. Foiles, S. M., Quong, A. (1998), Phys. Rev.
B 57, 11265.
The authors would like to thank Prof. R. Auger, P., Danoix, F., Menand, A., Bourgoin, J.,
Bormann and Dr H. Mertins for their criti- Guttmann, M. (1990), Mat. Sci. Techn. 6, 301.
cal comments on this chapter and Mrs E. Auger, P., Pareige, P., Akamatsu, M., Blavette, D.
(1995), J. Nucl. Mat. 225, 225.
Schröder, A. Conrad-Wienands and B. Bartel, T. L., Rundman, K. B. (1975), Metall. Trans.
Feldmann for their kind endurance and as- 6 A, 1887.
sistance in preparing the manuscript. The Bates, F. S., Wiltzius, P. (1989), J. Chem. Phys. 91,
3258.
support of this work by the Deutsche For- Beaven, P. A., Frisius, F., Kampmann, R., Wagner, R.
schungsgemeinschaft (Leibniz-Programm) (1986), in: Atomic Transport and Defects in Metals
is also gratefully acknowledged. by Neutron Scattering: Janot, C., Petry, W., Rich-
ter, D., Springer, T. (Eds.). Berlin: Springer Verlag,
p. 228.
Becker, R., Döring, W. (1935), Ann. Phys. 24, 719.
Beddoe, R., Haasen, P., Kostorz, G. (1984), in: De-
5.11 References composition of Alloys: the Early Stages: Haasen,
P., Gerold, V., Wagner, R., Ashby, M. F. (Eds.).
Aalders, C., Van Dijk, S., Radelaar, S. (1984), in: De- Oxford: Pergamon Press, pp. 233 – 238.
composition of Alloys: The Early Stages: Haasen, Biehl, K.-E., Wagner, R. (1982), in: Solid – Solid
P., Gerold, V., Wagner, R., Ashby, M. F. (Eds.). Phase Transformations: Aaronson, H. I., Laugh-
Oxford: Pergamon Press, pp. 149–155. lin, D. E., Sekerka, R. F., Wayman, C. M. (Eds.).
Aaron, H. B., Fainstain, D., Kotler, G. R. (1970), J. Warrendale: The Metallurgical Society of AIME,
Appl. Phys. 41, 4404. pp. 185 – 189.
Aaronson, H. I., Russell, K. C. (1982), in: Solid-State Bigot, A., Auger, P., Chambreland, S., Blavette, D.,
Phase Transformations: Aaronson, H. I., Laugh- Reeves, A. (1997), Microsc. Microanal. Microstr.
lin, D. E., Sekerka, D. F., Wayman, C. M. (Eds.). 8, 103.
Warrendale: The Metallurgical Society of AIME, Binder, K. (1977), Phys. Rev. B 15, 4425.
pp. 371–399. Binder, K. (1980), J. Phys. (Paris) C 4, 51.
Aaronson, H. I., Kinsman, K. R., Russell, K. C. Binder, K. (1983), J. Chem. Phys. 79, 6387.
(1970 a), Scripta Metall. 4, 101. Binder, K. (1984), Phys. Rev. A 29, 341.
Aaronson, H. I., Laird, C., Kinsman, K. R. (1970 b), Binder, K. (1987), Advances on Phase Transitions
in: Phase Transformations: Aaronson, H. I. (Ed.). and Disorder Phenomena: Busiello, G., De Ce-
Metals Park, Ohio: Am. Soc. of Metals, pp. 313 – sare, L., Mancini, F., Marinaro, M. (Eds.). Singa-
396. pore: World Scientific, pp. 1– 71. (Kinetics of First
Acuna, R., Craievich, A. F. (1979), J. Non-Cryst. Order Transitions)
Solids 34, 13. Binder, K., Heermann, D. W. (1985), in: Scaling
Agarwal, S. C., Herman, H. (1973), in: Phase Transi- Phenomena in Disordered Systems: Pynn, R.,
tions and Their Application in Materials Science: Skjeltorp, T. (Eds.). New York: Plenum Press, pp.
Heinisch, H. K., Roy, R. (Eds.). New York: Per- 207 – 230.
gamon, pp. 207 – 222. Binder, K., Stauffer, D. (1974), Phys. Rev. Lett. 33,
Akaiwa, N., Voorhees, P. W. (1994), Phys. Rev. E 49, 1006.
3860. Binder, K., Stauffer, D. (1976), Adv. Phys. 25, 343.
Al-Kassab, T., Wollenberger, H., Blavette, D. (1997), Binder, K., Fratzl, P. (2001), in: Phase Transforma-
Z. Metallkd. 88, 2. tions in Materials: Kostorz, G. (Ed.). Weinheim:
Alkemper, J., Snyder, V. A., Akaiwa, N., Voorhees, Wiley-VCH, Ch. 6.
P. W. (1999), Phys. Rev. Lett. 82, 2725. Binder, K., Billotet, C., Mirold, P. (1978), Z. Physik
Ardell, A. J. (1967), Acta Metall. 15, 1772. B 30, 183.
Ardell, A. J. (1968), Acta Metall. 16, 511. Binder, K., Kalos, M. H., Lebowitz, J. L., Marro, J.
Ardell, A. J. (1970), Met. Trans. 1, 525. (1979), Colloid Interface Sci. 10, 173.
Ardell, A. J. (1972), Acta Metall. 20, 61. Blaschko, O., Fratzl, P. (1983), Phys. Rev. Lett. 51, 288.
5.11 References 403
Blaschko, O., Ernst, G., Fratzl, P., Auger, P. (1982), Danoix, F., Auger, P., Bostel, A., Blavette, D. (1991),
Acta Metall. 30, 547. Surf. Sci. 246, 260.
Blavette, D., Bostel, A., Sarran, J. M., Deconihout, Davies, C. K. L., Nash, P., Stevens, R. N. (1980),
B., Menand, A. (1993), Nature 363, 432. Acta Metall. 28, 179.
Blavette, D., Deconihout, B., Chambreland, S., Bos- Davies, D. M., Ralph, B. (1972), J. Microscopy 96,
tel, A. (1998), Ultramicroscopy 70, 115. 343.
Bonfiglioli, A. F., Guinier, A. (1966), Acta Metall. de Fontaine, D. (1969), Trans. AIME 245, 1703.
14, 1213. de Fontaine, D. (1971), J. Appl. Crystallogr. 4, 15.
Bouchard, M., Thomas, G. (1975), Acta Metall. 23, de Fontaine, D. (1973), J. Phys. Chem. Sol. 34, 1285.
1485. de Fontaine, D. (1982), in: Metallurgical Treatises:
Brailsford, A. D., Wynblatt, P. (1979), Acta Metall. Tien, J. K., Elliott, J. F. (Eds.). Warrendale, PA:
27, 489. The Metallurgical Society of AIME, pp. 423 – 444.
Bratland, D. H., Grong, O., Shercliff, H., Myhr, (Kinetics of First Order Transitions)
O. R., Tjotta, S. (1997), Acta Mater. 45, 1. Doherty, R. D. (1982), Metal. Science 16, 1.
Brenner, S. S., Camus, P. P., Miller, M. K., Soffa, W. Doherty, R. D. (1983), in: Physical Metallurgy:
A. (1984), Acta Metall. 32, 1217. Cahn, R. W., Haasen, P. (Eds.), Vol. 2. Amsterdam:
Cahn, J. W. (1962), Acta Metall. 10, 179, 907. North Holland Physics Publishing, pp. 933 – 1030.
Cahn, J. W. (1965), J. Chem. 42, 93. Doi, M., Miyazaki, T. (1986), Mater. Sci. Eng. 78,
Cahn, J. W. (1966), Trans. AIME 242, 166. 87.
Cahn, J. W., Charles, R. J. (1965), Phys. Chem. Doi, M., Miyazaki, T., Wakatsuki, T. (1984), Mater.
Glasses 6, 181. Sci. Eng. 67, 247.
Cahn, J. W., Hilliard, J. E. (1958), J. Chem. Phys. 28, Doi, M., Fukaya, M., Miyazaki, T. (1988), Phil. Mag.
258. A 57, 821.
Cahn, J. W., Hilliard, J. E. (1959 a), J. Chem. Phys. Dutta, I., Allen, S. M. (1991), J. Mater. Sci. Lett. 10,
31, 539. 323.
Cahn, J. W., Hilliard, J. E. (1959 b), J. Chem. Phys. Eckerlebe, H., Kampmann, R., Wagner, R. (1986),
31, 688. Atomic Transport and Defects in Metals by Neu-
Calderon, H. A., Voorhees, P. W., Murray, J. L., Kos- tron Scattering: Janot, C., Petry, W., Richter, D.,
torz, G. (1994), Acta Metall. Mater. 42, 991. Springer, T. (Eds.). Berlin: Springer Verlag,
Cerri, A., Schmelczer, R., Schwander, P., Kostorz, pp. 66 – 72.
G., Wright, A. F. (1987), in: Proc. MRS Symp., Vol. Edwards, G. A., Stiller, K., Dunlop, G. L., Couper,
82: Characterization of Defects in Materials: Sie- M. J. (1998), Acta Mater. 46, 3893.
gel, R. W., Weertman, J. R., Sinclair, R. (Eds.). Enomoto, Y., Kawasaki, K., Tokuyama. M. (1987),
Pittsburgh: Materials Research Society, p. 169. Acta Metall. 35, 907, 815.
Cerri, A., Schönfeld, B., Kostorz, G. (1990), Phys. Epperson, J. E. (1988), in: Dynamics of Ordering
Rev. B 42, 958. Processes in Condensed Matter: Komura, S., Fu-
Chan, K. S., Lee, J. K., Shiflet, G. J., Russell, K. C., rukawa, H. (Eds.). New York: Plenum Press, pp.
Aaronson, H. I. (1978), Metall. Trans. 9 A, 1016. 245 – 250.
Che, D. Z., Spooner, S., Hoyt, J. J. (1997), Acta Ernst, F., Haasen, P. (1988), Phys. Stat. Sol. (a), 104,
Mater. 45, 1167. 403.
Chen, H., Polat, S., Epperson, J. E. (1988), in: Dy- Ernst, F., Xiao, S. Q., Haider, F., Wilbrandt, P.-J.
namics of Ordering Processes in Condensed Mat- (1987), Scripta Metall. 21, 1189.
ter: Komura, S., Furukawa, H. (Eds.). New York: Eshelby, J. D. (1957), Proc. R. Soc. A 241, 376.
Plenum Press, pp. 245 – 250. Fährmann, M., Fratzl, P., Paris, P., Fährmann, E.,
Cho, J. H., Ardell, A. J. (1997), Acta Mater. 45, Johnson, W. C. (1995), Acta Metall. Mater. 43,
1393. 1007.
Christian, J. W. (1975), The Theory of Transforma- Feder, J., Russell, K. C., Lothe, J., Pound, G. M.
tions in Metals and Alloys, 2nd edn. Oxford: Per- (1966), Adv. Phys. 15, 111.
gamon Press. (General Diffusive Phase Transfor- Ferguson, P., Jack, K. H. (1984), Phil. Mag. A 50,
mations in Solids) 221.
Cockayne, D. J. H., Gronsky, R. (1981), Phil. Mag. Ferguson, P., Jack, K. H. (1985), Phil. Mag. A 52,
A 44, 159. 509.
Cook, H. E. (1970), Acta Metall. 18, 297. Forouhi, A. R., de Fontaine, D. (1987), Acta Metall.
Craievich, A. (1975), Phys. Chem. Glasses 16, 133. 35, 1863.
Craievich, A. F., Olivieri, J. R. (1981), J. Appl. Cryst. Fratzl, P., Penrose, O. (1994), Phys. Rev. B 50, 3477.
14, 444. Fratzl, P., Penrose, O. (1996), Acta Metall. 44, 3025.
Craievich, A., Sanchez, J. M. (1981), Phys. Rev. Lett. Fratzl, P., Lebowitz, J. L., Marro, J., Kalos, M. H.
47, 1308. (1983), Acta Metall. 31, 1849.
Craievich, A. F., Sanchez, J. M., Williams, C. E. Frenkel, J. (1939), J. Phys. (USSR) 1, 315.
(1986), Phys. Rev. B 34, 2762. Furukawa, H. (1981), Phys. Rev. A 23, 1535.
Furusaka, M., Ishikawa, Y., Yamaguchi, S., Fujino, Hirata, T., Kirkwood, D. H. (1977), Acta Metall. 25,
Y. (1986), J. Phys. Soc. Japan 55, 2253. 1425.
Gaskell, D. R. (1983), in: Physical Metallurgy, Vol. Hobbs, L. W., Westmacott, K. W., Williams, D. B.
1: Cahn, R. W., Haasen, P. (Eds.). Amsterdam: (Eds.) (1986), Proc. of the Materials Research
North Holland Physics Publishing, pp. 271 – 326. Society Symposium, Vol. 62: Materials Problem
(Metallurgical Thermodynamics) Solving with the Transmission Electron Micro-
Gaulin, B. D., Spooner, S., Morii, Y. (1987), Phys. scope. (Experimental Techniques; CTEM)
Rev. Lett. 59, 668. Horiuchi, S., Izumi, F., Kikuchi, T., Uchida, K.
Gerling, R., Schimansky, F.-P., Wagner, R. (1988), (1984), Phil. Mag. A 50, L 29.
Acta Metall. 36, 575. Hornbogen, E. (1967), Aluminium 43, 115.
Gerold, V., Merz, W. (1967), Scripta Metall. 1, 33. Hoyt, J. J. (1989), Acta Metall. 37, 2489.
Glatter, O. (1982), in: Small Angle X-Ray Scattering: Hoyt, J. J., de Fontaine, D. (1989), Acta Metall. 37,
Glatter, O., Kratky, O. (Eds.). London: Academic, 1611.
p. 167. Hoyt, J. J., Sluiter, M., Clark, B., Kraitchman, M., de
Gleiter, H. (1983), in: Physical Metallurgy: Cahn, R. Fontaine, D. (1987), Acta Metall. 35, 2315.
W., Haasen, P. (Eds.), Vol. 1. Amsterdam: North Hoyt, J. J., Clark, B., de Fontaine, D. (1989), Acta
Holland Physics Publishing, pp. 649 – 713. Metall. 37, 1597.
Goldstein, M. (1965), J. Am. Ceram. Soc. 48, 126. Hütten, A., Haasen, P. (1986), J. Phys. (Paris) C 7,
Grong, O., Myhr, O. R. (2000), Acta Mater. 48, 205.
445. Izumitani, T., Hashimoto, T. (1985), J. Chem. Phys.
Gronsky, R., Okada, M., Sinclair, R., Thomas, G. 83, 3694.
(1975), 33rd Ann. Proc. E.M.S.A., 22. Jantzen, C. M. F., Herman, H. (1978), in: Phase Dia-
Grüne, R. (1988), Acta Metall. 36, 2797. grams: Materials Science and Technology, Vol. 5:
Grüne, R., Oehring, M., Wagner, R., Haasen, P. Alper, A. M. (Ed.). New York: Academic Press,
(1985), in: Rapidly Quenched Metals, Vol. 1: pp. 128 – 185.
Steeb, S., Warlimont, H. (Eds.). Amsterdam: North Johnson, W. C. (1984), Acta Metall. 32, 465.
Holland, pp. 761 – 765. Johnson, W. C., Cahn, J. W. (1984), Acta Metall. 23,
Gunton, J. D. (1984), in: Decomposition of Alloys: 1839.
the Early Stages: Haasen, P., Gerold, V., Wagner, Johnson, W. C., Abinandanan, T. A., Voorhees, P. W.
R., Ashby, M. F. (Eds.). Oxford: Pergamon Press, (1990), Acta Metall. 38, 1349.
pp. 1–10. Kalos, M., Lebowitz, J. L., Penrose, O., Sur, A.
Gunton, J. D., Droz, M. (1984), Lecture Notes in (1978), J. Stat. Phys. 18, 39.
Physics, Vol. 183 – Introduction to the Theory of Kampmann, R., Wagner, R. (1984), in: Decomposi-
Metastable and Unstable States. Berlin: Springer tion of Alloys: the Early Stages: Haasen, P., Ge-
Verlag. (Kinetics of First Order Transitions) rold, V., Wagner, R., Ashby, M. F. (Eds.). Oxford:
Gupta, A. K., Lloyd, D. J. (1992), in: Proc. 3rd Int. Pergamon Press, pp. 91 – 103.
Conf. on Alluminium Alloys: Their Physical and Kampmann, R., Wagner, R. (1986), in: Atomic Trans-
Mechanical Properties, Vol. 2: Arnberg, L. (Ed.). port and Defects in Metals by Neutron Scattering:
Trondheim: Norwegian Institute of Technology Janot, C., Petry, W., Richter, D., Springer, T.
and SINTEF Metallurgy, p. 21. (Eds.). Berlin: Springer Verlag, pp. 73 – 77.
Guyot, P., Simon, J. P. (1981), in: Solid – Solid Phase Kampmann, R., Eckerlebe, H., Wagner, R. (1987),
Transformation: Aaronson, H. I., Laughlin, D. E., in: Phase Transitions in Condensed Systems – Ex-
Sekerka, R. F., Wayman, C. M. (Eds.). New York: periments and Theory: Cargill, G. S., Spaepen, F.,
Metallurgical Society of A.I.M.E., p. 325. Tu, K.-N. (Eds.). MRS-Symp. Proceedings, Vol.
Haasen, P., Wagner, R. (1985), Ann. Rev. Mat. Sci. 57: Materials Research Society, Pittsburgh, PA, pp.
15, 43. 525 – 542.
Hardy, S. C., Voorhees, P. W. (1988), Metall. Trans. Kampmann, R., Ebel, T., Haese, M., Wagner, R.
19 A, 2713. (1992), Phys. Stat. Sol. B. 172, 295.
Hashimoto, T., Itakura, M., Hasegawa, H. (1986 a), Katano, S., Iizumi, M. (1982), J. Phys. Soc. Japan
J. Chem. Phys. 85, 6118. 51, 347.
Hashimoto, T., Itakura, M., Hasegawa, H. (1986 b), Katano, S., Iizumi, M. (1984), Phys. Rev. Lett. 52, 835.
J. Chem. Phys. 85, 6773. Kaufman, L. (Ed.) (1977), CALPHAD, 1. Oxford:
Hennion, M., Ronzaud, D., Guyot, P. (1982), Acta Pergamon Press. (Volumes are published annually,
Metall. 30, 599. from 1977 onwards.)
Herring,, C. (1953), in: Structure and Properties of Kaufman, L., Bernstein, H. (1970), Computer Calcu-
Solid Surfaces: Gomer, R., Smith, C. S. (Eds.). lations of Phase Diagrams. New York: Academic
Chicago, IL: Univ. of Chicago Press, pp. 5 – 149. Press.
Hilliard, J. E. (1970), in: Phase Transformations: Kaufmann, M. J., Voorhees, P. W., Johnson, C.,
Aaronson, H. I. (Ed.). Metals Park, Ohio: Am. Soc. Biancaniello, F. S. (1989), Metall. Trans. 20 A,
of Metals, pp. 497 – 560. 2171.
5.11 References 405
Kawasaki, K. (1972), in: Phase Transitions and Crit- LeGoues, F. K., Lee, Y. W., Aaronson, H. I. (1984 b),
ical Phenomena, Vol. 2: Domb, C., Green, M. S. Acta Metall. 32, 1845.
(Eds.). New York: Academic Press, p. 2. LeGoues, F. K., Wright, R. N., Lee, Y. W., Aaronson,
Khachaturyan, A. G. (1983), Theory of Phase Trans- H. I. (1984 c), Acta Metall. 32, 1865.
formations in Alloys. New York: John Wiley. Leo, P. H., Mullins, W. W., Sekerka, R. F., Vinals, J.
Khachaturyan, A. G., Airapetyan, V. M. (1974), (1990), Acta Metall. Mater. 38, 1573.
Phys. Stat. Sol. 26, 61. Lifshitz, I. M., Slyozov, V. V. (1961), Phys. Chem.
Khachaturyan, A. G., Semenovskaya, S. V., Morris, Solids 19, 35.
J. W. (1988), Acta Metall. Mater. 36, 1563. Liu, Z. G., Wagner, R. (1984), J. Phys. Colloq. C 9,
Kirkwood, D. H. (1970), Acta Metall. 18, 563. 441.
Kostorz, G. (Ed.) (1979), Neutron Scattering. New Livak, R. J., Thomas, G. (1974), Acta Metall. 22,
York: Academic, p. 227. (Experimental Techniques; 589.
SANS) Lyon, O., Simon, J. P. (1987), Phys. Rev. B 35, 5164.
Kuo, S. Y., Virkar, A. V. (1987), J. Am. Ceram. Soc. Lyon, O., Simon, J. P. (1988), J. Phys. F, Metal Phys-
70, C-125. ics 18, 1787.
Langer, J. S. (1971), Ann. Phys. 65, 53. Maheshwari, A., Ardell, A. J. (1992), Scripta Metall.
Langer, J. S. (1973), Acta Metall. 21, 1649. 26, 347.
Langer, J. S. (1975), in: Fluctuations, Instabilities, Mandyam, H., Glicksman, M. E., Helsing, J., Marsh,
and Phase Transitions: Riste, T. (Ed.). New York: S. P. (1998), Phys. Rev. E 58, 2119.
Plenum Press, pp. 19 – 42. Marder, M. (1987), Phys. Rev. A 36, 858.
Langer, J. S., Schwartz, A. J. (1980), Phys. Rev. A 21, Marqusee, J. A., Ross, J. (1984), J. Chem. Phys. 79,
948. 373.
Langer, J. S., Bar-On, Miller, H. D. (1975), Phys. Marqusee, J. A., Ross, J. (1984), J. Chem. Phys. 80,
Rev. A 11, 1417. 536.
Langmayr, F., Fratzl, P., Vogl, G. (1994), Phys. Rev. Marro, J., Bortz, A. B., Kalos, M. H., Lebowitz, J. L.
B 49, 11759. (1975), Phys. Rev. B 12, 2000.
La Salle, J. C., Schwartz, L. H. (1984), in: Decompo- Marro, J., Bortz, A. B., Kalos, M. H., Lebowitz, J. L.
sition of Alloys: the Early Stages: Haasen, P., Ge- (1977), Phys. Rev. B 15, 3014.
rold, V., Wagner, R., Ashby, M. F. (Eds.). Oxford: Marsh, S. P., Glicksman, M. E. (1996), Acta Mater.
Pergamon Press, pp. 104 – 109. 44, 3761.
Laughlin, D. E. (1976), Acta Metall. 24, 53. Martin, G. (1978), in: Solid State Phase Transforma-
Laughlin, D. E., Cahn, J. W. (1975), Acta Metall. 23, tions in Metals and Alloys: Aussois: Les Editions
329. de Physique, pp. 337 – 406. (Kinetics of Phase
Lê, T. N., Barbu, A., Maury, F. (1992), Scripta Met- Separation in Solids)
all. Mater. 26, 771. Martin, J. W., Doherty, R. D. (1976), Stability of Mi-
Lebowitz, J. L., Kalos, M. H. (1976), Scripta Metall. crostructure in Metallic Systems. Cambridge: Uni-
10, 9. versity Press, p. 163.
Lebowitz, J. L., Marro, J., Kalos, M. H. (1982), Acta Miller, M. K., Brenner, S. S., Burke, M. G., Soffa, W.
Metall. 30, 297. A. (1984), Scripta Metall. 18, 111.
Lee, J. K. (1997), Mater. Sci. Eng, A 238, 1. Miller, M. K., Horton, L. L., Spooner, S. (1986), in:
Lee, J. K. (2000), Theor. Appl. Fract. Mech. 33, Proc. 32nd Intl. Field Emission Symp.: J. Phys.
207. (Paris) 47, Coloque C 2: Miller, M. K., Brenner,
Lee, J. H., Johnson, W. C. (1982), in: Solid-State S. S. (Eds.). Paris: Les Editions de Physique, pp.
Phase Transformations: Aaronson, H. I., Laugh- 409 – 416.
lin, D. E., Sekerka, D. F., Wayman, C. M. (Eds.). Miller, M. K., Cerezo, A., Hetherington, M. G.,
Warrendale: The Metallurgical Society of AIME, Smith, G. D. W. (1996), Atom-Probe Field Ion Mi-
pp. 127 – 150. croscopy, Oxford: Oxford Science Publications.
Lee, J. K., Barnett, D. M., Aaronson, H. I. (1977), (Tomographic Atom-Probe)
Metall. Trans. 8 A, 963. Mirold, P., Binder, K. (1977), Acta Metall. 25, 1435.
Lee, Y. W., Aaronson, H. I. (1980), Acta Metall. 28, Miyazaki, T., Doi, M. (1989), Mater. Sci. Eng. A 110,
539. 175.
LeGoues, F. K., Aaronson, H. I. (1984), Acta Metall. Miyazaki, T., Imamura, H., Kozakai, T. (1982), J.
32, 1855. Mater. Sci. Eng. 54, 9.
LeGoues, F. K., Aaronson, H. I., Lee, Y. W., Fix, G. Miyazaki, T., Seki, K., Doi, M., Kozakai, T. (1986),
J. (1982), in: Solid – Solid Phase Transformations: J. Mater. Sci. Eng. 77, 125.
Aaronson, H. I., Laughlin, D. E., Sekerka, R. F., Morral, J. E., Cahn, J. W. (1971), Acta Metall. 19,
Wayman, C. M. (Eds.). Warrendale: The Metallur- 1037.
gical Society of AIME, pp. 427 – 431. Mueller, R., Gross, D. (1998), Comp. Mat. Sci., 11,
LeGoues, F. K., Lee, Y. W., Aaronson, H. I. (1984 a), 25.
Acta Metall. 32, 1837. Mullins, W. W. (1986), J. Appl. Phys. 59, 1341.
Murata, Y., Iwama, Y. (1981), Trans. Japan. Inst. of Sato, K., Tagawa, K., Inoue, Y. (1989), Mater. Sci.
Metals 22, 433. Engin. A 111, 45.
Myhr, O. R., Grong, O. (1991 a), Acta Metall. 39, Sauthoff, G., Pitsch, W. (1987), Phil. Mag. B 56, 471.
2693. Schmidt, I., Gross, D. (1997), J. Mech. Phys. Solids
Myhr, O. R., Grong, O. (1991 b), Acta Metall. 39. 45, 1521.
2703. Schmidt, I., Mueller, R., Gross, D. (1998), Mech.
Neilson, G. F. (1969), Phys. Chem. Glasses 10, 54. Mater. 30, 181.
Nishi, T., Wang, T. T., Kwei, T. K. (1975), Macro- Schmitz, G., Haasen, P. (1992), Acta Metall. Mater.
molecules 8, 227. 40, 2209.
Nishimori, H., Onuki, A. (1990), Phys. Rev. B 42, 980. Sequeira, A. D., Calderon, H. A., Kostorz, G., Peder-
Okada, M., Han, C. C. (1986), J. Chem. Phys. 85, sen, J. S. (1995), Acta Metall. Mater. 43, 3427.
5317. Servi, I. S., Turnbull, D. (1966), Acta Metall. 14, 161.
Onsoien, M. I., Grong, O., Gundersen, O., Skaland, Shercliff, H. R., Grong, O., Myhr, O. R., Ashby,
T. (1999), Metall. Mater. Trans. 30 A, 1069. M. F. (1992), in: Proc. 3rd Int. Conf. on Aluminium
Onuki, A., Nishimori, H. (1991), Phys. Rev. B 43, Alloys – Their Physical, Mechanical Properties,
13649. Vol. 3: Arnberg, L. (Ed.). Trondheim: Norwegian
Pareige, P., Soisson, F., Martin, G., Blavette, D. Institute of Technology and SINTEF Metallurgy,
(1999), Acta Mater. 47, 1889. p. 357.
Paris, O., Langmayr, F., Vogl, G., Fratzl, P. (1995), Z. Shercliff, H. R., Ashby, M. F. (1990 a), Acta Metall.
Metallkd. 86, 860. 38, 1789.
Park, M. W., Mitchell, T. E., Heuer, A. H. (1975), J. Shercliff, H. R., Ashby, M. F. (1990 b), Acta Metall.
Am. Ceram. Soc. 58, 43. 38, 1803.
Park, M. W., Mitchell, T. E., Heuer, A. H. (1976), Shewmon, P. G. (1965), Trans. TMS-AIME 233, 736.
J. Mater. Sci. 11, 1227. Shiflet, G. J., Lee, Y. W., Aaronson, H. I., Russell, K.
Penrose, O. (1978), in: Lecture Notes in Physics, Vol. C. (1981), Scripta Metall. 15, 719.
84: Garrido, L., Seglar, P., Shepherd, P. J. (Eds.). Si-Qun Xiao, Haasen, P. (1991), Acta Metall. 39, 65.
Heidelberg: Springer Verlag, pp. 210 – 234. Simon, J. P., Lyon, O. (1989), Acta Metall. 37, 1727.
Penrose, O., Lebowitz, J. L. (1979), in: Studies in Simon, J. P., Guyot, P., DeSalva, A. G. (1984), Phil.
Statistical Mechanics, Vol. 7: Lebowitz, J. L., Mag. A 49, 151.
Montroll, E. (Eds.). Amsterdam: North Holland, Sinclair, R., Thomas, G. (1974), 32nd Ann. Proc.
pp. 293 – 340. (Kinetics of First Order Transitions) E.M.S.A., 500.
Penrose, O., Lebowitz, J. L., Marro, J., Kalos, M. H., Singh, J., Lele, S., Ranganathan, S. (1980), J. Mater.
Sur, A. (1978), J. Stat. Phys. 19, 243 – 267. Sci. 15, 2010.
Pike, B. C., Messoloras, S., Stewart, R. J. (1989), Singhai, S. P., Herman, H., Kostorz, G. (1978), J.
Scripta Metall. 23, 983. Appl. Cryst. 11, 572.
Piller, J., Wagner, W., Wollenberger, H., Mertens, P. Skripov, V. P., Skripov, A. V. (1979), Sov. Phys. Usp.
(1984), in: Decomposition of Alloys: the Early 22, 389.
Stages: Haasen, P., Gerold, V., Wagner, R., Ashby, Sluiter, M., Kawazoe, Y. (1996), Phys. Rev. B 54,
M. F. (Eds.). Oxford: Pergamon Press, pp. 156 – 10381.
164. Smith, D. J. (1983), Helv. Phys. Acta 56, 463. (Ex-
Polat, S., Marsh, C., Little, T., Ju, C. P., Epperson, J. perimental Techniques; HREM)
E., Haydn Chen (1986), Scripta Metall. 20, 1759. Snyder, H. L., Meakin, P. (1983 a), J. Chem. Phys.
Polat, S., Haydn Chen, Epperson, J. E. (1989), Met. Soc. 79, 5588.
Trans. 20 A, 699. Synder, H. L., Meakin, P. (1983 b), Macromolecules
Purdy, G., Bréchet, Y. J. M. (2001), in: Phase Trans- 16, 757.
formations in Materials: Kostorz, G. (Ed.). Wein- Snyder, H. L., Meakin, P. (1985), J. Polym. Sci. 73,
heim: Wiley-VCH, Chap. 7. 217.
Radmilovic, V., Fox, A. G., Thomas, G. (1989), Acta Soisson, F., Barbu, A., Martin, G. (1996), Acta
Metall. 37, 2385. Mater. 44, 3789.
Rikvold, P. A., Gunton, J. D. (1982), Phys. Rev. Lett. Staron, P., Kampmann, R. (2000 a), Acta Mater. 48,
49, 286. 701.
Rindone, G. E. (1975), Bull. Cent. Glass Ceram. Re- Staron, P., Kampmann, R. (2000 b), Acta Mater. 48,
search Inst. Calcutta 22, 119. 713.
Rioja, R. J., Laughlin, D. A. (1977), Metall. Trans. Stauffer, D. (1979), Phys. Rep. 54, 1.
8 A, 1257. Steiner, D., Beddoe, R., Gerold, V., Kostorz, G.,
Rundman, K. B., Hilliard, J. E. (1967), Acta Metall. Schmelzcer, R. (1983), Scripta Metall. 17, 733.
15, 1025. Stubican, V. S., Schultz, A. H. (1970), J. Am. Ceram.
Russell, K. C. (1980), Adv. Colloid Interface Sci. 13, Soc. 53, 211.
205. Su, C. H., Voorhees, P. W. (1996), Acta Metall.
Sato, R., Han, C. C. (1988), J. Chem. Phys. 88, 2057. Mater. 44, 1987.
5.11 References 407
Sur, A., Lebowitz, J. L., Marro, J., Kalos, M. H. morrow – Technology, Products, Research: Cere-
(1977), Phys. Rev. B 15, 3014. sara, S. (Ed.). Barga die Lucca, pp. 119 – 127.
Taylor, K. A. (1985), Sc.D. Thesis Aging Phenomena Wahi, R. P., Stajer, J. (1984), in: Decomposition of
in Ferrous Martensites. MIT, Cambridge, MA. Alloys: the Early Stages: Haasen, P., Gerold, V.,
Thompson, M. E., Voorhees, P. W. (1999), Acta Wagner, R., Ashby, M. F. (Eds.). Oxford: Per-
Mater. 47, 983. gamon Press, pp. 165 – 169.
Thompson, M. E., Su, C. S., Voorhees, P. W. (1994), Wang, Y., Khachaturyan, A. G. (1995), Acta Metall.
Acta Metall. 42, 2107. Mater. 43, 1837.
Tien, J. K., Shewmon, P. G., Foster, J. S. (1973), Wang, Y., Chen, L.-Q., Khachaturyan, A. G. (1991),
Scripta Metall. 7, 1171. Scripta Metall. Mater. 25, 1387.
Tokuyama, M., Kawasaki, K. (1984), Physica A 123, Wang, Y., Chen, L.-Q., Khachaturyan, A. G. (1992),
386. Phys. Rev. B 46, 11194.
Tsumuraya, K., Miyata, Y. (1983), Acta Metall. 31, Weins, J. J., Cahn, J. W. (1973), in: Sintering and Re-
437. lated Phenomena, Kuczynski, G. C. (Ed.). Lon-
Ujihara, T., Osamura, K. (2000), Acta Mater. 48, don: Plenum, p. 151.
1629. Wendt, H., Haasen, P. (1983), Acta Metall. 31, 1649.
Vintaikin, Y. E., Dmitriyev, V. B., Udovenko, V. A. West, A. W., Kirkwood, D. H. (1976), Scripta Metall.
(1979), Phys. Met. Metall. 46, 97. 10, 681.
Volmer, M., Weber, A. (1926), Z. Phys. Chem. 119, Williams, D. B., Carter, C. B. (1996), Transmission
277. Electron Microscopy. New York and London: Ple-
von Alvensleben, L., Wagner, R. (1984), in: Decom- num Press, p. 683.
position of Alloys: the Early Stages: Haasen, P., Wiltzius, P., Bates, F. S., Dierker, S. B., Wignall, G.
Gerold, V., Wagner, R., Ashby, M. F. (Eds.). Ox- D. (1987), Phys. Rev. A 36, 2991.
ford: Pergamon Press, pp. 143 – 148. Wiltzius, P., Bates, F. S., Heffner, W. R. (1988), Phys.
Voorhees, P. W., Glicksman, M. E. (1984), Acta Rev. Lett. 60, 1538.
Metall. 32, 2001, 2013. Wood, J. V., Mills, P. F., Bingham, J. K., Bee, J. V.
Voorhees, P. W., Johnson, W. C. (1988), Phys. Rev. (1979), Metall. Trans. 10 A, 525.
Lett. 61, 225. Yaldram, K., Binder, K. (1991 a), Z. Phys. B 82, 405.
Voorhees, P. W., Schaefer, R. J. (1987), Acta Metall. Yaldram, K., Binder, K. (1991 b), Acta Metall. Mater.
35, 327. 39, 707.
Wagner, C. (1961), Z. Elektrochem. 65, 581. Yao, J. H., Elder, K. R., Guo, H., Grant, M. (1993),
Wagner, R. (1982), Field Ion Microscopy in Ma- Phys. Rev. B 47, 14110.
terials Science, Vol. 6, Crystals – Growth, Proper- Yin, F., Oksawa, Y., Sato, A., Kawakara, K. (2000),
ties and Applications: Freyhardt, H. C. (Ed.). Ber- Acta Mater. 48, 1273.
lin: Springer Verlag. (Experimental Techniques; Yoo, Y. S., Yoon, D. Y., Henry, M. F. (1995), Metals
AFIM) and Mater. 1, 47.
Wagner, R., Brenner, S. S. (1978), Acta Metall. 26, Yoshida, S., Fukaya, M., Miyazaki, T. (1987), J. Ja-
197. pan. Inst. Metals 51, 18.
Wagner, W., Poerschke, R., Wollenberger, H. (1984), Zeldovich, J. B. (1943), Acta Physicochim. (USSR)
in: Decomposition of Alloys: the Early Stages: 18, 1.
Haasen, P., Gerold, V., Wagner, R., Ashby, M. F. Zener, C. (1949), J. Appl. Phys. 20, 950.
(Eds.). Oxford: Pergamon Press, pp. 170 – 179. Zhu, F., Haasen, P., Wagner, R. (1986), Acta Metall.
Wagner, R., Kampmann, R., Jiang, B., Beaven, P. A. 34, 457.
(1988), in: Proc. Int. Conf. Cu ’86 – Copper To-
6 Spinodal Decomposition
Kurt Binder
Institut für Physik, Johannes Gutenberg-Universität Mainz,

Mainz, Federal Republic of Germany
Peter Fratzl
Erich-Schmid-Institut der Österreichischen Akademie der Wissenschaften

und Montan-Universität Leoben, Leoben, Austria

6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 414
6.2 General Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 416
6.2.1 Phenomenological Thermodynamics of Binary Mixtures
and the Basic Ideas of Phase Separation Kinetics . . . . . . . . . . . . . . 416
6.2.2 The Cahn – Hilliard – Cook Nonlinear Diffusion Equation . . . . . . . . . . 420
6.2.3 Linearized Theory of Spinodal Decomposition . . . . . . . . . . . . . . . 421
6.2.4 Spinodal Decomposition of Polymer Mixtures . . . . . . . . . . . . . . . 426
6.2.5 Significance of the Spinodal Curve . . . . . . . . . . . . . . . . . . . . . 428
6.2.6 Towards a Nonlinear Theory of Spinodal Decomposition
in Solids and Fluids . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 437
6.2.7 Effects of Finite Quench Rate . . . . . . . . . . . . . . . . . . . . . . . . 441
6.2.8 Interconnected Precipitated Structure Versus Isolated Droplets,
and the Percolation Transition . . . . . . . . . . . . . . . . . . . . . . . . 442
6.2.9 Coarsening and Late Stage Scaling . . . . . . . . . . . . . . . . . . . . . 445
6.2.10 Effects of Coherent Elastic Misfit . . . . . . . . . . . . . . . . . . . . . . 447
6.3 Survey of Experimental Results . . . . . . . . . . . . . . . . . . . . . . 450
6.3.1 Metallic Alloys . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 451
6.3.2 Glasses, Ceramics, and Other Solid Materials . . . . . . . . . . . . . . . . 454
6.3.3 Fluid Mixtures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 455
6.3.4 Polymer Mixtures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 457
6.4 Extensions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 460
6.4.1 Systems Near a Tricritical Point . . . . . . . . . . . . . . . . . . . . . . . 460
6.4.2 Spontaneous Growth of Ordered Domains out of Initially Disordered
Configurations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 462
6.4.3 Phase Separation in Reduced Geometry and near Surfaces . . . . . . . . . 467
6.4.4 Effects of Quenched Impurities; Vacancies; Electrical Resistivity
of Metallic Alloys Undergoing Phase Changes . . . . . . . . . . . . . . . 468
6.4.5 Further Related Phenomena . . . . . . . . . . . . . . . . . . . . . . . . . 470

ISBN: 3-527-30256-5
410 6 Spinodal Decomposition
6.5 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 471

6.7 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 474

a lattice spacing
A, B, … symbols for chemical elements
a (t) structural relaxation variable
A, B symbols for Landau expansion coefficients or other constants
a0 , …, a3 constants
B̂ critical amplitude of the order parameter
c (x), ci local concentration, concentration of site i
Ĉ critical amplitude of the scattering function
cBs , csp (T ) concentration at the spinodal curve
ccoex concentration at the coexistence curve
c– average concentration
ccrit critical concentration
d spatial dimensionality
D0 diffusion constant
D2 damping coefficient of second sound
Dt tracer diffusion coefficient
e electron charge
eij elastic strain tensor
eij0 elastic misfit tensor
Eact activation energy
F thermodynamic free energy
DᏲ free energy function
DF* nucleation free-energy barrier
F* MF mean-field result for the nucleation barrier
f0 parameter or universal constant of order unity
fcg free energy density (coarse-grained)
fD Debye function
g cooling rate
Ᏼ Hamiltonian
ä Planck constant
J concentration current density
J interaction strength of the Ising model
K elastic modulus
k wavevector, scattering vector in diffraction experiments
k modulus of k
kc critical wavenumber in Cahn’s theory
kF radius of Fermi sphere
km wavenumber of maximal growth
L length scale of coarse-graining cell
l number of atoms within a cluster
ᏸ Liouville operator
M mobility
meff effective mass of an electron

N chain length of a polymer
N number of atoms per cm3
nL concentration of a cluster containing L atoms
PL (M) magnetization distribution function
q rescaled wavenumber
QB superstructure Bragg spot
r range of effective interaction
r radius of the grain
·R 2 Ò mean-square end-to-end distance of a polymer chain
·R 2gyr Ò gyration radius of a polymer chain
R (k) rate factor
r0 phenomenological coefficient
S (k, t) equal-time structure factor
s (x, t) entropy density
S̃ rescaled structure factor
T temperature
t time
T0 temperature at which a quench is started at t = 0
Tab Oseen tensor (component)
Ti , Tf initial, final temperature
tij elastic stress tensor
Tt tricritical point
u phenomenological coefficient of the Landau expansion
u2 second sound velocity
V volume
v phenomenological coefficient of the Landau expansion
w elastic energy density
W (l, l¢) cluster reaction matrix
WA , WB time constant of element A, B
x Lifshitz – Slyozov exponent
x position vector
x, y, z spatial coordinates
XA , XB degree of polymerization of a polymer chain of type A, B
ᒗ scaling variable
Z-A%, Z-B% atomic numbers of constituents A, B
a phenomenological coupling constant

b critical exponent of the order parameter
g critical exponent of the structure factor
g phenomenological coefficient
G, g rate factors
h viscosity
ha change of lattice constant with alloy composition
hT random force
Q temperature at which behavior of polymer in solution is ideal
k factor describing screening of Coulomb interaction
l wavelength
lc critical wavelength in Cahn’s theory
lijmn elastic stiffness tensor
L generalized Onsager coefficient
m chemical potential difference
m elastic modulus
n critical exponent of the correlation length
nP Poisson coefficient in elasticity theory
x interfacial width; correlation length
xˆ critical amplitude of the correlation length
xcoex correlation length at the coexistence curve
r1, r 2 , … probability distributions
s subunit length of a polymer
s ij elastic stress tensor
t rescaled coordinate (time)
t time constant
j pair interaction energy
F volume fraction
Fc volume fraction at the critical point
c Flory – Huggins parameter
cs Flory – Huggins parameter at the spinodal curve
c n–1 phenomenological coefficient
y complex order parameter
ABV element-A – element-B – vacancy model

CHC Cahn – Hilliard – Cook (method)
ESA European Space Association
LBM Langer – Baron – Miller (method)
LSW Lifshitz – Slyozov – Wagner (method)
m (index) maximum
MF mean field
p (index) polarization
V vacancy (also as index)
6.1 Introduction only two values: ci = 1, if the site is taken

by a B atom, and ci = 0, if it is taken by an
This chapter deals with the dynamics of A atom in an AB mixture. Instead here we
phase changes in materials, which are are interested in an averaged concentration
caused by transferring the material into an field obtained by “coarse graining” over a
initial state that is not thermodynamically cell of size L3 (or L2 if we consider a two-
stable (e.g. by rapid cooling (“quenching”) dimensional layer):
or, occasionally, rapid heating; in fluids the
system may also be prepared by a rapid c ( x ) = L− 3 ∑ ci (6-1)
i Œ cell
pressure change). It is generally believed
(Gunton et al., 1983; Binder, 1984 a, 1987 a, where x is the center of gravity of the cell
1989; Kostorz, 1991; Wagner and Kamp- over which ci is averaged. This intermedi-
mann, 2000) that in homogeneous materi- ate length scale L must be large in compar-
als such phase changes may be initiated by ison with the lattice spacing, since only if
two different types of mechanism, corre- the cell contains many lattice sites is a con-
sponding to two different types of statisti- tinuum description as anticipated in Fig.
cal fluctuations, namely “homophase fluc- 6-1 warranted. On the other hand, L must
tuations” and “heterophase fluctuations”. be small in comparison with the length
Fig. 6-1 illustrates these fluctuations qual- scales of the statistical fluctuations that we
itatively for the example of a binary mix- wish to consider, i.e., the wavelength l of
ture, where the variable to consider is a lo- the concentration wave in Fig. 6-1 a, or the
cal concentration variable c (x). By “local” width x of the interface between a droplet
we do not mean the scale of a lattice site i of the new phase and its environment in
(in a crystalline solid), since then the asso- Fig. 6-1 b. As we shall see, the existence of
ciated concentration variable ci could take this intermediate length scale L severely re-
Figure 6-1. Schematic diagram of unstable thermodynamic fluctuations in the two-phase regime of a binary
mixture AB at a concentration cB (a) in the unstable regime inside two branches cBs of the spinodal curve and (b)
in the metastable regime between the spinodal curve cBs and the coexistence curve c coex (1)
. The local concentration
c (r) at a point r = (x, y, z) in space is schematically plotted against the spatial coordinate x at some time t after
the quench. In case (a), the concentration variation at three distinct times t1 , t2 , and t3 is indicated. The diame-
ter of the critical droplet, whose cross-section is shown in case (b), is denoted by 2 R*, and the width of the inter-
facial region by x . Note that the concentration profile of the droplet reaches the other branch of the coexistence
(2) (1)
curve c coex in the droplet center only for weak “supersaturations” of the mixture, where cB – c coex O cBs – cB and
R* o x ; for the sake of clarity, the figure is therefore not drawn to scale (Binder, 1981).
stricts the quantitative validity of the theo- chapter, and the spontaneous growth of
retical concepts developed in the present thermal fluctuations dominating phase
chapter (see also Binder, 2000). This chap- changes of unstable but nearly ideal (i.e.,
ter will exclusively consider the mecha- defect-free) systems will be emphasized.
nism of Fig. 6-1 a, where in a thermody- Although most of the discussion refers to
namically unstable initial state long-wave- binary mixtures, an extension to ternary
length delocalized small-amplitude statis- mixtures is often possible.
tical fluctuations grow spontaneously in Unmixing of binary or ternary mixtures
amplitude as the time after the quench in- is not the only phase change where sponta-
creases. For a binary mixture, this mecha- neous growth of statistical fluctuations oc-
nism is called spinodal decomposition curs. Consider, for example, an alloy A – B,
(Cahn, 1961, 1965, 1966, 1968); the rele- which in thermal equilibrium undergoes an
vant fluctuations can then be considered as order – disorder phase transition form a dis-
a wavepacket of “concentration waves”, ordered state, where the two species of at-
and one such wave is shown in Fig. 6-1 a. oms A and B are distributed at random over
As will be discussed below, this mecha- the available lattice sites, to an ordered ar-
nism should not occur inside the whole rangement. There A and B atoms preferen-
two-phase coexistence region of the phase tially occupy sites on sublattices (e.g., b-
diagram of the mixture, but rather only in- CuZn, Cu3Au, CuAu, Fe3Al, and FeAl). If
side a smaller region, the boundary of we quench such an alloy from the disor-
which is given by the spinodal curve (Cahn dered regime to a state which in equilib-
and Hilliard, 1958, 1959). Between the spi- rium should be ordered, the unstable disor-
nodal curve cBs and the coexistence curve dered initial state may also decay by spon-
(1)
c coex (or in between the other branch of the taneous growth of fluctuations. The dis-
(2)
spinodal and c coex the mechanism of drop- tinction from Fig. 6-1 a is that the wave-
let nucleation and growth is involved (Fig. length of the growing concentration wave
6-1 a). This latter mechanism is discussed is not large but coincides with the lattice
in the chapter by Wagner et al., 2001). spacing of the superstructure. Neverthe-
Apart from these phase transformation less, the theoretical treatment of this spino-
mechanisms triggered by spontaneous ther- dal ordering (De Fontaine, 1979) is similar
mal fluctuations, heterogeneous nucleation to the theory of spinodal decomposition,
processes must also be considered (Zettle- and will also be discussed in this chapter,
moyer, 1969); near inhomogeneities in which emphasizes the theoretical aspects
solids such as grain boundaries, disloca- (see Wagner and Kampmann (2000) for a
tions, external surfaces, or point defects complementary treatment emphasizing the
such as substitutional or interstitial impur- point of view of the experimentalist).
ities or clusters thereof, microdomains of One of the basic questions in the applica-
the new phase may already be formed be- tions of materials is to understand the pre-
fore the quench, or at least their formation cipitated microstructure that forms in the
after the quench is greatly facilitated. Gen- late stages of a phase separation process.
eral theoretical statements about these One of the main characteristics of this mi-
heterogeneous mechanisms, however, are crostructure concerns its morphology: the
hardly possible without a detailed discus- minority phase may either form compact
sion of the nature of such defects. Such islands well separated from each other or it
problems will not be discussed in this may form an irregular interconnected net-
Figure 6-2. Snapshot pic-

tures of a two-dimensional
system undergoing phase
separation with a volume
fraction of (a) F = 0.21 and
(b) F = 0.5. The pictures
were obtained by numerical
solution of the non-linear
Cahn – Hilliard equation, Eq.
(6-14) (Rogers and Desai,
1989).
work-like structure (Fig. 6-2). Obviously, a 6.2 General Concepts

critical volume fraction exists where this
interconnected net first appears, a so-called 6.2.1 Phenomenological Thermodynamics
percolation threshold (Stauffer, 1985). Of- of Binary Mixtures and the Basic Ideas
ten it is assumed that the morphology of of Phase Separation Kinetics
well separated islands must have formed We first consider the dynamics of phase
by nucleation and growth, while the perco- transformations induced by an instantane-
lating morphology is taken as the signature ous quench from the initial temperature T0
of the spinodal curve. However, such an to a final temperature T (the idealization of
identification is misleading, as will be dis- this infinitely rapid cooling will be relaxed
cussed in detail in this chapter: the spinodal in Sec. 6.2.7). The simplest case where
curve at best has a meaning for the initial such a sudden change of external parame-
homogeneous state and controls the initial ters causes a phase transition is a system
stages of phase separation; in the late undergoing an order – disorder phase tran-
stages, however, a generally universal sition at a critical temperature Tc with
coarsening behavior occurs (which is sen- T < Tc < T0 (Fig. 6-3 a). The initially disor-
sitively affected by defects!), irrespective dered system is then unstable, and immedi-
of whether the morphology is intercon- ately, small ordered domains form. As the
nected or not. Since a closely related uni- time t after the quench passes, the order pa-
versal coarsening behavior also occurs in rameter in these domains must reach its equi-
fluid mixtures and can be studied fairly librium value (± y), and the domain size
well in these systems, we include a discus- must ultimately grow to macroscopic size.
sion of phase separation in fluid mixtures Alternatively, we consider a solid or liq-
throughout this chapter. Moreover, “poly- uid mixture A – B with a miscibility gap
mer alloys” can be conveniently produced (Fig. 6-3 b). Quenching the system at time
by quenching from the fluid phase of poly- t = 0 from an equilibrium state in the single
mer mixtures, and have become a practi- phase region to a state underneath the co-
cally important class of materials (Hashi- existence curve leads to phase separation;
moto, 1993). in thermal equilibrium, macroscopic re-
gions of both phases with concentrations
(1) (2)
c coex and c coex coexist.
6.2 General Concepts 417
Figure 6-3. (a) Order parameter y of a second-order phase transition vs. temperature, assuming a two-fold
degeneracy of the ordered state (described by the plus and minus signs of the order parameter). The quenching
experiment is indicated. (b) Phase diagram of a binary mixture with a miscibility gap ending in a critical point
(Tc , cBcrit ) of unmixing, in the temperature – concentration plane. Again the quenching experiment is indicated,
and the quenching distances from the coexistence curve (dT ) and from the critical point (DT ) are indicated.
(c) Free energy plotted versus composition at temperature T (schematic). For further explanations, see text
(Binder, 1981).
Of course, phase transformations com- rium, we have a linear variation of F with

bining an ordering process (such as in Fig. composition; this reflects the fact that the
6-3 a) with an unmixing process (such as in amounts of the two coexisting phases (with
(1) (2)
Fig. 6-3 b) also occur; as we are interested concentrations c coex , c coex ) change linearly
mainly in the basic concepts of these pro- with cB according to the lever rule.
cesses, we will not consider these more Mean-field type theories (Gunton et al.,
complex phenomena here, but will return 1983; Binder, 1987 a) suggest that a free
to this problem in Sec. 6.4.1. energy F¢ Ù F can also be introduced de-
The basic concepts of the kinetic mech- scribing single-phase states within the two-
anisms of the processes indicated in Fig. phase region. Since F¢ must coincide with
(1) (2)
6-3 a and b invoke the idea of extending the F for c coex and c coex , inevitably a double-
concept of the thermodynamic Helmholtz well structure results, as indicated by the
free energy F to states out of equilibrium dash-dotted curve in Fig. 6-3 c. This hypo-
(Gunton et al., 1983; Binder, 1987 a). In thetical free energy F¢ now allows a further
the single-phase region of the A-rich alloy, distinction to be made: states where
F first decreases with increasing concen- (∂2 F¢/∂c B2 )T > 0 are called metastable, and
tration of B atoms cB , owing to the entropy states where (∂2 F¢/∂c B2 )T < 0 are called un-
of mixing. Similarly, in the B-rich single- stable. The locus of inflection points in the
phase region, F decreases with increasing (T, cB ) plane,
concentration of A atoms, cA = 1 – cB . In the
two-phase coexistence region in equilib- (∂2 F¢/∂c B2 )T = 0 (6-2)
defines the spinodal curve c = cBs (T ). This (Binder et al., 1978). As will be discussed
distinction is now linked to the two trans- in more detail in Sec. 6.2.5, the spinodal
formation mechanisms shown in Fig. 6-1; curve cannot be unambiguously defined.
it is the unstable regime inside the two At this point, we recall that the “weak delo-
branches of the spinodal curve, where calized long-wavelength fluctuations” can-
long-wavelength fluctuations spontane- not be identified in terms of the atomic
ously grow rather than decay. At a later concentration variable ci , which undergoes
stage, the inhomogeneous concentration rapid large-amplitude (ci = 0 to ci = 1!) vari-
distribution thus generated will coarsen. In ations from one lattice site to the next, but
the metastable regime, the system is stable implies the introduction of the coarse-
against such weak (small-amplitude) fluc- grained concentration field c (r), Eq. (6-1).
tuations, and localized large-amplitude The free energy F¢ of “homogeneous”
fluctuations (“droplets” of the new phase) states in the two-phase region depends on
must form in order to start the transforma- the length scale L over which short-wave-
tion. length concentration fluctuations have
This idea implies, as will be explained been integrated (Fig. 6-4). This “coarse-
below, a singular transition in the kinetic grained” free-energy density fcg is not pre-
transformation mechanism at the spinodal cisely identical to the true free-energy den-
curve. However, this spinodal singularity is sity f in the single-phase region, since con-
really only an artefact of an over-simplified centration variations with wavelengths ex-
theoretical picture: apart from the very ceeding L contribute to f but are excluded
special limit of infinitely weak, infinitely from fcg . However, this difference is minor,
long-ranged forces in which mean-field and in the single-phase region we may con-
theory becomes correct (Penrose and Lebo- sider the limit L Æ ∞ and then fcg tends to-
witz, 1971; Lebowitz and Penrose, 1966; wards f uniformly. This is not possible in
Binder, 1984 b), the transition from the nu- the two-phase region, however, where fcg
cleation mechanism to the spinodal decom- describes homogeneous states only if
position mechanism is completely gradual LÁ x , the interfacial width; if L Ô x then
Figure 6-4. Schematic plot of

the coarse-grained free-energy
density fcg (f) as a function of the
order parameter f = (c – ccrit )/ccrit
in a first-order transition from
f 1coex to f 2coex , for a “symmetric”
situation with f (f 1coex ) = f (f 2coex )
as in the Ising model. Spinodals
f s(1, 2) (L) defined from inflection
points of fcg (f) depend distinctly
on the coarse-graining length L
and the interfacial width x (Bin-
der, 1987 a).
the states that yield dominating contribu- The term –12 r2 [—c]2 in Eq. (6-4) accounts
tions to fcg are phase separated on a local for the free energy cost of inhomogeneous
scale, and therefore fcg tends smoothly to- concentration distributions. Here the
wards the double-tangent construction as coarse-grained free-energy density fcg (c)
L Æ ∞ . Hence there is no unique theoreti- resulting from Eqs. (6-3) and (6-1) by car-
cal method of calculating a spinodal curve; rying out the restricted trace over the {ci}
remember that mean-field theories, which for a fixed concentration field {c (x)} is
yield spinodal curves easily, are only inac- very difficult to obtain in practice; qualita-
curate descriptions of real systems. Simi- tively the behavior of fcg (c) should be sim-
larly, extrapolation procedures by which ilar to the mean-field (MF) result for the
spinodal curves are extracted from experi- free energy of a binary mixture,
mental data also involve related ambigui-
1
ties, as will discussed in Sec. 6.2.5. fMF ( c ) = c ln c + (1 − c ) ln (1 − c )
Hence the coarse-graining alluded to kB T
T MF
above means that a microscopic Hamil- + 2 c c (1 − c ) (6-6)
tonian Ᏼ {ci} of the binary mixture is re- T
placed by a so-called free-energy function with
DᏲ {c (x)}. For example, the Hamiltonian
may correspond to a standard Ising-type kB TcMF ≡ ∑ Dj ( xi − x j ) (6-7)
j
pairwise interaction model (Binder, 1986;
De Fontaine, 1979): Near TcMF we can replace fMF (c) by its Lan-
dau expansion,
Ᏼ {ci } = 1
2 ∑ {j BB ( xi − x j ) ci c j (6-3)
i≠ j
fMF ( c ) = f0 + A ( c − ccrit )2
+ j AB ( xi − x j ) [ci (1 − c j ) + c j (1 − ci )]
+ j AA ( xi − x j ) (1 − ci ) (1 − c j )} + B ( c − ccrit ) 4 + … (6-8)
where j is the interaction energy between a where A ~ (T/TcMF – 1) < 0 for T < TcMF and
pair of atoms. B > 0. It is well known that the actual
Carrying out the coarse-graining defined critical temperature Tc does not coincide
in Eq. (6-1) we expect that Ᏼ {ci}/kBT will with the mean-field prediction. Therefore,
be replaced by it is usually assumed that the parameters
appearing in the actual coarse-grained
DᏲ {c ( x )} free energy fcg (c) are not the mean-field
(6-4)
kB T parameters A, B, … , but rather these pa-
= ∫ d d x { fcg [c ( x )]/ kB T + 12 r 2 [∇c ( x )]2 } rameters are “renormalized” due to short-
wavelength fluctuations, and also r (Eq.
where d is the spatial dimensionality of the (6-5)) may thus be modified. Therefore,
system and r is the range of the effective fcg (c) and r are not calculated from micro-
interaction in Eq. (6-3) (Dj (xi – xj) ∫ j AA scopic models such as Eq. (6-3), but are
+ j BB – 2 j AB): treated as phenomenological input parame-
ters of the theory, which are fitted to ex-
∑ ( xi − x j )2 Dj ( xi − x j ) periment.
j
r2 = (6-5)
2 d ∑ Dj ( xi − x j )
j
6.2.2 The Cahn – Hilliard – Cook and using this result in the continuity rela-
Nonlinear Diffusion Equation tion, Eq. (6-10), we obtain the Cahn – Hil-
liard nonlinear diffusion equation (Cahn,
Since in the total volume V the average
1961):
concentration,
c– = (1/V ) Ú dx c (x, t) ∂c ( x, t )
(6-9) = M ∇2 (6-14)
∂t
is conserved, the time-depent concentra-
⎧⎪⎛ ∂ fcg [c ( x, t )] ⎞ ⎫⎪
tion field c (x, t) satisfies a continuity equa- × ⎨⎜ ⎟ − r kB T ∇ c ( x , t ) ⎬
2 2
tion, ⎪⎩⎝ ∂c ⎠T ⎪⎭
∂c ( x, t ) One immediately obvious defect of this
+ ∇ ⋅ j ( x, t ) = 0 (6-10)
∂t equation is its completely deterministic
where j (x, t) is the concentration current character, which implies that random sta-
density. Following standard nonequilib- tistical fluctuations are disregarded (apart
rium thermodynamics (de Groot and Ma- from fluctuations included in the initial
zur, 1962), j (x, t) is assumed to be propor- condition, the state at temperature T0 where
tional to the gradient of the local chemical the quench starts). This defect can be rem-
potential difference m (x, t): edied, following Cook (1970), by adding a
random force term h T (x, t) to Eq. (6-14):
j (x, t) = – M —m (x, t) (6-11)
∂c ( x, t )
where M is a mobility that is discussed be- = M ∇2 (6-15)
low. ∂t
In the thermal equilibrium the chemical ⎧⎪⎛ ∂ fcg [c ( x, t )] ⎞ ⎫⎪
× ⎨⎜ ⎟ − r kB T ∇ c ( x , t ) ⎬
2 2
potential difference is given as a partial de-
⎩⎪⎝ ∂c ⎠T ⎭⎪
rivative of the Helmholtz energy F (c, T ):
+ h T ( x, t )
m = (∂F/∂c)T ; (6-12)
Here h T (x, t) is assumed to be delta-corre-
remember that the condition for two-phase lated Gaussian noise, and the mean-square
coexistence, the equality of chemical po- amplitude ·h 2T ÒT is then linked to the mobil-
tential differences ity M via a fluctuation-dissipation relation.
m1 = ( ∂F / ∂c )T | ccoex
( 1 ) = m 2 = ( ∂F / ∂c ) T | ( 2 )
ccoex ·h T (x, t) h T (x¢, t¢)ÒT
is the physical content of the double-tan- = ·h T2 ÒT —2 d(x – x¢) d(t – t¢) (6-16)
gent construction shown in Fig. 6-3 c. Eq. ·h T2 ÒT = 2 kB T M (6-17)
(6-12) is generalized to an inhomogeneous
nonstationary situation far from equilib- Eqs. (6-15) – (6-17) constitute the main re-
rium, where both c (x, t) and m (x, t) depend sults of this section, on which all further
on space and time, by defining m (x, t) as treatment presented here is based. At this
a functional derivative of the Helmholtz point we stress the main assumptions that
energy functional DᏲ in Eq. (6-4): have been made either explicitly or tacitly:
m (x, t) ∫ d (DᏲ {c (x, t)})/dc (x, t) (6-13) (i) Effects due to the lattice anisotropy of
the solid have been ignored. In a crystalline
Inserting Eq. (6-14) into Eq. (6-13) yields
solid the interfacial free energy between
m (x, t) = (∂ fcg /∂c)T – r2 kB T —2c (x, t) co-existing A-rich and B-rich phases will
depend on the orientation of the interface; that a coarse-graining is performed over a

only in the limit T Æ Tc does this interfa- length scale much larger than the lattice
cial free energy become isotropic (Wortis, spacing, but less than x, which again re-
1985). The isotropic form of Eq. (6-15) stricts the region of validity of the theory to
outside the critical region would only hold the critical region. On the other hand, fluc-
for phase separation in isotropic amor- tuations and nonlinear phenomena are
phous solids. known to lead to a breakdown of meanfield
(ii) It is assumed that the local concen- theory near the critical point, and nontrivial
tration field is the only slowly relaxing var- critical phenomena arise (Stanley, 1971;
iable whose dynamics must be considered Binder, 2001). As we shall see (Secs. 6.2.5
explicitly after the quench, while all other and 6.2.6) this also restricts the applicabil-
variables equilibrate instantaneously. This ity of the theory.
assumption is not true, of course, if phase-
separating fluids are considered where the
6.2.3 Linearized Theory
coupling of hydrodynamic variables and
of Spinodal Decomposition
the resulting long-range hydrodynamic
interactions need to be taken into account It may seem unsatisfactory that the basic
(Kawasaki and Ohta, 1978). However, in equation of the theory (Eq. (6-15)) is sup-
solids other slow variables may also occur, posed to hold only under fairly restrictive
particularly near the glass transition (in conditions, but what is even worse is the
amorphous solids) or near other phase fact that this equation completely with-
transformations. A treatment of spinodal stands an analytical solution, and brute
decomposition in the presence of a cou- force numerical approaches where Eq. (6-
pling to slowly relaxing structural vari- 15) is attacked by large-scale computer
ables is only possible in very simplified simulation are needed (Meakin et al., 1983;
cases (Binder et al., 1986). Petschek and Metiu, 1985; Milchev et al.,
(iii) It is assumed that the spatial concen- 1988; Gawlinski et al., 1989; Rogers et al.,
tration variations of interest are small, i.e., 1988; Toral et al., 1988; Oono and Puri,
in Eq. (6-4) terms of the order [—2c (x)]2 1988; Puri and Oono, 1988). We shall re-
(and higher) can be neglected in compari- turn to these approaches in Sec. 6.2.6, and
son with the lowest-order gradient term, also address the complementary approach
[—c (x)]2. This assumption is true near the of Monte Carlo simulations based directly
critical point of unmixing, where the inter- on the microscopic Hamiltonian, Eq. (6-3)
facial width x is much larger than the inter- (Bortz et al., 1974; Marro et al., 1975,
atomic spacing, but it is not true in the gen- 1979; Rao et al., 1976; Sur et al., 1977;
eral. Binder et al., 1979; Lebowitz et al., 1982;
(iv) The concentration dependence of the Fratzl et al., 1983; Heermann, 1984 a, b,
mobility M is neglected in the derivation 1985; Amar et al., 1988). Here we prefer to
yielding Eqs. (6-14) and (6-15). Again, this discuss the assumption that in the initial
approximation is valid near Tc , since then stages of unmixing the fluctuation
the relevant scale for concentration varia-
(2) (1) dc (x, t) ∫ c (x, t) – c–
tions becomes small ((ccoex – ccoex )/ccrit O 1),
but is not true in general. is small everywhere in the system. Since L
(v) The use of a continuum description cannot be made arbitrarily large, this as-
for a solid (Eqs. (6-1) and (6-4)) requires sumption is not typically true, as we shall
see below (Secs. 6.2.5 and 6.2.6), but nev- periment (Fig. 6-5 a, b) nor in simulations
ertheless it is instructive to study it! (Figs. 6-5 c, d) is such a time-independent
Given this assumption, we may linearize intersection point observed. Also in the
Eq. (6-15) (or Eq. (6-14) if we also neglect Cahn plot, R (k)/k2 plotted versus k2, in-
the thermal noise) in dc (x, t). Then Eq. (6- stead of the predicted linear behavior
14) becomes (Fig. 6-6 a), 2 R (k)/k2 = 2 D0 (1 – k2/kc2) with
a negative diffusion constant
∂
d c ( x, t ) = M ∇ 2 (6-18)
∂t D0 = – M(∂2 fcg /∂c2 )T, c– (uphill diffusion)
⎧⎛ ∂ 2 f ( c ) ⎞ ⎫
⎪ cg 2⎪ curvature is typically observed (Fig. 6-6 b).
× ⎨⎜ ⎟ − r 2
k B T ∇ ⎬ d c ( x, t )
⎪⎩⎝ ∂c
2
⎠ T, c ⎪⎭ There are several possible reasons why
the simple linearized theory of spinodal de-
and, introducing Fourier transformations composition as it has been outlined so far is
dck (t) ∫ ∫ dd x exp (i k · x) dc (x, t) (6-19) invalid:
(i) Fluctuations in the final state at the
Eq. (6-18) is solved by a simple exponen-
temperature T must be included (Cook,
tial relaxation,
1970). This problem will be considered at
dck (t) ∫ dck (0) exp [R (k) t] (6-20) the end of this subsection.
(ii) Nonlinear effects are important dur-
with the rate factor R (k):
(6-21) ing the early stages of the quench. This
R (k) ∫ M k2 [(∂2 fcg /∂c2 )T, c– + r 2 kB T k2 ] problem will be discussed in Secs. 6.2.5
and 6.2.6.
The equal-time structure factor S (k, t) at (iii) There is already an appreciable re-
time t after the quench: laxation of the structure factor occurring
S (k, t) ∫ ·dc– k (t) dck (t)ÒT (6-22) during the quench from T0 to T if the cool-
ing rate g = – dT/dt is finite (see Sec. 6.2.7).
where ·…ÒT denotes a thermal average, (iv) The concentration field is coupled to
then also exhibits a simple exponential re- another slowly relaxing variable (Binder et
laxation: al., 1986; Jäckle and Pieroth, 1988). Here
S (k, t) ∫ ST0 (k) exp [2 R (k) t] (6-23) we only very briefly outline the idea of the
approach of Binder et al. (1986). Accord-
Here the prefactor ing to Eqs. (6-23) and (6-21), the maximum
(6-24)
growth rate of the structure factor occurs at
ST0 (k) ∫ ·dc– k (0) dck (0)ÒT ∫ ·dc– k dck ÒT0
–
Rm = R (km), km = kc / ÷2 (6-26)
is simply the equal-time structure factor in
thermal equilibrium at temperature T0 be- Suppose now the concentration couples
fore the quench. Note that R (k) is positive linearly to a non-conserved variable a (t)
for 0 < k < kc , with describing, for example, structural relaxa-
(6-25)
tion, whose fluctuations in the absence of
kc ∫ 2 p/lc = [– (∂ fcg /∂c )T, c– /(r kB T )]1/2
2 2 2
any coupling would decay exponentially,
Thus, whereas the structure factor should proportional to exp (– g t). The decay of
exhibit exponential growth in this region, concentration fluctuations will be affected
for k = kc it should be time-independent, (i) if the coupling between the variables
S (kc , t) = S (kc , 0). However, neither in ex- a (t) and c (x, t) is sufficiently strong, and
Figure 6-5. (a) Neutron small-angle scattering intensity vs. scattering vector k (b : scattering angle) for an
Au – 60 at.% Pt alloy quenched to 550 °C (Singhal et al., 1978). (b) Time evolution of the structure factor
at 541 K in Al – 38 at.% Zn after subtraction of the prequench scattering. Solid lines are the best fit to the data
using the LBM theory, Eq. (6-60) (Mainville et al., 1997). (c) Time evolution of the structure factor S (k, t) ac-
cording to a Monte Carlo simulation of a three-dimensional nearest-neighbor Ising model of an alloy at critical
concentration and temperature T = 0.6 Tc . Due to the periodic boundary condition for the 30 ¥ 30 ¥ 30 lattice, k is
only defined for discrete multiples of (2 p)/30; these discrete values of S (k, t) are connected by straight lines
(Marro et al., 1975). (d) Time evolution of the normalized structure factor S̃ (k, t) vs. k for the discrete version
of the Ginzburg – Landau model (Eq. (6-4)), namely
Ᏼ / kB T = ∑ [ A ( cl − ccrit ) 2 + B ( cl − ccrit ) 4 ] + ∑ 1
2 C ( cl − c m ) 2
l 〈l m〉
where cl is a continuous variable representing the average concentration in the l th cell of size L ¥ L, and the
-
phenomenological constants A, B, and C have been chosen as A/C = – 2.292, ÷B/C = 0.972 (C can be scaled out
by redefining the cl s). Data are for an N = 40 ¥ 40 lattice with periodic boundary conditions at times t = 0, 10,
20, …, 90 Monte Carlo steps (MCS) per site. The arrow indicates the estimate for the wavenumber of maximum
- ----
growth, km (0) = kc / ÷2 = ÷ – A/C (Milchev et al., 1988).
Figure 6-6. Schematic “Cahn plot”,

i.e., R (k)/k2 vs. k2, (a) as predicted by
the linear theory and (b) as it is typi-
cally observed. Ideally, the “Cahn
plot” should yield a straight line: the
intercept with the abscissa occurs at
k c2, and the intercept with the ordinate
at D0 (Binder et al., 1986).
(ii) if the rates Rm and g are of the same or in fluid polymer mixtures near their
order. In the absence of any coupling glass tansition (e.g., Meier and Strobl,
between these variables their fluctuations 1988). The slow variables are then ex-
would decay independently of each other, pected to relax with a broad spectrum of
with decay rates being given as G+ (k) = g rates rather than with a single rate g. Fluc-
(relaxation of a (t) is then independent of k)
and G– (k) = – |D0 | k2 (1 – k2/kc2) (spinodal
decomposition). This mode spectrum of
uncoupled structural and concentration
fluctuations is shown by the dashed curves
in Fig. 6-7). However, both relaxation rates
become strongly modified if these vari-
ables are coupled. The strength of this
coupling can be related to 1 – D∞ /D0 , where
D0 and D∞ are the low- and high-frequency
limits of the diffusion constant. Full curves
in Fig. 6-7 show for three (arbitrary)
choices of parameters, the relaxation rates
G+ (k) and G– (k) for the case of coupled
variables. It is seen that a “mixing” of
interdiffusion and relaxation occurs; for
small k the interdiffusion (or spinodal de-
composition) is given by G– (k), for large k
by G+ (k), and for intermediate k both expo-
nentials exp [– G+ (k) t)] and exp [– G– (k) t)]
contribute to the growth and/or decay of
concentration fluctuations. Plotting the
mode G– (k) which describes spinodal de-
composition for small k in the form of a
Cahn plot, we encounter pronounced cur- Figure 6-7. Mode spectrum {G+ (k), G– (k)} (full
vature. curves) of an unmixing system coupled to a slow var-
Such a coupling where Rm and g are of iable plotted vs. k2/k c2 for three parameter choices. In
the absence of any coupling, the slow variable would
comparable size might occur if we study
relax with a rate G+ (k) = g (broken horizontal straight
spinodal decomposition in glasses (see for lines) and the unmixing system would relax with a
example Yokota, 1978; Acuña and Craie- rate G– (k) = 4 Rm (k2/k c2 ) (1 – k2/k c2 ) (broken curves).
vich, 1979; Craievich and Olivieri, 1981) All rates are normalized by Rm (Binder et al., 1986).
tuations in the final state also need to be in- now yields

cluded in this case (Jäckle and Pieroth, (6-29)
⎛ k2 ⎞
1988). Deff ( k , t ) ≡ − D0 ⎜ 1 − 2 ⎟ + kB T M / S ( k , t )
We now discuss the effect of fluctuations ⎝ kc ⎠
in the final state for the simplest case, Eq. Since in the linear theory S (k, t Æ ∞) Æ ∞,
(6-15), treated in the framework of the lin- Deff (k, t Æ ∞) reduces to the simple Cahn
earization approximation. After some sim- (1961) result quoted above, but this limit is
ple algebra we obtain from Eqs. (6-15) to never reached owing to nonlinear effects.
(6-22) the following equation (Cook, On the other hand, at t = 0, Eq. (6-29) leads
1970): to a linear relation between Deff (k, 0) and
d k2 again, because (6-30)
S (k, t ) = − 2 M k 2 (6-27)
dt ⎡ ∂2 f ⎤
[ S ( k , 0 )]−1 = ⎢ 2 + r 2 kB T k 2 ⎥ kB T0
cg
⎧⎡ ∂2 f ⎤ ⎫
⎪ ⎪ ⎢ ∂c ⎥
× ⎨⎢ 2 + r 2 kB T k 2 ⎥ S ( k , t ) − kB T ⎬
cg
⎣ T0 , c ⎦
⎪⎩ ⎢⎣ ∂c ⎥ ⎪⎭
T, c ⎦
is linear in k2 also. Here the point where
It is seen that an effective diffusion con- Deff (k, 0) changes sign is not kc , but shifts
stant for uphill diffusion, defined as to a larger value. This behavior is illus-
trated in the upper part of Fig. 6-8 where
1 d the (rescaled) structure factor S̃ (q, t) and
Deff ( k , t ) ≡ ln S ( k , t ) (6-28)
2 k 2 dt (rescaled) effective diffusion constant
Figure 6-8. Scaled structure

function (left) and normalized
diffusion constant D̃eff (q, t ) ∫
q2 d [ln S̃ (q, t )]/dt (right) vs. q
or q2, respectively, for an instan-
taneous quench from infinite
temperature to T /Tc = 4/9. Top,
Cahn – Hilliard – Cook (CHC)
approximation, Eqs. (6-27) and
(6-29), bottom based on the
Langer – Baron – Miller (LBM)
(1975) theory. Ten times t = 1,
2, …, 10 are shown after the
quench. Note that in the CHC
approximation we use the
renormalized value m – (Fig.
6-18) of (∂2 fcg /∂c 2 )T, c– instead
(∂2 fcg /∂c 2 )T, c– , m – (T, t = 0) = 0.65
in the present units (Carmesin
et al., 1986).
D̃eff (q, t) following from Eqs. (6-27) and distance and gyration radius of these coils
(6-29) are plotted against the (rescaled) then is (Flory, 1953; De Gennes, 1979):
square of the scattering vector, q2 (where
q = k/kc , t = 2 M r 2 T kc4 t, S̃ = r2 kc2 S). It is ·RA2 Ò = sA2 NA , ·RB2 Ò = s B2 NB , (6-31)
2
seen that on the level of the Cahn – Hilli- ·Rgyr, AÒ = –16 sA2 NA , 2
·Rgyr, BÒ = –16 s B2 NB
ard – Cook (CHC) approximation there is
initially some shift in the position at which In the Flory – Huggins (FH) approximation
S (k, t) has its maximum, although this shift (Flory, 1953; Koningsveld et al., 1987) the
is more pronounced when nonlinear effects expression corresponding to Eq. (6-4) be-
are taken into account, as is approximated comes (De Gennes, 1980)
by the theory of Langer et al. (1975), which DᏲ {F ( x )}
we shall refer to as the Langer – Baron – kB T
{
= ∫ d d x fFH [F ( x )]/ kB T (6-32)
Miller (LBM) approximation. It is seen that 1 1 ⎫
the main distinction between the CHC and + a2 [∇F ( x )]2 ⎬
2 F (1 − F ) ⎭
LBM approximations during the early stages
is the lack of a common intersection point in where F (x) is the volume fraction of A seg-
the LBM approximation. Also the growth of ments, 1 – F (x) is the volume fraction of B
S̃(q, t ) with t is generally slower (note the segments, the mixture is assumed to be
difference in the ordinate scales!). Qualita- incompressible, and the lattice spacing a of
tively, however, the behavior is similar, and the Flory – Huggins lattice model is given
this is also true for the behavior of D̃eff (q, t ). by a2/[F (1 – F )] ∫ sA2 /F + s B2 /(1 – F ). Note
Note that D̃eff (q, t) is distinctly curved in that unlike Eq. (6-4) the parameter a is not
both cases (apart from the limit t Æ 0). related to the range of interaction, but
rather reflects the random coil structure of
the polymer chains. Owing to the connec-
6.2.4 Spinodal Decomposition
tivity of the polymer chains, the entropy of
of Polymer Mixtures
mixing term is much smaller than in Eq. (6-
Here we consider the modifications nec- 6) (Flory, 1593):
essary for describing binary mixtures of
long flexible macromolecules. In a dense 1 F ln F
fFH (F ) = (6-33)
polymer melt, the configurations of these kB T NA
linear chain molecules (we disregard here (1 − F ) ln (1 − F )
+ + c F (1 − F )
star polymers, branched polymers, copol- NB
ymers, and also interconnected networks,
where the Flory – Huggins parameter c con-
although some aspects of the theory can be
tains all enthalpic contributions that lead to
extended to these more complex situations,
unmixing. If c does not depend on volume
see Binder and Frisch (1984)) are random
fraction, then the spinodal curve resulting
Gaussian coils interpenetrating each other.
from Eq. (6-33) is given by
In terms of subunits (“Kuhn segments”) of
lengths sA and sB for the two types of 2 cs (F ) = (F NA)– 1 + [(1 – F ) NB]– 1 (6-34)
polymers A and B, the degree of polymer-
ization XA and XB is expressed in terms of and the critical point occurs at
chain lengths NA and NB as XA = NA gA and
XB = NB gB, if each subunit contains gA and Fc = ( NA / NB + 1) −1
gB monomers. The mean square end-to-end c c = ( NA−1/ 2 + NB−1/ 2 )2 / 2
These results assume well-defined chain (6-6)) only in the long wavelength limit,
2 1/2
lengths (“monodisperse polymers”), but where kc ·Rgyr Ò < 1 (for both types of
can be generalized to a distribution chains). In this limit L k can be replaced
of chain lengths (“polydispersity”; see by a constant L 0 , and the results are just
Joanny, 1978). special cases of Eqs. (6-9) – (6-30); e.g., for
Another modification concerns the con- a symmetric mixture (NA = NB , sA = sB = a)
stitutive kinetic equation: instead of Eq. (6- we simply find:
11) the connectivity of the chains requires 1/ 2
a non-local relationship for the current den- ⎛ 18 ⎞
2 p / l c = kc = ⎜ ⎟ (1 − c s (F )/ c )1/ 2 ,
sity JF (x, t) relating to volume fraction F ⎝ N a2 ⎠
(De Gennes, 1980; Pincus, 1981), k m = kc / 2 (6-39)
JF (x, t) = – Ú L (x – x¢) —m (x¢, t) dx¢ (6-35) Hence the critical wavelength lc is rescaled
where L (x – x¢) is a generalized Onsager by a large prefactor, namely the coil gyra-
coefficient describing polymer – polymer tion radius, similar to the correlation length
interdiffusion. The rate R (k) in Eq. (6-21) xcoex of concentration fluctuations at the
is then expressed in terms of the Fourier coexistence curve,
transform L k of L (x – x¢), i.e., x coex = (N a2/36)1/2 (c /cc – 1)–1/2 (6-40)
2
R (k) = k L k [STcoll (k)]– 1 (6-36) Eq. (6-39), however, is only applicable for
where the expression for the effective col- “shallow” quenches, where c does not
lective structure factor STcoll (k) can be ex- greatly exceed c s (F). A different behavior
pressed in terms of the structure factors occurs (Binder, 1983) for deep quenches,
SA (k) and SB (k) of the single chains as where c o c s (F) (thus describing the un-
(Binder, 1983, 1984 c, 1987 b) mixing of two incompatible polymers):
1/ 2
[STcoll (k)]– 1 = [F SA (k)]– 1 ⎛ 6 ⎞
kc ≈ ⎜ ⎟ [2 c / c s (Φ )]1/ 2
+ [(1 – F ) SB (k)] – 2 c–1 eff
(F, T, k) (6-37) ⎝ N a2 ⎠
1/ 2
eff
where 2 c (F , T, k) is a wavevector- ⎛ 6 ⎞
km ≈ ⎜ ⎟ [2 c / c s (Φ )]1/ 4 (6-41)
dependent generalization of the term re- ⎝ N a2 ⎠
sulting from the second derivative of
Now the initial wavelength of maximum
∂2 [c F (1 – F )]/∂f2 in Eq. (6-33). The sin-
growth is smaller than the coil radius,
gle-chain structure functions SA (k) and
and km O kc , since the gradient-square ex-
SB (k) are expressed for Gaussian chains by
pansion, Eq. (6-32), is no longer appli-
the Debye function fD (x) as
cable.
SA (k) = NA fD (k2 ·Rgyr,
2
A Ò) As is obvious from Eq. (6-36), the relax-
SB (k) = NB fD (k2 ·Rgyr,
2 ation rate R (k) is the product of two fac-
BÒ) (6-38)
tors, a static factor [STcoll (k)]– 1 which con-
where fD (x) ∫ 2 [1 – (1 – e– x )/x]/x. tains the thermodynamic singularities at
Eqs. (6-36) and (6-37) result (Binder, the critical point (and along the spinodal
1983) from random phase approximations curve; cf Eq. (6-39)), and a kinetic factor
(De Gennes, 1979) and agree with the k2 L k, to which we now turn. Here the fac-
treatment as given by Eqs. (6-9) – (6-30) tor k2 simply reflects the conservation law
(but using Eq. (6-32) instead of Eqs. (6-4) – for the concentration, while L k reflects the
complicated dynamics of single polymer If DA B

t and Dt are of the same order, then
chains in an entangled polymer melt. De- Eqs. (6-42) and (6-43) also yield Dcoll of
spite some effort (De Gennes, 1980; Pin- the same order, and the order of magnitude
cus, 1981; Binder, 1983; Akcasu et al., of Dcoll (and hence R (k), Eq. (6-36)) is then
1986) there is no systematic method that predicted correctly from these equations. If
yields information on the k-dependence of the chain length NA (NB) is less than the
L k; the approximate theories quoted above chain length NAe (NBe ) between effective
all lead to a long-wavelength limit called entanglements, the single-chain dynamics
the slow-mode theory, which is believed to are simply given by the Rouse (1953)
be an incorrect description of interdiffusion model, which implies
because it disregards bulk flow (Kramer
t ≈ sA WA /NA ,
DA DBt ≈ sB WB /NB
et al., 1984). Remember that Eq. (6-36) for
k Æ 0 can be interpreted as R (k) = k2 Dcoll, where WA (WA) are time constants for the
where Dcoll = L 0 /STcoll (0), i.e., uphill diffu- reorientation of subunits. In contrast, for
sion. In the slow-mode theory, the interdif- NA p NAe , NB p NBe the reptation model
fusion coefficient of an ideal non-interact- (Doi and Edwards, 1986; De Gennes,
ing mixture is expressed in terms of tracer 1979) implies for strongly entangled chains
diffusion coefficients DA B
t and Dt of an A
t ≈ sA WA NA /NA ,
DA DBt ≈ sB WB NBe /NB2
e 2
chain (B chain) in a pure B (A) environ-
ment as (neglecting prefactors of order unity). A
problem which is not completely under-
(Dcoll)– 1 = (1 – F ) /DA B
t + F /Dt (6-42)
stood so far is the concentration depen-
i.e., the slow species controls the interdif- dence of the parameters NAe and NBe in
fusion coefficients. In contrast, the fast- blends. In view of all these uncertainties
mode theory (Kramer et al., 1984) yields about the validity of Eqs. (6-42) and (6-43)
the opposite result, that the fast species and the precise values of the constants DA t
controls interdiffusion: and DBt to be used in them, it is better to
consider L k in Eq. (6-36) as a phenomeno-
Dcoll = (1 – F ) DA B
t + F Dt (6-43)
logical coefficient, about which only the
Experimental evidence (for a review of order of magnitude can roughly be in-
polymer interdiffusion, see Binder and Sil- ferred.
lescu, 1989) seems to favor Eq. (6-43) over An effect disregarded by the theories but
Eq. (6-42), but we feel that a valid deriva- seen in computer simulations (Sariban and
tion of Eq. (6-43) is still lacking, and Binder, 1989 a) is the change in the chain
moreover neither Eq. (6-43) nor (6-42) de- linear dimensions (Eq. (6-31)) in quench-
scribe interdiffusion in lattice gas models ing experiments. This also means that the
correctly, as computer simulations show parameter a in Eqs. (6-32) and (6-39) to
(Kehr et al., 1989). Both Eqs. (6-42) and (6-41) should be treated as a c-dependent
(6-43) are based on a description of trans- quantity, which further complicates the
port in which in the constitutive equation quantitative analysis of experiments.
relating currents and chemical potential
differences (cf. Eq. (6-11)) off-diagnonal
6.2.5 Significance of the Spinodal Curve
Onsager coefficients are neglected; the
Monte Carlo simulation shows that this ap- From Eq. (6-25) it is evident that the
proximation is inaccurate. linear theory predicts a singular behavior
when the concentration c– approaches the tion, which according to Eqs. (6-21), (6-25)
concentration csp (T ) of the spinodal curve, and (6-26) can be written as
defined by
Rm = (1/4) M r2 kB T (2 p /lc )4 (6-47)
2 2
(∂ fcg /∂c )T, c– = csp (T ) = 0 (6-44) vanishes as c– Æ c (T ), and a similar “criti-
sp
Here the critical wavelength lc diverges to cal slowing down” (Hohenberg and Halpe-
infinity. For example, if we adopt Eq. (6-8) rin, 1977) would occur in the growth rate of
with A ~ (T/TcMF – 1), near the spinodal we a supercritical droplet. Thus, within the
obtain, omitting prefactors of order unity, framework of this theory, the spinodal
−1/ 2
curve plays the role of a line of critical
⎡ c − csp ( T ) ⎤ points. We now wish to investigate whether
lc ~ r (1 − T / TcMF ) −1/ 2 ⎢ ⎥
⎣ ccrit − csp ( T ) ⎦
there is a physical signficance to this singu-
c Ê csp ( T ) (6-45) lar behavior.
In Sec. 6.2.1 we emphasized that the def-
On the metastable side of the spinodal inition of fcg is not unique but really in-
curve a similar divergence occurs for both volves a length scale L over which a coarse-
the correlation length x (c– ) of concentra- graining of short-range fluctuations is per-
tion fluctuations and the radius R* of a crit- formed. This is best seen from the attempts
ical droplet, namely (Binder, 1984 b) to calculate fcg explicitly, which can be
approximated by using renormalization
x (c ) ~ R* ~ r (1 − T / TcMF ) −1/ 2 (6-46) group methods (Kawasaki et al., 1981) or
−1/ 2
⎡ csp ( T ) − c ⎤ by Monte Carlo simulation (Kaski et al.,
× ⎢ (2) (1) ⎥ , c Á csp ( T ) 1984). These treatments show that the posi-
⎣ ccoex − ccoex ⎦
tion of csp (T ) depends strongly on the
Although R* diverges as c– approaches length scale L (see Fig. 6-10 a), and there-
csp (T ), the associated free-energy barrier fore for systems with short-range forces, to
DF* against the formation of a critical which these treatments apply, there is no
droplet vanished there (Fig. 6-9). The max- physical significance to the spinodal singu-
imum growth rate of spinodal decomposi- larity (Eqs. (6-44) to (6-47)) whatsoever.
Figure 6-9. (a) Character-

istic lengths R*, x , l c and
(b) nucleation barrier DF*
vs. concentration cB (sche-
matic). Full curves are the
predictions of the linearized
Cahn – Hilliard mean-field
theory of nucleation. Dash-
dotted curves show, on a
different scale, the conjec-
tured smooth behavior of a
system with extremely
short-range interaction, for
which the spinodal singu-
larity is completely washed
out (Binder, 1981).
Figure 6-10. (a) Monte Carlo results for the dependence of the relative distance of the “spinodal” from the co-
existence curve on the size of the coarse-graining cell L. The results refer to a simple-cubic nearest-neighbor
Ising magnet in the critical region, and are obtained from sampling the magnetization distribution function
PL (M) in L ¥ L ¥ L sub-blocks of a 243 system. Here, Smax is the value at which ln PL (M) has its maximum, and
corresponds to the coexistence curve if we assume ln PL (M) ~ L3 fcg /kB T, with c = (1 – M)/2, and the “spinodal”
is estimated as inflection point Ms of ln PL (M). By scaling L with the correlation length x , all temperatures
superimpose on one “scaling function” (Kaski et al., 1984). (b) Extrapolation of the inverse collective structure
function versus inverse temperature to locate the spinodal temperatures Tsp (c–) or their inverse (arrows). Here
Monte Carlo simulation data for a polymer mixture are used, the polymers (A, B) being modelled as self- and
mutually avoiding random walks on the simple cubic lattice with NA = NB = N = 32 steps, at a concentration of
vacancies fv = 0.6. Values on the curves are the reduced volume fraction FA /(1 – Fv ) of monomers of A chains.
If two neighboring lattice sites are taken by monomers of the same kind, an energy e is obtained, and thus an en-
thalpic driving force for phase separation is created. From Sariban and Binder (1989 b). (c) Phase diagram of the
model for a polymer mixture as described in (b), displaying both the true coexistence curve (binodal) and the
extrapolated spinodal. From Sariban and Binder (1989 b).
Another way of showing this comes from Heermann, 1984 a, b). Although very close
a closer examination of the procedures to the critical point Tc such system behave
by which experimentalists locate spinodal qualitatively like short-range systems (this
curves. Eqs. (6-30) and (6-37) imply that is to be expected from the so-called “uni-
versality principle” (Kadanoff, 1976)), far-
[S coll (k Æ 0)]– 1 ~ (∂2 f /∂c2) – ~ T – T (c– )
T cg T, c sp
ther from Tc a well-defined mean-field crit-
where Tsp (c– ) is the inverse function of ical region exists. It is this region where
csp (T ) in the c – T plane. Thus a plot of both the coarse-graining defined in Eq. (6-
[STcoll (k Æ 0)] vs. temperature should allow 1) and the linearization approximation of
a linear extrapolation to locate Tsp(c– ) from Sec. 6.2.2 make sense, as we will now dis-
the vanishing of [STcoll (k Æ 0)] (cf. Fig. 6- cuss.
10 b). However, the extrapolated spinodal The argument is simply an extension of
determined in this way (e.g., Sariban and the Ginzburg (1960) criterion for the valid-
Binder, 1989 b) (Fig. 6-10 c), is not physi- ity of the mean-field theory for critical
cally meaningful as it crosses the true co- phenomena to nucleation and spinodal de-
existence curve (binodal) near Tc , which is composition (Binder, 1984 b): nonlinear
physically impossible. This happens be- terms in dc (x, t) can be neglected if their
cause, in reality, mean-square amplitude in a coarse-grain-
ing cell is small in comparison with the
[STcoll (k Æ 0)]c––1= ccrit ~ (T – Tc )g
concentration difference squares in the
with g ≈ 1.24, a critical exponent different system, over which relevant nonlinear ef-
from the mean-field result g = 1. Hence fects are felt:
a linear extrapolation fails near c– = ccrit . At
·[dc (x, t)]2 Ò  [c– – c (T )]2
T, L sp (6-48)
strongly off-critical concentrations this ex-
trapolation is also ambiguous, because usu- We estimate ·[dc (x, t)]2 ÒT, L as
ally actual data cannot be taken deep in the
·[dc (x, t)]2 ÒT, L ≈ ·[dc (x, 0)]2 ÒT, L exp[2 Rm t]
metastable phase for temperatures close to
Tsp(c– ). where Rm is the maximum growth rate de-
This situations is different, however, for fined in Eq. (6-26), and the initial mean-
systems with infinitely weak but infinitely square amplitude is related to the correla-
long-range interactions (Penrose and Lebo- tion function of concentration fluctuations
witz, 1971); then the mean-field theory is in the initial state at temperature T0 . Thus,
correct because statistical fluctuations are using Eq. (6-1):
suppressed. At the same time, the lifetime
〈[d c ( x, 0 )]2 〉 T , L = 〈[d c ( x )]2 〉 T0 , L
of metastable states is infinite because ho-
1
mogeneous nucleation is no longer pos- = 6 ∑ [ 〈 ci c j 〉 T0 − c 2 ]
sible (“heterophase fluctuations” are su- L i, j
pressed at the same times as “homophase 1
= 3 ∫ dx [ 〈 c ( 0 ) c ( x )〉 T0 − c 2 ] (6-49)
fluctuations”). The spinodal curve is the L
limit of metastability here.
the summations over i and j being restrict-
While such a system with infinitely
ed to sites within the cell of size L3 cen-
long-range interactions is clearly artificial,
tered at x and in the last step the sums are
it makes sense to consider systems with
converted into integrals (one sum Â is can-
long but finite range r of the interactions i
(Heermann et al., 1982; Binder, 1984 b; celled against a factor L3 making use of the
translational invariance of the correlation equalities which must be satisfied require

function). For distances | x |  x the correla- that
tion function is simply given by a power-
r 3 (1 – T /Tc )1/2 p1 (6-53)
law decay, ·c (0) c (x)ÒT0 – c– 2 ≈ r– 2 x– 1, and
therefore the order of magnitude of the in- This condition can be satisfied only for
tegral in Eq. (6-49) is estimated, yielding a large range r of interaction, and defines
the mean-field critical region (Ginzburg,
·[dc (x, 0)]2 ÒT0 , L ≈ r – 2 L – 1 (6-50) 1960). As long as Eq. (6-53) holds, the
As expected, the mean-square concentra- Landau mean-field theory of critical phe-
tion fluctuation is the smaller the larger is nomena (Stanley, 1971) is quite adequate.
the range of the interactions and the larger However, for systems with rather short-
is the coarse-graining length L. Since the range interactions Eq. (6-53) and hence
largest permissible value for L is L ≈ lc , Eqs. (6-51) and (6-52) can never be ful-
Eqs. (6-45), (6-48), and (6-50) readily filled.
yield (ii) Even for systems with a long but fi-
nite range r of interaction, close enough to
exp (2 Rm t) O r3 (1 – T /Tc )1/2 Tc (namely for r3 (1 – T /Tc )1/2 ≈ 1) a cross-
¥ [c– /csp (T ) – 1] 3/2 (6-51) over occurs to a non-mean-field critical be-
A similar self-consistency criterion can be havior, described by the same critical be-
formulated in the metastable region, where havior as short-range systems. This is ex-
the largest permissible choice for L in the pected from the principle of universality of
Ginzburg criterion critical behavior (Kadanoff, 1976). Then
the nonlinear character of Eq. (6-15) is im-
·[dc (x)]2 Ò O (c (T ) – c– )2
T, L sp portant during the initial stages of the
quench, and the linearization approxima-
now is L = x (c– ), as given in Eq. (6-46), and
tion (Eq. (6-18)) is never warranted.
it is found in full analogy with Eqs. (6-48)
(iii) For mean-field-like systems for
to (6-51) that
which Eq. (6-53) holds, Eqs. (6-51) and
1 O r 3 (1 – T /T )1/2 [1 – c– /c (T )] 3/2 (6-52)
c sp (6-52) hold only if we do not come too
close to the spinodal. The region excluded
It is also interesting that Eq. (6-52) can be
by these inequalities scales as
derived from a completely different argu-
ment, namely requiring that the free energy | c– /c (T ) – 1 | ~ r – 1 (1 – T /T )– 1/3
sp c
barrier of nucleation DᏲ *pkB T (Binder,
In this excluded region the singularities de-
1984 b) (for c– near csp (T ) the Cahn – Hil-
scribed by Eqs. (6-45) to (6-47) are essen-
liard (1959) mean-field theory of nuclea-
tially smeared out, and a gradual transition
tion predicts DᏲ */ kB T ~ r 3 (1 – T /Tc )1/2
from nucleation and growth to nonlinear
¥[1 – c– /csp (T )]3/2). From Eqs. (6-51) and
spinodal decomposition occurs. This grad-
(6-52) we draw the following conclusions:
ual transition can be understood qualita-
(i) For the linearized theory of spinodal tively from a cluster dynamics treatment of
decomposition to be self-consistent, and for phase separation (Binder and Stauffer,
the description of a metastable state near 1976 a; Binder, 1977, 1981; Mirold and
the spinodal to be self-consistent (and that it Binder, 1977; Binder et al., 1978). Figs.
has a long lifetime owing to a large barrier 6-11 and 6-12 summarize the main ideas
against homogeneous nucleation), the in- about this description which is closely
related in spirit to the Langer – Schwartz clusters of size l and l¢, and cl, l¢ is the rate
(1980) and Kampmann – Wagner (1984) factor for the inverse “coagulation” reac-
treatments of concomitant nucleation, tion. These rate factors are assumed to be
growth, and coarsening (see Wagner et al., independent of time t and hence Eq. (6-54)
2001). Each state of the system is charac- also describes the concentration fluctua-
terized by a cluster size distribution n–l (t), tions in thermal equilibrium, where de-
where l is the number of B atoms contained tailed balance must hold between splitting
within a cluster (Fig. 6-11 a) and the bar and coagulation reactions:
represents an average over other “cluster
Sl + l¢, l¢ nl + l¢ = cl, l¢ nl nl¢ ∫ W (l, l¢) (6-55)
coordinates” (cluster surface area, shape,
etc.) which are not considered explicitly. where nl is the cluster concentration in
The time evolution of the cluster size equilibrium and W (l, l¢) is a cluster reac-
distribution in a quenching experiment tion matrix in equilibrium. While nl (t = 0)
is described by a system of kinetic equa- for a random distribution of atoms in the
tions: alloy is just the cluster distribution in the
∞ well-known percolation problem (Stauffer,
d
nl (t ) = ∑ Sl + l ′, l ′ nl + l ′ (t ) 1985), nl and W (l, l¢) are not explicitly
dt l ′ =1 known, but can be fixed by plausible as-
l −1
1 sumptions (Binder, 1977; Mirold and
− ∑ Sl, l ′ nl (t ) Binder, 1977). Then Eqs. (6-54) and (6-55)
2 l ′ =1
l −1 can be solved numerically (see Fig. 6-11 b).
1
+ ∑ cl − l ′, l ′ nl ′ (t ) nl − l ′ (t ) What must happen is that n–l (t) for t Æ ∞
2 l ′ =1 develops towards nl , the cluster size distri-
∞
bution in the state at the coexistence curve
− ∑ cl, l ′ nl (t ) nl ′ (t ) (6-54) ∞
l ′ =1 (1)
c coex (1)
, with c coex = ∑ l nl . The excess
l =1
Here Sl + l¢, l¢ is a rate factor for a splitting ∞
reaction of a cluster of size l + l¢ into two concentration c– – c coex
(1)
= ∑ l [nl (0) – nl ] is
l =1
Figure 6-11. (a) “Clusters”

of B atoms in a binary AB
mixture defined from con-
tours around groups of B
atoms and their reactions.
(b) Cluster size distribution
n–l (t) for various times t after
a quench from infinite tem-
perature to T /Tc = 0.6 at
c– = 0.1 for parameters appro-
priate for a two-dimensional
Ising model of a binary al-
loy. These results were ob-
tained from a numerical so-
lution of Eq. (6-54), where
time units are rescaled by an
arbitrary rate factor. From
Mirold and Binder (1977).
redistributed into macroscopic B-rich do- lations (Marro et al., 1975; Rao et al.,
(2)
mains (of concentration c coex ), i.e., occurs 1976; Sur et al., 1977).
at cluster size l Æ ∞ for t Æ ∞. For interme- The description in terms of Eqs. (6-54)
diate times, we find a nonmonotonic clus- and (6-55) contains both nucleation and
ter size distribution (Fig. 6-11 b); a mini- growth and coagulation as special cases
mum occurs at the critical size l* of nucle- (Binder, 1977), and it can be used as a ba-
ation theory (Binder and Stauffer, 1976 a), sis for understanding the gradual transi-
while the broad maximum at larger sizes is tion from nucleation to spinodal decompo-
due to growing supercritical clusters which sition (Binder et al., 1978). In the meta-
have already been nucleated. For l < l* the stable regime the density of unstable fluc-
cluster size distribution is basically the tuations (“critical” and “supercritical” clus-
equilibrium size distribution of a slightly ters with l < l*) is very small (Fig. 6-12 a),
supersaturated solid solution. As time goes because the energy barrier D F* p kB T.
on, the peak in n–l (t), representing the Near the spinodal curve, on the other hand,
supercritical growing clusters, shifts to D F* ≈ kB T, and hence there is a high den-
larger and larger cluster sizes, and at the sity of unstable fluctuations: DF* here no
same time the supersaturation is dimin- longer limits the growth, but rather the
ished, until for t Æ ∞ the peak has shifted conservation of concentration. Near grow-
to l Æ ∞. This separation of clusters into ing clusters, c (x, t) locally decreases, and
two classes – small ones describing con- then no other cluster can grow there. Ow-
centration fluctuations in the supersatu- ing to this “excluded volume” interaction
rated A-rich background and large ones of clusters, a quasiperiodic variation of
describing the growing B-rich domains – concentration results (Fig. 6-12 a, bottom),
can also be shown analytically from Eqs. roughly equivalent to a wavepacket of
(6-54) and (6-55) (Binder, 1977). This pic- Cahn’s concentration waves. But the latter
ture of the phase separation process also are not growing independently, rather they
emerges very clearly from computer simu- are strongly interacting; hence it is more
Figure 6-12. (a) Schematic “snapshot

pictures” of fluctuations of an unmixing
system in the metastable regime (top)
and in the unstable regime (bottom).
Unstable fluctuations, which steadily
grow with increasing time (arrows) are
shaded. (b) Structure factor S (q, t) vs.
scaled wavevector q at various times t
after a quench from infinite temperature
to T = 0.6 Tc for c– = 0.1. Parameters are
chosen for a three-dimensional Ising
model of a binary alloy. These results
were obtained from a numerical solu-
tion of Eqs. (6-54) to (6-56). From
Binder et al. (1978).
reasonable to consider large clusters as sical and nonclassical “spinodal nuclea-

objects nearly independent of each other tion” is essentially given by the condi-
(apart from the reactions in Eq. (6-54)) tion that the free-energy barrier D f * to
rather than as these waves. form a critical nucleus is of the order of
The fact that the cluster picture and the r 3 (1 – T /Tc )1/2 p1 (Fig. 6-14 a). Therefore
concentration picture are simply two dual this regime of “spinodal nucleation” must
descriptions of the same physical situation disappear when the non-mean-field critical
becomes very apparent when the structure region is approached (Fig. 6-14 b).
factor S (k, t) is calculated from the cluster It should be emphasized that the phase
size distribution n–l (t). Introducing the con- diagram in Fig. 6-13 showing these various
ditional probability gl (x) that the site x + x¢ regimes in the temperature – concentration
is taken by a B atom if x¢ is taken by a B plane is relevant only in the very early
atom of an l-cluster, we find (Binder et al., stages of phase separation. This should be
1978): obvious from Eq. (6-51). The time range
∞ where it is valid is, at best, of the order of
S (k, t ) = ∑ l nl (t ) ∫ dx {gl ( x ) − c} Rm t ≈ l n r (r being measured in units of the
l =1
lattice spacing). Therefore the experimental
× exp ( i k ⋅ x ) (6-56)
verification of the linear theory of spinodal
Simple assumptions for gl (x) yield S (k, t) decomposition is a difficult problem (see
as shown in Fig. (6-12 b). The resulting Sec. 6.4). Clear confirmation of the ideas
S (k, t) is therefore in qualitative agreement described above has come from computer
with experimentation (Fig. 6-5 a, b), com- simulations of medium-range Ising models
puter simulation of microscopic models due to Heermann (1984 a) (see Figs. 6-15
(Fig. 6-5 c, d) and the approximate non- and 6-16). Since for the Ising model the re-
linear theory of spinodal decomposition sult for R (k) is readily worked out, a com-
of Langer et al. (1975) (see Fig. 6-8, bot- parison between theory and simulation is
tom). possible without any adjustable parameter
This gradual transition between nuclea- whatsoever! Fig. 6-15 shows that there is
tion and spinodal decomposition always indeed a regime where initially fluctuations
appears close to the critical point in the increase exponentially with time (Eq.
phase diagram (Fig. 6-13); for systems (6-23)) and the observed rates R (k) do
with a large interaction range r a mean- agree with the predictions, if we stay far off
field critical regime exists (Eq. (6-53)) the spinodal.
where this gradual transition is confined to It is not easy to identify physical systems
a narrow regime adjacent to the spinodal. that have a large but finite interaction
Outside this transition regime the linear- range. What is really needed is a large pre-
ized theory of spinodal decomposition is factor in the mean-field result for the corre-
expected to hold initially in the unstable relation length (Eq. (6-46)), and such a large
gime, while a nonclassical nucleation the- prefactor has in fact been identified for
ory (where ramified droplets, which must polymers with high molecular weight (Eq.
first be compacted before they can grow, (6-40)). This happens because for polymers
are nucleated (Klein and Unger, 1983; the coefficient of the gradient energy in Eq.
Unger and Klein, 1984; Heermann and (6-32) comes from the random-coil struc-
Klein, 1983 a, b)) holds in the metastable ture, and does not have the meaning of a
regime. This crossover line between clas- squared interaction range as in Eq. (6-4).
Figure 6-13. Various regimes in the phase diagram of a binary mixture AB, showing part of the plane formed
by temperature T and volume fraction F of the B component (only volume fractions F < Fcrit are shown, since
this schematic phase diagram is symmetric around the axis F = Fcrit ). The horizontal broken line separates the
non-mean-field critical regime (top) from the mean-field critical regime (bottom). The two solid curves are the
coexistence curve (left) and the spinodal curve (right). The two dash-dotted curves on both sides of the spino-
dal limit represent the regime where a gradual transition from nucleation to spinodal decomposition occurs. The
linearized theory of spinodal decomposition (Fig. 6-1 a) should hold to the right of these dash-dotted curves. The
regime between the coexistence curve and the left of the two broken curves is described by classical nucleation
theory (compact droplets, Fig. 6-1 b). The regime between the right broken curve and the left dash-dotted curve
is described by “spinodal nucleation” (ramified droplets). From Binder (1984 b).
Figure 6-14. Schematic plots of the nucleation free energy barrier for (a) the mean-field critical region
of a d-dimensional alloy system, i.e., r d (1 – T/ Tc ) (4 – d )/2 o 1 and (b) the non-mean-field critical region,
i.e., r d (1 – T/ Tc ) (4 – d )/2 < 1. The gradual transition from nucleation to spinodal decomposition occurs for
DF*/kB Tc ≈ 1. From Binder (1984 b).
However, we note that for NA = NB = N we

may map Eqs. (6-33) and (6-32) into Eqs.
(6-4) and (6-6) with the following identifi-
cations (Binder, 1984 b):
1
DᏲpolymer ≡ DᏲsmall molecules
N
T 1 2
c ≡2 c , r 2 ≡ N a2 (6-57)
T N 3
The condition that the nucleation barrier
D Ᏺ * /kB T p1 instead of Eq. (6-52) then
Figure 6-15. Time dependence of the logarithm of yields
(6-58)
spherically averaged structure factors of a 603 sim- 1/ 2
⎛ c ⎞
ple-cubic Ising lattice, where each spin interacts with 1 O N 1/ 2 ⎜ 1 − crit ⎟ |1 − c / csp ( T )| 3 / 2
q = 124 neighbors with equal interaction strength J, ⎝ c ⎠
quenched from infinite temperature to
T = (4/9) T MF = (4/9) q J/k and c– = 0.4
and a similar criterion holds on the un-
c B
stable side of the spinodal (Binder, 1983),
The five smallest wavenumbers kn = 2 p n/60 are dis-
played. Straight lines for short times indicate expo-
with 1 being replaced by exp (2 Rm t) on
nential growth and thus yield R (k). From Heermann the left-hand side of Eq. (6-58), as in Eq.
(1984 a). (6-51). It follows that for large chain length
N the spinodal is smeared out over a region
of width N –1/3.
In view of these results, it is gratifying to
note that for polymers, convincing experi-
mental demonstrations of the validity of
Cahn’s linearized theory of spinodal de-
composition have indeed been presented
(see Sec. 6.3.4).
6.2.6 Towards a Nonlinear Theory

of Spinodal Decomposition
in Solids and Fluids
In the last subsection it was shown that
the regime of times for which the linear
theory of spinodal decomposition holds is
extremely restricted, if it holds at all.
Therefore, a treatment of nonlinear effects
is necessary. A systematic approach is only
possible by an expansion in powers of 1/r
Figure 6-16. Cahn plot R (k)/k2 vs. k2 (R (k) is defor large r (Grant et al., 1985). This theory
noted as w (k) in this figure), as extracted from data is very complicated and has so far been
such as shown in Fig. 6-15. Crosses are the Monte
Carlo results, and straight lines are the predictions of
worked out only for c– = ccrit. It explicitly
the linearized theory. Note that csp (T ) ≈ 0.127 in this shows the coupling between concentration
case. From Heermann (1984 b). waves with different wavevectors.
So far the most popular (if approximate) that is similar to Eq. (6-27), namely,
approach is the decoupling approximation
d
suggested by Langer et al. (1975). Exact S ( k, t ) = − 2 M k 2 (6-60)
equations of motion are derived for the dt
probability distributions r1 [c (x)], r2 [c (x1), ⎧ ⎡⎛ ∂ 2 f ⎞ ⎤ ⎫
⎪ ⎪
c (x2)], etc. Here r1 is the probability den- × ⎨ ⎢⎜ 2 ⎟ + a (t ) + r 2 T k 2 ⎥ S ( k, t ) − kB T ⎬
⎪⎩ ⎢⎣ ⎝ ∂c ⎠ T, c ⎥⎦ ⎪⎭
sity that, at point x, the concentration c (x)
occurs, and r2 is the corresponding two-
where all nonlinear effects are now con-
point function. As expected, the equation
tained in a correction term a (t), which it-
of motion for r1 involves r2, and the equa-
self depends on S (k, t) in a nonlinear way.
tion of motion for r2 involves the three-
As noted above, the coefficients A, B,
point function r3 , etc., so that an infinite
and r in Eqs. (6-5) and (6-8) are adjusted
hierarchy of equations of motion is gener-
such that the critical behaviors of the co-
ated.
existence curve, critical scattering inten-
This hierarchy is decoupled by the fol-
sity, and correlation length (at the coexis-
lowing approximation for the two-point
tence curve) are reproduced:
function [d c (x) = c (x) – c– ]:
(6-59) (1)
(c coex – ccrit )/ccrit = B̂ (1 – T /Tc ) b (6-61 a)
r2 [c ( x1 ), c ( x2 )] = r1 [c ( x1 )] r1 [c ( x2 )] –g
ccoex = Ĉ (1 – T /Tc ) (6-61 b)
⎧ 〈d c ( x1 ) d c ( x2 )〉 T d c ( x1 ) d c ( x2 ) ⎫
× ⎨1 + ⎬ xcoex = x̂ (1 – T /Tc ) –n
(6-61 c)
⎩ 〈 (d c ) 2 〉 T 〈 (d c ) 2 〉 T ⎭
The motivation for Eq. (6-59) is the fol- where B̂, Ĉ, and x̂ are the appropriate criti-
lowing: if there were no correlation be- cal amplitudes and b, g, and n the asso-
tween concentrations at points x1 and x2, ciated critical exponents (Stanley, 1971;
the probability r2 would just be the product Binder, 2001). It now turns out that the
of the one-point probabilities. Therefore, strength of nonlinear effects is controlled
the correction of this factorization approxi- by the inverse of a parameter f0 which near
mation is made proportional to the two- Tc is expressed as (Billotet and Binder,
point correlation function ·dc (x1) dc (x2 )Ò T . 1979)
In this way, Eq. (6-59) yields a closed equa- f0 ~ x̂ d B̂ 2 Ĉ – 1 (1 – T /Tc )g + 2 b – dn (6-62)
tion of motion for the probability r1 . This
equation is then solved approximately, as- for a d-dimensional system. In the non-
suming that the coarse-grained free energy mean-field critical region, the hyperscaling
has the Landau form (Eqs. (6-5) and (6-8)). relation dn = g +2 b (Kadanoff, 1976) elimi-
The coefficients A, B and r are adjusted nates the temperature dependence from Eq.
self-consistently such that the resulting (6-62). In addition, two-scale factor uni-
state equilibrium is correctly described in versality (Stauffer et al., 1972) implies that
the non-mean-field critical region. the critical amplitude ratio xˆd B̂ 2/Ĉ and
This approach has also been worked out hence f0 is a universal constant of order
for the dynamics of non-conserved order unity (Billotet and Binder, 1979), i.e.,
parameters (Billotet and Binder, 1979) and f0 ≈ 9.45. On the other hand, if we consider
the validity of this approximation has been a quench into the mean-field critical re-
studied carefully (see also Binder et al., gion, again adjusting the coefficients A and
1978). The final result can be cast in a form B in Eq. (6-8) but now using mean-field
results in Eqs. (6-61 a – c), namely, b = –12 , However, for large values of f0 the linear
g = 1, and n = –12 , xˆ ~ r, we obtain theory does hold initially, consistent with
the simulations (Fig. 6-15). The same con-
f0 ~ r d (1 – T /Tc )(4 – d)/2 clusion emerges from the time dependence
for r d (1 – T /Tc )(4 – d)/2 p1 (6-63) of the rescaled effective second derivative
m – (T, t) in Eq. (6-60):
It is seen that f0 is now simply proportional
to the parameter appearing in the Ginzburg m – (T, t) ∫ 1 + a (t)/[∂2 f /∂c2]T, c (6-64)
criterion, and hence very large. For (see Fig. 6-18). Nonlinear effects are negli-
r d (1 – T /Tc )(4 – d)/2 ≈ 1 a crossover to the uni- gible as long as m – = 1, while the decrease
versal constant f0 ≈ 9.45 in the non-mean- in m – is a signal of nonlinear effects, since
field critical region occurs. a (t) is always negative. The physical
Fig. 6-17 shows the time evolution for significance of m – (T, t) is that it describes
the structure factor for two choices of f0. the ratio between the actual kc2 (t), where
Here again a rescaling of the structure fac- the growth rate in Eq. (6-60) changes sign,
tors Ŝ = r 2 kc2 S and of time t = 2 M r 2 T kc4 t and the corresponding prediction of the
was used, while q = 40 k /(2 p kc) here. It is Cahn – Hilliard theory. Fig. 6-18 implies
seen that for f0 ≈ 9.45 the linear theory is that even if the nonlinear effects are very
indeed invalid from the start, as expected. strong the Cahn – Hilliard prediction for kc2
differs from the actual kc2 (t) by about a fac-
tor of two at most; therefore, the Cahn –
Hilliard theory is certainly useful for esti-
Figure 6-18. Rescaled effective second derivative

Figure 6-17. Plot of ln {[S̃ (q, t ) – S̃ T (q)]/S̃ T0 (q)} vs. m – (Tf , t) (Eq. (6-64)) vs. scaled time t for four
scaled time t for five different values of q, for a choices of f0 . Only when m – (T, t ) ≈ 1 is the Cahn –
quench from infinite temperature at ccrit , using Hilliard – Cook approximation accurate. For f0 = 9.45
f0 = 9.45 (top) and f0 = 9450 (bottom). From Carmesin the initial value m – (T, 0) = 0.65. From Carmesin et al.
et al. (1986). (1986).
mating the order of magnitude of the range interaction:

of wavelengths that is unstable initially.
For systems with short-range forces, ᏸ = ᏸ A + ᏸ HD (6-65b)
where f0 ≈ 9.45 applies, the early-time d ⎡ d dF ⎤
ᏸ A = − M ∫ dx ∇2 ⎢ +
structure factor for critical quenches (Fig. d c (x) ⎣ d c ( x ) d c ( x ) ⎥⎦
6-8, bottom) is in reasonable agreement
with both experiment (Fig. 6-5 a, b) and d
ᏸ HD = 2 ∫ ∫ dx dx ′ ∇c ( x ) ⋅ T ( x − x ′ )
simulations (Figs. 6-5 c, d). However, the d c (x)
Langer – Baron – Miller (LBM) (1975) ap-
⎡ d dF ⎤
proximation gets worse for strongly off- × ∇′ c ( x ′ ) ⎢ + ⎥ (6-65c)
critical quenches (Binder et al., 1978). The ⎣ d c ( x ′) d c ( x ′) ⎦
theory again exhibits a spurious singularity and
at a spinodal curve, which does not coin-
1 ⎡ 1
cide with the mean-field spinodal of the Tab ( x − x ′ ) = ⎢ | x − x | dab (6-65d)
CHC approximation, but occurs at a renor- 8 ph⎣ ′
malized concentration closer to the coexis- 1 ⎤
tence curve. This shift of the spinodal is not + 3 ( xa − xa
′ ) ( xb − xb′ )⎥
| x − x′| ⎦
surprising, since the effective second de-
rivative of the potential is also renormal- where h is the shear viscosity. Kawasaki
ized (Fig. 6-18). This spinodal of the LBM and Ohta (1978) adapted the LBM decou-
approximation is completely spurious – its pling to binary fluids. Their results differ
precise location depends on details of the from the original LBM theory mainly
coarse-graining procedure. For concentra- where km (t)/km (0) 1; however, at such
tions between the coexistence curve and late times neither of these theories is valid,
this spinodal, i.e., the metastable regime, since they can only account for non-
the structure factor S (k, t) for T Æ ∞ satu- linearities during the early stages where
rates at a value S (k, ∞) = kB T /[∂2 f /∂x 2 )T, c km (t)/km (0) ≈ 1. No theory exists which re-
+ a (∞) + r 2 T k2 ], i.e., at the Ornstein – Zer- liably describes the crossover from these
nike result for scattering from fluctuations initial time regimes to the late stages,
–1
in metastable equilibrium (Binder et al., where km (t) ~ t is presumed to hold.
1978). An interesting extension is to consider
We now consider fluid binary mixtures. spinodal decomposition of fluid mixtures
Here the theoretical formulation is com- in the presence of flow, e.g., laminar shear
plicated by another long-range interaction, flow (Onuki and Kawasaki, 1978, 1979) or
namely the hydrodynamic backflow inter- turbulent flow (Onuki, 1984, 1989 c). In
action. If the Liouville equation is written weak shear, the scattering pattern is no
for the probability distribution r ({c (x)}, t) longer the ring pattern familiar from stan-
that a concentration field c (x) occurs, dard spinodal decomposition, since con-
centration fluctuations become anisotropic.
∂
r ({c ( x )}, t ) = ᏸ r ({c ( x )}, t ) (6-65 a) In polymer mixtures additional effects
∂t arise because the polymer coils become
the Liouville operator ᏸ contains a term stretched and oriented by the flow (Pistoor
ᏸA also present in solid alloys and another and Binder, 1988 a, b). No such phenomena
term ᏸ HD containing the Oseen tensor are considered here.
T = {Ta b } describing the hydrodynamic
6.2.7 Effects of Finite Quench Rate al., 1966; Carmesin et al., 1986). Here we
only quote a few results from the model
So far both the theory and the discussed calculation of Carmesin et al. (1986), since
simulations have always assumed instanta- the work by Houston et al. (1966) only
neous quenches from an initial temperature considers fluctuations in the initial state
Ti to a final temperature Tf . With regard to and not in the intermediate states visited in
actual experiments, this is extremely ideal- the quench, which is an approximation that
ized. Often the early stages of spinodal de- usually cannot be justified.
composition have already been passed dur- As an example, Fig. 6-19 shows a situa-
ing such a continuous quench where the tion similar to the quench treated in Fig. 6-
temperature is gradually lowered. While it 8, but here the quench is not carried out in-
is believed that the late stages are not af- stantaneously from infinite temperature to
fected by the “quenching history”, the lat- T /Tc = 4/9, but takes several steps: at t = 0
ter can have a drastic effect on both early the system is cooled instantaneously to
and intermediate stages of phase separa- T1 /Tc = 0.75185, at t = 1 to T2 /Tc = 0.67667,
tion. Unfortunately, in general, the problem at t = 2 to T3 /Tc = 0.60148, at t = 3 to
is complicated – the behavior of both the T4 /Tc = 0.5263, and at t = 4 to T /Tc = 4/9,
thermodynamic functions and of the mobil- where the system is later maintained. This
ity M (T ) in the full regime from Ti to Tf stepwise quenching is more readily acces-
may affect the phase separation behavior. sible to calculation than a fully continuous
Thus, relatively little theoretical effort has quench. A further simplification of the cal-
been devoted to this problem (Houston et culation is to neglect any temperature de-
Figure 6-19. Scaled structure

function (left) and normalized
diffusion constant D̃eff (q, t ) ∫
q2 d [ln S̃ (q, t )]/dt (right) vs. q
or q2, respectively, for the step-
wise quenching procedure de-
scribed in the text. Top, CHC
approximation; bottom, LBM
theory. Ten scaled times t are
shown. From Carmesin et al.
(1986).
pendence of the mobility. Comparing Fig.

6-19 with Fig. 6-8, characteristic differ-
ences are noted: even in the CHC approxi-
mation there is no longer a unique intersec-
tion point for S (q, t ) – S (q, 0) at different
times t ; apart from t = 0, D̃eff (q, t) in the
“Cahn plot” exhibits pronounced curva-
ture, although it depends relatively weakly
on time. Note that in the LBM approxima-
tion we can no longer see any distinct shift
of the maximum position of S̃ (q, t ) for the
times shown; this happens because first
(at the intermediate temperature steps)
smaller-q components become more ampli-
fied and later (at the final temperature) am-
plification occurs at larger values of q. This
behavior just happens to offset the coarsen-
ing tendency. This example shows that care
must be taken in drawing any conclusions
about the validity of the CHC approxima-
tion from experimental data – the effects of
fluctuations in the final state, the gradual
onset of nonlinearities, and finite quench Figure 6-20. Structure function S (q, t ) vs. scaled
rate effects are interwoven in a compli- wavevector q for scaled times t as indicated in the
cated manner, and very accurate measure- figure, for a continuous quench (linear cooling from
ments and detailed knowledge of the Ti = 1.01 Tc to Tf = 0.99 Tc during t = 0.001), using
system parameters are indispensable for Eact /kB T = 900. Top: CHC approximation; bottom:
LBM approximation. From Carmesin et al. (1986).
disentangling all these effects.
These effects are much more dramatic
if the mobility has a thermally activated
6.2.8 Interconnected Precipitated
behavior, M (T ) ~ exp (– Eact /kB T ), with
Structure Versus Isolated Droplets,
Eact Ô kB Tf . As a second example, we con-
and the Percolation Transition
sider a shallow continuous quench from
Ti = 1.01 Tc to Tf = 0.99 Tc during t = 0.001, Whereas so far the description of the
the temperature being lowered linearly structures formed by phase separation has
with time (Fig. 6-20). Although this is as focussed on the behavior of the equal-time
good an approximation for an instantane- structure factor S (k, t) at time t after the
ous quench as is experimentally feasible, quench, we now concentrate on the de-
nevertheless, for the choice Eact /kB Tc scription of these structures in real space.
= 900, more relaxation takes place during In Fig. 6-2 it was illustrated that during late
the quench than in the remaining early time or intermediate stages two different pat-
interval (t &1). The qualitative behavior of terns of behavior are present, depending
the CHC and LBM approximations is iden- on the volume fraction F of the minority
tical – only the absolute scales of the struc- phase. For small F, the minority phase is
ture factors differ significantly. confined to independent clusters well sep-
arated from each other; for larger F, we

have a percolating interconnected network
(in viewing Fig. 6-2 we expect an isotropic
system, such as a fluid or glassy mixture; in
crystalline solids the shape of the precipi-
tates reflects the anisotropy of the interfa-
cial free energy between the coexisting
phases, and elastic interactions may even
lead to a regular rather than a random ar-
rangement of these precipitates).
This difference in morphology of the
growing structures often – and erroneously
– is associated with the distinction between
nucleation and growth versus spinodal de-
composition; it is then claimed that the per-
colation of the growing structure is the Figure 6-21. Schematic phase diagram (temperature
vs. concentration c– ) of a three-dimensional, short-
hallmark of spinodal decomposition, while range Ising lattice model of a binary mixture. Since
it is assumed that well separated domains the diagram is symmetric along the line c– – ccrit = 0.5,
must have been formed by nucleation and if the meaning of A and B is exchanged, full informa-
growth. We maintain, however, that the tion is given only for the regime c– &0.5. Both the
distinction between nucleation and spino- coexistence curve (“binodal”) separating the single-
phase region from the two-phase region and the
dal decomposition, meaningful only for the mean-field spinodal curve separating metastable
earliest stages of phase separation, must from unstable states in mean-field theory are shown.
not be confused with this distinction in The mean-field spinodal is described here by the
morphology of the precipitated structures, equation (ccrit – c)/ccrit = ± (1 – T/ Tc )1/2 , Tc being the
which is relevant for intermediate and late actual critical temperature. The transition regime
from nucleation to spinodal decomposition for a
stages: even if well separated domains oc- short-range system, as discussed in the text, is also
cur, they may result from a spinodal de- indicated. The dash-dotted curves indicate percola-
composition mechanism, and even if a state tion transitions, as discussed in the text. From Hay-
decays by nucleation, it may correspond to ward et al. (1987).
an interconnected percolating microscopic
structure of B atoms in the A-rich matrix.
Hence, the transition regime between nu- mately by the region from DF* MF = 10 kB T
cleation and spinodal decomposition and to DF* MF = k B T, where DF*MF is the mean-
the line of percolation transitions separat- field result for the nucleation barrier (this
ing the regime of finite B clusters (Fig. shift of the transition region from the mo-
6-11 a) from the regime where a percolat- lecular-field result towards the binodal was
ing cluster occurs which reaches from one disregarded in Fig. 6-13). Two different
boundary of the system to the opposite one percolation transition lines are drawn,
must in fact cross each other (Fig. 6-21). which depend on the resolution with which
Fig. 6-21 shows both the molecular-field the system is observed. Suppose the resolu-
result for the spinodal and the actual re- tion is very fine, such that individual atoms
gime of gradual transition between nuclea- can be distinguished (this is the situation
tion and nonlinear spinodal decomposition envisaged in Fig. 6-11 a). Then in the sin-
of a short-range system, defined approxi- gle-phase region we encounter a percola-
tion transition of a correlated percolation tion is reached where t1 = t2 (for still lower
problem at cp(corr) (T ). This line starts out for concentrations there is no percolation at
T Æ ∞ at the percolation concentration for all), whereas for increasing concentration
random percolation (cp(random) ≈ 0.312 for another critical concentration is reached
the simple cubic lattice (Stauffer, 1985)) where t2 Æ ∞. For still higher concentra-
and bends over to the left, until it hits the tions the system percolates as the time
coexistence curve at the point T /Tc ≈ 0.96, exceeds t1 and then remains percolating
c– ≈ 0.22 (Müller-Krumbhaar, 1974). This throughout.
line continues in the two-phase region as A different behavior occurs if a system
a transient time-dependent phenomenon, with a much coarser resolution is studied;
cp(corr) (T, t) (see Fig. 6-22). Hayward et al. at late stages, where the system is phase-
(1987) and Lironis et al. (1989) have separated on a length scale km–1 (t) into the
shown that for certain concentrations the coexisting phases with concentrations
configuration is percolating for a time given by the two branches of the coexis-
(1) (2)
interval t1 < t < t2 , whereas it does not per- tence curve, c coex and c coex , respectively. It
colate for 0 & t1 and for t 7 t2 . If the con- now makes sense to consider percolation
centration decreases, a critical concentra- phenomena on much larger length scales
than the lattice spacing a. Suppose we di-
vide our system again into cells of linear
dimension L (Eq. (6-1)), with a  L  km–1 (t).
Most of these cells will then have concen-
(1) (2)
trations close to either c coex or c coex . We
may now define clusters consisting of
neighboring cells with concentrations in a
(2) (2)
given interval [c coex – dc/2, c coex + dc/2]
and we may ask whether these clusters are
well separated from each other or if they
form an infinite percolating network. Since
l (t Æ ∞) Æ ∞ we may also take L Æ ∞ in
this limit and therefore dc Æ 0. Hence there
is no longer any ambiguity in this coarse-
grained percolation problem. We expect
that this “macroscopic percolation” will
occur at a critical volume fraction f p of the
minority phase, which does not depend on
temperature. Therefore, the line of macro-
scopic percolation concentrations is simply
cp(macro) = c coex
(1) (2)
+ (c coex (1)
– c coex ) fp
and hence must end in the critical point.
Figure 6-22. Part of the phase diagram of the simple This line is also shown schematically in
cubic, nearest-neighbor lattice-gas model, showing
Fig. 6-21. Note that this line will be ob-
the percolation transition line cp(corr) (T, t), where the
time t refers to an average over the time interval from served experimentally by techniques which
t = 80 to t = 240 MCS per site during phase separa- are sensitive to the “contrast” (i.e., differ-
tion. From Hayward et al. (1987). ence in refractive index) between the two
coexisting phases, such as observations ume fraction for continuum percolation is

with light or electron microscopes. about 0.16 (Scher and Zallen, 1970), we
This phenomenon of “gelation of clus- expect all these theories only to be reliable
ters” at cp(corr) (T, t) into an infinite percolat- for F  0.16. This percolative behavior
ing net has a pronounced effect on the clus- n–l (t) ~ l – t is in conflict with the scaling be-
ter size distribution n–l (t), of course, which havior that is described in the next section,
was used in the “cluster dynamics” model- namely (Binder, 1977, 1989), n–l (t) = l –1
ing of Eq. (6-54). This is shown in Fig. ñ (l t – d/3 ), where ñ (ᒗ) is a scaling function.
6-23 where n–l (t) is plotted against l for The extent to which this percolation transi-
c– = 0.156 in a quench to T = 0, where the tion affects S (k, t) is unclear.
system develops towards a frozen-in clus-
ter size distribution. The curvature of n–l (0) 6.2.9 Coarsening and Late Stage Scaling
on the log – log plot reflects the exponential
variation, ln n–l (0) ~ l, while the straight- From Figs. 6-5 and 6-12 it is evident that
line behavior occurring at later times is the peak position km (t) of the structure fac-
characteristic of the power law of percola- tor S (k, t) decreases with increasing time
tion, n–l (t) ~ l – t . This percolation problem after the quench. This decrease already re-
is usually disregarded in treatments based flects the onset of a coarsening behavior of
on Eq. (6-54) or related models (Langer the domains of A-rich and B-rich phases
and Schwartz, 1980). Certainly this behav- that have formed (Fig. 6-2). For large
ior also makes theories of coarsening, enough times, the domains have grown to a
where the growing droplets are modeled as size much larger than all “microscopic”
essentially spherical, doubtful at volume lengths (such as the interfacial width).
fractions F close to Fp ; note that from Then a simple power law should hold,
Fig. 6-22 and the fact that the critical vol- km (t) ~ t – x , tÆ∞ (6-66)
and the structure factor should satisfy a
scaling hypothesis (Binder and Stauffer,
1974, 1976 b; Binder et al., 1978; Furu-
kawa, 1978; Marro et al., 1979)
S (k, t ) – kB T /[(∂2 f /∂c 2 )T, c(1,
coex
2 2
2) + r T k ]
~ [km (t)] –d S̃ {k/km (t)} (6-67)

The term subtracted on the left-hand side
of Eq. (6-67) represents the scattering from
concentration fluctuations within the do-
mains, and S̃ (z ) is a scaling function which
will be discussed later.
Understanding the growth law, Eq. (6-
Figure 6-23. Cluster size distribution for a system 66), and predicting the associated scaling
of L3 lattice site with L = 40. Full symbols give the function S̃ (z ) has been a longstanding
cluster size distribution n–l (0) corresponding to the problem that is still not completely solved
random percolation problem at c– = 0.156 at the sim-
ple cubic lattice, and open symbols denote the cluster
(see Bray, 1994, for a recent review). It is
size distribution averaged over the time intervals now thought that for solid mixtures, both in
shown. From Hayward et al. (1987). d = 2 and in d = 3 dimensions, but in the ab-
sence of any elastic interactions (see Sec. Binder, 1977; and Binder et al., 1978) and
6.2.10), a result originally derived by Lif- various extensions (e.g., Langer and
shitz and Slyozov (1961) and Wagner Schwartz, 1980; Kampmann and Wagner,
(1961) holds 1984; Wagner et al., 2001) incorporate the
LSW growth law, nucleation and coagula-
x = 1/3 (6-68)
tion in a phenomenological way. But they
This LSW theory is essentially a mean- do not take into account the statistical fluc-
field theory valid in the limit of zero tuations and the correlations in the diffu-
volume fraction F of the new phase sion field around growing clusters.
[F ∫ (c– – c coex
(1) (2)
)/(c coex (1)
– c coex )] and consid- Eq. (6-68) has now been confirmed by
ers the cluster size distribution n–l (t) that approaches based on scaling ideas (Furu-
was discussed in Sec. 6.2.5, showing that kawa, 1978, 1984, 1985 a, 1988; Bray,
for F = 0 there exists a solution nl (t) = l –1 1998), renormalization group concepts
ñ (l t – d/3 ), which implies that the mean (e.g., Lai et al., 1988; Bray, 1990, 1994),
–
cluster “size” (i.e., volume) l (t) scales as and theories considering fluctuating ran-
–
l (t) ~ t d/3 for t Æ ∞ (for a more detailed dom interfaces (Mazenko, 1994; Mazenko
outline of this theory, see Wagner et al., and Wickham, 1995). However, the most
2001). Thus the cluster linear dimension convincing evidence that Eq. (6-68) is true
–
scales as [l (t)]1/d ~ t1/3, and if we can ex- both for critical and for off-critical
tend this result to nonzero F we would ex- quenches in d = 2 and in d = 3 comes from
–
pect km (t) ~ [l (t)]1/d and hence Eq. (6-68) computer simulations (Amar et al., 1988;
results. However, despite numerous at- Gunton et al., 1988; Gawlinski et al., 1989;
tempts (e.g., Tokuyama and Enomoto, Huse, 1986; Rogers et al., 1988; Rogers
1993; Akaiwa and Voorhees, 1994) even and Desai, 1989; Chakrabarti et al., 1993).
the extension of LSW theory to the case of We emphasize that Eq. (6-68) holds both
small F is only approximately possible, in d = 2 and in d = 3 dimensions. The situa-
and the accuracy of these extensions is tion is different for fluid binary mixtures:
open to doubt (Mazenko and Wickham, the droplet diffusion – coagulation mecha-
1995). Different power laws for not so late nism (also called “Brownian coalescence”)
stages have also been proposed (e.g., predicts (Binder and Stauffer, 1974)
Binder, 1977, invoking a cluster diffusion
km (t) ~ t –1/d (6-69)
and coagulation mechanism, and To-
kuyama and Enomoto, 1993), and observed while mechanisms invoking hydrodynamic
in computer simulations where atomic dif- flow of interconnected structures imply
fusion is mediated by a single vacancy (Siggia, 1979)
moving through the lattice (Fratzl and Pen-
km (t) ~ t –1 , d=3 (6-70)
rose, 1994, 1997). Only for deep quenches
where we must take into account that the and km (t) ~ t –1/2, d = 2 (San Miguel et al.,
mobility M in Eq. (6-11) is itself concen- 1985), in the “viscous hydrodynamic re-
tration-dependent and (almost) vanishing gime” (Bray, 1994), while km (t) ~ t – 2/3 in
in the pure phases, diffusion along inter- the “inertial hydrodynamic regime” (Furu-
faces results in a slower growth law, x = 1/4 kawa, 1985 c).
(Puri et al., 1997). A theoretical problem that is still out-
“Cluster dynamics” approaches such as standing is the calculation of the scaling
Eq. (6-54) (see Binder, 1977; Mirold and function S̃ (z ) in Eq. (6-67) (see Bray
(1994), Puri et al. (1997) and Mazenko call that for fluid binary mixtures even the
(1994, 1998) for recent discussions). We exponent x is different in these two regimes
mention here only a few approaches very (Eqs. (6-69), (6-70)). For the percolative
briefly. Rikvold and Gunton (1982) used regime of fluid mixtures, Furukawa (1984)
Eq. (6-56), following Binder et al. (1978) has proposed an approximate relation for
but treating the depletion zones around qm vs. t
growing clusters more realistically. Clearly,
their model is only qualitative. More ambi- ( qm−1 − 1) [ A*/ B* arctan ( B*/ A* qm )
tious early attempts are due to Furukawa − arctan B*/ A*] = B* t (6-73)
(1984, 1985 a, b), Ohta (1984), Tomita
(1984) and Tokuyama et al. (1987). An where A* and B* are adjustable constants.
equation due to Furukawa (1984) has been Again a more precise treatment has to rely
extensively compared with experimental on numerical calculations (Koga and Ka-
data on Al – Zn alloys (Komura et al., 1985) wasaki, 1991, 1993; Puri and Dünweg,
1992; Valls and Farrell, 1993; Shinozaki
˜
S̃ (z ) ~ z 2/(g̃ 2/2 + z 2 +g ) (6-71) and Oono, 1993; Alexander et al., 1993;
where g̃ = d + 1 for strongly off-critical Bastea and Lebowitz, 1995).
quenches (cluster regime) and g̃ = 2 d for
critical quenches (percolative regime). 6.2.10 Effects of Coherent Elastic Misfit
S̃ (z ) thus exhibits “Porod’s law” (see e.g.,
When phase separation occurs on a crys-
Glatter and Kratky (1982))
talline lattice it is often the case that the
S (k, t) ~ k – (d +1) (6-72) two phases differ slightly in their crystal
in the cluster regime only. Eq. (6-71) also structure or their lattice constants, thus in-
fails to reproduce the exactly established troducing elastic strains in the crystal. The
(Yeung, 1988) behavior for small z , resulting elastic interaction is long-range
S̃ (z ) ~ z 4. For systems without hydrody- and typically anisotropic and may consid-
namics, good empirical forms for S̃ (z ) erably change the phase separation pro-
describing the limiting behavior both for cess, e.g., in metal alloys (for a recent re-
small z and for large z correctly and ac- view, see Fratzl et al., 1999; an extensive
counting both for simulation and experi- treatment of this problem is also given in
ment have been constructed by Fratzl et al. Khachaturyan, 1983).
(1991), but the derivation of such functions The coherent misfit strain tensor eij0 is
from more fundamental theories is still not the strain required to transform the (undis-
fully solved (Mazenko, 1994, 1998; Ma- torted) lattice of one phase into the (undis-
zenko and Wickham, 1995; Bray, 1994). torted) lattice of the other. In metals, parts
Fig. 6-27 shows a comparison of some of this strain can be relaxed by misfit dislo-
theoretical predictions for the halfwidth of cations disrupting the continuity of the lat-
the scaling function S̃ with corresponding tices between the two phases. This process
experiments (Kostorz, 1991). It is intri- will not be discussed here. In cases of non-
guing to note that the behavior of S̃ (z ) does zero misfit strain, eij0 ≠ 0, an influence on
not change much when the volume fraction spinodal decomposition is expected if one
f changes from the “cluster regime” at of the following conditions is violated:
smaller f through a percolation transition (i) both phases have the same elastic
to the interconnected regime at large f . Re- stiffness tensor;
(ii) the misfit strain is purely dilatational; with

(iii) the elastic stiffness tensor is isotropic;
∂w 1 dl ijmn
(iv) the crystal can be considered infi- = ∑ Deij Demn
nitely large; ∂c 2 ijmn dc
de 0
(v) the stress depends linearly on the − ∑ l ijmn Deij mn (6-77)
strains. ijmn dc
If all five conditions are satisfied, then The first term arises from a dependence of
phase separation can proceed indepen- the elastic constants on composition and
dently of the misfit strains (Bitter – Crum the second one from the composition de-
theorem, see, for example, Cahn and pendence of the misfit strain.
Larché (1984)). Otherwise, we can expect The main difficulty in the use of Eq. (6-
changes in the shape of the single-phase 77) is the determination of Deij . Since the
domains, e.g., from spherical to cuboidal or strains in the alloy can be assumed to relax
plate-like shapes but also in their spatial ar- much faster to their equilibrium values
rangement and coarsening kinetics. Typi- than the concentration profiles, Deij will al-
cally the tendency towards shape changes ways be given by the elastic equilibrium
increases when single-phase droplets be- condition
come larger because the elastic energy (be- ∂tij
ing proportional to the droplet volume) in- ∑ =0 (6-78)
j ∂x j
creases faster with the radius than the sur-
face energy. where the stress tensor is defined by
The introduction of elastic misfit effects Hooke’s law
into the theory of spinodal decomposition
means substituting fcg by fcg + w in Eq. (6- tij = l ijmn Demn (6-79)
15). The function w is the elastic energy As a consequence, the equilibrium strain at
density stored in the lattice each point in the material is a (non-local)
1 functional of the entire concentration pro-
w= ∑ l ijmn Deij ( x ) Demn ( x ) (6-74) file c (x, t). In the special case when the
2 ijmn
elastic constants l ijmn are independent of
where l ijmn is the elasticity tensor or stiff- composition, the elastic problem can be
ness tensor (which may depend on the alloy solved by Fourier transformation (Khacha-
composition). turyan, 1966, 1983) to give
1
Deij (x) = eij (x) – eij0 (x) (6-75) w=
2
∫ ∫ Vel ( x − y ) (c ( x, t ) − c )
is the difference between the strain at posi- × ( c ( y, t ) − c ) d 3 x d 3 y (6-80)
tion x in the elastic equilibrium, eij (x), and or
the misfit strain, eij0 (x). Inserting this into (6-81)
∂w
the nonlinear Cahn – Hilliard equation (Eq. ( x, t ) = ∫ Vel ( x − y ) ( c ( y, t ) − c ) d y
3
(6-14)), we obtain (Larché and Cahn, 1982; ∂c

Onuki, 1989 a – c) which enters the equations (6-76). Vel (u) is
an elastic potential that depends on the
∂c ⎧ ∂ fcg ∂w ⎫ stiffness constants l ijmn and the misfit
= M ∇2 ⎨ + − r 2 kB T ∇ 2 c ⎬ (6-76)
∂t ⎩ ∂c ∂c ⎭ strain eij0 (Khachaturyan, 1983). The aver-
age concentration within the specimen is and Onuki, 1990). A discretized version of
called c–. similar equations was used by Khachatur-
In the case of isotropic elasticity and yan and coworkers to study the behavior of
misfit strain, Eq. (6-81) reduces to (Onuki, single precipitates (Wang et al., 1991) or
1989 a – c) the evolution of precipitate morphologies
(Wang et al., 1992, 1993). Moreover,
∂w 2K 2
= h a [c ( x, t ) − c ] (6-82) Monte Carlo simulations of phase separa-
∂c 1 − n P tion have also been performed, including
where h a is the change of lattice constant elastic misfit interactions between un-
with concentration c, that is equally sized atoms on a lattice (Fratzl and
Penrose, 1995, 1996; Laberge et al., 1995,
eij0 (c) = h a (c – c– ) d ij (6-83) 1997; Lee, 1997, 1998; Gupta et al., 2001).
A further approach is the simulation of the
K and nP are Young’s modulus and the
coarsening of particles with sharp inter-
Poisson coefficient of the elastic matrix,
faces to the matrix, describing the elastic
and d ij = 0 if i ≠ j, d ii = 1 (Kronecker delta).
misfit interaction in the framework of mac-
Eq. (6-82) has been introduced by Cahn
roscopic elasticity theory. Recent examples
(1961). The consequence of this equation is
are the studies by Su and Voorhees (1996),
a shift of the spinodal line towards lower
Abinandanan and Johnson (1993, 1996) or
temperatures because Eq. (6-44) has to be
Jou et al. (1997) (see also the review by
replaced by
Fratzl et al. (1999)). The common observa-
∂ 2 fcg ∂ 2 w ∂ 2 fcg 2K 2 tions in all these approaches are that the
2 + 2 = 2 + ha = 0 domains become very anisotropic and typi-
∂c ∂c ∂c 1 − nP
cally orient parallel to crystallographic di-
Hence, the coherent elastic misfit between rections of the alloy crystal (Fig. 6-24 d, e).
the phases may stabilize the solid solution, Moreover, the spatial arrangement of the
even though the demixing into two inco- domains becomes progressively more peri-
herently separated phases would decrease odic. Despite these enormous changes with
the total energy (Cahn, 1961). The reason respect to the case without elastic misfit
is that the coherency condition forces the interactions, the growth law of a typical do-
alloy to store a considerable amount of main size is often still described by the re-
elastic energy in the lattice, which could be sult of the LSW theory (Eq. (6-68)). If ex-
released by creating an incoherent boun- ternal stress is applied, an additional reor-
dary between two regions both having the ientation of the domains either parallel or
(undistorted) equilibrium lattice structure. perpendicular to the applied stress is ob-
For anisotropic stiffness constants l ijmn served (Laberge et al., 1995, 1997; Weinka-
or anisotropic misfit strains eij0 , the chemi- mer et al., 2000). Many of these effects are
cal potential due to elastic interactions, observed in real alloys, most notably for the
∂w/∂c, will depend on the direction in the technically important nickel-based superal-
crystal. This has already been recognized loys (Maheswari and Ardell, 1993; Conley
by Cahn (1962) (see also Khachaturyan et al., 1989; Sequeira et al., 1995; Paris et
(1983) and Onuki (1989 a – c)). Numerical al., 1995, 1997; Fährmann et al., 1995).
solutions of Eq. (6-76) combined with Eq. If the elastic stiffness constants l ijmn de-
(6-81) have shown the development of pend on alloy composition (see the first
strongly anisotropic domains (Nishimori term in Eq. (6-77)), this can result in anom-
Figure 6-24. Typical snapshot pictures using the

Cahn – Hilliard equation with elastic interactions (Eq.
(6-77)). (a – c) is the case of isotropic elasticity but
where the elastic stiffness depends on composition.
The stiffer phase is shown white with volume frac-
tions of 0.3, 0.5 and 0.7, respectively (from Onuki
and Nishimori, 1991). Note that the softer phase al- Figure 6-25. Transmission electron micrograph
ways “wraps” stiffer particles. (d, e) is the case taken in the (001)-plane of a Ni – Al – Mo where
where the elastic stiffness has cubic anisotropy. The the lattice spacing between matrix and precipitates
volume fraction of the white phase is 0.5 and 0.7, re- (gamma-prime phase) is different by h a = – 0.5%
spectively (from Nishimori and Onuki, 1990). (aged for 5 h at 1253 K). Cube-like precipitates can
be seen, aligned along the elastically soft directions,
[010] and [100]. (b) Results from computer simula-
alously slow precipitate growth. The most tions of an Ising model with elastic interactions on a
spectacular observation, however, is that simple square lattice and with repulsive interaction
(at sufficiently late times in the coarsening of like atoms on nearest-neighbor sites (with interac-
tion energy J) and attractive interaction of like atoms
process) the softer phase always “wraps”
on next-nearest-neighbor sites (energy J/2). The dis-
precipitates of the harder phase (that is, the ordered phase (containing mostly A atoms) is shown
phase with the higher stiffness) and, therein black and the ordered phase (consisting of about
fore, stays percolated even if it is the mi- 50% A and 50% B atoms) is shown in white. The
nority phase (Onuki and Nishimori, 1991; overall concentration of B atoms was 0.35 and the
run was performed at a temperature of T = 0.567 J/kB
Sagui et al., 1994; Leo et al., 1998; Orli-
on a 128 ¥ 128 lattice and stopped after 106 Monte
kowski et al., 1999), see Fig. 6-24 a – c. Carlo steps. (c) The same alloy and heat treatment as
Most recently, the possibility of atomic in (a), but now with an external compressive load of
ordering within the precipitates has also 130 MPa applied along the vertical [010]-direction.
been included in the theory outlined in this (d) The same model, temperature and annealing time
as in (b), but with an additional external load along
section (Sagui et al., 1994; Wang and
the vertical direction. Experimental data (a and c) are
Khachaturyan, 1995), as well as in Monte from Paris et al. (1997) and simulation data (b and d)
Carlo simulations on an elastic lattice (Nie- from Nielaba et al. (1999).
laba et al., 1999). This amounts to studying
decomposition close to a tricritical point,
including the effects of elastic misfit inter- 6.3 Survey of Experimental
action. With these features, the theoretical Results
models are now capable of describing a sur-
prising number of details in the evolution of The number of experimental studies de-
alloy microstructures (e.g., Li and Chen, voted to spinodal decomposition is enor-
1998). An example is shown in Fig. 6-25. mous: first, it is a widespread phenomenon
6.3 Survey of Experimental Results 451
that occurs in diverse systems; second, Paris et al., 1995, 1997; Sequeira et al.,
there has been much interest from a theo- 1995). In all these systems, the structure
retical point of view in this phenomenon, factor looks very similar to the early data
and therefore many experimental studies on Au – Pt alloys (Singhal et al., 1978) and
have been carried out in an attempt to test those of Al – Zn (Mainville et al., 1997) re-
some of the theoretical concepts. produced in Fig. 6-5. These results are in
This section cannot present a complete qualitative agreement with the nonlinear
overview of all these experiments. More LBM theory and the computer simulations.
detailed reviews of various aspects of the The hallmarks of the linearized Cahn the-
experimental work can be found in Haasen ory (exponential increase in intensity with
et al. (1984), De Fontaine (1979), Gerold time, time-independent intersection point
and Kostorz (1978), Gunton et al. (1983), of S (k, t), maximum position km of S (k, t)
Goldburg (1981, 1983), Beysens et al. independent of time) are never found. In a
(1988), Kostorz (1988, 1994), Hashimoto few cases it has been found that at early
(1987, 1988, 1993), and Nose (1987), times km (t) is nearly independent of time,
among others. Here we attempt only to e.g., in Mn67Cu33 (Morii et al., 1988), and
give a few representative examples to illus- this fact has been taken as evidence for the
trate some of the points discussed in the validity of the Cahn–Hilliard–Cook (CHC)
theoretical section, and at the same time approximation. In view of the fact that
show similarities – as well as differences – km (t) almost independent of t can also re-
between different systems. sult as an effect of continuous rather than
instantaneous quenching (Sec. 6.2.7) from
a nonlinear theory, we feel that the question
6.3.1 Metallic Alloys
of whether the CHC theory applies to any
Some systems in which spinodal decom- of the metallic systems quantitatively is
position has been very extensively studied still unanswered. Certainly we expect that
are Al – Zn alloys (e.g., Hennion et al., the CHC theory gives a useful order of
1982; Guyot and Simon, 1982, 1988; Si- magnitude estimate of km (0) and the initial
mon et al., 1984; Osamura, 1988; Komura growth rate Rm , as it does for the theoreti-
et al., 1985, 1988; Mainville et al., 1997), cal models (Fig. 6-5 d), provided a spinodal
ternary Al – Zn – Mg alloys (Komura et al., curve is not close. Experiments by Guyot
1988; Fratzl and Blaschko, 1988), Ni- and Simon (1982) and Hennion et al.
based alloys such as Ni – Al, Ni – Ti, Ni – Cr, (1982) were deliberately carried out choos-
and Ni – Mo (Kostorz, 1988; see also ing temperatures both above and below the
Kampmann and Wagner, 1984), Al – Li al- spinodal curve of Al – Zn, for several con-
loys (Furusaka et al., 1985, 1986, 1988; Li- centrations. The structure factor in both
vet and Bloch, 1985; Tomokiyo et al., cases is qualitatively the same, as was pre-
1988; Che et al., 1997; Schmitz et al., dicted (see Sec. 6.2.5). This classical set of
1994; Hono et al., 1992), Mn – Cu alloys experiments in our view definitely shows
(Gaulin et al., 1987), Ni – Si alloys (Chen et that the spinodal singularity does not play
al., 1988; Cho and Ardell, 1997), Fe – Cr al- any role in the unmixing kinetics of Al – Zn
loys (Katano and Iizumi, 1984), and vari- alloys. Note that it is clear that the effective
ous other ternary alloys, e.g., Cu – Ni – Fe potential in this alloy is predominantly
(Lyon and Simon, 1987; Lopez et al., 1993) short range; however, this alloy does have
or Ni – Al – Mo (Fährmann et al., 1995; a significant misfit of atomic radii between
Al and Zn, and thus the “coherent phase di- t is curved and a well-defined exponent x
agram” (where elastic strain fields are not cannot be identified, to the final Lifshitz –
released) is depressed by about 30 K below Slyozov behavior, x = 1/3. Katano and Ii-
the thermodynamic (incoherent) miscibil- zumi (1984) interpret the result x = 1/6 in
ity gap, and the spinodal curves are then terms of the mechanism (Binder and Stauf-
similarly shifted, providing an example fer, 1974) that B-rich clusters in a solid
of the effects of elastic interactions (Sec. solution show a random diffusive motion
6.2.10). Despite these significant long- since B atoms evaporate randomly from the
range elastic forces, the data for Al – Zn cluster and re-impinge at another boundary
spinodal decomposition are surprisingly position, thus leading to a small shift in
similar to the Monte Carlo results on sim- the cluster center of gravity. The resulting
ple nearest-neighbor Ising models of al- cluster diffusivity decreases strongly with
loys, where all such elastic effects are not increasing cluster size. Assuming then that
included. In other cases, such as many bi- two clusters coalesce when they meet in
nary and ternary Ni alloys (see, for exam- their random motions, we arrive at km (t) ~
ple, Fig. 6-25) elastic misfit interactions t –1/(d +3) in d dimensions. Although such a
lead to highly anisotropic precipitate mor- mechanism certainly exists, it is not clear
phologies and to ordered arrangements of whether it ever dominates during a well-
precipitates (see Sec. 6.2.10). defined time interval, since the Lifshitz –
There is ample experimental evidence Slyozov – Wagner (1961) mechanism com-
for the scaling of the structure factor at late petes with it and should dominate, at least
stages (Eq. (6-67)), and the validity of for long times. Monte Carlo simulations
the Lifshitz – Slyozov exponent x = 1/3 (Eq. considering diffusion via vacanies indicate
(6-68)). Sometimes smaller exponents are that cluster – diffusion – coagulation could
found, e.g., km–1 (t) ~ t x with x ≈ 0.13 to be important at low temperatures (Fratzl
x ≈ 0.2 (see, for example, Furusaka et al., and Penrose, 1997). This cluster – diffu-
1986; Osamura, 1988), but typically data sion – coagulation mechanism was origi-
extending over only 1.5 decades in time are nally proposed to explain the correspond-
available. A general problem is that some ing small values of the exponent x seen in
misfit between matrix and precipitates is Monte Carlo simulations (Bortz et al.,
present in many real alloys. Moreover, va- 1974; Marro et al., 1975). However, more
cancy concentrations may be larger than accurate simulations (Huse, 1986; Amar et
the equilibrium value after the quench and al., 1988) are rather consistent with a grad-
then decrease gradually. Both effects could ual approach to the asymptotic Lifshitz –
affect the exponent in real alloys. Katano Slyozov – Wagner (1961) law without an
and Iizumi (1984), Furusaka et al. (1988) intermediate regime characterized by a
and Forouhi and De Fontaine (1987) found well-defined different exponent x. In fact,
a regime of 1.5 decades in time where the data can usually be fitted to an equation
x = 1/6 whereas after a rather sharp cross- derived by Huse (1986):
over x = 1/3 at later times. Other experi-
ments (e.g., Morii et al., 1988) where the km–1 (t) = A + B t 1/3 (6-84 a)
km–1 (t) vs. t relationship is also measured while the original treatment of Lifshitz and
over about three decades in time find a Slyozov (1961) yielded
much smoother crossover from the initial
stages, where the log – log plot of km–1 (t) vs. km–3 (t) = A¢ + B¢ t (6-84 b)
where A, B, A¢, and B¢ are constants. Eq. (6-

84 a) can also be interpreted by an effective
exponent in the growth law,
Xeff (t) ∫ d {log km–1 (t)}/d (log t)

= –13 [1 – A km (t)]
Evidence for this law is found in various

simulations (Amar et al., 1988; Gunton et
al., 1988) and some experiments (Morii et
al., 1988; Chen et al., 1988; Alkemper et
al., 1999).
There have also been numerous attempts
to characterize experimentally the scaling
function S̃ (ᒗ) in Eq. (6-67). Typical data for
Al– Zn alloys are shown in Fig. 6-26. Al-
though the agreement with Eq. (6-71) looks
impressive, the theory (Furukawa, 1984)
neglects anisotropy, and so does the data
analysis on polycrystalline samples of Ko-
mura et al. (1985). Single-crystal work (Si-
mon et al., 1984), however, shows pro-
nounced anisotropy. Anisotropy was also
seen, for example, in scattering from Ni-
based alloys as a result of the coherent
elastic misfit (Kostorz, 1988; Fährmann et
al., 1995; Sequeira et al., 1995). In early
stages of phase separation, the anisotropy
increases roughly linearly with the mean
radius of precipitates, R µ l 1/3 (Paris et al.,
1995). The reason is that the elastic misfit
energy (which favors anisotropy) is pro-
portional to the volume of the precipitate, l,
while the surface energy (which favors
round shapes) depends on l 2/3.
The scaling function has been deter-
mined experimentally for a number of al-
loys and there is a general trend that the
Figure 6-26. (a) Scattering cross-section measured
width of the scaling function decreases for neutron small-angle scattering from Al – 10 at.%
with increasing volume fraction of precipi- Zn polycrystals at 18 °C (Komura et al., 1985). (b)
tate phase. A large collection of data is Scaling plot of the data shown in (a). Full curve is a
shown in Fig. 6-27, mostly taken from the fit to Furukawa’s (1984) function, Eq. (6-71). From
Komura et al. (1985).
review by Kostorz (1991). This decrease
can be understood qualitative as follows: if
L is a typical period in the spatial arrange-
features of the linear theory can be demon-

strated quantitatively in these materials,
and thus no real evidence for the signifi-
cance of a spinodal curve is present. While
the isotropy of these systems is clearly a
simplifying feature, coupling to structural
relaxation sometimes needs to be consid-
ered (Yokota, 1978), which implies a sig-
nificant complication (Binder et al., 1986).
An investigation of late stages was
Figure 6-27. Full width at half maximum (FWHM)
of the scaling function normalized such that the posi- carried out by Craievich and Sanchez
tion of the maximum is located at 1. Black stars are (1981) for the B2O3 – PbO(A2O3 ) glass at
for Al – Ag alloys (Langmayr et al., 1992), all other T = 0.65 Tc where the critical point of un-
data points are taken from the review by Kostorz mixing is Tc ≈ 657 °C. The structure factor
(1991) and correspond to Al – Zn, Pt – Au, Cu – Mn,
S̃ (k, t) was obtained in a time range of
Fe – Cr and Al – Li. Full and broken lines indicate pre-
dictions from various models (RG: Rikvold and Gun- about 12 to 400 min, and Eq. (6-66) was
ton (1982), TEK: Tokuyama et al., (1987), FL: Fratzl found to be obeyed with km–1 (t) ~ t x,
and Lebowitz (1989)). x ≈ 0.23. The early stages were studied by
Stephenson et al. (1991) and were found to
qualitatively agree with the Cahn – Hilli-
ment of the precipitates, the maximum po- ard – Cook theory (Eq. 6-27). Qualitatively,
sition of the structure function varies as the behavior is very similar to the results
L–1. For spherical precipitates of radius R, for metallic alloys (Sec. 6.3.1) and to com-
the width of the structure function varies puter simulations.
roughly as R –1. Hence, the width of the At this point, we also mention the appli-
scaling function behaves approximately as cation of spinodal decomposition of boro-
R/L µ F –1/3 for small volume fractions F. silicate glass-forming melts to produce
A similar argument may also be developed porous Vycor glass; on cooling the melt be-
for larger volume fractions, where the mi- low its demixing temperature it decom-
crostructure corresponds to percolated do- poses into an SiO2-rich phase and a B2O3-
mains instead of isolated droplets (Fratzl, alkali-oxide-rich phase. The latter is acid
1991), leading to the prediction shown in soluble and can be leached out with suit-
Fig. 6-27 (broken line). able solvents leaving a fully penetrable mi-
croporous glass. The small-angle scattering
from such materials has been examined and
6.3.2 Glasses, Ceramics,
interpreted in the framework of the theory
and Other Solid Materials
of spinodal decomposition (Wiltzius et al.,
Early experimental data on glassy 1987).
systems such as the Na2O – SiO2 system Qualitative observations of phase separ-
have been reviewed by Jantzen and Her- ation on a local scale attributed to spinodal
man (1978). Some data have been inter- decomposition have been reported for a va-
preted in the framework of the linearized riety of glassy materials and ceramics. Ex-
theory of spinodal decomposition (e.g., amples include rapidly quenched Al2O3 –
Yokota, 1978; Acuna and Craievich, 1979; SiO2 – ZrO8 (McPherson, 1987), La – Ni – Al
Craievich and Olivieri, 1981), but not all amorphous alloys (Okamura et al., 1993),
and SiC – AlN ceramics (Kuo et al., 1987). diffusion in fluid mixtures proceeds much
The crystallization behavior of bulk metal- faster than in solids, we must consider
lic glasses, such as Zr – Ti – Cu – Ni – Be, has quenches in a narrow region just below Tc ,
been found to be strongly influenced by de- in order to take advantage of critical slow-
composition (e.g., Wang et al., 1998 b). ing down. This implies that we must al-
Of course, spinodal decomposition is ways work in the non-mean-field critical
also expected to occur in various nonmetal- region (in the phase diagram in Fig. 6-13),
lic mixed crystals, e.g., oxides such as the i.e., nonlinear effects are very strong, and
TiO2 – SnO2 system (Flevaris, 1987; Taka- so it is not expected that the linearized the-
hashi et al., 1988) and semiconducting ory of spinodal decomposition will account
GaInAsP epitaxial layers (Cherns et al., for these systems. In addition, very early
1988; Mcdevitt et al., 1992). Often these stages are not observable because the un-
situations are difficult because of lower mixing proceeds too fast. This is best seen
crystal symmetry (the SnO2 – TiO2 system when working with rescaled variables
is tetragonal) and strong elastic lattice mis- (Chou and Goldburg, 1979, 1981), defining
fit effects (Flevaris, 1987).
qm (t ) ∫ km x (t = 0)
and
6.3.3 Fluid Mixtures
t = t D (t = 0)/ x 2 (t = 0)
Binary fluids containing small mole-
cules such as lutidine – water (Goldburg et where x (t = 0) and D (t = 0) are the correla-
al., 1978; Goldburg, 1981, 1983; Chou and tion length and the interdiffusion constant
Goldburg, 1979, 1981) and isobutyric in the initial state, respectively. Early times
acid – water (Wong and Knobler, 1978, then mean a t of the order of unity. Fig. 6-
1979, 1981; Chou and Goldburg, 1979, 28 shows that only scaled times t Ê6 are
1981) are classic systems where nonlinear accessible, whereas earlier times are ac-
spinodal decomposition was observed via cessible for fluid polymer mixtures (see
light scattering techniques. Since the inter- Sec. 6.3.4). The growth rate exponent x
Figure 6-28. Log – log

plot of qm vs. t found in
the polymer mixture poly-
styrene (PS) – poly(vinyl
methyl ether) (PVME)
(data labelled system K),
compared with isobutyric
acid – water (l) and 2,6-luti-
dine – water mixtures (m).
Systems (K, m) are data
from Chou and Goldburg
(1979). From Snyder and
Meakin (1983).
(qm–1
(t ) ~ t x ) crosses over from x near methanol and partially deuterated cyclo-
x ≈ 1/3 for t = 10 to x ≈ 1 at very late times, hexane, a perfect density matching at Tc is
as expected (Siggia, 1979) (see Eq. (6-70)). possible; see Houessou et al., (1985)) or
–1
The behavior qm (t ) ~ t –1/3, which for off- space experiments are performed, where
critical quenches where the structure corre- the gravitational effect can be reduced by a
sponds to well separated clusters is be- factor of 104. Beysens et al. (1988) showed
lieved to be the true asymptotic behavior that such experiments agree with the re-
for t Æ ∞ (Siggia, 1979), is attributed to a sults obtained from isodensity systems. By
cluster diffusion and coagulation mecha- such means not only can the accuracy of
nism (Binder and Stauffer, 1974; Siggia, data of the sort shown in Fig. 6-28 be sub-
1979). Note that for fluid droplets the stantially improved, but several decades
diffusion constants decrease in proportion can be added to a plot of qm vs. t (Fig. 6-
to their radii with increasing droplet size, 29). The full curve in this plot is the func-
due to Stokes’ law, and hence the cluster tion (Eq. (6-73)) proposed by Furukawa
diffusion – coagulation mechanism never (1984), which reduces to Eq. (6-70) for
becomes negligible in comparison with large t , i.e., the result of Siggia (1979).
the Lifshitz – Slyozov (1961) evaporation – More recently, a growth of the mean drop-
condensation mechanism. let size with the power-law exponent 1/3
Fig. 6-28 shows, however, that the cross- could be followed over more than seven
over to the asymptotic power law is grad- decades in a microgravity experiment (Per-
ual. This smooth behavior is approximately rot et al., 1994).
described by the nonlinear theory of spino- Interesting extensions involve spinodal
dal decomposition of Kawasaki and Ohta decomposition in fluid mixtures under
(1978). However, this theory can give only weak steady-state shear (Chan et al., 1988;
a poor account of the full structure func- Krall et al., 1992; Lauger et al., 1995;
tion. This is not surprising, since it is only Hashimoto et al., 1995; Hobbie et al.,
expected to be accurate for t Á10 (see Sec. 1996) or periodically applied shear (Bey-
6.2.6). Again, the problem of calculating sens and Perrot, 1984; Joshua et al., 1985),
the scaled structure function S̃ (ᒗ) (Eq. (6- as well as in strongly stirred mixtures
67)) arises and is as difficult as in the case where a turbulent suppression of spinodal
of solids (Furukawa, 1985 a). decomposition occurs (Pine et al., 1984;
A problem for binary fluids is the effect Chan et al., 1987; Easwar, 1992). Many of
of gravity. Since the two coexisting phases these observations can be explained on the
usually differ in density, one phase must go basis of the theoretical considerations of
to the top and the other to the bottom of the Onuki (1984, 1986 a, 1989 c).
container. This effect (for a more detailed Finally, we draw attention to phase sep-
discussion see Beysens et al., (1988)) im- aration phenomena in more exotic fluids
plies a smearing of the critical region for such as surfactant micellar solutions (Wil-
the unmixing critical point of a sample of coxon et al., 1988, 1995), polymer solu-
finite height, and also affects the late stages tions near their Q -point, e.g., polystyrene
of phase separation (where even hydrody- in methyl acetate (Chu, 1988), gel – gel
namic instabilities may set in; see Chan transitions (such a transition occurs, for ex-
and Goldburg (1987)). The effect of gravity, ample, in N-isopropylacrylamide gel with
however, can be strongly reduced if iso- water as a solvent, where this system phase
density systems are used (using mixtures of separates into two states with different sol-
Figure 6-29. Behavior of the scaled wavevector qm (t ) versus scaled time t . Full curve is from Furukawa
(1985), data points refer to mixtures of methanol and cyclohexane in space-flight experiments (F) or to isoden-
sity mixtures of methanol and partially deuterated cyclohexane for various quench depths, as indicated. The
late-time behavior, where qm (t ) ~ t –1 is observed, is magnified in the inset. From Beysens et al. (1988).
vent concentrations and hence a different spinodal curve can be defined; (iii) chang-
swelling ratio of the gel; see Hirotsu and ing the molecular weights while other
Kaneki (1988)), and lipid membranes with parameters (in particular, intermolecular
dissolved protein. forces!) remain constant allows a more
stringent test of theories than for other
systems. For example, it has been possible
6.3.4 Polymer Mixtures
recently to obtain by laser scanning confo-
As pointed out in Secs. 6.2.4 and 6.2.5, cal microscopy (Jinnai et al., 1997) three-
mixtures of long flexible macromolecules dimensional images of phase-separated
are particularly well suited model systems polymer blends that look very similar to
for the study of phase separation kinetics: pictures obtained from theory (Fig. 6.30).
(i) owing to the mutual entanglement of the Therefore, it is gratifying that quantita-
random polymer coils, their interdiffusion tive agreement with the linear theory of
is very slow for high molecular weights, so spinodal decomposition was obtained in
the early stages can be studied; (ii) polymer careful experiments with various poly-
mixtures exhibit a well-defined mean-field mer blends, e.g., polystyrene – poly(vinyl
critical regime, where the linear theory of methyl ether) (Okada and Han, 1986; Han
spinodal decomposition should hold, and a et al., 1988; Sato and Han, 1988), polybu-
Figure 6-30. Three-dimensional representation of a

bicontinuous phase-separated polymer blend. Inves-
tigation of the blend by laser scanning confocal mi-
croscopy gives results very similar to numerical solu-
tions of the nonlinear theory of spinodal decomposi-
tion (Jinnai et al., 1997).
tadiene and styrene – butadiene random co-

polymer mixtures (Hashimoto and Izumi-
tani, 1985), and mixtures of deuterated and
protonated polybuta-1,4-diene (Bates and
Wiltzius, 1989), among others. As an ex-
ample, Fig. 6-31 now provides an experi-
mental counterpart to Figs. 6-15, 6-16,
which were obtained from simulation. Note
that here Eq. (6-23) is used, i.e., fluctua-
tions in the final state are neglected – there-
fore the Cahn plot bends over for large Figure 6-31. (a) Early stage growth of concentration
wavevectors, and R (q)/q 2 never becomes fluctuations in a mixture of protonated and deuter-
ated polybuta-1,4-diene, with polymerization index
negative, owing to the second term on the NH = 3180, ND = 3550, at critical volume fraction
right-hand side of Eq. (6-29). Moreover, by f D = 0.486 and T = 322 K (Tc = 334.5 K). Four repre-
carrying out quenches to different tempera- sentative scattering vectors are shown. From the
2 straight-line fits to these semi-logarithmic plots the
tures it is shown that both qm (0) and the ef-
fective diffusion constant D0 vary linearly amplification factor R (q) is extracted, which is
shown as a Cahn plot in (b). From Bates and Wiltzius
with (1 – T /Tc ), as expected from Sec. 6.2.5. (1989).
The onset of nonlinear effects can be iden-
tified, but it is not accurately described by
LBM-type theories (Sato and Han, 1988).
At intermediate time a behavior sharp interfaces between the single-phase

km–1 (t) ~ t x is observed, where x varies with domains. The exponent x = 4 at small ᒗ has
quench depth. This behavior is not under- been proposed by Yeung (1988) from an
stood theoretically. In this regime of times, analysis of the Cahn – Hilliard equation. Up
nonlinear effects gradually become impor- to now, there is no theoretical expression
tant. A transition stage then occurs, before for the scaling function that is capable of
the full scaling behavior of the structure reproducing all the features in Fig. 6-33.
factor is observed in the final stage. The The model of Fratzl and Lebowitz (1989)
scaling plot for qm (t ) vs. t is presented in uses the correct exponents at low ᒗ and at
Fig. 6-32, and the scaling function for the large ᒗ and predicts a volume fraction de-
structure factor is shown in Fig. 6-33. It is
seen that Furukawa’s (1984) function gives
a reasonable overall account of the data.
The fact that the curves for cyclohexane
methanol and for poly(vinyl methyl
ether) – polystyrene (PVME – PS) mixtures
are offset is not considered to be very seri-
ous, because there are some uncertainties
involved in the rescaling of k and t in order
to obtain the scaled variables q and t ; only
in terms of the scaled variables does it
make sense to compare different materials
with each other and to discuss whether a
truly universal behavior exists.
A possible interpretation of the devia-
tions from universality in the “intermedi-
ate” and “transition” stage in Fig. 6-32 is
the fact that Bates and Wiltzius (1989) con-
sider deep quenches, for which 1 – T /Tc is
no longer very small. Only for 1 – T /Tc  1
can we expect a truly universal behavior. In
any case, comparison of Fig. 6-29 and Fig.
Figure 6-32. Log – log plot of qm (t– ) versus scaled
6-32 with Fig. 6-28 clearly demonstrates time t– for the polymer mixtures in Fig. 6-28 (dif-
the experimental progress made during less ferent symbols represent different samples) at
than a decade; the accuracy of the data in T = 298 K. Solid curve is a fit of Furukawa’s (1984)
Fig. 6-32 is substantially better and the function, Eq. (6-73), to the data in the “final stage
range of t is nearly one decade larger. region”. The poly(vinyl methyl ether) – polystyrene
(PVME – PS) curve represents results of Hashimoto
Figure 6-33 shows the behavior of the et al. (1986), and the curve for the cyclohex-
scaling function at small and at large ᒗ for a ane – methanol small-molecule mixture is taken from
polymer blend (a) and for a metal alloy (b). data of Guenoun et al. (1987). The crossover time
This behavior is of the form ᒗ x with x = – 4 from the intermediate stage to the transition stage is
at large ᒗ. At small ᒗ, x is close to 4 in both denoted as t–s , from the transition stage to the final
stage t– w . Note that the scaled time t– for the small-
polymers and metal alloys. The decay at molecule mixtures is defined differently, since
large ᒗ follows directly from Porod’s law –
Deff = D0 is not well defined. From Bates and Wilt-
(Porod, 1951) which is a consequence of zius (1989).
pendence in reasonable agreement with ex-

periments (Fig. 6-27), but it fails to repro-
duce the shoulder in the scaling function
appearing on the right side of the maximum
for polymers (Fig. 6-33 a) and even some
metal alloys (Fig. 6-33 b). Earlier theories
also fail to produce this shoulder, except
for the approach by Ohta and Nozaki
(1989), which predicts a function qualita-
tively similar to the data in Fig. 6-33 a, but
is still quantitatively off.
6.4 Extensions
Basic aspects of the theory of spinodal
decomposition were treated in Sec. 6.2.
Only the generic phase diagram (Figs. 6-
3 b, 6-13, and 6-21) of a binary mixture
with a miscibility gap has been considered.
In this section, we briefly mention the re-
lated phenomenon of ordering kinetics
(Fig. 6-3 a) and also discuss cases where
formation of order and unmixing compete.
The effects of certain complications (im-
purities, effects of finite size and free sur-
faces) are briefly assessed.
6.4.1 Systems Near a Tricritical Point

If a binary system exhibits a line of crit-
ical points Tc (c– ) for an order – disorder
Figure 6-33. (a) Log – log plot for the structure-fac-
transition, a tricritical point Tt (c–t ) may oc-
tor scaling function in the late stages of a quench to cur where this critical line stops, and below
313 K for the polymer mixture in Fig. 6-31. The solid this tricritical point both ordering and
curve is the theoretical prediction of Ohta and No- phase separations occur simultaneously
zaki (1989). The straight line I (q) ~ q – 4 demonstrates (Griffiths, 1970; Lawrie and Sarbach,
the validity of Porod’s law. The inset shows low-q
behavior, as measured in the intermediate stage,
1984). Two order parameters, y (x) and
where a power law I (q) ~ q 4.5 is found. From Bates c (x), are simultaneously needed to de-
and Wiltzius (1989). (b) Similar plot for the scaling scribe such phenomena. Examples of this
function in the metallic alloy Al – 15 at.% Ag (Lang- situation occur in some magnetic alloys
mayr et al., 1992). The full line corresponds to the where the order – disorder transition de-
model function of Fratzl and Lebowitz (1989) for a
–
volume fraction of F = 0.27. This function S (ᒗ) in-
scribes a magnetic ordering, 3He – 4He mix-
creases with ᒗ at small ᒗ and decreases with ᒗ– 4 at
4 tures (here y describes the superfluid order
large ᒗ (Porod’s law). parameter and hence must be taken as a
6.4 Extensions 461
complex variable), and also crystallo- from some reference value that fixes the
graphic order – disorder phase transitions in constant D 0 , and m 3 and m 4 are the chemi-
alloys and in adsorbed layers at surfaces cal potentials of 3He and 4He, respectively.
are known to exhibit such tricritical points. For fixed c, T, and D = m 3 – m 4 + D 0 the
In alloys, the first discussions of phase sep- static equilibrium follows from seeking the
aration in tricritical systems come from Al- minimum of
len and Cahn (1976, 1979 a, b) in the con- (6-86)
1 1 1 1
text of Fe – Al alloys. An elegant extension F = r˜ |y |2 + u˜ |y |4 + v |y |6 + c n D2
of the linearized theory of spinodal decom- 2 4 6 2
position to tricritical 3He – 4He mixtures where r̃ = r0 + 2 g Dc n , ũ = u – 2 c n g 2, and
was given by Hohenberg and Nelson c = c n [D – g | y |2 ] was eliminated from the
(1979). Nonlinear effects have mostly been equation. For ũ > 0 there is a second-order
studied for the model with simple relaxa- transition from normal fluid to superfluid
tional dynamics of the order parameter at r̃ = 0, whereas for ũ < 0 the transition is
(Dee et al., 1981; San Miguel et al., 1981). of first order. The critical point occurs for
Here we briefly mention the case of ũ = 0.
3
He – 4He mixtures, since the most detailed Whereas for the binary mixture a single
experiments exist for these systems (Hoffer Langevin equation for the concentration
et al., 1980; Sinha and Hoffer, 1981; Benda field c (x, t) and y (x, t) had to be derived
et al. 1981, 1982; Alpern et al., 1982). (Eq. (6-15)), here we need two equations.
3
He – 4He mixtures have no practical appli- In fact, the entropy density S (x, t) must
cations in materials science, but we men- also be added (Hohenberg and Nelson,
tion them here as a model system for the 1979), and thus the calculation becomes
study of phase separation near a tricriti- complicated. However, since y is not a con-
cal point. The concepts developed (and served quantity, it relaxes much more rap-
checked experimentally) for this model idly, hence the assumption is made that it al-
system can be carried over to more com- ways adjusts to the local equilibrium corre-
plex materials such as ternary mixtures, sponding to the local concentration c (x, t).
magnetic alloys, etc. Instead of Eqs. (6-4) Instead of the simple exponential function
and (6-8), the expression for the free en- in Eq. (6-23), additional oscillating terms
ergy functional is now due to the “second-sound” mode are found:
DᏲ
= ∫ dx ⎧⎨ r0 |y |2 + u |y |4
2
1 1 S (k, t) = a1 e2 R (k) t + a2 e[R (k) – D2 k ]t
(6-87)
kB T ⎩2 4 ¥ cos (u2 k t) + a3 e
2
– 2 D2 k t
cos (2 u2 k t) + a0
1 1
+ v |y |6 + r 2 | ∇y ( x )|2 where a0 , a1 , a2 , and a3 are constants, u2 is
6 2
1 −1 2 the second-sound velocity and D2 its damp-
+ c n c + g c |y |2 − (m 3 − m 4 + D0 ) c ing coefficient. In the spinodal region, R (k)
2
is positive, and not too close to the tricriti-
+ l02 | ∇c ( x )|2 ⎫⎬
1
(6-85) cal point for small k, R (k) – D2 k2 should
2 ⎭
also be positive. Therefore in addition to
where r0 , u, v, c n–1, l0 , and g are pheno- the first term on the right-hand side of Eq.
menological coefficients that should de- (6-87), which is the analog of Eq. (6-23),
pend only on temperature and not on the lo- there is another exponentially growing
cal 3He concentration c, which is measured term oscillating in time. This would imply
a “flickering” component in the scattering. driving force leading to coarsening of the

However, the linearized theory of spinodal domain structure, similar to the coarsening
decomposition is not expected to hold for discussed for spinodal decomposition (Sec.
3
He – 4He mixtures, since neither of the 6.2.9).
interaction ranges r and l0 in Eq. (6-85) are The description of the initial stages of
expected to be large. Note, however, that this domain growth process closely par-
the Ginzburg (1960) criterion for tricritical allels the treatment presented in Eqs. (6-4)
systems shows that mean-field theory is es- to (6-26). Let us assume a free energy func-
sentially correct for tricritical points; hence tional similar to Eq. (6-85), with pheno-
nonlinear effects do not become worse menological constants r0 , u, v, and r:
when T Æ Tt , and unlike in Eqs. (6-51) and
DᏲ
= ∫ dx ⎧⎨ r0 y 2 + uy 4
(6-52), we expect no factor depending on 1 1
(6-88)
the temperature distance from Tt under the kB T ⎩2 4
conditions of validity of the linear theory.
+ v y 6 + r 2 [ ∇y ( x )]2 ⎫⎬
1 1
For tricritical systems, both experiments 6 2 ⎭
and simulations (Sahni and Gunton, 1980;
Sahni et al., 1982; Ohta et al., 1988; Sagui and we emphasize that the order parameter
et al., 1994; Wang et al., 1998 a; Gorents- y is not a conserved quantity. Thus instead
veig et al., 1997; Nielaba et al., 1999) have of Eqs. (6-10) and (6-11) we write:
shown a nonlinear behavior for all times ∂y ( x, t )
accessible to study, and a scaling behavior = − G m˜ ( x, t )
∂t
of the structure factor as in Eq. (6-67) again
m˜ ( x, t ) = dDᏲ (y ( x, t ))/ dy ( x, t ) (6-89)
applies to the late stages.
describing a simple relaxational approach
6.4.2 Spontaneous Growth of Ordered towards equilibrium, with G being an ap-
Domains out of Initially Disordered propriate rate constant. Again, fluctuations
Configurations must be added to this equation as in Eqs.
(6-14) to (6-17), so that the final result is
There are many examples where the dy-
namics of a phase change involve the spon- ∂y ( x, t ) ⎧⎪ ⎡ ∂ fcg [y ( x, t )] ⎤
taneous formation of ordered structures. =G ⎨⎢ ⎥ (6-90 )
∂t ⎩⎪ ⎣ ∂y ⎦T
Suppose an alloy such as b-CuZn, which is
a prototype realization of Ising-model type }
− r 2 kB T ∇ 2 y ( x , t ) + h T ( x , t )
ordering (Als-Nielsen, 1976), is quenched
from T0 > Tc to T < Tc (Fig. 6-3 a). Then where the random force h T (x, t) now satis-
again the initial homogeneously disordered fies the relationship
state is thermodynamically unstable, and
·h T (x, t) h T (x¢, t¢)Ò
we expect that ordered domains will form.
Since no sign of the order parameter is pre- = ·h T2 ÒT d(x – x¢) d(t – t¢)
ferred, and there is no symmetry-breaking ·h T2 ÒT = 2 kB T G (6-91)
–
field, the average order parameter y in the
–
system, y = (1/V ) ∫ dx y (x, t), will remain Again the first step of the analysis is a lin-
–
y = 0: domains with both signs of the order earization approximation, similar to Eqs.
parameter form equally often. Again, the (6-18) to (6-26). For the structure factor,
(unfavorable) domain wall energy acts as a which we now define as follows,
6.4 Extensions 463
S (k, t) = ·d y– k (t) d y k (t)Ò Fontaine and Kikuchi, 1978) to the meso-

d y k (t) ∫ ∫ dx [exp (i k · x)] y (x, t) (6-92) phase separation transition of block copol-
ymers (Fredrickson and Binder, 1989). It is
the result is analogous to Eq. (6-23) (De important to realize, however, that the “or-
Fontaine and Cook, 1971): dering spinodal” is also a mean-field con-
cept, suffering from the same “ill-defined-
S (k, t) = ST0 (k) exp [2 R (k) t] (6-93)
ness” as the unmixing spinodal. Taking
with into account both statistical fluctuations
(6-94)
⎡⎛ ∂ 2 f ⎞ ⎤ and nonlinear effects, as they are both con-
R ( k ) ≡ − G ⎢⎜ 2⎥ tained in Eqs. (6-90) and (6-91), we again
cg
+ r 2
k T k
⎢⎝ ∂y 2 ⎟⎠ T , y = 0
B
⎥ expect to find a smooth transition between
⎣ ⎦
the nucleation of order and the “spinodal
Note that k in Eqs. (6-92) to (6-94) de- ordering” mechanism (Cook et al., 1969),
scribes the distance in reciprocal space as described by Eqs. (6-93) and (6-94). In
from the superstructure Bragg spot Q B cor- addition, nonlinear effects will limit the
responding to the considered ordering. predicted exponential growth, as they do in
Suppose now fcg has the form assumed the case of spinodal decomposition.
in Eq. (6-88), with u > 0, v > 0, and This similarity of the behavior also car-
r0 = r¢(T – T0 ) changes sign at T0 (this is the ries over to the late stages of growth, where
standard Landau description of a second- a scaling similar to Eq. (6-67) applies: but
order transition; see Stanley (1971) or now km (t) denotes the half-width of the
Binder (2000)). Then peak (unlike spinodal decomposition, the
– – maximum growth always occurs at the
(∂2 fcg /∂y 2 )y– = r0 + 12 u y 2 + 30 v y 4
Bragg position Q B describing the order,
–
is equal to r0 < 0 for y = 0, i.e., the initial i.e., k ∫ 0). Since the order parameter y in
configuration is always unstable, and all Eq. (6-90) is not conserved, a faster growth
modes y k (t) grow in the interval from law than Eq. (6-68) is predicted (Lifshitz,
0 < k < kc , with kc still being given by Eq. 1962; Allen and Cahn, 1979 b; Ohta et al.,
(6-25). However, Eq. (6-26) does not hold 1982; Bray, 1994):
here and the maximum growth rate occurs
km–1 (t) ~ t 1/2 (6-95)
for k = 0, as is obvious from Eq. (6-94).
In a case where u < 0, however, the phase In a binary alloy undergoing an order – dis-
transition does not occur for T = T0 where order transition, or in a layer adsorbed at a
r0 changes sign, but rather at a higher surface at constant coverage, the conserva-
temperature Tc = T0 + 3 u2/32 r¢ v (Binder, tion of concentration (or density, respec-
–
1987 a). For T0 < T < Tc the state with y = 0 tively) may change the growth law (Eq. (6-
is metastable, and the ordering reaction 95)): Sadiq and Binder (1984) suggested
again needs a nucleation process (Chan, that the excess concentration (or excess
1977; Fredrickson and Binder, 1989). The density, respectively) contained in certain
temperature T0 hence again plays the role types of domain walls could again lead to a
of a spinodal point (“ordering spinodal”; mechanism of the Lifshitz – Slyozov type,
De Fontaine, 1979). Such spinodals for i.e., Eq. (6-68). The validity of this sugges-
thermally driven first-order transitions tion is not yet confirmed (Binder and Heer-
have been calculated for diverse systems, mann, 1985; Binder et al., 1987). Figs.
from alloys such as Cu – Au systems (De 6-34 and 6-35 give some examples of the
behavior found in computer simulations of

a square lattice model of a two-dimen-
sional alloy, where there is a repulsive
interaction between AB pairs on both near-
est- and next-nearest-neighbor sites. For
c– = 0.5 the ordering is the two-component
(2 ¥1) structure: there are four kinds of do-
mains, where in the ground state full rows
of A atoms alternate with full rows of B
atoms, and these rows may be oriented in
either the x-direction or y-direction for a
square lattice. Fig. 6-34 shows the growth
of these ordered domains in the simulation
of a quenching experiment for this model,
and Fig. 6-35 the resulting growth of the
superstructure Bragg intensity. The scaling
behavior of Eq. (6-67) is well verified.
Similar results have also been found for
simulations of various other models (see
Binder et al., (1987), Furukawa (1985 a),
Bray (1994), for review.
Experiments on the kinetics of the for-
mation of ordered superstructures have
been carried out both for two-dimensional
monolayers adsorbed at surfaces (Wang
and Lu, 1983; Wu et al., 1983; 1989; Trin-
gides et al., 1986, 1987; Henzler and
Busch, 1990), three-dimensional metallic
alloys (e.g., Hashimoto et al., 1978; Nishi-
hara et al., 1982; Noda et al., 1984; Takeda
et al., 1987; Katano and Iizumi, 1988; Ko-
nishi and Noda, 1988), dielectric materials
such as K2Ba(NO2 )4 (Noda, 1988) or
K2ZnCl 4 (Mashiyama, 1988), martensitic
materials such as KD3(SeO3 )2 (Yagi and Lu,
Figure 6-34. “Snapshot pictures” of the computer simulation of the ordering process of a 120 ¥ 120 square lat-
tice model of an alloy with repulsive interactions between A – B pairs on nearest- and next-nearest-neighbor
sites, e nn = e nnn . A quench from a disordered configuration at infinite temperature to kB T/e nn = 1.33 is performed
for c– = 0.5 (a second-order transition to the (2 ¥ 1) structure occurs in equilibrium at kB Tc /e nn ≈ 2.1). Time evo-
lution occurs via random nearest-neighbor A – B exchanges (i.e., the model of Kawasaki (1972)). Times shown
refer to 10 MCS after the quench (upper part), 200 MCS (middle) and 1700 MCS (bottom). Only B atoms are
shown (using four different symbols to indicate whether a B atom belongs to a domain of type 1, 2, 3, or 4, see
Fig. 6-35 a), A atoms are not shown. From Sadiq and Binder (1984).
6.4 Extensions 465
1988), graphite intercalation compounds

such as transition metal chlorides in graph-
ite (Matsuura, 1988), molecular crystals
such as CN1–x Clx -adamantane (Descamps
and Caucheteux, 1988), etc. Of course,
there are also cases where phase separation
and ordering occur simultaneously, a clas-
sic example being Fe – Al alloys where
Fe3 Al domains precipitate (Allen, 1977;
Oki et al., 1973; Eguchi et al., 1984).
Again, no attempt at completeness is made
here, we simply quote a number of exam-
ples to emphasize the universality of the
phenomenon.
As expected from the above discussion,
we must distinguish the cases where the
“parent phase” from which the ordered
structure forms is metastable or unstable;
in the latter case, the structure forms by
“spinodal ordering” (De Fontaine, 1979)
whereas in the former case it forms by nu-
cleation and growth, although again this
distinction is not very sharp. An example
of the latter case is Mg3In, where the order
is of Cu3Au type but the transition is very
strongly of first order (Konishi and Noda,
1988). The volume fraction of the ordered
phase is then found to obey the well-known
Avrami (1939), Johnson – Mehl (1939)
equation
F (t) = 1 – exp [– (t /t )n ] (6-96)
where n is a constant characteristic of the
type of nucleation process and t a time
constant that depends on the degree of
Figure 6-35. (a) Four types of domains (1, 2, 3, and 4) in the (2 ¥ 1) structure of a binary alloy AB on the square
lattice (B atoms are indicated by black circles and A atoms by white circles). (b) Structure factor S (q, t) of
superlattice Bragg scattering, at times t after the quench. Since the lattice spacing is set to unity, the Bragg po-
sitions of the (2 ¥1) structure are (±p, 0) and (0, ± p). Only the variation with qx near qx = p is shown. Times are
measured in units of Monte Carlo steps per site. Data are for a computer simulation of a quench from T = • to
kB T/e nn = 1.33 as in Fig. 6-34. (c) Scaling plot of the data shown in (b), with the structure factor S (q, t) being
rescaled with the maximum intensity S (p, t) and the relative distance q/p – 1 from the Bragg position being re-
scaled with the half-widths s (t) of the peaks in (b). From Sadiq and Binder (1984).
undercooling of the transition. Here we are find a law km–1 (t) ~ t 1/4 (Noda, 1988), for
more interested in the “spinodal ordering” Ni3Mn a crossover from km–1 (t) ~ t 1/4 at
as occurs in Cu3Pd (Takeda et al., 1987), short times to km–1 (t) ~ t 1/2 at later times is
Ni3Mn (Katano and Iizumi, 1988), Cu3Au found (Katano and Iizumi, 1988), and for
(Hashimoto et al., 1978; Nishihara et al., AuCu3 a law km–1 (t) ~ t 1/2 (i.e., Eq. (6-95))
1982; Noda et al., 1984), and K2Ba(NO2 )4 is established (Noda et al., 1984). Whereas
(Noda, 1988). the t 1/2 law is expected from various theo-
Fig. 6-36 shows examples of growing ries (Lifshitz, 1962; Allen and Cahn,
Bragg peaks, i.e. the experimental counter- 1979 b; Ohta et al., 1982), a t 1/4 law occur-
parts to simulation data such as shown ring over a transient period of time can per-
in Fig. 6-35. Whereas for K2Ba(NO2 )4 we haps be attributed to the “softness” of the
domain walls between the growing ordered
regions (Mouritsen, 1986; Mouritsen and
Praestgaard, 1988). An interesting obser-
vation of Lifshitz – Slyozov type coarsen-
ing of the structure factor is also reported
for colloid crystallization (Schätzel and
Ackerson, 1993).
In the two dimensional case, Wu et al.
(1989) measured a growth exponent x =
0.28 ± 0.05 for O on W (110). Since the
p (2 ¥1) phase is believed to have an eight-
fold ground-state degeneracy in this sys-
tem, whereas the theories mentioned above
refer to a two-fold degenerate ordering
only, the interpretation of this result is not
obvious. We note, however, that reasonable
scaling of the structure factor is observed
(Fig. 6-37). For the system silver on Ge
(111), a growth exponent x = 1/2 is found
(Henzler and Busch, 1990) and also good
scaling of the structure factor is seen. In
this system, on the other hand, a crossover
to slower growth occurs at later times; the
reason for this behavior is not clear – per-
haps it is due to pinning of domain walls at
defects (see Sec. 6.4.4). At this point, we
note that a similar slowing down of the
growth where km–1 (t) basically stops grow-
ing further, has also been seen in Ni3Mn
(Katano and Iizumi, 1988) and in the phase
Figure 6-36. (a) Time evolution of the 211 super-
separation of mixtures of flexible and
lattice peak of Ni3Mn annealed at 470 °C for times
up to 34 h. From Katano and Iizumi (1988). (b) Time semi-rigid polymers (Hasegawa et al.,
evolution of the (–12 , x , –52 ) superlattice peak of 1988). In neither of these systems is the
K2Ba(NO2 )4 annealed at 190 K. From Noda (1988). very slow growth at late stages understood.
6.4 Extensions 467
6.4.3 Phase Separation in Reduced

Geometry and Near Surfaces
Various cases of reduced geometry may

be of interest; for example, spinodal de-
composition of fluid mixtures confined to
pores in porous media like Vycor glass, or
spinodal decomposition or ordering of thin
films or adsorbed monolayers at surfaces,
and phase separation in small particles or
grains in inhomogeneous materials. All
these problems have seen much recent ac-
tivity, and lack of space allows us to give
only a very brief introduction.
In a porous medium (or gel), a binary
Figure 6-37. (a) Dynamic scaling in the growth of
mixture at critical composition experiences
Bragg peaks for W (110) p (2 ¥1) – O at coverage a random chemical potential difference (De
F = 0.5 and T = 297 K. From Wu et al. (1989). (b) Gennes, 1984). Using concepts from the
-
Scaled LEED intensities of the ÷3 peak of Ag ad- random field Ising model (Villain, 1985;
sorbed on Ge (111) at T = 111 °C. From Henzler and Huse, 1987) we expect both a change in the
Busch (1990).
phase diagram and slow relaxation (Gold-
burg, 1988; Goldburg et al., 1995; Falicov
In domain growth studies of O on W (112) and Berker, 1995). However, there is also
Zuo et al. (1989) also found scaling behav- the view that we must rather focus on the
ior at early times but slowing of domain wetting behavior on the walls inside a
growth attributed to random field effects at straight cylindrical pore (Liu et al., 1990;
later times. Monette et al., 1992; Tanaka, 1993; Zhang
and Chakrabarti, 1994, 1995). Of course, 6.4.4 Effects of Quenched Impurities;

ultimately we should combine this single- Vacancies; and Electrical Resistivity
pore behavior with effects due to the ran- of Metallic Alloys Undergoing Phase
dom structure of the pore network. Changes
From Sec. 6.2 it should be clear that the In this subsection, we draw attention to a
behavior in bulk two and bulk three dimen- variety of topics that cannot be discussed
sions is qualitatively similar, although non- here in any depth, owing to lack of space.
linear phenomena are usually more impor-
tant for d = 2 than for d = 3. In particular, (i) Quenched Impurities
for polymer mixtures different coils can Throughout this chapter, only ideal
interpenetrate each other for d = 3 but not systems have been considered where phase
for d = 2; therefore a mean-field critical re- separation is triggered by spontaneous
gime does not exist for two-dimensional fluctuations. However, real materials al-
polymer mixtures. ways contain impurities, whereas in fluids
For thin films that are many atomic di- these impurities are mobile (“annealed
ameters thick, there is an interesting inter- defects”) and hence act like a dilute addi-
play between finite size effects (Binder et tional component of a multi-component
al., 1987) and surface effects. Usually al- mixture, in solids such impurities are often
ready in equilibrium at the surface of a immobile (“quenched defects”) at the tem-
mixture, the concentration of one compo- peratures of interest. This frozen-in disor-
nent of a mixture will be enhanced in com- der may have two effects: in a metastable
parison with the bulk concentration (“sur- region, free energy barriers for nucleation
face enrichment” occurs because the inter- are reduced and hence heterogeneous nu-
actions between a wall and the two compo- cleation is facilitated (Zettlemoyer, 1969).
nents A and B differ from each other). At In the late stages of spinodal decomposi-
the coexistence curve, this preference of tion or domain growth, there is an impor-
the surface for, say, the B component may tant interaction between such defects and
lead to the formation of a wetting layer at domain walls. Whereas so far it has been
the surface (Dietrich, 1988). Inside the assumed that domain walls may diffuse
two-phase coexistence region, surfaces freely owing to their capillary wave excita-
may hence have a profound effect on the tions, the randomly spaced defects act like
kinetics of phase separation: surface-di- a random potential in which the interface
rected concentration waves (with wavevec- moves. Since thermal activation is now
tor normal to the surface plane) dominate required to overcome the barriers of this
the initial growth in the region near the sur- potential, at low temperatures the domain
face (first observed by Jones et al. (1991) wall motion is dramatically slowed owing
for a polymer mixture). In later stages of to such impurities. Simulations of this
the coarsening, we have an interesting problem have been carried out by Grest and
competition between the domain growth in Srolovitz (1985) and Srolovitz and Grest
the bulk and the possible propagation of a (1985). More work has been devoted to the
wetting layer from the surface into the bulk related problem of the dynamics of the ran-
(see Krausch, 1995; Puri and Frisch, 1997; dom field Ising model (see Villain, 1985;
and Binder, 1998, for recent reviews). and Nattermann, 1998, for reviews). In all
these systems, we expect that the growth
law of the ideal system (Eqs. (6-68) or
6.4 Extensions 469
(6-95)) will hold until some characteristic A simulation of the early stages of spino-
length lc is reached, which depends on the dal decomposition in such an ABV model
concentration ci of the quenched impurities (Yaldram and Binder, 1991) shows that the
(lc ~ ci–1/d in d dimensions). At times t sig- general features of the structure factor
nificantly exceeding the time given by the S (k, t) are almost the same as those of the
condition km–1 (t) lc ≈ 1, a logarithmic growth direct-exchange AB model (Fig. 6-5), and
law (Villain, 1985) is expected: the two models can be approximately
mapped onto each other by adjusting the
km–1 (t) ~ ln t (6-97)
time scales. In real systems, however, the
A similar crossover in the relaxation from a behavior may be more complicated; the va-
fast to a slow growth law may also be cancy concentration does not need to re-
caused by interface pinning at extended de- main constant, and it may be that many va-
fects (e.g., dislocations, grain boundaries) cancies are created during the quench
rather than point defects. which later are annealed out by migration
to surfaces or recombination with intersti-
(ii) Vacancies tials. If this happens, the effective mobility
In solid mixtures, vacancies V are cru- M would itself depend on the time t after
cial for a microscopic description of inter- the quench. Also the vacancy concentra-
diffusion, which occurs via a vacancy tions in equilibrium may be different in A-
mechanism rather than by direct A – B ex- rich and B-rich domains, or may become
change (Flynn, 1972; Manning, 1986). Mi- preferentially enriched in domain boundar-
croscopically we therefore need concentra- ies. This did not occur in the simulation of
tions ciA (t), ciB (t), and ciV (t) for A atoms, B Yaldram and Binder (1991), since there the
atoms, and vacancies as dynamic variables, static properties were assumed with perfect
respectively, rather than the single concen- symmetry between A and B and there was
tration variable ci or concentration field no energy parameter associated with the
c (x, t) (Eq. (6-1)). Since ciV (t) 1, it may vacancies, but this model is still a gross
be possible to reduce the problem to Eq. over-simplification of reality. Fratzl and
(6-15), where the mobility M then needs to Penrose (1994, 1997) found that vacancies
be related to the jump rates GA and GB of may speed up the coarsening by changing
A and B atoms to vacant sites (and the va- the mechanism. Note that for very low va-
cancy concentration). However, owing to cancy content but very late stages a faster
correlation effects in the vacancy motion, growth law (x = 1/2) due to cluster – cluster
this is a difficult problem even in the non- aggregation has been suggested (Mukher-
interacting case (Kehr et al., 1989). There jee and Cooper, 1998). These problems
Eqs. (6-6) and (6-14) would yield an inter- may confuse the proper interpretation of
diffusion coefficient Dint = M/[c (1 – c)], real experiments.
i.e., M can be found if Dint can be simply
related to GA and GB and the average va- (iii) Electrical Resistivity of Metallic
cancy concentration c– V. Although M obvi- Alloys Undergoing Phase Changes
ously is proportional to c– V for small c– V, In metallic alloys the quasi-free elec-
computer simulations (Kehr et al., 1989) trons responsible for electrical conduction
show that the simple relationships pro- are scattered from the atomic disorder. In
posed in the literature to relate M to GA and the framework of the Born approximation,
GB are inaccurate. treating the scattering as elastic, the excess
resistivity Dr due to concentration fluctua- time-dependent cluster distribution func-

tions can be represented as a convolution tion (see Sec. 6.2.5), the problem of the
of the structure factor ·d y– k (t) d y k (t)Ò and electrical resistivity can be rephrased in
the Fourier transform of the effective po- terms of electron scattering from clusters
tential that the quasi-free electrons feel, (Hillel et al., 1975; Edwards and Hillel,
namely (Binder and Stauffer, 1976 b), 1977). Experimental resistivity data for
Al – Zn alloys can be interpreted qualita-
e 2 ( ZB − ZA )2 meff
2
tively along such lines (Luiggi et al., 1980).
Dr = N
( 4 p) neff a kF
2 3 4 3
At this point, we also mention that elec-
k trical currents have been found to affect the
× ∫ dk (1 + k 2 / k 2 )2
0 < k < 2 kF phase separation behavior of Al – Si alloys
× [ 〈d c− k (t ) d ck (t )〉 − c 2 d ( k )] (6-98) (Onodera and Hirano, 1986, 1988) and
other alloys. At present, the explanation of
where e is the charge of an electron, meff its these phenomena is unclear.
effective mass, N is the number of atoms
per cm3, ZA and ZB are the atomic numbers
of the two constituents, ä is Planck’s con- 6.4.5 Further Related Phenomena
stant, and k describes the screening of the As a first point of this subsection, we
Coulomb interaction. In Eq. (6-98) it is as- draw attention to peculiar morphologies of
sumed that the Fermi sphere (whose radius structures that may form via phase separa-
is kF ) lies completely within the first Bril- tion in fluid mixtures with very strong
louin zone, and that there is no sublattice dynamic asymmetry, e.g., solution of poly-
ordering. mers which are frozen into a glass-like
Using the structure factor from the com- structure at high density. The resulting vis-
puter simulations of Marro et al. (1975), coelastic phase separation leads to the for-
Binder and Stauffer (1976 b) showed that mation of sponge-like patterns (Tanaka,
during spinodal decomposition a resistivity 1994, 1996; Taniguchi and Onuki, 1996) or
maximum occurs. This can be understood rigid foams (Hikmet et al., 1988).
since in Eq. (6-98) the main contribution Another recent topic of interest is phase
comes from the large k (k of order k or of separation induced by temperature gra-
order kF ). As the coarsening proceeds, dients (Kumaki et al., 1996) and velocity
however, the structure factor is large only gradients, i.e. shear (Onuki et al., 1997).
at much smaller values of k. Scaling con- The effect of shear flow on the unmixing of
siderations (Eq. (6-67)) suggest that fluids is in fact three-fold, as recently re-
Dr (t) – Dr (∞) ~ km (t) ~ t –1/3 viewed by Onuki (1997): the phase dia-
gram is modified (shear-induced mixing
This treatment can again be generalized for due to a depression of the critical tempera-
the kinetics of ordering alloys. We refer to ture may occur), the rheology of the un-
Binder and Stauffer (1976 a) for a discus- mixed two-phase states is profoundly al-
sion of this problem, and of previous theo- tered, and the morphology of the patterns
ries and related early experiments. that form is changed.
Just as an alternative description for con- Finally we mention the interplay of
centration inhomogeneities in terms of phase separation and chemical reactions
“concentration waves” and their mean- (Huberman, 1976; Glotzer et al., 1995; Le-
square amplitude S (k, t) in terms of the fever et al., 1995). Chemical reactions may
6.5 Discussion 471
terminate the coarsening process and thus understand the crossover from these early
freeze in an inhomogeneous pattern. stages to the late stage of phase separation
where the structure factor develops to-
wards a scaling limit. It is not clear under
which conditions power-law behavior
6.5 Discussion km (t) ~ t – x can be observed at intermediate
stages, and what the appropriate interpreta-
The understanding of the dynamics of
tion of the exponent x is.
phase changes of materials has advanced
(iv) In the late stages the structure factor
considerably in recent years; the theoreti-
obeys the scaling behavior first suggested
cal results now present a fairly clear picture
by Binder and Stauffer (1974, 1976 b),
of at least some basic questions, and also
quantitatively reliable experiments are S (k, t) ~ [km (t)] –d S̃ {k/km (t)}
available for many systems. This chapter
While the scaling function S̃ (ᒗ) proposed
has emphasized the theoretical aspects; its
by Furukawa (1984) seems to account well
main conclusions can be summarized as
for the general shape of experimental data
follows:
on solid or polymer mixtures, several de-
(i) The linearized theory of spinodal details are not reproduced correctly. Good ev-
composition holds only for mean-field type idence for Porod’s law, S̃ (ᒗ) ~ ᒗ– (d +1) for
systems (with long ranges of interactions) ᒗ  1, has been provided. The width of S̃ (ᒗ)
or in systems that are equivalent to them, increases when the volume fraction f of
such as polymer mixtures. precipitate phase decreases from that of the
(ii) Whereas a spinodal curve is well de- critical mixture (Fratzl and Lebowitz,
fined in the mean-field limit or for polymer 1989). There we find, as a function of tem-
mixtures in the limit of chain length N Æ∞, perature and f , both morphologies, inter-
in all other cases the spinodal is an ill-de- connected domains and well separated
fined concept. This implies that the transi- droplets. Such a transition can also be ob-
tion from nucleation to spinodal decompo- served as a function of time in computer
sition is gradual. For polymer mixtures, the simulations (Hayward et al., 1987; Lironis
width of this transition region may be nar- et al., 1989) and experiments on polymer
row, as estimations from Ginzburg criteria mixtures (Hasegawa et al., 1988) (see Fig.
show. Experiments are still needed to 6-38).
check the latter point, whereas the gradual (v) There is now agreement that in solid
transition from nucleation to spinodal de- mixtures the behavior at late times is given
composition in metallic alloys over a broad by the Lifshitz – Slyozov law, km–1 (t) ~ t 1/3,
temperature region has been established even in the percolative regime (Amar et al.,
experimentally. 1988). This has recently also been verified
(iii) Nonlinear spinodal decomposition in a solid – liquid system that satisfies all
during the early stages can be successfully assumptions of the LSW theory (Alkemper
described qualitatively by the Langer – et al., 1999). For critical quenches of fluid
Baron – Miller theory (for solid mixtures) mixtures at late times, Siggia’s (1979) re-
and the Kawasaki – Ohta theory (for fluid sult km–1 (t) ~ t holds, but further research is
mixtures). Recent experiments, however, needed to achieve an understanding of the
show that these theories are not quantita- exponents for the growth of ordered struc-
tively accurate. More work is required to tures in cases with a higher degeneracy of
the ordered phase, and to understand how

these growth laws are affected by defects.
Other questions concern the suggestion
(Langer et al., 1975; Binder, 1977) that by
expressing km and t as suitably scaled
quantities qm and t, e.g., qm (t ) = km (t) x
and t = D0 t/x 2, where x is the correlation
length at phase coexistence and D0 the
interdiffusion constant, a material-inde-
pendent universal function should be ob-
tained (which must depend, however, on
the volume fraction of the phases coexist-
ing in the final equilibrium state). A coun-
ter-example for this universality seems to
be the “N branching” found for polymer
mixtures (Onuki, 1986 b; Hashimoto,
1988), and more work is needed to under-
stand this problem. Clearly such a univer-
sality, if it exists, would be useful as it al-
lows the phase separation behavior of other
Figure 6-38. Polarizing optical microscope images materials to be predicted if properties such
obtained from the same area of a cast film of a binary as x and D0 are known. Of course, we must
mixture of poly(ethylene terephthalate) and a copoly- distinguish between different “universality
ester composed of 60 mol% p-oxybenzoate and
classes” (Binder, 1977) for solids and fluid
40 mol% ethylene terephthalate at 50% relative con-
centration during isothermal heat treatment at mixtures (cf. Figs. 6-28 and 6-39).
270 °C. Times at which the images are taken are indi- (vi) Apart from crude “cluster dynam-
cated (in seconds). From Hasegawa et al. (1988). ics” models, there is no theoretical ap-
Figure 6-39. Comparison of qm vs.

t behavior found in metallic alloys
(a) – (h) and in organic glasses (i), (j).
Here the scaling is done differently to
that in the text, namely qm = km (t)/km (0)
and t = – 2 t R [km (0)]. Systems used are
Au – 60 at.% Pt at 550 °C (a) (Singhal et
al., 1978), Al – 6.8 at.% Zn at T = 108 °C
(b), 116 °C (c), 129 °C (d) (Laslaz et al.,
1977), Al – 5.3 at.% Zn at T = 20 °C (e),
Al – 6.8 at.% Zn at 20 °C (f), 90 °C (g),
110 °C (h) (Hennion et al., 1982),
76 B2O3 – 19 PbO – 5 Al2O3 at 450 °C (i)
(Zarzycki and Naudin, 1969), and
13.2 Na2O – SiO2 at 560 °C (j) (Tomo-
zawa et al., 1970). From Synder and
Meakin (1983).
6.6 Acknowledgements 473
proach to describe the behavior in the tran-

sition region from spinodal decomposition
to nucleation. This problem, and theories
connecting consistently the early-time spi-
nodal decomposition to the scaling behav-
ior at late stages, still needs more research.
So far much progress has been due to large-
scale computer simulation, e.g., Amar et al.
(1988) (see Figs. 6-40 and 6-41); it is still a
challenge to explain quantitatively simula-
tion data such as shown in these figures
or experimental data as shown in Fig. 6-32
by analytical theories.
(vii) Finally, we draw attention to the
Figure 6-41. Log – log plot of the characteristic lin-
fact that the concepts developed in the
ear dimensions RG (t) and RE (t) at time t after the
present context can also be extended to quench, for the simulation shown in Fig. 6-40. RG (t)
very different physical phenomena. For is extracted from the correlation function and RE (t)
example, the Lifshitz – Slyozov mechanism from the energy relaxation. The full curve is a fit to
can be identified to describe processes such Eq. (6-84 a). From Amar et al. (1988).
as the healing of rough surfaces at low tem-
perature (Villain, 1986; Selke, 1987).
6.6 Acknowledgements
This work has profited considerably
from a longstanding and stimulating inter-
action of K. B. with D. Stauffer and D. W.
Heermann. This author thanks them and
also other co-workers, namely C. Billotet,
H.-O. Carmesin, H. L. Frisch, J. D. Gun-
ton, S. Hayward, J. Jäckle, M. H. Kalos, K.
Kaski, K. W. Kehr, J. L. Lebowitz, G. Liro-
nis, A. Milchev, P. Mirold, S. Reulein, A.
Sariban, and K. Yaldram, for fruitful col-
laboration.
The other author (P. F.) is particularly in-
debted to O. Penrose and J. L. Lebowitz for
a longstanding and fruitful collaboration.
Figure 6-40. “Snapshot pictures” of the two-dimen- Furthermore he is indebted to H. Gupta, C.
sional nearest-neighbor Ising model of a phase-sep- A. Laberge, F. Langmayr, P. Nielaba, O.
arating mixture evolving after a critical quench to Paris, G. Vogl, and R. Weinkamer for fruit-
T = 0.5 Tc at times t = 5000 MCS (a), t = 105 MCS (b) ful interactions.
and t = 9.8 ¥ 105 MCS (c). B atoms are shown
in black, A atoms are not shown. Data are for a
512 ¥ 512 lattice. From Amar et al. (1988).
6.7 References Binder, K. (1987 a), Rep. Prog. Phys. 50, 783.
Binder, K. (1987 b), Colloid Polym. Sci. 265, 273.
Binder, K. (1989), in: Alloy Phase Stability: Gonis, A.,
Abinandanan T. A., Johnson W. C. (1993), Acta Stocks, L. M. (Eds.). Dordrecht: Kluwer, p. 232.
Metall. Mater. 41, 17; 27. Binder, K. (1998), J. Non-Equilib. Thermodyn. 23,
Abinandanan T. A., Johnson W. C. (1995), Mat. Sci. 114.
Eng. B 32, 169. Binder, K. (2001), in: Phase Transformations in Ma-
Acuña, R. J., Craievich, A. F. (1979), J. Non-Cryst. terials: Kostorz, G. (Ed.). Weinheim: Wiley-VCH,
Solids 34, 13. Chap. 4.
Akaiwa, N., Voorhees, P. W. (1994), Phys. Rev. E 49, Binder, K., Frisch, H. L. (1984), J. Chem. Phys. 81,
3860. 2126.
Akcasu, A. Z., Benmouna, M., Benoit, H (1986), Binder, K., Heermann, D. W. (1985), in: Scaling
Polymer 27, 1935. Phenomena in Disordered Systems: Pynn, R.,
Alexander, F. J., Chen, S., Grunau, D. W. (1993), Skjeltorp, T. (Eds.). New York: Plenum Press,
Phys. Rev. B 48, 634. p. 207.
Alkemper, J., Snyder, V. A., Akaiwa, N., Voorhees, Binder, K., Sillescu, H. (1989), in: Encyclopedia of
P. W. (1999), Phys. Rev. Lett. 82, 2725. Polymer Science, Vol. 2: Kroschwitz, J. (Ed.). New
Allen, S. M. (1977), Philos. Mag. 36, 181. York: Wiley, p. 297.
Allen, S. M, Cahn, J. W. (1976), Acta Metall. 24, 425. Binder, K., Stauffer, D. (1974), Phys. Rev. Lett. 33,
Allen, S. M., Cahn, J. W. (1979 a), Acta Metall. 27, 1006.
1017. Binder, K., Stauffer, D. (1976 a), Adv. Phys. 25, 343.
Allen, S. M., Cahn, J. W. (1979 b), Acta Metall. 27, Binder, K., Stauffer, D. (1976 b), Z. Phys. B 24, 407
1085. Binder, K., Billotet, C., Mirold, P. (1978), Z. Phys.
Alpern, P., Benda, Th., Leiderer, P. (1982), Phys. B 30, 183.
Rev. Lett. 49, 1267. Binder, K., Kalos, M. H., Lebowitz, J. L, Marro, J.
Als-Nielsen, J. (1976), in: Phase Transitions and (1979), Adv. Colloid Interface Sci. 10, 173.
Critical Phenomena, Vol. 5 a: Domb, C., Green, M. Binder, K., Frisch, H. L., Jäckle, J. (1986), J. Chem.
S. (Eds.). New York: Academic Press, p. 88. Phys. 85, 1505.
Amar, J. G., Sullivan, F. E., Mountain, R. D. (1988), Binder, K., Heermann, D. W., Milchev, A, Sadiq, A.
Phys. Rev. B 37, 196. (1987), in: Proceedings of Heidelberg Colloquium
Avrami, M. (1939), J. Chem. Phys. 7, 1103. on Glassy Dynamics: Van Hemmen, L., Morgen-
Bastea, S., Lebowitz, J.L. (1995), Phys. Rev. E52, 3821. stern, I. (Eds.). Berlin: Springer, p. 154.
Bates, F. S., Wiltzius, P. (1989), J. Chem. Phys. 91, Bortz, A. B., Kalos, M. H., Lebowitz, J. L., Zendejas,
3258. M. A. (1974), Phys. Rev. B 10, 535.
Benda, Th., Alpern, P., Leiderer, P. (1981), Physica Bray, A. J. (1990), Phys. Rev. B 41, 6274.
107 B, 157. Bray, A. J. (1994), Adv. Phys. 43, 357.
Benda, Th., Alpern, P., Leiderer, P. (1982), Phys. Bray, A. J. (1998), Phys. Rev. E 58, 1508.
Rev. B 26, 1450. Cahn, J. W. (1961), Acta Metall. 9, 795.
Beysens, D., Perrot, F. (1984), J. Phys. (Paris) Lett. Cahn, J. W. (1962), Acta Metall. 10, 179.
45, L-31. Cahn, J. W. (1965), J. Chem. Phys. 42, 93.
Beysens, D., Guenoun, P., Perrot, F. (1988), in: Dy- Cahn, J. W. (1966), Acta Metall. 14, 1685.
namic of Ordering Process in Condensed Matter: Cahn, J. W. (1968), Trans. Metall. Soc. AIME 242,
Komura, S., Furukawa, H. (Eds.). New York: 166.
Plenum Press, p. 373. Cahn, J. W., Hilliard, J. E. (1958), J. Chem. Phys. 28,
Billotet, C., Binder, K. (1979). Z. Phys. B 32, 195. 258.
Binder, K. (1977), Phys. Rev. B 15, 4425. Cahn, J. W., Hilliard, J. E. (1959), J. Chem. Phys. 31,
Binder, K. (1981), in Stochastic Nonlinear Systems 688.
in Physics, Chemistry and Biology: Arnold, L., Cahn, J. W., Larché, F. C. (1984), Acta Metall. 32,
Lefever, R. (Eds.). Berlin: Springer, p. 62. 1915.
Binder, K. (1983), J. Chem. Phys. 79, 6387. Carmesin, H.-O., Heermann, D. W., Binder, K.
Binder, K. (1984 a), in: Condensed Matter Research (1986), Z. Phys. B 65, 89.
Using Neutrons: Lovesey, S. W., Scherm, R. Chakrabarti, A., Toral, R., Gunton, J. D. (1993),
(Eds.). New York: Plenum Press, p. 1. Phys. Rev. E 47, 3025.
Binder, K. (1984 b), Phys. Rev., A 29, 341. Chan, S. K. (1977), J. Chem. Phys. 67, 5755.
Binder, K. (1984 c), in: Kinetics of Aggregation and Chan, C. K., Goldburg, W. I. (1987), Phys. Rev. Lett.
Gelation: Family, F., Landau, D. P. (Eds.). Amster- 48, 674.
dam: Elsevier, p. 209. Chan, C. K., Goldburg, W. I., Maher, J. V. (1987),
Binder, K. (1986), in: Festkörperprobleme – Ad- Phys. Rev. A 35, 1756.
vances in Solid State Physics, Vol. 26: Grosse, P. Chan, C. K., Perrot, F., Beysens, D. (1988), Phys.
(Ed.) Braunschweig: Vieweg, p. 133. Rev. Lett. 61, 412.
6.7 References 475
Che, D. Z., Spooner, S., Hoyt, J. J. (1997), Acta Falicov, A., Berker, A. N. (1995), Phys. Rev. Lett. 74,
Mater. 45, 1167. 426.
Chen, H., Rolat, S., Epperson, J. E. (1988), in: Dy- Flevaris, N. K. (1987). J. Am. Ceram. Soc. 70, 301.
namics of Ordering Processes in Condensed Mat- Flory, P. J. (1953), Principles of Polymer Chemistry.
ter: Komura, S., Furukawa, H. (Eds.). New York: Ithaka: Cornell University Press.
Plenum Press, p. 245. Flynn, C. P. (1972), Point Defects and Diffusion. Ox-
Cherns, D., Greene, P. D., Hainsworth, A., Preston, ford: Clarendon Press.
A. R. (1988), in: Microscopy of Semiconducting Forouhi, A. R., De Fontaine, D. (1987), Acta Metall.
Materials. Bristol: Institute of Physics, p. 83. 35, 1863.
Cho, J. H., Ardell, A. J. (1997), Acta Mater. 45, 1393. Fratzl, P. (1991), J. Appl. Cryst. 24, 593.
Chou, Y. C., Goldburg, W. I. (1979), Phys. Rev. A 20, Fratzl. P., Blaschko, O. (1988), in: Dynamics of Or-
2105. dering Processes in Condensed Matter: Komura,
Chou, Y. C., Goldburg, W. I. (1981), Phys. Rev. A 23, S., Furukawa, H. (Eds.). New York: Plenum Press,
858. p. 223.
Chu, B. (1988), in: Dynamics of Ordering Processes Fratzl, P., Lebowitz, J. L. (1989), Acta Metall. 12,
in Condensed Matter: Komura, S., Furukawa, H. 3245.
(Eds.). New York: Plenum Press, p. 463. Fratzl, P., Penrose, O. (1994). Phys. Rev. B 50, 3477.
Conley, J. G., Fine, M. E., Weertmann, J. R. (1989), Fratzl, P., Penrose, O. (1995). Acta Metall. Mater. 43,
Acta Metall. 37, 1251. 2921.
Cook, H. E. (1970), Acta Metall. 18, 297. Fratzl, P., Penrose, O. (1996). Acta Metall. Mater. 44,
Cook, H. E., De Fontaine, D., Hilliard, J. E. (1969), 3227.
Acta Metall. 17, 765. Fratzl, P., Penrose, O. (1997). Phys. Rev. B 55,
Craievich, A. F., Olivieri, J. R. (1981), J. Appl. Crys- R 6101.
tallogr. 14, 444. Fratzl, P., Lebowitz, J. L., Marro, J., Kalos, M. H.
Craievich, A. F., Sanchez, J. M. (1981), Phys. Rev. (1983), Acta. Metall. 31, 1849.
Lett. 47, 1308. Fratzl, P., Lebowitz, J. L., Penrose, O., Amar, J.
Dee, G., Gunton, J. D., Kawasaki, K. (1981), Prog. (1991), Phys. Rev. B 44, 4794.
Theor. Phys. 65, 365. Fratzl, P., Penrose, O., Lebowitz, J. L. (1999), J. Stat.
De Fontaine, D. (1979), in: Solid State Physics, Vol. Phys. 95, 1429.
34: Ehrenreich, H., Seitz, F., Thurnbull, D. (Eds.). Fredrickson, G. H., Binder, K. (1989), J. Chem.
London: Academic Press, p. 73. Phys. 91, 7265.
De Fontaine, D., Cook, H. E. (1971), in: Critical Furukawa, H. (1978), Prog. Theor. Phys. 59, 1072.
Phenomena in Alloys, Magnets and Super Con- Furukawa, H. (1984), Physica 123 A, 497.
ducters: Miller, R. E., Ascher, E., Jaffee, I. (Eds.). Furukawa, H. (1985 a), Adv. Phys. 34, 703.
New York: McGraw-Hill, p. 257. Furukawa, H. (1985 b), Prog. Theor. Phys. 74, 174.
De Fontaine, D., Kikuchi, R. (1978), NBS Spec. Publ. Furukawa, H. (1985 c), Phys. Rev. A31, 1103.
No. 496, 976. Furukawa, H. (1988), in: Dynamics of Ordering
De Gennes, P. G. (1979), Scaling Concepts in Poly- Processes in Condensed Matter: Komura, S., Fu-
mer Physics. Ithaca: Cornell University Press. rukawa, H. (Eds.). New York: Plenum Press,
De Gennes, P. G. (1980), J. Chem. Phys. 72, 4756. p. 35.
De Gennes, P. G. (1984), Phys. Chem. 88, 6469. Furusaka, M., Ishikawa, Y., Mera, M. (1985), Phys.
De Groot, S. R., Mazur, P. (1962), Non-Equilibrium Rev. Lett. 54, 2611.
Thermodynamics. Amsterdam: North-Holland. Furusaka, M., Ishikawa, Y., Yamaguchi, S., Fujino,
Descamps, M., Caucheteux, C. (1988), in: Dynamics Y. (1986), J. Phys. Soc. Jpn. 55, 2253.
of Ordering Processes in Condensed Matter: Ko- Furusaka, M., Fujikawa, S., Sakauchi, M, Hirano, K.
mura, S., Furukawa, H. (Eds.). New York: Plenum (1988), in: Dynamics of Ordering Processes in
Press, p. 257. Condensed Matter: Komura, S., Furukawa, H.
Dietrich, S. (1988), in: Phase Transitions and Criti- (Eds.). New York: Plenum Press, p. 281.
cal Phenomena, Vol. 12: Domb, C., Lebowitz, S. Gaulin, B. D., Spooner, S., Morii, Y. (1987), Phys.
L. (Eds.). New York: Academic Press, p. 1. Rev. Lett. 59, 668.
Doi, M., Edwards, S. F. (1986), The Theory of Poly- Gawlinksi, E. T., Gunton, J. D., Vinãls, J. (1989),
mer Dynamics. Oxford: Clarendon Press. Phys. Rev. B 39, 7266.
Easwar, N. (1992), Phys. Rev. Lett. 68, 186. Gerold, V., Kostorz, G. (1978), J. Appl. Crystallogr.
Edwards, J. T., Hillel, A. J. (1977), Philos. Mag. 35, 11, 376.
1231. Ginzburg, V. L. (1960), Sov. Phys. Solid State 2,
Eguchi, T., Oki, K., Matsumura, S. (1984), Mater. 1824.
Res. Soc. Symp. Proc. 21, 589. Glatter, O., Kratky, O. (1982), Small Angle X-Ray
Fährmann, M., Fratzl, P., Paris, O., Fährmann, E., Scattering. London: Academic Press.
Johnson, W. C. (1995), Acta Metall. Mater. 43, Glotzer, S. C., Di Marzio, E. A., Muthukumar, M.
1007. (1995), Phys. Rev. Lett. 74, 2034.
Goldburg, W. I. (1981), in: Scattering Technique Hashimoto, T. (1993), in: Materials Science and
Applied to Supramolecular and Nonequilibrium Technology, Vol. 12: Cahn, R. W., Haasen, P.,
Systems: Chen, S. H., Chu, B., Nossal, R. (Eds.). Kramer, E. J. (Eds.). Weinheim: VCH, p. 251.
New York: Plenum Press, p. 383. Hashimoto, T., Izumitani, T. (1985), J. Chem. Phys.
Goldburg, W. I. (1983), in: Light Scattering Near the 83, 3694.
Phase Transition: Cummins, H. Z., Levanyuk, A. Hashimoto, T., Nishimura, K., Takeuchi, Y. (1978),
P. (Eds.). Amsterdam: North-Holland, p. 531. J. Phys. Soc. Jpn. 45, 1127.
Goldburg, W. I. (1988), Dynamics of Ordering Hashimoto, T., Itakuri, M., Shimidzu, N. (1986),
Processes in Condensed Matter: Komura, S., J. Chem. Phys. 85, 6773.
Furukawa, H. (Eds.). New York: Plenum Press, Hashimoto, T., Matsuzaka, K., Moses, E., Onuki, A.
p. 361. (1995). Phys. Rev. Lett. 74, 126.
Goldburg, W. I., Shaw, C. H., Huang, J. S., Pilant, M. Hayward, S., Heermann, D. W., Binder, K. (1987),
S. (1978), J. Chem. Phys. 68, 484. J. Stat. Phys. 49, 1053.
Goldburg, W. I., Aliev, F., Wu, X. L. (1995), Physica Heermann, D. W. (1984 a), Phys. Rev. Lett. 52, 1126.
A 213, 61. Heermann, D. W. (1984 b), Z. Phys. B 55, 309.
Gorentsveig, V. I., Fratzl, P., Lebowitz, J. L. (1997), Heermann, D. W. (1985), Z. Phys. B 61, 311.
Phys. Rev. B 55, 2912. Heermann, D. W., Klein, W. (1983a), Phys. Rev. Lett.
Grant, M., San Miguel, M., Vinals, S., Gunton, J. D. 50, 1962.
(1985), Phys. Rev. B 31, 302. Heermann, D. W., Klein, W. (1983 b), Phys. Rev.
Grest, G. S., Srolovitz, D. J. (1985), Phys. Rev. B 32, Lett. B 27, 1732.
3014. Heermann, D. W., Klein, W., Stauffer, D. (1982),
Griffiths, R. B. (1970), Phys. Rev. Lett. 24, 715. Phys. Rev. Lett. 49, 1262.
Guenoun, P., Gastand, R., Perrot, F., Beysens, D. Hennion, M., Ronzaud, D., Guyot, P. (1982), Acta
(1987), Phys. Rev. A 36, 4876. Metall. 30, 599.
Gunton, J. D., San Miguel, M., Sahni, P. S. (1983), Henzler, M., Busch, H. (1990), Phys. Rev. B 41, 4891.
in: Phase Transition and Critical Phenomena, Vol. Hikmet, R. M., Callister, S., Keller, A. (1988), Poly-
8: Domb, C., Lebowitz, J. L. (Eds.). London: Aca- mer 29, 1378.
demic Press, p. 267. Hillel, A. J., Edwards, J. T., Wilkes, P. (1975), Phi-
Gunton, J. D., Gawlinski, E. T., Kaski, K. (1988), in: los. Mag. 32, 189.
Dynamics of Ordering Processes in Condensed Hirotsu, S., Kaneki, A. (1988), in: Dynamics of Or-
Matter: Komura, S., Furukawa, H. (Eds.). New dering Processes in Condensed Matter: Komura,
York: Plenum Press, p. 101. S., Furukawa, H. (Eds.). New York: Plenum Press,
Gupta, H., Weinkamer, R., Fratzl, P., Lebowitz, J. L. p. 481.
(2001), Acta Mater. 49, 53. Hobbie, E. K., Kim, S. H., Han, C. C. (1996), Phys.
Guyot, P., Simon, J. P. (1982), in: Solid State Phase Rev. E 54, R 5909.
Transformations: Aaronsen, H. I., Laughlin, D. E., Hoffer, J. K., Campbell, L. J., Bartlett, R. J. (1980),
Schuka, R. F., Wayman, C. M. (Eds.). New York: Phys. Rev. Lett. 45, 912.
AIME, p. 325. Hohenberg, P. C., Halperin, B. I. (1977), Rev. Mod.
Guyot, P., Simon, J. P. (1988), in: Dynamics of Or- Phys. 49, 435.
dering Processes in Condensed Matter: Komura, Hohenberg, P. C., Nelson, D. (1979), Phys. Rev. B 20,
S., Furukawa, H. (Eds.). New York: Plenum Press, 2665.
p. 211. Hono, K., Babu, S. S., Hiraga, K., Okano, R., Saku-
Haasen, P., Gerold, V., Wagner, R., Ashby, M. F. rai, T. (1992), Acta Mater. 40, 3027.
(Eds.) (1984), Decomposition of Alloys: The Early Houessou, C., Guenoun, P., Gastaud, R., Perrot, F.,
Stages. Oxford: Pergamon Press. Beysens, D. (1985), Phys. Rev. A 32, 1818.
Han, C. C., Okada, M., Sato, T. (1988), in: Dynamics Houston, E. L., Cahn, J. W., Hilliard, J. E. (1966),
of Ordering Processes in Condensed Matter: Ko- Acta Metall. 14, 1685.
mura, S., Furukawa, H. (Eds.). New York: Plenum Huberman, B. A. (1976), J. Chem. Phys. 65, 2023.
Press, p. 433. Huse, D. A. (1986), Phys. Rev. B 34, 7845.
Hasegawa, H., Shiwaku, T., Nakai, A., Hashimoto, T. Huse, D. A. (1987), Phys. Rev. B 36, 5383.
(1988), in: Dynamics of Ordering Processes in Izumitani, T., Takanaka, M., Hashimoto, T. (1989),
Condensed Matter: Komura, S., Furukawa, H. J. Chem. Phys. 92, 3213.
(Eds.). New York: Plenum Press, p. 457. Jäckle, J., Pieroth, M. (1988), Z. Phys. B 72, 25.
Hashimoto, T. (1987), in: Current Topics in Polymer Jantzen, C. M. F., Herman, H. (1978), in: Phase Dia-
Science, Vol. 2: Ottenbrite, R. M., Itracki, L. A., grams: Material Science and Technology, Vol. 5:
Inoue, S. (Eds.). Munich: Hanser, p. 199. Alper, A. A. (Ed.). New York: Academic Press,
Hashimoto, T. (1988), in: Dynamics of Ordering p. 127.
Processes in Condensed Matter: Komura, S., Fu- Jinnai, H., Hashimoto, T., Lee, D., Chen, S.-H.
rukawa, H. (Eds.). New York: Plenum Press, (1997), Macromolecules 30, 130.
p. 421. Joanny, J. F. (1978), C. R. Acad. Sci. Ser. B 286, 89.
6.7 References 477
Johnson, W. A., Mehl, R. F. (1939), Trans. AIME Krall, A. H., Sengers, J. V., Hamano, K. (1992),
135, 416. Phys. Rev. Lett. 69, 1963.
Jones, R. A. L., Norton, L. J., Kramer, E. J., Bates, F. Kramer, E. J., Green, P., Palmstrom, C. (1984). Poly-
S., Wiltzius, P. (1991), Phys. Rev. Lett. 66, 1326. mer 25, 473.
Joshua, M., Goldberg, W. I., Onuki, A. (1985), Phys. Krausch, G. (1995). Mat. Sci. Eng. Rep. 14, 1.
Rev. Lett. 54, 1175. Kumaki, J., Hashimoto, T., Granick, S. (1996), Phys.
Jou, H. J., Leo, P. H., Lowengrub, J. S. (1997), J. Rev. Lett. 77, 1990
Comp. Phys. 131, 109. Kuo, S. Y., Viskar, A. V (1987), J. Am. Ceram. Soc.
Kadanoff, L. P. (1976), in: Phase Transitions and 70, C 125.
Critical Phenomena, Vol. 5 a: Domb, D., Green, Laberge, C., Fratzl, P., Lebowitz, J. L. (1995). Phys.
M. S. (Eds.). London: Academic Press, Ch. 1. Rev. Lett. 75, 4448.
Kampmann, R., Wagner, R. (1984), in: Decomposi- Laberge, C., Fratzl, P., Lebowitz, J. L. (1997). Acta
tion of Alloys: the Early Stages: Haasen, P., Ge- Mater. 45, 3946.
rald, V., Wagner, R., Ashby, M. F. (Eds.). Oxford: Lai, Z. W., Mazenko, G. F., Valls, O. T. (1988), in:
Pergamon Press, p. 91. Dynamics of Ordering Processes in Condensed
Kaski, K., Binder, K., Gunton, J. D. (1984), Phys. Matter: Komura, S., Furukawa, H. (Eds.). New
Rev. B 29, 3996. York: Plenum Press, p. 45.
Katano, S., Iizumi, M. (1984), Phys. Rev. Lett. 52, Langer, J. S., Schwartz, A. J. (1980), Phys. Rev. A 21,
835. 948.
Katano, S., Iizumi, M. (1988), in: Dynamics of Or- Langer, J. S., Baron, M., Miller, H. D. (1975), Phys.
dering Processes in Condensed Matter: Komura, Rev. A 11, 1417.
S., Furukawa, H. (Eds.). New York: Plenum Press, Langmayr, F., Fratzl, P., Vogl, G. (1992), Acta Met-
p. 321. all. Mater. 40, 3381.
Kawasaki, K. (1972), in: Phase Transitions and Crit- Larché, F. C., Cahn, J. W. (1982), Acta Metall. 30,
ical Phenomena. Vol. 2: Domb, C., Green, M. S. 1835.
(Eds.). New Academic Press, Ch. 11. Laslaz, G., Guyot, P., Kostorz, G. (1977), J. Phys.
Kawasaki, K., Ohta, T. (1978), Prog. Theor. Phys. (Paris) C 7, 406.
59, 362. Lauger, J., Laubner, C., Gronski, W. (1995). Phys.
Kawasaki, K, Imeada, T., Gunton, J. D. (1981), in: Rev. Lett. 75, 3576.
Perspectives in Statistical Physics: Raveché, H. J. Lawrie, I. D., Sarbach, S. (1984), in: Phase Transi-
(Ed.). Amsterdam: North -Holland, p. 203. tions and Critical Phenomena, Vol. 9: Domb, C.,
Kehr, K. W., Binder, K., Reulein, S. M. (1989), Phys. Lebowitz, J. L. (Eds.). New York: Academic Press,
Rev. B 39, 4891. p. 1.
Khachaturyan, A. G. (1966), Sov. Phys. Solid State 8, Lebowitz, J. L. Penrose, O. (1966), J. Math. Phys. 7,
2163. 98.
Khachaturyan, A. G. (1983), Theory of Structural Lebowitz, J. L. Marro, J., Kalos, M. H. (1982), Acta
Transformations in Solids. New York: Wiley. Metall. 30, 297.
Klein, W., Unger, C. (1983), Phys. Rev. B 28, 445. Lee, S. K. (1997), Mater. Sci. Eng. A 238, 1.
Koga, T., Kawasaki, K. (1991), Phys. Rev. A 44, R 817. Lee, S. K. (1998), Trans. JIM 39, 114.
Koga, T., Kawasaki, K. (1993), Physica A 196, 389. Lefever, R., Carati, D., Hassani, N. (1995), Phys.
Komura, S., Osamura, K., Fujii, H., Takeda, T. Rev. Lett. 75, 1674.
(1985), Phys. Rev. Lett. B 11, 1278. Leo, P. H., Lowengrub, J. S., Jou, H. J. (1998), Acta
Komura, S., Takeda, T., Osamura, K., Okuda, K. Mater. 46, 2113.
(1988), in: Dynamics of Ordering Processes in Li, D. Y., Chen, L. Q. (1998), Acta Mater. 46, 639.
Condensed Matter: Komura, S., Furukawa, H. Lifshitz, I. M. (1962), Sov. Phys. JETP 15, 939.
(Eds.). New York: Plenum Press, p. 251. Lifshitz, I. M., Slyozov, V. V. (1961), J. Phys. Chem.
Konigsveld, R. Kleintjens, L. A., Nies, E. (1987), Solids 19, 35.
Croat. Chem. Acta 60, 53. Lironis, G., Heermann, D. W., Binder, K. (1989),
Konishi, H., Noda, Y. (1988), in: Dynamics of Order- J. Phys. A: Math. Gen. 23, L329.
ing Processes in Condensed Matter: Komura, S., Liu, A. J., Durian, D. J., Herbolzheimer, E. Safran, S.
Furukawa, H. (Eds.). New York: Plenum Press, p. A. (1990), Phys. Rev. Lett. 65, 1897.
309. Livet, F., Bloch, D. (1985), Scr. Metall. 10, 19.
Kostorz, G. (1988), in: Dynamics of Ordering Pro- Lopez, V. M., Sano, N, Sakurai, T, Hirano, K. (1993).
cesses in Condensed Matter: Komura, S., Furu- Acta Metall, Mater. 41, 265.
kawa, H. (Eds.). New York: Plenum Press, p. 199. Luiggi, N., Simon, J. P., Guyot, P. (1980), Acta
Kostorz, G. (1991), J. Appl. Cryst. 24, 444. Metall. 28, 1115.
Kostorz, G. (1994), Solid – Solid Phase Transforma- Lyon, O., Simon, J. P. (1987), Phys. Rev. B 35,
tions, PTM’94: Johnson, W. C., Howe, J. M., 5164.
Laughlin, D. E., Soffa, W. A. (Eds.). The Metals, Maheswari, A., Ardell, A. J. (1993), Phys. Rev. Lett.
Minerals and Materials Society: Warrendale, p. 133. 70, 2305.
Mainville, J., Yang, Y. S., Elder, K. R., Sutton, M., Ohta, T., Enomoto, Y., Kawasaki, K., Sato, A.
Ludwig Jr., K. F., Stephenson, G. B. (1997), Phys. (1988), in: Dynamics of Ordering Processes in
Rev. Lett. 78, 2787. Condensed Matter: Komura, S., Furukawa, H.
Manning, J. R. (1986), Diffusion Kinetics for Atoms (Eds.). New York: Plenum Press, p. 127.
in Crystals. Princeton: Van Nostrand. Ohta, T., Jasnow, D., Kawasaki, K. (1982), Phys.
Marro, J., Bortz, A. B., Kalos, M. H., Lebowitz, J. L Rev. Lett. 49, 1223.
(1975), Phys. Rev. B 12, 2000. Okada, M., Han, C. C. (1986), J. Chem. Phys. 85,
Marro, J., Lebowitz, J. L., Kalos, M. H. (1979), Phys. 5317.
Rev. Lett. 43, 282. Oki, K., Hasaka, M., Eguchi, T. (1973), Jpn. J. Appl.
Mashiyama, H. (1988), in: Dynamics of Ordering Phys. 12, 1522.
Processes in Condensed Matter: Komura, S., Fu- Okumura, H., Inoue, A., Masumoto, T. (1993), Acta
rukawa, H. (Eds.). New York: Plenum Press, p. Metall. Mater. 41, 915.
355. Onodera, Y., Hirano, K. (1986), J. Mater. Sci. Lett. 5,
Matsuura, M. (1988), in: Dynamics of Ordering Pro- 1048.
cesses in Condensed Matter: Komura, S., Furu- Onodera, Y., Hirano, K. (1988), in: Dynamics of Or-
kawa, H. (Eds.). New York: Plenum Press, p. 349. dering Processes in Condensed Matter: Komura,
Mazenko, G. F. (1994), Phys. Rev. E 50, 3485. S., Furukawa, H. (Eds.). New York: Plenum Press,
Mazenko, G. F. (1998), Phys. Rev. E 58, 1543. p. 293.
Mazenko, G. F., Wickham, E. (1995), Phys. Rev. Onuki, A. (1984), Phys. Lett. A 101, 286.
E 51, 2886. Onuki, A. (1986 a), Phys. Rev. A 34, 3528.
Mcdevitt, T. L., Mahajan, S., Laughlin, D. E., Bon- Onuki, A. (1986 b), J. Chem. Phys. 85, 1122.
ner, W. A., Keramidas, V. G. (1992), Phys. Rev. Onuki, A. (1989 a), J. Phys. Soc. Japan 58, 3065.
B 45, 6614. Onuki, A. (1989 b), J. Phys. Soc. Japan 58, 3069.
McPherson, R. (1987), J. Mater. Sci. Lett. 6, 795. Onuki, A. (1989 c), Int. J. Thermophys. 10, 293.
Meakin, P., Metiu, H., Petschek, R. G., Scalapino, D. Onuki, A. (1997), J. Phys.: Condens Matter 9, 6119.
J. (1983), J. Chem. Phys. 69, 1948. Onuki, A., Kawasaki, K. (1978), Prog. Theor. Phys.
Meier, H., Strobl, G. R. (1988), Macromolecules 20, Suppl. 64, 436.
649. Onuki, A., Kawasaki, K. (1979), Ann. Phys. 121,
Milchev, A., Heermann, D. W., Binder, K. (1988), 456.
Acta Metall. 36, 377. Onuki, A., Nishimori, H. (1991), Phys. Rev. B 43,
Mirold, P., Binder, K. (1977), Acta Metall. 25, 1435. 13649.
Monette, L., Liu, A. J., Grest, G. S. (1992), Phys. Onuki, A., Yamamoto, R., Taniguchi, T. (1997),
Rev. A 46, 7664. Phys. II (France) 7, 295.
Morii, Y., Gaulin, B. D., Spooner, S. (1988), in: Dy- Oono, Y., Puri, S. (1988), Phys. Rev. A 38, 434.
namics of Ordering Processes in Condensed Mat- Orlikowski, D., Sagui, C., Somoza, A., Roland, C.
ter: Komura, S., Furukawa, H. (Eds.). New York: (1999), Phys. Rev. B59, 8646.
Plenum Press, p. 239. Osamura, K. (1988), in: Dynamics of Ordering Pro-
Mouritsen, O. G. (1986), Phys. Rev. Lett. 56, 850. cesses in Condensed Matter: Komura, S., Furu-
Mouritsen, O. G., Praestgaard, E. (1988), Phys. Rev. kawa, H. (Eds.). New York: Plenum Press, p. 187.
B 38, 2703. Paris, O., Fährmann, M., Fratzl, P. (1995), Phys. Rev.
Mukherjee, S., Cooper, B. R. (1998), Mat. Sci. Eng. Lett. 75, 3458.
A 248, 101. Paris, O., Färmann, M., Fährmann, E., Pollock, T. E.,
Müller-Krumbhaar, H. (1974). Phys. Lett. 50 A, 27. Fratzl, P. (1997), Acta Mater. 45, 1085.
Nattermann, T. (1998), in: Spin Glasses and Random Penrose, O., Lebowitz, J. L. (1971), J. Stat. Phys. 3,
Fields: Young, A. P. (Ed.). Singapore: World Sci- 211.
entific, p. 277. Perrot, F., Guenoun, P., Baumberger, T., Beysens, D.,
Nielaba, P., Fratzl, P. Lebowitz, P. (1999), J. Stat. Garrabos, Y., Lenemdre, B. (1994), Phys. Rev.
Phys. 95, 23. Lett. 73, 688.
Nishihara, S., Noda, Y., Yamada, Y. (1982), Solid Pincus, P. (1981), J. Chem. Phys. 75, 1996.
State Commun. 44, 1487. Pine, D. J., Easwer, N., Maher, J. V., Goldburg, W. I.
Nishimori, H., Onuki, A. (1990), Phys. Rev. B 42, 980. (1984), Phys. Rev. A 29, 308.
Noda, Y. (1988), in: Dynamics of Ordering Processes Pistoor, N., Binder, K. (1988 a), Colloid Polym. Sci
in Condensed Matter: Komura, S., Furukawa, H. 266, 132.
(Eds.). New York: Plenum Press, p. 373. Pistoor, N., Binder, K. (1988 b), in: Polymer Motion
Noda, Y., Nishihara, S., Yamada, Y. (1984). J. Phys. in Dense Systems: Richter, D., Springer, T. (Eds.).
Soc. Jpn. 53, 4241. Berlin: Springer, p. 285.
Nose, T. (1987), Phase Transit. 8, 245. Porod, G. (1951), Kolloid Z. 124, 83; 125, 51.
Ohta, T. (1984), Prog. Theor. Phys. 71, 1409. Puri, S., Dünweg, B. (1992), Phys. Rev. A45, R 6977.
Ohta, T., Nozaki, H. (1989), preprint referred to by Puri, S., Frisch, H. L. (1997), J. Phys.: Cond. Matt. 9,
Bates and Wiltzius (1989). 2109.
6.7 References 479
Puri, S., Oono, Y. (1988), Phys. Rev. A 38, 1542. Sur, A., Marro, J., Lebowitz, J. L., Kalos, M. H.
Puri, S., Bray, A. J., Lebowitz, J. L. (1997), Phys. (1977), Phys. Rev. B 15, 535.
Rev. E 56, 758. Takahashi, J., Kuwayama, M., Kamiya, H., Takatsu,
Rao, M., Marro, J., Kalos, M. H., Lebowitz, J. L. M. (1988), Mater. Sci. 23, 337.
(1976), Phys. Rev. B 13, 2000. Takeda, S., Kulik, J., De Fontaine, D. (1987), Acta
Rikvold, P. A., Gunton, J. D. (1982), Phys. Rev. Lett. Metall. 35, 2243.
49, 226. Tanaka, H. (1993), Phys. Rev. Lett. 70, 53.
Rogers, T. M., Desai, R. C. (1989), Phys. Rev. B 39, Tanaka, H. (1994), J. Chem. Phys. 100, 5323.
11956. Tanaka, H. (1996), Phys. Rev. Lett. 76, 787.
Rogers, T. M., Elder, K. R., Desai, R. C. (1988), Taniguchi, T., Onuki, A. (1996), Phys. Rev. Lett. 77,
Phys. Rev. B 37, 9638. 4910.
Rouse, P. E. (1953), J. Chem. Phys. 21, 1273. Tokuyama, M., Enomoto, Y. (1993), Phys. Rev. E 47,
Sadiq, A., Binder, K. (1984), J. Stat. Phys. 35, 1156.
517. Tokuyama, M., Enomoto, Y., Kawasaki, K. (1987),
Sagui C., Somoza, A. M., Desai, R. (1994), Phys. Physica A 143, 187.
Rev. E 50, 4865. Tomita, H. (1984), Prog. Theor. Phys. 71, 1405.
Sahni, P. S., Gunton, J. D. (1980), Phys. Rev. Lett. 45, Tomokiyo, Y., Yahiro, K., Matsumura, S., Oki, K.,
369. Epudi, T. (1988), in: Dynamics of Ordering Pro-
Sahni, P. S., Gunton, J. D., Katz, S., Timpe, R. cesses in Condensed Matter: Komura, S., Furu-
(1982), Phys. Rev. B 25, 389. kawa, H. (Eds.). New York: Plenum Press, p. 233.
San Miguel, M., Gunton, J. D., Dee, G., Sahni, P. S. Tomozawa, M., MacCrone, R. K., Herman, H.
(1981), Phys. Rev. B 23, 2334. (1970), Phys. Chem. Glasses 11, 136.
San Miguel, M., Grant, M., Gunton, J. D. (1985), Toral, R., Chakrabarti, A., Gunton, J. D. (1988),
Phys. Rev. A 31, 1001. Phys. Rev. Lett. 60, 2311.
Sariban, A., Binder, K. (1989 a), Polym. Commun. Tringides, M. C., Wu, P. K., Moritz, W., Lagally,
30, 205. M. G. (1986), Ber. Bunsenges. Phys. Chem. 90, 277.
Sariban, A., Binder, K. (1989 b), Colloid Polym. Sci. Tringides, M. C., Wu, P. K., Lagally, M. G. (1987),
267, 469. Phys. Rev. Lett. 59, 315.
Sato, R., Han, C. C. (1988), J. Chem. Phys. 88, 2057. Unger, C., Klein, W. (1984), Phys. Rev. Lett. B 29,
Schätzel, K., Ackerson, B. J. (1993), Phys. Rev. E 48, 2698.
3766. Valls, O. T., Farrell, J. E. (1993), Phys. Rev. E 47,
Scher, M., Zallen, R. (1970), J. Chem. Phys. 53, 3759. R 36.
Schmitz, G., Hono, K., Haasen, P. (1994), Acta Villain, J. (1985), in: Scaling Phenomena in Disor-
Metall. Mater. 42, 201. dered Systems: Pynn, R., Skjeltorp, A. (Eds.). New
Selke, W. (1987), J. Phys. B 20, L 455. York: Plenum, p. 423.
Sequeira, A. D., Calderon, M. A., Kostorz, G., Peder- Villain, S. (1986), Europhys. Lett. 2, 531.
sen, J. S. (1995), Acta Metall. Mater. 43, 3427; 3441. Wagner, C. (1961), Z. Elektrochem. 65, 243.
Shinozaki, A., Oono, J. (1993), Phys. Rev. E 48, Wagner, R., Kampmann, R., Vorhees, P. W. (2001),
2622. in: Phase Transformations in Materials: Kostorz,
Sigga, E. (1979), Phys. Rev. A 20, 595. G. (Ed.). Weinheim: Wiley-VCH, Chap. 5.
Simon, J. P., Guyot, P., Ghilarducca de Silva (1984), Wang, Y., Khachaturyan, A. G. (1995), Acta Metall.
Philos. Mag. A 49, 151. Mater. 43, 1837.
Singhal, S. P., Herman, H., Kostorz, G. (1978), J. Wang, G. C., Lu, T. M. (1983), Phys. Rev. Lett. 50,
Appl. Crystallogr. 11, 572. 2014.
Sinha, D. N. Hoffer, J. K. (1981), Physica 107 B, 157. Wang, Y., Chen, L. A., Khachaturyan, A. G. (1992),
Snyder, H. L., Meakin, P. (1983), J. Chem. Phys. 79, Phys. Rev. 46, 11194.
5588. Wang, Y., Chen, L. A., Khachaturyan, A. G. (1993),
Srolovitz, D. J., Grest, G. S. (1985), Phys. Rev. B 32, Acta Metall. Mater. 41, 279.
3021. Wang, Y., Banarjee, D., Su, C. C., Khachaturyan, A.
Stanley, H. E. (1971), An Introduction to Phase Tran- G. (1998 a), Acta Mater. 46, 2983.
sitions and Critical Phenomena. Oxford: Oxford Wang, W. H., Wei, Q., Friedrich, S. (1998 b), Phys.
University Press. Rev. B 57, 8211.
Stauffer, D. (1985), An Introduction to Percolation Weinkamer, R., Gupta, H., Fratzl, P., Lebowitz, J. L.
Theory. London: Taylor and Francis. (2000), Europhys. Lett. 52, 224.
Stauffer, D., Ferer, M., Wortis, M. (1972), Phys. Rev. Wilcoxon, J. P., Schaefer, D. W., Kaler, E. W. (1988),
Lett. 29, 245. Phys. Rev. Lett. 60, 333.
Stephenson, G. B., Warburton, W. K., Haller, W., Wilcoxon, J. P., Martin, J. E., Odinek, J. (1995),
Bienenstock, A. (1991), Phys. Rev. B 43, 3417. Phys. Rev. Lett. 75, 1558.
Su, C. H., Voorhees, P. W. (1996), Acta Mater. 44, Wiltzius, P., Bates, F. S., Dierker, S. B., Wignall, G.
1987; 2001. D. (1987), Phys. Rev. A 36, 2991.
Wong, N. C., Knobler, C. M. (1978), J. Chem. Phys. rukawa, H. (Eds.). New York: Plenum Press, p.
69, 725. 343.
Wong, N. C., Knobler, C. M. (1979), Phys. Rev. Lett. Yaldram, K., Binder, K. (1991), J. Stat. Phys. 63, 161.
43, 1733. Yeung, C. (1988), Phys. Rev. Lett. 61, 1135.
Wong, N. C., Knobler, C. M. (1981), Phys. Rev. A 24, Yokota, H. (1978), J. Phys. Soc. Jpn. 45, 29.
3205. Zarzycki, J., Naudin, F. (1969), J. Non-Cryst. Solids.
Wortis, M. (1985), in: Fundamental Problems in Sta- 1, 215.
tistical Physics, Vol. VI: Cohen, E. G. D. (Ed.). Zettlemoyer, A. C. (1969), Nucleation. New York:
Amsterdam: North-Holland, p. 87. Dekker.
Wu, P. K., Perepezko, J. H., McKinney, J. T., Lagally, Zhang, Z., Chakrabarti, A. (1994), Phys. Rev. E 50,
M. G. (1983), Phys. Rev. Lett. 51, 1577. R 4290.
Wu, P. K., Tringides, M. C., Lagally, M. G. (1989), Zhang, Z., Chakrabarti, A. (1995), Phys. Rev. E 52,
Phys. Rev. B 39, 7595. 2736.
Yagi, T., Lu, Z.-L. (1988), in: Dynamics of Ordering Zuo, J.-K., Wang, G.-C., Lu, T. M. (1989), Phys. Rev.
Processes in Condensed Matter: Komura, S., Fu- B 40, 524.
7 Transformations Involving Interfacial Diffusion
Gary R. Purdy
Department of Materials Science and Engineering, McMaster University, Hamilton,

Ontario, Canada
Yves J. M. Bréchet
Laboratoire de Thermodynamique et Physico-Chimie Métallurgiques, E.N.S.E.E.G.,

Institut National Polytechnique de Grenoble, France

7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 484
7.2 Equilibrium Properties of Solid – Solid Interfaces . . . . . . . . . . . . 484
7.2.1 Structures of Grain Boundaries . . . . . . . . . . . . . . . . . . . . . . . 484
7.2.2 Structures of Interphase Boundaries . . . . . . . . . . . . . . . . . . . . 486
7.2.3 Segregation to an Interface . . . . . . . . . . . . . . . . . . . . . . . . . 489
7.3 Interfacial Diffusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 490
7.4 Forces for Interface Migration . . . . . . . . . . . . . . . . . . . . . . 492
7.4.1 Capillary Forces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 492
7.4.2 Chemical Forces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 493
7.4.3 Mechanical Forces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 493
7.4.3.1 Elastic Strain Energy . . . . . . . . . . . . . . . . . . . . . . . . . . . . 493
7.4.3.2 Plastic Response . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 494
7.4.4 Frictional Forces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 496
7.4.4.1 Solute Drag . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 496
7.4.4.2 Particle Pinning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 498
7.5 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 498
7.5.1 The Massive Transformation . . . . . . . . . . . . . . . . . . . . . . . . 498
7.5.2 Chemically-Induced Grain Boundary Migration . . . . . . . . . . . . . . 503
7.5.3 Discontinuous Precipitation . . . . . . . . . . . . . . . . . . . . . . . . . 504
7.5.3.1 Initiation of Discontinuous Precipitation . . . . . . . . . . . . . . . . . . 505
7.5.3.2 Theories of Steady Cooperative Growth . . . . . . . . . . . . . . . . . . 507
7.5.3.3 Experimental Observations . . . . . . . . . . . . . . . . . . . . . . . . . 510
7.5.4 Interface Migration in Multilayers . . . . . . . . . . . . . . . . . . . . . 514
7.6 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 516
7.7 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 516

ISBN: 3-527-30256-5
482 7 Transformations Involving Interfacial Diffusion

A interfacial area
b magnitude of the Burgers vector
C0 average concentration
Ca concentration
DC local difference from the average concentration C
DV /v ratio of the volume diffusion constant and the interfacial velocity
–
Db average grain boundary diffusion coefficient
dE/dx measure of force between solute atoms and boundary
DFch molar Helmholtz energy change accompanying complete transformation
DFch ¢ molar Helmholtz energy difference across an interface
0
DFch fraction of the total available Helmholtz energy dissipated in volume diffusion
f (C ) Helmholtz energy density functional
f0 measure of curvature of Helmholtz energy– composition relationship
gB shear modulus
H 3 k a /(3 k a + 4 gB)
JB flux of solute atoms
K = dA/dV interface curvature
L half-thickness of a foil specimen
La0 , Lb0 regular solution parameters for the crystal and boundary phases
l correlation length
M mobility
NV number of atoms per unit volume
naA mole number
pch interfacial driving force/chemical force
ph residual chemical force derived from coherency strain
pi solute drag force
pW virtual mechanical force
r1 , r2 radii of curvature
ri radius of facet
S spacing between precipitate lamellae
s equilibrium ratio of solute concentration in a grain boundary and that in adja-
cent crystal
T /Tm reduced temperature
Tm melting temperature
v boundary velocity
Vf volume fraction of a precipitate
Vm molar volume
Vmb molar volume of the boundary phase
W applied load per unit area
x, y, z coordinates
Y elastic constant
Z applied tensile stress
a crystalline phase
a, b included, external phase
ga activity coefficient
GB surface excess concentration of solute B per unit area
d boundary thickness
e misfit
ea interaction parameter; a measure of the non-ideality of the a solid solution
h coefficient of lattice parameter change with composition
q angle of tilt
k gradient energy coefficient
ka bulk modulus of precipitate
l spacing of parallel edge dislocations
mA, mB solvent, solute chemical potentials
W alignment parameter
s specific interfacial Helmholtz energy
sA , s B specific grain boundary Helmholtz energy in pure A, B
s ab specific interfacial Helmholtz energy of a/b interface
j phase field parameter
y order parameter

CIGM chemically induced grain boundary migration
DP discontinuous precipitation
DC discontinuous coarsening
h.c.p. hexagonal close-packed
STEM scanning transmission electron microscopy
7.1 Introduction or to interfacial precipitation. Before dis-

cussing these processes, a brief review will
Many microstructure-determining reac- be given of the equilibrium properties of
tions in solids are similar in the sense that transformation interfaces and of the dy-
they occur primarily at an interface, which namics of interfacial diffusion and inter-
traverses a volume of solid material, and face migration.
leaves that volume altered structurally or
chemically, or both. Indeed, most solid –
solid transformations fall into this set; in 7.2 Equilibrium Properties
this chapter, we will restrict discussion in
the following ways:
of Solid – Solid Interfaces
1) We will consider only those processes 7.2.1 Structures of Grain Boundaries
whose main characteristics are determined
Grain boundaries constitute a well stud-
by diffusion within the transformation
ied class of solid state defects which are
front, which may be a grain boundary or an
structural in nature, and which depend for
interphase boundary.
their existence on a misorientation between
2) We will further restrict attention to
otherwise identical crystals. If we restrict
processes that begin mainly at crystal
attention to macroscopic variables only,
boundaries, and that lead to heterogene-
the properties of a planar boundary are ex-
ously transformed regions. This restriction
pected to depend on five geometric vari-
implies that the transformation interface
ables, three of which define the relative or-
will be of relatively high energy, and there-
ientations of the two crystals, and a further
fore that the transformation will often be
two are required to specify the orientation
initiated with some difficulty.
of the boundary plane. Three additional de-
The subset defined above excludes those grees of freedom, associated with sublat-
homogeneous transformations which are tice translations, exist on the microscopic
initiated everywhere within a parent crystal scale appropriate to the modeling of grain
(Chap. 5; Wagner et al., 2001), and truly boundary structure.
diffusionless transformations (Chap. 9; De- Much of what is known or inferred about
laey, 2001). However, the massive trans- grain boundary structure is obtained from
formation survives, as do the processes of simulations using molecular statics or mo-
peritectoid precipitation, chemically in- lecular dynamics, supported in some cases
duced grain boundary migration, and dis- by high resolution microscopic techniques
continuous precipitation. The latter two (Smith, 1986).
(and similar diffusional processes) are The three main modeling techniques that
the subject of an excellent monograph by have been used for surfaces, grain bound-
Pawlowski and Ziȩba (1991), which is, un- aries and (to a lesser extent) interphase
fortunately, not widely available. It has re- boundaries, are energy relaxation methods,
cently been observed that the study of these molecular dynamics, and Monte Carlo sim-
and similar phenomena may be advanced ulations (Sutton and Balluffi, 1995). The
through the use of artificially prepared information expected from these simula-
multilayered structures (Klinger et al., tions concerns both the structural aspects
1997 b, c, 1998), which may be subject, for (molecular relaxation, molecular dynam-
example, to discontinuous homogenization ics), chemical aspects such as segregation
7.2 Equilibrium Properties of Solid – Solid Interfaces 485
(Monte Carlo), and dynamic aspects such Monte Carlo methods) still requires further
as diffusion (molecular dynamics). The development.
structural information stemming from A surprisingly large amount of our infor-
these methods is very much dependent on mation relates to a rather special kind of
the interatomic potential chosen, and espe- interface, the symmetric tilt grain bound-
cially on the anharmonic part, since the ary. As suggested by Fig. 7-1, low angle
structure of the interface is usually rather boundaries are made up of arrays of dislo-
distorted. The dynamic aspects of inter- cations, and the low angle symmetric tilt
faces include the interface diffusivity and boundary is composed of parallel edge dis-
the interface mobility. While the diffusivity locations, whose spacing l is given by
is accessible to these simulations (via mo-
l = b/ 2 sin (q / 2 ) (7-1)
lecular dynamics), the mobility is more of a
problem, because the system size is typi- where b is the magnitude of the Burgers
cally constrained by available computa- vector, and q the angle of tilt. Studies by
tional resources. This imposes, for exam- Krakow et al. (1986) suggest that this de-
ple, very large curvatures as driving forces, scription in terms of individual disloca-
and results in high interface velocities tions can be extrapolated, with some mod-
which might not be realistic reflections of ification, to high angles, e.g., to 26.5°.
experimental situations (Sutton, 1995). More generally, the grain boundary region
The chemical aspects accessible to is expected to possess the following char-
Monte Carlo simulation also possess two acteristics:
facets: the thermodynamic equilibrium to a) The high angle grain boundary is thin,
be attained and the kinetics required to perhaps two or three atomic diameters in
reach it. Provided that the interatomic po- width. To a first approximation, it may be
tential is reliable, the Monte Carlo method considered a high energy “phase”, con-
gives an accurate description of chemical strained to have constant volume.
segregation (Treglia and Legrand, 1998). b) The high angle grain boundary is peri-
However, the simulation of the kinetics odic in structure, like its low angle counter-
needed to reach this state encounters a new part. The geometric repeating units are not
problem of coupling the bulk diffusion to
the diffusion close to the interface. Monte
Carlo simulations with time residence al-
gorithms (Martin et al., 1998) have allowed
the inclusion of the role of vacancies in
bulk diffusion, but the treatment of vacan-
cies in solids with free surfaces or internal
interfaces remains a source of problems
(Delage, 1998).
The whole field of surface and interface
thermodynamics and kinetics of metallic
alloys with tendencies to ordering or un-
mixing has profited from Monte Carlo
simulations. The next step (to introduce Figure 7-1. A symmetric low angle tilt grain bound-
simultaneous structural relaxations and ary. The tilt angle q and the edge dislocation spacing
chemical aspects, e.g., via off-lattice l are related by Eq. (7-1).
lattice dislocations, but are capable of de-

scription in terms of nearly perfect com-
pact polyhedra (Ashby et al., 1979). Depar-
tures from symmetric orientations are often
accommodated by mixing two or more
types of structural units within the bound-
ary. It is the view of Sutton and Balluffi
(1987) that no general criterion for low
grain boundary energy can be developed on
the basis of simple geometry.
c) Metastable states, and even degener-
ate states of different structure but identical
energy, may be formed by the relative mi-
croscopic translation of one of the crystals
(Vitek, 1984).
The computed structures are of great Figure 7-2. Top: A schematic concentration (C )
conceptual value, yielding information that profile through a diffuse coherent interface (compo-
nents A and B) in which excess chemical Helmholtz
is inaccessible or accessible only with dif-
energy, represented by the quantity D m , is every-
ficulty through experiment; they should be where balanced against gradient Helmholtz energy
treated with a measure of caution however, (F: Helmholtz energy, m eA and m eB : equilibrium chem-
as the computed results depend to some ex- ical potentials). Bottom: Finite thickness d of equi-
tent on the choice of potential. Further, librium interface, corresponding to concentration
profile.
many (but not all) of these simulations are
performed for a temperature of 0 K. Never-
theless, in combination with experiment,
ters. The first applies to chemically hetero-
they form a valuable tool in the study of
geneous solids, and the second to structu-
this complex field.
rally heterogeneous materials. The first al-
lows us to derive a diffusion equation for the
7.2.2 Structures of Interphase Boundaries
evolution of concentrations, and the other an
These internal surfaces, unlike grain equation for the evolution of order parame-
boundaries, are necessary elements of an ters. These two classes of problems, coupled
equilibrium multiphase system; they are with intermediate situations, cover the range
not defects. They separate phases which in of problems involving relaxed interfaces.
general may differ both chemically and The overall method is to write a Helm-
structurally from one another. holtz energy functional involving not only
The description of an interface using the dependence of the local composition C
continuum concepts dates from the seminal or order parameter y , labeled f (C, y ), but
papers by Cahn and Hilliard (1958) and Al- also those gradient terms compatible with
len and Cahn (1977). The Cahn – Hilliard the symmetry of the problem. This leads to
model of a stationary coherent interphase terms in (—C )2 or (—y )2. The interfacial
boundary is illustrated in Fig. 7-2. The energy between the two regions of different
Cahn – Hilliard and Allen – Cahn treatments prescribed order parameter or composition
deal with two broad cases involving con- is then computed in two steps: the concen-
served and non-conserved order parame- tration or order parameter which minimizes
the excess Helmholtz energy is computed, the interfaces is thus related to the coupling
and this excess energy is then taken as the term weighting the square gradients.
interfacial energy. In a single model we These continuum descriptions are in
therefore get both a “continuum structure” principle well suited for application to dif-
for the interface, and the specific interfa- fuse interfaces. Their facility of implemen-
cial Helmholtz energy. tation in computer simulations (since they
The time-dependent version of these do not involve front tracing methods) has
equations allows us to treat the motion of triggered much interest under the name of
the interface under a prescribed driving “phase field approaches” (Carter et al.,
force, provided that a relation is assumed 1997). They have recently been applied to
between the fluxes and the driving forces. various problems involving sharp inter-
For a conserved order parameter C (such faces using the following systematics: the
as concentration), the kinetic equation interface is artificially made diffuse with
takes the form: respect to a “phase field parameter” j . (For
instance, this parameter can be taken equal
∂C
= M [∇ 2 f ( C ) − k 2 ∇ 4 C ] (7-2) to 1 in a solid and to 0 in a liquid if we wish
∂t to address a solidification problem involv-
where k is a gradient energy coefficient, ing both solute and heat flow (Wheeler et
and for a non-conserved order parameter al., 1992). It can be taken to vary continu-
(such as a long-range order parameter, or a ously from +1 to –1 through a grain bound-
crystal orientation), we obtain: ary, mimicking the orientation change
across the boundary for application to a
∂y
= M [k 2 ∇ 2 y − f ′ (y )] (7-3) problem in grain growth (Chen et al.,
∂t 1998).) A Helmholtz energy equation is
The interfacial Helmholtz energy for a dif- proposed which involves both the concen-
fuse interface between two concentrations tration and the phase field parameter and
C1 and C2 is given by their spatial derivatives. The dynamic equa-
tions for the concentrations and the phase
C2
fields are solved using continuum methods
s = 2 Nv ∫ k 2 f (C ) dC (7-4)
C1
and the coupling term containing the phase
field gradient is set equal to zero so that the
Similarly, assuming a quadratic expan- sharp interface can be recovered.
sion of the Helmholtz energy with two min- These methods have been applied to
ima for order parameters y = 1 and y = – 1 solidification, phase transformations, grain
gives the following expression for the inter- growth, and chemically induced grain
facial Helmholtz energy of an antiphase boundary migration. They possess the ad-
boundary: vantage of relative ease of implementation
using classical numerical techniques, and a
8
s= Nv k 2 f ′′ (7-5) great flexibility in terms of the introduction
3 of such features as anisotropy. They share
In each case, the width of the interface scales the drawback of all Ginzburg – Landau type
as k /f 1/2: the stronger the gradient term com- phenomenological equations, i.e., that the
pared with the local chemical Helmholtz precise atomistic interpretation of both the
energy contribution, the more diffuse will mobility and the coupling terms is not al-
be the interface. The “structural width” of ways clear.
When a structural difference coexists orientation relationship or set of orienta-

with the chemical difference, the situation tion relations, considered to be established
becomes more complex. However, if the at nucleation. In these cases, we search for
difference between phases is small, as is a reason for the strong anisotropy of inter-
the case for structurally similar phases with facial Helmholtz energy commonly ob-
similar orientations and lattice parameters, served in such systems. The most plausible
the differences may be accommodated by approach involves the search for optimum
an elastic component or by an array of mis- matching of atomic positions at the inter-
fit dislocations. In the case of g precipitates face, that is a maximum in coherency, cor-
in b brass, the interfacial energy has been responding to a minimum in the structural
taken as the sum of structural (dislocation) component of the energy. A particularly
and chemical energies. This approach per- simple example is found in the f.c.c./h.c.p.
mits the prediction of weak anisotropies, system Al – Ag, where the (111) f.c.c. habit
for example those in Fig. 7-3, after Ste- plane corresponds to the structurally simi-
phens and Purdy (1975). lar (001) h.c.p.
More complex systems, e.g., interfaces Figure 7-4 illustrates a model of the
joining f.c.c. and b.c.c. phases, are gener- f.c.c./b.c.c. interface due to Rigsbee and
ally discussed with respect to a particular Aaronson (1979), in which a macroscopi-
cally irrational habit plane interface is
made up of rational areas with a high de-
gree of coherency, separated by “structural
ledges” which serve to displace the inter-
face plane by a few atomic dimensions,
thereby to increase the degree of cohe-
rency. Aaronson and his co-workers em-
phasize that these structural ledges are con-
sidered intrinsic to the structure of the
interface, and do not provide a mechanism
for interface motion. Other studies have
utilized a more formal geometrical ap-
proach (Bollman, 1970; Dahmen, 1981;
Zhang and Purdy, 1993 a, b) to seek inter-
faces of optimal matching, with similar
conclusions. Based on evidence from high
resolution electron microscopy (e.g.,
Zhang et al., 1998), it has become apparent
that the typical interface between phases of
dissimilar structure is often faceted on a
microscopic scale, and that invariant
Figure 7-3. The morphologies of equilibrated g pre- planes or invariant lines of the transforma-
cipitates (solid curves) formed in b brass, and the tion, when they exist, are clearly important
corresponding Wulff plots (dashed), for the tempera-
in the selection of precipitate habits.
tures indicated. The anisotropies are consistent with
a model that treats the interfacial Helmholtz energy
as a sum of structural (dislocation) and chemical
terms (Stephens and Purdy, 1975).
Figure 7-4. Rigsbee and

Aaronson’s model for the
f.c.c./b.c.c. interface, in
which structural ledges are
introduced which maximize
the degree of coherence
within the interface. Re-
printed with permission
from Rigsbee and Aaronson
(1979), copyright Pergamon
Press.
7.2.3 Segregation to an Interface (s B − s A ) Vmb La − Lb

ln s = + (7-8)
In system of two or more components, it RT d RT
is often found that the equilibrium compo- where sA and s B are the specific grain
sition at an internal surface is different boundary Helmholtz energies in pure A
from that of the surrounding phase(s). As and B, Vmb the molar volume of the boun-
before, it is necessary to distinguish be- dary “phase”, and La and Lb the regular so-
tween grain and interphase boundaries; for lution parameters for the crystal and boun-
the case of a grain boundary in a two-com- dary phases, respectively.
ponent solid solution, we may consider that A more comprehensive treatment of
the high energy defect has its own thermo- phase equilibria of grain boundary struc-
dynamic solution properties. A constrained tures is given by Cahn (1982). The pos-
equilibrium is then possible between the sibility of faceting transitions is high-
boundary “phase”, which must have an ap- lighted in this work and supported experi-
proximately fixed volume, and the adjacent mentally by the studies of Ferrence and
crystalline phase (a). The constant volume Balluffi (1988) and Hsieh and Balluffi
condition requires, for a virtual change in (1989).
mole number n aA ,
dn aA = – dn A
b
; b
dn A = – dn Bb ;
b a
dn B = – dn B (7-6)
and the constrained equilibrium is defined
by a line tangential to the a Helmholtz-en-
ergy curve at the bulk composition C0 , and
a parallel line tangential to the boundary
Helmholtz-energy curve (Hillert, 1975 a).
This is illustrated in Fig. 7-5. We then ob-
tain for the distribution coefficient Figure 7-5. Graphical interpretation of the con-
strained equilibrium between a crystalline solid solu-
CBb tion, a, and a grain boundary phase, b. (For details
s= (7-7)
CBa see text.)
For interphase boundaries, segregation Grain boundary diffusion is commonly

is also expected, as first shown by Gibbs investigated experimentally in one of two
(1906). His expression for the relationship ways:
between the rate of change of interfacial
Helmholtz energy with bulk composition a) “Random” polycrystalline specimens
and the interface excess solute concentra- are exposed to a source of diffusant (e.g.,
tion, GB , the Gibbs absorption isotherm, a planar plated or implanted layer), an-
may be written as nealed, then serially sectioned. If the dif-
fusant is a tracer, the technique can be
ds = – GB dm B (7-9)
used to give a sensitive evaluation of the
– –
where GB is the surface excess concentra- product D bd, where D b is the average
tion of solute B per unit area of interface. grain boundary diffusion coefficient, and
(The solvent surface excess is zero by defi- d the boundary thickness, assumed to
nition.) represent a region of uniformly enhanced
A growing body of experimental evidence diffusivity. If a solute is diffused into the
demonstrates that both grain boundaries polycrystal, the usual experimentally ac-
–
and interphase boundaries are generally cessible quantity is the product {s D b d }
sites for complex interactions among chem- (Levine and MacCallum, 1960). It is only
ical species. Guttman (1995) has reviewed in exceptional cases, where the volume dif-
the elementary mechanisms of intergranu- fusion contribution to some part of the pen-
lar segregation in multi-component sys- etration profile is negligible (type “C” be-
tems, and concludes that site competition, havior) that we can obtain in principle a
as well as attractive and repulsive chemical more direct evaluation of the boundary dif-
interactions, are all important in determin- fusivity.
ing the degree of interfacial segregation. b) Bicrystals of controlled misorienta-
tion are sectioned such that the grain boun-
dary plane is perpendicular to a source of
7.3 Interfacial Diffusion diffusant, which may be of constant com-
position, or finite. In most cases, as above,
Solid – solid interfaces often provide ex- both volume diffusion and boundary diffu-
ceptionally rapid paths for diffusion. Pro- sion will contribute to the penetration pro-
vided that there is a significant structural file, as schematized in Fig. 7-6. In this
difference, so that there is a disordered re- more controlled case, the mathematical
gion linking nearly perfect crystalline re- solutions of Whipple (1954) or Suzuoka
gions, this is easy to understand in broad (1964) (depending on whether the source is
terms. The mechanisms and the details of fixed in composition or finite) are used to
grain and interphase boundary diffusion unfold the grain boundary transport coeffi-
are not fully understood at present, al- cient from serial sectioning data. The dif-
though there exists a wealth of experimen- ference here lies in the opportunity to cor-
tal information on the former. The reviews relate misorientation (and, by inference,
by Peterson (1983) and Kaur and Gust structure) with diffusion rates. Again, a
(1988) for grain boundary diffusion, and large amount of precise data has been ac-
that of LeClaire (1986) for the related quired, mainly for the special case of the
problem of diffusion along dislocations are symmetric tilt boundary. This was re-
recommended. viewed by Peterson (1983).
7.3 Interfacial Diffusion 491
e) The mechanism of grain boundary dif-

fusion in close-packed metallic systems
is thought to involve vacancy exchange.
This is a tentative conclusion based on
the experimental evidence of Martin et al.
(1967), who studied the pressure depen-
dence of grain boundary diffusion and de-
duced activation volumes from their data,
and of Robinson and Peterson (1973), who
studied isotope effects in silver polycrys-
tals and bicrystals, and demonstrated that
Figure 7-6. Schematic composition (C) distribution their results were consistent with a vacancy
in the neighborhood of a grain boundary, normal to a mechanism. In their computer simulation
surface of fixed composition, C0 . d : boundary thick- studies, Balluffi et al. (1981) and Faridi
ness. and Crocker (1980) have provided support
for this idea by showing that the formation
From the consideration of grain bound- and migration energies for vacancies and
ary diffusion data for metals, we can draw interstitials in the grain boundary favored a
a few general conclusions: vacancy exchange process.
a) The activation enthalpy for grain Having summarized briefly the present
boundary self-diffusion is of the order of level of understanding of grain boundary
half that for volume diffusion. Indeed, a diffusion, it is perhaps worth emphasizing
good empirical correlation exists between that the correlation of structure with diffu-
grain boundary diffusivity and reduced sivity remains incomplete, and that much re-
temperature T /Tm for different metal struc- mains to be learned about the fundamental
tures. For f.c.c. metals, Gust et al. (1985) processes of interfacial diffusion in solids.
obtained (7-10) Interphase boundary diffusion is less
– well studied, to the extent that very few
d D b = 9.7 ¥ 10–15 exp (– 9.07 Tm /T ) m3/s
reliable data are available in the litera-
b) For the low angle symmetric tilt ture. Kaur and Gust (1988) cite only one
boundary, rates of diffusion are aniso- system for which extensive measurements
tropic, with faster diffusion occurring par- have been reported, the Sn – Ge/In system
allel to the dislocation lines. Undissociated (Straumal et al., 1981). One reason for the
dislocations have higher diffusivities than lack of direct information about this impor-
dissociated ones; the diffusivity increases tant aspect of interfacial diffusion may lie
linearly with dislocation density (and therein the fact that, unlike grain boundaries, in-
fore with tilt angle q ) for angles up to terphase boundaries can support local equi-
about 10°. librium concentration gradients only in the
c) Anisotropic behavior is also obtained presence of gradients of curvature, temper-
for higher angle symmetric tilt boundaries. ature or stress. The isothermal planar inter-
d) Low angle twist boundaries, com- phase boundary is normally isoconcentrate.
posed of screw dislocations, are less effi- Mullins (1957) has considered a number
cient in transporting material than their tilt of important cases of interfacial diffusion
counterparts. in response to variations in curvature. For
the case of substitutional diffusion, and Helmholtz energy s A, where s is the spe-
under the simplifying assumptions of con- cific interfacial Helmholtz energy and A
stancy of composition of the two phases a the interfacial area. For constant s , the
and b and site conservation at the interface, force can be expressed as
we may write for the interfacial flux:
ps = K s (7-12)
A
Jab =− B
Jab =− Lab ∇ (m ab
A
− B
m ab ) which is the product of the surface Helm-
Lab Vm s ab holtz energy and the interface curvature
=− ∇Kab (7-11) K = dA/dV.
ca − cb
For smoothly curved interfaces the curva-
Here, L represents the mobility of atoms ture is the sum of the reciprocals of the two
along the interface, Vm is a molar volume, principal radii of curvature [(1/r1) + (1/r 2 )].
the m’s are chemical potentials within the For strongly anisotropic or faceted inter-
interface, s ab is the specific interfacial faces, an effective capillary force may still
Helmholtz energy and K ab is the interfacial be defined, through the derivative dA/dV, al-
curvature. though care must be taken in its formulation.
For the special case of an equilibrium
particle, the capillary force gives rise to
7.4 Forces for Interface Migration a variation of the specific Helmholtz
energy of the included phase: the resulting
It is useful to define and classify a set of relation between the zero-curvature equi-
generalized “forces” for interface migra- libria (C0 ) and the equilibrium composi-
tion. In most cases, the driving and retard- tions in the presence of curvature is sum-
ing forces are expressed in terms of a nor- marized in the Gibbs – Thompson relation-
mal force per unit area of interface, and ship. For two components we obtain the
therefore have the units of pressure. linearized expressions (Purdy, 1971)
It is conventional to separate the various
forces according to their origin (e.g., surface C0a (1 − C0a ) Vm ⎛ K s ⎞
DC a = (7-13 a)
Helmholtz energy, elastic energy, chemical ( C0a − C0b ) ⎜⎝ RT eb ⎟⎠
Helmholtz energy) and to some extent the
division is arbitrary. Its utility lies in the C0b (1 − C0b ) Vm′ ⎛ K s ⎞
DC b = (7-13 b)
ability to visualize the interplay of different ( C0a − C0b ) ⎜⎝ RT eb ⎟⎠
forces on a moving transformation interface.
It is also true that the definition of the where a is the included phase and b the
“force” for boundary migration is only a external phase, Vm is the molar volume of
part of the solution to the migration prob- the a phase, V m′ the molar volume of the
lem. The response function (or mobility) of a phase extrapolated to the composition of
the interface must also be known, and this ⎛ ∂ ln g b ⎞
the b phase and e b = ⎜ 1 + , where
∂ ln Cb ⎟⎠
function is expected to differ greatly from
one interface type to the next. ⎝
g b is an activity coefficient. This last term
is a measure of the departure of the parent
7.4.1 Capillary Forces
phase from ideal solution behavior.
These forces originate from the system’s An equivalent graphical construction
tendency to reduce its total interfacial due to Hillert (1975 a) is shown in Fig. 7-7.
7.4 Forces for Interface Migration 493
7.4.2 Chemical Forces

The chemical force for interface migra-
tion should be distinguished from the
chemical driving force for the transforma-
tion DFch , defined by Eq. (5-9) in the chap-
ter by Wagner et al. (2001). In this section
the case where the chemical force acts di-
rectly across the transformation interface is
considered. Its magnitude is given by
DFch′
pch = (7-15)
Figure 7-7. Effect of interfacial curvature, K, on the Vm
two-phase equilibrium in a binary system; the com-
mon tangent line, which generates equilibrium com- where DFch ¢ is the molar Helmholtz energy
positions at its points of tangency for the planar difference acting directly across the inter-
phase interface, is replaced by two nonparallel tan- face, and is generally less than DFch , the
gent lines, whose intercepts differ by the amounts Helmholtz energy difference accompany-
shown.
ing the complete transformation. The force
is derived through a virtual work argument,
in which the energy change in displacing
The construction is of great conceptual
the interface through unit normal distance
value, and permits the straightforward ex-
is equated to a work term.
tension to the case of three components
The distinction between total and local
with different partial molar volumes. Mor-
chemical driving forces is best seen in the
ral and Purdy (1995) have also developed
composition-invariant transformation il-
compact expressions for the general multi-
lustrated in Fig. 7-8. Here, the Helmholtz
component case. For example, the solubil-
energy difference for interface migration is
ity change of the included a phase (a
given by DF ¢, and some fraction DF 0 of
column vector |DC bj |) with curvature can
the total available Helmholtz energy is dis-
be written in terms of second derivatives
sipated in the volume diffusion field. In the
∂2 G j often assumed limit of local equilibrium,
of the Gibbs energy, , of the par-
∂Ck2 DF ¢ is implicitly set equal to zero, and all
ent b phase, the molar volume of the a of the Helmholtz energy difference for the
phase, Vm , and the equilibrium concentra- transformation is used in the volume diffu-
tion differences between the a and b sion field.
phases, the row vector |DC jab |: 7.4.3 Mechanical Forces
∂2 G j These forces fall into two categories:
DC ab
j DC bj = K s Vm (7-14) those that may simply be treated as part of
∂Ck 2
the chemical Helmholtz energy of the
For faceted equilibrium shapes, the cur- strained phase(s) and those that cannot.
vature may be replaced by a constant term
s i /r i , where the subscript indicates facet i, 7.4.3.1 Elastic Strain Energy
and r i is the minimum radial distance to the As noted by Wagner et al. (2001, Chap. 5
ith facet. of this volume), elastic strain energy is an
The strain energy density in a coherent

solute profile will be given as an intensive
variable by Eq. (5-12), which therefore has
application to the diffusional growth of a
precipitate in a binary solid solution, and
also to the problem of the stability of thin
epitaxial deposited layers.
For the more complex case of the strain
energy associated with the formation of a
second phase inclusion, Eq. (5-14) holds,
independently of inclusion shape, when the
transformation strain is a pure dilation and
the elastic constants of the two phases are
similar. For spherical coherent transformed
regions, the combined effects of surface
energy and (dilatational) transformation
strain may be incorporated in the Gibbs –
Thomson equation by adding 6 H e 2 gB to
the numerator of the term in braces in Eqs.
(7-13). Here H = 3 k a /(3 k a + 4 gB ) where
k a is the bulk modulus of the precipitate,
gB is the shear modulus of the matrix, and e
is the misfit (Rottman et al., 1988).
Figure 7-8. Helmholtz energy and interfacial com- 7.4.3.2 Plastic Response
position relationships for a composition-invariant
phase transformation. The available Helmholtz en- It is important to distinguish between
ergy for the transformation is used in part to drive the mechanically derived forces which origi-
interface, DF ¢, and the balance, DF 0, is dissipated in
nate from elastic strain energy, and those
the volume-diffusion field.
which result in the plastic relaxation of the
loaded specimen. In the latter case, the
additive term in the Helmholtz energy of a elastic energy may remain approximately
phase. Thus, for homophase variations, the constant during the course of the transfor-
local elastic energy density is given by mation, but the interfaces can experience
h 2 DC 2 Y as indicated in Eq. (5-12). virtual forces, in the sense that the Helm-
This term generally reduces the total holtz energy of the loading system is re-
driving force for a phase transformation. In duced by interface motion.
Eq. (5-12), DC is the local difference from As an example, consider the symmetric
the average concentration C0 . The equation tilt boundary of Fig. 7-9, which is capable
therefore has application to any case where of motion in response to the applied load.
a gradient of misfitting solute exists in an Studies of this type of boundary have
otherwise perfect crystal, and in particular shown that the dislocation boundary under-
to the case where a thin layer of altered goes normal migration, that is, synchro-
composition is in coherent contact with an nous glide, under an appropriate load. We
infinite bulk phase of composition C0 . can consider that there is a virtual mechan-
Another class of strain energy stored in

bulk material is that associated with struc-
tural defects such as dislocations. These
defects may be generated and stored during
plastic deformation, and their effect on the
Helmholtz energy of the solid provides the
basis for the driving force for recrystalliza-
tion. Dislocations may also occur due to
plastic relaxation of misfit stresses (as seen
in some cases of discontinuous precipita-
tion or in some transformations in steels).
The excess Helmholtz energy is tradi-
tionally taken to be proportional to the total
dislocation density and is added as a bulk
driving force in the overall thermodynamic
balance. This is probably acceptable for
Figure 7-9. A tilt wall (tilt angle q ) is subject to a
misfit-generated dislocations but, for the
virtual force, p, derived from the work done on the case of dislocations stored during plastic
specimen by the loading system. deformation, this approximation neglects
the formation of dislocation substructures
such as subgrains. This relaxation phenom-
ical force on the interface equal to pw = W enon decreases the stored Helmholtz en-
tan q where W is the applied load per unit ergy and renders its increase with disloca-
area and q is the angle of tilt. In this exam- tion density less than linear due to self-
ple, the applied stress can be thought of screening effects (Verdier et al., 1997).
as acting directly on the interface. The ex- Again the question of the contribution of
perimental response of the symmetric tilt this stored energy to the overall kinetics of
boundary is linear, and a simple statement microstructure development is two-fold:
of the mobility of the boundary summa- we may consider the nucleation step from
rizes the kinetics of boundary motion. an initially homogeneous situation, or the
If a coherent twin boundary were to re- growth regime for which the stored energy
place the symmetric tilt boundary in Fig. is different on either side of the moving
7-9, the transformation strain would again interface. A simple approximation for the
be a shear. It is clear, however, that the syn- growth process would be the addition of an
chronous glide process would not occur. extra driving force when the stored ener-
The propagation of the twin will require gies differ on either side of the interface
the lateral motion of twinning dislocations, (e.g. due to different stages of deformation
and the interface response function will be in two neighboring grains). However, this
fundamentally different. This simple ex- approach is unlikely to be sufficient to ex-
ample illustrates the general principle that plain the nucleation of recrystallization,
a knowledge of the force for boundary mi- which is known to occur at grain boundar-
gration is only part of the empirical de- ies, and at localized shear or transition
scription of boundary motion; evidently the bands (Humphreys and Hatherly, 1995). As
response characteristics of the boundary proposed initially by Bailey and Hirsch
must also be known. (1962), nucleation seems to be a random
event associated with local fluctuations in boundary migration are not fully under-
subgrain configuration; it is unlikely to be stood.
successfully modeled either by a standard When two or more components are in-
nucleation theory or by an averaging ap- volved, this picture must be extended.
proach for the stored energies. Information Superimposed on the intrinsic response of
on local misorientations (leading to differ- the interface, and frequently masking it, we
ences in local mobilities and local driving find the effects of solutes which are prefe-
forces) may be crucial, as has been sug- rentially attracted to or repelled from the
gested using vertex-based numerical simu- interface. The experiments of Aust and
lations (Weygand, 1998). Rutter (1959) first demonstrated conclu-
sively the powerful retarding effects of
trace amounts of solute on grain boundary
7.4.4 Frictional Forces
migration.
7.4.4.1 Solute Drag The idea of a “solute drag” perhaps orig-
inated with Cottrell (1953), who noted that
Each type of mobile interface will pos-
solute atoms would be attracted to disloca-
sess its own characteristic response func-
tions, and would therefore be required to
tion. For the simple tilt wall subject to a
diffuse along with the dislocations, or to be
mechanical force, we have noted that the
left behind by dislocations that had broken
force – velocity relationship is linear over a
free of their solute atmospheres. Similar
range of forces. If a considerable recon-
ideas were put forward by Lücke and his
struction of the crystal structure is required
co-workers (1957) for application to the
for the motion of an interface, as in the mo-
case where a solute is preferentially segre-
tion of a high-angle boundary in a single-
gated to grain boundaries. The most com-
component system, we expect that cross-
plete and informative theoretical treat-
boundary diffusive motion will play a sig-
ments of this effect are those of Cahn
nificant role. Turnbull’s (1951) expression
(1962) and Hillert and Sundman (1976).
for the intrinsic mobility, M, of a grain
The latter is based on a treatment put for-
boundary
ward earlier by Hillert (1969). The basic
d D b Vm reason for the solute drag force is simple: if
M= (7-16) solute is attracted to the boundary, it will
b 2 RT
tend to diffuse along with the moving
is based on a model in which atom jumps boundary. Depending on the relative rates
across the boundary are independent of one of boundary motion and solute diffusion,
another. This relation has not been con- the solute distribution may become asym-
firmed by experiment, and this is taken to metric, with more solute trailing the
suggest that more complex, cooperative boundary than leading it. Because of the
movements of atoms are generally in- mutual attraction between the boundary
volved in interface migration. Again refer- and the solute, the asymmetric solute dis-
ring to experiment, there exists a consider- tribution leads to a retarding force of the
able database drawn from measurements form
of grain boundary motion in high purity
pi = NV Ú (C – C0 ) (dE/dx) dx (7-17)
metals. Thus the intrinsic response of
high-angle grain boundaries is rather well where NV is the number of atoms per unit
documented, even if the mechanisms of volume, and dE/dx is a measure of the
force between the solute atoms and the is not restricted to dilute solutions, nor is it
boundary. The effects of small amounts of restricted to grain boundaries. Figure 7-11
solute can be profound. The profiles in Fig. uses the simple square well to illustrate the
7-10 are due to Cahn (1962). They are effect of cross-boundary diffusion in set-
based on a triangular interaction potential ting the degree of asymmetry of the solute
between solute and boundary, and corre- profile. The drag term may be written in
spond (a) to a slowly moving boundary, (b) several ways. Among the most convenient,
to a boundary experiencing maximum so- Hillert and Sundman (1976) find:
lute drag, and (c) to a fast-moving bound-
ary. That is, the drag force goes through a RT v ( c − c0 )2
pi =
Vm
∫ Dc (1 − c ) dx (7-18)
maximum as the velocity is increased. If an
initial intrinsic force – velocity relationship In their approach, the ramps (zones 1 and 3
is assumed linear, and the solute drag term in Fig. 7-12), as well as the centre of the
is added, the total response function is boundary (zone 2) and the parent phase re-
quickly rendered nonlinear. Under more gion (zone 4), contribute to the drag. In-
severe conditions, the system can even be- deed, it is possible to separate out the con-
come unstable, displaying a region of ve-
∂p
locities where is negative.
∂v
Hillert’s formulation uses a different
interaction potential, initially square
(rather than triangular), and later a square-
topped potential with ramps at the edges.
The drag is evaluated as a rate of dissipa-
tion of Helmholtz energy due to diffusion
in and in front of the moving interface. The
model lends itself to numerical analysis; it
Figure 7-10. Computed solute profile in the vicinity

of a moving grain boundary: a) at equilibrium, b) at a
velocity corresponding to maximum solute drag, and Figure 7-11. A square-well model for the moving
c) at higher velocity. After Cahn (1962). grain boundary. After Hillert (1969).
interface from moving: a pinning force.

The idea of pinning a grain boundary with
foreign particles was put forward initially
by Zener (Smith, 1948). The grain bound-
ary will be able to move only if the gradient
of bulk stored energy across the interface
is sufficient to move the interface to an
untrapped configuration. Zener assumed a
purely geometrical distribution of particles
along the grain boundary; this led to a fric-
tion force to be overcome that scales as
Vf /r, where Vf is the volume fraction of
particles and r their radius. A more refined
Figure 7-12. The Hillert – Sundman (1976) model of approach due to Hazzledine and Oldershaw
a moving boundary, showing the interaction profile, (1990) allows us to treat the collective pin-
and some computed solute profiles for v d /2 D = 0.001
(1), 1.61 (2), and 10 (3).
ning problem in a consistent manner (i.e.,
to treat both the flexibility of the interface
and the number of particles along the inter-
tribution of any slice taken parallel to the face). Hazzledine’s result, recently con-
interface, and Sundman and Hillert have firmed by computer simulations (Weygand,
presented their results in this way, as com- 1998), indicates that a scaling law in Vf1/2/r
ponents of the total solute drag. They find, might be more appropriate.
as we might expect, that the drag is reduced The consequence of these pinning forces
for given velocity by increasing the cross- on the overall kinetics of interface migra-
boundary diffusion coefficient, or by re- tion can be two-fold: on the one hand, the
ducing the depth of the potential well seen particles will impose a threshold driving
by the solute. The model is capable of gen- force for boundary migration; and on the
erating a variety of informative results. It other hand, they will impose a constant re-
is not restricted to grain boundary motion, tarding force to be subtracted from the
but also finds application to steady state available Helmholtz energy. The standard
phase transformations, as will be seen later. method is to equate the retarding force with
It is interesting to note that solute drag the threshold force, but this identification
effects have been (in some sense) better is by no means obvious. Numerical experi-
studied and quantified than intrinsic bound- ments on grain growth kinetics do however
ary friction forces. The solute effects are indicate that this simple approach gives a
thought to be additive, and to dominate and good description of the overall kinetic be-
even to mask the intrinsic properties of havior (Weygand, 1998).
grain boundaries for many cases of practi-
cal interest.
7.5 Examples
7.4.4.2 Particle Pinning 7.5.1 The Massive Transformation
Whereas the contribution of solute drag Massive transformations are considered
as a retarding force is by definition of a vis- to include all inhomogeneous, noncoher-
cous nature, another term may prevent the ent, thermally activated, composition-in-
7.5 Examples 499
variant, solid – solid transformations (Mas-

salski, 1958). Massive solidification reac-
tions are excluded by this definition, but
their literature is relevant, in part because
of advances in understanding of the rapid
solidification of silicon-based alloys (Aziz
and Kaplan, 1988).
The massive transformation generally in-
volves a major structural change; it normal-
ly begins with the heterogeneous nuclea-
tion of a thermodynamically stable or meta-
stable daughter phase, at imperfections or
grain boundaries in a supercooled parent, and
proceeds by the thermally activated migra-
tion of a mobile transformation interface.
Plichta et al. (1984) reviewed the avail-
able information on massive transforma- Figure 7-13. Helmholtz energy relationships for the
tion nucleation, and concluded that the allotropic transformation of a pure element.
nucleation event is structurally identical
with that expected for diffusional transfor-
mations among dissimilar phases; grain For two components, the composition-
boundary and triple junction sites are ener- invariant condition need not apply to mas-
getically preferred, and an orientation rela- sive nucleation (although it must hold
tionship between the nucleus and at least overall for massive growth). In the exam-
one parent grain is generally set on nuclea- ple in Fig. 7-14, drawn for a temperature
tion. Since interfacial torques are likely to below T0 , the temperature for which the a
be present during the nucleation event, the and b phases of the bulk composition C0
construction of Hoffman and Cahn (1972) have the same molar Helmholtz energy, it
may be expected to provide guidance in the is clear that the formation of a nucleus of
modeling of the critical nucleus. Experi- composition C0 would be accompanied by
ence has shown, however, that the search a volume Helmholtz energy change
for plausible nucleus shapes in such heter- DF1 /Vm . However, the volume Helmholtz
ogeneous systems seldom leads to unam- energy change is maximized (DF2 /Vm ) if
biguous results. the nucleus takes composition C ¢. Even in
The chemical Helmholtz energy change, the single-phase supersaturated region, a
DFch , for nucleation in a single-component similar argument shows that the volume
system that undergoes an allotropic trans- Helmholtz energy change will be max-
formation is obtained directly from the imized for a composition other than C0 .
undercooling below the equilibrium tem- Turning to the process of massive prod-
perature, as suggested by Fig. 7-13. For uct growth, if we again consider the allo-
small undercoolings, DT, this may be ex- tropic transformation of pure element, it is
pressed as clear that the rate of transformation is de-
termined by the undercooling (which sets
DFch = DH DT /T0 (7-19)
the interfacial driving force pch ), and by the
with DH as the change of enthalpy. migration characteristics of the transforma-
Figure 7-15. Diffusional and diffusionless growth

of a from g iron, calculated assuming that carbon dif-
fusion is negligible at high growth rates. Reprinted
with permission from Hillert (1975 b), p. 12.
Figure 7-14. Helmholtz energy relationship for the

isothermal nucleation of a massive product in a bi- For richer alloys, massive growth could,
nary system. in principle, be entirely composition-invar-
iant. However, some results (Singh et al.,
1985) suggest that solute diffusion ahead
tion interface. As in the case of the high an- of the interface can play a major role in
gle grain boundary, the details of the cross- steady interface migration. Referring to
boundary atomic transfer process are not Fig. 7-16, which is based on the simple
well understood. As in the grain boundary model of an infinitely thin interface, it is
case, it has proved difficult to separate the clear that a solute build-up in front of the
effects of trace impurities from the intrinsic interface will result in a reduction of the
migration properties of the noncoherent chemical force on the interface. Having
interphase boundary. Hillert (1975 b) has conceptually released the concentration in
estimated the mobility of grain boundaries the interface region from its value far from
in pure iron from a series of limiting argu- the interface (C0 ), the description of two-
ments as component massive growth becomes a free
boundary problem, in which the interfacial
M = 0.035 exp (– 17 700/T ) m4 /(J s) (7-20)
driving force and the velocity must be si-
He then used this expression in his anal- multaneously determined. It is likely that
ysis of the rates of the a/g transformation this type of approach to local equilibrium
of iron with very low carbon contents, as occurs at higher relative temperatures,
investigated experimentally by Bibby and where the possibility of volume-diffusion
Parr (1964) and Ackert and Parr (1971). loss is strongest, and where, for example,
Using a model in which a fraction of the Widmanstätten growth competes with the
chemical force is dissipated by solute dif- massive reaction.
fusion ahead of the interface, Hillert was Much of our current knowledge of mas-
able to predict a number of characteristics sive growth derives from the pioneering
of the massive transformation in pure and work of Massalski and co-workers (Mas-
nearly pure iron, as summarized in Fig. 7-15. salski, 1958; Barrett and Massalski, 1966).
7.5 Examples 501
38 at.% Zn alloy, which had previously

been quenched to retain the b phase. They
found:
i) that the transformation occurred only
within the single phase a region of the
phase diagram and
ii) that steady massive growth (at con-
stant isothermal velocity of order 1 cm/s)
developed after a delay time of several mil-
liseconds.
Both the limited temperature range for
growth and the delay time were attributed
to the pre-existence of solute fields around
small a particles. Only in the single-phase
a region would such solute fields be con-
sumed during initial growth, thereby per-
mitting the development of composition-
invariant (massive) growth. They noted
that massive growth in the two-phase a + b
region (below T0 ) is possible, provided that
some massive a has first been formed by
Figure 7-16. Solute build-up in front of the massive pulsing into the single-phase region. With
interface will result in a reduction of the chemical further heating these massive regions con-
force for interface migration. tinued to growth into the a + b region, thus
reinforcing the idea that the inhibition of
such growth lies in the initiation stage.
In early papers, it was implicitly assumed In the analysis of growth rates, it was as-
that a pseudo-unary cross-boundary diffu- sumed that massive growth occurred with-
sion process controlled the rate of massive out local composition change. However, in
growth; later (Massalski, 1984) it was discussion, impurity drag effects and par-
noted that local changes of composition in ent-phase solute fields were admitted as
the region of the interface, consistent with possibilities.
overall composition invariance, are pos- A comprehensive theoretical treatment
sible, even probable, under certain condi- of the two-component massive transforma-
tions. tion, which applies to a range of transfor-
Because of the speed of transformation, mation conditions, will deal with solute
massive reactions are often studied under diffusion within the interface and also in
conditions of continuous cooling. The most the thin region ahead of it. The model of
informative kinetic studies, however, are Hillert and Sundman (1976), an extension
those that yield rates of isothermal trans- of their treatment of solute drag at grain
formation, as exemplified by the work of boundaries, finds application here. Their
Karlyn et al. (1969); they used pulse-heat- formulation of the solute drag involves the
ing and rapid quenching to evaluate the computation of the dissipation of Helm-
rates of formation of massive a in a Cu – holtz energy due to diffusion within the
interface region (Eq. (7-18)). The solute Helmholtz energy defines the most prob-
field is schematized in Fig. 7-17 (a) for a able velocity. A second treatment of the so-
solute that is attracted to the interphase lute drag in massive growth (Bréchet and
boundary. The cross-boundary diffusion of Purdy, 1992) extended Cahn’s (1962) anal-
solute introduces a component of drag (re- ysis, and demonstrated that a finite drag
gions 1, 2 and 3), and diffusion ahead of term will always be present, however small
the boundary provides a further contribu- the velocity. This analysis, like that of Hil-
tion (region 4), which dominates the total lert and Sundman, therefore suggests that a
drag as the velocity approaches zero. The threshold driving force is a natural charac-
vd teristic of such transformations. However,
total drag is optimized for ≈ 1, and it
2D if the solute field in front of the interface is
may be that the optimum of dissipation of accompanied by a misfit, in the sense of
Eq. (5-12), we expect an additional term in
the force balance, a “pulling” force due to
the elastic energy contained in the coherent
composition gradient.
The above discussion is based on the
premise that the migration characteristics
of the massive front are those of a non-co-
herent interface. This is borne out by nu-
merous observations (Massalski, 1984)
which indicate, for example, that the trans-
formation interface is able to cross grain
boundaries in the parent phase without a
change in velocity or morphology. Dy-
namic observations indicate, however, that
the motion of the front is irregular, and of-
ten accomplished by a lateral process, in
which diffuse steps move parallel to the
interface plane (Perepezko, 1984). In the
view of Menon et al. (1988), lateral pro-
cesses are the rule in massive growth.
Hence models based solely on diffusion
normal to the boundary may need to be
modified. However, it is interesting that
Perepezko (1984) demonstrated a scaling
relationship for a wide range of alloy
systems, which yields a composite en-
Figure 7-17. a) Schematic concentration profile thalpy of activation for massive propaga-
through a massive transformation interface, i) at tion of 94 Tm J/mol. This value is similar to
equilibrium, and ii) in motion. b) Computed solute that for grain boundary diffusion, which
drag for the massive transformation in part a, for a suggests that diffusion within the interface
constant value of D. XA· and X B· represent the initial
composition of the parent phase expressed as mole is a common rate-determining feature for
fractions of the two components A and B. After Hil- all such transformations.
lert and Sundman (1976).
7.5 Examples 503
7.5.2 Chemically-Induced Grain

Boundary Migration
Chemically-induced grain boundary mi-
gration (CIGM), or diffusion-induced grain
boundary migration, is a rather recently
recognized phenomenon (den Broeder,
1972; Hillert and Purdy, 1978). The pro-
cess is one in which diffusion along a grain
boundary, to or from a sink or source of so-
lute, causes the boundary to move. The
source may be in the solid, liquid or gas
state. A composition change results (Fig.
7-18).
The process is widespread in binary me- Figure 7-19. Grain boundary bowing between pre-
tallic systems. King (1987) lists 30 systems cipitates, which act as solute sinks, and serve to pin
in which it has been detected. The process the grain boundary.
is capable of generating substantial inter-
mixing where little or none might be ex-
pected in the absence of grain boundary
CIGM is often seen as symmetry break-
motion.
ing. The initiation process is not well docu-
A variation on the conditions in Fig.
mented or understood. It is clear that diffu-
7-18 is obtained for a supersaturated solid
sion along an initially stationary boundary
solution: an initially planar boundary bows
will lead to symmetric diffusion profiles,
out between grain boundary precipitates,
as illustrated in Fig. 7-6, and this has led to
and simultaneously sweeps solute to the
the suggestion that the region next to the
precipitates (Fig. 7-19). This type of mi-
boundary will become one of increased
crostructural development was reported by
elastic strain energy, provided that a solute
Hillert and Lagneborg (1971), and was re-
misfit exists, and provided that the solute-
cently supported by microanalytical evi-
enriched or solute-depleted regions remain
dence for solute depletion in the volume
coherently connected with their respective
swept by the moving boundary by Solor-
grains. The strain energy density will then
zano et al. (1984).
be given by Eq. (5-12). Because the elastic
constant Y is a function of orientation, the
strain energy density will in general be dif-
ferent on the two sides of the boundary, and
this may lead to boundary displacement.
Tashiro and Purdy (1987), in a survey of
binary metallic systems, found only one
system in which CIGM could not be in-
duced; this was the system with the lowest
misfit, Ag – Mn.
Figure 7-18. Schematic representation of chemi-
cally-induced grain boundary migration in a thin Further evidence for the importance of
sample exposed to a vapor source of solute. The front solute misfit in the initiation of CIGM is
is often observed to bow against its curvature. found in the work of Rhee and Yoon
(1989), who systematically varied the mis- and unstrained product crystal. These con-
fit parameter in a ternary system, and siderations are contained in the phenomen-
showed that the phenomenon is suppressed ological treatment due to Bréchet and
when the misfit is brought to zero. An ap- Purdy (1992), in which the driving force is
proximate parabolic dependence of veloc- evaluated over a correlation length (l) on
ity on misfit was also found, consistent either side of the boundary. The steady
with Eq. (5-12). This suggests that elastic concentration profile is C (z). The force
strain energy also plays a role in propaga- then becomes:
tion of the grain boundary. (7-21)
l
In practice it is difficult to distinguish 1 ⎡1 ∂ F 2
⎤
∫ ( C ( z ) − C0 )2 + Y h 2 C ( z )2 ⎥ dz
the possible driving forces for the process. l 0 ⎢⎣ 2 ∂C 2 ⎦
At the highest temperatures, it is likely that
solute field stresses will be dominant. This where the second derivative is a measure of
appears to be the case for the study of Rhee the Helmholtz energy composition relation.
and Yoon (1989). In the early work of Hil- This approach has the virtue of including
lert and Purdy (1978), in which thin poly- both possible contributions to the driving
crystalline iron films were exposed to zinc force; it results in the conclusion that, as in
vapor, it was evident that the volume diffu- the massive transformation, a threshold ex-
sion penetration of the parent grains, in- ists, below which no motion is possible.
dexed by DV /v, was of atomic dimensions. An ambitious treatment of the problem
Hence, it was assumed that the driving has been put forward by Cahn et al. (1998).
force was entirely chemical. A comprehen- Their analysis is based on a phase-field
sive model should take into account all treatment of the grain boundary. They look
possible sources of the driving force. for travelling-wave solutions of the equa-
The moving grain boundary is subject to tions of motion, and treat their existence
a set of forces, which may in general in- as a requirement for forces capable of
clude chemical, elastic, frictional and cap- coupling with the boundary motion. They
illary forces. As in the massive transforma- find that coupling with the elastic field is
tion, there exists in principle a degree of indeed possible. However, their treatment
freedom corresponding to the concentra- gives no indication that a purely chemical
tion in the parent phase immediately adja- force is effective in moving the boundary.
cent to the boundary. In the conceptual
limit where no volume-diffusion penetra-
7.5.3 Discontinuous Precipitation
tion exists in advance of the boundary, the
concentration profile will be a step, and the Discontinuous precipitation, like chemi-
full chemical force pch will act across the cally-induced grain boundary migration,
interface. In a second limiting case, a con- involves the lateral diffusion of solute
centration gradient exists in front of the within a sweeping grain boundary. The dif-
boundary and an elastically-derived resid- ference lies in the nature and spacing of the
ual force is determined by the coherency solute sources/sinks. In the case of discon-
field. Intermediate cases would then corre- tinuous precipitation, these are members of
spond to a higher or lower degree of relax- a regular array of precipitates, whose spac-
ation of the concentration at the leading ing is a free variable, capable of internal
edge of the boundary towards a constrained adjustment. The reaction is found in a wide
equilibrium between strained parent crystal variety of precipitation systems, often at
7.5 Examples 505
low homologous temperatures where grain

boundary diffusion is expected to be a
dominant mechanism of material transport.
Like CIGM, discontinuous precipitation
can in principle occur in the absence of
volume diffusion. The reaction is capable
of destabilizing microstructure at relatively
low temperatures, and it is therefore of
practical interest.
Unlike eutectoid reactions, which yield
morphologically similar products, discon-
tinuous precipitation reactions are not as-
sociated with a particular feature of the
phase diagram, although supersaturation is
an obvious prerequisite. The question of Figure 7-20. Scanning electron micrograph of a dis-
which systems will give rise to discontinu- continuous precipitation colony formed in Al – 22%
ous precipitation, and which will not, will Zn at 478 K. The original grain boundary positions is
perhaps be answered by reference to and indicated by arrows. After Solorzano et al. (1984).
detailed understanding of the initiation and
growth processes.
ature. Gust (1984) found a correlation with
the solvus temperature rather than the melting
7.5.3.1 Initiation of Discontinuous
temperature. Duly and Bréchet (1994) de-
Precipitation
termined that the proportion of double-seam
A nucleation event, in the classical morphologies decreased from more than
sense, is not required. As in CIGM, a pre- 1/2 to 0 as the temperature was increased.
existing grain boundary is caused to move, In free-boundary initiation, the initial
and eventually to become the steady reac- grain boundary displacement is thought to
tion front, as suggested by Fig. 7-20. The be caused by ordinary recrystallization or
mechanisms proposed for the initial stages grain growth forces (Fournelle and Clark,
of grain boundary displacement fall into 1972), or by forces derived from solute
two broad classes: free-boundary mecha- segregation (Meyrick, 1976). The subse-
nisms and precipitate-assisted mecha- quent stages must then include the nuclea-
nisms. Baumann et al. (1981) have indi- tion of precipitates at the moving boundary
cated that free-boundary initiation is domi- and the evolution of a steady state in which
nant at higher homologous temperatures, the precipitates acquire a uniform spacing,
leading eventually to “single-seam” mor- as depicted in Fig. 7-21. Cu – Co alloys
phologies. At lower temperatures, it is ex- seem to require this type of initiation, and
pected that boundary motion is initiated by readily undergo discontinuous precipita-
precipitate – boundary interactions, and that tion only when cold-worked or when
these lead to a preponderance of double- treated to give a small parent grain size
seam morphologies. The correlation be- (Perovic and Purdy, 1981).
tween temperature and morphology ap- Precipitate – boundary interactions were
pears to be quite general, with the break oc- first studied in detail by Tu and Turnbull
curring at half the absolute melting temper- (1969), who showed that boundary torques,
Figure 7-21. Free-boundary initiation

and early development of discontinuous
precipitation. Reprinted with permission
from Fournelle and Clark (1972), p. 2762.
developed when a regular array of precipi-

tates, formed on the boundaries in Pb – Sn
bicrystals, were responsible for the initial
displacement of the boundary (Fig. 7-22).
Eventually, boundary breakaway was en-
visaged, with the subsequent embedding of
the precipitates in the advancing grain. Aa-
ronson and Aaron (1972) extended these
ideas to include a range of possible geome-
tries, each of which resulted in an initial
boundary displacement due to capillary
forces associated with the formation of
equilibrium precipitate nuclei at the grain
boundaries.
A further class of precipitate – boundary
interactions, based on CIGM, was pro-
posed by Purdy and Lange (1984). In this
case, initial displacement can occur against
capillary forces, in response to chemical
forces. If the initial precipitates are closely
spaced, the initial displacement may be
preceded by a period of precipitate coars-
ening. The process is suggested by Fig.
7-19. Several reports of grain boundaries
bowing between fixed precipitates can be
Figure 7-22. Formation of grain boundary precipi-
found in the literature (Hillert and Lagne- tates at an initially static boundary (a), and boundary
borg, 1971; Solorzano et al., 1984). Mi- displacements in response to capillary forces (b). Re-
chael and Williams (1986) found that so- printed with permission from Tu (1972), p. 2773.
7.5 Examples 507
lute-depleted volumes were left in the the steady state has attracted the attention
wake of bulging grain boundaries in super- of successive generations of theorists. In
saturated Al – 4.7 wt.% Cu. Fonda et al. this section, we review only the more ad-
(1998) and Mangan and Shiflet (1997) vanced theories, while acknowledging
studied the initiation of discontinuous pre- their geneology.
cipitation in Cu – 3% Ti alloys, and demon- Before proceeding, the question of the
strated that, at low supercooling, the initial steady state should be pursued in more de-
boundary motion is due to the presence of tail. It appears that the steady state can be
Widmanstätten precipitates, which grow achieved fairly generally, and that it can be
into one grain, and cause boundary dis- approached from higher or lower supersat-
placement into the other. Duly and Bréchet uration, such that the system has little or no
(1994) examined the initiation of discon- memory, and the steady state is characteris-
tinuous precipitation over a wide range of tic of the isothermal reaction conditions
temperatures, initial grain sizes and com- only. Nevertheless, the constancy of reac-
positions in Mg – Al alloys. They conclude tion front velocity has recently been ques-
that the Tu – Turnbull and Fournelle – Clark tioned by a number of authors, including
mechanisms dominate at low and high tem- Kaur and Gust (1988), Mangan and Shiflet
peratures respectively, and that the Purdy – (1997), and Fonda et al. (1998). This point
Lange mechanism may be important at is discussed in Sec. 7.5.3.3.
intermediate temperatures. It is convenient to divide the theoretical
It is now clear that a range of mecha- treatments into two classes depending on
nisms exist, each mechanism capable of in- whether the details of the transformation
itiating boundary motion. At lower temper- front are predicted as part of the develop-
atures, precipitation on the static boundary ment. In the “global” approach of Cahn
is a common precursor of discontinuous (1959), which builds on the ideas implicit
precipitation; the initial precipitates are ca- in Turnbull’s (1951) and Zener’s (1946)
pable of acting either directly, to pull the treatments, the reaction front is treated as a
boundary from its initial location, or indi- planar high-diffusivity path whose mobil-
rectly through CIGM to accomplish a simi- ity is rate determining. Thus the important
lar result. We conclude that, for most situa- input quantities to the theory include the
tions where the supersaturation is signifi- overall Helmholtz energy change accom-
cant, there will be no difficulty in initiating panying the passage of the front and the
grain boundary motion. interface mobility. Cahn (1959) first em-
phasized the importance of storage of
Helmholtz energy in the product phase, and
7.5.3.2 Theories of Steady Cooperative
showed how to estimate it from a knowl-
Growth
edge of the interphase boundary energy
The development of a steady or near- s a, b , the grain boundary diffusivity and
steady growth front, characterized by a distribution coefficient, s, and the interla-
regular spacing between precipitate lamel- mellar spacing S and velocity v. The Helm-
lae (or rods), S, and a constant velocity, v, holtz energy balance is written
is not immediate upon initiation, but is pre-
DF ≤ = DFch – 2 s ab Vm /S (7-22)
ceded by extended transient growth re-
gions. However, steady or near-steady con- Here DFch is the total chemical Helm-
ditions appear eventually to prevail, and holtz energy difference between the parent
and product phases, evaluated some dis- Certain aspects of the theory are amenable
tance from the interface. In determining to experimental verification, for example,
this quantity, the stored Helmholtz energy the concentration profiles left in the a
in the product a phase must be calculated product lamellae have been measured by
from a knowledge of the composition pro- Porter and Edington (1977) and by Solor-
file in the a lamellae. This is determined zano and Purdy (1984) using high resolu-
in theory from a knowledge of the grain tion elemental analysis, to give the trans-
boundary transport properties and the ve- port properties of the reaction front,
locity and spacing through {s D b d }, through Eq. (7-23). These same
measurements allow the evaluation of the
C0 − C ( z ) cosh ( z a / Sa )
= (7-23) amount of the total available Helmholtz
C0 − C3 cosh ( a / 2 ) energy for the transformation retained as
segregation in the product; this latter quan-
where C0 is the initial concentration in the
tity is also available, albeit in averaged
parent phase, C (z) is the local concentra-
form, through X-ray measurements of the
tion in the a phase behind the transforma-
lattice parameter of the product phase.
tion front, C3 is the concentration in the a
A heuristic description of discontinuous
phase in contact with the b lamellae and
precipitation, after Petermann and Horn-
v S2 bogen (1968), utilizes a rate equation of the
a= (7-24) type of Eq. (7-25) coupled with an approx-
s Db d
imate evaluation of the relaxation time for
Eq. (7-23) is a solution of the moving- grain boundary diffusion to the b lamellae,
boundary diffusion equation, for constant to give a velocity expression which is di-
v, S, and boundary concentration C3 ; the mensionally similar to that of Cahn.
solution reduces in approximation to a sine A second class of theoretical description
function when the centerline composition has been developed by Hillert (1969, 1972,
C (z = 0) is closer to C3 than to C0 . 1982). This approach has been termed “de-
e
Cahn (1959) next chose C3 = C ab , the tailed”, in the sense that a much deeper
equilibrium composition of a in contact knowledge of the transformation front is
with b at a planar interface. He was then implicit, even required, for its application.
able to evaluate the stored chemical Helm- The basis for this treatment is the applica-
holtz energy in the product phases from Eq. tion of a local force balance at every point
(7-23). To complete the description, he as- along the interface. Thus the possibility is
sumed a relationship between the total raised of determining the interface shape
Helmholtz energy change DF ≤ and the ve- for steady growth, as a consequence of the
locity of the form interplay of capillary, chemical, elastic,
and frictional forces, as defined in Sec. 7.4.
v = M≤ DF ≤ (7-25)
In simplified form, for application to the
Here, M≤ is a “global” mobility that dif- case where the composition profile (step),
fers from the intrinsic grain-boundary mo- DC, measured normal to the interface is not
bility, e.g., of Eq. (7-16). relaxed by volume diffusion, and where
The theory described above has been elastic and solute drag effects are negli-
modified by others, for example by Aaron- gible, the interfacial force balance becomes
son and Liu (1968), to take partial account
of capillary forces acting on the b lamellae. v/M = pch – ps (7-26)
7.5 Examples 509
which holds at every point. The solute pro- solute drag forces and severely nonplanar
file in the boundary will be given by Eq. shapes, thus allowing for consideration of
(7-23), and the concentration step (DC) at morphological stability of the transforma-
the interface will take a minimum value at tion front.
the centres of the lamellae, varying contin- Neither of the approaches described
uously to a maximum value at the a/a0 above is sufficient to predict the transfor-
interface immediately adjacent to the b la- mation state that will occur at given super-
mellae. Hillert has chosen the limiting saturation in a particular system. Each is
value of the solute concentration at this capable of generating a unique relationship
point, C3 in Eq. (7-23), as the capillary- between v and S, however.
modified solubility in a in equilibrium In considering the spacing problem for
with b under curvature 2/Sb . lamellar eutectoids, Zener (1964) set a
For a steady transformation, the right- thermodynamic limit, a minimum spacing
hand side of Eq. (7-26) is constant along for which all of the available Helmholtz
the interface, and the curvature must vary energy for the transformation is stored as
to match the chemical force, indexed by the interfacial Helmholtz energy in the prod-
composition step DC. uct. This minimum also exists for discon-
A further local force balance is required tinuous products:
at the a0 /a/b junction. This is depicted
by a vectorial balance of surface energies 2 s ab Vm
Smin = (7-27)
in Fig. 7-23. Thus the angles of intersec- DFch
tion and local curvatures are set at the junc-
tion, and the variation of the grain bound- At this (virtual) state, the system is in
ary curvature is required to match the vari- equilibrium and no growth is possible.
ation of chemical force along the boundary. Larger spacings corresponds to finite
The theory therefore allows the self-con- rates of growth. It has been suggested that
sistent calculation of the interface shape the steady spacing is one which maximizes
for given values of the spacing S, the veloc- the growth rate (Hillert, 1972) or the inte-
ity v, and the interfacial transport property gral rate of dissipation of Helmholtz
s D b d. energy (Cahn, 1959). Solorzano and Purdy
Sundquist (1973) extended Hillert’s (1984) put this latter hypothesis to the test,
treatment to include the possibility of and found a reasonable level of confirma-
Figure 7-23. Schematic (plan)

view of a steady discontinuous
growth front, defining the
spacings, S, Sa , Sb , relevant to
the problem, and indicating the
vectorial balance of interfacial
energies, sa/b , sa0/a , sa0/b , at
the three-grain junction.
tion for two well-characterized systems, out modification; however, the theories
Al – 22 at.% Zn and Mg – 9 at.% Al. need to take into account the possibility of
Building upon the local approach of Hil- lateral migration of the growth interface for
lert and co-workers, Klinger et al. (1996) lower supersaturation. A recent treatment
have proposed a solution which leads to a by Klinger et al. (1997 a) takes this into ac-
selection of the growth velocity without the count by describing the motion of the inter-
need for an optimization principle. This face as intermittent.
treatment relies on the cooperative nature
of the growth of the a and b lamellae, and
7.5.3.3 Experimental Observations
on the different characters of the interfaces
between the a and b phases and the parent The majority of experimental studies
a phase: one is a grain boundary and the have focused on microstructures developed
other is an interphase boundary. Assuming through steady isothermal growth. They
a definite composition for the b phase leads are therefore capable of description in
to a Mullins (1957) equation for transport terms of the theoretical treatments of the
along the a0/b interface determined by gra- previous section. The types of information
dients of curvature. Transport along the accessible to experiment are:
a0/a is considered to be driven by concen-
a) quantities that may be derived from
tration gradients, according to Cahn’s
conventional metallographic methods: av-
(1959) formulation. The condition at the
erage velocity v and spacing S;
triple junction is taken as a local equilib-
b) chemical information obtained either
rium, corrected for Gibbs – Thompson ef-
from averaging processes such as X-ray
fects. The requirement that the a and b
diffraction measurements of lattice param-
growth velocities match in order to have a
eters of product lamellae or from high res-
cooperative growth process is sufficient to
olution microanalyses of the product;
select the spacing and velocity of the front.
c) structural information relating to the
This approach has the advantage of avoid-
growth interface, as obtained for example
ing the introduction of optimization princi-
by transmission electron microscopy.
ples, and it has been shown to accurately
describe some experimental results for the In addition, for the test of theory we re-
Al – Zn system. It also predicts that steady- quire solution thermodynamic data, inter-
state solutions with invariant front shapes facial energies and interfacial diffusion co-
are possible within a limited range of driv- efficients. Many alloy systems have been
ing forces, suggesting the possibility of investigated. In this section we will focus
morphological instabilities outside this on the systems Al – Zn and Mg – Al, for
range. which much of the necessary data are avail-
All of these theories of steady growth as- able (e.g., Rundman and Hilliard, 1967;
sume that the motion of the interface is Cheetham and Sale, 1974; Hassner, 1974).
continuous. This assumption is probably Yang et al. (1988) collected velocity and
acceptable when the growth front is well- spacing data for Al – Zn alloys, and for a
rounded, but there is substantial evidence range of reaction temperatures and spac-
for faceted interfaces at low supersatura- ings, as summarized in Figs. 7-24 and 7-25.
tions, as discussed in detail in the next sec- Figure 7-26 gives the average composition
tion. In the high-supersaturation regime, of the a lamellae superimposed on the Al-
the continuum models seem to apply with- rich portion of the phase diagram. These
7.5 Examples 511
composite figures include results from the

work of Ju and Fournelle (1985), Suresh
and Gupta (1986), Cheetham and Sale
(1974), Razik and Maksoud (1979), Wata-
nabe and Koda (1965), Gust et al. (1984)
and Yang et al. (1988). Perhaps the most
striking feature of these data is the extent
of residual supersaturation left in the a
phase behind the transformation front; a
significant amount of the total Helmholtz
energy for the transformation is stored in
the product phase.
This stored Helmholtz energy, first em-
phasized by Cahn (1959), must be esti-
mated in order to evaluate the driving force
for the transformation. This estimation is
Figure 7-25. Interlamellar spacing measurements

for discontinuous precipitation (DP) products, from
the same sources as for Fig. 7-24. Reprinted with per-
mission from Yang et al. (1988).
readily performed using data of the type

shown in Fig. 7-26. This stored Helmholtz
energy provides a portion of the driving
force for subsequent discontinuous coars-
ening processes, which generally increase
the interlamellar spacing and deplete the
product a to (near) the solubility limit
(Yang et al., 1988).
High resolution microanalysis is capable
of providing further insights into the stored
Helmholtz energy term (Ziȩba and Gust,
1998). The elemental trace of Fig. 7-27
was obtained from a STEM (scanning
transmission electron microscope) X-ray
microanalysis of a thin foil of Al – 22 at.%
Figure 7-24. Grain-boundary velocities for a range Zn transformed at 428 K. Measurements of
of reaction temperatures and for several Al – Zn al- this type give both local and integral values
loys (index DP denotes discontinuous precipitation). of the stored Helmholtz energy. In addition
The data are from studies by Yang et al. (1988), Ju to the information concerning the solute
and Fournelle (1985) [1], Suresh and Gupta (1986) concentration distribution in the a phase,
[2], Cheetham and Sale (1974) [9], Razik and Mak-
soud (1979) [10], Gust et al. (1984) [18], and Wata-
these profiles can be analyzed to yield
nabe and Koda (1965) [19]. Reprinted with permis- rather directly the product {s D b d } via Eq.
sion from Yang et al. (1988). (7-23), (Porter and Edington, 1977; Solor-
Cahn’s equation with constant values of

–
* and ÷a.
Cab
b) However, within a given nodule, the
value of (a/Sa2 ) was generally not constant,
varying by more than an order of magni-
tude. This quantity is proportional to the
instantaneous interfacial velocity, and its
variation is taken to mean that, on an
atomic length scale, the interface velocity
is irregular. On a larger length scale, typi-
cally 0.1 µm, an average velocity can be
defined that is consistent with the steady
diffusion analysis of Cahn.
Figure 7-26. Average zinc concentrations in the a A major implication of these findings is
phase for discontinuous precipitation (DP) and for that the transformation interface moves by
discontinuous coarsening (DC) reactions in various a lateral growth mechanism, resulting in an
Al – Zn alloys. Reprinted with permission from Yang
average velocity much less than the instan-
et al. (1988).
taneous velocity accompanying the pas-
sage of a growth step. It is here that the
recent studies of Shiflet and his co-workers
zano et al., 1984; Duly et al., 1994 a, b).
(Fonda and Shiflet, 1990; Fonda et al.,
However, in their studies of regular growth
1998) on Cu – Ti alloys are relevant. Their
in Mg – Al alloys, Duly et al. determined
results indicate a strong and consistent ten-
that:
dency for the transformation front to facet,
a) In certain cases, the concentration particularly at low undercoolings, and this
profiles between parallel lamellae (similar implies a lateral displacement process,
to Fig. 7-27) were well described by consistent with the findings of Duly et al.
Figure 7-27. STEM microanalysis

of an a lamella in Al – 22 at.% Zn,
formed at 428 K. After Solorzano
and Purdy (1984).
7.5 Examples 513
chemical (evaluated from microanalyses),

capillary and friction forces, realized at
every point on the transformation front. It
therefore appears that at some stage an ef-
fective transition occurs where the accu-
mulation of growth ledges is sufficient to
mimic the normal migration of the front,
and the classical theories of Cahn and Hil-
lert then hold to a good approximation.
There is a further set of experimental
data on discontinuous precipitation that re-
lates to the effect of an applied stress on the
transforming system. Sulonen (1964 a, b)
showed that in some alloy systems, fronts
with normals parallel to the tensile axis
moved more slowly, and those with nor-
mals perpendicular to the stress axis more
quickly than those in unstressed alloys. In
other alloy systems, the opposite behavior
was found. The effect for Cu – Cd alloys is
shown in Fig. 7-29. Sulonen proposed an
Figure 7-28. Observed and calculated interface

shapes for an a lamella formed at 400 K in Al –
22 at.% Zn. The numbers on the curves represent
different amounts of Helmholtz energy lost owing to
continuous precipitation in the parent phase.
(1994 a, b). At lower reaction temperatures,

the growth interfaces in Cu – Ti become
smoothly curved. This is attributed to the
rapid accumulation of growth ledges with
increasing undercooling.
The smoothly curved interfaces and
steady migration rates seen in high resolu-
tion in situ studies of discontinuous precip-
itation in Al – Zn (see Fig. 7-28, Tashiro
and Purdy, 1989) suggest that, at lower
temperatures, the interface can be consid- Figure 7-29. Effect of an applied tensile stress on
the rates of growth of the discontinuous product in
ered to move continuously. Indeed, the Cu – Cd alloys (1 kp/mm2 ≈ 9.81 ¥ 106 N/m2 ). Re-
interface shapes observed could be ration- printed with permission from Sulonen (1964 b),
alized in terms of a local balance between copyright Pergamon Press.
explanation based on the elastic interaction 7-29), and give the correct sign of the ef-
between the applied and coherency (solute) fect for the other five systems investigated,
stress fields. Hillert (1972) indicated how a for which no quantitative data were ob-
quantitative treatment of such an effect tained. We conclude that it is the plastic (or
could be formulated. Sulonen’s result and creep) response of the specimen which is
Hillert’s analysis have been widely quoted more plausibly coupled to the migration
in support of the existence of a solute (co- rates of differently oriented transformation
herency) field in front of the discontinuous interfaces.
transformation front, even at temperatures
where the calculated diffusion distance is
7.5.4 Interface Migration in Multilayers
negligible.
Dryden and Purdy (1990) reconsidered Multilayers have been utilized for criti-
the problem, and included the possible ef- cal experiments since the seminal work of
fect of a volume misfit in the transformed Hilliard and his co-workers (1954). They
region. It was found, in agreement with provide an ideal tool for the investiga-
Hillert, that the sign of the effect could be tion of the thermodynamics and kinetics of
predicted on the basis of an elastic solute heterogeneous systems far from equilib-
field interaction; however, the predicted ef- rium. Much of the work to date has been
fects on interface velocities are too small concerned with the approach to equilib-
by four orders of magnitude. The more rium via bulk diffusion. The recently de-
likely cause of the coupling is found in the veloped techniques for precisely controlled
plastic response of the dead-loaded speci- growth (e.g., molecular beam epitaxy)
men. The reduction in Helmholtz energy of prompt the exploration of the approach to
the loading device results in a virtual force equilibrium via interfacial diffusion and
on the transformation interface, as dis- migration. The theoretical investigations
cussed in Sec. 7.4.3.2. For the case of inter- summarized in this section deal with pos-
faces with normals parallel to the x, y, and sible effects, many of which are still to be
z (tensile) axes, these virtual forces are re- observed.
lated by Several classes of problems can be de-
(7-28)
⎛ 1 + 3W ⎞ Ze veloped, beginning with an A – B multi-
px = py = − p = (1 + 3W ) layer grown on a bicrystalline substrate
⎝ 2 ⎠ z 2
such that a grain boundary penetrates the
where px , py , and pz are forces per unit whole structure and provides a possible
area, Z is the applied tensile stress, e is the fast diffusion path. The simplest case oc-
stress-free strain in the transformed vol- curs when A and B are fully miscible.
ume, and W is an alignment parameter Homogenization can then occur by grain
which accounts for the possibility that the boundary motion, either via a cooperative
lamellae in transformed regions bounded mechanism, or by a “fingering solution” at
by x and y interfaces are aligned with re- the former A/B interfaces as illustrated in
spect to the z axis. Figure 7-30 (Klinger et al., 1997 a). In each
It is found that the virtual forces so de- case, the shape of the moving boundary, as
rived, in conjunction with independently well as its velocity and the concentration
derived grain-boundary mobilities, are ca- profile left in its wake, can be computed in
pable of explaining the sign and magnitude terms of driving forces, interface energies
of Sulonen’s data for Cu – Cd alloys (Fig. and diffusion coefficients.
7.5 Examples 515
Figure 7-30. Illustrating three different possibilities for the discontinuous homogenization of a monophase
multilayer containing a mobile grain boundary, assumed initially to bisect the multilayer. Situations correspond-
ing to (a) the steady state motion of the boundary; (b) an initially sinusoidal instability of the grain boundary,
and (c) a “fingering” instability of the moving boundary, after Klinger et al. (1997 b).
The next situation to be considered is search for such interface-mediated structu-

when A and B are reactive, and form a stoi- ral evolution in controlled multilayered
chiometric compound w (or, conversely, structures can play a key role in the deeper
when such a compound decomposes to understanding of the more frequently en-
form A and B). Again, for this case, the ve- countered phenomena described in the rest
locity of the cooperative moving front for of this chapter.
reaction (or dissolution) can be computed
as a function of the energies and mobilities
involved (Klinger et al., 1997 c). A solution
involving a reaction product layer at the
interface rather than a cooperative growth
front can also be considered (Fig. 7-31),
and this may help to explain certain pecu-
liarities found in kinetics in reactive multi-
layers, in terms of diffusion barrier effects
(Klinger et al., 1998; Emeric, 1998).
These theoretical investigations, many
still awaiting experimental confirmation,
suggest a wide variety of possible kinetic
paths toward thermodynamic equilibrium
(involving pattern and velocity selection
as well as morphological instabilities) in
well-controlled systems. In these systems, Figure 7-31. Schematic representation of the pos-
sible reactions of A and B to form a stoichiometric
a number of classical hypotheses such as product phase w. (a) A cooperative reaction at a sin-
that of local equilibrium could be checked gle front; (b) the growth of a product layer at the
quantitatively. It is suggested that the interface (after Klinger et al., 1997 c).
7.6 Conclusions transformations will need to be reformu-

lated, and this is only beginning to take
In this chapter we have been concerned place.
with those phase transformations that de- The thermodynamic structure remains a
pend on diffusion within a transformation valuable synthetic framework within which
front. The massive transformation, chemi- to place advances in the understanding of
cally-induced grain boundary migration, this fascinating field.
discontinuous precipitation and transfor-
mations in multilayers represent different
and distinct facets of the general class of 7.7 References
interface-diffusion controlled transforma-
tions. Aaronson, H. I., Aaron, H. B. (1972), Metall. Trans.
In each case, the description of the trans- A 3, 2743.
Aaronson, H. I., Liu, Y. L. (1968), Scripta Metall. 2,
formation in terms of a balance of forces on 1.
the moving interface has proved a unifying Ackert, R. J., Parr, J. G. (1971), J. Iron Steel Inst.
concept. For the massive transformation, it 209, 912.
Ashby, M. F., Spaepen, F., Williams, S. (1979), Acta
is believed that the major forces at play are Metall. 26, 235.
the chemical driving force and the oppos- Aust, K. J., Rutter, J. W. (1959), Trans. AIME 215,
ing frictional forces. In the description of 119.
Aziz, M. J., Kaplan, T. (1988), Acta Metall. 36, 2335.
chemically-induced grain boundary migra- Bailey, J., Hirsch, P. B. (1962), Proc. Roy. Soc. A287,
tion, especially at higher temperatures, an 11.
elastic (coherency) term due to a gradient Balluffi, R. W., Kwok, T., Bristowe, P. S., Brokman,
A., Ho, P. S., Yip, S. (1981), Scripta Metall. 15,
of misfitting solute atoms in the parent 951.
grain must be added to the other two types Barrett, C. S., Massalski, T. B. (1966), in: Structure
of force. For discontinuous precipitation, of Metals, 3rd ed. New York: McGraw-Hill.
Baumann, S. F., Michael, J., Williams, D. B. (1981),
capillary forces must be introduced to op- Acta Metall. 29, 1343.
pose the chemical driving force. We have Bibby, M. J., Parr, J. G. (1964), J. Iron Steel Inst.
also considered transformations in multi- 202, 100.
Bollmann, W. (1970), Crystal Defects and Crystal-
layer heterostructures, where capillary line Interfaces, Berlin: Springer-Verlag.
terms must be accounted, and where inter- Bréchet, Y. J. M., Purdy, G. R. (1989), Acta Metall.
facial diffusion is often driven by interfa- 37, 2253.
Bréchet, Y., Purdy, G. R. (1992), Scripta Metall. 27,
cial curvature. 1753.
The description offered has been pre- Cahn, J. W. (1959), Acta Metall. 7, 8.
dominantly thermodynamic, rather than Cahn, J. W. (1962), Acta Metall. 10, 789,
Cahn, J. W. (1982), J. Physique 43, Suppl. 12, C 6.
deeply mechanistic. Much remains to be Cahn, J. W., Allen, S. (1977), J. Phys. C7, 12; 57.
learned about the mechanisms of accom- Cahn, J. W., Fife, P., Penrose, O. (1998), Acta Mater.
modation of diffusing atoms within the 45, 4397.
Cahn, J. W., Hilliard, J. E. (1958), J. Chem. Phys. 28,
transformation interface, about the cou- 258.
pling of the driving forces with interface Carter, W. C., Taylor, J. E., Cahn, J. W. (1997), J. O.
response, and about the interrelationships M., December, 30.
Cheetham, D., Sale, F. R. (1974), Acta Metall. 33, 71.
between interfacial structure and kinetics. Chen, L-Q., Fan, D. N., Tikare, V. (1998), in: Grain
It has become increasingly clear that many Growth in Polycrystalline Materials, Weiland, H.,
transformation interfaces are faceted, and Adams, B., Rollett, T. (Eds.), Warrendale, PA,
USA: TMS, p. 137.
remain faceted during migration. In such Chen, J. K., Purdy, G. R., Weatherly, G. C., Kroupa,
cases, the classical models for diffusional A. (1998), Metall. Mater. Trans. A 29, 2049.
7.7 References 517
Cottrell, A. H. (1953), Dislocations and Plastic Flow Humphreys, F. J., Hatherly, M. (1995), Recrystalliza-
in Crystals, Oxford: Oxford University Press. tion and Related Annealing Phenomena. Oxford:
Dahmen, U. (1981), Scripta Metall. 15, 77. Pergamon, Elsevier Science Ltd., p. 212.
Delaey, L. (2001), in: Phase Transformations in Ma- Ju, C. P., Fournelle, R. A. (1985), Acta Metall. 33,
terials: Kostorz, G. (Ed.). Weinheim: Wiley-VCH, 71.
Chap. 9. Karlyn, D. A., Cahn, J. W., Cohen, M. (1969), Trans.
Delage, S. (1998), Ph.D. thesis, University of Paris V. TMS-AIME. 245, 197.
den Broeder, F. J. A. (1972), Acta Metall. 20, 319. Kaur, I., Gust, W. (1988), Fundamentals of Grain
Dryden, J. R., Purdy, G. R. (1990), Acta Metall. 38, and Interphase Boundary Diffusion. Stuttgart:
1255. Ziegler Press.
Duly, D., Bréchet, Y. (1994), Acta Metall. Mater. 42, King, A. H. (1987), Internat. Mater. Rev. 32, 173.
3035. Klinger, L., Bréchet, Y., Purdy, G. R. (1996), Acta
Duly, D., Cheynet, M. C., Bréchet, Y. (1994 a) Acta Mater. 45, 5005.
Metall. Mater. 42, 3843. Klinger, L., Bréchet, Y., Duly, D. (1997 a), Scripta
Duly, D., Cheynet, M. C., Bréchet, Y. (1994 b) Acta Mater. 37, 1237.
Metall. Mater. 42, 3855. Klinger, L., Bréchet, Y., Purdy, G. R. (1997 b), Acta
Emeric, E. (1998), Ph.D. thesis, Marseille. Mater. 45, 325.
Faridi, B. A., Crocker, A. G. (1980), Phil. Mag. A 41, Klinger, L., Bréchet, Y., Purdy, G. R. (1997 c), Acta
137. Mater. 45, 4667.
Ferrence, T. G., Balluffi, R. W. (1988), Scripta Klinger, L., Bréchet, Y., Purdy, G. R. (1998), Acta
Metall. 22, 1929. Mater. 46, 2617.
Fonda, R. W., Mangan, M. A., Shiflet, G. J. (1998), Krakow, W., Wetzel, J. T., Smith, D. A. (1986), Phil.
Metall. Mater. Trans. A 29, 2101. Mag. A 53, 739.
Fournelle, R. A., Clark, J. B. (1972), Metall. Trans. LeClaire, A. D. (1986), in: Solute-Defect Interac-
A 3, 2757. tions: Saimoto, S., Purdy, G. R., Kidson, G. V.
Gibbs, J. W. (1906), Scientific Papers, Vol. 1, Ther- (Eds.). Oxford: Pergamon, p. 251.
modynamic. New York: Dover Publications. Levine, H. S., MacCallum, C. J. (1960), J. Appl.
Gust, W. (1984), in: Decomposition of Alloys, The Phys. 31, 595.
Early Stages: Haasen, P., Gerold, V., Wagner, R., Lücke, K., Detert, K. (1957), Acta Metall. 5, 517.
Ashby, M. F. (Eds.). Oxford: Pergamon, p. 221. Mangan, M. A., Shiflet, G. J. (1997), Scripta Mater.
Gust, W., Hintz, B. B., Lucic, R., Predel, B. (1984), 37, 517.
in: Phase Transitions in Solids: Tsakalakos, T. Martin, G., Blackburn, D. A., Adda, Y. (1967), Phys.
(Ed.). New York: Elsevier, p. 513. Stat. Sol. 23, 223.
Gust, W., Mayer, S., Bogel, A., Predel, B. (1985), Martin, G., Athenes, M., Desgranges, C., Nasbur, M.,
J. Physique C 4 46, 537. Soissons, F. (1998), in: Diffusion Mechanisms in
Guttman, M. (1995), J. Phys. C7, 85. Crystalline Solids, Mishin, Y., Cowern, N. E. B.,
Hassner, A. (1974), Kristall. Tech. 9, 1371. Catlow, C. R. N., Farkas, D., Vogl, G. (Eds.). MRS
Hazzledine, P. M., Oldershaw, R. D. (1990), Phil. Symposium Proceedings, vol. 527, Warrendale,
Mag. A 61, 579. PA, USA: MRS, p. 35.
Hillert, M. (1969), in: The Mechanism of Phase Massalski, T. B. (1958), Acta Metall. 6, 243.
Transformations in Crystalline Solids. London: In- Massalski, T. B. (1984), Metall. Trans. 15 A, 421.
stitute of Metals, Monograph No. 33, p. 231. Menon, E. S. K., Plichta, M. R., Aaronson, H. I.
Hillert, M. (1972), Metall. Trans. 3 A, 2729. (1988), Acta Metall. 36, 321.
Hillert, M. (1975 a), in: Lectures on the Theory Meyrick, G. (1976), Scripta Metall. 10, 649.
of Phase Transformations: Aaronson, H. I. (Ed.), Michael, J. R., Williams, D. B. (1986), in: Interface
New York: AIME, p. 1. Migration and Control of Microstructure: Pande,
Hillert, M. (1975 b), Metall. Trans. 6 A, 5. C. S., Smith, D. A., King, A. H., Walter, J. (Eds.).
Hillert, M. (1982), Acta Metall. 30, 1689. Metals Park, OH, USA: ASM, p. 73.
Hillert, M., Lagneborg, R. (1971), J. Mater. Sci. 6, Morral, J. E., Purdy, G. R. (1995), J. Alloys and Com-
208. pounds 220, 132.
Hillert, M., Purdy, G. R. (1978), Acta Metall. 26, Mullins, W. W. (1957), J. Appl. Phys. 28, 333.
333. Pawlowski, A., Ziȩba, P. (1991), Phase Transforma-
Hillert, M., Sundman, B. (1976), Acta Metall. 24, tions Controlled by Diffusion at Moving Boundar-
731. ies, Krakow, Poland: Polish Academy of Sciences.
Hilliard, J. E., Averbach, B. L., Cohen, M. (1954), Perepezko, J. H. (1984), Metall. Trans. 15 A, 437.
Acta Metall. 2, 621. Perovic, A., Purdy, G. R. (1981), Acta Metall. 29,
Hoffman, D. W., Cahn, J. W. (1972), Surface Science 53.
31, 368. Petermann, J., Hornbogen, E. (1968), Z. Metallk. 59,
Hsieh, T. E., Balluffi, R. W. (1989), Acta Metall. 37, 814.
2133. Peterson, N. L. (1983), Internat. Met. Rev. 28, 65.
Plichta, M. R., Clark, W. A. T., Aaronson, H. I. Sutton, A., Balluffi, R. W. (1995), Interfaces in Crys-
(1984), Metall. Trans. A 15, 42. talline Metals, Oxford: Oxford University Press.
Porter, D. A., Edington, J. W. (1977), Proc. Roy. Soc. Suzuoka, T. (1964), J. Phys. Soc. (Japan) 19, 839.
A 358, 335. Tashiro, K., Purdy, G. R. (1983), Scripta Metall. 17,
Purdy, G. R. (1971), Met. Sci. J. 5, 81. 455.
Purdy, G. R., Lange, N. (1984), in: Decomposition of Tashiro, K., Purdy, G. R. (1987), Scripta Metall. 21,
Alloys, The Early Stages: Haasen, P., Gerold, V., 361.
Wagner, R., Ashby, M. F. (Eds.). Oxford: Per- Tashiro, K., Purdy, G. R. (1989), Metall. Trans. A 20,
gamon, p. 214. 1593.
Razik, N. A., Maksoud, S. A. (1979), Appl. Phys. 19, Treglia, G., Legrand, B. (1998), in: Tight Binding
331. Approach to Computational Materials Science:
Rhee, W.-H., Yoon, D. K. (1989), Acta Metall. 37, Turchi, P. E., Gonis, A., Columbo, L. (Eds.). MRS
221. Symposium Proceedings, vol. 491, Warrendale,
Rigsbee, J. M., Aaronson, H. I. (1979), Acta Metall. PA, USA: MRS, p. 275.
27, 351, 365. Tu, K. N. (1972), Metall. Trans. A 3, 2769.
Robinson, J. T., Peterson, N. L. (1973), Acta Metall. Tu, K. N., Turnbull, D. (1969), Acta Metall. 17,
21, 1181. 1263.
Rottman, C., Voorhees, D. W., Johnson, W. C. Turnbull, D. (1951), Trans. AIME 191, 661.
(1988), Scripta Metall. 22, 293. Verdier, M., Groma, I., Flandin, L., Lendvaii, J.,
Rundman, K. B., Hilliard, J. E. (1967), Acta Metall. Bréchet, Y., Guyot, P. (1997), Scripta Mater. 37,
15, 1025. 449.
Singh, J., Purdy, G. R., Weatherly, G. C. (1985), Vitek, V. (1984), in: Dislocations 1984: Veyssiere, P.,
Metall. Trans. A 16, 1363. Kubin, L., Castaing, J. (Eds.). Paris: Editions du
Smith, C. S. (1948), TMS-AIME 175, 15. CNRS, p. 435.
Smith, D. A. (1986), Proc. Int. Conf. on Solute-De- Wagner, R., Kampmann, R., Vorhees, P. W. (2001),
fect Interactions: Saimoto, S., Purdy, G. R., Kid- in: Phase Transformations in Materials: Kostorz,
son, G. V. (Eds.). Oxford: Pergamon, p. 128. G. (Ed.). Weinheim: Wiley-VCH, Chap. 5.
Solorzano, I. G., Purdy, G. R. (1984), Metall. Trans. Wanatabe, R., Koda, S. (1965), Trans. Natl. Res. Inst.
A 15, 1055. Metals (Japan) 7, 87.
Solorzano, I. G., Purdy, G. R., Weatherly, G. C. Weygand, D. (1998), Ph.D. thesis, Grenoble, INPG.
(1984), Acta Metall. 32, 1709. Weygand, D., Bréchet, Y., Lepinoux, J. (1999), Acta
Stephens, D. E., Purdy, G. R. (1975), Acta Metall. 23, Mater. 47, 961.
1343. Wheeler, A., Boettinger, W. J., McFadden, G. B.
Straumal, B. B., Bokshtein, B. S., Klinger, L. M., (1992), Phys. Rev. A 45, 7425.
Shvindlerman, L. S. (1981), Scripta Metall. 15, Whipple, R. (1954), Phil. Mag. 45, 1225.
1197. Yang, C. F., Sarker, G., Fournelle, R. A. (1988), Acta
Sulonen, M. S. (1964 a), Acta Polytechnica Scandi- Metall. 36, 1511.
navica 28, 5. Zener, C. (1946), Trans. AIME 167, 550.
Sulonen, M. S. (1964 b), Acta Metall. 12, 749. Zhang, W-Z., Purdy, G. R. (1993 a), Phil. Mag. A 68,
Sundquist, B. F. (1973), Metall. Trans. A 4, 1919. 279.
Suresh, V., Gupta, S. (1986), Z. Metallk. 77, 529. Zhang, W-Z., Purdy, G. R. (1993 b), Phil. Mag. A 68,
Sutton, A. (1995), in: Computer Simulation in Mate- 291.
rials Science, NATO ASI vol. 308, Pontikis, V., Zhang, W-Z., Perovic, V., Perovic, A., Weatherly, G.
Kirchner, H., Kubin, L. (Eds.), Dordrecht (Nether- C., Purdy, G. R. (1998), Acta Mater. 46, 3443.
lands): Kluwer, p. 163. Ziȩba, P., Gust, W. (1998), Internat. Mater. Rev. 43,
Sutton, A. P., Balluffi, R. W. (1987), Acta Metall. 35, 70.
2177.
8 Atomic Ordering
Gerhard Inden
Max-Planck-Institut für Eisenforschung GmbH, Düsseldorf, Federal Republic of Germany

8.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 523
8.2 Definition of Atomic Configurations . . . . . . . . . . . . . . . . . . . 523
8.2.1 Configurational Variables . . . . . . . . . . . . . . . . . . . . . . . . . . 523
8.2.2 Point Variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 525
8.2.3 Point Correlation Functions and Point Probabilities . . . . . . . . . . . . 526
8.2.4 Pair Variables, Correlation Functions, and Probabilities . . . . . . . . . . 527
8.2.5 Generalized Cluster Variables, Correlation Functions, and Probabilities . . 529
8.2.6 Short-Range-Order (sro) Configurations –
Long-Range-Order (lro) Configurations . . . . . . . . . . . . . . . . . . 529
8.3 The Existence Domain and Configuration Polyhedron . . . . . . . . . 531
8.3.1 F.C.C. Structure, First Neighbor Interactions . . . . . . . . . . . . . . . . 532
8.3.2 F.C.C. Structure, First and Second Neighbor Interactions . . . . . . . . . 535
8.3.3 B.C.C. Structure, First and Second Neighbor Interactions . . . . . . . . . 537
8.4 Ground States . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 538
8.4.1 Pair Interactions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 538
8.4.1.1 Ground State Energies . . . . . . . . . . . . . . . . . . . . . . . . . . . 540
8.4.1.2 F.C.C. Structure, First Neighbor Interactions . . . . . . . . . . . . . . . . 540
8.4.1.3 F.C.C. Structure, First and Second Neighbor Interactions . . . . . . . . . 543
8.4.1.4 B.C.C. Structure, First and Second Neighbor Interactions . . . . . . . . . 544
8.4.1.5 Energy Minimum at Constant Composition . . . . . . . . . . . . . . . . . 544
8.4.1.6 Canonical Energy of lro States . . . . . . . . . . . . . . . . . . . . . . . 546
8.4.1.7 Relevant Literature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 547
8.4.2 Effective Cluster Interactions (ECIs) . . . . . . . . . . . . . . . . . . . . 548
8.5 Phase Equilibria at Finite Temperatures . . . . . . . . . . . . . . . . . 551
8.5.1 Cluster Variation Method . . . . . . . . . . . . . . . . . . . . . . . . . . 551
8.5.2 Calculation of Phase Diagrams with the CVM . . . . . . . . . . . . . . . 554
8.5.3 Phase Diagram Calculation with the Monte Carlo Method . . . . . . . . . 554
8.5.4 Examples of Prototype Diagrams . . . . . . . . . . . . . . . . . . . . . . 555
8.5.4.1 F.C.C. Structure, First Neighbor Interactions . . . . . . . . . . . . . . . . 555
8.5.4.2 F.C.C. Structure, First and Second Neighbor Interactions . . . . . . . . . 557
8.5.4.3 B.C.C. Structure, First and Second Neighbor Interactions . . . . . . . . . 558
8.5.4.4 Hexagonal Lattice, Anisotropic Nearest-Neighbor Interactions . . . . . . 558
8.5.5 The Cluster Site Approximation (CSA) . . . . . . . . . . . . . . . . . . . 560
8.6 Application to Real Systems . . . . . . . . . . . . . . . . . . . . . . . . 561
8.6.1 The Au – Ni System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 563

ISBN: 3-527-30256-5
520 8 Atomic Ordering
8.6.2 The Thermodynamic Factor of Ordered Phases . . . . . . . . . . . . . . . 565

8.6.2.1 The B.C.C. Fe –Al System . . . . . . . . . . . . . . . . . . . . . . . . . 565
8.6.2.2 The F.C.C. Ni –Al System . . . . . . . . . . . . . . . . . . . . . . . . . . 567
8.6.3 Ternary Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 569
8.6.3.1 B.C.C. Fe –Ti –Al . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 569
8.6.3.2 B.C.C. Ferromagnetic Fe – Co – Al . . . . . . . . . . . . . . . . . . . . . 570
8.6.4 H.C.P. Cd – Mg . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 573
8.6.5 Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . 574
8.7 Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 575
8.9 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 578

a, a1 , … basic vectors of the structure
aa CVM exponents
b, b1 , … basic vectors of the structure
c, c1 , … basic vectors of the structure
d distance of hyperplane to the origin
Ẽ part of internal energy that depends only on the configurational variables
F Helmholtz energy
i number identifying the type of atom
k number identifying a neighbor distance
K number of constituents
ma number of a -clusters per point
N number of lattice sites
n, m number identifying a lattice site
n normal vector
Ni number of atoms i
N (s r ) number of r-clusters with occupation s r
Nr number of equivalent r-site clusters
pn(i) site-occupation operator for atom i on site n
ij
pnm occupation operator for atom pair i, j on sites n, m
r number of lattice points within a cluster
S entropy
T temperature
U, Ũ internal energy, grand canonical internal energy
V volume
Vijnm pair energy of atoms i, j on sites n, m
W number of possible arrangements of clusters formed for given values of the
correlation functions
Wij(k) pair exchange energy of k-th neighbor atoms I and J
W (k) binary case: indices i j omitted
Wij12……l r r-site cluster exchange energy correction term
Vn cluster expansion coefficient
wniÆ j transition probabilities
x hyperplane position vector
xv vertex vector
xi mole fraction of atom of type i
z (1 … r) coordination number of r-site clusters
a number identifying a cluster

a nij– m Cowley –Warren sro parameters
mi chemical potential of atoms i
n dimension of configurational space
n tot , n p total number of configurational probabilities, number of configurational prob-

abilities related by partial summations
z1 , z 2 , z 3 lro parameters
r (s r ) N (s r )/Nr
ij
r nm probability of having atoms i, j on sites n, m
s vector with site operators as components
sn site operator or spin variable
tk values of the site operator
j arbitrary function
Fi , F energy variables, defining dual space of the configurational variables, their
vector hyperplane
W grand potential
CVM cluster variation method

CSA cluster site approximation
EC cluster expansion
ECI effective cluster interaction
lro long-range order
n–m neighbor distance between sites n, m (equivalent to number k)
MC Monte Carlo simulation
sro short-range order
1 kB-unit = 1 kB K = 13.8 ¥ 10 – 24 J = 8.6 ¥ 10 – 2 meV

8.2 Definition of Atomic Configurations 523
8.1 Introduction fects. First-principles calculations of total

energies, of lattice relaxations and of local
Ordered structures occur frequently in relaxations were still in their infancy ten
solid solutions and exhibit interesting years ago, but such theoretical calculations
physical and mechanical properties. This are now available. They will be discussed
explains the continuous interest in ordering in relation to the treatment of real alloys.
reactions by theoreticians and experimen- The variety of materials exhibiting or-
talists over recent decades. There are some dering is too great to be dealt with com-
excellent overview articles and books (e.g., pletely. The present chapter will therefore
Ducastelle, 1991; Turchi and Gonis, 1994; be limited to metallic substitutional alloys,
de Fontaine, 1979, 1994; Khachaturyan, despite the fact that the techniques dis-
1978) on the theoretical aspects of this sub- cussed have also been used extensively in
ject, while the experimental and materials the field of interstitial alloys, carbides, ni-
science aspects can be found in articles by trides, oxides, and semiconductor systems.
Kear et al. (1970), Warlimont (1974), Koch The chapter is organized as follows. In
et al. (1985), Stoloff et al. (1987), Liu et al. Sec. 8.2, a general formalism for describ-
(1989, 1992), and Whang et al. (1990). ing and characterizing atomic configura-
The ordered phases can be classified tions in multicomponent substitutional al-
under the family of intermetallic phases. loys is presented. The concept of correla-
As such they have attracted interest in the tion functions as independent variables for
development of materials for special appli- the definition of the configurations is intro-
cations (e.g., materials for use at high tem- duced and used in Sec. 8.3 to derive exis-
peratures). Thus there is an increasing de- tence domains; they limit the range of nu-
mand for quantitative descriptions of the merical values that can be scanned by the
variations in their properties. Furthermore, correlation functions for topologically ex-
materials with practical relevance rarely isting configurations. These existence do-
consist of binary systems. Treatments of mains are used in Sec. 8.4 to determine the
multicomponent systems are thus espe- ground states. In Sec. 8.5, equilibrium at fi-
cially needed. The most fundamental prob- nite temperatures is discussed in terms of
lems that need to be solved at the start of the cluster variation method (CVM) and
any optimization of materials involve the the Monte Carlo simulation (MC). Finally,
type of ordered structure, the variation of in Sec. 8.6, the application to real systems
the atomic distribution with temperature is discussed.
and composition, and the phase equilibria.
This chapter is an updated version of the
previous contribution by Inden and Pitsch 8.2 Definition of Atomic
(1991). The tutorial aspects have been Configurations
maintained as much as possible. It is for
this reason that the basic ideas are still 8.2.1 Configurational Variables
presented in terms of a pair interaction
Let us consider a crystalline system with
scheme, which can only provide prototype
N lattice sites and K constituents i = A,
results. An up-to-date treatment of real
B, … . Defining the number of atoms of
systems requires more sophisticated ap-
type i by Ni gives Â Ni = N and the mole
proaches that take into account previously i
neglected but very important physical ef- fractions xi = Ni /N. The distribution of the
atoms on the lattice sites defines the con- by s r = (s 1 , s 2 , …, s r ). The smallest clus-
figuration. ter is a point, the next one a pair, then a
In order to specify a configuration, we triplet, and so on. On an r-site cluster there
need an operator which identifies unequiv- are K r configurations, a much smaller num-
ocally the atomic species on an arbitrary ber than K N. The configurations of the N-
site n. A convenient means of identifica- point system can then be classified into
tion is to associate an integral number with groups with the same number of clusters
each constituent and to define a site opera- N (s r ) with the configuration s r . Instead of
tor s n , which takes these integral values N (s r ), it is preferable to work with the re-
corresponding to the constituent on site n. duced number r (s r ) = N (s r )/Nr , where Nr
Different choices are possible, e.g., is the number of equivalent r-site clusters
contained in the system. These fractions
⎧ K ⎛ K − 1⎞ are called cluster probabilities. They spec-
⎪ 2 ⎝ 2 ⎠ ify the configuration in the r-point cluster
⎧ 0 ⎧1 ⎪
⎪ 1 ⎪2 ⎪ approximation and constitute the most im-
⎪ ⎪ ⎪ 1 portant configurational variables. The ap-
s n = ⎨ 2 or ⎨ 3 or ⎨ (0)
−1 proximation depends on the size of the
⎪ ⎪ ⎪
⎪ K −1 ⎪K ⎪ largest cluster taken into account. For a
⎩ ⎩ ⎪ − K ⎛ − K + 1⎞ three-dimensional (3-dim.) lattice, it is
⎪ 2 ⎝ 2 ⎠
⎩ necessary to include at least one 3-dim.
cluster, otherwise the topological connec-
⎧ t 1 for i = A tion of the clusters for space filling cannot
⎪ t for i = B be taken into account correctly. For in-
or generally ⎨ 2
stance, if only pairs are considered, it is not
⎪
⎩ K for i = K
t possible to distinguish a 3-dim. configura-
tion from a 2-dim. Bethe lattice with the
Any configuration is then specified by the same coordination number.
vector s = (s 1 , s 2 , …, s N ). In total, there So far r defines a particular set of points
are K N different configurations. For binary 1, 2, …, r, and Nr is the number of clusters
alloys, the choice s n = ⎧⎨ B was first sug-
x having the same orientation in space, thus
−
⎩ xA differing from this particular set by a trans-
gested by Flinn (1956). The operator s n is lation in the lattice, or Nr = N. In this in-
sometimes called the spin variable because stance, we speak of oriented clusters and
of its correspondence with the Ising model oriented-cluster probabilities. The N-point
for binary alloys if we take s n = ± 1. Any system usually exhibits more symmetry
function of s n including s n itself is called a elements than translation, and clusters of
configurational variable. different orientation then become equiva-
Because N is a very large number, it is lent. Bringing these clusters together gives
not possible to handle such a large amount Nr > N.
of information. Therefore, we are forced to Suppose that an r-point cluster has been
work with a reduced amount of information selected for the description of the config-
by considering the configurations of much urations. Then the thermodynamic func-
smaller units called clusters. A cluster is tions derived, for example, with CVM, de-
defined by a set of lattice points 1, 2, …, r, pend on the probabilities of this cluster and
and a configuration on this cluster is given also on the probabilities of all subclusters.
Therefore, the total number of configura- ever, to introduce Lagrange parameters in

tional probabilities n tot adds up to the minimization of Gibbs energy in order
to take account of the n p consistency rela-
n tot = ⎛ ⎞ K + ⎛ ⎞ K 2 + … + ⎛ ⎞ K r
r r r tions of Eq. (8-1), and it is not straightfor-
⎝ 1⎠ ⎝ 2⎠ ⎝ r⎠ ward to make use of efficient minimization
= ( K + 1)r − 1 algorithms. Therefore, effort has been con-
centrated in defining a set of independent
because for a point, there are K choices of
variables (or a basis) in the n -dim. space
an element and ⎛ ⎞ choices of a point in
r such that all the configurational variables
⎝ 1⎠ can be expressed in terms of this basis. For
the cluster, for a pair, there are K 2 choices binary alloys, Sanchez and de Fontaine
of two elements and ⎛ ⎞ choices of a pair
r (1978) used multisite correlation func-
⎝ 2⎠ tions as an extension of the pair correlation
of points in this cluster, and so on. How- functions introduced by Clapp and Moss
ever, these n tot probabilities are not all in- (1966). An extension to multicomponent
dependent: if the probabilities of the largest systems was first proposed by Taggart
cluster are given, then the probabilities of (1973) using the spin concept, after which
all subclusters can be derived by partial Sanchez et al. (1984) suggested Chebychev
summation, e.g., for a 3-point cluster polynomials as a basis of the configura-
tional space. In the following another basis
r (s 1 , s 2 , s 3 ) is developed by a method that is equivalent
= ∑ r (s 1 , s 2 , s 3 , s 4 , …, s r ) (8-1) to the one by Taggart. This basis is simpler
s 4 …s r and better suited to numerical applications
than that by Sanchez et al. (1984).
The number of these partial summations in The procedure is as follows. First, we de-
fine a basis in the space of point variables.
np = ⎛ ⎞ K + ⎛ ⎞ K2 + … + ⎛
r r r ⎞ r −1
K This yields (K – 1) functions of s n . Then,
⎝ 1⎠ ⎝ 2⎠ ⎝ r − 1⎠
we consider the (K 2 – 1)-dim. space of the
Moreover, the sum of the probabilities of two-point variables and define a basis by
the largest cluster is equal to 1 by defini- taking products of the (K – 1) basic point
tion. Therefore the number of independent variables, and so on. This explains why the
probabilities is most important step is an appropriate
choice of the basis for point variables.
n = n tot – n p – 1 = K r – 1
that is, n is the dimension of the configura-
8.2.2 Point Variables
tional space. When r and K increase, the
difference n tot – n p becomes very large. In the case of point variables, r corre-
Of course any choice of n probabilities sponds to one point, and the vector s re-
out of the whole set n tot may serve as a set duces to one element s n . In order to define
of independent configurational variables. a basis of configurational point variables,
A convenient choice is any selection of n in particular for the point probabilities
probabilities out of the K r probabilities of r (s n ), it is helpful to introduce a second
the largest cluster. This is usually done in operator pn(i) which allows us to count the
Kikuchi’s natural iteration method (Kiku- number of sites n with the same type of
chi and Sato, 1974). It is necessary, how- atom i for taking averages. This operator
pn(i) is called the site-occupation operator Det M ≠ 0 and so we can invert the ma-
(Clapp and Moss, 1966) and is defined as trix M and define R = (R ik ) = M –1. From
follows: Eq. (8-2) we arrive at
⎧1 if an atom of type i occupies site n K
pn( i ) = ⎨ pn(1) = ∑ R1k s nk − 1
⎩0 otherwise k =1
K
Because i goes from 1 to K we have K op-
erators. Using this defintion we can imme-
pn( 2 ) = ∑ R2 k s nk − 1
k =1
diately write the following equations:
K
⎧ K
⎫ ⎛ ⎞ pn( K ) = ∑ RK k s nk − 1
⎪1 = ∑ pn
(i )
⎪ ⎜ 1 ⎟ k =1 (8-3)
⎪ i = 1K ⎪ ⎜ ⎟
⎪ ⎪ ⎜ ⎛ pn(1) ⎞ ⎛ 1 ⎞
⎪s n = ∑ t i pn
(i )
⎪ sn ⎟
i =1 ⎜ ⎟ ⎜ pn( 2 ) ⎟ ⎜ sn ⎟
⎪ ⎪ ⎜
⎨ 2 K
⎬ or ⎜ ⎟ or
⎜ pn( 3) ⎟⎟= R ⎜ s n2 ⎟
⎜ ⎟
s = ∑ t 2 (i )
p ⎜ s 2 ⎟
⎪ n i =1 i n ⎪ ⎜
n
⎟ ⎜⎜ ⎟⎟ ⎜ ⎟
⎪ ⎪ ⎜ ⎝ pn( K ) ⎠ ⎝ s nK − 1 ⎠
⎪ ⎪ ⎟
⎜ ⎟
⎪ K −1 K
K −1 (i ) ⎪ ⎜ K −1⎟
⎪s n = ∑ t i
This is the relation that connects the site-
pn ⎪ ⎜s n ⎟
⎩ i =1 ⎭ ⎝ ⎠ (8-2) occupation operators with the site opera-
tors. Because all the elements of the first
⎛ 1 1 1 … 1 ⎞ ⎛ pn ⎞ (1)
row of M are unity, the inverse matrix R
⎜ t1 t2 t 3 … t K ⎟ ⎜ pn( 2 ) ⎟
= ⎜ t12 t 32 … t K2 ⎟ ⎜ pn( 3) ⎟
exhibits the following properties: the ele-
t 22
⎜ ⎟ ⎜ ⎟ ments of the first column of R add up to
⎜⎜ K⋅− 1 K⋅− 1 K⋅− 1 … K⋅− 1 ⎟⎟ ⎜⎜ ⋅ ⎟⎟ unity and the elements of all other columns
⎝ t1 t2 t3 … t K ⎠ ⎝ pn( K ) ⎠ K
where the upper indices signify powers

add up to zero, i.e., ∑ Ri k = d1k .
i =1
of s n or t i . This notation will be used
throughout this chapter. To distinguish an
upper index from a power, the index will be 8.2.3 Point Correlation Functions
put in parentheses, except if it is a double and Point Probabilities
or multiple index, which cannot be con- So far we have considered the point var-
fused with a power. Let us call M the ma- iables for one arbitrary point n out of the N
trix of this system of equations. Its deter- lattice points of the crystal. As mentioned
minant is the van der Monde determinant, before, we want to reduce the number of
which is given by parameters describing an atomic configura-
K r tion by considering averages over equiva-
Det M = ∏ ∏ (t j − t i ) lent clusters, or in this case, points. The
j = i +1 i =1
space group of the structure will define the
Because all the t i are different numbers, equivalence of points. We may enumerate
this determinant is different from zero, and the classes of equivalent points by 1, 2, …
the equations are linearly independent. and define the number of lattice points in
This confirms that the K – 1 functions {s n , each class by N (1), N (2), …, N (L). The aver-
s n2 , …, s nK –1} are linearly independent. age of an arbitrary function j (s n , s n2 , …)
of the point variables s n , s n2 , … is defined tation, i.e., if n and m are non-equivalent

by points. In this instance, we speak of
N (n) oriented pair probabilities. In the opposite
1
〈j (s n , s n2 , …)〉 = ∑ j (s s , s s2 , …) case, we speak of isotropic pair probabil-
N (n) s =1
ities:
The index n now stands for one representa- K K
tive point out of the corresponding class. ij
rnm = 〈 pn( i ) pm( j ) 〉 = ∑ ∑ Rih R j k
The averages of the site-occupation opera- k =1 h =1
h −1 k −1
tors pn(i) yield the probabilities of finding an × 〈s n s m 〉 (8-5)
atom A, B, …, K on a point n:
The pair probabilities can thus be ex-
K pressed in terms of the already introduced
r (s n = t i ) = 〈 pn( i ) 〉 = ∑ Ri k 〈s nk − 1 〉 point correlation functions, written here in
k =1
i = 1, …, K (8-4) the form ·s nh –1 s mk –1Ò with h = 1 and k = 2, 3,
…, K (or k = 1 and h = 2, 3, …, K ) together
In the limit of N Æ ∞, this average cor- with a second set of functions, ·s nh –1 s mk –1Ò,
responds to the thermodynamic site-occu- with h, k = 2, …, K called pair correlation
pation probability. The (K – 1) functions functions. The number of pair correlation
·s nK –1Ò are called point correlation func- functions is (K – 1)2 if n and m are non-
tions. These functions constitute the basis equivalent sites. The total number of point
of the configurational space of point vari- and pair correlation functions adds up to
ables, and Eq. (8-4) yields the components 2 (K – 1) + (K – 1)2 = K 2 – 1. This is exactly
of the point probabilities with respect to the dimension of the configurational space
this basis. if we assume the pair to be the basic clus-
ter. Point and pair correlation functions to-
8.2.4 Pair Variables, Correlation Func- gether constitute a basis in this instance.
tions, and Probabilities By virtue of the relation Â pn(i) pm( j) = pn(i)
j
Consider a two-point cluster r = n, m. We the pair probabilities are consistent with
can immediately introduce a pair-occupa- the point probabilities.
tion operator that takes the value 1 if an For the purposes of illustration, let us
atom of type i occupies site n, and if type j consider two examples:
occupies site m. This operator is simply the
product of the two previously introduced Binary alloy
ij
site-occupation operators pnm = pn(i) pm( j).
Using the expressions of the site operators, For a binary alloy, K = 2 and t 1 = 1 for
Eq. (8-3), we obtain the first element A, and t 2 = – 1 for the sec-
ond element B. The point probabilities fol-
K K low from
ij
pnm = ∑ ∑ Rih Rj k s nh − 1 s mk− 1
k =1 h =1 1 − 1 − 1⎞
M=⎛
1 1⎞
and R = − ⎛
The pair probabilities ij
rnm
are obtained by ⎝ 1 − 1⎠ 2 ⎝ − 1 1⎠
ij
taking the average of the operator pnm over which yields
all equivalent pairs in the crystal. It is
⎛ 1 ⎞ = ⎛ 1 1⎞ ⎛ pn ⎞
(1)
worth mentioning that the pairs may be
non-equivalent if they differ by their orien- ⎝ s n ⎠ ⎝ 1 − 1⎠ ⎝ pn( 2 ) ⎟⎠
⎜
and Ternary alloy

⎛ pnA ⎞ ⎛ pn(1) ⎞ 1 ⎛ 1 1⎞ ⎛ 1 ⎞ For a ternary alloy, K = 3 and t 1 = 1 for
⎜ B ⎟ = ⎜ ( 2 ) ⎟ = ⎝ 1 − 1⎠ ⎝ s ⎠
⎝ pn ⎠ ⎝ pn ⎠ 2 n element A, t 2 = 0 for B, and t 3 = – 1 for C.
In this instance, the point variables follow
and consequently from
(8-6)
⎧r n(1) = 12 (1 + 〈s n 〉 ) ⎛ 1 1 1⎞
⎨ (2) 1 and 〈s n 〉 = r n − r n( 2 )
(1)
M = ⎜ 1 0 − 1⎟
⎩r n = 2 (1 − 〈s n 〉 ) ⎜ ⎟
⎝ 1 0 1⎠
For the pair probabilities, we obtain, ac-
cording to Eq. (8-5), and
2 2 1 ⎛ 0 1 1⎞
ij
r nm = ∑ ∑ Rih Rjk 〈s nh −1 s mk−1 〉 R= ⎜ 2 0 − 2⎟
2 ⎜ ⎟
k =1 h =1 ⎝ 0 − 1 1⎠
or
According to Eqs. (8-2) and (8-5), we ob-
⎛ rnm
AA
⎞ ⎛ rnm
11
⎞ tain
AB ⎟
⎜ rnm 12 ⎟
⎜ rnm
⎜ BA ⎟ = ⎜ 21 ⎟ (8-7) 〈s n 〉 = rn(1) − rn( 3)
⎜ rnm ⎟ ⎜ rnm ⎟ 〈s n2 〉 = rn(1) + rn( 3)
⎜ BB ⎟ ⎜ 22 ⎟
⎝ rnm ⎠ ⎝ rnm ⎠
rnA = rn(1) = 12 ( 〈s n 〉 + 〈s n2 〉 ) (8-8)
⎛ 1 1 1 1⎞ ⎛ 1 ⎞
rnB = rn( 2 ) = (2 −
1
2 〈s n2 〉 )
1 ⎜ 1 1 − 1 − 1⎟ ⎜ 〈s n 〉 ⎟ 2
= ⎜ ⎟⎜ ⎟
4 ⎜ 1 − 1 1 − 1⎟ ⎜ 〈s m 〉 ⎟ rnC = rn( 3) = ( − 〈s n 〉 + 〈s n2 〉 )
1
2
⎝ 1 − 1 − 1 1⎠ ⎝ 〈s n s m 〉⎠
For the pair probabilities, we obtain, ac-
cording to Eq. (8-5),
⎛ rnm
AA
⎞ ⎛ rnm11
⎞ ⎛0 0 0 0 1 1 0 1 1⎞ ⎛ 1 ⎞
⎜ rnm ⎟ ⎜ rnm
AB 12 ⎟ ⎜0 2 2 0 0 0 0 − 2 − 2 ⎟ ⎜ 〈s n 〉 ⎟
⎜ AC ⎟ ⎜ 13 ⎟ ⎜ ⎟ ⎜ ⎟
1 ⎟ ⎜ 〈s n 〉 ⎟
2
⎜ rnm ⎟ ⎜ rnm ⎟ ⎜0 0 0 0 −1 −1 0 1
⎜ rnm
BA ⎟ ⎜ rnm
21 ⎟ ⎜0 0 0 2 0 −2 2 0 − 2 ⎟ ⎜ 〈s m 〉 ⎟
⎜ BB ⎟ ⎜ 22 ⎟ 1 ⎜ ⎟ ⎜ ⎟
⎜ rnm ⎟ = ⎜ rnm ⎟ = 4 ⎜ 4 0 −4 0 0 0 −4 0 4 ⎟ ⎜ 〈s n s m 〉 ⎟ (8-9)
⎜ rnm
BC ⎟ ⎜ rnm
23 ⎟ ⎜0 0 0 −2 0 2 2 0 − 2 ⎟ ⎜ 〈s n s m 〉⎟
2
⎜ CA ⎟ ⎜ 31 ⎟ ⎜ ⎟
1 ⎟ ⎜ 〈s m 〉 ⎟
2
⎜ rnm ⎟ ⎜ rnm ⎟ ⎜0 0 0 0 −1 1 0 −1
⎜ 2 ⎟
⎜ rnm
CB
⎟ ⎜ rnm32
⎟ ⎜0 −2 2 0 0 0 0 2 − 2 ⎟ ⎜ 〈s n s m 〉 ⎟
⎜ CC ⎟ ⎜ 33 ⎟ ⎜ ⎟ ⎜ 2 2 ⎟
⎝ rnm ⎠ ⎝ rnm ⎠ ⎝0 0 0 0 1 −1 0 −1 1 ⎠ ⎝ 〈s n s m 〉⎠
8.2.5 Generalized Cluster Variables, correlation functions constitutes a basis of

Correlation Functions, and Probabilities this configurational space. If all sites are
The point and pair variables are the sim- equivalent, the number of correlation func-
plest examples of the general cluster vari- tions is much smaller.
ables. Consider an oriented cluster defined
by r points. As before, we introduce a clus- 8.2.6 Short-Range Order (sro)
ter-occupation operator by the product of Configurations – Long-Range Order
site-occupation operators p1(i1) p2(i2 ) … pr(ir ). (lro) Configurations
The cluster correlation functions are the
A configuration is called short-range or-
averages over all equivalent clusters in the
dered if the whole lattice constitutes one
crystal. Thus the result for the cluster prob-
class of lattice points (i.e., N (1) = N). As a
abilities is
consequence, all the cluster correlation
i1 i2 … ir
… r = 〈 p1 p2 … pr( ir ) 〉
( i1 ) ( i2 ) functions are isotropic. A configuration is
r12
K K K called long-range ordered if at least two
= ∑ ∑ … ∑ Ri1 k1 Ri2 k2 … Rir kr classes of points are to be distinguished by
k1 = 1 k 2 = 1 kr = 1 their different average occupation. These
k1 − 1 k 2 − 1
× 〈s 1 s 2 … s rk r − 1 〉 (8-10) classes of lattice points are usually called
r
sublattices. In Fig. 8-1, the face-centered
The parameters ∏ s nkn −1 include the r- cubic unit cell is shown with a subdivision
n =1 of the lattice sites into four simple cubic
point correlation functions (point, pair, sublattices labeled 1 to 4 with N/4 points
etc.) which provide the characterization of each. With these sublattices, the super-
the configuration by means of r-site clus- structures L12 , L10 and P4/mmm can be
ters, for example, the point correlation described. Each point on a sublattice is sur-
functions are coded in the form ·s m Ò = rounded in its first-neighbor shell by four
·s 10 … s 1m … s r0 Ò, ·s m 2
Ò = ·s 10 … s m
2
… s r0 Ò, points corresponding to each of the other
etc. and, correspondingly, for the pair cor-
relation functions, they are coded ·s m s n Ò =
·s 10 … s 1m s 1n … s r0 Ò, etc. The r-site cluster
correlation functions are obtained when
kn ≠ 1 for all values of kn . If all the r sites
are not equivalent, the total number of
r-point correlation functions is (K – 1)r.
The total number of correlation functions
Nc required for an r-point cluster treatment,
including all the subcluster correlation
functions, is
Nc = ⎛ ⎞ ( K − 1) + ⎛ ⎞ ( K − 1)2 + …
r r
⎝ 1⎠ ⎝ 2⎠
+ ⎛ ⎞ ( K − 1)r = ( K − 1 + 1)r − 1 = K r − 1
r Figure 8-1. Subdivision of the face-centered cubic
⎝ r⎠ (f.c.c.) unit cell into four simple cubic sublattices for
a characterization of the superstructures L12 , L10 and
This is the dimension of the configura- P4/mmm. The occupations of the sublattices for
tional space, and the whole set of cluster these structures are given in Table 8-1.
Table 8-1. F.c.c. and b.c.c. superstructures and values of their lro parameters.
Designation/Spacegroup Point probabilities lro parameter Values for maximum

lro (binary)
A1 (Cu) A2 (Fe) r1(i) = r 2(i) = r 3(i) = r 4(i) x1 = x 2 = x 3 = 0 r1A = 1 – xB

L10 (CuAu) B2 (CsCl) r1(i) = r 2(i) ≠ r 3(i) = r 4(i) x1 ≠ 0, x 2 = x 3 = 0 r1A = 1, r 3A = 1 – 2 xB
L12 (Cu3Au) r1(i) ≠ r 2(i) = r 3(i) = r 4(i) x1 ≠ x 2 ≠ 0, x 3 = 0 r1A = 1 – 4 xB , r 2A = 1
D03 (Fe3Al) r1(i) ≠ r 2(i) ≠ r 3(i) = r 4(i) x1 ≠ x 2 ≠ 0, x 3 = 0 r1A = 1 – 4 xB , r 2A = 1,
xB Ù 0.25 r 3A = 1
P4/mmm D03 (Fe3Al) r1(i) ≠ r 2(i) ≠ r 3(i) = r 4(i) x1 ≠ x 2 ≠ 0, x 3 = 0 r1A = 0, r 2A = 2 – 4 xB ,
0.25 Ù xB Ù 0.5 0.25 Ù xB Ù 0.5 r 3A = 1
B32 (NaTl) r1(i) = r 3(i) ≠ r 2(i) = r 4(i) x1 = 0, x 2 ≠ x 3 ≠ 0 r1A = 1, r 2A = 1 – 2 xB
–
F4 3m r1(i) ≠ r 2(i) ≠ r 3(i) ≠ r 4(i) x1 ≠ x 2 ≠ x 3 ≠ 0
0.25 Ù xB Ù 0.5
sublattices. Therefore these structures are

called ordered in the first shell. The sublat-
tice occupation for these structures is given
in Table 8-1.
In the body-centered cubic structure,
four face-centered cubic sublattices with a
parameter a = 2 a0 can be defined as shown
in Fig. 8-2. With these sublattices, the
–
superstructures B2, D03 , F43m, and B32
can be described. These structures are or-
dered in the first two neighbor shells. The
occupation of the sublattices is again given
in Table 8-1. The complete crystallo- Figure 8-2. Subdivision of the body-centered cubic
graphic specification of these superstruc- (b.c.c.) unit cell into four f.c.c. sublattices with twice
the lattice constant, 2 a0 , for a characterization of the
tures is given in Sec. 8.7, Table 8-13. superstructures B2, D03 and B32. The occupation of
We must therefore consider the follow- the sublattices for these structures are given in Table
ing variables: the points ·s 1 Ò, ·s 2 Ò, ·s 3 Ò, 8-1. The (irregular) tetrahedron cluster 1234 used in
·s 4 Ò; the pairs ·s 1 s 3 Ò, ·s 1 s 4 Ò, ·s 2 s 3 Ò, the CVM calculations is also shown.
·s 2 s 4 Ò, ·s 1 s 2 Ò, ·s 3 s 4 Ò; the triplets
·s 1 s 2 s 3 Ò, etc.; and, finally, the tetrahedron
·s 1 s 2 s 3 s 4 Ò – in total, (24 – 1) = 15 vari- Among these variables, we need four vari-
ables. Of course, the variables have differ- ables to define the isotropic f.c.c. sro state:
ent meanings in the two structures. In the ·s 1 Ò, ·s 1 s 2 Ò, ·s 1 s 2 s 3 Ò, and ·s 1 s 2 s 3 s 4 Ò;
f.c.c. lattice, all the pairs are nearest neigh- one more variable, ·s 1 s 3 Ò, is needed to
bors, while in the b.c.c. structure, the pairs characterize the isotropic b.c.c. sro state.
of type 1– 2 and 3 – 4 are second-nearest- The remaining variables characterize the
neighbor pairs. A corresponding distinc- lro state. For many applications, it is suffi-
tion among the triplets also has to be made. cient to identify the existence of lro by the
8.3 The Existence Domain and Configuration Polyhedron 531
difference between the point correlation In practice, the values of the pair correla-
functions or linear combinations between tion functions for the sro states are ob-
them (e.g., the same combinations as those tained from diffraction experiments. For
entering the structure factor for X-ray dif- this purpose, we usually introduce the
fraction). These linear combinations of the Cowley –Warren sro parameters a nij– m
point correlation functions are called lro (Cowley, 1950), which are directly related
parameters. In the case of the b.c.c. struc- to the measured intensities. They are de-
ture, we usually define the lro parameters fined as the deviation from the random
state:
x1 = ·s 1 Ò + ·s 2 Ò – ·s 3 Ò – ·s 4 Ò
x2 = ·s 1 Ò – ·s 2 Ò r nij− m = r nm
ij ji
+ r nm
x3 = ·s 3 Ò – ·s 4 Ò = [r n( i ) r m( j ) + r n( j ) r m( i ) ] (1 − anij− m )
Table 8-1 lists the b.c.c. superstructures, The index n – m stands for the neighbor dis-
together with the values of the parameters. tance between the positions n and m. Tak-
If there is no long-range order, then all ing again the t i values as before (i.e., ± 1
the positions are equivalent and the point for the binary case and 1, 0, – 1 for the ter-
probability takes one value r n(i) = x i . Taking nary case), we obtain the isotropic Cow-
ley –Warren sro parameters:
〈s n s m 〉 − 〈s n 〉 〈s m 〉
binary alloys: an − m =
1 − 〈s n 〉 〈s m 〉
〈s n s m2 〉 + 〈s n2 s m 〉 + 2 〈s n2 s m2 〉 − 4 xA (1 − xB )
ternary alloys: anAB
−m =
4 xA x B
− 〈s n2 s m2 〉 + 〈s n s m 〉 + 4 xA xC
anAC
−m =
4 xA x C
− 〈s n2 s m 〉 − 〈s n s m2 〉 + 2 〈s n2 s m2 〉 − 4 x C (1 − xB )
anBC
−m =
4 x B xC
the same t i values as in the examples 8.3 The Existence Domain

treated previously in Sec. 8.2.3, we can and Configuration Polyhedron
use Eq. (8-6) and obtain the following
values for the point correlation functions: In the preceding section, we developed a
·s n Ò = ·s 1 Ò = xA – x B in the case of a binary set of correlation functions (point, pair,
alloy and ·s 1 Ò = xA – x C , ·s 12 Ò = xA + x C in triplet, etc.) for characterizing atomic con-
the case of a ternary alloy. figurations in multicomponent systems.
If there is not even any short-range or- These correlation functions are indepen-
der, we consider the completely disordered dent variables that define the configura-
state. All the n-point correlation functions tional space, but they can only vary within a
then take a limiting value; for example, in restricted range of numerical values due to
a binary alloy, ·s 1 … s r Ò takes the value certain consistency conditions, which will
(xA – x B )r . now be discussed. Outside this restricted
range, the numerical values do not define Adopting again the values t i = ± 1, which
actually existing atomic configurations. were already used in the examples of Sec.
Therefore, this restricted range of values 8.2, the variables can both vary in the inter-
has been called the existence domain, or val [– 1, 1] (see Fig. 8-3): x1 takes the value
configuration polyhedron (Kudo and Kat- 1 for pure A and – 1 for pure B, x2 takes 1
sura, 1976). The dimension of the configu- for each pure component and – 1 for an al-
rational space depends either on the num- loy composed only of A– B pairs. These
ber of correlations which are taken into ac- ranges, however, are not fully accessible,
count or, equivalently, on the size of the for example because of the obvious con-
largest cluster. In the following sections the straints
existence domains will be investigated for
r1AA
2 7 0, r1BB
2 7 0, r1AB
2 70 (8-11)
some simple examples. The method ap-
plied closely follows the procedure devel- The constraints in Eq. (8-11) yield the fol-
oped by Finel (1987). For the sake of sim- lowing inequalities:
plicity, the analysis will be restricted to bi-
r1AA
2 7 0 fi 1 + 2 x1 + x2 7 0
nary alloys. For multicomponent systems,
AB
the procedure is analogous but has to be r1 2 7 0 fi 1 – x2 7 0 (8-12)
BB
carried out on a computer. r1 2 7 0 fi 1 – 2 x 1 + x 2 7 0
Existence domains are most useful for an
analysis of ground states. This will be shown
in Sec. 8.4. The ground-state analysis will be
limited to a finite range of interactions (e.g.
pair interactions between first and second
nearest neighbors). Therefore, we are partic-
ularly interested in the part of the existence
domain that is the subspace of correlation
functions required for a treatment with fi-
nite-range interactions. This existence do-
main in the configurational subspace will
be investigated in detail in this section.
8.3.1 F.C.C. Structure,

First-Neighbor Interactions
Figure 8-3. Projection of the configuration polyhe-
The first step in this analysis is to make a dron of a regular tetrahedron into the plane of corre-
choice for the basic cluster. This choice lation functions x1 = ·s 1Ò = xA – xB and x2 = ·s 1 s 2 Ò for
then determines the dimension of the con- different choices of clusters: (1) domain abg: nearest-
neighbor pair; (2) domain abfh: nearest-neighbor tri-
figurational space to be dealt with. It is angle; (3) shaded area abcde: nearest-neighbor tetra-
natural to start with the simplest possible hedron. The vertices a and b correspond to the pure
basic cluster, the nearest-neighbor pair. The components A and B, c and e correspond to L12 with
configurational variables are then the point compositions B3A and A3B, and d to L10 , composition
correlation function x1 = ·s 1 Ò = xA – x B , AB. The dashed line indicates the random configura-
tions. This line separates the existence domain into re-
which defines the composition, and the gions of (ordered) configurations with preferential for-
nearest-neighbor pair correlation function mation of unlike pairs (lower part) and configurations
x 2 = ·s 1 s 2 Ò = 1 – 4 r1AB
2 (see Eq. (8-7)). with preferential formation of like pairs (upper part).
If the relations in Eq. (8-12) are taken as new variable x3 = ·s 1 s 2 s 3 Ò must be intro-
equalities, they define a triangle in the duced, and the configurational space is
space (x1 , x2 ), as shown in Fig. 8-3. Any now three-dimensional. The constraints for
point inside the triangle abg fulfills these the triplet probabilities are
relations. However, these relations are (8-14)
AAA
not sufficient to define the existence do- r12 3 7 0 fi ⎛ 1 3 3 1⎞ ⎛ 1 ⎞
main. In fact, the vertex g, for instance,
AAB
r12 3 7 0 fi ⎜ 1 1 − 1 − 1⎟ ⎜ x1 ⎟
r12 3 7 0 fi ⎜⎜ 1 − 1 − 1 1⎟⎟ ⎜⎜ x2 ⎟⎟
70
corresponds to r1AA BB
2 = r1 2 = 0 (introduce
ABB
x1 = ·s 1 Ò = 0 and x2 = ·s 1 s 2 Ò = – 1 into Eq. 3 7 0 fi ⎝1 − 3

BBB
r12 3 − 1⎠ ⎝ x3 ⎠
(8-7)), which defines a configuration built
up with A– B pairs only. Such a struc- Because we are only interested in the do-
ture cannot be formed in the f.c.c. lattice. main spanned in the subspace (x1 , x2 ), we
In this structure, the nearest-neighbor have to eliminate the variable x3 . In order
bonds form a triangular network, and it is to converse the positive value of all the ex-
not possible to form a triangle with only pressions, this elimination can only be per-
A– B pairs. Point g in Fig. 8-3 thus corre- formed by additive combinations of the in-
sponds to a physically impossible state. equalities. This yields the three expressions
Obviously, there must be further restricting in Eq. (8-12) already obtained from the
relations. pairs, plus one more inequality:
The conditions r1i j2 &1 do not introduce
1 + 3 x2 7 0
new constraints because they are automati-
cally fulfilled together with the inequalities Taken as an equality, this equation defines
in Eq. (8-11). For the present choice of t i the line f – h in Fig. 8-3, and we have to ver-
values, this is immediately seen from the ify again if the new vertices f and h corre-
definition in Eq. (8-7). The general case spond to physically accessible states. Point
follows from Eq. (8-2), which yields the re- f corresponds to a state with no triangle of
lation: type AAA, AAB, BBB (i.e., a state built up
only with triangles of type ABB). It is easy
r1AA BB AB BA
2 + r1 2 + r1 2 + r1 2 = 1 (8-13)
to see that this is impossible in the f.c.c.
If the constraints in Eq. (8-11) are fulfilled, structure. Thus this procedure has to be
then Eq. (8-13) is a sum of positive terms continued in order to find further restrict-
which can only be equal to one if each term ing conditions.
is less than or equal to one. The next higher cluster that may be con-
Hence, we have to deduce further in- sidered is the nearest-neighbor tetrahedron
equalities from higher-order clusters. Let r = 1234, see Fig. 8-1. In this case, the new
us consider the nearest-neighbor triangle variable x 4 = ·s 1 s 2 s 3 s 4 Ò needs to be intro-
123 as the next cluster, see Fig. 8-1. The duced. The tetrahedron probabilities result
in the following consistency relations:
AAAA
r12 34 7 0 fi ⎛1 4 6 4 1⎞ ⎛ 1 ⎞
AAAB
r12 3 4 7 0 fi ⎜1 2 0 − 2 − 1⎟ ⎜ x1 ⎟
⎜ ⎟⎜ ⎟
⎜1 0 − 2
AABB
r12 34 7 0 fi 0 1⎟ ⎜ x2 ⎟ 7 0 (8-15)
⎜1 − 2 2 − 1⎟ ⎜ x3 ⎟
ABBB
r12 34 7 0 fi 0
⎜ ⎟⎜ ⎟
⎝1 − 4 6 − 4 1⎠ ⎝ x 4 ⎠
BBBB
r12 3 4 7 0 fi
Taken as equalities, these equations define an increasing number of correlation func-

a closed polyhedron in the space (x1, …, x 4 ). tions and, consequently, with more in-
Because we are still interested in the equalities that have to be taken into ac-
subspace (x1 , x2 ) we have to eliminate x3 count in order to define the existence do-
and x 4 . This is done by taking suitable ad- main. Therefore we look for a simplifica-
ditive combinations of the inequalities in tion. At first we notice that each one of the
Eq. (8-15). The result is the polygon abcde consistency relations, when taken as an
in Fig. 8-3, which differs from the pre- equality, defines a face of a polyhedron.
vious one in that the vertices f and h In the cases studied before (see, for exam-
are removed, and the vertices c, d, e are ple, Eq. (8-14) or (8-15)), the number of
added. It is easy to see that all these ver- consistency relations was one more than
tices correspond to possible atomic con- the dimension of the configurational space.
figurations in the f.c.c. structure; c and e The polyhedron was thus built by d + 1
correspond to the L12 structure with stoi- faces. Therefore, the polyhedron is a sim-
chiometric compositions B3A and A3B, re- plex, which is necessarily closed and con-
spectively, and d corresponds to the L10 vex. Generally, if d is the dimension of
structure with composition AB. The exis- the space, a simplex is a polyhedron with
tence domain is given by the shaded area in d + 1 faces. The intersection of d faces
Fig. 8-3. defines a vertex, a simplex thus has d – 1
The regular tetrahedron is thus the small- vertices. The coordinates of the vertices
est basic cluster that allows us to define the are obtained by solving the systems of d
existence domain in the configuration equations which can be chosen out of the
space of an f.c.c. lattice with nearest-neigh- whole set of d + 1. The simplification
bor interactions. Consequently, this tetra- comes from a very convenient way of ob-
hedron is also the smallest basic cluster taining these coordinates: Finel (1987)
upon which a statistical treatment should showed that the values obtained from
be based. It is possible, of course, to take a these solutions are the same as those ob-
higher cluster containing the tetrahedron as tained simply by dividing the elements in
a subcluster. In such a case, the dimension each column of the matrix by the cor-
of the configurational space is increased responding values of the first row. The
but the projection of the domain into the coordinates of the vertices are the elements
subspace of point correlations and nearest- in each row following the first column,
neighbor pair correlations remains the which always contains unity and does not
same. If we had considered a basic cluster define a coordinate. Performing this proce-
that does not contain the tetrahedron, we dure with the matrix of Eq. (8-14), we
would have obtained a polygon with some directly obtain the coordinates given in
vertices, which again would not corre- Table 8-2. Taking the projection into the
spond to accessible states. The procedure subspace (x1 , x2 ) means we conserve only
would have to be continued until a cluster the first two coordinates. This exactly de-
containing the tetrahedron were finally se- fines the polygon abfh, which was found
lected. before by means of the elimination proce-
The procedure of obtaining the vertices dure.
of the configuration polyhedron outlined Applying the same procedure to Eq.
above becomes more and more inconven- (8-15) immediately yields the coordinates
ient with increasing cluster size, i.e., with of the configuration polyhedron in the
Table 8-2. Coordinates of the vertices shown in Fig. 8.3.2 F.C.C. Structure, First and
8-3. Second Neighbor Interactions
Vertex x1 x2 x3
First and second neighbor interactions
a 1 1 1 correspond more closely to the situation
h 1/3 – 1/3 –1 encountered in real alloys (Inden, 1977 a)
f – 1/3 – 1/3 1 than the previous case does. The internal
b –1 1 –1 energy contains one more term, the second-
nearest neighbor interaction. Consequently
we will finally be interested in the exis-
tence domain within the subspace of point
Table 8-3. Coordinates of the vertices of the config- and first and second neighbor pair correla-
uration polyhedron shown in Fig. 8-3. tion functions. The interesting part of the
existence domain will be a polyhedron in
Configuration Correlation functions
this three-dimensional space.
Vertex x1 x2 x3 x4 The analysis begins with the octahedron
r = 234567 in Fig. 8-4, which is defined as
Pure A a 1 1 1 1 a basic cluster because it includes both
L12 (A3B) e 1/2 0 – 1/2 –1 types of interaction. For this cluster, nine
L10 (AB) d 0 – 1/3 0 1 correlation functions have to be intro-
L12 (AB3 ) c – 1/2 0 1/2 –1 duced:
Pure B b –1 1 –1 1
x1 = ·s 2 Ò x6 = ·s 2 s 3 s 4 s 7 Ò
x2 = ·s 2 s 3 Ò x7 = ·s 3 s 4 s 5 s 6 Ò
x3 = ·s 2 s 7 Ò x8 = ·s 2 s 3 s 4 s 5 s 7 Ò
four-dimensional space (x1 , …, x 4 ), see x4 = ·s 2 s 3 s 4 Ò x9 = ·s 2 s 3 s 4 s 5 s 6 s 7 Ò
Table 8-3. The vertices abcde in Fig. 8-3 of x5 = ·s 2 s 3 s 7 Ò
the two-dimensional polygon in the sub-
space (x1 , x2 ) can immediately be taken x1 again corresponds to xA – x B . The exis-
from Table 8-3. tence domain will be a polyhedron in 9-
dim. space (x1 , …, x 9 ), which is defined by
the consistency relations for the various
configurations on the octahedron cluster:
i j… k
AAAAAA : ⎛ 1 6 12 3 8 12 12 3 6 1⎞ ⎛ 1⎞
BAAAAA : ⎜1 4 4 1 0 0 −4 −1 −4 − 1⎟ ⎜ x1 ⎟
BBAAAA : ⎜ 1 2 0 −1 0 − 4 0 −1 2 1⎟ ⎜ x2 ⎟
BAAAAB : ⎜1 2 − 4 3 − 8 4 −4 3 2 1⎟ ⎜x ⎟
⎜ ⎟ ⎜ 3⎟
BBBAAA : i j …k 1 ⎜ 1 0 0 − 3 0 0 0 3 0 − 1⎟ ⎜ x 4 ⎟ 70
r =
BBAAAB : 23…7 2 6 ⎜ 1 0 − 4 1 0 0 4 −1 0 − 1⎟ ⎜ x5 ⎟
BBBAAB : ⎜ 1 −2 −4 3 8 −4 −4 3 −2 1⎟ ⎜ x6 ⎟
ABBBBA : ⎜1 − 2 0 −1 0 4 0 −1 −2 1⎟ ⎜x ⎟
⎜ ⎟ ⎜ 7⎟
BBBBAB : ⎜ 1 − 4 4 1 0 0 −4 −1 4 − 1⎟ ⎜ x8 ⎟
BBBBBB : ⎝ 1 − 6 12 3 − 8 − 12 12 3 −6 1⎠ ⎝ x9 ⎠
Figure 8-4. Tetrahedron and octahedron clusters in

the face-centered-cubic unit cell.
Figure 8-5. Projection of the configuration polyhe-
dron of an octahedron cluster r = 234567 (see Fig. 8-
4) into the subspace of point and first and second
Taken as equalities, these 10 equations de- neighbor pair correlation functions: x1 = ·s 2 Ò,
fine a simplex in the 9-dim. configurational x2 = ·s 2 s 3Ò, x3 = ·s 2 s 7 Ò, respectively. Not all the ver-
tices correspond to existing states. The shaded area
space. From these equations, the vertices of
represents the section for x3 = 0, i.e. plane (x1 , x2 ).
the simplex can be calculated either by Comparison with Fig. 8-3 reveals that this section de-
solving nine selected equations out of the fines a larger domain. This shows that the octahedron
whole set of 10 or obtained directly by does not impose the same constraints as the tetrahe-
Finel’s method outlined above. dron cluster.
Let us consider the projection of this
simplex into the subspace (x1 , x2 , x3 ). This
projection is obtained from columns 2 – 4
of the matrix. The result is shown in Fig.
8-5. A verification of whether or not all the
vertices correspond to possible atomic ar-
rangements is necessary. For example, ver-
tex (1/3, – 1/3, 1) does not correspond to an
existing state because it would be built
only with octahedra of type BAAAAB.
This is topologically impossible in the
f.c.c. structure. Therefore, we have to
consider further constraints, for instance
those resulting from the nearest-neighbor
tetrahedron, analyzed before. We can thus
use the previous results. In the space
(x1 , x2 , x3 ), the previously obtained exis-
tence domain would be the rectangular Figure 8-6. Configuration polyhedron of a nearest-
prism built on the basis given by the shaded neighbor tetrahedron cluster 1234 (Fig. 8-4) in the
space (x1 , x2 , x3 ). This cluster does not contain sec-
area in Fig. 8-3 (see also Fig. 8-6). If we ond-nearest neighbor pairs. The polyhedron is thus a
now take into account the constraints im- rectangular prism built with the existence domain in
posed by both the tetrahedron and the octa- Fig. 8-3 as basis.
Table 8-4. Coordinates of the vertices of the config-

uration polyhedron shown in Fig. 8-7.
x1 x2 x3
Pure A (A1) 1 1 1
A5B, AB5 ±2/3 1/3 1/3
A3B, AB3 (L12 ) ±1/2 0 1
A3B, AB3 (D022 ) ±1/2 0 2/3
A2B, AB2 (Pt2Mo) ±1/3 – 1/9 1/9
A2B, AB2 ±1/3 0 – 1/3
AB (L10 ) 0 – 1/3 1
AB (L11) 0 0 –1
A 2B 2 0 – 1/3 1/3
Pure B (A1) –1 1 1
Figure 8-7. Intersection of the two configuration
polyhedra corresponding to the tetrahedron and the
octahedron, Fig. 8-4 and Fig. 8-5. All vertices of this
polyhedron correspond to existing configurations on statement made in the previous paragraph
the f.c.c. lattice. The corresponding superstructures that if we reduce the interaction range to
are indicated. Notice the symmetry with respect to
the stoichiometric composition AB. The vertical sec-
nearest neighbors and keep the two basic
tions represent the existence domains for a fixed al- clusters, the projected existence domain in
loy composition, here AB and A3B. the space (x1 , x2 ) does not change due to
the introduction of the octahedron, a larger
cluster than the tetrahedron.
hedron, we will have to intersect the two
polyhedra. The result of the intersection of
8.3.3 B.C.C. Structure, First and
Fig. 8-5 and Fig. 8-6 is shown in Fig. 8-7,
Second Neighbor Interactions
and the coordinates of the vertices are
given in Table 8-4. It is easy to show that The treatment of the b.c.c. structure is
all the vertices of this polyhedron corre- much simpler than that of the f.c.c. struc-
spond to existing atomic configurations. ture because an irregular tetrahedron as
Consequently, in the f.c.c. structure with a basic cluster already contains first and
first and second nearest-neighbor interac- second neighbor distances, as shown in
tions, we have to introduce at least the reg- Fig. 8-2. With this cluster, we must in-
ular tetrahedron and the octahedron as ba- troduce the variable x1 = ·s 1 Ò, x2 = ·s 1 s 3 Ò,
sic clusters. Fig. 8-7 also illustrates the x3 = ·s 1 s 2 Ò, x 4 = ·s 1 s 2 s 3 Ò, x5 = ·s 1 s 2 s 3 s 4 Ò.
This consistency relations are
ijkl
AAAA : ⎛ 1 4 4 2 4 1⎞ ⎛ 1 ⎞
BAAA : ⎜ 1 2 0 0 − 2 − 1⎟ ⎜ x1 ⎟
BABA : i j k l 1 ⎜ 1 0 0 − 2 0 1⎟ ⎜ x2 ⎟
r1 2 3 4 = 4 ⎜ ⎟ ⎜ ⎟ 70
BBAA : 2 ⎜ 1 0 − 4 2 0 1⎟ ⎜ x3 ⎟
BBBA : ⎜ 1 − 2 0 0 2 − 1⎟ ⎜ x 4 ⎟
BBBB : ⎝ 1 − 4 4 2 − 4 1⎠ ⎝ x5 ⎠
Taken as equalities, these equations define 8.4 Ground States

a simplex in the 5-dim. configurational
space. All the vertices correspond to exist- The description of the configurations
ing atomic configurations. Its projection will now be used to derive the stability of
into the subspace (x1 , x2 , x3 ) defines the the configurations at T = 0 K. The most
configuration polyhedron shown in Fig. stable state, the ground state, is given by
8-8. The coordinates of the vertices are the minimum of the (grand canonical)
given in Table 8-5. internal energy Ũ (V, T, mA , m B , …) for
given values T, V, m i obtained by variation
of the internal configurational variables,
i.e. point, pair, triplet, etc. correlation func-
tions. In practice it is preferred to consider
the (canonical) internal energy U (T, V, NA ,
NB , …) with the particle numbers Ni as
variables. The grand canonical energy is
then obtained by the Legendre transforma-
tion of U:
K
Ũ = U + ∑ mi N i
i =1
The general expression for Ũ can be

written
(8-16)
U = ∑ ∑ pn pm Vij + ∑ ∑ pn
˜ ( i ) ( j ) nm (i )
( nm ) i , j ( nmq ) i , j, k
Figure 8-8. Projection of the configuration polyhe- × pm( j ) pq( k ) Vijnmq

k +…+ ∑ ∑ pn( i ) mi
dron corresponding to the irregular tetrahedron n i
r = 1234 of the b.c.c. lattice (see Fig. 8-2) into the The summations have to be taken over all
subspace of point and first and second neighbor pair
correlation functions x1 = ·s 1Ò = xA – xB , x2 = ·s 1 s 3Ò,
equivalent points, pairs, etc. In this formu-
x3 = ·s 1 s 2 Ò, respectively. All vertices correspond to lation the higher-order cluster interactions
existing ordered atomic configurations. (beyond pairs) are not the total energies of
these clusters, but correction terms. De-
pending on the structure, the clusters of a
given class may share subclusters. This
Table 8-5. Coordinates of the vertices of the config-
uration polyhedron shown in Fig. 8-8.
sharing of subclusters has to be taken into
account if total energies are used.
x1 x2 x3 x4 x5 8.4.1 Pair Interactions
Pure A (A2) 1 1 1 1 1
The concept of pair interactions has a
A3B, AB3 (D03) ±1/2 0 0 ⫿1/2 –1
long tradition and has been supported by
AB (B2) 0 –1 1 0 1
the theoretical work of Gautier (1984),
AB (B32) 0 0 –1 0 1
Bieber and Gautier (1984 a, b, 1986, 1987),
Pure B (A2) –1 1 1 –1 1
and Turchi et al. (1991 a, b), who came to
the conclusion that pair interactions play a
8.4 Ground States 539
dominant role in alloys, while higher-order from sro by energetic arguments. In fact,
cluster interactions are of minor impor- lro results from a topological constraint.
tance. This appears to be a good approxi- Beyond a critical number of unlike bonds,
mation for the class of alloys with negli- it is no longer possible to arrange these
gible relaxation effects. Furthermore, pair bonds on a lattice with the constraint to
correlations can be determined by diffrac- maintain the equivalence of all lattice
tion experiments (see, e.g., Schönfeld, points.
1999). In Eq. (8-17), no reference state is speci-
Grouping together all the energy contri- fied. We will refer the energy to the me-
butions from positions n m with the same chanical mixture of the pure components
distance k, we obtain the following results in the same crystal structure as the alloys
for the internal energy: and obtain the following expression from
L Eq. (8-17):
∑ ∑ ⎧⎨ z ( k ) N (n ) ∑ 〈 pn( i ) pn( kj ) 〉 Vij( k )
1
U˜ =
⎩2 1
n =1 k i, j
DU˜ = U˜ − N ∑ z(k ) (8-18)
⎫ 2
+N ∑ 〈 pn( i ) 〉 m i ⎬
(n ) k
i ⎭ × ( xA VAA +
(k ) (k )
xB VBB )− N ∑ mi0 r0( i )
L i = A, B
1
= ∑ ∑ Vij( k ) ∑ zn( k ) N (n ) rnijn k N N
= ( DmA + DmB ) + ( DmA − DmB ) 〈s 0 〉
2 k i, j n =1
2 8
L
+ ∑ mi N N
∑ N (n ) rn( i ) − ∑ z ( k ) W ( k ) + ∑ z ( k ) W ( k ) 〈s 0 s k 〉
i n =1 8 k 8 k
where n k defines a point in the k-th neigh-
where D m i = m i – m 0i and W (k) = – 2 VAB (k)
+
bor shell of an arbitrary position in sublat- (k) (k) (k)
VAA + VBB . The W are called exchange-
tice n taken as the origin, and z(k)
n is the cor-
energy parameters. They take positive val-
responding coordination number. We ob-
ues for an ordering tendency in the k-th
tain
shell and negative values for a separation
L
tendency. In order/disorder problems, we
∑ zn( k ) N (n ) rnijn k = z ( k ) N r ij0 k are not usually interested in the size of the
n =1
system, and the quantity DŨ/N can then be
and treated. If we fix temperature and pressure
L and consider only equilibrium states, the
∑ N (n ) rni = N r (0i ) number of independent chemical potentials
n =1
is reduced by one. For convenience we often
where position 0 stands for any position in take D mA + D m B = 0 and define an effective
the crystal taken as the origin. Therefore, chemical potential m * = –12 (D mA – D m B ).
the energy is Finding the ground states is equivalent
(8-17)
1 to finding the minimum of Ẽ, which is de-
U˜ = N ∑ ∑ z ( k ) Vij( k ) r ij0 k + N ∑ mi r (0i ) fined as the part of the internal energy that
2 k ij i
depends only on the configurational vari-
This expression contains only the isotropic ables:
pair probabilities because we assumed k +1
DU˜ 1
isotropy for the pair interactions. Conse- E˜ = + ∑ z ( k ) W ( k ) = ∑ xs Fs (8-19)
quently, it is not possible to distinguish lro N 8 k s =1
with xs = ·s 0 Ò, ·s 0 s 1 Ò, …, ·s 0 s k Ò and that in this instance the planes of constant

1 1 Ẽ are parallel to each other, and it follows
Fs = m *, z (1) W (1) , …, z ( k ) W ( k ) .
8 8 immediately from Eq. (8-20) that the ex-
trema of Ẽ must correspond to extreme val-
8.4.1.1 Ground State Energies ues of x. Because the origin is inside the
existence domain for the present choice of
Eq. (8-19) shows how the energy Ẽ de-
t i = ±1, the value Ẽ min must be negative.
pends on the configurational variables xs ,
Searching the ground state thus becomes
that is, on the composition by x1 = ·s 0 Ò =
equivalent to finding, for a given alloy
xA – x B , and on the pair correlation func-
(i.e., for a given n = F /|F |), the values of x
tions by x2 , x3 , … . Among the variety of
that yield the most negative value for the
possible states, there are also those with the
distance d from the origin. If we recall that
lowest energy Ẽ min . These states will now
the existence domain is a closed polyhe-
be determined using a geometrical inter-
dron, it follows that the ground states must
pretation of Eq. (8-19).
correspond to vertices of the polyhedron.
Firstly, Eq. (8-19) is a linear relation. This
This will now be illustrated with some ex-
implies that all those values xs which fulfill
amples that have already been considered
Eq. (8-19) define a (hyper)plane in the con-
in previous sections.
figurational space. It will now be shown
that the energy Ẽ is proportional to the dis-
8.4.1.2 F.C.C. Structure,
tance of this (hyper)plane from the origin.
First-Neighbor Interactions
To demonstrate this, we introduce the
vector x = (x1 , x2 , …, xk –1) and a unit vec- With first-neighbor interactions in the
tor n = F /|F | = |F | –1 ( m *, z (1) W (1)/8, …, f.c.c. structure, the energy Ẽ depends only on
z (k) W (k)/8). Eq. (8-19) can then be writen the compositional parameter x1 = xA – x B ,
in vector form: and on one configurational parameter, the
˜ correlation function of nearest-neighbor
( x | n) = E = d (8-20) AB
pairs x2 = ·s 0 s 1 Ò = 1 – 4 r 01 (see Eq. (8-
|F |
7)), where position 1 is a nearest-neighbor
where d is the distance of the hyperplane
site to a central site 0. Therefore, the exis-
from the origin. This distance takes posi-
tence domain in Fig. 8-3 is relevant to this
tive or negative values depending on the
case. It follows from the foregoing discus-
directions of the two vectors n and x. The
sion that the ground states are given by the
factor |F | depends only on the chemical
vertices in Fig. 8-3 (i.e., pure A, pure B,
potential, the exchange energies and the
L12 with compositions A3B and AB3 , and
coordination numbers, that is, it defines the
L10 with composition AB). Fig. 8-9 illus-
alloy under consideration. If we change the
trates this point with some examples.
chemical potential (which controls the
If, for example, F 1 = m */|F | = 0, then the
composition) and/or the energy parame-
vector F = (0, F 2 ) and, consequently, n is
ters, then the direction of n changes (ex-
parallel or antiparallel to the x2-axis, de-
cept if we change all the components Fi by 12
pending on the sign of F 2 = –– W (1). The
a common factor), as does the value of d. 8
lines Ẽ = const. are then parallel to the x1-
If, on the other hand, we consider a
axis. This can also be seen from Eq. (8-20),
given alloy, which means we keep vector F
which reduces to
fixed but vary the values of x, then the nor-
mal vector n does not change. This means F 2 x2 = Ẽ
Figure 8-9. Lines of constant grand canonical energy E˜ (Eq. (8-20)) for given sets of values F1 = m * =
–12 ( m A – m B ) and F2 = –34 W (1) in the case of the f.c.c. lattice and nearest-neighbor interactions. The configuration
polygon is that of Fig. 8-3. Extrema of the energy correspond to lines passing through the vertices of the poly-
gon. Depending on the sign of F1 and F2 the line through a vertex corresponds to a minimum or a maximum.
The vertices thus define the ground states. For particular F1 , F2 values the lines coincide with the sides of the
polygon. For these values, the ground state is degenerate.
These lines are shown in Fig. 8-9. If we ideal solution. The lines Ẽ = const. are par-
consider an alloy with ordering tendency allel to the x2-axis in this case, because n is
(W (1) > 0 ¤ F 2 > 0), the value Ẽ min is parallel to the x1-axis. Eq. (8-20) now re-
reached for the most negative x2 value, duces to
which is x2 = – 1/3. This point corresponds
F1 x 1 = Ẽ
to the L10 structure AB. If we consider an
alloy with separation tendency (W (1) < 0 For a positive value of F1 = m * =
¤ F 2 < 0), the value Ẽ min is reached for the –12 (D mA – D m B ), the most negative value
most positive x2 value, which is x2 = 1. In of Ẽ is obtained for x1 = (xA – x B ) = – 1, i.e.,
this case, the line Ẽ = Ẽ min coincides with a corresponding to pure B, and a negative
side of the configuration polygon. This line value of F1 corresponds to pure A.
joins the vertices corresponding to the pure The procedure outlined above can be re-
components A and B. The ground state is peated for any direction of the vector F ,
degenerate: it could be pure A and pure B see Fig. 8-9. As long as the lines of con-
in any proportion juxtaposed to each other, stant Ẽ pass through a vertex of the poly-
or any other arrangement of A and B gon, a unique ground state can be defined.
which, in the limit of N Æ ∞, does not If these lines coincide with the sides of the
form A– B bonds (e.g., constant number of polygon, the ground state is degenerate. It
2-dim. boundaries). could then be any mixture of the two states
Next, we may consider the case corresponding to the two vertices defining
12
F 2 = ––8
W (1) = 0, which corresponds to an the side of the polygon. It was mentioned
before that the proportion of the two states When higher-order interactions are in-
differing in composition is not defined volved, as in the next paragraph, the deter-
because in these cases of degeneracy the mination of the ground-state diagram is not
overall composition of the system is not that simple, and it is useful to have a more
defined by a fixed value of the chemical analytical method for its determination.
potential. We can say, however, that the Such a method will now be introduced and,
state is two-phase. for the purpose of illustration, will be ap-
Finally, we can determine a so-called plied to the present very simple case.
ground-state diagram, which is shown in The ground-state diagram displays the
Fig. 8-10. It is worth noting that the single- ranges of F1 and F 2 values for which a
phase fields are sectors in the (F1, F 2 ) given state, e.g., L10 , is stable. We thus
plane limited by lines originating from the interpret Eq. (8-19) inversely to that of Eq.
origin. It is thus sufficient to know the di- (8-20): given a vertex in the configura-
rections of those lines. Furthermore, the ditional space, e.g., vertex x v = (– 1 3, 0) cor-
agram is symmetric with respect to F1 = 0. responding to L10 , we define a unit vector
n* = x v /| x v |. Eq. (8-19) can then be rewrit-
ten as
(F |n*) = Ẽ /| x v | = d* (8-21)
In a method similar to that used with Eq.
(8-20), we interpret Eq. (8-21) as the equa-
tion of a line in the space (F1, F 2 ) with a
distance d* = Ẽ /| x v | from the origin. If we
fix the value of Ẽ, then all vectors F ful-
filling Eq. (8-21) must end in the line with
the normal n* and with distance d*. In Fig.
8-10, the result is shown for Ẽ = – 1 (energy
units). It shows that the vectors fulfilling
Eq. (8-21) describe a polygon. The coordi-
nates of the vertices of this polygon are
given in Table 8-6. This polygon is the dual
polygon to the configuration polygon
(Finel, 1987), and the space (F1, F 2 ) is
called the dual space. A face of the dual
Figure 8-10. Ground-state diagram and dual poly- polygon corresponds to each vertex of the
gon (to Fig. 8-9) for the case of an f.c.c. lattice with configuration polygon in the direct space
nearest-neighbor interactions. The axes are
(x1 , x2 ). All vectors F pointing to the same
F1 = m * = –12 ( m A – m B ) and F2 = –34 W (1). The solid lines
are radius vectors which subdivide the plane into re- face of the polygon define alloys that ex-
gions with a defined ground state. The directions of hibit the same ground state given by the as-
these vectors are defined by the vertices of the dual sociated vertex of the configuration poly-
polygon to Fig. 8-3. It corresponds to a fixed value of gon. If we vary Ẽ, then the polygon shrinks
the energy, here E˜ = – 1 (energy units), and is ob-
or blows up, respectively, and the vertices
tained from Eq. (8-21). The dual polygon already
contains complete information on the ground-state of the dual polygon vary along the lines
diagram. For F1 , F2 values corresponding to these that define the two-phase states in Fig. 8-
lines, the ground state is degenerate. 10. It is easy to verify that these lines are
12
vertical to the faces of the configuration Let us first consider F 3 = –– 8
W (2) = 0.
polygon in Fig. 8-3. The normal vector n corresponding to the
The generalization of these results to planes of constant energy Ẽ is perpendicu-
higher dimensions is straightforward. A dual lar to the axis x3 in this case. If we consider
polyhedron defined by Eq. (8-21) is asso- an alloy with an ordering tendency in the
12
ciated with the (direct) configuration poly- first shell, F 2 = –– 8
W (1) > 0, then n is
hedron in such a way that each vertex of the oriented in the half space x2 > 0, and the
direct polyhedron corresponds to a face of the planes corresponding to Ẽ min are those
dual polyhedron, and vice versa. In order to passing through the vertices pure A, L12
get the vertices of the dual polyhedron, we with compositions A3B or AB3 , L10 with
have to solve Eq. (8-21) simultaneously for composition AB, and pure B, depending on
the number of vertices of the direct polyhe- the value of x1 = m * = –12 (D mA – D m B ). This
dron required to define the (hyper)plane. is in accordance with the results already
The application of this in three dimensions obtained in Sec. 8.4.3 (see Fig. 8-10).
will be shown in the next two paragraphs. However, an additional result can be de-
duced from Fig. 8-7: the energy planes pass
simultaneously through the vertices corre-
8.4.1.3 F.C.C. Structure, First and
sponding to the structures D022 and L12
(both either A3B or AB3 ), as well as L10
With first and second neighbor interac- and A2B2 . The ground states are thus de-
tions in the f.c.c. structure, the energy Ẽ generate in these instances.
depends on one more configurational pa- The complete ground-state diagram is
rameter, the second neighbor pair corre- obtained from the determination of the dual
AB
lation function x3 = ·s 0 s 2 Ò = 1 – 4 r 02 (see polyhedron associated with the existence
Eq. (8-7)). The relevant existence domain domain of Fig. 8-7. This dual polyhedron is
is now that shown in Fig. 8-7. We can again shown in Fig. 8-11. All F vectors pointing
proceed as previously. to (or piercing) a given face of the dual
Figure 8-11. Dual polyhedron (to Fig. 8-7) for the case of an f.c.c. lattice with first and second neighbor inter-
actions. The axes are F1 = m * = –12 ( m A – m B ), F2 = –34 W (1), and F3 = –34 W (2). The dual polyhedron defines the
ground-state diagram and corresponds to a constant energy E˜ = – 1 (energy units). All radius vectors pointing to
the same face correspond to the same ground state, which is labeled for each face, except the top, which is L1 1 .
Vectors coinciding with the boundaries of the faces correspond to degenerate ground states.
polyhedron define alloys with the same coordinates of the dual vertices are given in
ground state, which is defined by the corre- Table 8-6. They have been derived by
sponding vertex of the configuration poly- inserting the vertices in Table 8-3 into Eq.
hedron. The ground-state diagram thus di- (8-21) for Ẽ min = – 1.
vides the dual space into (hyper)cones
formed by the vectors from the origin to the
8.4.1.4 B.C.C. Structure, First and
borderlines of the faces. Each cone defines
a ground state. Due to graphical limita-
tions, those cones are not shown in Fig. In the first and second neighbor interac-
8-11, but the ground states are labeled tions of the b.c.c. structure, the energy Ẽ
at each face of the dual polyhedron. The depends on the same variables (x1 , x2 , x3 )
as in the preceding example, and the rele-
vant existence domain is that shown in Fig.
Table 8-6. Coordinates of the dual polyhedra shown 8-8. We can now construct the dual polyhe-
in Figs. 8-10, 8-11 and 8-12. dron, solving Eq. (8-21) for Ẽ = – 1 and us-
Tetra- Tetrahedron – Tetrahedron
ing the vertex coodinates in Table 8-4. The
hedron Octahedron b.c.c. coordinates of the dual vertices are given in
f.c.c. f.c.c. Table 8-6, and the dual polyhedron is
shown in Fig. 8-12.
F1 F2 F1 F2 F3 F1 F2 F3
2 1 ±2 1 0 0 0 –1 8.4.1.5 Energy Minimum at Constant

2 3 ±2 3 0 ±2 1 0 Composition
–2 3 ±18/7 12/7 3/7 ±2 2 1
–2 1 ±2 0 1 ±2 0 1 From a metallurgical point of view it
0 –1 ±2 4 1 0 –2 1 seems more natural to consider the com-
0 –2 1
0 0 –1
position of a system an independent vari-
able which we can control, rather than the
Figure 8-12. Dual polyhedron (to Fig. 8-8) for the case of an b.c.c. lattice with first and second neighbor inter-
actions. The axes are F1 = m * = –12 ( m A – m B ), F2 = W (1), and F3 = –34 W (2). The dual polyhedron defines the
ground-state diagram and corresponds to a constant energy E˜ = – 1 (energy units). All radius vectors pointing to
the same face of the polyhedron correspond to the same ground state, which is labeled for each face, except the
top one, which is B32.
chemical potential. However, this is not The states corresponding to L and D

possible. We can, of course, control the (L12 and D022 ) are ground states because
overall composition of our system by tak- they are vertices of the configuration poly-
ing the appropriate amounts of the pure hedron. S is not a vertex in this figure, but
substances, closing the system against ma- lies on the line joining the states pure A
terials exchange, and producing a certain and pure B. This line corresponds to con-
configuration (e.g., by heat treatment). If figurations that do not contain any A– B
we thereby equilibrate the system at a pairs. There is no other configuration pos-
given temperature, the system will attain sible except for pure A and pure B taken to-
its energy minimum by internal reactions. gether in the proportion A3B. Of course,
These reactions can be the formation of or- the planar interface, which is necessarily
dered atomic arrangements (homogeneous formed, contains A– B pairs. In the thermo-
single-phase state) or the formation of het- dynamic limit of N Æ ∞, however, the rela-
erogeneous microstructures. In the latter tive number of these pairs approaches zero
case, we have to introduce new variables in and can be neglected.
addition to the configurational variables, The point K lies on the line joining the
for example, the volume of produced vertices pure A and L11 (AB). Here again
phases, describing the advance of these re- the two limiting phases can be taken to-
actions. At equilibrium, the driving force gether in the right proportion to build up a
for the reactions must be zero, that is, the two-phase configuration at point K. The
chemical potentials of all the reaction prod- possibility of another homogeneous ar-
ucts must be the same. Because the overall rangement needs to be verified, though. But
composition is fixed, the amount of the K is not a vertex of the configuration poly-
phases present is also fixed. This is differ- hedron in Fig. 8-7. Hence, an additional
ent from a grand canonical treatment. The configuration cannot be characterized
energy in this case is given by the canoni- within the configurational space created
cal energy E, which is given by the expres- by the basic clusters of the octahedron and
sion tetrahedron.
k +1 The point M is on the line joining the
E = E˜ − x 1F1 = ∑ xs Fs (8-22) vertices A2B and A5B. Hence, at first, a
s=2
mixture of the two limiting phases can be
This equation is similar to Eq. (8-19), ex- attributed to this point. But a homogeneous
cept that the number of variables is reduced configuration also lies here, the structure
by one. The geometrical considerations can classified Amm2 in the appendix, which
be applied as before. In Fig. 8-7, two sec- has been observed in Monte Carlo simu-
tions corresponding to the compositions lations performed at fixed composition
x1 = xA – x B = 0 and 0.5 (AB and A3B) are (Gahn, 1982). This structure is built up
shown as shaded areas. Since in this in- with alternating cube planes which are oc-
stance we are confined to a planar section, cupied either by pure A or by an equal
the loci of constant energy are lines, and amount of A and B in an arrangement iden-
the minimum energy corresponds to the tical to the cube planes in the L11 structure,
lines passing through the vertices of the see Fig. 8-13. It is interesting to notice the
polygon. We will consider the case x1 = 0.5 structural relationship between this Amm2
in detail, and we thus have to consider the structure at point M and the arrangement at
polygon SLDMK in Fig. 8-7. K (pure A plus L11 (AB)): at M the stack-
Table 8-7. Numerical values of the correlation func-

tions for the b.c.c. superstructures in their most per-
fect lro state.
B2 D03 D03 B32

(xB Ù 0.5) (xB Ù0.25) (0.25 ÙxB Ù0.5) (xB Ù 0.5)
·s 1Ò 1 1 1 1
·s 2 Ò 1 1 1 1– 4 xB
·s 3 Ò 1– 4 xB 1 3 – 8 xB 1
·s 4 Ò 1– 4 xB 1– 8 xB –1 1– 4 xB
Figure 8-13. Atomic distribution on four subse- ·s 1 s 3 Ò 1– 4 xB 1 3 – 8 xB 1
quent (0 0 1) planes of the superstructures L11 (AB) ·s 1 s 4 Ò 1– 4 xB 1– 8 xB –1 1– 4 xB
and Amm2 (A3B). The structure Amm2 is composed ·s 2 s 3 Ò 1– 4 xB 1 3 – 8 xB 1– 4 xB
of alternating (0 0 1) planes of pure A and L11 . ·s 2 s 4 Ò 1– 4 xB 1– 8 xB –1 (1– 4 xB )2
·s 1 s 2 Ò 1 1 1 1– 4 xB
·s 3 s 4 Ò (1– 4 xB)2 1– 8 xB – (3 – 8 xB) 1– 4 xB
ing sequence is one cube plane of pure A

followed by one cube plane of L11 . If we
increase the stacking distance (i.e., the Eq. (8-5). These values are given in Table
number of cube planes of each phase), the 8-7 for the b.c.c. structures. Introducing
corresponding configurational point moves these values into the expression for the ca-
from M toward K. At K, the thickness of nonical energy
each package of cube planes approaches
infinity. The displacement of the configu- DU = DU˜ − N x 1F1
rational point is caused by the decreasing N
= − [8 W (1) + 6 W ( 2 ) ] (8-23)
number of AB pairs that cross the “inter- 8
faces” between the two structural units. N
+ W (1) [ 〈s 1 s 3 〉 + 〈s 1 s 4 〉
4
+ 〈s 2 s 3 〉 + 〈s 2 s 4 〉 ]
8.4.1.6 Canonical Energy of lro States 3 N (2)
+ W [ 〈s 1 s 2 〉 + 〈s 3 s 4 〉
Instead of imposing a given composi- 8
tion, we may also impose the state of order yields the energies of formation of the var-
(e.g., the most perfect lro) and calculate the ious lor phases. In Fig. 8-14 a, b, the ener-
canonical energy for these states as a func- gies of formation of b.c.c. structures are
tion of composition. These are most stable shown for two particular choices of inter-
states under the constraint of homogeneity. change energies. It can be seen in Fig.
Consider the f.c.c. structure with nearest- 8-14 a that the canonical energy of the most
neighbor interactions and the b.c.c. struc- stable lro phase for an interchange energy
ture with first and second neighbor interac- ratio W (2)/W (1) = 0.5, namely off-stoichio-
tions as simple examples. In Table 8-1, the metric D03 , is a polygon. This means that
point probabilities for most perfect lro (at all the homogeneous ordered states that can
off-stoichiometric composition) are given. be formed at every composition are degen-
From these probabilities, the numerical erate with the two-phase mixtures A + D03
values of the correlation functions for the and D03 + B2. This is different from Fig.
different superstructures in their most per- 8-14 b where W (2)/W (1) = – 1. The canoni-
fect lro state can be derived directly using cal energy of the most stable lro phase, off-
Figure 8-14. Canonical energy DU of different

b.c.c. and f.c.c. long-range ordered structures as a
function of composition for fixed values of the inter-
change energies. (a) b.c.c., ordering tendency be-
tween first and second neighbors: W (1) = 2 W (2) > 0.
(b) b.c.c., ordering tendency between first neighbors
and separation tendency between second neighbors:
W (1) = – W (2) Û 0. (c) f.c.c. ordering tendency be-
tween nearest neighbors: W (1) > 0, W (2) = 0.
stoichiometric B2, is a concave curve. In

this instance, the most stable state is a two-
phase mixture of A + B2. From the canoni-
cal energy, we are thus given a hint of
whether the ground state will be two-phase,
or if there is a degeneracy with single-
phase states. The canonical energy for the
f.c.c. lro structures L12 and L10 can be de-
rived in a similar way. The result is shown
in Fig. 8-14 c for the ratio W (2)/W (1) = – 1.
Here again, the canonical energy of the
most stable lro phase is concave, and only
the two-phase mixtures A + L12 and
L12 + L10 are ground states.
8.4.1.7 Relevant Literature

The ground state analysis of f.c.c. and
b.c.c. alloys with first and second neigh-
bors was done by Kanamori (1966), Rich-
ards and Cahn (1971), Allen and Cahn
(1972, 1973), and de Ridder et al. (1980).
Multiplet interactions were included by
Cahn and Kikuchi (1979), Sanchez and de
Fontaine (1981). The ground states of the
f.c.c. lattice with up to fourth neighbor
interactions were determined by Kanamori
and Kakehashi (1977), fifth neighbor inter-
actions were treated by Kanamori (1979),
Finel and Ducastelle (1984) and Finel
(1984, 1987). The ground states of the hex-
agonal lattice with anisotropic interactions
have been deduced by Crusius and Inden
(1988) and Bichara et al. (1992 a) from the
known ground states of the planar hexago-
nal lattice with up to third neighbor interac-
tions (Kudo and Katsura, 1976; Kanamori, It may be useful to spell out explicitly
1984). the pure configurational part of the Vn . In
the case of pair interactions the Vn with
8.4.2 Effective Cluster Interactions n ≥ 2 are equivalent to the pair exchange
(ECIs) z(n) (n)
energies, i.e., Vn = W . In this case
Up to this point only pair interactions 8
the specific lattice type is only introduced
have been considered with the exchange
in terms of the coordination numbers.
energies W (k) as dummy parameters. The
In the case of higher-order cluster inter-
internal energy DŨ, with reference to the
actions the lattice type is not only specified
pure components in the same crystal struc-
by the coordination numbers of the basic
ture, has been written in terms of an expan-
cluster and its subclusters, but most impor-
sion of these W (k) (Eq. (8-19)). This is a
tantly by the type of clusters they share
special case of the more general cluster ex-
with each other. Subclusters which are not
pansion (CE) of the internal energy DŨ in
shared define no new correlations. They are
terms of the correlation functions:
already taken into account by those of their
DŨ a parent cluster. We could thus expect that the
= ∑ Vn xn (8-24) Vn of such clusters do not show up in the
N 0
expansion, for example triplet correlations
where n refers to subclusters of the basic in the tetrahedron expansion of the f.c.c.
cluster a (n = 1 for point, n = 2 for pair lattice. This, however, is not true because
clusters, etc.) (Sanchez et al., 1984). This the Vk in the expansion, Eq. (8-24), not only
equation is exact if a represents the entire contain terms from the k-point clusters, but
system. It is expected, however, that Eq. also terms from higher clusters (n > k).
(8-24) converges rapidly such that the se- This becomes transparent if the Vn are
ries can be truncated. For a general discus- expressed in terms of the cluster exchange
sion of the method see Laks et al. (1992), energies, defined as
de Fontaine (1994) and Sanchez (1996).
l
The parameters Vn are supposed to em- Wij12……l r = − r Vij …l +
brace all possible energy sources that depend
∑ Vk k …k
k =i
on configuration, including all ground- 1234
(for example, WABBB = – 4VABBB + VAAAA +
state features like electronic energy, long-
3VBBBB). It should be noted that, unlike the
range elastic and Coulombic interactions.
pairs, these cluster-exchange energies do
But they might also contain configuration-
not represent the total energy of the clus-
dependent excitation energies such as those
ters, only successive corrections to the
arising from vibrational and electronic ex-
cluster energies. If total energies were con-
citations when finite temperatures are con-
sidered, the sharing of subclusters would
sidered, making the model interaction en-
have to be taken into account.
ergies temperature dependent. It should be
Considering the f.c.c. lattice and a trun-
appreciated that the parameters Vn repre-
cation at the regular tetrahedron cluster
sent energy contributions, not only from di-
(a = 4), Eq. (8-24) gives:
rect interactions within the cluster, but also
from the interactions of much longer range DU˜ (j )
outside the cluster range. They are thus = V0 + V1 ( xA − xB )
N
called effective cluster interactions (ECIs). + V2 x2(j ) + V3 x3(j ) + V4 x 4(j ) (8-25)
where j indicates specific states, e.g. the z triplet

five vertices a – e of the configuration poly- − ( 2 WABA + WAAB
72
hedron in Table 8-3. The Vn expressed in + WBBA + 2 WBAB )
terms of the cluster exchange energies are z tetra
given by − ( 4 WBAAA + 4 WBABA
256
+ 2 WBBAA + 4 WBBBA )
1 z (1) (1)
V0 = ( Dm A + Dm B ) − W
2 8
1 z triplet
z triplet V1 = ( Dm A − Dm B ) −
− (WBAA + WBBA ) 2 72
24
⋅ ( 2 WABA + WAAB − WBBA − 2 WABB )
z tetra ⎛
WBAAA + WBBAA + WBBBA ⎞
3
− z tetra
64 ⎝ 2 ⎠ − (8 WBAAA − 8 WBBBA )
256
1 z triplet
V1 = ( Dm A − Dm B ) − (WBAA − WBBA )
2 24 z (1) (1) z triplet
z tetra V2 = W + ( 2 WAAB + 2 WBBA )
− ( 2 WBAAA − 2 WBBBA ) 8 AB 72
64 z tetra
+ (8 WBBAA )
1 (1) (1) z triplet 256
V2 = z W + (WBAA + WBBA )
8 24
z tetra z ( 2 ) ( 2 ) z triplet
+ (3 WBBAA ) V3 = WAB +
64 8 72
⋅ ( 2 WABA − WAAB − WBBA + 2 WBAB )
z triplet
V3 = (WBAA − WBBA ) z tetra
24 + (8 WBABA − 4 WBBAA )
z tetra 256
+ ( 2 WBAAA − 2 WBBBA ) (8-26)
64
z triplet
z tetra ⎛ V4 =
WBAAA − WBBAA + WBBBA ⎞
3 72
V4 =
64 ⎝ 2 ⎠
⋅ ( 2 WABA + WAAB − WBBA − 2 WAAB )
Taking for example all the triplet terms z tetra
Wijk = 0 does not mean V3 = 0, because + (8 WBAAA − 8 WBBBA )
256
there remain tetrahedron terms.
In the irregular tetrahedron approxima-
z tetra
tion of the b.c.c. lattice, the cluster expan- V5 = ( 4 WBAAA − 4 WBABA
sion of the internal energy is written as: 256
− 2 WBBAA + 4 WBBBA ) (8-28)
DU˜ (j ) These relations show that cluster interac-
= V0 + V1 ( xA − xB ) + V2 x2(j )
N tions not only influence the correlations of
+ V3 x3(j ) + V4 x 4(j ) + V5 x5(j ) (8-27) their own cluster but also those of the sub-
clusters. The system of equations, Eq. (8-
where the CE parameters Vn correspond to:
24), can be solved for the different Vn . The
1 idea is that the ECIs derived from the ener-
V0 = ( Dm A + Dm B )
2 gies of ordered configurations are appli-
z (1) (1) z ( 2 ) ( 2 ) cable to any configuration, i.e. for lro, sro
− W − WAB
8 AB 8 and for the disordered state. This is equiva-
lent to saying, in the pair approximation, molar volumes are not constant, and “local
that the W (k) can be used for all configura- relaxation” when atoms deviate from their
tions. Since the only source of error in Eq. ideal lattice positions. Ferreira et al. (1987,
(8-24) is the truncation, it is imperative to 1988) and Wei et al. (1990) have shown
carefully select the truncation. that, if the volumes depend on composi-
The ingredients for the determination of tion but only weakly upon configuration,
the ECIs are values of DŨ(j ), which may the energy of formation can be split into
be obtained from various sources (see two additive terms, the elastic energy re-
e.g. Zunger, 1994). The simplest approach quired to hydrostatically deform the pure
is to take experimental data of the corre- constituents from their equilibrium vol-
sponding quantities (e.g. Inden (1975 a, b, umes to the alloy volume, and the pure
1977 a, b; Oates et al., 1996)). This requires configurational or “chemical” term. The
a truncation at a level imposed by the avail- ECIs can be determined either from a set
ability of data and in most cases does not of as many ordered compounds as there are
provide satisfactory results. unknowns, or from a larger set, in which
The second approach is to derive DŨ(j ) case the ECIs are obtained by a fitting pro-
from first principles calculations, treating cedure (Lu et al., 1991, 1995). It must be
the random alloy by the Coherent Potential emphasized that the ECIs are “effective”
Approximation (CPA) and the electronic and their values depend on the level of
band structure with the Tight-Binding truncation.
method (Gautier et al., 1975; Ducastelle Cluster expansions for systems with
and Gautier, 1976; Treglia and Ducastelle, strong lattice relaxations converge slowly
1980; de Fontaine, 1984; Sigli and San- in real space. This has led to an alternative
chez, 1988; Sluiter and Turchi, 1989 a, b) route, treating the CE in the reciprocal
or with the Korringa, Kohn and Rostoker space (e.g. Laks et al., 1992).
method (Gyorffy and Stocks, 1983; John- Finally, it should be mentioned that pair
son et al., 1990). The approaches include interactions may also be obtained from the
the Generalized Perturbation Method and diffuse scattering of X-rays or neutrons
the Concentration Wave Method. The or- (for a recent review, see Schönfeld, 1999).
dering contributions are treated as pertur- The procedure is the inverse of the CVM
bations of the disordered states. The pertur- (see Sec. 8.5.1) or the MC technique (see
bation method is limited to alloys with sim- Sec. 8.5.3). Instead of calculating equilib-
ilar atomic species and provides essentially rium configurations using given energy pa-
pair interactions. rameters, experimental data of sro are taken
The third approach is the inversion as input in order to obtain the interaction
method of Connolly and Williams (1983) parameters. The methods have been pre-
(Ferreira et al., 1988, 1989; Wei et al., sented by Priem et al. (1989 a, b) for CVM,
1990, 1992; Terakura et al., 1987) . In this and Gerold and Kern (1987) and Livet
approach special atomic configurations (1987) for MC.
like in ordered compounds are selected and
their total energy is obtained from direct
electronic structure calculations.
It has been pointed out by Zunger (1994)
that it is necessary to account for two types
of relaxation, “volume relaxation” when
8.5 Phase Equilibria at Finite Temperatures 551
8.5 Phase Equilibria 8.5.1 Cluster Variation Method

at Finite Temperatures In the CVM, the entropy is evaluated ac-
cording to S = kB ln W, where W is the
Two ways of determining the equilibria number of arrangements that can be formed
at finite temperatures have proven to be for given values of the correlation func-
most useful (see also the chapter by Binder, tions. The CVM develops an approximate
2001), CVM introduced by Kikuchi (1950, expression for W, taking into account only
1951) and the MC technique (Binder, 1979, the correlations up to the basic cluster size.
1986; Binder and Stauffer, 1984; Mourit- Therefore, the basic clusters are considered
sen, 1984; Binder and Heermann, 1988). to be uncorrelated independent species.
The CVM is based on an analytical cal- The first step in determining the entropy is
culation of the configurational entropy S. to write W as the number of possible ar-
The equilibrium states at constant volume rangements of this basic cluster:
and temperature are obtained by a mini-
mization of the Helmholtz energy Na !
W= = {a}Na (8-29)
F (V, T, Ni ) = U – T S ∏ [ Na (s )]! def
s
for the equilibrium at fixed composition, or where Na is the total number of a -clusters
by minimization of the grand potential in a system with N points, and Na (s ) is the
number of a -clusters with a given configu-
W (V, T, m i ) = U – T S – Â m i Ni ration specified by s ; the term {a}Na is a
i
short notation which will be used hereafter
for the equilibrium with exchange of atoms, for abbreviation. The ratio Na (s )/Na will
i.e., for given values of the chemical poten- be indicated by ra(s ).
tials. Eq. (8-29) overestimates the number of
The MC method simulates the configu- arrangements: two overlapping a -clusters
rations in a computer crystal. At a given cannot be permuted independently because
temperature and fixed chemical potentials, they must fit together with their over-
atoms are exchanged with a reservoir of at- lapping units. This becomes obvious if
oms with a probability defined in such a we look at the high temperature limit of Eq.
way that the equilibrium state is reached (8-29), which is W∞ = {1}Na , while the
after a sufficient number of atomic replace- exact expression of the limit of W is given
ments. In the case of fixed composition, at- by the number of configurations of N
oms are selected pairwise and interchanged points: W∞ = {1}N .Therefore, a correction is
with each other according to an equivalent needed in order to obtain the correct high-
probability. This method yields the equilib- temperature limit. Correcting for the points
rium configuration but not the thermody- would correspond to the generalized quasi-
namic functions. On the other hand, we ob- chemical approximation (Yang, 1945; Yang
tain complete and fine-scale information and Li, 1947, 1949). Based on geometrical
about the atomic configurations and the considerations, Kikuchi (1950, 1951) intro-
correlations at large distances, the limita- duced a correction in such a way that the
tion being imposed only by the size of the high temperature limit is not only obtained
computer crystal. Both methods will be ap- for the point cluster but also for all subclus-
plied in this section. ters of a. A short time later Barker (1953)
arrived at the same result by a more mathe- The same reasoning now needs to be fol-
matical treatment, and the continued inter- lowed with the next subcluster a – 2. For
est in the derivation of the entropy formula this cluster, the number of arrangements of
is manifested in a number of papers, e.g., the basic cluster a, as well as the correction
Hijmanns and de Boer (1955), Burley (1972), term for a – 1, have already been counted.
Sanchez and de Fontaine (1978), Gratias et We can therefore write
al. (1982) and Finel (1987). For a dis- ma − 1 aa − 1
W = ({a}N ) ma aa ({a − 1}N )
cussion of the hierarchy of the cluster ap- m a
proximations, see Schlijper (1983, 1984, × ({a − 2}N ) a − 2 a − 2
1985). with aa – 2 determined from the equation
Following Barker, we define ma as the
number of a -clusters per point: Na = ma N. ma − 2 = ma naa − 2 aa + ma −1 naa −−12 aa −1
Then we can write (in Stirling’s approxi- + ma − 2 aa − 2
mation): Continuing this reasoning leads to
ma
( ma N )! ( N !) a
{a}Na = = W = ∏ ({n }N ) mn an (8-30)
∏ ra(s ) ma N )!
( ∏ ra(s ) N !)ma
((
n =1
s s
= ({a}N ) ma and to the following recursion scheme for
the exponent an , using the identities nnn = 1
We now want to correct for the overlap and aa = 1:
with the first subcluster a – 1, which has
one point less than a. Due to the indepen- (basic) cluster a : ma = ma naa aa
dent variation of the a -clusters, the over- a − 1: ma −1 = ma −1 naa −−11 aa −1
lapping cluster a – 1 has also been counted
a −1 + ma naa −1 aa
({a − 1}N ) ma na times, where naa −1 is the
number of times the a – 1 cluster is con- a − 2: ma − 2 = ma − 2 naa −−22 aa − 2
tained in a. The correct number of times + ma −1 naa −−12 aa −1 + ma naa −1 aa
the a – 1 cluster should be counted, pre-
a −n
suming it were an independent species, is
m
({a − 1}N ) a − 1. Therefore, we must correct
n : mn = ∑ ma − n nan − m aa − m with 1 Ù n Ù a
m =0
Eq. (8-29) as follows:
We thus obtain for the entropy in Stirling’s
ma − 1
({a − 1}N ) approximation
W = ({a}N ) ma a −1
({a − 1}N ) ma na a
S = − kB N ∑ mn an ∑ rn(s ) ln rn(s ) (8-31)
ma − 1 − ma naa − 1
= ({a}N ) ma ({a − 1}N ) n =1 s
The values of ma and naa ––mn are to be de-

Using the identity aa = 1, which is valid for rived by geometrical considerations of the
the basic cluster, and aa –1 determined from lattice. This procedure is straightforward
the equation even for large clusters. For the purpose of
illustration, the full set of these values and
ma −1 aa −1 = ma −1 − ma naa −1
the “CVM exponents” an are given in Table
W can be written as 8-8 both for the f.c.c. structure in the tetra-
hedron–octahedron approximation and for
ma − 1 aa − 1
W = ({a}N ) ma aa ({a − 1}N ) the b.c.c. structure in the irregular tetra-
Table 8-8. Numerical values of ma , naa ––nm , and the CVM exponents an , for the f.c.c. structure in the tetrahe-
dron – octahedron approximation and for the b.c.c. structure in the irregular tetrahedron approximation.
F.C.C. Tetrahedron – Octahedron Fig. 8-4
Cluster 23 … 7 23456 3456 2347 1234 237 123 27 12 1

a= 6 5 4a 4b 4c 3a 3b 2a 2b 1
ma = 1 6 3 12 2 12 8 3 6 1
n a6 = 1
n a5 = 6 1
n a4a = 3 1 1
n a4b = 12 4 0 1
n a4c = 0 0 0 0 1
n a3a = 12 6 4 2 0 1
n a3b = 8 4 0 2 4 0 1
n a2a = 3 2 2 1 0 1 0 1
n a2b = 12 8 4 5 6 2 3 0 1
n a1 = 6 5 4 4 4 3 3 2 2 1
aa = 1 0 0 0 1 0 –1 0 1 –1
B.C.C. Irregular Tetrahedron Fig. 8-2
Cluster 1234 123 12 13 1

a= 4 3 2a 2b 1
ma = 6 12 3 4 1
aa = 1 –1 1 1 –1
hedron approximation. In the latter case, b.c.c.

the naa ––mn have been omitted for the sake of {1234}6N {12}3N {13}4N
S = kB ln
brevity. {123}12
N {1}N
1
[
For the entropy, we thus obtain the fol-
lowing:
= − kB N 6 ∑ r1234
ij k l ij k l
ln r1234
{ij k l}
f.c.c. − 12 ∑ r123
ij k ij k
ln r123 (8-33)
{23 … 7}1N {1234}2N {12}6N {ij k}
S = kB ln
{123}8N {1}1N + 3 ∑ r12
ij ij
ln r12
{ij}
[
ij … k ij … k
= − kB N ∑
∑ r13i k ln r13i k − ∑ r1i ln r1i ]
r23 … 7 ln r 23… 7
{ij … k} +4
{i k} {i}
+2 ∑ ij k l
r1234 ij k l
ln r1234 (8-32)
{ij k l} The entropy equations (8-32) and (8-33)
−8 ∑ ij k
r123 ij k
ln r123 are valid for the sro state. In the lro state,
{ij k} the equivalence of certain clusters is
+ 6 r12
{ij}
∑ ij
ln r12ij
∑− r1i
{i}
ln r1i ] broken. For example, Eq. (8-33) has to be
written in the lro state with four sublattices
(Fig. 8-2):
{1234}6N {12}3N/ 2 {34}3N/ 2 {13}1N {14}1N {23}1N {24}1N

S = kB ln (8-34)
{123}3N {124}3N {134}3N {234}3N {1}1N/ 4 {2}1N/ 4 {3}1N/ 4 {4}1N/ 4
8.5.2 Calculation of Phase Diagrams terwards by the superposition of all the

with the CVM two-phase equilibria. More sophisticated
techniques have not yet been applied be-
Once the grand canonical energy Ũ, Eq.
cause phase diagrams of multicomponent
(8-18), and the entropy, Eq. (8-31) have
systems beyond ternary have not yet been
been derived, the equilibrium at any tem-
analyzed. A second-order transition is
perature and the chemical potentials can be
more difficult to define. In this instance,
determined by minimizing the grand poten-
there is no metastable extension of the ex-
tial W = Ũ – T S with respect to the configu-
istence range of the most-ordered phase be-
rational variables. Two different tech-
yond the transition point, and the grand po-
niques of minimization can be found in the
tentials meet at this point with a continuous
literature: the natural iteration method,
slope. In such a case, the second Hessian of
introduced by Kikuchi (1974), in which
the grand potential has to be analyzed
minimization is achieved with respect to
(Kikuchi, 1987). Often this is circum-
the occupation probabilities of the largest
vented by showing that there is no meta-
cluster; and minimization by means of
stable extension of the most-ordered phase
the Newton – Raphson and steepest-descent
within given accuracy limits.
techniques with respect to the correlation
functions as independent variables (e.g.
8.5.3 Phase Diagram Calculation
Sanchez and de Fontaine, 1978; Mohri et
with the Monte Carlo Method
al., 1985; Finel, 1987). The natural itera-
tion method converges even with starting In the MC method the configuration of a
values that are far from the equilibrium crystal with a limited number of points
values, but it is slow. The second method is (typically several 104 to 105) is stored in a
more sensitive to the quality of the starting computer. In order to minimize the boun-
values, but it is much faster. dary effects of the finite size of this crystal,
A phase diagram is determined by calcu- periodic boundary conditions are usually
lating the equilibrium state for two phases applied. These boundary conditions must
and comparing the values of the corre- be consistent with the superstructures we
sponding grand potentials. In this way, the want to treat so that an integer number of
most stable phase is obtained. The transi- unit cells fits into the crystal. This deter-
tion between different lro states of from lro mines both the shape we have to give to the
to sro can be of first order or higher (usu- computer crystal as well as the number of
ally second) order. A first-order transition points. In the b.c.c. structure with first and
is detected by the intersection of the grand second neighbors, the superstructures are
potentials of two phases for a certain value cubic, and there is no difficulty. In the f.c.c.
of the chemical potentials (e.g. Kikuchi, structure, the superstructures have different
1977 b, 1987). This value yields a different symmetries and the unit cells have differ-
composition for each phase which defines ent extensions in different crystal direc-
the tie-line of the two-phase equilibria. tions, as shown in the appendix. The phase
Three-phase equilibria are recognized af- diagram calculation with the MC method is
usually carried out in the grand canonical ⎧i ⎫

scheme, i.e., for a given value of the chem- where W ⎨ ⎬ is the grand canonical energy
⎩n ⎭
ical potentials. This ensures a homogene- when an atom of kind i occupies site n. Be-
ous single-phase equilibrium state. For this cause the sum of the probabilities over all
single-phase state, the configuration can be components adds up to 1, the interval [0, 1]
analyzed, and a composition determined. can be subdivided into corresponding inter-
Even in this simple case, problems may vals. The atom i on site n is then replaced
appear due to defects such as antiphase by the atom j, which corresponds to the
boundaries (APBs), which are produced in interval selected by a random number be-
the equilibration process in the same man- tween 0 and 1. If the random number se-
ner as in real alloys (see, e.g., Gahn, 1986; lects the interval corresponding to i, no ex-
Ackermann et al., 1986; Crusius and Inden, change is made. In a Monte Carlo simula-
1988). tion, the above-mentioned steps have to be
The Monte Carlo process can be started performed many times, and it is necessary
from any configuration and composition, to use efficient programming techniques
even from the pure components. An arbi- for data storage and saving of computer
trary site n is selected at random and the time (for general aspects see Binder (1976,
atom i on this position is replaced by an 1979, 1985), Binder and Stauffer (1984);
atom j according to a transition probability for multispin coding see Jacobs and Rebbi
wniÆ j which has to fulfill certain criteria in (1981), Zorn et al. (1981) and Kalle and
order to guarantee the convergence toward Winkelmann (1982)).
the equilibrium state from any initial con-
figuration (Chesnut and Salsburg, 1963;
Fosdick, 1963; Binder, 1976). A sufficient 8.5.4 Examples of Prototype Diagrams
condition for this convergence is the fulfill- The following examples represent proto-
ment of the detailed balance equation and type diagrams. The interchange energies
of normalization of the transition probabil- W (k) used in the calculations are selected in
ity: such a way that the phase diagrams display
typical features. Calculations for real al-
r ni wni→ j = r nj wnj → i (8-35 a)
loys will be presented in the next section.
K We will start with the systems treated in the
∑ wni→ j = 1 (8-35 b)
previous sections.
j =1
Different expressions for the transition

8.5.4.1 F.C.C. Structure,
probability which fulfill this criterion have
First Neighbor Interactions
been suggested. The choice is made ac-
cording to the most efficient expression In this instance we take the case of an
in terms of computer time. The following ordering tendency in the first shell, i.e.,
expression fulfills the conditions of Eq. W (1) > 0, W (2) = 0. The phase diagram, as
(8-35): obtained by CVM in the tetrahedron – octa-
(8-36)
hedron approximation (Finel and Ducas-
⎡ 1 ⎛ ⎧ j⎫ ⎧ i ⎫⎞ ⎤
exp ⎢ − ⎜ W ⎨ ⎬ − W ⎨n ⎬⎟ ⎥ telle, 1986), together with the results from
i→ j ⎣ kB T ⎝ ⎩ n ⎭ ⎩ ⎭⎠ ⎦ the MC method (Ackermann et al., 1986;
wn = K
⎡ 1 ⎛ ⎧ j⎫ ⎧ i ⎫⎞ ⎤ Diep et al., 1986; Gahn, 1986), are shown
∑ exp ⎢ − ⎜ W ⎨n ⎬ − W ⎨n ⎬⎟ ⎥
⎝
j =1 ⎣ B k T ⎩ ⎭ ⎩ ⎭⎠ ⎦ in Fig. 8-15. The two diagrams do not coin-
Figure 8-15. F.c.c. structure: calculated prototype phase diagrams for the case of nearest-neighbor interactions,
W (1) > 0, W (2) = 0. (a) CVM calculation in the tetrahedron – octahedron approximation (Finel and Ducastelle,
1986), (b) Monte Carlo simulation (Ackermann et al., 1986).
cide quantitatively, but their main features The CVM diagram in the tetrahedron –
agree qualitatively. In particular, they agree octahedron approximation, Fig. 8-15, dif-
as to the existence of a triple point at finite fers only slightly from that previously cal-
temperature, i.e. at culated in the tetrahedron approximation
(van Baal, 1973; Kikuchi, 1974), which
CVM
t tri CVM
= kB Ttri /(W (1)/4) ≈ 1.5
gave a triple point at t tri CVM
≈ 1.6. The
(Finel and Ducastelle, 1986) and at higher-order cluster approximation does
MC
t tri ≈ 0.8 (Ackermann et al., 1986) and 0.9 not lead to a strong shift of the triple point
(Diep et al., 1986). This point gave rise to to lower temperatures, contrary to an ear-
some controversy (discussed by Kikuchi, lier result by Sanchez et al. (1982) who
1986), caused by an earlier MC study
CVM
found t tri ≈ 1.2. The difference results
(Binder, 1980; Binder et al., 1981) which only from certain approximations in the
indicated that the phase boundaries extra- numerical treatment made by Sanchez et
polate to 0 K, and that a triple point does al., which were avoided by Finel (1987). In
not exist. The existence of this triple point Finel’s work, a more sophisticated CVM
is now well confirmed and has been further calculation was also made using the tetra-
corroborated by the studies of Lebowitz et hedron – octahedron for the ordered phases
al. (1985) and Finel (1994). The diagram in and the quadrupole tetrahedron for the sro
Fig. 8-15 replaces that of the earlier MC by states. In that approximation. Finel ob-
Binder. The MCs were not performed at tained complete agreement with the MC
sufficiently low temperatures to detect the work.
P4/mmm phase obtained in the CVM
(called L¢ by Finel, 1984).
8.5.4.2 F.C.C. Structure, First and agree fairly closely. Phase diagrams for
Second Neighbor Interactions other values of the ratio W (2)/W (1) are
given by Mohri et al. (1985) and Binder
In this section, the case of an ordering et al. (1983).
tendency in both shells will be treated first, Fig. 8-17 shows the results for an order-
namely, W (1) > 0 and W (2)/W (1) = 0.25. The ing tendency between first neighbors and a
results of the tetrahedron – octahedron separation tendency between second neigh-
CVM (Sanchez and de Fontaine, 1980) and bors, W (1) > 0, W (2) = – W (1). This situation
of MC simulations (Bond and Ross, 1982) corresponds to the calculation presented
are shown in Fig. 8-16. The two diagrams previously, in Fig. 8-14 c. As expected
Figure 8-16. F.c.c. struc-

ture: calculated prototype
phase diagrams for the case
of first and second neighbor
ordering interactions,
W (1) = 4 W (2) > 0. (a) CVM
calculation in the tetrahe-
dron – octahedron approxi-
mation (Sanchez and de
Fontaine, 1980). (b) Monte
Carlo simulation (Bond and
Ross, 1982).
Figure 8-17. F.c.c. structure:

calculated prototype phase dia-
grams for the case of an order-
ing tendency between first
neighbors and a separation ten-
dency between second neigh-
bors, W (1) = – W (2) > 0. (a) CVM
calculation in the tetrahedron –
octahedron approximation
(Mohri et al., 1985). (b) Monte
Carlo simulation (Binder et al.,
1983).
from this diagram the low temperature tions on the shape of the miscibility gap is
states are two-phase states and the phase small. The shape obtained differs signifi-
boundaries extrapolate to the pure compo- cantly from the miscibility gap which is
nents and to the stoichiometric ordered usually calculated with the regular solution
phases. A complete series of prototype (i.e., point approx.) model.
diagrams calculated with the tetrahedron –
octahedron CVM is presented by Mohri et 8.5.4.3 B.C.C. Structure, First and
al. (1985). Second Neighbor Interactions
The case of a separation tendency be-
A series of prototype phase diagrams,
tween first and second neighbors has been
calculated with the tetrahedron CVM for
analyzed with the MC technique: Kutner
varying strengths of ordering tendency in
et al. (1982) analyzed the case W (1) < 0,
both neighbor shells, was first presented by
W (2) = 0 with the grand canonical simula-
Golosov and Tolstick (1974, 1975, 1976).
tion, while Gahn et al. (1984) analyzed the
Simultaneously, Kikuchi and van Baal
cases W (2) = 0 and W (2) ≠ 0 with a spe-
(1974) presented a diagram corresponding
cial canonical simulation. The results for
to the ratio W (2)/W (1) = 0.5, which is close
W (2) = 0 are identical in both treatments.
to the situation encountered in Fe – Si and
The resulting miscibility gaps are shown in
Fe – Al. Figs. 8-19 a, b and c display a series
Fig. 8-18 (on a reduced scale in order to
of diagrams calculated with the tetrahedron
show the variation in shape). It is found
CVM and with the MC method (Acker-
that the effect of second neighbor interac-
mann et al., 1989) for different strengths
and signs of the interchange energies. The
results of both methods are in good agree-
ment. In these diagrams, second-order tran-
sitions are indicated by a hachure. These
second-order transitions turn into first-or-
der transitions at tricritical points, and the
topology of the phase boundaries close to
these points exhibit the characteristics de-
rived by Allen and Cahn (1982).
8.5.4.4 Hexagonal Lattice, Anisotropic

Nearest-Neighbor Interactions
The ordering reactions in the hexagonal
crystal structure have been studied with
the MC method (Crusius and Inden, 1988;
Bichara et al., 1992 b). In order to simulate
the situation for c/a ≠ 1.633, i.e., for non-
close packing, the interchange energies
Figure 8-18. Miscibility gaps according to MC cal- between nearest neighbors within a basal
culations for various values of the interchange ener- plane, W (11), and between two such planes,
gies between first and second neighbors. The dia-
gram is symmetric with respect to the equiatomic W (12), were given different values. The
composition, W (1) < 0, W (2) = 0 (Kutner et al., 1982). resulting phase diagrams for two sets of
W (2) = 0 and W (2) ≠ 0 (Gahn et al., 1984). interchange energies are shown in Figs.
8-20 a and b. The phase A2B is a two-

dimensional structure built with three sub-
sequently ordered planes, but with no cor-
relation between them. Therefore a two-di-
mensional characterization has been given
in Sec. 8.7 in addition to the three-dimen-
sional one that is to be considered for cases
with W (12) ≠ 0 (see Appendix, Table 8-13).
In the case W (12) = 0, the phase A2B is
the only stable superstructure. The transi-
tion A2B ´ A3 is second order in this in-
stance. For m * = 0 (xB = 0.5), no phase tran-
sition has been observed down to the re-
duced temperature t = kB T /(W (11)/4) = 0.6.
The extrapolation of the transition temper-
ature goes to 0 K at this composition. This
is consistent with the exact solution which
Figure 8-19. B.c.c. structure: calculated prototype phase diagrams for ordering or separation tendencies
between first or second neighbors. The lines correspond to the CVM calculation in the tetrahedron approxima-
tion, the points correspond to the Monte Carlo simulation (Ackermann et al., 1989). The hachure indicates a
second-order transition. (a) W (1) = 2 W (2) > 0, (b) W (1) = – W (2) > 0, (c) W (1) = – W (2) < 0.
8.5.5 The Cluster Site Approximation

(CSA)
Notwithstanding the successes of the
CVM in the calculation of phase diagrams
and thermodynamic properties, a major
disadvantage is the large number of con-
figurational variables, and thus of non-
linear equations, to be solved in order to
minimize the Helmholtz energy. The num-
ber of independent variables in a K-com-
ponent system and an r-site CVM is
K r – 1. This high number increases strongly
with the size of the basic cluster, even for
binary systems. Real systems, however,
may involve many more than three compo-
nents.
The CSA suggested by Oates and Wenzl
(1996) is a revival of the quasi-chemical
tetrahedron approximation (Yang and Li,
1947). A system with N points is decom-
posed into Na clusters of type a in such a
way that the clusters share only points. The
important result of this assumption is that
only point correlation functions are used in
the Helmholtz energy minimization. The a
Figure 8-20. Hexagonal structure: calculated proto- cluster probabilities are obtained after the
type phase diagrams for the case of anisotropic near- calculation from the quasi-chemical equi-
est-neighbor interactions, W (11) within the basal librium between the atoms (points) and
plane and W (12) between basal planes. Monte Carlo
calculations (Crusius and Inden, 1988). The hachure
molecules (clusters). If a contains r points,
indicates a second-order transition. (a) W (12) > 0, the number of independent variables is
W (12)/W (11) = 0.8, (b) W (11) > 0, W (12) = 0. (K – 1) (r – 1). Thus with increasing cluster
size or number of components the number
is known for m * = 0 (Houtappel, 1950; Ne- of independent variables in the CSA is sig-
well, 1950; Wannier, 1950). The diagram nificantly smaller than in the case of CVM.
for W (12) = W (11) corresponds to the hexag- This is the main advantage.
onal close-packed structure, see Bichara Following the concept of the CVM, the
et al. (1992 b). This diagram is not shown entropy of the CSA in the r-site cluster ap-
because it is exactly the same as the one in proximation can be written
Fig. 8-15, except that the phases L12 and r
L10 have to be replaced by the phases D019 ({12…r}N )g ∏ {n }N
n =1
and B19, respectively. The reason for this S = kB ln r
is that the f.c.c. and hexagonal close pack- ∏ ({n }N )g
ings cannot be distinguished by nearest- n =1
neighbor interactions only. = g Sg − (g r − 1) S1 (8-37)
8.6 Application to Real Systems 561
where g is the coordination number of tet-

rahedra per lattice point. In the case of the
f.c.c. lattice and tetrahedron cluster this
number is g = 1, i.e. half the value used in
the CVM.
The principal reason for the lack of at-
tention to the CSA is its inability to obtain
the correct topology of prototype phase di-
agrams as obtained from CVM or MC cal-
culations. The tetrahedron approximation
of the f.c.c. lattice is a prominent example.
The CSA with g = 1 gives lro regions L12
and L10 around the stoichiometric compo-
sitions, but between these regions the dis-
ordered phase is predicted to be stable
down to 0 K. Oates et al. (1999) have
shown that a simple correction of the value
of g allows us to compensate for this. The
authors were able to reproduce the f.c.c.
prototype phase diagram of the tetrahedron Figure 8-21. Cu – Au phase diagrams calculated
CVM by taking g = 1.42. with the CSA using the constant entropy correction
Several successful results have been ob- factor g = 1.42 (Oates et al., 1999). a) Calculated
tained with the CSA; those for the Cu – Au without any size mismatch using the pair and CuCu-
CuAu and CuAuAuAu tetrahedron interactions de-
system will be mentioned here. Using the
rived by Kikuchi et al. (1980) using the tetrahedron
value g = 1.42 as a constant entropy correc- CVM. The phase diagram is the same as that of the
tion term for the tetrahedron treatment of CVM. b) Calculated with size mismatch using the
f.c.c. alloys, Oates et al. (1999) applied the many body interactions and the elastic energy terms
CSA to the Cu – Au system using the tetra- derived by Ferreira et al. (1987) using the tetrahe-
dron CVM. The phase diagram is the same as that of
hedron interactions from Kikuchi et al.
the CVM.
(1980). They obtained exactly the same
phase diagram as Kikuchi with the CVM.
The diagram is shown in Fig. 8-21 a. Fur- grams in Figs. 8-15, 8-16 and 8-19 a where
thermore, taking the many-body interac- such effects were not included.
tions and also the elastic energy arising These results indicate that the CSA may
from atomic size mismatch from Ferreira et be a very useful tool when it comes to tech-
al. (1987), Oates et al. obtained again the nological problems. It remains to be seen
same phase diagram as Ferreira with the whether the CSA can be applied as suc-
CVM, see Fig. 8-21 b. This diagram also cessfully to other structures and to larger
shows the effect of the elastic terms: they clusters.
contribute positive corrections to the con-
figurational energy, thus removing the de-
generacy of the ground states. Therefore, at 8.6 Application to Real Systems
0 K, the phase boundaries meet at the stoi-
chiometric compositions and go to the pure In contrast to prototype systems, real
components, contrary to the prototype dia- systems exhibit very complex properties
even at T = 0 K. With the progress made in ond neighbor pair interactions and used a
first-principles calculations, it has become power series expansion in composition for
obvious that higher-order pair and multisite the disordered state. No ternary interac-
cluster interactions have to be used in the tions were included. The agreement of the
configurational part of enthalpy, and that it results for the limitrophe binary systems
is mandatory to take lattice and local relax- with experimental data is satisfying only in
ations into account. This will be illustrated a few of the cases studied, and very little is
using the Au – Ni system as an example. presented with respect to ternary isother-
When it comes to T > 0 K the cluster inter- mal sections. McCormack et al. (1996,
actions have to embrace all configuration- 1997) studied the systems Cd – Ag – Au
dependent excitation energies such as arise (1996) and Cu – Al – Mn (1997) using first
from thermal vibrations and electronic ex- and second neighbor pair interactions, but
citations. The inclusion of these effects failed to present phase equilibria for ter-
means that the cluster interactions become nary and limitrophe binary systems that
temperature dependent. There is still some could be checked by experiments. In
disagreement as to whether vibrational the following the two ternary systems
contributions can (Ozoliņš et al., 1998 b; Fe –Ti – Al and Fe – Co – Al will be dis-
van de Walle et al., 1998) or cannot (Craie- cussed. The Fe – Co – Al system will be
vich and Sanchez, 1997; Craievich et al., treated as a magnetic system with spin 1/2
1997 a, b) be included in the cluster expan- given to Fe, Co and Al. This leads in fact to
sion. a six-component system.
However, even the apparently most care- In multicomponent systems the number
ful first-principles calculations (Ozoliņš et of points within a chosen basic cluster may
al., 1998 a) remain insufficiently accurate not be large enough to accommodate all
for technological purposes. For the fore- components. Then the question arises as to
seeable future it seems clear that simpler what extent this is detrimental to the qual-
approaches will continue to be important, ity of the approximation. In order to give
e.g. for technological phase diagram calcu- some idea of this effect, a magnetic spin
lations and also for other applications re- 7/2 system with f.c.c. structure has been
lated to the thermodynamic properties. The analyzed by Schön and Inden (2001) using
systems Fe – Al and Ni – Al will be taken the tetrahedron CVM and Monte Carlo
here as examples for calculations of phase simulations. This spin problem is equiva-
diagrams and of the thermodynamic factor lent to an eight-component alloy problem.
for diffusion. The calculation of phase equilibria and
Surveys of the abundant literature de- thermodynamic properties of multicompo-
scribing applications to mainly binary al- nent systems is of paramount importance
loy systems can be found in the reviews by for technological design and materials de-
Inden and Pitsch (1991) and by de Fontaine velopment. The use of both a cluster ex-
(1994). pansion for the energy and the CVM for
At this stage multicomponent systems configurational entropy introduces severe
present apparently insurmountable prob- computational problems because of the
lems for first-principles calculations. This large number of cluster types that must be
even holds for ternary systems. Rubin and considered in both formalisms. It also has
Finel (1993) studied ternary Ti – Al – X to be emphasized that in practical applica-
(X = W, Nb, Mo) systems with first and sections with coexisting phases of different
crystal structures, the calculation of the 8.6.1 The Au – Ni System

atomic configurations has to be performed
This system has been analyzed inten-
for each of the structures, and within each
sively over the last decade (Renaud et al.,
structure with the same level of approxima-
1987; Eymery et al., 1993; Wolverton and
tion for all subsystems in order to achieve
Zunger, 1997; Wolverton et al., 1998;
consistency. That is to say, isolated sophis-
Ozoliņš et al., 1998; Colinet and Pasturel,
ticated treatments of particular systems
2000). It has attracted much interest because
will provide deep insight into the physics,
of a phase separation tendency and positive
but they are only of limited use for solving
enthalpies of mixing at low temperatures,
technological problems.
ordering type of short-range order at high
Recently, the quasi-chemical tetrahedron
temperatures, and a large lattice mismatch
approximation has been adapted to over-
of about 15% between the constituents. It
come the problems of obtaining the cor-
may be considered a key system for check-
rect topology of order – disorder phase dia-
ing the quality of the theoretical methods.
grams for f.c.c. systems and to treat lattice
EXAFS experiments (Renaud et al.,
relaxations and excitations. This method
1987) have shown that the atoms are
seems to offer the possibility of handling
strongly displaced relative to the regular
order – disorder effects with sufficient qual-
undistorted lattice. The distribution of
ity within a format that can be used in ther-
Au – Au distances was found to be narrow
modynamic databases of multicomponent
and weakly asymmetric, while those for
systems. The Cd – Mg system will be taken
Ni – Ni were found to be widely distributed
as an example for illustration, because it
and highly asymmetric. Eymery et al.
has been extensively analyzed: all possible
(1993) came to the same conclusion on the
contributions to the ground states have
basis of their theoretical work. From the
been treated by first-principles calculation,
calculated average nearest-neighbor dis-
and the phase diagram has been calculated
tances, partial tetrahedral volumes were
by CVM (Asta et al., 1993).
calculated as a function of composition.
The results are shown in Fig. 8-22. The size
Figure 8-22. Variation of the

average tetrahedral volumes
Au4 , Au3Ni, AuNi, AuNi3 and
Ni4 in random Au1–x Nix solid
solutions at T = 0 K versus Ni
concentration [redrawn from
Eymery et al. (1993)]. The vol-
umes were calculated from the
calculated average nearest-
neighbor distances. The differ-
ence between the various vol-
umes and their variation with
composition reflects the lattice
relaxations. The calculations
were based on a tight-binding
second moment approximation.
space representation, and the constituent

strain energy. The authors concluded that
pair interactions up to about the 15th shell
are needed. Ozoliņš et al. (1998 a, b) exam-
ined the same question for the series of no-
ble metal alloys Cu – Au, Ag – Au, Cu – Ag
and Au – Ni. While in Ag – Au and Ag – Cu
the first three neighbor pair interactions are
dominant, the same is not true for the other
systems, particularly not for Au – Ni, where
not only pair interactions up to the 10th
shell, but also triplet and four-point cluster
Figure 8-23. Histogram of the calculated partial tet- interactions in increasing distances have to
rahedral volumes Au4 , Au3Ni, AuNi, AuNi3 and Ni4 be taken into account.
in a random alloy at T = 0 K with xNi = 0.5 [redrawn
Most recently, Colinet et al. (2000) took
from Eymery et al. (1993)]. The volume distributions
are broad and partially asymmetric. The average vol- up to fourth neighbor pair and triplet inter-
umes, taken from Fig. 8-22, are shown by the arrows. actions into account, the tetrahedron inter-
The difference between these average values and the actions turning out to be almost negligible
centers of gravity of the distributions is remarkable in their treatment. They achieved good
showing the importance of local relaxations.
agreement in the entropy of mixing, fairly
good agreement in the enthalpy and Gibbs
energy of mixing and calculated the phase
distribution of the tetrahedral volumes was diagram by means of the tetrahedron – octa-
also calculated. The results for the equi- hedron CVM. The comparison between the
atomic composition are shown in Fig. 8-23. calculated and experimental miscibility
The local relaxations become evident by gap is shown in Fig. 8-24.
the deviation of the average positions of
these distributions from the corresponding
average volumes in Fig. 8-22, which are
shown in Fig. 8-23 by the arrows. These re-
sults clearly indicate that both lattice and
local relaxations must not be ignored. This
fact becomes also visible in the enthalpy of
formation of the random alloys differing by
a factor of two between unrelaxed and re-
laxed states.
Wolverton et al. (1997, 1998) considered
the Au – Ni system and analyzed the ques-
tion of up to what order the pair interaction
scheme has to be driven in order to repro-
duce experimental findings. The CE was
Figure 8-24. Miscibility gap of the Au – Ni system.
separated into three parts: the pair interac-
Experiments from Bienzle et al. (1995). The calcula-
tions with arbitrary distance summed in the tions were performed using pair and triplet interac-
reciprocal space representation (Laks et al., tions in the tetrahedron – octahedron CVM (Colinet et
1992), the multi-atom interactions in real- al., 2000).
8.6.2 The Thermodynamic Factor following parameters are obtained (in units
of Ordered Phases of kB K)
8.6.2.1 The B.C.C. Fe – Al System V0 = – 2190, V1 = 0, V2 = 1680,
Iron aluminides such as Fe3Al and FeAl V3 = 457, V4 = 0, V5 = 52.5
have received considerable attention as or equivalently
candidates for high temperature structural (8-38)
(1) (2)
materials due to their low cost, high W = 1680, W = 740, W 1234
FeAlFeAl =– 140
strength and good oxidation resistance. With these parameters the irregular tetrahe-
Schön and Inden (1998) assessed the sys- dron CVM yields the b.c.c. phase diagram
tem in order to derive ECIs from experi- shown in Fig. 8-25. The calculated critical
mental enthalpies of formation, corrected temperatures match the high temperature
for b.c.c reference states and converted into experimental data. However, at lower tem-
the paramagnetic state: peratures, the tricritical point and the two-
phase region A2 + B2 are not obtained and
j DH j [J/mol of atoms] the agreement with the experimental phase
boundaries A2 + D03 is poor. This is not
Fe3Al (D03) – 18 650 surprising in view of the large relaxation
FeAl (B2) – 27 940 effects, which are to be expected in this
FeAl(B32) – 21 570 system as seen from the variation of lattice
FeAl3(D03) – 18 650 parameter as a function of composition and
of state of order, Fig. 8-26. On the other
Introducing these enthalpies and the values hand, recent experimental data on chemical
of the correlation functions of the configu- potential measurements at 1000 K (Kley-
rations (see Table 8-5) into Eq. (8-24), the kamp and Glasbrenner, 1997) are very well
Figure 8-25. The b.c.c. phase dia-

gram of Fe – Al calculated with the
irregular tetrahedron CVM using
the parameters given in Eq. (8-38)
(Schön and Inden, 1998). Second-
order transitions are indicated by
broken lines. Three temperatures
are indicated at which the thermo-
dynamic factor of diffusion is
shown below in Fig. 8-28. Experi-
mental points: neutron diffraction
and cp from Inden and Pepperhoff
(1990), dilatometry from Köster
and Gödecke (1980), TEM and
DTA from Ohnuma et al. (1998 a).
Figure 8-26. Variation of the

lattice parameter of Fe –Al alloys
as a function of composition and
state of order. Experiments: Lihl
and Ebel (1961).
Figure 8-27. Chemical potential

of Fe and Al at 1000 K in b.c.c.
Fe – Al alloys calculated with the
tetrahedron CVM using the
parameters given in Eq. (8-38)
(Schön and Inden, 1998). Experi-
mental data from Kleykamp and
Glasbrenner (1997).
reproduced by the calculations, see Fig. in ordered alloys. Fig. 8-28 shows this vari-
8-27. ation at three different temperatures. Close
A very important quantity for the treat- to stoichiometric compositions of ordered
ment of diffusion is the thermodynamic phases, F varies dramatically. The amount
factor defined for a binary system A – B as of variation depends on the degree of lro,
reaching up to an order of magnitude at
x B dm B
F= least in almost fully lro states, as observed
dx B at T = 650 K for B2 at the composition
This quantity is smooth in a random solid xAl = 0.5. At second-order transition points
solution, but shows a complicated variation F changes discontinuously.
Figure 8-28. Thermodynamic factor of diffusion in b.c.c. Fe – Al alloys, calculated as a function of composi-
tion at three temperatures, 650 K, 1000 K and 1400 K. The calculations were performed with the irregular
tetrahedron CVM using the parameters given in Eq. (8-38) (Schön and Inden, 1998). At second-order transi-
tions, F changes by a step. Close to the stoichiometric compositions xAl = 0.25 and 0.5 the value of F increases
as a function of the degree of lro. For example, the temperature of T = 650K is only slightly below the D03 /B2
transition temperature, see Fig. 8-25. Therefore, at xAl = 0.25 the increase is small compared to xAl = 0.5 where
F increases by more than an order of magnitude because the distance from the B2/A2 transition temperature is
large and the degree of lro almost maximum. With increasing temperature this effect becomes smaller.
8.6.2.2 The F.C.C. Ni – Al System Introducing these enthalpies and the values
The variation of the thermodynamic fac- of the correlation functions of the configu-
tor is even more pronounced in the Ni – Al rations (see Table 8-3) into Eq. (8-24), the
system because there the order – disorder following parameters are obtained (in units
transitions occur at much higher tempera- of kB K)
tures than in Fe – Al, as shown in the cal- V0 = – 4630, V1 = – 1563, V2 = 5051,
culated f.c.c. phase diagram in Fig. 8-29
(Schön and Inden, 1998). The ECIs were V3 = 1563, V4 = – 421
derived from the ground-state energies of or equivalently
L12 – Ni3Al, L12 – NiAl3 , and L10 – NiAl,
obtained by Pasturel et al. (1992) from W (1) = 3370, W 1234
NiAlAlAl = – 4810,
first-principles calculations: W 1234
NiNiNiAl = 1440 (8-39)
j DH j [J/mol of atoms] The calculated thermodynamic factor of
diffusion is shown in Fig. 8-30 for the two
Ni3Al (L12) – 48 000
temperatures indicated in Fig. 8-29. Due to
NiAl (L10) – 56 000
the high degree of lro, the variation of F in
NiAl3(L12) – 22 000
the range of stoichiometric composition is
Figure 8-29. Phase diagram

of the f.c.c. Ni – Al system
calculated with the regular
tetrahedron CVM (Schön
and Inden, 1998) using the
parameters given in Eq.
(8-39). Two temperatures are
indicated for which the ther-
modynamic factor has been
calculated, Fig. 8-30.
Figure 8-30. Thermody-

namic factor of diffusion in
f.c.c. Ni – Al alloys, calcu-
lated as a function of com-
position at two tempera-
tures, 1500 K and 2350 K
(Schön and Inden, 1998).
The calculations were per-
formed with the regular
tetrahedron CVM using
the parameters given in Eq.
(8-39). At the stoichiomet-
ric compositions, F in-
creases by more than an
order of magnitude.
particularly pronounced. Variations of the usually observed at phase boundaries. The

same order of magnitude are obtained for electron microprobe analysis reveals a
the B2 phase. steep change of composition at this boun-
It is worth mentioning that these dra- dary, which again could be misinterpreted
matic variations have repercussions in the as a tie-line. This is an important fact to be
composition profiles of diffusion couples. realized, because phase diagram determi-
An example is shown in Fig. 8-31, where a nations in multicomponent systems are
contrast has been observed in the micro- most conveniently performed by means of
graph within the B2 phase region as it is diffusion couple experiments.
parameters given in Table 8-9. The param-

eters for Fe – Al were taken as given in Eq.
(8-38).
The calculated isothermal sections at
1173 and 1073 K are shown in Figs. 8-32
and 8-33. From the metallurgical point of
Table 8-9. Atomic exchange energy parameters in

units of kB K (Ohnuma et al., 1998 b).
(1) (2) 1234

A–B WAB WAB W ABAB
Ti – Fe 1580 – 1050 1800

Al – Ti 2420 1200 0
Figure 8-31. Diffusion couple Ni77Al23 /Ni40Al60

annealed at 1273 K for 100 h (Kainuma et al., 1997).
a) Microstructure showing a change in contrast
within the B2 phase field. The regions labelled b1
and b2 do not represent two different phases. The
pseudo-boundary between b1 and b2 comes from a
steep change in composition. b) Composition profile
measured by electron probe microanalysis.
8.6.3 Ternary Systems

The b.c.c. states of the two systems
Fe – Ti – Al and Fe – Co – Al will now be
treated by taking the energy description
from the limitrophe binaries in order to see
what can be obtained by extrapolating from
the binaries into the ternary system without
ternary interaction terms.
8.6.3.1 B.C.C. Fe – Ti – Al
Ohnuma et al. (1998 b) studied ordering
and phase separation in the b.c.c. phase of
the ternary system, with particular empha-
sis on ternary miscibility gaps between dis- Figure 8-32. Calculated isothermal section of the
ordered and ordered phases. The theoreti- b.c.c. Fe – Ti – Al system at 1173 K using the energy
parameters in Table 8-9. Second-order transitions are
cal analysis was based on the irregular tet- shown as broken lines. The enlarged part of the sec-
rahedron CVM. The analysis of the Fe – Ti tion shows the good agreement obtained between ex-
and Ti – Al binary systems led to the energy periments and calculation (Ohnuma et al., 1998 b).
8.6.3.2 B.C.C. Ferromagnetic Fe–Co–Al

(six-component system with spin 1/2
for Fe, Co and Al)
This system has been studied by Colinet
et al. (1993) in the irregular tetrahedron
approximation. The binary system Fe – Al
was discussed in Sec. 8.6.2.1, but without
taking magnetic effects into account. Be-
cause the Curie temperature of metastable
b.c.c. Fe – Co alloys goes up to about
1500 K, the magnetic effects cannot be
disregarded in this ternary system. A spin
1/2 treatment has been taken in this in-
stance. The magnetic exchange energies
are defined in the same way as the atomic
≠Ø ≠≠ ØØ
equivalents: JAB = – 2 JAB + JAB + JAB , with
≠Ø
A, B Œ{Fe, Co, Al} and the JAB etc. being
proportional to the corresponding ex-
change integrals. The energy parameters
are given in Table 8-10.
The phase diagram of the b.c.c. Fe – Co
system calculated with the energy parame-
ters given in Table 8-10 is shown in Fig.
8-34. All transitions are of second order
Figure 8-33. Calculated isothermal section of the and have been obtained with the second
b.c.c. Fe – Ti – Al system at 1073 K using the energy Hessian method. The calculation without
parameters given in Table 8-9. Second-order transi-
tions are shown as broken lines. The enlarged part of
magnetic interactions is also shown. In this
the section shows the good agreement obtained be- non-magnetic case the critical temperature
tween experiments and calculation (Ohnuma et al., of the B2/A2 transition is lower than in the
1998 b). ferromagnetic case. The magnetic interac-
tions strengthen the atomic ordering in this
instance.
view it is interesting to see the opening of
In the Co – Al system, the b.c.c. phase is
ternary miscibility gaps between ordered
stable only around the equiatomic compo-
and disordered phases. This allows us to
sition, and there the B2 structure is stable
produce coherent two-phase equilibria with
interesting mechanical properties. Guided
by the calculations, Ohnuma et al. (1998 b)
also performed experiments in order to Table 8-10. Atomic and magnetic exchange energy
parameters in units of kB K (Colinet et al., 1993).
confirm the existence of the two-phase
fields experimentally. The results are A–B (1)
WAB (2)
WAB (1)
JAA (2)
JBB (1)
JAB
shown in the enlarged parts of the sections
of Figs. 8-32 and 8-33. The calculated tie- Fe – Co 500 0 – 163 – 218 – 24
Fe – Al 1680 740 – 163 0 – 38
lines agree very well with the experimental
Co – Al 3600 1500 – 218 0 – 38
results.
Figure 8-35. Phase diagram of the Co – Al system.

Second-order transitions are indicated by a hachure.
Heavy lines: CVM calculation in the (irregular) tetra-
hedron approximation using the interchange energies
in Table 8-10 (Colinet et al., 1993). Light lines:
Phase diagram according to Hansen and Anderko
Figure 8-34. Phase diagram of the Fe – Co system. (1958).
Heavy lines: CVM calculation in the (irregular) tetra-
hedron approximation using the interchange energies f.c.c. phase, and intermetallic compounds.
given in Table 8-10 (Colinet et al., 1993); lower The B2 phase exists up to very high tem-
curve: no magnetic interactions, upper curve: with
magnetic interactions. Experiments: (˘) Masumoto
peratures in a metastable state, far within
et al. (1954), (Ñ) Eguchi et al. (1968), (¸) Oyedele the liquid phase. This tendency is consis-
and Collins (1977). tent with the trend obtained experimentally
from data on ternary alloys (Ackermann,
1988).
up to the melting point. The D03 structure Figs. 8-36 and 8-37 show the isothermal
has not been observed within the b.c.c. sections of the ternary system Fe – Co – Al
phase field. This imposes an upper bound at 1000 K and at 600 K. The ordered re-
(2)
as to the value of WCoAl that controls the gions are separated by first-order and sec-
value of the critical temperature D03 /B2. ond-order transitions. At 600 K the mag-
Based on these considerations and on ex- netic interactions become important, but
perimental values for the critical tempera- they are still small compared with the
ture B2/A2 in some ternary Fe – Co – Al al- chemical interactions. Prior to these calcu-
loys, the values in Table 8-10 have been lations, Ackermann (1988) performed an
derived by Ackermann (1988). MC calculation of the isothermal section at
The b.c.c. phase diagram calculated in 700 K using the same chemical interaction
the tetrahedron approximation (Colinet et parameters as those in Table 8-10. Her dia-
al., 1993) is shown in Fig. 8-35, super- gram and the corresponding CVM diagram
posed to the stable diagram. This includes agree very well (see Inden and Pitsch,
the equilibria with the liquid phase, the 1991).
Ackermann (1988) also analyzed the

Fe – Co – Al system experimentally. Here
only one comparison will be shown. Fig.
8-38 shows the calculated and experimen-
tal miscibility gap in a vertical section
which happened to be quasibinary (the tie-
lines are within the section).
The agreement between experiment and
calculation is surprisingly good, though not
really satisfactory. The calculated miscibil-
ity gap ends with a “horn” at the tricritical
point. This is not obtained in the experi-
ments. However, we should recall that the
two coexisting phases A2 and B2 are
Figure 8-36. Isothermal section of the phase equi- coherent with different lattice parameters.
libria of ternary b.c.c. Fe – Co – Al alloy at It has been shown by Williams (1980,
T = 1000 K, calculated with the CVM in the (irregu- 1984) and Cahn and Larché (1984) that the
lar) tetrahedron approximation using the energy pa- boundary of so-called “coherent” miscibil-
rameters given in Table 8-10 (Colinet et al., 1993).
ity gaps (with elastic contributions) is
lower than the so-called “incoherent” gaps
(no elastic contributions). Therefore, elas-
tic effects due to the precipitation of coher-
ent particles depress the phase boundary.
This is only one of the effects to be taken
into account. A second one is due to the
lattice and local relaxations within both of
these phases, as already pointed out in Sec.
8.6.2.1 for Fe – Al. The arguments present-
ed for Fe – Al are also valid for the
Fe – Co – Al system with Fe and Co being
very similar. These relaxation effects tend
to shift the miscibility gap to higher tem-
peratures, opposite to the first effect. A full
calculation is needed to evaluate the com-
bined result.
Simultaneously with the work by Acker-
mann (1988) the miscibility gap was ex-
perimentally observed by Miyazaki et al.
Figure 8-37. Isothermal section of the phase equi- (1987) using transmission electron micros-
libria of ternary b.c.c. Fe – Co – Al alloys at copy (TEM). These authors, however,
T = 700 K, calculated with the CVM in the (irregular) speculated that the miscibility gap was due
tetrahedron approximation using the energy parame-
to magnetic effects and assumed the tie-
ters given in Table 8-10 (Colinet et al., 1993). Sec-
ond-order transitions are indicated by a hachure. At lines to be oriented towards the Fe – Co
this temperature ferromagnetic (fm) and paramag- binary system, i.e. perpendicular to those
netic (pm) states have to be distinguished. shown in Fig. 8-37. The contradiction be-
Figure 8-38. Vertical section of

the phase diagram showing the
ternary miscibility gap as a func-
tion of composition according
to experiment and MC (Acker-
mann, 1988), (a) Experiments
(¯) miscibility gap, (É) critical
temperature of lro. (b) Calcula-
tion with the MC method: (¯)
miscibility gap and critical tem-
perature of lro.
tween the results of Miyazaki et al. and hedron – octahedron CVM to calculate the
those from the CVM and MC pointed out phase diagram, which is shown in Fig. 8-
by Inden and Pitsch (1991) was removed 39 a. The topology of the diagram is correct
by the TEM – EDS analysis of Kozakai and but it does not have the accuracy necessary
Miyazaki (1994) confirming the calculated for technological purposes.
direction of the tie-lines. Zhang et al. (2000) tried the CSA in the
tetrahedron approximation. Because the
c/a ratio is not ideal (f.c.c. and h.c.p. lat-
8.6.4 H.C.P. Cd – Mg
tices become equivalent in the ideal case),
The Cd – Mg system is a prototype order- two irregular tetrahedra should be consid-
ing system for h.c.p. alloys in the same way ered (Onodera et al., 1994). Zhang et al.
as Cu – Au plays this role for f.c.c. alloys. (2000) tried the irregular tetrahedra, but
This system has been studied using first- found that the regular tetrahedron CSA
principles methods by Asta et al. (1993). gave an equally good description. The
The configurational energy was evaluated value of g was varied in such a way that the
by a cluster expansion up to seven-point two congruent maxima which appear at the
clusters, as required for a treatment with compositions A3B and AB3 for g = 1.42
the octahedron – tetrahedron CVM. Lattice (valid for the ideal h.c.p. and f.c.c., see Sec.
relaxation and vibrational energy were 8.5.5) moved towards the mid-composition
taken into account, but no local relaxations. until, at g = 1.8, the maxima disappeared,
The lattice relaxation cluster expansion just as observed in the Cd – Mg system.
was based on experimental results for dis- Starting with energies taken from Asta et
ordered alloys, and the other ECIs were ob- al. (1993), which proved to be excellent,
tained from total-energy calculations on or- only slight changes were applied in order
dered phases. In total 32 energy terms were to get an optimum description of all the
used. The ECIs were then used in the tetra- properties that can be checked with avail-
Figure 8-39. Calculated phase diagrams of the hexagonal Cd – Mg system. a) Phase diagram calculated from
“first principles” (redrawn from Asta et al., 1993). b) Phase diagram calculated with the CSA using almost the
same parameters as in a) and a value g = 1.8. Experimental data from Frantz and Gantois (1971).
able experimental data: the phase diagram, At present, it can be concluded that the
the enthalpy of mixing of the disordered CVM and MC techniques supply a treat-
state, the enthalpies of formation of or- ment of sufficient sophistication to cor-
dered phases and the chemical potentials of rectly handle the statistical aspects of the
Cd and Mg. The cluster energies used were equilibria in solid solutions.
(the values of Asta et al. are given in brack- Much progress has been made in the
ets): field of first-principles calculations of total
energies, including lattice and local relaxa-
Cd3Mg: – 9.3 (–10.3); CdMg: –13.7 (–13.5);
tions, and sometimes including excitations.
CdMg3: – 9.95 (– 9.5) in kJ/(mol of atoms). From this the energy parameters of the sta-
tistical models can be calculated. The re-
The calculated phase diagram is shown
sults for binary systems are numerous, but
in Fig. 8-39 b. The agreement between cal-
the field of multicomponent systems is still
culation and experiment is very good, not
to be explored. With increasing numbers of
only for the phase diagram but also for all
components the cluster size also has to be
the other properties mentioned above.
increased.
For the solution of metallurgical prob-
8.6.5 Concluding Remarks lems, the phase equilibria between all
phases have to be considered. The order –
The examples treated above represent disorder equilibria within a given crystal
the large group of real systems that can be structure are only one part of this task.
analyzed using these techniques. Similar Other phases, such as the liquid phase or
findings were obtained for oxide systems, intermetallic compounds, have to be in-
e.g., by Burton (1984, 1985), Burton and cluded. These aspects have been discussed
Kikuchi (1984), Kikuchi and Burton (1988), at recent workshops on the thermodynamic
Burton and Cohen (1995), Tepesch et al. modeling of solutions and alloys (e.g. Cac-
(1995, 1996), Kohan and Ceder (1996). ciamani et al., 1997).
8.7 Appendix 575
8.7 Appendix
Table 8-11. Superstructures of the f.c.c. lattice. 1
Designation 2 Spacegroup Basis 3 Equivalent positions Occupation
Positions
A1 (Cu) Fm3m a = a0 {1 0 0} (0 0 0; 1/2 1/2 0; 1/2 0 1/2; 0 1/2 1/2) A/B

–
A5B B2/m a = a0 /2 [1 1 2] (0 0 0; 1/2 0 1/2) +
–
Fig. 8-40 b = a0 /2 [2 2 2] (0 0 0) 2B
c = a0 /2 [3 3 0] (0 1/2 0) 2A
(0 0 ±1/3) 4A
(0 1/2 ± 1/3) 4A
–
D1a (Ni4Mo) I4/m a1 = a0 /2 [3 1 0] (0 0 0; 1/2 0 1/2) +
Fig. 8-41 a2 = a0 /2 [1 3 0] (0 0 0) 2B
c = a0 /2 [0 0 2] (x y 0; x– y– 0; y– x 0; y x– 0) 8A
ææ æ
x = a0 ÷ 2/5 y = a0 / ÷10
L12 (Cu3Au) Pm3m a = a0 {1 0 0} ––
(0 0 0) B
(1/2 1/2 0; 1/2 0 1/2; 0 1/2 1/2) 3A
–– P4/mmm a1 = a0 [1 0 0] ––
a2 = a0 [0 1 0] (0 0 0) B
c = a0 [0 0 1] (1/2 1/2 0) AB
(1/2 0 1/2; 0 1/2 1/2) A
D022 (Ti3Al) I4/mmm a1 = a0 /2 [0 1 0] (0 0 0; 1/2 1/2 1/2) +
Fig. 8-42 a2 = a0 /2 [0 0 1] (0 0 0) 2B
c = a0 [2 0 0] (0 0 1/2) 2A
(0 1/2 1/4; 1/2 0 1/4) 4A
A3B Amm2 a = a0 /2 [0 2 2] (0 0 0; 1/2 0 1/2) +
–
b = a0 /2 [0 1 1] (0 0 0) 2B
c = a0 /2 [0 0 2] (0 0 1/2) 2A
(1/4 1/2 1/4; 1/4 1/2 3/4) 4A
–
A2B (Pt2Mo) Immm a = a0 /2 [1 1 0] (0 0 0; 1/2 1/2 1/2) +
Fig. 8-43 b = a0 [1 0 0] (0 0 0) 2B
c = a0 /2 [3 3 0] (0 0 ±1/3) 4A
L10 (CuAu) P4/mmm a1 = a0 /2 [1 1 0] ––
–
a2 = a0 /2 [1 1 0] (0 0 0) B
c = a0 [0 0 1] (1/2 1/2 1/2) A
–
L11 (CuPt) R 3m a1 = a0 /2 [2 1 1] ––
Fig. 8-44 a2 = a0 /2 [1 2 1] (0 0 0) B
a3 = a0 /2 [1 1 2] (1/2 1/2 1/2) A
A 2B2 I41/amd a1 = a0 [0 1 0] (0 0 0; 1/2 1/2 1/2) +
Fig. 8-45 a2 = a0 [0 0 1] (0 0 0; 0 1/2 1/4) B
c = a0 [2 0 0] (0 0 1/2; 0 1/2 3/4) A
1 Figures 8-40 to 8-45 represent the original f.c.c. unit cell and the unit cells of the superstructures; 2 “Struk-
turbericht” designation; 3 In terms of vectors of the f.c.c. structure.
Figure 8-40. Original f.c.c. unit cell and unit cell of Figure 8-43. Original f.c.c. unit cell and unit cell of
the superstructure A5B (B2/m), indicated by bold the superstructure A2B (Pt2Mo), indicated by bold
lines. lines.
the superstructure D1a (Ni4Mo), indicated by bold the superstructure L11 (CuPt), indicated by bold
lines. lines.
the superstructure D022 (Ti3Al), indicated by bold the superstructure A2B2 (I4/amd), indicated by bold
lines. lines.
8.7 Appendix 577
Table 8-12. Superstructures of the b.c.c. lattice.
Positions
A2 (Fe) Fd3c a = a0 {1 0 0} –– A/B

D03 (Fe3Al) Fm3m a1 = a0 [2 0 0] ––
a2 = a0 [0 2 0] (0 0 0) 4B
a3 = a0 [0 0 2] (1/2 1/2 1/2) 4A
(1/4 1/4 1/4; 3/4 3/4 3/4) 8A
–
F43m a1 = a0 [2 0 0] ––
a2 = a0 [0 2 0] (0 0 0) 4B
a3 = a0 [0 0 2] (1/2 1/2 1/2) 4A
(1/4 1/4 1/4) 4 A/B
(3/4 3/4 3/4) 4A
B2 (CsCl) Pm3m a = a0 {1 0 0} ––
(0 0 0) A
(1/2 1/2 1/2) B
B32 (NaTl) Fd3m a1 = a0 [2 0 0] (0 0 0; 0 1/2 1/2; 1/2 0 1/2; 0 1/2 1/2) +
a2 = a0 [0 2 0] (0 0 0; 1/4 1/4 1/4) 8A
a3 = a0 [0 0 2] (1/2 1/2 1/2; 3/4 3/4 3/4) 8B
1 “Strukturbericht” designation; 2 In terms of vectors of the b.c.c. structure.
Table 8-13. Superstructures of the hexagonal lattice.
Positions
A3 F63 /mmc a = a0 [1 0 0] ––
b = a0 [0 1 0] (1/3 2/3 3/4) 2 (A/B)
c = a0 [0 0 1]
D019 (A3B) P63/mmc a = a0 [2 0 0] ––
b = a0 [0 2 0] 6 h (5/6 2/3 1/4) 6A
c = c0 [1 0 0] 2 c (1/3 2/3 1/4) 2B
–
B19 Pmma a = c0 [0 0 1] ––
b = a0 [0 1 0] 2 f (1/4 1/2 5/6) 2A
c = a0 [2 1 0] 2 e (1/4 0 1/3) 2B
–
A 2B P21/m a = a0 [1 1 0] 2 e (1/2 1/6 1/4) A
b = a0 [1 2 0] 2 e (1/6 1/2 1/4) A
c = c0 [0 0 1] 2 e (5/6 5/6 1/4) B
–
A 2B p6m a = a0 [1 1] 2 b (1/3 2/3) A
(2-dim.) b = a0 [1 2] 1 a (0 0) B
1 “Strukturbericht” designation; 2 In terms of vectors of the h.c.p. structure.

8.8 Acknowledgements Binder, K. (1980), Phys. Rev. Lett. 45, 811.

Binder, K. (1981), Z. Physik B 45, 61.
Binder, K. (1985), J. Comput. Phys. 59, 1.
In writing this chapter the author has Binder, K. (1986), in: Advances in Solid State Phys-
benefited from the work of past students, ics, Vol. 26: Braunschweig: Vieweg, p. 133.
Claudio G. Schön, Sabine Crusius and Binder, K. (2001), in: Phase Transformations in Ma-
terials: Kostorz, G. (Ed.). Weinheim: Wiley-VCH,
Helen Ackermann, and from intense co-op- Ch. 4.
eration with colleagues, Ryoichi Kikuchi, Binder, K., Heermann, D. W. (1988), Monte Carlo
Catherine Colinet and Christophe Bichara. Simulations in Statistical Physics, Springer Series
Solid State Sci., Fulde, P. (Ed.), Vol. 80. Heidel-
This is gratefully acknowledged. The con- berg: Springer.
cept of multicomponent correlation func- Binder, K., Stauffer, D. (1984), in: Applications of
tions was worked out in co-operation with the Monte Carlo Method in Statistical Physics:
Binder, K. (Ed.). Berlin: Springer, p. 1.
Alphonse Finel when writing the version Binder, K., Kinzel, W., Selke, W. (1983), J. Magn.
for the first edition of this volume (Inden Magn. Mater. 31 – 34, 1445.
and Pitsch, 1991). Thanks are due to Alan Binder, K., Lebowitz, J. L., Phani, M. K., Kalos, M.
H. (1981), Acta Metall. 29, 1655.
Oates for very helpful and enlightening Bond, R. A., Ross, D. K. (1982), J. Phys. F 12, 597.
discussions. Burley, D. M. (1972), in: Phase Transitions and Crit-
ical Phenomena, Vol. 2: Domb, C., Green, M. S.
(Eds.). New York: Academic Press, p. 329.
Burton, B. (1984), Phys. Chem. Minerals 11, 132.
8.9 References Burton, B. (1985), Am. Mineralogist 70, 1027.
Burton, P. B., Cohen, R. E. (1995), Phys. Rev. B 52,
Ackermann, H. (1988), Thesis, RWTH Aachen, Ger- 792.
many. Burton, B., Kikuchi, R. (1984), Am. Mineralogist 69,
Ackermann, H., Crusius, S., Inden, G. (1986), Acta 165.
Metall. 34, 2311. Cacciamani, G., Chang, Y. A., Grimvall, G., Franke,
Ackermann, H., Inden, G., Kikuchi, R. (1989), Acta P., Kaufman, L., Miodownik, P., Sanchez, J. M.,
Metall. 37, 1. Schalin, M., Sigli, C. (1997), Calphad 21, 219.
Allen, S. M., Cahn, J. W. (1972), Acta Metall. 20, Cahn, J. W., Kikuchi, R. (1979), Acta Metall. 27,
423. 1329.
Allen, S. M., Cahn, J. W. (1973), Scripta. Metall. 7, Cahn, J. W., Larché, F. (1984), Acta Metall. 32, 1915.
1261. Chesnut, D. A., Salsburg, Z. W. (1963), J. Chem.
Allen, S. M., Cahn, J. W. (1982), Bulletin Alloy Phys. 38, 2861.
Phase Diagrams 3, 287. Clapp, P. C., Moss, S. (1966), Phys. Rev. 142, 418.
Asta, M., McCormack, R., de Fontaine, D. (1993), Clapp, P. C., Moss, S. (1968), Phys. Rev. 171, 754.
Phys. Rev. B 48, 748. Colinet, C., Pasturel, A. (2000), J. Alloys Comp. 296,
van Baal, C. M. (1973), Physica 64, 571. 6.
Barker, J. A. (1953), Proc. Roy. Soc. A 216, 45. Colinet, C., Inden, G., Kikuchi, R. (1993), Acta Met-
Bichara, C., Crusius, S., Inden, G. (1992 a), Physica all. Mater. 41, 1109.
B 182, 42. Connolly, J. W. D., Williams, A. R. (1983), Phys.
Bichara, C., Crusius, S., Inden, G. (1992 b), Physica Rev. B 27, 5169.
B 179, 221. Cowley, J. M. (1950), Phys. Rev. 77, 669.
Bieber, A., Gautier, F. (1984 a), Z. Physik B 57, 335. Craievich, P. J., Sanchez, J. M. (1997), Comp. Mater.
Bieber, A., Gautier, F. (1984 b), J. Phys. Chem. Soc. Sci. 8, 82.
Japan 53, 2061. Craievich, P. J., Sanchez, J. M., Watson, R. E., Wei-
Bieber, A., Gautier, F. (1986), Acta Metall. 34, 2291. nert, M. (1997), Phys. Rev. B 55, 787.
Bieber, A., Gautier, F. (1987), Acta Metall. 35, 1839. Crusius, S., Inden, G. (1988), in: Proc. Int. Symp.
Bienzle, M., Oishi, T., Sommer, F. (1995), J. Alloys Dynamics of Ordering Processes in Condensed
Comp. 220, 182. Matter, Kyoto 1987: Komura, S., Furukawa, H.
Binder, K. (1976), in: Phase Transitions and Critical (Eds.). New York: Plenum, p. 138.
Phenomena, Vol. 5 b: Domb, C., Green, M. S. Diep, H. T., Ghazali, A., Berge, B., Lallemand, P.
(Eds.). New York: Academic Press, p. 2. (1986), Europhys. Lett. 2, 603.
Binder, K. (1979), in: Monte Carlo Methods in Sta- Ducastelle, F. (1991), in: Cohesion and Structure,
tistical Physics: Binder, K. (Ed.). Berlin: Springer, Vol. 3: de Boer, F. R., Pettifor, D. G. (Eds.). Am-
p. 1. sterdam: North-Holland, p. 1.
8.9 References 579
Ducastelle, F., Gautier, F. (1976), J. Phys. F 6, 2039. Gyorffy, B. L., Stocks, G. M. (1983), Phys. Rev. Lett.
Eymery, J., Lançon, F., Billard, L. (1993), J. Phy- 50, 374.
sique I France 3, 787. Hansen, M., Anderko, K. (1958), Constitution of Bi-
Ferreira, L. G., Mbaye, A. A., Zunger, A. (1987), nary Alloys. New York: McGraw-Hill.
Phys. Rev. B 35, 6475. Hijmans, J., de Boer, J. (1955), Physica 21, 471.
Ferreira, L. G., Mbaye, A. A., Zunger, A. (1988), Houtappel, R. M. F. (1950), Physica 16, 425.
Phys. Rev. B 37, 10 547. Inden, G. (1975 a), Z. Metallkde. 66, 577.
Ferreira, L. G., Wei, S.-H., Zunger, A. (1989), Phys. Inden, G. (1975 b), Z. Metallkde. 66, 648.
Rev. B 40, 3197. Inden, G. (1977 a), Z. Metallkde. 68, 529.
Finel, A. (1987), Thesis, University Pierre et Marie Inden, G. (1977 b), J. de Physique 38, C7, 373.
Curie, Paris. Inden, G., Pepperhoff, W. (1990), Z. Metallkde. 81,
Finel, A. (1994), in: Statics and Dynamics of Alloy 770.
Phase Transformations: Turchi, P. E. A., Gonis, A. Inden, G., Pitsch, W. (1991), in: Materials Science
(Eds.). NATO ASI Series B, Physics, Vol. 319. and Technology, Vol. 5: Phase Transformation in
Dordrecht: Kluwer, p. 495. Materials: Haasen, P. (Ed.). New York: VCH
Finel, A., Ducastelle, F. (1984), in: Mat. Res. Soc. Press, p. 497.
Symp. Proc. 21: Phase Transformations in Solids, Jacobs, L., Rebbi, C. (1981), J. Comput. Physics 41,
Maleme (Crete) 1983. Amsterdam: North-Holland 203.
Publ., p. 293. Kainuma, R., Ikenoya, H., Ohnuma, I., Ishida, K.
Finel, A., Ducastelle, F. (1986), Europhys. Lett. 1, (1997), Defect and Diffusion Forum Vols. 143 –
135 and Erratum p. 543. 147, 425.
Finel, A., Gratias, D., Portier, R. (1984), in: L’ordre Kalle, C., Winkelmann, V. (1982), J. Stat. Phys. 28,
et le Desordre dans les Materiaux, Reynaud, F., 639.
Clément, N., Couderc, J. J. (Eds.). Ecole d’Hiver Kanamori, J. (1966), Prog. Theor. Phys. 35, 16.
Aussois 1984. Les Ulis: Editions de Physique, Kanamori, J. (1979), in: Int. Conf. on Modulated
p. 9. Structures: Cowley, J. M., Cohen, J. B., Salomon,
Flinn, P. A. (1956), Phys. Rev. 104, 350. M. B., Wuensch, B. J. (Eds.). AIP Conf. Proc. 53,
de Fontaine, D. (1979), in: Solid State Physics, Vol. p. 117.
34: Ehrenreich, H., Seitz, F., Turnbull, D. (Eds.). Kanamori, J. (1984), J. Phys. Soc. Jpn. 53, 250.
London: Academic Press, p. 73. Kanamori, J., Kakehashi, Y. (1977), J. de Physique
de Fontaine, D. (1984), in: High Temperature Or- 38, C-7, 274.
dered Intermetallic Alloys, Mat. Res. Soc. Proc., Kear, B. H., Sims, C. T., Stoloff, N. S., Westbrook, J.
Vol. 39: Koch, Liu, Stoloff (Eds.). Pittsburgh: Ma- H. (Eds.) (1970), Ordered Alloys. Proc. 3rd. Bol-
terials Res. Soc., p. 43. ton Landing Conf. Baton Rouge: Claitors Publ.
de Fontaine, D. (1994), Solid State Phys. 47, 33. Khachaturyan, A. G. (1978), in: Progress in Materi-
Fosdick, L. D. (1963), Methods Comp. Phys. 1, 245. als Science, Vol. 22. New York: Pergamon Press,
Frantz, C., Gantois, M. (1971), J. Appl. Cryst. 4, p. 1.
387. Kikuchi, R. (1950), Phys. Rev. 79, 718.
Gahn, U. (1982), J. Phys. Chem. Solids 43, 977. Kikuchi, R. (1951), Phys. Rev. 81, 988.
Gahn, U. (1986), J. Phys. Chem. Solids 47, 1153. Kikuchi, R. (1974), J. Chem. Phys. 60, 1071.
Gahn, U., Sauthoff, G., Pitsch, W. (1984), in: Decom- Kikuchi, R. (1977 a), Acta Metall. 25, 195.
position of Alloys, the Early Stages, Acta-Scripta Kikuchi, R. (1977 b), J. de Physique 38, C7-307.
Metallurgica Conf. Proc.: Haasen, P., Gerold, V., Kikuchi, R. (1986), Progr. Theor. Phys. Suppl. 87,
Wagner, R., Ashby, M. F. (Eds.). New York: Per- 69.
gamon Press, p. 16. Kikuchi, R. (1987), Physica 142 A, 321.
Gautier, F. (1984), in: High Temperature Alloys: The- Kikuchi, R., van Baal, C. M. (1974), Scripta Metall.
ory and Design: Stiegler, P. O. (Ed.). New York: 8, 425.
Am. Inst. Min. Engrs., p. 163. Kikuchi, R., Burton, B. (1988), Physica B 150, 132.
Gautier, F., Ducastelle, F., Giner, J. (1975), Philos. Kikuchi, R., Sato, H. (1974), Acta Metall. 22, 1099.
Mag. 31, 1373. Kikuchi, R., de Fontaine, D., Murakami, M., Naka-
Gerold, V., Kern, J. (1987), Acta Metall. 35, 393. mura, T. (1977), Acta Metall. 25, 207.
Golosov, N. S., Tolstick, A. M. (1974), J. Phys. Kikuchi, R., Sanchez, J. M., de Fontaine, D., Yama-
Chem. Solids 36, 899. uchi, H. (1980), Acta Metall. 28, 651.
Golosov, N. S., Tolstick, A. M. (1975), J. Phys. Kleykamp, H., Glasbrenner, H. (1997), Z. Metallkde.
Chem. Solids 36, 903. 88, 230.
Golosov, N. S., Tolstick, A. M. (1976), J. Phys. Koch, C. C., Liu, C. T., Stoloff, N. S. (Eds.) (1985),
Chem. Solids 37, 273. High Temperature Ordered Intermetallic Alloys I,
Gratias, D., Sanchez, J. M., de Fontaine, D. (1982), Mat. Res. Soc. Symp. Proc., Vol. 39. Pittsburgh:
Physica A 113, 315. Materials Res. Soc.
Köster, W., Gödecke, T. (1980), Z. Metallkde. 81, Ozoliņš, V., Wolverton, C., Zunger, A. (1998 b),
770. Phys. Rev. B 57, 5897.
Kohan, A. F., Ceder, G. (1996), Phys. Rev. B 53, Pasturel, A., Colinet, C., Paxton, A. T., van Schilf-
8993. gaarde, M. (1992), J. Phys.: Condens. Matter 4,
Kozakai, T., Miyazaki, T. (1994), J. Mater. Sci. 29, 945.
652. Priem, T., Beuneu, B., Finel, A., Livet, F., de Novion,
Kudo, T., Katsura, S. (1976), Prog. Theor. Phys. 56, C. H. (1989 a). J. de Physique 50, 2217.
435. Priem, T., Beuneu, B., de Novion, C. H., Chevrier, J.,
Kutner, R., Binder, K., Kehr, K. W. (1982), Phys. Livet, F., Finel, A., Lefevbre, S. (1989 b). Physica
Rev. B 26, 2967. B 156 + 157, 47.
Laks, D. B., Ferreira, L. G., Froyen, S., Zunger, A. Renaud, G., Motta, N., Belakhovsky, M., Lançon, F.,
(1992), Phys. Rev. B 46, 12 587. Billard, L. (1987), Solid State Comm. 63, 569.
Lebowitz, J. L., Phani, M. K., Styer, D. F. (1985), J. Richards, M. J., Cahn, J. W. (1971), Acta Metall. 19,
Stat. Phys. 38, 413. 1263.
Li, Y. Y. (1949), Phys. Rev. 76, 972. de Ridder, R., van Dyck, D., Amelinckx, S. (1980),
Li, Y. Y. (1949), J. Chem. Phys. 17, 447. Phys. Stat. Solidi A 61, 231.
Lihl, F., Ebel, H. (1961), Arch. Eisenhüttenwes. 32, Rubin, G., Finel, A. (1993), J. Phys.: Condens. Mat-
483. ter 5, 9105.
Liu, C. T., Taub, A. I., Stoloff, N. S., Koch, C. C. Sanchez, J. M. (1996), in: Theory and Applications
(Eds.) (1989), High Temperature Ordered Inter- of the Cluster Variation and Path Probability
metallic Alloys III, Mat. Res. Soc. Symp. Proc., Methods: Moran-Lopez, J. L., Sanchez, J. M.
Vol. 133, Pittsburgh: Materials Res. Soc. (Eds.). New York: Plenum Press, p. 175.
Liu, C. T., Cahn, R. W., Sauthoff, G. (Eds.) (1992), Sanchez, J. M., de Fontaine, D. (1978), Phys. Rev.
Ordered Intermetallics – Physical Metallurgy and B 17, 2926.
Mechnical Behaviour, Nato ASI Series Vol. 213. Sanchez, J. M., de Fontaine, D. (1980), Phys. Rev.
Dordrecht: Kluwer, p. 525. B 21, 216.
Livet, F. (1987), Acta Metall. 35, 2915. Sanchez, J. M., de Fontaine, D. (1981), in: Structure
Lu, Z. W., Wei, S. H., Zunger, A., Frota-Pessoa, S., and Bonding in Crystals, Vol. II: O’Keefe, M.,
Ferreira, L. G. (1991), Phys. Rev. B 44, 512. Navrotsky, A. (Eds.). New York: Academic Press,
Lu, Z. W., Klein, B. M., Zunger, A. (1995), Model. p. 917.
Sim. Mater. Sci. Eng. 3, 753. Sanchez, J. M., de Fontaine, D. (1982), Phys. Rev.
McCormack, R., de Fontaine, D. (1996), Phys. Rev. B 25, 1759.
B 54, 9746. Sanchez, J. M., de Fontaine, D., Teitler, W. (1982),
McCormack, R., Asta, M., Hoyt, J. J., Chakoumakos, Phys. Rev. B 26, 1465.
B. C., Misture, S. T., Althoff, J. D., Johnsson, Sanchez, J. M., Ducastelle, F., Gratias, D. (1984),
D. D. (1997), Comp. Mater. Sci. 8, 39. Physica A 128, 334.
Miyazaki, T., Isobe, K., Kozakai, T., Doi, M. (1987), Schlijper, A. G. (1983), Phys. Rev. B 27, 6841.
Acta Metall. 35, 317. Schlijper, A. G. (1984), J. Stat. Phys. 35, 285.
Mohri, T., Sanchez, J. M., de Fontaine, D. (1985), Schlijper, A. G. (1985), Phys. Rev. B 31, 609.
Acta Metall. 33, 1171. Schön, C. G., Inden, G. (2001), Comp. Mater. Sci.,
Mouritsen, O. G. (1984), Computer Studies of Phase 20, 98.
Transitions and Critical Phenomena. Berlin: Schön, C. G., Inden, G. (1998), Acta Mater. 46, 4219.
Springer. Schönfeld, B. (1999), in: Progress in Materials Sci-
Newell, G. F. (1950), Phys. Rev. 79, 876. ence, Vol. 44. New York: Pergamon Press, p. 435.
Oates, W. A., Wenzl, H. (1996), Scripta Mater. 35, Sigli, C., Sanchez, J. M. (1988), Acta Metall. 36,
623. 367.
Oates, W. A., Spencer, P. J., Fries, S. G. (1996), Sluiter, M., Turchi, P. E. A. (1989 a), in: Alloy Phase
CALPHAD 20, 481. Stability: Stocks, G. M., Gonis, A. (Eds.). Dor-
Oates, W. A., Zhang, F., Chen, S-L., Chang, Y. A. drecht: Kluwer Acad. Publ., p. 521.
(1999), Phys. Rev. B 59, 11 221. Sluiter, M., Turchi, P. E. A. (1989 b), Phys. Rev. B 40,
Ohnuma, I., Ikeda, O., Kainuma, R., Sundman, B., 215.
Ishida, K. (1998 a), presented at the CALPHAD Sluiter, M., Turchi, P. E. A., Fu, Zezhong, de Fon-
Meeting, Beijing. taine, D. (1986), Phys. Rev. Lett. 60, 716.
Ohnuma, I., Schön, C. G., Kainuma, R., Inden, G., Stoloff, N. S., Koch, C. C., Liu, C. T., Izumi, O.
Ishida, K. (1998 b), Acta Mater. 46, 2083. (Eds.) (1987), High Temperature Ordered Inter-
Onodera, H., Abe, T., Yokokawa, T. (1994), Acta metallic Alloys III, Mat. Res. Soc. Symp. Proc.,
Metall. 42, 887. Vol. 81.
Ozoliņš, V., Wolverton, C., Zunger, A. (1998 a), Taggart, G. B. (1973), J. Phys. Chem. Solids 34,
Phys. Rev. B 57, 6427. 1917.
8.9 References 581
Tepesch, P. D., Garbulski, G. D., Ceder, G. (1995), Warlimont, H. (Ed.) (1974), Order – Disorder Trans-
Phys. Rev. Lett. 74, 2274. formations in Alloys. Berlin: Springer.
Tepesch, P. D. (1996), J. Am. Ceram. Soc. 79, 2033. Wei, S.-H., Ferreira, L. G., Zunger, A. (1990), Phys.
Terakura, K., Oguchi, T., Mohri, T., Watanabe, K. Rev. B 41, 8240.
(1987), Phys. Rev. B 35, 2169. Wei, S.-H., Ferreira, L. G., Zunger, A. (1992), Phys.
Treglia, G., Ducastelle, F. (1980), J. Phys. F 10, Rev. B 45, 2533.
2137. Whang, S. H., Liu, C. T., Pope, D. P., Stiegler, J. O.
Turchi, P. E. A., Gonis, A. (Eds.) (1994), Statics and (Eds.) (1990), High Temperature Aluminides and
Dynamics of Alloy Phase Transformations: NATO Intermetallics. Warrendale: The Minerals Metals
ASI Series B, Physics, Vol. 319. Dordrecht: and Materials Soc.
Kluwer, p. 361. Williams, R. O. (1980), Metall. Trans. A 11, 247.
Turchi, P. E. A., Sluiter, M., Pinski, F. J., Johnson, D. Williams, R. O. (1984), CALPHAD 8, 1.
D. (1991a), in: Mat. Res. Soc. Symp. Proc., Vol. Wolverton, C., Zunger, A. (1997), Comp. Mater. Sci.
186: Stocks, G. M., Giamei, A. F., Pope, D. P. 8, 107.
(Eds.). Pittsburgh: Materials Res. Soc., p. 59. Wolverton, C., Ozoliņš, V., Zunger, A. (1998), Phys.
Turchi, P. E. A., Sluiter, M., Pinski, F. J., Howell, R. Rev. B 57, 4332.
H., Wachs, A. L., Fluss, M. L., Johnson, D. D., Yang, C. N. (1945), J. Phys. Chem. 13, 66.
Stocks, G. M., Nicholson, D. M., Schweika, W. Yang, C. N., Li, Y. Y. (1947), Chin. J. Phys. 7, 59.
(1991b), in: Mat. Res. Soc. Symp. Proc., Vol. 166, Zhang, J., Oates, W. A., Zhang, F., Chen, S-L., Chou,
Shapiro, S. M., Moss, S. C., Jorgensen, J. D. K-C., Chang, Y. A. (2001), Intermetallics 9, in print.
(Eds.). Pittsburgh: Materials Res. Soc., p. 231. Zorn, R., Heermann, H. J., Rebbi, C. (1981), J.
Vaks, V. G., Samolyuk, G. D. (1999), J. Exp. Theor. Comp. Phys. 23, 337.
Phys. 88, 89. Zunger, A. (1994), in: Statics and Dynamics of Alloy
van de Walle, A., Ceder, G., Waghmare, U. V. Phase Transformations: Turchi, P. E. A., Gonis, A.
(1998), Phys. Rev. Lett. 80, 4911. (Eds.), NATO ASI Series B, Physics, Vol. 319.
Wannier, G. H. (1950), Phys. Rev. 79, 357. Dordrecht: Kluwer, p. 361.
9 Diffusionless Transformations
Luc Delaey
Departement Metaalkunde en Toegepaste Materiaalkunde,

Katholieke Universiteit Leuven, Heverlee-Leuven, Belgium

9.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 587
9.2 Classification and Definitions . . . . . . . . . . . . . . . . . . . . . . . 590
9.3 General Aspects of the Transformation . . . . . . . . . . . . . . . . . 593
9.3.1 Structural Relations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 593
9.3.2 Pre-transformation State . . . . . . . . . . . . . . . . . . . . . . . . . . 597
9.3.3 Transformation Mechanisms . . . . . . . . . . . . . . . . . . . . . . . . 599
9.3.4 Microstructures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 600
9.3.5 Shape Changes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 600
9.3.6 Transformation Thermodynamics and Kinetics . . . . . . . . . . . . . . . 604
9.4 Shuffle Transformations . . . . . . . . . . . . . . . . . . . . . . . . . . 607
9.4.1 Ferroic Transformations . . . . . . . . . . . . . . . . . . . . . . . . . . . 609
9.4.2 Omega Transformations . . . . . . . . . . . . . . . . . . . . . . . . . . . 610
9.5 Dilatation-Dominant Transformations . . . . . . . . . . . . . . . . . . 610
9.6 Quasi-Martensitic Transformations . . . . . . . . . . . . . . . . . . . 611
9.7 Shear Transformations . . . . . . . . . . . . . . . . . . . . . . . . . . 613
9.8 Martensitic Transformations . . . . . . . . . . . . . . . . . . . . . . . 615
9.8.1 Crystallography of the Martensitic Transformation . . . . . . . . . . . . . 615
9.8.1.1 Shape Deformation and Habit Plane . . . . . . . . . . . . . . . . . . . . 615
9.8.1.2 Orientation Relationship . . . . . . . . . . . . . . . . . . . . . . . . . . 616
9.8.1.3 Morphology, Microstructure and Substructure . . . . . . . . . . . . . . . 618
9.8.1.4 Crystallographic Phenomenological Theory . . . . . . . . . . . . . . . . 620
9.8.1.5 Structure of the Habit Plane . . . . . . . . . . . . . . . . . . . . . . . . . 623
9.8.2 Thermodynamics and Kinetics of the Martensitic Transformation . . . . . 624
9.8.2.1 Critical Driving Force and Transformation Temperatures . . . . . . . . . 624
9.8.2.2 Nucleation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 630
9.8.2.3 Growth and Kinetics . . . . . . . . . . . . . . . . . . . . . . . . . . . . 632
9.8.2.4 Transformation Hysteresis . . . . . . . . . . . . . . . . . . . . . . . . . 634
9.9 Materials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 634
9.9.1 Metallic Materials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 634
9.9.1.1 Ferrous Alloys . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 635
9.9.1.2 Non-Ferrous Alloys . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 637
9.9.2 Non-Metals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 639
9.10 Special Properties and Applications . . . . . . . . . . . . . . . . . . . 641
9.10.1 Hardening of Steel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 641

ISBN: 3-527-30256-5
584 9 Diffusionless Transformations
9.10.2 The Shape-Memory Effect . . . . . . . . . . . . . . . . . . . . . . . . . 641

9.10.3 High Damping Capacity . . . . . . . . . . . . . . . . . . . . . . . . . . . 647
9.10.4 TRIP Effect . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 647
9.11 Recent Progress in the Understanding of Martensitic Transformations 649
9.13 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 652

a length
A factor (containing elastic terms)
a, b, c constants
Ad retransformation temperature (deformation induced)
Aj amplitude of perturbation with polarization j
As starting temperature for austenite formation
B pure strains associated with lattice correspondence
C lattice correspondence
C cubic sequence
C¢ elastic shear constant
c* size of critical nucleus
Cij eigenvalues of elasticity tensor
c/a axial ratio
c/r thickness/radius ratio of nucleus
e order parameter
E electric field
e1 , e2 , e3 principal strains
Ee strain energy
F applied force
DG difference in chemical Gibbs energy
DG* Gibbs energy of nucleation
Dgs surface Gibbs energy per unit volume
G* elastic state function
G a, G b, G g Gibbs energy of phases a, b, g
Gc , Gchem chemical Gibbs energy
Gelast elastic Gibbs energy
Gsurf surface Gibbs energy
Gtot total Gibbs energy
Gd defect Gibbs energy
Gi interaction Gibbs energy
H magnetic field
H* elastic state function
DH, DH* enthalpy change
l molar length
Md deformation-induced martensitic temperature
Mf martensite finishing temperature
Ms martensite starting temperature
P inhomogeneous lattice-invariant deformation
q wave vector
r radius of a plate
r lattice vector
R rigid body rotation
R rhombohedral sequence
r* size of critical nucleus

S strain matrix
DS entropy change
T temperature
T0 equilibrium temperature
Tc critical transition temperature
TN Neel temperature
TTW temperature at which twins appear
v volume of a plate
Vm molar volume
x atomic fraction of elements
x, y lattice vectors
a name of phase
b name of phase
g name of phase
G interfacial energy
d0 shear strain
e name of phase
e0 strain associated with the transformation
j surface to volume ratio
s stress
sa applied stress
ASM American Society for Materials

b.c.t. body-centered tetragonal
f.c.t. face-centered tetragonal
G–T Greninger – Troiano
h.c.p. hexagonal close packed
HIDAMETS high-damping metals
HP habit plane
HRTEM high-resolution transmission electron microscopy
IPS invariant plane strain
K–S Kurdjurnov – Sachs
LOM light optical microscopy
N –W Nishiyama –Wassermann
PTFE polytetrafluoroethylene
SMA shape-memory alloys
SME shape-memory effect
TRIP transformation-induced plasticity
9.1 Introduction transformations have been observed are

presented.
Diffusionless solid-state phase transfor- A martensitic transformation can be de-
mations, as the name suggests, do not re- tected by a number of techniques; some are
quire long-range diffusion during the phase in situ methods, whereas others are step-
change; only small atomic movements over by-step measurements. The results are usu-
usually less than the interatomic distances ally plotted as a change in a physical prop-
are needed. The atoms maintain their rela- erty versus temperature (see the schematic
tive relationships during the phase change. representation in Fig. 9-2), from which the
Diffusionless phase transformations there- transformation temperatures can be deter-
fore show characteristics (such as crystal- mined. Some of these plots can be trans-
lographic, thermodynamic, kinetic and mi- lated into the volume fraction of martensite
crostructural) very different from those of formed versus the temperature. Such
diffusive phase transformations. curves allow us to determine the transfor-
Martensitic transformations, because mation temperatures as indicated in these
some of the properties associated with them figures.
sometimes lead to specialized applications, In situ detection becomes limited if the
are considered to be an extreme class of transformation temperature is above room
diffusionless phase transformation and we temperature, and dilatometry then seems to
therefore in this chapter concentrate on be the most appropriate technique provided
martensite. Because of the similarity of that quenching – which is needed to avoid
some of the transformation characteristics, alterations in the sample due to diffusion –
a number of other diffusionless solid-state is possible inside the dilatometer. There are
phase transformations have sometimes far more ways of following the transforma-
been designated erroneously as marten- tion when the transformation temperature
sitic. To avoid misinterpretations, Cohen is below room temperature – preparing the
et al. (1979) proposed a classification sample and carrying out the measurements
scheme that identifies broad categoris of can take some time and at room tempera-
displacive transformations showing feature diffusion is then almost negligible.
tures in common with martensitic transfor- During slow cooling after water quench-
mations, but distinct from them. Their clas- ing, the techniques frequently used include
sification scheme, reproduced in Fig. 9-1, dilatometry, electrical resistivity and mag-
will largely be followed here. Martensitic netic measurements, calorimetry, in situ
transformations are here only a subclass of microscopy, acoustic emission, elastic and
the broader class of displacive/diffusion- internal friction measurements, positron
less phase transformations. annihilation, and Mössbauer spectroscopy.
The classification proposed by Cohen et Some of the less common techniques used
al. (1979) is discussed first, and subsequent to study martensitic transformations were
sections deal with general aspects of the reviewed by Fujita (1982).
crystallography, thermodynamics and ki- If crystallographic information is re-
netics of the different displacive transfor- quired, X-ray and electron or neutron dif-
mations. Although it is not the purpose to fraction are used. X-ray diffraction meas-
give full details of all materials that exhibit urements by Fink and Cambell in 1926 (lat-
this type of transformation, the most im- tice parameter of C-steel martensite), by
portant material systems in which such Kurdjurnov and Sachs in 1930, by Ni-
Figure 9-1. Classification scheme for the displacive/diffusionless phase transformations as proposed by Cohen
et al. (1979).
shiyama in 1934 and by Greninger and If a sample is polished into the marten-
Troiano in 1949 (orientation relationship site (= parent phase), a surface relief ap-
between austenite – the parent phase – and pears. An edge-on section of such a sample
martensite) represent breakthroughs in the is shown in Fig. 9-2 a (see Hsu, 1980). The
study of martensitic transformations. origin of the surface relief is indicated by
The most frequently used techniques the white arrows. Owing to the macro-
will now be briefly discussed and illus- scopic martensite shear (the two thinner ar-
trated. rows), a surface relief is obtained. This is
Figure 9-2. Some examples of how to see or measure the presence and growth of martensite (see text for de-
tails).
explained further in Sec. 9.8.1.1 (Fig. also the specific volume, which allows the
9-19). transformation temperatures to be deter-
The electrical resistance shows at the mined by dilatometry (Fig. 9-2 e, from
transformation temperatures a deviation Yang and Wayman, 1993).
from linearity versus temperature. This is Changes in mechanical properties are
shown in Fig. 9-2 b for the martensitic also measured while the sample is trans-
transformation in an Fe – Ni alloy and a forming. The Young’s modulus exhibits a
Au – Cd alloy. The difference in the resis- dip between the two transformation tem-
tance ratio for the two different materials is peratures Ms and Mf , as clearly visible in
obvious and remarkable (see Otsuka and Fig. 9-2 f for four different alloys (see Su-
Wayman, 1977). Measuring the electrical gimoto and Nakaniwa, 2000).
resistance while cooling or heating a sam-
ple is a very convenient and relatively easy
and accurate technique of determining the 9.2 Classification and
transformation temperatures Ms , Mf , As Definitions
and Af .
The heat exchanged with the surround- A structrual change in the solid state is
ings is becoming a more popular method of termed “displacive” if it occurs by coordi-
determining the transformation tempera- nated shifts of individual atoms or groups
tures. An example is shown in Fig. 9-2 c of atoms in organized ways relative to their
(Nakanishi et al., 1993). A DSC (differen- neighbors. In general, this type of transfor-
tial scanning calorimetry) curve allows any mation can be described as a combination
particular behavior of the martensitic sam- of “homogeneous lattice-distortive strain”
ple to be detected (for example, effects oc- and “shuffles”.
curring during heat treatments and/or def- A lattice-distortive deformation is a ho-
ormation steps). mogeneous strain that transforms one lat-
A number of martensitic transformations tice into another; examples are shown in
and materials are characterized by a so- Fig. 9-3. The homogeneous strain can be
called shape-memory effect (see Sec. represented by a matrix according to
9.11.2). Figure 9-2 d (courtesy of Memory
y = Sx (9-1)
Europe) shows the displacement of a
spring made of a NiTi shape-memory alloy. where the strain S deforms the lattice vec-
The spring controls a small valve in a cof- tor x into a lattice vector y. This strain
fee-making machine. At the temperature As is homogeneous because it transforms
the hot water starts to drop onto the coffee straight lines into other straight lines. A
powder. This valve is completely open as spherical body of the parent phase will thus
soon as the temperature Af is reached. The be transformed into another sphere or into
temperature range between As and Af seems an ellipsoidal body. The actual shape of the
to be most suitable for making the best cup ellipsoid depends on the deformation in the
of coffee. On cooling, the valve closes three principal directions. If a spherical
again. The “displacement – temperature” body is completely embedded inside the
curve measured on cooling does not coin- matrix phase and is undergoing the strain
cide with the heating curve. S, the volume and shape change associated
During a martensitic transformation, not with this deformation will cause elastic
only is the shape of the sample changed but and, sometimes, plastic strains in the parent
9.2 Classification and Definitions 591
Figure 9-3. Examples of the lattice-distortive defor-

mations of a cubic lattice: (1) a dilatation in the three
principal directions transforms the lattice into an-
other cubic lattice with larger lattice parameters; (2)
a shear along the (001) plane leads to a monoclinic
lattice, and (3) an extension along the [001] axis
combined with a contraction along the [100] and
[010] axis results in an orthorhombic lattice.
and/or product phases. The lattice-distor-

tive displacements therefore give rise to
elastic strain energy. In addition, an inter-
face separating the phases is created, gen-
erating an interfacial energy. As is obvious
from Fig. 9-1, the relative values of these
energies play an important role in the clas-
sification scheme.
A shuffle is a coordinated movement of
atoms that produces, in itself, no lattice-
distortive deformations but alters only the
symmetry or structure of the crystal; a
sphere before the transformation remains Figure 9-4. Examples of shuffle displacements in:
the same sphere after the transformation. ⶿⶿⶿ ti-
(a1) strontium titanate; 쎲 oxygen, 쑗 strontium, 쑗
Shuffle deformations produce, in the ideal tanium; (a2) the displacement of some of the oxygen
case, no strain energy and thus only interfa- atoms can be represented by an alternating clockwise
and anti-clockwise rotation around the titanium at-
cial energy. Two examples of the shuffle oms; and (b) the (111) planes in a b (b.c.c.) lattice
displacement are given in Fig. 9-4. Shuffle and the collapsed (0001) planes in the hexagonal w
deformations can be expressed by “lattice lattice (Sikka et al., 1982).
wave modulations” as On the other hand, the ellipsoid obtained

after a pure shear intersects the original
D x = Aj (q) exp (i q · r) (9-2)
sphere; hence a set of vectors exist, whose
where r is a lattice vector, q is the wave lengths remain unchanged. Such a defor-
vector giving the direction and inverse mation is said to be characterized by an
wavelength of the modulation, A is the am- “undistorted line“. An undistorted line can
plitude of the perturbation, and j denotes only result from a homogeneous lattice
the polarization of the wave. An alternative deformation if the deviatoric or shear com-
description is represented by relative dis- ponent sufficiently exceeds the dilatational
placements of the various atomic sub-lat- component.
tices that specify the structures of the two Cohen et at. (1979) thus consider a trans-
phases in terms of corresponding unit cells, formation as deviatoric-dominant if an in-
which are not necessarily primitive. variant line exists. A further subdivision
Cohen et al. (1979) subdivided the dis- was made between phase transformations
placive transformations into two main with and without an invariant line, or
groups, according to the relative contribu- between “dilatation-dominant” and “devi-
tion of the two above-mentioned atom dis- atoric-dominant” transformations.
placements and hence the ratio of the inter- If the magnitude of the lattice-distortive
facial/strain energy. In this context, they displacements is large in relation to that of
distinguish between “shuffle transforma- the lattice vibrational displacements, high
tions” and “lattice-distortive transforma- elastic strain energies are involved. How-
tions”. Since the latter give rise to elastic ever, if they are comparable, the strain
strain energy and the former only to inter- energies will be small and hence will not
facial energy, major differences are found dominate the kinetics and morphology of
in the kinetic and morphological aspects of the transformation. In the latter case we
the transformation, which justifies the dis- deal with “quasi-martensitic transforma-
tinction. Shuffle transformations are not tions”. The former, i.e., the “martensitic
necessarily pure; small distortive deforma- transformations”, are therefore those dis-
tions may additionally occur. They there- placive or diffusionless phase transforma-
fore also include those transformations in- tions where the lattice-distortive displace-
volving dilatational displacements, in addi- ments are large enough to dominate the ki-
tion to the pure shuffle displacements, pro- netics and the morphology of the transfor-
vided that they are small enough not to al- mation. “Martensite” is the name now
ter significantly the kinetics and morphol- given to the product phase resulting from a
ogy of the transformation. martensitic transformation. Because of the
The lattice-distortive transformations volume change and the strain energy in-
themselves are subdivided according to the volved with the transformation, martensitic
relative magnitudes of the two components transformation requires heterogeneous nu-
of the homogeneous lattice deformation, cleation and passes through a two-phase
i.e., the dilatational and the deviatoric mixture of parent and product; it is a first-
(shear) components (see Fig. 9-3). The in- order diffusionless phase transformation.
itial and the isotropically dilated spheres Consequently, the forward and reverse
have no intersection and it is therefore not transformations are accompanied by an ex-
possible to find a vector whose length has othermic and endothermic heat effect, re-
not been changed by the transformation. spectively, and forward and reverse trans-
9.3 General Aspects of the Transformation 593
formation paths are separated by a hystere- sive transformation as a non-martensitic

sis. composition-invariant reaction involving
Among the various diffusionless phase diffusion at the interfaces (see also the
transformations which exist in solid-to- chapter by Purdy and Bréchet (2001)).
solid phase transformations, martensitic “Bainite transformations” are also not
transformations have received much atten- treated in this chapter, although they do
tion in the past. Historically, the term “mar- show some martensitic characteristics, but
tensite” was suggested by Osmond in 1895, combined with diffusional processes. For
in honor of the well-known German metal- further information, the reader is referred
lurgist Adolph Martens, as the name for the to Aaronson and Reynolds (1988) for an in-
hard product obtained during the quench- troductory review and to Krauss (1992).
ing of carbon steels. It was found that the
transformation to martensite in steel was
associated with a number of distinctive 9.3 General Aspects
characteristic structural and microstructu- of the Transformation
ral features. During the last few decades it
was recognized that martensite also forms The various diffusionless phase transfor-
in numerous other materials, such as super- mations have a number of features in com-
conductors, non-ferrous copper-based al- mon, such as the crystallographic aspects
loys, zirconia, which recently became a of the structural changes, the pretransfor-
popular research subject for ceramists, mation state, the transformation mecha-
physicists, chemists and polymeric and bi- nisms, the microstructure and the shape
ological scientists. changes that result from the transforma-
Martensitic transformations have been tion, and the thermodynamic and kinetic
the subject of a series of international con- aspects. The more general aspects are
ferences held in various places throughout treated in the following section before dis-
the world. A list is given at the end of this cussing separately each subclass of trans-
chapter. formation.
This increasing interest not only has aca-
demic origins but can to a large extent also
9.3.1 Structural Relations
be attributed to a number of industrial ap-
plications such as maraging, TRIP (trans- This section is concerned with some
formation-induced plasticity) and dual- crystallographic aspects of the structural
phase steels, applications involving the changes. It is always useful to first deter-
shape-memory effect, the high damping ca- mine a unique relationship – a lattice corre-
pacity, and the achievement of transforma- spondence (C ) – between any vector in the
tion toughening in ceramics. initial lattice and the vector that it becomes
It may also be of interest to draw atten- in the product lattice. A lattice correspon-
tion here to the “massive phase transforma- dence thus defines a structural unit in the
tions”. Although this type of phase transparent phase that, under the action of a ho-
formation, which occurs upon fast cooling, mogeneous deformation, is transformed
is composition-invariant and the transfor- into a unit of the product phase. Such a cor-
mation interface has a relatively rapid respondence therefore tells us which vec-
movement, it does not fall into the present tors, planes and unit cells of the product
category. Massalski (1984) defines mas- phase are derived from particular vectors,
planes and cells of the parent phase, with-

out regard to their mutual orientation. For
every structural change there exist many
ways of producing a lattice correspon-
dence; the one that involves minimal
atomic displacements and which reflects
the experimentally observed orientation re-
lationships most closely should be se-
lected.
The actual relationship between labelled
vectors, planes, etc., before and after trans-
formation (including their mutual orienta-
tion) is given by the lattice deformation.
Mathematically this lattice deformation is
factorized into a pure strain and a pure
rotation, so the correspondence indicates
what is the pure strain. Knowing the princi-
pal axes of the strain ellipsoid, the direc-
tions of undistorted lines, if any exist, can
then easily be found. In 1924, Bain pro-
posed such a lattice correspondence for the
f.c.c.-to-b.c.c. (or b.c.t.) transformation in
iron alloys; it is referred to in the literature
as the prototype Bain correspondence and/
Figure 9-5. Some examples of lattice correspon-
or Bain strain. Since then, lattice corre- dence and homogeneous deformation (expansions
spondence values (C ) and their associated and contractions) for (a) f.c.c. to b.c.c. or b.c.t. (after
pure strains (B) have been published for a Bain, 1924) and (b) NaCl- to CsCl-type structures
number of other structural changes; some (the Shôji–Buerger lattice deformation) (Kriven,
examples are given in Fig. 9-5. 1982).
In the original Bain strain, a tetragonal
cell is delineated into two adjacent f.c.c.
unit cells. Then, it is contracted along z by A special situation arises for some mate-
about 20% and is expanded along x¢ and y¢ rials, when the structural change can be
by about 12%. In another example, the achieved by a pure deformation that leaves
transformation from a NaCl-type structure one of the principal directions unaltered.
into a CsCl-type structure, a contraction of Such a situation is found in some polymers.
40% along the [111] body diagonal and a It occurs, for example, in polyethylene,
19% isotropic expansion in the perpendicu- which has an orthorhombic and a mono-
lar (111) plane is needed; the volume clinic polymorph with chains parallel to the
change is about 17%. z-axis; these strong covalently bonded
Homogeneous strains alone, however, do chains are unlikely to be distorted by the
not always describe the structural transfor- transformation; consequently, the pure def-
mation completely. Additional shuffles ormation along the z-axis (e3 = 0) is zero.
may be needed to obtain the exact atom ar- During transformation the chains are dis-
rangements inside the deformed unit cell. placed transversely in such a way that one
of the remaining principal strains becomes lattice hase already been generated by the
positive (e2 > 0) and the other negative Bain strain, this additional strain should be
(e1> 0) (Bevis and Allan, 1974). a “lattice-invariant strain”. Slip and twin-
The strain ellipsoid for the above exam- ning in the product phase or in both phases
ple has a special shape. The cone of undis- are typical lattice-invariant strains; both
torted vectors of the product phase degen- deformation modes are shown schemati-
erates into a pair of planes, which rotate in cally in Fig. 9-6. The diffusionless phase
opposite directions in the pure strain. transformation can in this case be repre-
Hence either of them may be invariant if sented by an analog consisting of a pure
the total deformation of the lattice includes lattice strain (B), a rigid lattice rotation (R)
a suitable rotation. Because all the vectors and an inhomogeneous lattice-invariant
in this plane are undistorted, the transfor- deformation (P). The last factor is also as-
mation is said to be an “invariant plane sociated with a shape change, which mac-
strain (IPS)” type. The pure Bain strain is roscopically can be considered as homoge-
then equivalent to a simple shear on that in- neous. Such a combination is typical of
variant plane. Because this invariant plane martensite and is discussed in Sec. 9.8.1.4.
is also a matching plane between the ma- In cases where the lattice-invariant shear
trix and the product phase and both phases is twinning (as opposed to faulting or slip),
have to be present at the same time, a rigid type I twinning, where the twin plane orig-
body rotation (R) over an angle q is re- inates from a mirror plane in the parent
quired in order to bring the product and the phase, has been assumed. Otsuka (1986)
parent phases in contact along that plane, carefully analyzed for a number of systems
the habit plane. the possibility of a type II twinning as an
A similar situation is found in structu- alternative inhomogeneous shear. In type II
rally less sophisticated systems, namely the twinning, the shear direction stems from a
f.c.c. to h.c.p. transformation in cobalt. two-fold symmetry direction of the parent
Both phases are close packed and a simple phase. In a table, Otsuka (1986) compiled
shear on the basal plane transforms the cu- all the twinning modes observed in marten-
bic stacking into a hexagonal stacking. Be-
cause the atomic distances in the basal
plane do not change significantly during
the transformation, the plane of simple
shear is the plane of contact or the habit
plane (HP). This is the case, however, only
if there is zero volume change in the trans-
formation, i.e., in the case quoted above if
the h.c.p. phase has an ideal axial ratio of
1.633.
The situation becomes more complicated
if none of the principal strains is zero, but
of mixed sign. To achieve matching along
the plane of contact in cases where the two
Figure 9-6. Schematic representation of (a) the ho-
phases coexist, a deformation is needed ad- mogeneous lattice deformation, (b) the inhomogene-
ditional to the pure Bain strain in order to ous lattice-invariant deformation (slip and twinning),
have an invariant plane. Because the final and (c) the lattice rotation.
site and found that most of them are type I et al. (1998) carried out computer simula-
or compound but that type II twinning had tions by using the molecular dynamics
only recently been observed. According to method. The result showed that the irra-
Nishida and Li (2000), five different twin- tional thin boundary did not show any
ning modes exist in TiNi and other shape steps. Thus, the above interpretation for an
memory alloys such as Cu-Al-Ni, Cu-Sn irrational twin boundary is justified. Ot-
etc., namely the {111} type I, the {011} suka and Ren (1999) have pointed out
type I, the 〈011〉 type II, the (100)-com- again the importance of type II twinning in
pound and the (001)-compound twins. the crystallographic aspects of martensite.
Type II twinning has recently received They also stress the role that martensite ag-
much attention as a mechanism for lattice ing has on the rubber-like behaviour of
invariant shear in some alloys. Since martensite, a point that has been a long-
type II twins have irrational twin boundar- standing unsolved problem. They showed
ies, the physical meaning of an irrational that the point defects play a primordial
boundary is still a controversial problem. It role. The deformation mechanisms of the
has proposed that an irrational boundary cold deformation of NiTi martensite have
consists of rational ledges and steps, the been thoroughly analyzed by Liu et al.
average being irrational. Thereafter, Hara (1999a, b). They also found an interplay
et al. (1998) carried out a careful study to between type I and type II twinning.
observe 〈111〉 type II twin boundaries in a The cubic to tetragonal transformations,
Cu-Al-Ni alloy by HRTEM, but they could which occur in a number of metallic and
not observe any ledges or steps. The boun- non-metallic systems, need some special
dary is always associated with dark strain attention. The volume change with these
contrast, and the lattice is continuous transformations is sometimes very small or
through the irrational boundary. Nishida even absent, and the c/a ratio does not
and Li (2000) also made extensive studies change abruptly but progressively (Fig.
on 〈011〉 type II twin boundaries in TiNi by 9-7); the transformation is then said to be
HRTEM, but they did not observe ledges or “continuous”. The c/a ratio can be smaller
steps either. Based on these experimental or larger than unity, depending on compo-
results, it is thus most likely that the type II sition and temperature. The shape change
thin boundary is irrational even on a micro- associated with the transformation is small
scopic scale, and the strains at the boun- enough in many systems, especially in
dary are elastically relaxed with wide twin those belonging to the quasi-martensites,
width. To confirm this interpretation, Hara for elastic accommodation alone to be suf-
Figure 9-7. Temperature

dependence of c/a as mea-
sured during the cubic to
tetragonal transformation;
(a) second and (b) first-or-
der phase transformation.
Table 9-1. A schematic overview of some characteristics typical of the various types of diffusionless phase
transformations.
Characteristics Structural change Pure lattice deformation
Type of Principal strains Volume change

diffusionless
transformation type * order ** Sign Value
Shuffle C or D F All zero Zero Zero up to

S 10 – 5
Dilatational D F All positive or Large Large:
All negative 10 –1
Quasi-martensitic C or D F Mixed sign Small Small:
S 10 – 4 – 10 – 3
Martensitic D F Mixed sign or Large Small or large:
Zero and +, – 10 – 2 – 10 –1
* C: continuous, D: discontinuous
** F: first order, S: second order
ficient for lattice matching. It is, however, phase transformations. A soft mode is,
possible for c and a to change abruptly with in simple terms, a vibrational mode, the
zero volume change. square of whose frequency tends toward
Based on the crystallographic aspects zero as the temperature approaches that of
discussed above, a list of the most typical the phase transition. The average static
characteristics of the diffusionless phase atom displacements resemble the frozen-in
transformations can be compiled (Table pattern of the vibrational displacements of
9-1). a certain vibrational mode. According to
Vallade (1982), “the crystal lattice vibra-
tions can within the harmonic approxima-
9.3.2 Pre-transformation State
tion be separated into independent plane
Diffusionless structural changes are waves (phonons) characterized by a set
achieved by atom displacements, such as of collective atomic displacements corre-
shuffles and shears. The new atomic con- sponding to a well defined frequency. The
figuration is already prepared in some ma- energy involved is a function of the squares
terial systems at temperatures above the of the momentum and of the eigenfrequen-
transition temperature. Atoms in the parent cies of the mode. The eigenfrequencies de-
phase then become displaced more easily pend only on the mass of the atoms and on
towards their positions in the new phase the force constants. It is clear that the van-
because the restoring force that is felt by ishing of one eigenfrequency corresponds
the displaced atoms diminishes on cooling. to the lack of restoring force for the mode:
In certain cases the restoring force even the amplitude can then grow without any
vanishes at the phase transition tempera- limit and the lattice is mechanically un-
ture. stable. Stability can be recovered only
Certain shuffle transformations result by changing atomic equilibrium positions
from a vibrational instability of the parent which, in turn, changes the force con-
phase and are therefore called “softmode” stants“.
Figure 9-8. Temperature

dependence of the squared
frequency of the softening
mode for (a) a second-order
and (b) a first-order transfor-
mation, Tc and Tt being the
critical and the transforma-
tion temperatures, respec-
tively. (c) Phonon energy of
SrTiO3 measured below and
above Tc (after Rao and Rao,
1978).
As regards SrTiO3, the rotation-vibra- The tendency toward mechanical in-

tional modes of the oxygen atoms are fro- stability can also be studied through exam-
zen into the low-temperature positions be- ination of the phonon dispersion curves,
low 110 K. The temperature dependence of which gives the relationship between the
the softening, expressed by the square of wavevector q of the vibrational mode and
the eigenfrequency of the mode, is repre- its eigenfrequency. The lattice instability
sented schematically in Fig. 9-8 for a sec- can correspond to uniform (q = 0) or mod-
ond- and first-order phase transformation. ulated (q = non-zero) atom displacements
Usually, the low-temperature phase also and the soft phonon may belong to an optic
shows a soft mode as the temperature is or an acoustic branch; an example of a
raised towards Tc. measured dispersion curve is given in Fig.
Lattice softening can also be treated in 9-9. The longitudinal acoustic branch in
terms of a static approach in which the zirconium shows a dip at about 2/3 [111],
stability of the lattice is examined when which is the mode needed to transform the
submitted to small static or quasi-static ho- high-temperature b.c.c. structure of Zr into
mogeneous strains. The free energy is then the omega structure. The slope of the trans-
expressed as a function of the elastic converse acoustic branch of Nb3Sn is very flat
stants; for a lattice to be stable when sub- at the origin on approaching the transition
mitted to small homogeneous strains, the temperature of 46 K (Shapiro, 1981). This
free elastic energy must increase for all corresponds fairly well with the experi-
possible strains. For a cubic crystal this is mental observation of a vanishing value of
mathematically expressed by saying that the elastic shear constant C ¢= (C11 – C12)/2.
all the eigenvalues of the elasticity tensor The atom displacements induced by this
must be positive, in other words C44 > 0, soft mode coincide exactly with those as-
(C11 –C12) > 0, and (C11 + 2 C12) > 0. sociated with the deformation from cubic
to tetragonal structure. For Cu–Zn–Al no

soft mode is present at 2/3 [111], although
the LA branch shows an anomalous dip;
the branch measured perpendicular to it
proves that the point in the reciprocal space
is a saddle point and not a minimum. The
–
branch TA2 [110] (polarization [110]),
however, shows a small slope correspond-
ing to a low value of C ¢ (Guénin 1982).
Transformation models have been pro-
posed for Cu–Zn–Al taking into account Figure 9-9. Phonon dispersion curves for b.c.c. Zr.
both the anomalous dip and the small A pronounced dip occurs in the longitudinal (l) pho-
non dispersion curve in the vicinity of q = 2/3 [111]
slope. (Sikka et al., 1982).
In a number of materials undergoing
a cubic to tetragonal transformation, a
tweed-like pattern is observed in the parent duce a localized lattice instability (Guénin
phase by transmission electron micros- and Gobin, 1982), which may trigger the
copy. This tweed contrast is characterized nucleation of martensite on further cooling
by a ·100Ò direction of the modulation, a or stressing.
–
type of {110} ·110]Ò shear strain and a Pre-transformational lattice instabilities
modulation which is incommensurate with and soft modes and their relation to diffu-
the parent phase. It is still debated whether sionless phase transformations have been
all the pre-transformational or precursor reviewed by Delaey et al. (1979), Nakani-
effects are evidence of stable or metastable shi (1979), Vallade (1982), Nakanishi et al.
modulated phases or whether they are well- (1982), and Barsch and Krumhansl (1988).
defined artefacts determined by the kinet- The validity of the soft phonon or soft elas-
ics of nucleation and the growth process. tic stiffness approach to martensite is a dif-
A long-standing issue with b Cu-, Ag- ficult and somewhat controversial subject.
and Au-base alloys that has now been re-
solved is the appearance of extra maxima 9.3.3 Transformation Mechanisms
in the electron diffraction patterns of the
parent phase from quenched alloys. Over It should be emphasized that the pure lat-
the years these maxima have been given tice distortions considered above do not
various interpretations, often with an over- necessarily imply the actual path the atoms
emphasis as possible pre-martensitic ef- follow during the transformation. For sec-
fects. Systematic investigation, however, ond-order phase transformations, there is a
established that these effects are in fact ob- continuous change throughout the crystal
tained in martensitic structures located on with decreasing temperature starting at Tc .
the surface of the thin foil and extending Following the soft-mode concept, the ap-
inwards to a depth of 1 µm. pearance of the new phase is considered as
When considering martensitic transfor- the freezing of a particular wavelength vi-
mations, the role played by the combina- bration. The Bain-type strain for a second-
tion of lattice defects and of lattice instabil- order cubic to tetragonal transformation,
ities is of particular interest; the large defor- for example, is equivalent to two {110}
–
mations present around the defects may in- ·110Ò shear strains whose corresponding C¢
shear constant vanishes at Tc . The trans- the fully formed product. The exact mecha-
formation mechanism is therefore not a nisms for the various types of martensitic
combination of expanding and contracting transformations are still under debate.
atom movements, but a lock-in of long-
wavelength shear-type movements on
– 9.3.4 Microstructures
{110} planes in the ·110Ò directions in this
scheme. The microstructures that result from
As for the martensitic transformations, diffusionsless phase transformations show
the situation is not so straightforward. The typical features, which are related to the
Bain-type strains are concerned only with crystallograph of the transformation. The
the correspondence between initial and fi- transformation is associated with a reduc-
nal lattices; they do not give the actual ob- tion in symmetry; consequently, different
served crystal orientation relationships equivalent orientational states of the prod-
between them. Based on the experimen- uct phase are formed. A single crystal of
tally determined orientation relationships, the parent phase thus transforms to a col-
different transformation mechanisms have lection of the product-phase crystals, called
been proposed, such as shears on the planes variants, that are separated by interfaces.
and along the directions involved in the The higher the symmetry of the parent
orientation relationship. However, these phase and the lower the symmetry of the
shear mechanisms have been found to be product phase, the greater is the number of
too simple to be consistent with the experi- equivalent paths of transformation. The
mental facts. More recently, a transforma- number of equivalent orientations or vari-
tion mechanism has been proposed for ants is also determined by the symmetry
martensitic transformations of b.c.c. to elements that are maintained or broken due
close-packed structures; a condensing state to the Bain strain. The collection of vari-
of some soft phonon modes combined with ants constitutes the microstructure.
a homogeneous shear explains the variety The order of the transformation (whether
of structures that are found. For the same first or second order) also determines the
transformations, Ahlers (1974) proposed a microstructure: in the former case parent/
two-shear mechanism; the first shear product or heterophase interfaces in addi-
creates the close-packed planes, whereas tion to product/product or homophase inter-
the close-packed structure is obtained by faces are created, whereas in the latter only
the second shear. product/product interfaces are formed. In
Martensitic transformations are first- the former case the first plates formed can
order phase transformations that occur by grow to a larger extent than those formed
nucleation and growth. The growth stage later, which can lead, for example, to mi-
generally takes place by the motion of crostructures with fractal characteristics.
interfaces converting the parent phase to Fig. 9-10 shows a selection of character-
the fully formed product phase. Two types istic microstructures obtained through dif-
of paths have to be considered for the case fusionless phase transformations.
of nucleation, the “classical” and the “non-
classical” nucleation paths (Olson and Co-
9.3.5 Shape Changes
hen, 1982). The latter involves a continu-
ous change in structure whereas the former If we could transform a single crystal of
involves a nucleus of the same structure as the parent phase into a single crystal of the
Figure 9-10. A selection of

microstructures obtained by
diffusionless phase transfor-
mations: (i) schematic repre-
sentation of the microstruc-
ture of (a) martensite and
austenite, (b) the Dauphiné
twins between the low tem-
perature a- and the high
temperature b-phase of
quartz, and (c) the zig-zag
domain structure in neody-
mium pentaphosphate which
undergoes an orthorhombic
to monoclinic transforma-
tion (after James, 1988);
(ii) optical and transmission
electron micrographs of (a, b)
the twinned orthorhombic
YBa2Cu3Ox high-tempera-
ture superconductor (cour-
tesy H. Warlimont, 1989),
(c) the domain in SiO2 at the
transition temperature a to b
(846 K) (courtesy Van Ten-
deloo, 1989), and (d) the
fractal nature of the marten-
site microstructure in steel
(courtesy Hornbogen, 1989).
product phase, the macroscopically visible

shape change would clearly reflect the
Bain-type strain; it is then the maximum
transformation-induced shape change that
can be achieved. Depending on the symme-
try relationship, this deformation can be
obtained in as many orientations as product
variants exist.
In a martensitic transformation, the mac-
roscopic shape change associated with the
formation of a single martensite plate is not
only the result of the Bain strain but also of
a lattice-invariant deformation. The total
macroscopic shape change is mainly a
shear deformation along the habit plane of
the martensite variant. The martensite plate
contains either a large number of stacking
faults or has twins inside. It is therefore not
a single crystal. If the single martensite
plate has twins inside and is subjected after
the transformation to an externally applied
stress, an additional shape change is ob-
tained by detwinning. Only then is the final
product a single crystal of the product
a phase. Fig. 9-11a shows the changes in
shape after transforming a b-Cu–Zn–Al
single crystal into a single martensite vari-
ant and Fig. 9-11b shows the shape change
after partially transforming an iron whisker.
The transformed sample usually contains
a very large number of single-product do-
mains arranged in a special configuration.
In some systems the domains are arranged
such that the shape changes are mutually
accommodated. Because each product
variant is associated with a differently
oriented shape change, applying a mechan-
b
ical stress during the transformation will
Figure 9-11. Macroscopic shape change associated
with martensite: (a) a Cu–Zn–Al single crystal before
promote the formation of those variants
and after transforming to martensite and (b) a par- that accommodate the applied stress. This
tially transformed Fe whisker (courtesy Wayman, provides a resolved shape change in the di-
1989). rection of the applied stress. This is the
fundamental concept of the shape-memory
effect, as will be explained further in Sec.
9.10.2.
Figure 9-12. A series of macrographs representing the shape change while mechanically straining a Cu–Al–Ni
martensite single crystal; (a) to (e) increasing with time (Ichinose et al., 1985).
If a single crystal of the product phase is temperature, similar shape changes are ob-
mechanically strained it can either be trans- served, provided that the reverse transfor-
formed to another or be deformed to a dif- mation is also diffusionless. The situation
ferently oriented single crystal of the prod- for second-order phase transformations is
uct phase (Fig. 9-12). Similar behaviour is straightforward; the sample whose shape is
also typical of a number of ferroelastic changed during the forward transformation
materials; the reorientation is there referred and possibly after deformation below Tc re-
to as “switching” (Wadhawan, 1982). The verts back to its original shape above Tc in
switching force in these materials is not a homogeneous and continuous way. For
only a mechanical stress but can also be an first-order transformations, the reverse
electric or magnetic field, the domains be- transformation is more complex and not
ing either electrically or magnetically po- yet well understood. Much depends on
larized. whether the forward transformation is
In first-order phase transformations, as completed or not, and whether the growth
shown above, the full transformation shape of the martensite plate occurs by bursts or
change is induced locally and is gradually under thermoelastic conditions (see be-
spread over the whole sample within a low). Occurrence of the reverse transfor-
small temperature interval, whereas in a mation does not necessarily imply that the
second-order phase transformation the original shape is restored. Depending on
sample changes its shape homogeneously the crystal symmetry of the product phase,
and continuously as soon as the critical more than one crystallographically equiva-
transition temperature Tc is reached. lent path can be followed for the reverse
Until now, shape changes have been dis- transformation. The shape changes that oc-
cussed that are induced by the forward cur during the reverse transformation are at
transformation. It is evident that if a sam- the origin of the shape-memory effect and
ple of the low-temperature phase, a single are discussed in Sec. 9.10.2.
crystal or a polyvariant, is heated to tem-
peratures above the reverse transformation
9.3.6 Transformation Thermodynamics

and Kinetics
A diffusionsless phase transformation

may be of second or first order. The former
is generally dealt with in the phenomeno-
logical Landau theory, while the latter
is treated by classical thermodynamics
(Kaufman and Cohen, 1958). The Landau
theory has, however, been extended to also
cover first-order phase transformations
(the Devonshire–Ginzburg–Landau theory)
and has been applied by Falk (1982) to
martensitic transformations. The reader is
referred to the chapter by Binder (2001) for
an introduction to those theories.
The chemical driving force occupies a
key position in the classical thermodynam-
ics of first-order diffusionless phase trans-
formations, a subject that is introduced in the
first chapter of this volume (Pelton, 2001). In
the following section, therefore, only those
aspects directly relevant to diffusionless
phase transformations will be dealt with.
Because no chemical composition Figure 9-13. Schematic representation of the molar
change is associated with a diffusionless Gibbs energy (a) as a function of temperature but
phase transformation, the parent and prod- constant composition and (b) as a function of compo-
uct phases have the same homogeneous sition for an Fe–Ni alloy with T4 > T3 > T2 > T1 and
T2 = T0 for XNi = X (after Mukherjee, 1982).
chemical composition and hence they are
treated as a single-component system. For
those phase transformations whose structu-
ral change is easily described by a dis- 9-13 for first-order diffusionless phase
placement parameter, a phenomenological transformations between a parent phase,
description of the free energy as a function denoted P, and a product phase, denoted M.
of the order parameter in terms of the Lan- The product phase may be one of the low-
dau theory leads to some interesting con- temperature equilibrium phases or a meta-
clusions. In the following, the free energy stable phase.
is discussed as a function of temperature Taking again the martensitic transforma-
and composition. Other possible intensive tion as an example, the transformation
thermodynamic state variables include ex- starts at Ms , which is lower than T0 , and
ternal pressure, mechanical stress, and finishes at Mf . This means that a higher
magnetic and electrical field strength. driving force is needed to complete the
The changes in chemical Gibbs energy, transformation. On heating a fully marten-
DG, as a function of temperature and com- sitic stress-free single crystal, the reverse
position are shown schematically in Fig. transformation sets in at a temperature As ,
which is higher than T0 . The difference tributions to the Gibbs energy change,
between the forward and the reverse trans- DGnon-chem, and Eq. (9-3) then becomes
formation temperatures is the transforma-
DGtot = DGchem + DGnon-chem (9-4)
tion hysteresis. The true first-order equilib-
rium temperature, T0 , which is calculated The transformation then proceeds until
from DG = 0, can thus only be bracketed DGtot becomes minimum or, if the phase
from experimental data for the forward and boundary is mobile, until the total force at
the reverse transformation temperatures, the parent-to-product interface is zero. If
and is not necessarily halfway between Ms the advancing direction of the interface is
and As. x, we can write
Strain energy resulting from the transfor-
[∂ (DGtot)/∂x] dx = 0 (9-5)
mational shape change and interfacial en-
ergy have been omitted from the free-en- or
ergy curves in Fig. 9-13. These two non- [∂DGchem /∂x] dx + [∂DGnon-chem /∂x] dx = 0
chemical-energy terms have to be consid-
ered, however, in the overall free-energy The sum of the non-chemical restoring
balance. The strain energy associated with forces is then identical with the chemical
the formation of a single domain of the driving forces. The difficult task now is
product phase is proportional to the volume to find expressions representing the non-
of that domain. The interfacial energy is chemical terms. Three thermodynamic ap-
not directly related to the volume of the proaches have been worked out, dealing es-
transformed domain but merely to its sur- sentially with the influence of the two non-
face-to-volume ratio, and, in the case of an chemical contributions on the transforma-
anisotropic interfacial energy, also to the tion behavior (Roitburd, 1988; Ball and
orientation of the interface. Both quantities James, 1988; Shibata and Ono, 1975, 1977).
are positive and thus consume part of the According to Roitburd (1988), the strain
chemical driving force for a forward trans- energy, which arises in crystals owing to a
formation. Both terms will, however, in- diffusionless phase transformation, can de-
crease the driving force for the reverse crease if the crystals are subdivided into
transformation, provided that the inter- domains arranged such that a maximum
facial coherence is not lost. The reverse compensation of the individual strain fields
transformation might start below T0 if a is achieved. In order to determine which ar-
negligible net driving force is required for rangements are energetically most favor-
nucleation. able, Roitburd calculates the strain energy
Considering the Gibbs energy per unit for arbitrary domain arrangements, and
molar volume, the total Gibbs energy then minimizes this energy. The formula-
change per unit molar volume for the for- tion of this problem is complex and can
mation of a single domain of the product hardly be solved in general, but he suc-
phase embedded in the matrix phase is ceeded for some specific cases.
given by Ball and James (1988) do not assume
any geometric restrictions on the shape or
DGtot = DGchem + (DGelast + j DGsurf) (9-3)
arrangements of the domains; they found
where j is the surface-to-volume ratio this necessary to determine microstructures
of the single domain. The two terms in occurring in complex stress fields, or to
parentheses are then the non-chemical con- explore new and unusual domain arrange-
ments. The general aim of their work was Shibata and One (1975, 1977) use the
to develop mathematical models, using cal- Eshelby theory; the principle of their calcu-
culus of variations, capable of predicting lation is in a corrected version (Christian,
the microstructure, especially the micro- 1976) illustrated schematically in Fig.
structural details at the interface between 9-14. An embedded part of the parent
the parent and the product phases. At- phase is cut out (step a) and is allowed to
tempts have been made to predict the pos- transform stress-free into the product phase
sible interfaces between austenite and mar- (step b). A lattice-invariant deformation is
tensite from a minimization of a Gibbs en- applied (step c) and the transformed crystal
ergy function, which depends on the defor- is subjected to forces along its surface such
mation gradients of all possible domain that it is deformed to the original shape
variants and on temperature. A deforma- (step d). The thus deformed part of the
tion or domain is then termed stable if product phase is introduced in the empty
it minimizes the total energy. They show, space of the parent phase (step e), and the
among others, that a martensite–austenite forces are removed, creating internal
interface can exist as an energy-minimiz- stresses in both the product and the parent
ing sequence of very fine twins. A further phase. The total energy is then calculated
example of an intriguing application is as a function of all possible lattice orienta-
the formation of triangular Dauphiné twins tions, taking into account the actual elastic
in quartz, which become finer and finer in constants and the modes of lattice-invari-
the direction of increasing temperature. A ant deformation, twinning or slip.
Gibbs energy function accounting for this The total Gibbs energy of the system is
behaviour could be constructed. therefore a function not only of the intrin-
sic energies of the stress- and defect-free
parent and product phases, but also of the
arrangement of the domains. The non-
chemical component of the total Gibbs en-
ergy of the transforming system is lowered
by an appropriate rearrangement of the mi-
crostructure and/or by irreversible plastic
deformation.
If the structural change can be repre-
sented by an order parameter e, the Gibbs
energy of the system can then, according to
the theory of Landau–Devonshire, be rep-
resented by
G = G0 + a (T – T1) e2 – B e4 + Ce6 (9-6)
where a, B and C are constants and T1 > 0.
It can be shown that the high-temperature
phase becomes unstable with respect to
any fluctuation of e, as soon as the temper-
Figure 9-14. The necessary steps in calculating the ature reaches T1 on cooling, and hence is
elastic stresses induced by a transforming ellipsoid thermodynamically metastable between T0
(Christian, 1976). (For details see text.) and T1. Accordingly, the low-temperature
9.4 Shuffle Transformations 607
phase cannot exist at temperatures higher The most relevant thermodynamic data
than T2, which is the temperature above for the various diffusionless phase transfor-
which the low-temperature phase becomes mations are presented in Table 9-2.
unstable with respect to any fluctuation in e.
That additional undercooling is needed
for further transformation below Ms is due 9.4 Shuffle Transformations
(in part) to the non-chemical contributions,
which increase with increasing volume Shuffle transformations from a distinct
fraction of transformed product. class of diffusionless phase transitions. At
An interesting aspect of the diffusionless the unit-cell level the atom displacements
phase transformations that are accompa- are intercellular with little or no pure strain
nied by a volume and shape change is the of the lattice. The role of elastic strain en-
role played by external stresses, e.g., hy- ergy in shuffle-phase transformations is
drostatic or uniaxial. Both thermodynamics sufficiently small that the transformation
and experiments show that the transforma- can either occur continuously from the par-
tion temperatures are affected by the appli- ent to the product phase or that it is com-
cation of stresses. According to Wollants et pletely controlled by interfacial energy. In
al. (1979), the relationship between a uni- the former case the transformation is sec-
axially applied stress s and the transfor- ond order whereas in the latter it is a first-
mation temperature T depends on the order phase transformation.
transformation entropy and the transfor- Cohen et al. (1979) gave three examples
mational strain in the direction of the ap- which clearly illustrate the shuffle transfor-
plied stress. This relationship, the Clau- mations. The displacive transformation in
sius–Clapeyron equation for uniaxially strontium titanate is the prototype example
stressed diffusionless first-order phase of a pure shuffle transformation. The asso-
transformations, is ciated strain energy is so small that the
transformation occurs continuously. The
ds /dT = – D S/e = – DH*/[T0 (s ) e] (9-7)
b-to-w transformation in some Ti and Zr
where DH* = DH – FDl = DH – s e Vm = alloys shows, in addition to the shuffle
T0 (s ) D S is itself a function of the applied displacements, small homogeneous lattice
load, e = Dl/l, l is the total “molar length” distortions. These distortions are small
of the sample, and F is the applied load enough for the transformation mechanism
(s = F/A). This equation is similar to that and the resulting microstructure to be dom-
relating the equilibrium temperature to the inated only by shuffling. In ferroelectric
hydrostatic pressure, except for the nega- transformations, which are accomplished
tive sign on the right-hand side of Eq. (9-7). by shuffling, the interfacial energy is con-
This relationship between ds and dT is ex- stituted largely by electrostatic interaction
perimentally constant for most of the diffu- energies and is therefore dependent on the
sionless transformations, which means that orientation of the domain interfaces. The
the thermodynamic quantity DS is, within interfacial energy in those materials is
the experimental scatter, independent of strongly anisotropic and controls the poly-
temperature and stress. Knowing the trans- domain structure.
formation strain, uniaxial tensile tests are Phase transformations that can be en-
very useful for determining the transforma- tirely described by shuffle diplacements
tion entropy. are often found where the change in crystal
Table 9-2. The elastic shear constant and some thermodynamic data characterizing the diffusionless phase
transformation (Delaey et al., 1982 b).
Elastic shear constant C ¢ Trans- Thermodynamic quantities

near the Ms temperature formation
strain Heat of Change Chemical Transfor-
C¢ (1/C ¢) (dC ¢/dT ) transforma- in driving mation
(1010 Pa) (10–4 K–1) tion entropy force tempera-
(J/mol) (J/(K · mol)) (J/mol) ture hys-
teresis (K)
Ferrous g Æ a¢ 2–3 negative ≈ 10–1 ≈ 2000 – 300 5.8 150 – 450 200 – 400
gÆe 3–10 (positive 600 – 1800
for Ni > 30%)
Co alloys negative ≈ 10–3 ≈ 400 – 500 ≈ 0.2 4 – 16 40 – 80

rare-earth
alloys
Ti and Zr 0.1 negative ≈ 2 ¥ 10–2 ≈ 4000 ≈ 1.0 ≈ 25 –

alloys
b Cu–Ag–Au 0.5 – 1 4 – 20 ≈ 10–2 ≈ 160 – 800 0.2 – 3.0 ≈ 8 – 20 10 – 50

alloys (positive)
In alloys 0.05–01 ≈ 1000 ≈ 10–3 ≈0 – ≈ 1.5 1 – 10

(positive)
Mn alloys positive ≈ 10–3
(strongly)
A 15 com- 0.5 ≈ 1000 – 3000 ≈ 10–4
pounds (positive)
Fe–Pt ≈ 320 – ≈ 16 ≈ 20 – 200

(ordered) (ordered)
Fe–Pd ≈1 ≈100 ≈10–3 ≈1200 ≈1200
alloys (positive) –10–2 (disordered) (disordered)
structure is such that the point group to placement. For strontium titanate, the order
which the crystal structure of the product parameter would then simply be the rota-
phase belongs is a subgroup of that of the tion angle that describes the displacement
parent phase. In other words, some symme- of the oxygen atoms around the titanium
try elements of the high-temperature phase atoms (see Fig. 9-4). For convenience, the
are lost on cooling below the transition order parameter is taken as zero for the
temperature Tc . Because of this group/sub- high-temperature configuration and as non-
group relationship, the product phase pos- zero for the low-temperature phase. The
sesses two or more equally stable orienta- majority of such transitions are found in
tional states in the absence of any external chemical compounds (e.g., Rao and Rao,
field. The change in crystal structure can 1978). As soon as the critical temperature
easily be described by an order parameter Tc is reached on cooling, the order parame-
which itself is related to the shuffle dis- ter changes continuously. The thermody-
9.4 Shuffle Transformations 609
Figure 9-15. Temperature de-

pendences of the lattice
parameters of the different
phases of BaTiO3 .
namics of such transformations are then in point-group symmetry are called ferroic
the temperature range close to Tc , which is phase transitions. We refer to a crystal as
dealt with by a Landau approach. being in a ferroic phase if that phase results
from a symmetry-lowering ferroic phase
transition”.
9.4.1 Ferroic Transformations
Not all ferroelastic phase transitions be-
Usually a phase transition that is domi- long to shuffle transformations as defined
nated by shuffling is associated with a in Fig. 9-1. Indeed, in addition to shuffle
change in some physical properties, such as displacements, as for example those in-
spontaneous electrical polarization, strain volved in the cubic to tetragonal transition
and magnetization. Because the crystal in barium titanate, the lattice may become
symmetry of the parent phase decreases homogeneously distorted. For the example
during the phase transition, two or more considered here, the lattice distortion oc-
equivalent configurations of the product curs discontinuously at the transition tem-
phase are formed. In the absence of any ex- perature; the lattice parameters change
ternal field, the average polarization of the abruptly (Fig. 9-15). Even below this tran-
product phase is zero. However, under a sition temperature, the lattice continues to
suitably chosen driving force, which may be homogeneously distorted. In cases
be an electrical field (E ), a mechanical where this lattice is tetragonal, the c/a ratio
stress (s), or a magnetic field (H), the do- steadily increases. For barium titanate the
main walls of the product phase move, change in c/a continues until the tempera-
switching the crystal from one domain or- ture for another first-order phase transition
ientation to the other. Owing to the applica- is reached. Many such phase transforma-
tion of a uniaxial stress, for example, one tions are encountered in chemical com-
orientation state can be trensformed repro- pounds. In some cases the amount of spon-
ducibly into the other, and the crystal is taneous strain is not large enough to con-
then said to be “ferroelastic”. The materials trol the microstructure. In others, the strain
exhibiting this property are called ferro- energy associated with the transformation
elastic materials. Similarly, we can define will be dominant. The transformation is
ferroelectric and ferromagnetic materials. then, according to Fig. 9-1, quasi-marten-
According to Wadhawan (1982), “phase sitic or martensitic. In ferroelectric materi-
transitions accompanied by a change of the als, the interfacial energy also has to be
leaving the two adjacent planes unaltered

(Fig. 9-4 b). The collapse can be repre-
sented as a short-wavelength displacement
of atoms. The displacement of the atoms
occurs over a distance approximately equal
to 2/3 ·111Ò. Each lattice site can thus be
associated with a forward, zero or back-
ward displacement that can be represented
by a sinusoidal wave dividing the repeat
distance along a [111] direction into six
parts. The collapse is not always complete,
and then results in a “rumpled” plane. If the
collapse is incomplete the crystal structure
of the w-phase is trigonal; if the collapse is
complete, it is hexagonal. The figure also
shows that reversing the direction of the dis-
Figure 9-16. Gibbs energy change for the b.c.c. to w
placement will not lead to a collapse of the
transformation as a function of the order parameter {111} planes. Moreover, it can be shown
for various reduced temperatures (after de Fontaine, that a 2/3 [111] displacement wave is equiv-
1973).
–
alent to a 1/3 [112] displacement wave.
If the displacement is taken as the order
taken into account and may even become parameter in a Landau-type approach, the
the dominant parameter controlling the transformation is seen to be first order and
microstructure. A ferroic ferroelastic phase the Gibbs energy as a function of this order
transformation can thus be a shuffle, a parameter has an asymmetric shape. Con-
quasi-martensitic, or a martensitic phase sequently, no negative values of the order
transformation, a clear discrimination is parameter are then allowed (Fig. 9-16).
only possible by analyzing all the transfor- The b-to-w transformation can also be par-
mation characteristics, and this not after aphrased in terms of a soft mode. The lat-
but during the transformation. tice tends to a mechanical instability for a
2/3 ·111Ò longitudinal mode. This tendency
can be shown when measuring the phonon
9.4.2 Omega Transformations
dispersion curves by inelastic neutron scat-
The omega transformation is known to tering. Such curves are reproduced in Fig.
occur as a metastable hexagonal or trigonal 9-9 a for zirconium; a clear dip is visible at
phase in certain Ti, Zr and Hf alloys on the 2/3 [111] position.
cooling from the high-temperature b.c.c b- The omega transformation has been re-
phase solid solution or as a stable phase viewed by Sikka et al. (1982).
under the influence of high hydrostatic
pressures or shock waves. The w-phase
cannot be suppressed by quenching and 9.5 Dilatation-Dominant
forms as small cuboidal or ellipsoidal par- Transformations
ticles with a diameter of 10–20 nm. Its lat-
tice is obtained by collapse of one pair of A transformation is regarded by Cohen
(111) planes of the parent b.c.c. b-phase, et al. (1979) as dilatation dominant if no
9.6 Quasi-Martensitic Transformations 611
undistorted line can be found in the lattice- are very similar (large plates, occurrence of
distortive deformation. The f.c.c.-to-f.c.c.¢ variants and twins). A distinction between
transformation in cerium is considered as the two transformations cannot be made by
the prototype for a dilatational dominant simply judging only the product morphol-
transformation. Below 100 K cerium un- ogy, but rather a knowledge is required of
dergoes a pure volume contraction of about the morphological relationships between
16%; the ellipsoid of the f.c.c.¢ phase thus parent and product phases during the trans-
falls completely inside that of the high- formation itself. It may be adequate to say
temperature f.c.c. phase. The low-tempera- first what a quasi-martensitic transforma-
ture cubic to tetragonal transformation in tion is: a quasi-martensitic transformation
tin also appears to be dominated by dilata- is not a martensitic transformation, which
tion, although some deviatoric components itself is “a first-order phase transformation,
are present; the volume expansion is about that undergoes nucleation, passes through a
27%. The deviatoric component is not two-phase mixture of the parent and prod-
large enough to let the original sphere uct phases, and which product grows with a
intersect with the dilated ellipsoid. transformation front in a plate-like or lath-
The name “dilatational diffusionless like shape being indicative of a tendency
phase transformation” has been used by toward an invariant-plane interface” (Co-
Buerger (1951) but with a different mean- hen et al., 1979). If a deviatoric dominant
ing. In the systems he considers, for exam- transformation does not satisfy the above
ple the CsCl-to-NaCl transitions in many criterion, it should not be designated as
alkali metal halides, he defines the term martensitic but as quasi-martensitic.
dilatational as follows: “the transformation Three aspects are common to most of the
can be achieved by a differential dilatation materials that transform quasi-martensiti-
in which the structure expands along the cally: (1) the lattice distortion is small and
trigonal axis and contracts at right-angles deviatoric dominant and the change in lat-
to the axis”. Although the volume change tice distortion is continuous or nearly con-
in these and other related inorganic sys- tinuous; (2) a banded internally twinned
tems may be very large (up to 17%), the microstructure gradually builds up on cool-
transformation is, in the context of Fig. 9-1, ing below Tc ; and (3) a mechanical lattice
clearly not dilatation dominant but devi- softening is expressed by elastic shear
atoric dominant. See Kriven (1982) for a constants approaching zero as Tc is ap-
more detailed review of these dilatational proached. Because of the small lattice dis-
dominant transformations. tortion at the transformation, the ratio of
the strain energy to the driving energy for
transformation is small; this ratio has been
used by Cohen et al. (1979) as an alterna-
9.6 Quasi-Martensitic tive index to differentiate quasi-martensitic
Transformations transformations from martensitic.
The three aspects are now illustrated by
The quasi-martensitic and the marten- taking the manganese-based magnetostric-
sitic transformation are both deviatoric tive antiferromagnetic alloys as an example
dominant and are characterized by an un- (see Delaey et al., 1982a). One of the four
distorted line. The morphologies of the polymorphic states of manganese is the
product phases of the two transformations gamma f.c.c. phase which is stable only at
high temperatures. Alloying with elements schematic representation of the phase dia-
such as Cu, Ni, Fe, Ge, Pd and Au stabi- gram of the Mn-based alloys is given in
lizes the f.c.c phase and the latter can be re- Fig. 9-17 a, showing the temperature–com-
tained by quenching. However, owing to position areas in which the various crystal
the antiferromagnetic ordering, the lattices structures and microstructures are ob-
become homogeneously distorted. This or- served. The accompanying variation in the
dering to the Mn atoms starts at a temper- lattice parameters as a function of tempera-
ature TN , which is the Néel temperature ture for the three classes of Mn-based al-
for the paramagnetic to antiferromagnetic loys is given in Fig. 9-17 b.
transition. The transformed product phase The changes in lattice parameters show
has a banded microstructure containing that the transformation is almost second or-
fine twins. The temperature at which this der, except for some alloys of class I and III
banded microstructure is formed does not where the transformation is weakly first
always coincide with the transformation order. A phase transformation is called
temperature TN. Vintaikin et al. (1979) di- “weakly first order” whenever the height of
vide these antiferromagnetic alloys into the discontinuous jump in the correspond-
three classes according to the relative posi- ing thermodynamic property is very small.
tions of the temperature TN and the temper- The formation of the twinned banded mi-
ature TTW. The latter is the temperature at crostructure extends over the entire volume
which the banded microstructure sets in. of the sample quasi-instantaneously and is
Depending on the type of lattice distortion, visible in polarized light because of the
the alloys are grouped into three classes, non-cubic structure of the product phase.
each class being characterized by the rela- Similar microstructures are observed in
tive positions of the two temperatures. A other quasi-martensitic product phases
such as V54 – x Ru46Osx (Oota and Müller,
1987). The microstructure, if properly
oriented with respect to the prepolished
surface, exhibits a surface relief effect that
is enhanced as the temperature decreases
below TTW. This surface relief proves that
the transformation is accompanied by a
shape change associated with each domain.
Because of the continuously changing lat-
tice parameters, accommodation stresses
are built up as the temperature decreases.
An appropriate arrangement of these do-
mains reduces the overall stored elastic en-
ergy; further changes in microstructure are
therefore expected to occur even below the
transition temperature.
Class II alloys do not exhibit the twinned
banded microstructure immediately below
Figure 9-17. Schematic representation of (a) the
phase diagram and (b) the variation of the lattice pa-
TN. In the temperature region between TN
rameters for the three classes of Mn-based alloys and TTW, broadening of some of the X-ray
(Delaey et al., 1982 a). diffraction peaks is observed, which is
9.7 Shear Transformations 613
attributable to a chaotic distribution of the gradually disappear and the product phase
a and c axes with small undercooling. At becomes a single domain maximizing the
TTW the banded structure becomes visible shape change. The shape change thus ob-
(point A in Fig. 9-17) and a tetragonal tained is gradually recovered on heating
structure can now be clearly detected by X- the sample and is completely recovered at
ray diffraction. If the sample is now heated, TN and not (as in the case of Mn-based
only the banded microstructure disappears, alloys of class II) at TTW, but at the point B
not at A but at a temperature B that coin- in Fig. 9-17. The quasi-martensitic alloys
cides with the Néel temperature. This thus also exhibit the shape-memory effect.
proves that on cooling, very small, submi- Some of the materials characterized by
croscopic tetragonal regions are first shuffle displacements during the phase
formed as soon as TN is reached. Hocke and transition may develop elastic strains as
Warlimont (1977) have shown that when transformation proceeds. As in ferroelec-
the distortion |c/a – 1| becomes greater trics, for example, in addition to the elastic
than 0.005, a critical value is obtained at strain energy, the dipole interaction energy
which the elastic strain is relaxed through also contributes to the polydomain forma-
coalescence of the small distorted regions tion. If the elastic strain energies are only
into large banded twinned regions. Thus, at slightly dominating, the transformation is
TTW there is not a phase transformation but quasi-martensitic; if, however, the elastic
a stress relaxation in the microstructure, strain energy is largely dominating, the
which results in a twinned microstructure. transformation can be martensitic.
The lattice distortive phase transformation Sometimes it becomes difficult to differ-
itself occurs at TN, followed immediately entiate between martensite and quasi-mar-
(class I) or after some undercooling (class tensite, as for example in In-based alloys.
II) by a domain rearrangement and macro- In particular, if quasi-martensitic samples
scopic twinning. are cooled in such a way that a temperature
Similar conclusions can be drawn for gradient is created across the sample, the
other quasi-martensitic transformations, as product phase and the parent phase then
for example in the iron – palladium alloys; coexist and apparently the transformation
the Pd-rich f.c.c. phase transforms on cool- goes through a two-phase region, the two
ing first to an f.c.t phase and at lower tem- regions being separated by a blurred or
peratures to a b.c.t. phase. The f.c.c.-to- planar interface. Such observations do not,
f.c.t. transformation, although sometimes of course, facilitate the distinction between
regarded as martensitic, shows all the char- quasi-martensite and true martensite.
acteristics of a quasi-martensitic transfor-
mation.
Because the formation of each single do- 9.7 Shear Transformations
main is associated with a shape change and
thus with accommodation stresses, the ap- In this section we discuss a special group
plication of an external stress to the trans- of phase transformations, the so-called
formed product will result in a macro- shear transformations or polytypic transi-
scopic shape change. As the domain boun- tions, which strictly belong to the marten-
daries, which for the Mn-based alloys coin- sitic transformations. According to Verma
cide with the antiferromagnetic boundar- and Krishna (1966), “polytypism may be
ies, are mobile, the banded structure will defined, in general, as the ability of a sub-
stance to crystallize into a number of dif-

ferent modifications, in all of which two
dimensions of the unit cell are the same
while the third is a variable integral multi-
ple of a common unit. The different poly-
typic modifications can be regarded as
built-up of atom layers stacked parallel to
each other at constant intervals along the
Figure 9-18. Mechanism of the f.c.c.-to-h.c.p. trans-
variable dimension. The two unit-cell di- formation (Nishiyama, 1978).
mensions parallel to these layers are the
same for all the modifications. The third
dimension depends on the stacking se- placements or the movement of partial dis-
quence, but is always an integral multiple loctions along the basal plane. For exam-
of the layer spacing. Different manners of ple, the transition between a 2 H and a 3 C
stacking these layers may result in struc- stacking is easily performed by a shear,
tures having not only different morpholo- whereby the basic units are kept together in
gies but even different lattice types and pairs (Fig. 9-18). This shear results in a
space groups”. Some random disorder of large deviatoric shape change. It should be
layers (faulted sequences) is almost always kept in mind that the interlayer spacing
present. Polytypic transitions are then tran- need not be constant, as is observed in the
sitions among different polytypes; the f.c.c.–h.c.p. changes in metals; the trans-
movement of partial dislocations along the formation may involve small changes and
basal plane constitutes the transition mech- thus be IPS (invariant plane strain) rather
anism, thereafter the name shear trans- than simple shear transitions.
formations. Polytypic transformations are The same transitions can be achieved by
found in a variety of inorganic compounds the generation and movement of closely
and also in metals and alloys. spaced and repeatedly arranged partial dis-
Polytypic phases are constructed by locations. The passage of a positive partial
stacking basic units in a cubic, hexago- dislocation shifts the crystal in the direc-
nal or rhombohedral sequence. The stack- tion X Æ Y Æ Z Æ X and negative partial
ing sequences are described by three key dislocations shifts the crystal in the direc-
layer positions, X, Y and Z; a cubic se- tion X Æ Z Æ Y Æ X. The following dis-
quence (C) is represented by the sequence tribution of partial dislocations on the unit
XYZXYZ …, a hexagonal (H) by, for layers in the direction perpendicular to the
example, XYXZ … or XYXZXYXZ …, layers is proposed by Liao and Allen
and a rhombohedral (R) by, for example, (1982) (a layer without a partial dislocation
XYZYZXZXYXYZYZXZXY … . Many is denoted by a dot):
other stacking variants are possible and · – + · for 2 H Æ 4 H
unit cells containing as many as 126 or 144
· – – – · · for 6 H Æ 3 C
layers have been reported. Each unit itself
can contain a single layer, as in cobalt and The polytypic shear transformation from
its alloys, or two as in silicon carbide. one modification to the other is thus ac-
Transitions between different modifica- complished by a coordinated propagation
tions can be achieved either by a simple of groups of partials along the interface
shear, a shear combined with shuffle dis- between the two phases. The lateral dis-
9.8 Martensitic Transformations 615
structure, specific morphology or proper-

ties), but rather on how it forms.
The crystallographic and thermody-
namic aspects are fully discussed in the lit-
erature and have already been introduced
in a more general context in the above sec-
tions; only a brief overview is given here.
9.8.1 Crystallography of the Martensitic

Transformation
Figure 9-19. Two different mechanisms, (a) and (b),
to transform an h.c.p. to an f.c.c. structure (Bidaux, The relevant experimental observable
1988). parameters of the martensitic transforma-
tion are the shape deformation, the habit
plane, the crystallographic orientation rela-
placement of the interface, which is the tionships, and the characteristic micro-
thickening of the new phase generated by structures.
the movement of those partials, is then due
to the formation of partial dislocations and 9.8.1.1 Shape Deformation
their outward movement. During the tran- and Habit Plane
sition from a cubic to a hexagonal 2 H se-
quence, it has been implicitly assumed When a sample of the parent phase is
above that the glide of the layers would al- cooled to below Ms , a relief gradually ap-
ways occur in the same direction. There ex- pears on a prepolished surface of the par-
ist, however, three different directions for ent-phase crystal. The surface relief disap-
transforming an X stacked unit into a Z pears on heating to temperatures above As ,
stacked unit. If the glide occurs alternately provided that no diffusion-controlled trans-
in these three directions, no shape change formation interferes.
results from such a mechanism, as shown The martensite phase usually takes the
in Fig. 9-19 (Bidaux, 1988). form of plates; the plane of contact be-
tween the parent and the martensite phases
is called the “habit plane”. A schematic
9.8 Martensitic Transformations representation of such a martensite plate
embedded in the matrix is shown in Fig.
The characteristics necessary and suffi- 9-20. During the formation of martensite,
cient for defining a martensitic transforma- straight lines (for example, scratches on the
tion are (a) displaciveness of the lattice- prepolished surface) are transformed into
distortive type involving a shear-dominant other straight lines and planes are trans-
shape change, (b) diffusion not required for formed into other planes. No discontinu-
the transformation, and (c) sufficiently ities are observed at the points of deflec-
high shear-strain energy in the process to tion. This distortion can thus be repre-
dominate the kinetics and morphology dur- sented as a “linear homogeneous transfor-
ing the transformation (Cohen, 1982). The mation” of vectors and can be expressed by
definition is thus not based on the identity a matrix formulation. The macroscopic
of the transformation product itself (its shape deformation can be decomposed into
with the formation of martensite is thus an

“invariant plane strain” deformation, ab-
breviated to IPS. The most general invari-
ant plane strain deformation, as observed
in most martensitic transformations, can be
achieved by combining an extension and a
simple shear.
The habit plane and the direction of mac-
roscopic shear are, with few exceptions,
not simple low-indexed crystallographic
planes or directions of the parent or prod-
uct phase. They are usually represented in
a stereographic projection as shown sche-
matically in Fig. 9-20.
9.8.1.2 Orientation Relationship
The next most important observable pa-

rameter is the crystallographically well-de-
fined “orientation relationship” that exists
between the lattices of the parent and the
martensite phases. It is described either by
the angles between certain crystallographic
directions in both phases or by specifying
the parallelism between certain planes and
directions. This parallelism does not need
Figure 9-20. Schematic representation of (a) a sin- to be rigorous, however, experimental re-
gle martensite plate embedded in a single crystal of sults usually deviate slightly. Nevertheless,
the matrix phase, (b) a twinned plate and the position
of habit and twin plane, and (c) their stereographic
the fact that such crystallographic parallel-
representation. ism is observed yields important informa-
tion concerning the possible mechanisms
explaining the change in crystal structure.
a component normal to the habit plane and Some of those relationships observed in
a shear component parallel to a shear direc- steels received great attention in the early
tion located in this interface. The latter is martensite literature. Depending on the
called “macroscopic shear” and quantifies alloy composition, the f.c.c. austenite in
the shape deformation, whereas the former steels transforms either to a b.c.c. or b.c.t.
represents the volume change associated martensite or to an h.c.p. martensite, which
with the transformation. A careful analysis itself may further transform into b.c.c. mar-
of the surface relief reveals that the habit tensite. As regards the transformation of
plane itself is unrotated and that any vector f.c.c. austenite to b.c.c. or b.c.t. martensite,
in this interface is also left unrotated and the orientation relationships are as follows:
undistorted by the shape change. The habit
plane is thus essentially “undistorted” and – the Kurdjumov – Sachs (K – S) relations:
– –
the macroscopic shape change associated (111)P //(011)a¢ and [011]P //(111)a¢
Table 9-3. The crystallographic observables of the martensitic transformations in some metals and alloys
(courtesy G. Guénin et al. 1979*).
Alloy system Structural change Composition wt.% Orientation relationship Habit plane
Fe–C f.c.c. 0–0.4% C (111)PΩΩ(101)M (111)P

Ø [11̄0]PΩΩ[11̄1]M
b.c.tetr. K–S relationship
0.55–1.4% C K–S relationship (225)P
1.4–1.8% C Idem
Fe–Ni f.c.c. 27–34% Ni (111)PΩΩ[101]M

Ø [12̄1]PΩΩ[101̄]M ≈ (259)P
b.c.c. N-relationship
Fe–C–Ni f.c.c. 0.8% C–22% Ni (111)P ≈ 1° of (101)M (3, 10, 15)P

Ø [12̄1]P ≈ 2° of [101̄]M
b.c.tetr. G–T relationship
Fe–Mn f.c.c. 13 to 25% Mn (111)PΩΩ(0001)e (111)P

Ø [11̄0]PΩΩ[12̄10]e
h.c.p. (e-phase)
Fe-Cr-Ni f.c.c. 18% Cr, 8% Ni (111)PΩΩ(0001)eΩΩ(101)a¢ e (111)P
Ø [11̄0]PΩΩ[12̄10]eΩΩ[11̄1]a¢ a¢ (111)P
h.c.p. (e), b.c.c. (a¢)
––
Cu–Zn b b.c.c.Æ 9 R 40% Zn (011)PΩΩ?(114)M ≈ (2, 11, 12)P
Cu–Sn idem 25.6% SN [11̄1]PΩΩ[1̄10]M ≈ (133)P
Cu–Al b.c.c. 11.0 to 13.1% Al (101)P at 4° of (0001)M 2° of (133)P

Ø [111]PΩΩ[101̄0]M
h.c.p. distorted 12.9 to 14.7% Al (101̄)PΩΩ(101̄1)M 3° of (122)P
[111]PΩΩ[101̄0]M
Pure Co f.c.c. (111)PΩΩ(0001)M (111)P

Ø ·110ÒPΩΩ[112̄0]M
h.c.p.
Pure Zr b.c.c. (101)PΩΩ(0001)M (596)P

Ø [111]PΩΩ[112̄0]M (8, 12, 9)P
Pure Ti h.c.p. (334)P
(441)P
Pure Li Burgers relations
* Gobin, P. F., Guénin, G., Morin, M., Robin, M. (1979), in Transformations de Phases à l’État Solide-Trans-
formations Martensitiques. Lyon: Dep. Gènie Phys. Mat., INSA
– the Nishiyama – Wassermann (N – W) re- – the Greninger – Troiano (G – T) relations

lations: (here the planes and directions are no
– – longer exactly parallel):
(111)P //(101)a¢ and [121]P //(101)a¢
– ––
(111)P ≈ (011)a¢ and [101]P ≈ [1 11]a¢
Concerning the transformation of f.c.c. to tion. This morphology is easily observed

h.c.p. austenite and that of h.c.p. to b.c.c. by light optical microscopy (LOM) and the
martensite (the b.c.c. to h.c.p. relation is mutual arrangement of these regions con-
known as the Burgers relation), the follow- stitutes the microstructure at the LOM
ing relations apply: level. Electron microscopic analysis re-
veals that also at the submicroscopic level
(111)P //(0001)e //(101)a¢
martensite is characterized by a typical
and substructure. The morphological, micro-
– – – structural and substructural aspects of mar-
[11 0]P //[1210]e //[111]a¢
tensite are briefly discussed below.
Taking the N – W relations as an exam- The martensite regions are generally
ple, any one of the four crystallographi- plate-shaped, i.e. one lateral dimension is
cally equivalent {111} austenite planes, much smaller than the other two. If the two
– – –
(111), (111), (111) and (111), can be the larger dimensions are nearly equal they are
plane of parallelism. In each such plane called “plates”, and if they are very un-
–
any one of the three ·121Ò directions, which equal “laths”. A typical lath in low-carbon
happen to be directions of the Burgers vec- steel (with a carbon content less than 0.4%)
tors, can be chosen. This therefore results has dimensions 0.3 ¥ 4 ¥ 200 µm3. How-
in 12 different orientations of an a¢-crystal ever, martensite formed into the parent
in one austenite crystal. These differently phase does not always appear as a geomet-
oriented martensite crystals are called rically well-shaped plate. Martensite plates
“variants”. It can easily be shown that the that form near a free surface or in a single
K – S relations lead to 24 variants. crystal as a result of a single-interface
Orientation relationships and the orien- transformation may show the idealized
tation of the habit plane change from one plate-like shape. In such a single-interface
alloy system to another, and within a given transformation the habit plane extends
alloy system from one composition to an- from one side of the crystal to the other
other. The observable crystallographic pa- (see Fig. 9-11). Because of the shape
rameters are summarized in Table 9-3 for change and the high elastic stresses that are
a large number of alloy systems; a more created, a thick plate cannot terminate in-
complete list of these and other crystallo- side a parent crystal. As is frequently ob-
graphic characteristics of various marten- served, lenticular shapes or groupings of
sites is given by Nishiyama (1978). differently oriented martensite plates will
reduce these elastic stresses.
In the case of lenticular martensite, the
9.8.1.3 Morphology, Microstructure
habit plane is no longer a plane but a
and Substructure
curved surface and the normal average of
Because the martensitic transformation the lenticular plate is then taken as the or-
is a first-order phase transformation, both ientation of the habit plane. Sometimes this
phases, the parent and the martensite orientation is visible as a “midrib” in some
phase, coexist on cooling in a temperature martensites (Fig. 9-21). It is believed that
range between Ms and Mf and on heating the martensite could grow to a certain ex-
between As and Af . Martensite thus occurs tent as a plate, but that lenticular shapes are
in physically isolated regions, the morphol- formed owing to the high elastic stresses
ogy of which is typical of the transforma- that are building up. The high stresses may
trigger other plates to form in the vicinity

of the plate formed earlier, giving rise to an
“autocatalytic growth” of martensite.
A multivariant martensite arrangement
is the most commonly observed micro-
structure. Often variants are arranged in
some recognizable patterns and at times
numerous variants present in a regular ar-
ray give the impression of a martensite col-
ony. The latter is typical of the “massive”
microstructure, consisting of a packet of
parallel martensite “laths” separated by
more or less wavy interfaces. Each lath in
the packet maintains the same variant or
orientation relationship with the parent
crystal. A single grain of the parent phase
can transform into one or more such pack-
ets. The “plate” martensite arrangement
which is observed in the same alloys differs
from the lath configuration, because adja-
cent martensite plates are generally not
parallel to each other.
Diagrams have been constructed for
cases where there is a variety in morphol-
ogy, as for example for Fe – Ni – C alloys.
Maki and Tamura (1987) showed that the
morphology of the a¢-martensite in these
alloys is related to the transformation tem-
perature and the carbon content (Fig. 9-
22).
Distinct martensite plate arrangements
can also be recognized in alloys pertaining
to the b-Hume – Rothery alloys. Schroeder
and Wayman (1977) classified these ar-
Figure 9-21. Transmission electron micrograph of a
rangements into spear, fork, wedge and di- martensite in steel showing the twinned midrib
amond forms. Each representation carries (courtesy C. M. Wayman, 1989, University of Illi-
with it a definite crystallographic relation- nois, Urbana (IL)).
ship between the variants constituting
the arrangement. Grouping of martensite
plates in such arrangements will lead to a group formation is then called “self-ac-
serious reduction in the elastic stresses. By commodation”.
analysis of the crystallography of the plates In these b-Hume – Rothery alloys, three
in a single group, it can be shown that the types of martensite form, the 3 R-, 9 R- and
respective macroscopic shape changes an- 2 H-types. A detailed analysis of the micro-
nihilate each other (Tas et al., 1973). Such structure reveals that the martensite vari-
deformation, strain accommodation is then

accomplished not only by elastic but also by
plastic deformation in one or both phases.
Partitioning of the parent crystal, with
finer plates forming subsequently in the
partitioned region, frequently occurs and
illustrates the fractal nature of the transfor-
mation (Fig. 9-10). It should be mentioned,
however, that not all martensite micro-
structures show fractal characteristics
(Hornbogen, 1988).
Until now only the more macroscopic
observable features of the microstructure
of martensite have been discussed. Trans-
mission electron microscopy reveals that
the substructure of martensite is also char-
acteristic. It consists, depending on the al-
loy system and alloy composition, of regu-
larly spaced stacking faults (e.g. Cu-base
b¢-type martensite), twins with a constant
Figure 9-22. Relationship between a¢-martensite thickness ratio (e.g. Fe – 30% Ni), disloca-
morphology and Ms temperature as a function of car- tions (e.g. Fe – 20% Ni – 5% Mn), stacking
bon content in Fe – Ni – C alloys (Maki and Tamura, faults and twins in the same martensite
1987).
plate (e.g. Cu – Ga), or twins in the midrib
region surrounded by dislocations.
ants form in six different groups, each
group consisting of four variants. The habit
9.8.1.4 Crystallographic
planes of these four variants are located
Phenomenological Theory
around the same {110}b pole, whereas each
basal plane is located close to one of four The formal phenomenological theories
other {110}b poles. The total macroscopic of martensite formation predict the crystal-
deformation of such a group is to a first ap- lographic characteristics, such as the shape
proximation completely compensated. A deformation, the orientation of the habit
more complete reduction in the three-di- plane, the orientation relationship between
mensional strain is obtained if the total parent and product phase, and the ampli-
transformation strain is calculated for the tude of lattice invariant deformation. This
six groups as an entity. As the martensite prediction is obtained from the sole knowl-
transformation involves a very small vol- edge of the structures and lattice parame-
ume change in all b-Hume – Rothery alloys, ters of the two phases and with the basic as-
the strain accommodation is thus almost sumption that the interface between parent
complete. phase and martensite is undistorted on a
The shape change associated with a mar- macroscopic scale.
tensite plate creates stresses in both the The observation of the K – S and N – W
parent and the martensite phases. If these orientation relationships led us to origi-
stresses exceed the flow stress for plastic nally believe that a martensite was formed
by shear on those planes and directions martensite would be a vector of the same
specified in the orientation relationships. size and the same orientation in the parent
However, it was found that the shear mech- phase before transformation. As indicated
anisms proposed by the K – S and N – W in Sec. 9.3.1, the macroscopic shape
relations are not consistent with these exchange of an invariant-plane transforma-
perimental observations. The observations tion can be represented by a combination
made by Greninger and Troiano (1949) on of a pure lattice deformation (B), the so-
Fe–22% – Ni–0.8% C martensite were the called Bain strain, a rigid lattice rotation
key to the mathematical development of (R), and an inhomogeneous lattice-invari-
the crystallographic theory of martensite. ant deformation (P). The pure lattice defor-
They found that martensite plates exhibited mation either increases or decreases some
a surface relief that can be described by a vectors in length. According to Wayman
homogeneous shear along the habit plane, (1964), “the essence of the crystallographic
but this homogeneous shear could not theory of martensitic transformations is to
transform the f.c.c. lattice of the parent find a simple shear (of a unique amount, on
phase into the b.c.t. lattice of the marten- a certain plane, and in a certain direction)
site. If the f.c.c. lattice had undergone the such that vectors which are increased in
same homogeneous deformation, the struc- length due to the lattice deformation are
ture of the martensite would have been correspondingly decreased in length due to
trigonal. They therefore suggested that two the simple shear, and vice versa. Such vec-
types of shear are involved in the marten- tors which remain invariant in length to
sitic transformation: a “first” simple shear these operations define the potential habit
which is responsible for the macroscopic plane. Physically speaking, the ellipsoid
shape change, and a “second” shear which generated from the initial sphere by the lat-
needs to be added to obtain the structural tice deformation is distorted by the simple
change but which should produce no ob- shear into another ellipsoid which becomes
servable macroscopic change in shape. tangential to the initial sphere, the points of
Two years later, Bowles (1951) showed tangency being related along a diameter.”
that the shape deformation may be any in- This is clearly illustrated in Fig. 9-23,
variant plane strain. This opened the way to where the problem becomes two-dimen-
the formulation of the general theory of the sional, because one of the principal axes of
crystallography by Wechsler et al. (1953) the lattice deformation is taken as normal
and, independently, by Bowles and Mack- to the plane of shear.
enzie (1954). Almost equivalent theories Some complementary remarks concern-
were later developed by Bullough and ing the crystallographic theory should be
Bilby (1956) and Bilby and Frank (1960). made. The input data for the calculations
The reader may consult the following more are (i) the lattice parameters of the parent
elaborate reviews of these theories: Way- and martensite phases, (ii) the lattice corre-
man (1964), Christian (1965), Nishiyama spondence, and (iii) the lattice-invariant
(1978) and Ahlers (1982). shear. The output of the calculations is then
The basic assumption in the crystallo- the amount of inhomogeneous shear re-
graphic theories is that the interface be- quired to obtain the invariant plane condi-
tween the product and the parent phases is tion, the macroscopic shape change, and
undistorted, which means that any vector the orientation relationship. Because of the
that lies in this interface on the side of the lattice symmetries, differently oriented
ments have been observed in other alloy

systems. To test critically the validity of
the crystallographic theories, all crystallo-
graphic parameters should be measured
and compared with the theoretical predic-
tions. Agreement should be obtained for
the complete set of parameters. For the
martensites that are twinned, this includes
a careful determination of the normal to the
twinning plane K1 relative to the parent lat-
tice. In Cu – Al – Ni, for example, inconsis-
tencies up to 12.5° have been found.
In cases where the lattice-invariant shear
Figure 9-23. Production of an undistorted plane by is twinning (as opposed to faulting or slip),
shear such that the shape ellipsoid touches the unit
type I twinning, where the twin plane orig-
sphere along one of its principal axes (Christian,
1965). inates from a mirror plane in the parent
phase, has been assumed. Otsuka (1986)
carefully analyzed for a number of systems
Brain relations and inhomogeneous shear the possibility of a type II twinning as an
systems lead to a number of crystallo- alternative inhomogeneous shear. In type II
graphically equivalent solutions. Because twinning, the shear direction stems from a
of the observed orientation relationships, two-fold symmetry direction of the parent
the Bain relationship is fixed for most phase. Otsuka (1986) compiled all the
cases. However, larger unit cells are some- twinning modes observed in martensite
times chosen, especially for those marten- into a table and found that most of them are
sites where the crystal structure has a large type I or compound but that type II twin-
unit cell compared with that of the parent ning had only recently been observed.
phase. The only variable in these calcula- According to Nishida and Li (2000), five
tions is the choice of the inhomogeneous different twinning modes exist in TiNi
shear system. The orientation of the habit and other shape-memory alloys such as
plane is found to be very sensitive to the Cu – Al – Ni, Cu – Sn etc., namely the {111}
choice that is made. For the f.c.c. to b.c.c. type I, the {011} type I, the ·011Ò type II,
or b.c.t. transformation, the twin shear the (100)-compound and the (001)-com-
–
(112)M [111 ]M gives a (3 15 10)P habit pound twins. Type II twinning has recently
––
plane, whereas a (011)M [111]M shear re- received much attention as a mechanism
sults in a (111)P habit plane. for lattice-invariant shear in some alloys.
The phenomenological theory as ex- Because type II twins have irrational twin
plained above is based on one active shear boundaries, the physical meaning of an ir-
system. However, for some alloy systems rational boundary is still a controversial
this is not adequate. For example, a single problem. It has been proposed that an irra-
(112)M twinning system is not able to ex- tional boundary consists of rational ledges
plain the {225}P habit plane in some steels, and steps, the average being irrational.
and even two inhomogeneous shear sys- Thereafter, Hara et al. (1998) carried out a
tems do not give agreement with the ex- careful study using HRTEM, to try and ob-
perimental observations. Similar disagree- serve ·111Ò type II twin boundaries in a
Cu – Al – Ni alloy, but they were unable to even at very low temperatures. This obser-
observe any ledges or steps. The boundary vation shows that the interface migration
is always associated with dark strain con- must be accomplished without appreciable
trast, and the lattice is continuous through thermal activation. The interface is thus
the irrational boundary. Nishida and Li “glissile”. In searching for models to ex-
(2000) also carried out extensive studies on plain the structure and mobility of the
·011Ò type II twin boundaries in TiNi using interface, we are concerned with the ideal
HRTEM, but they did not observe ledges or and the actual interface morphology. A
steps either. Based on these experimental careful experimental analysis of the inter-
results, it is thus most likely that the type II face structure is therefore required if we
thin boundary is irrational even on a micro- want to verify the various models that have
scopic scale, and the strains at the boun- been proposed. As the models treat the
dary are elastically relaxed with wide twin interface on an atomistic scale, the sub-
width. To confirm this interpretation. Hara structure of the interface should be studied
et al. (1998) carried out computer simula- by conventional and by high-resolution
tions by using the molecular dynamics transmission electron microscopy. The
method. The result showed that the irra- same holds for martensite-to-martensite
tional thin boundary did not show any interfaces, which in some alloys are also
steps. Thus, the above interpretation for an mobile. Recently, atomistic imaging of the
irrational twin boundary is justified. Ot- martensite/austenite and martensite/mar-
suka and Ren (1999) have pointed out tensite interfaces have been obtained. It is
again the importance of type II twinning in therefore not surprising that both aspects,
the crystallographic aspects of martensite. the observation of interface substructures
They also stress the role that martensite and the atomistic models, are treated
aging has on the rubber-like behavior of jointly in the literature. For further reading
martensite, a point that has been a long- concerning the interface structures and the
standing unsolved problem. They showed growth mechanism of martensite we refer
that the point defects play a fundamental to the review papers by Christian (1982),
role. The deformation mechanisms of the Christian and Knowles (1982), and Olson
cold deformation of NiTi martensite have and Cohen (1986). A summary of these pa-
been thoroughly analyzed by Liu et al. pers is given below.
(1999 a, b). They also found an interplay Let us first introduce the kinds of mar-
between type I and type II twinning. tensite interfaces concerned: glissile and
As already mentioned in Sec. 9.3.6, a non-glissile martensitic interfaces, with the
better and more complete agreement can be latter subdivided into the coherent and
achieved when the strain energy terms, semi-coherent interfaces. The two struc-
both bulk and interfacial, are included in tures, martensite and the parent phase, are
the calculation. said to be “fully coherent” if both lattices
have a matching plane parallel to the inter-
face. If a fully coherent interface is dis-
9.8.1.5 Structure of the Habit Plane
played, the crystal undergoes a shape de-
In a number of alloys, especially those in formation leaving all vectors in the inter-
which the so-called thermoelastic marten- face invariant. In general, the two phases
sites are formed, the interface between do not have a plane of atomic fit, so that
martensite and the parent phase is mobile, fully coherent martensite interfaces are ex-
Figure 9-24. Three-dimen-

sional representation of a
semi-coherent martensite
interface; the vectors OA are
distorted into O¢A¢ but the
large vectors OZ = O¢Z¢ are
invariant (Christian, 1982).
ceptional. A fully coherent martensitic in- electron micrographs show that the inter-
terface is, for example, that between f.c.c. face between martensite and the parent
and h.c.p. structures with lattice parame- phase and also the intervariant interfaces
–
ters such that a (f.c.c.) = ÷2 a (h.c.p.); the and the interfaces between the internal
atomic arrangement in the basal planes, twins in one martensite plate contain dis-
which constitute the interface between the continuities (“steps”) on an atomic scale,
two structures, are identical. Such transfor- the nature of which has not yet been com-
mations are found in Co and its alloys and pletely unravelled. These steps can be con-
in some Fe-based alloys. The situation at sidered as resulting from a small deviation
semi-coherent interfaces becomes more of the ideal habit plane, and would then be
complex. The models predict the presence comparable to those observed along the
of dislocations to correct the mismatch interfaces of tapered twins.
along the interface. If this coherent inter- An exact understanding of the structure
face moves, it is suggested that not all vec- of the interfaces involved in the martensitic
tors are left invariant and that the move- transformation (the parent – martensite, the
ment of dislocations causes shear in the intervariant, and the twin/twin interfaces)
product phase. Fig. 9-24 shows the slip as- is therefore essential in determining the
sociated with the interface dislocations. mechanism of transformation and the mo-
Internally twinned martensite has been bility of the interfaces.
reported to show a zig-zag parent – marten-
site interface, as observed by conventional
9.8.2 Thermodynamics and Kinetics
electron microscopy in, for example,
of the Martensitic Transformation
Ti – Mn and Cu – Al – Ni. Fine parallel stria-
tions have been observed in the interface
9.8.2.1 Critical Driving Force
between austenite and both the b¢-type and
and Transformation Temperatures
the g¢-type Cu – Al – Ni-martensite. These
striations have been accounted for in terms A quantitative thermodynamic treatment
of interfacial dislocations resulting from of the martensitic transformation requires a
random faulting on the basal plane of the precise knowledge of the thermodynamic
b¢-type martensite and the twinning planes equilibrium temperature T0 and of the
of the g¢-type martensite. High-resolution change in Gibbs energy at the transforma-
tion temperature Ms . Both can be in the g-phase. Because of the shape and
calculated and/or derived from measured volume changes associated with the trans-
data, as is shown here for two examples: formation, elastic strain energy also has to
the martensitic transformation in Fe – X be considered. The Gibbs energy is thus
(X = Ni, Ru, …) and in Cu – Zn – Al alloys. composed of chemical Gibbs energy, Gc ,
In the former example, both the parent and and strain energy, Ee , so that the Gibbs en-
the martensite phases have the same struc- ergy change accompanying the transforma-
ture as the equilibrium phases and hence tion may be written as
the data for the equilibrium phases can be
taken. In the latter example, both structures Gg Æ M = DGcg Æ a + DEea Æ M (9-10)
differ from that of the equilibrium phases, At temperatures below Ms , where both
which requires a more elaborate calculation. phases coexist and thus are in equilibrium,
For the Fe – X alloys, the Gibbs energy DG g Æ M | T = 0. DGcg Æ a | T is then exactly
per mole of the parent austenite phase g equal, but opposite in sign, to the sum of all
and of the martensite phase a are G g and non-chemical energies DGnc aÆM
| T . If the
G a, respectively. The change in Gibbs en- surface energies are neglected in compari-
ergy per mole, DG g Æ a, which for a mar- son with the high strain energies, the non-
tensitic transformation g Æ a is available chemical energy equals DEea Æ M | T , and
to the system at any temperature T, is then approaches zero at T = Ms . The strain en-
ergy stored in the material is the sum of
DG g Æ a | T = G a – G g (9-8) that produced by shearing and by volume
change. The former depends on the
This quantity is negative for temperatures
strength of the parent phase and thus also
at which the a-phase is the more stable and
on the grain size, hence Ms also depends on
positive for temperatures at which the g-
the grain size, as shown by Hsu and Xiao-
phase is the more stable. There is a charac-
wang (1989).
teristic temperature T0 corresponding to the
The necessary undercooling (T0 – Ms )
thermodynamic equilibrium between both
and superheating (As – T0 ) vary for differ-
phases, such that
ent alloy systems, and for certain materials
DG g Æ a | T = T0 = 0 (9-9) even with composition. A precise thermo-
dynamic definition of Ms and As cannot be
Because the transformation creates interfa- given, however, if the non-chemical ener-
aÆM
cial and elastic energies, the martensitic gies, DGnc , are not known. We can then
transformation g Æ a or a Æ g does not only say that Ms or As is the temperature at
start at T0 , but at a temperature below or which the quantity DG g Æ a (at T = Ms or
above T0 , respectively. It is therefore nec- T = As , respectively) is sufficiently nega-
essary to undercool or overheat, respective or positive, respectively, to have a rea-
tively, until Ms or As is reached. At these sonable chance of nucleation.
temperatures the Gibbs energy change Two approaches are found in the litera-
DG g Æ a is sufficiently large to induce the ture for calculating the critical chemical
forward or reverse transformation, respec- driving force. The first is based on the ex-
tively. DG g Æ a (at Ms ) is then the critical perimentally determined Ms temperatures
chemical driving force. (Kaufman, 1965) and the other on a theo-
The martensite phase, represented by M, retical model for the non-chemical energies
is to be regarded as the a-phase embedded (Hsu, 1985).
Figure 9-25. Schematic dia-

grams for ferrous alloys that
form a g-loop (Fe – Cu, Cr,
Mo, Sn, V, W) and that g-
loop forms a stabilized aus-
tenite phase (Fe – C, Ir, Mn,
N, Ni, Pt, Ru); (a) equilib-
rium diagrams; (b) Ms tem-
perature diagrams (Krauss
and Marder, 1971).
a a
The first approach has been used for fer- where GFe and GA are the Gibbs energies of
rous alloys, which can be classified into pure iron and pure element A as a b.c.c. a-
a
two systems: those with g-loops and those phase, respectively, and Gm is the Gibbs
with stabilized g-phases (see Fig. 9-25). energy of mixing of the martensite phase.
Figure 9-25 also gives the Ms temperatures The change in chemical Gibbs energy ac-
for the a- and e-martensite. If A is the al- companying the martensitic transformation
loying element for iron, the molar chemical g Æ a then becomes
Gibbs energy for the austenite phase (G g ) gÆa gÆa gÆa
DG = (1 – x) DGFe + x DGA
can be written as gÆa
g g g g + DGm (9-13)
G = (1 – x) GFe + x GA + Gm (9-11)
gÆa
The quantity DGFe represents the Gibbs
where x represents the atomic fraction of energy change for transformation g Æ a of
the element A in solid solution in the g- pure iron and can be assessed experimen-
austenite, (1 – x) the atomic fraction of tally from the measured heat of transforma-
g
iron, GFe the chemical Gibbs energy of tion and the specific heat of both phases.
g
pure iron as f.c.c. g-phase, GA the chemical gÆa
The quantity DGA cannot usually be ob-
Gibbs energy of element A as f.c.c. phase, tained from experiments because the ele-
g
and Gm the Gibbs energy of mixing of the ment A does not always exist in the two
g-phase. Similarly, the Gibbs energy of the modifications g and a; it must therefore be
a-phase can be given as estimated from thermodynamic models for
a a a a gÆa
G = (1 – x) GFe + x GA + Gm (9-12) solid solutions. The quantity DGm is the
difference in Gibbs energy of mixing and

can in principle be measured experimen-
tally through activity measurements; if not,
it must also be estimated.
The equilibrium temperature T0 and the
critical driving force and Ms and As can be
calculated from Eq. (9-12). Such calcula-
tions have been performed by Kaufman
(1965) for the iron – ruthenium alloy sys-
tem, which is of particular interest because
the g-phase transforms martensitically into
two phases, the a-b.c.c. and the e-hexago-
nal phases. Both phases also occur as equi-
librium phases, as shown in Fig. 9-26 to-
gether with observed Ms and As tempera-
tures and the calculated T0 temperatures.
The undercooling for the a-martensite
formation is strongly composition depen-
dent, whereas it is independent of composi-
tion for the hexagonal martensite. The
computed T0 curves are seen to lie between
the appropriate transformation temperature
curves. The calculated driving forces,
DG gÆa and DG gÆe, for both transforma-
tions are plotted as a function of temperature
for various compositions. The intersections
of these curves with the temperature axis
correspond to the theoretically deduced T0
temperatures for the appropriate composi-
tions. When the appropriate experimentally
derived Ms and As are cross-plotted, the crit-
ical driving forces for the g/a and g/e mar-
tensitic transformations are obtained. The
latter is seen to be smaller, which is consis-
tent with the closer lattice correspondence
of the former transformation.
Hsu (1985) presented a model by which
more accurate computations of the non-
chemical part DG a Æ M are possible for
Fe – C, Fe – X and Fe – C – X alloys. This en-
Figure 9-26. (a) The iron – ruthenium phase diagram
abled him to obtain from Eq. (9-9) the and (b) the T0 and Ms and As temperature diagrams
theoretical Ms temperatures, which are in (after Kaufman, 1965).
good agreement with the observed values.
In the second example, martensite for-
mation in Cu – Zn alloys, the change in
Gibbs energy can, according to Hsu and It is known that martensite may also be
Zhou (1989), be described as induced by an external stress at tempera-
tures above Ms . The problem now is to cal-
DG b¢ – M = DG b¢ – b + DG b – a + DG a – a¢ culate the change in T0 due to changes in
+ DG a¢ – M (9-14) stress. As a first approximation, it is as-
PÆM
sumed that the driving force DGm | T = Ms
where b¢ – M represents the transformation
required for nucleation remains constant
from the the ordered b.c.c. phase to the or-
with temperature and thus independent of
dered 9 R-type of martensite, b¢ – b the or-
stress. Patel and Cohen (1953) calculated
der – disorder transition, b – a the transfor-
the work done on the stressed specimen;
mation from the disordered b.c.c. to the
their treatment provides a good under-
disordered f.c.c. phase having the same
standing of how an applied stress that is de-
composition, a – a¢ the disorder – order
composed into a shear stress along the
transformation in the f.c.c. phase, and
habit plane and a normal stress perpendicu-
a¢ – M the transition from the ordered f.c.c.
lar to it, affects the transformation temper-
phase to the ordered martensite phase. As-
ature. At Mss , which is the martensite start
suming a simplified relationship between
temperature when cooling under an applied
the degree of ordering and temperature,
stress s , the chemical Gibbs energy change
Hsu and Zhou found good agreement
equals the transformation work of the ex-
between the calculated and observed Ms .
ternal stress:
Their calculations show that ordering of
the parent phase, which cannot be sup- DGsP Æ M = 1/2 s a (9-15)
pressed even by severe quenching, strongly ¥ [d 0 sin 2 q ± e0 (1 + cos 2 q )] Vm
influences T0 .
where d 0 is the shear strain, s a the applied
stress, q the angle between the stress axis
and the normal to the operative shear plane,
e0 the corresponding strain associated with
the transformation, and Vm the molar vol-
ume. The quantity DH P Æ M can be meas-
ured by calorimetry and DS P Æ M can be
evaluated from stress-induced transfor-
mation experiments or calculated from Eq.
(9-16). The temperature T0 can be calcu-
lated thermodynamically or obtained more
or less accurately from the relationships
DG P Æ M = DH P Æ M – T DS P Æ M
(9-16)
DH P Æ M = T0 DS P Æ M
and
T0 = (As – Ms )/2 = (Af – Mf )/2 (9-17)
Figure 9-27. Gibbs energy G* versus temperature
and force for stressed samples: P and M represent However, it should be noted that the deter-
free energy surfaces for parent and martensite, re- mination of T0 does not always obey these
spectively (Wollants et al., 1979). simple relationships and that the calorimet-
rically measured heats do not always re-

flect the exact heats of transformation.
As stress itself is also a state variable in-
dependent of temperature, it should be con-
sidered in the thermodynamic treatment as
explained by Wollants et al. (1979). To de-
scribe the thermodynamic state of a uniax-
ially stressed crystal, they introduced the
“elastic” state functions H* and G*, which
incorporate the effect of stress as follows:
H* = U + P V – F l = H – F l
= H – s e Vm (9-18)
G* = U + P V – T S – F l = G – F l
= G – s e Vm (9-19)
where F is the applied force and l the “mo-
lar length” of the crystal. Fig. 9-27 illus- Figure 9-28. Results of tensile tests for inducing
martensite in a Cu – 34.1 Zn – 1.8 Sn (at.%) alloy
trates how the equilibrium temperature and (Pops, 1970).
force change when one of the variables is
changed; P and M represent the Gibbs energy
surfaces of the parent phase and of marten-
site, respectively. At the two-phase equilib-
rium G*P = G* M and if, at constant hydro-
static pressure, the intensive variables F
(or s ) and T are changed in such a way
that there is thermodynamic equilibrium
between martensite and the parent phase,
then dG*P = dG* M, or
– S P dT – l P dF = – S M dT l M dF
so that dF/dT = – [DS/Dl] P Æ M, or, since
the molar work F DlP Æ M = s e P Æ M Vm and
DS P Æ M = DH* (s )/T0 (s ), it also follows
Figure 9-29. Orientation dependence of stress –
that
(9-20) strain curves for martensite formation in a
ds /dT = – [DH* (s )]/[T0 (s ) e P Æ M Vm] Cu – Al – Ni alloy (Horikawa et al., 1988).
where Vm is the molar volume (Vm = V P =

V M ). Eq. (9-20) is the Clausius – Clapeyron atures (see Fig. 9-28). The elongation re-
equation for a uniaxial stress, which is sim- sulting from the transformation is orienta-
ilar in form to that for hydrostatic pressure, tion dependent (Fig. 9-29). From data such
except for the negative sign. as those shown in Figs. 9-28 and 9-29, we
The change in critical stress necessary to can calculate DS P – M . It is evident that for
induce martensite can be obtained from each crystal orientation the slope ds /dT is
tensile tests carried out at different temper- different.
concerning both static and dynamic lattice

instability. All nucleation models can fur-
ther be divided into classical and non-clas-
sical. The former model involves lattice
perturbations of fixed amplitude and vary-
ing size, whereas the latter considers per-
turbations of varying size (Olson and Co-
hen, 1982 b).
In the classical nucleation theory; mar-
tensite nuclei form along a path of constant
composition and structure and the state of
the nucleus is given by its size. Because the
martensitic transformation involves shear
strains, it can be shown that the strain en-
ergy is minimized for a disc-like nucleus,
Figure 9-30. Shape of a nucleus of a martensite
but then the surface energy becomes very
plate. large. The critical nucleus, assuming an
oblate spheroidal shape (Fig. 9-30), will
then have an aspect ratio (c/r) such that for
Similarly, martensite can be induced by any change in shape, the decrease in strain
magnetic fields. By taking into account the energy will be exactly balanced by an in-
composition, the influence of grain boun- crease in interfacial energy. The interfacial
daries, and crystal orientation, the mag- Gibbs energy per plate is
netic invar effect and the austenite magne-
v D gs = 2 p r 2 G (9-21)
tism, Shimizu and Kakeshita (1989) pro-
posed an equation that describes the shift where v is the volume of the plate, Dgs the
of Ms as a function of the magnetic field. surface Gibbs energy per unit volume, and
G the interfacial energy. The strain energy
per plate is
9.8.2.2 Nucleation
v D ge = (4/3) p r 2 c (A c/r) (9-22)
Martensitic transformations are first-or-
der phase transformations and hence occur where D ge (= A c/r) is the strain energy per
by nucleation and growth. In most in- unit volume and A is a factor to be deduced
stances, except in the case of thermoelas- from linear elasticity and thus a function of
ticity (see below), the growth of a marten- the elastic constants and of the shear and
site plate proceeds so rapidly that the trans- dilatational strains. The chemical Gibbs
formation kinetics are dominated by the energy change per plate is
nucleation event. Various mechanisms of v D gc or (4/3) p r 2 c D gc (9-23)
martensite nucleation have in the past been
proposed and can be considered under two If the nucleation occurs at a lattice defect,
subheadings. In the first group of models, we have to consider also the Gibbs energy
the so-called localized nucleation models, Gd due to the defect and the nucleus – de-
concepts of diffusional nucleation kinetics fect interaction energy Gi . According to
are applied; the second group of models is Olson and Cohen (1982), the total Gibbs
based on lattice instability considerations energy describing the formation of a classi-
Figure 9-31. Schematic

nucleus Gibbs energy (G)
curves for nucleation via a
classical path: (a) homoge-
neous, (b) heterogeneous,
and (c) barrierless nuclea-
tion (Olson and Cohen,
1982 b).
cal martensitic nucleus becomes recognized that homogeneous nucleation

(9-24)
of martensite is impossible. Recently,
G (r, c) = Gd + Gi + v (Dgc + Dge + D gs )
much progress has been made in the under-
and is given schematically in Fig. 9-31. standing of nucleus formation at lattice de-
Three cases are considered in calculating fects. DG* and therefore also the critical
the critical free energy for nucleation DG* size of the nucleus can be reduced by as-
and the critical nucleus size r* and c*. suming the nucleation at a defect. Under
In the case of homogeneous nucleation, certain special conditions, this heterogene-
Gd and Gi are zero, and on inserting the ous nucleation may even be barrierless.
necessary quantities into Eq. (9-24) we Such a case applies to the f.c.c.-to-h.c.p.
find that the barrier DG* is too high by sev- transformation, which may take place by
eral orders of magnitude. Even assuming dissociation of a number of properly
local compositional fluctuations or the ex- spaced total dislocations present in the ma-
istence of pre-existing embryos does not trix phase into partial dislocations separ-
give full satisfaction. It was therefore soon ated by stacking faults. The stacking-fault
Figure 9-32. Electron mi-

crographs of the nucleation
and early growth stage
around inclusion particles in
a Ti – Ni – Cu alloy (Saburi
and Nenno, 1987).
energy is temperature dependent and be- 9.8.2.3 Growth and Kinetics

comes positive below T0 , which results in a
barrierless nucleation. A distinction is made between the kinet-
In a number of alloy systems, softening ics of a single martensite plate and the glo-
of certain elastic constants is observed and bal kinetics, which expresses the volume
it is then argued that, although the homoge- fraction of the parent phase that is trans-
neous soft-mode concept is definitely not formed. According to the observed kinet-
adequate to describe the nucleation of mar- ics, martensitic transformations can be di-
tensite, stresses and strains present around vided into two distinct classes: athermal
defects of the lattice can induce a local me- and isothermal martensite. In athermal mar-
chanical instability. Such a model is called tensite, the transformation progresses with
“the localized soft-mode concept” (Guénin decreasing temperature, whereas in iso-
and Clapp, 1986). In this model the lattice thermal martensite, the transformation pro-
Gibbs energy is a function of pure strains gresses with time at a constant temperature.
and therefore of second- and third-order The growth may be “thermoelastic” or of
elastic constants. The third-order constants the “burst” type. The latter is the more
(which relate the strain energy to the common mode. It consists of the formation
amount of strain) introduce anharmonic of comparatively large amounts of marten-
terms into the strain energy and may lead to site (typically 10 – 30 vol.%) in “bursts”
mechanical instability. The region of me- that are caused by autocatalytic nucleation
chanical instability, or the “strain spino- and rapid growth of numerous plates. Each
dal”, is so defined that any further increase individual martensite plate is completely
in strain will make the lattice unstable with formed with a speed higher than 105 cm/s
respect to a decomposition into strained re- and the transformation progresses by the
gions. In these zones a nucleus can develop formation of new plates. The global kinet-
without generating any strain energy, and ics of the transformation are therefore es-
the only resisting term remains the surface sentially controlled by the nucleation fre-
energy. This results in a reduced critical quency. The thermoelastic growth mode is
size of the nucleus, which is further de- characterized by the formation of thin, par-
creased as the temperature is lowered ow- allel-sided plates or wedge-shaped pairs of
ing to the increase in chemical driving plates (Fig. 9-33), which form and grow
force. progressively as the temperature is lowered
In situ electron microscope observations below Ms and which shrink and disappear
have been made on the nucleation and early on reversing the temperature change. This
stages of growth of martensite, as shown in behavior arises because the matrix accom-
Fig. 9-32. Martensite nucleates at stress modates the shape deformation of the mar-
concentrations, the nucleation takes place tensite plate elastically, so that at a speci-
repeatedly at the same place, and the strain fied temperature the transformation front
contrast disappears as nucleation and of the plate and the matrix are in thermody-
growth proceed and reappears when mar- namic equilibrium. Any change in temper-
tensite disappears. ature displaces this equilibrium and, there-
At present, the nucleation models are before, the plate grows or shrinks. A com-
ing further refined by molecular dynamic plete mechanical analog of this thermoelas-
calculations. tic behavior is the pseudoelastic behavior.
The growth or shrinkage of individual mar-
Heating 9.8 Martensitic Transformations 633
Cooling
Figure 9-33. Thermoelastic behavior in Ag – Cd alloys showing the

growth of self-accommodating groups of martensite plates (Delaey
et al., 1974).
Figure 9-34. Schematic representation of some relevant features (volume-transformed product or transforma-
tion strain) experimentally observed in hysteresis curves corresponding to thermally induced and stress-induced
thermoelastic transformations: (a, e) single interface transformation in a single crystal; (b, f) multiple interface
transformation; (c, g) discontinuous jumps (bursts), (d, h) partial cycling behavior.
tensite plates is then a direct function of the ing the control parameter, return to their
increase or decrease in stress. More elab- original values, a closed loop is formed
orate thermodynamic treatments of ther- (Fig. 9-34). The loop is always contoured in
moelasticity can be found in the papers by such a sense that it encloses a positive area,
Delaey et al. (1974), Salzbrenner and Co- representing the energy lost in the cyclic
hen (1979), Ling and Owen (1981) and Or- process. Therefore, hysteretic behavior is
tin and Planes (1989). always related to an energy-dissipative pro-
The best quantitative understanding of cess. The dissipated energy is much smaller
the kinetics of the martensitic transforma- in thermoelastic martensitic transforma-
tion is obtained from isothermal transfor- tions than in burst-type transformations.
mations, because they permit both the nu-
cleation and the transformation rates to be
determined. In those alloys exhibiting iso- 9.9 Materials
thermal martensite (Thadhani and Meyers,
1986), it is shown that at each temperature
9.9.1 Metallic Materials
the transformation starts in the austenite
and proceeds as a function of time. The A classification of the diffusionless dis-
transformation exhibits a C-curve behav- placive transformations in metallic materi-
ior. Isothermal martensitic transformation als is given in Table 9-4, where the alloy
kinetics consist of two effects: an initial in- systems are subdivided into three groups.
crease in the total volume fraction of mar- The origin of the martensitic transforma-
tensite, which is attributed to an autocata- tion in the first group lies in the allotropic
lytic nucleation of new martensite plates, transformation of the pure solvent. The
followed by a decrease due to the compart- parent phase of the alloys of this group thus
mentalization of the austenite into smaller does not show any remarkable mechanical
and smaller areas. instability. The second group consists of
the b b.c.c. Hume – Rothery alloys, which
are characterized by a moderate lattice in-
9.8.2.4 Transformation Hysteresis
stability in the temperature range above
Hysteresis behavior is one of the pecu- Ms . The third group is characterized by a
liar characteristics of both the thermal and drastic mechanical instability of the parent
stress-induced martensitic transformations. phase. Because the transformation is only
In several studies the origin of the fric- weakly first order (by this we mean a dis-
tional resistance opposing the interfacial continuous jump in the corresponding ther-
motion of martensite plates has been inves- modynamic property whose height, how-
tigated and described. From a practical ever, is very small) or even second order, it
point of view, the hysteresis phenomenon is in this group of alloy systems that we
is an important problem in the application find, in addition to the martensitic, the
of shape-memory alloys. In general hyster- quasi-martensitic transformations.
esis appears when, on passing through a lo- Traditionally, the ferrous and non-fer-
cal extreme value (maximum or minimum) rous martensites have been treated separ-
of any control parameter such as tempera- ately in the literature. Before going into de-
ture or stress, one or more state variables tail it will be an advantage to first compare
do not follow the original path in state and contrast ferrous and non-ferrous mar-
space. When all the state variables, includ- tensites and to do it in such a way that we
9.9 Materials 635
Table 9-4. Classification of metallic alloy systems showing diffusionless displacive transformations (Delaey
et al., (1982).
1. Martensite based on allotropic transformation of solvent atom
1. Iron and iron-based alloys

2. Shear transformation, close packed to close packed
1. Cobalt and alloys f.c.c. Æ h.c.p., 126 R SF*
2. Rare earth and alloys f.c.c., h.c.p., d.h.c.p., 9 R
(3. MnSi, TiCr2 NaCl Æ NiAs, Laves)
3. Body centered cubic to close packed
1. Titanium, zirconium and alloys b.c.c. Æ h.c.p., orth. f.c.c tw, d*
2. Alkali and alloys (Li) b.c.c. Æ h.c.p.
3. Thallium b.c.c. Æ h.c.p.
4. Others: plutonium, uranium, mercury, Complex structures
etc. and alloys
2. b-b.c.c. Hume–Rothery and Ni-based martensitic shape-memory alloys
1. Copper-, silver-, gold-, b-alloys

(disord., ord.) b.c.c. AB, ABABCBCAC, ABAC
2. Ni–Ti–X b-alloys b.c.c. Æ 9 R, AB tw, SF*
Nickel b-alloys (Ni–Al) b.c.c. Æ ABC tw, SF*
Ni3–xMxSn (M = Cu, Mn) b.c.c. Æ AB tw*
(Cobalt b-alloys, Ni–Co–X)
3. Cubic to tetragonal, stress-relaxation twinning or martensite
1. Indium-based alloys f.c.c. Æ f.c.t., orth. tw, tws *

2. Manganese-based alloys f.c.c. Æ f.c.t., orth. tws *
3. A 15 compounds, LaAgxIn1–x b-W. Æ tetr.
4. Others: Ru–Ta, Ru–Nb, YCu, LaCd
* SF: stacking faults; tws : (stress relaxation) twins; d: dislocated
can establish criteria to differentiate the (1982). Depending on alloy composition, a

two groups. Although not every detail of distinction is made among the various mar-
transformation behavior will be discussed, tensites, based either on the crystallogra-
a set of criteria have been chosen as shown phy, morphology or growth characteristics.
in Table 9-5. A more detailed review of Essentially, three different crystal struc-
martensite in metallic systems is given by tures appear: the b.c.c. or b.c.t. a¢-marten-
Nishiyama (1978). site, the h.c.p. e-martensite, and the long-
range ordered f.c.t. martensite. In plain car-
bon steels martensite is regarded as a
9.9.1.1 Ferrous Alloys
supersaturated, interstitial solid solution of
Martensitic transformations in ferrous carbon in b.c.c. iron (ferrite), with a crystal
alloys have been studied extensively, espe- structure that is a tetragonally distorted
cially the crystallography and morphology version of the ferrite structure. The tetrago-
which have been reviewed by Muddle nality is linearly dependent on the carbon
Table 9-5. A qualitative comparison between ferrous and non-ferrous martensites (Delaey et al., 1982b).
Ferrous martensite Non-ferrous martensite
Interstitial and/or substitutional Nature of alloying Substitutional
Martensitic state in interstitial Hardness Martensitic state is not much

ferrous alloys is much harder harder and may even be softer
than the austenite state than the austenite state
Large Transformation hysteresis Small to very small
Relatively large Transformation strain Relatively small
High values near the Ms Elastic constants of Low values near the Ms
the parent phase
Negative near the Ms Temperature coefficient Positive near the Ms

in most cases of elastic shear constant in many cases
High Transformation enthalpy Low to very low
Large Transformation entropy Small
Large Chemical driving force Small
Self-accommodation is not obvious Growth character Well developed self-accommodating

variants
High rate, “burst”, athermal Kinetics Slower rate, no “burst”,

and/or isothermal transformation no isothermal transformation,
thermoelastic balance
No single interface Growth front Single interface possible

transformation observed
Low and non-reversible Interface mobility High and reversible
Low Damping capacity High

of martensite
content. This a¢-martensite is also found in In some long-range ordered alloys, as in

a number of substitutional ferrous alloys, Fe – Pt and Fe – Pd, f.c.t. in addition to b.c.t.
the martensite being either b.c.c. or b.c.t. In martensite is observed.
certain alloy systems with an austenite As far as the morphology is concerned,
phase of low stacking fault energy, a mar- plate, lath, butterfly, lenticular, banded,
tensitic transformation to a fully coherent thin-plate and needle-like martensite can
h.c.p. product (e-martensite) is observed. be distinguished.
9.9 Materials 637
Table 9-6. Summary of substructure, habit plane cant deviations from the above equation.
(H.P.) and orientation relationship (O.R.) for the four The tetragonality is abnormally lower for
types of a¢-martensite (Maki and Tamura, 1987).
X = Mn or Re and abnormally higher for
Morphology Substructure H.P. O.R.* Ms X = Al or Ni. Heating to room temperature
of the latter martensite results in a lowering
Lath (Tangled) (111)A K–S High of the tetragonality. The formation of do-
dislocations mains or microtwinning in the former al-
Butterfly (Straight) (225)A K–S
loys and ordering of the Al atoms in the lat-
dislocations ter have been put forward as the origin of
and the abnormal c/a ratio. This behavior has
twins moreover been related to the martensite
plate morphology (Kajiwara et al., 1986,
Lenticular (Straight) (259)A N
dislocations
1991). Kajiwara and Kikuchi found that in
and or or Fe – Ni – C alloys the tetragonality is abnor-
twins (3 10 15)A mally large and depends on the microstruc-
(Mid-rib) G–T ture. It is very large for a plate martensite,
while it is normal or not so large for a len-
Thin-plate Twins (2 10 15)A G–T Low
ticular martensite. They conclude that “the
* K–S: Kurdjumow–Sachs relationship, N: Nishiyama martensite tetragonality is dependent on
relationship, G–T: Greninger–Troiano relationship the mode of the lattice deformation in the
martensitic transformation. If the lattice
deformation is twinning, the resulting c : a
A distinct substructure, crystallographic is large, while in the case of slip it is small”
orientation of the habit plane and austenite (Kajiwara and Kikuchi, 1991).
to martensite orientation relationship are
associated with each morphology, as sum-
9.9.1.2 Non-Ferrous Alloys
marized in Table 9-6.
Because the carbon atom occupies octa- A classification base of the non-ferrous
hedral interstices in the austenite f.c.c. lat- alloy systems exhibiting martensite is
tice, special attention is drawn to the
Fe – X – C martensite. In the martensite lat-
tice, those interstitial positions are defined
by the Bain correspondence (Fig. 9-35).
Only those at the midpoints of cell edges
parallel to [001]B and at the centers of the
faces normal to [001]B are permitted. This
preferred occupancy affords an explanation
of the observed tetragonality c/a, the de-
gree of which is a function of the carbon
content:
c/a = 1 + 0.045 (wt.% C) (9-25) Figure 9-35. Schematic representation of the Bain
correspondence for the f.c.c. to b.c.t. transformation.
Careful X-ray diffraction of martensite, The square symbols represents the possible occupied
freshly quenched and maintained at liquid positions of the interstitial carbon atoms (Muddle,
nitrogen temperature, has shown signifi- 1982).
given in Table 9-4, while the alloying ele-

ments are given in Table 9-7. A review of
the non-ferrous martensites was given by
Delaey et al. (1982 a).
Two typical examples of the first group
are the cobalt and titanium alloys. The
structure of the cobalt-based martensites
is in general hexagonal close-packed, but
more complex close-packed layered struc-
tures have been reported, such as the 126 R,
84 R and 48 R structures observed in
Co – Al alloys. Because the transformation
is a result of an f.c.c. to h.c.p. transforma-
tion, the basal planes of the martensite Figure 9-36. Critical stresses as a function of tem-
phase are parallel with the (111) planes of perature for the various stress-induced martensite
transformations in a Cu – Al – Ni alloy (Otsuka and
the parent phase and constitute the habit Shimizu, 1986).
plane. The structure of the martensite in Ti-
based alloys is also hexagonal but that of
the high-temperature phase is b.c.c. Both A typical example of the second group
plate and lath morphologies are encoun- are the copper-, silver- and gold-based
tered in titanium, and also in the similar alloys, which have been extensively re-
zirconium-based alloys. Slip is suggested viewed by Warlimont and Delaey (1974).
as the lattice-invariant deformation mode Depending on the composition, three types
in lath martensite, whereas the twining of close-packed martensite are formed
mode is observed in the plate martensities. from the disordered or ordered high-tem-
perature b.c.c. phase, either by quenching
or by stressing. The factors determining the
Table 9-7. Schematic representation of some non- exact structure are the stacking sequence of
ferrous martensitic alloy systems; the Co-, Ti- and the close-packed structure, the long-range
Zr-based terminal solid solutions, the intermetallic order of the martensite as derived from the
Ni-, Cu-, Ag- and Alu-based alloys, the antiferro- parent b-phase ordering, and the deviations
magnetic Mn-based alloys and the In-based alloys
from the regular hexagonal arrangements
(Delaey et al., 1982).
of the martensite. The last factor is due to
differences in the sizes of the constituent
atoms. The stacking sequence of the main
three phases are ABC, ABCBCACAB and
AB, respectively.
One of the interesting findings is the suc-
cessive stress-induced martensitic transfor-
mations in some of the b-phase alloys dis-
cussed above, as shown clearly if we plot
the critical stresses needed for the transfor-
mation (Fig. 9-36). Stressing a single crys-
tal of Cu – Al – Ni at, for example, 320 K
we find the parent b1-to-b¢1 , the b¢1 – g¢1 ,
9.9 Materials 639
the g¢1 – b≤1 , and finally the b≤1 – a¢1 marten- bohedral (the so-called R-phase) structural
site. change. At still lower temperatures the
The other typical example of the second rhombohedral R-phase transforms into a
group is the Ni – Ti-based alloy system; monoclinically distorted martensite. The
both prototype alloy systems constitute the R-phase also forms displacively and can be
shape-memory alloys (SMA). The occur- stress-induced, and shows all the character-
rence of a so-called “pre-martensitic” R- istics of a reversible transformation.
shape has long obscured the observations. Concerning the third group, in only a
The review by Wayman (1987), illustrates few cases, as in In – Tl, has definite proof
the complexity of the transformation be- been provided to justify the conclusion that
havior. During cooling, the high-tempera- the transformation is martensitic. Most of
ture ordered b.c.c. phase (P) transforms the transformations in these systems have
first to an incommensurate phase (I) and on to be classified as quasi-martensitic.
further cooling to a commensurate phase
(C), and finally to martensite. The P-to-I
9.9.2 Non-Metals
transformation is second order, whereas the
I-to-C transformation is a first-order phase Inorganic compounds exhibit a variety
transformation involving a cubic-to-rhom- of crystal structures owing to their diverse
Table 9-8. Non-metals with lattice deformational transformations (Kriven, 1982).
Inorganic compounds
Alkali and ammonium halides MX, NH4X (NaCl-cubic ¤ CsCl-cubic)
Nitrates RbNO3 (NaCl-cubic ¤ rhombohedral ¤ CsCl-cubic)
KNO3, TlNO3, AgNO3 (Orthorhombic ¤ rhombohedral)
Sulfides MnS (Zinc-blende-type ¤ NaCl-cubic)
(Wurtzite-type ¤ NaCl-cubic)
ZnS (Zinc-blende-type ¤ wurtzite-type)
BaS (NaCl-type ¤ CsCl-type)
Minerals
Pyroxene chain silicates Enstatite (MgSiO3) (Orthorhombic ¤ monoclinic)
Wollastonite (CaSiO3) (Monoclinic ¤ triclinic)
Ferrosilite (FeSiO3) (Orthorhombic ¤ monoclinic)
Silica Quartz (Trigonal ¤ hexagonal)
Tridymite (Hexagonal, wurtzite-related)
Cristobalite (Cubic ¤ tetragonal, zinc blende-related)
Ceramics
Boron nitride BN (Wurtzite type ¤ graphite-type)
Carbon C (Wurtzite type ¤ graphite)
Zirconia ZrO2 (Tetragonal ¤ monoclinic)
Organics
Chain polymers Polyethylene (CH2–CH2)n (Orthorhombic ¤ monoclinic)
Cement
Belite 2 CaO · SiO2 (Trigonal ¤ orthorhombic ¤ monoclinic)
chemistry and bonding. Compared with

metals, the relatively low-symmetry parent
structures have fewer degrees of freedom
on transforming to even lower symmetry
product structures, or vice versa. Many of
these transformations involve changes in
electronic states with relatively small volume
changes. They tend to proceed by shuffle-
dominated mechanisms. However, shear
transformations involving large structural
changes in terms of coordination number
or volume changes have also been reported
in inorganic and organic compounds, min-
erals, ceramics, organic compounds, and
some crystalline compounds of cement.
Some of the most prominent examples are
given in Table 9-8 (Kriven, 1982, 1988).
Because of its technological interest as a
toughener for brittle ceramic materials, zir-
conia is considered as the prototype of
martensite in ceramic materials. On cool-
ing, the high-temperature cubic phase of
zirconia transforms at 2370 °C to a tetrago-
nal phase. On further cooling, bulk zirconia
transforms at 950 °C to a monoclinic phase
with a volume increase of 3%. The latter
transforms on heating at about 1170 °C.
The monoclinic to tetragonal phase trans-
formation is considered to be martensitic.
The Ms temperature can be lowered sub-
stantially even below room temperature by
alloying or by reducing the powder size.
Small particles of zirconia, embedded in a
single-crystal matrix of alumina, remain
metastable (= tetragonal) at room tempera- Figure 9-37. (a) Helix structure of the a- and b-
ture for particle diameters less than a criti- modification of PTFE, and (b) dilatometric deter-
cal diameter (Rühle and Kriven, 1982). mination of relative volume changes and tempera-
ture range of transformations of PTFE (Hornbogen,
These metastable particles can transform to 1978).
the monoclinic phase under the action of an
applied stress, and it is this property that is
exploited in toughening brittle ceramics depends strongly on thermal activation.
(see Becher and Rose (1994)). However, in PTFE (polytetrafluoroethy-
Polymorphism is known to occur in lene) the conditions for no or weak thermal
several crystalline polymeric materials. In activation are fulfilled, and the transforma-
most of these systems the transformation tion can then be regarded as a diffusionless
9.10 Special Properties and Applications 641
or martensitic transformation (Hornbogen, Table 9-9. Contribution to as-quenched martensite

1978). This polymer crystallizes as parallel strength in 0.4 wt.% C steel (Williams and Thomp-
son, 1981).
arrangements of molecular chains parallel
to the c-axis. The atoms along the chains Boundary strengthening 620 MPa
are arranged as helices; the period along Dislocation density 270 MPa
the c-axis in the a-modification is 13 C2F4 Solid solution of carbon 400 MPa
units while it is 15 units in the b-modifica- Rearrangement of C in quench 750 MPa
Other effects 200 MPa
tion. The transformation from the a- to the
b-helix occurs at about 19 °C (Fig. 9-37). 0.2% yield strength 2240 MPa
Relaxation of the helix during this transfor-
mation does not lead to an extension of the
specific length of the molecules in the c-di-
liams and Thompson (1981) consider mar-
rection. The diameters of the molecule and
tensite as one of the most complex cases of
thus the lattice parameters in the a-direc-
combined strengthening. The hardness of
tion increase, which leads to an increase
martensite in as-quenched carbon steel de-
of about 1% in the specific volume. The
pends very much on the carbon content. Up
observed shape change can be increased if
to about 0.4 wt.% C the hardening is lin-
the molecules have been aligned by plastic
ear; retained austenite is present in steel
deformation. An analysis of the shape
containing more carbon, which reduces the
changes leads to the conclusion that the
rate of hardening. Solute solution harden-
PTFE transformation is diffusionless by a
ing by the interstitial carbon atoms is very
free volume shear, a type of transformation
substantial, whereas substitutional solid
not yet known in metallic and inorganic
solution hardening is low. For example,
materials.
Fe – 30 wt.% Ni martensites, where the car-
Biological materials consisting of crys-
bon content is very low, are not very hard.
talline proteins also undergo martensitic
The hardening of martensite is not due only
transformations in performing their life
to interstitial solute solution hardening,
functions. In a review entitled “Martensite
however. The martensite contains a large
and Life”, Olson and Hartman (1982) dis-
number of boundaries and dislocations,
cuss some examples. The tail-sheath con-
and the carbon atoms may rearrange during
traction in T 4 bacteriophages can be de-
the quench forming clusters that cause ex-
scribed as an irreversible strain-induced
tra dislocation pinning. The various contri-
martensitic transformation, while polymor-
butions to the strength of a typical C-con-
phic transformations in bacterial flagellae
taining martensite are given in Table 9-9,
appear to be stress-assisted and exhibit a
from which it becomes evident that inter-
shape-memory effect.
stitial solid solution hardening is not the
most important cause. Because many mar-
9.10 Special Properties tensitic steels are used after tempering,
and Applications lower strengths than those shown in Table
9-9 are found.
9.10.1 Hardening of Steel
9.10.2 The Shape-Memory Effect
Much of the technological interest con-
cerns martensite in steels. In a review on A number of remarkable properties have
strengthening of metals and alloys, Wil- their origin in a martensitic phase trans-
formation, such as the shape-memory ef-

fect, superelasticity, rubber-like behavior
and pseudoelasticity. The most fascinating
property is undoubtedly the shape-memory
effect. Review articles were published by
Delaey et al. (1974), Otsuka and Shimizu
(1986), Schetky (1979), and Junakubo
(1987). Recently, Van Humbeeck (1997) Figure 9-38. Schematic illustration of the shape-
prepared a review on Shape Memory Mate- memory effect: (a) and (e) parent phase; (b), (c) and
rials: “State of the Art and Requirements (d) martensite phase (Otsuka and Shimizu, 1986).
for Future Applications”. His review con-
tains 104 references to recent articles on
the topic. fect”. A more visual description of these
A metallic sample made of a common two effects is given in Fig. 9-38 and a clar-
material (low-carbon steel, 70/30 brass, ifying example is shown in Fig. 9-39,
aluminum, etc.) can be plastically de- where the applicability of the one-way
formed at room temperature. The macro- shape-memory effect is given for a space-
scopic shape change resulting from this craft antenna.
deformation will remain unchanged if the The shape that has to be remembered
sample is heated to higher temperatures. must, first of course, be given to the speci-
The only observable change in property men. This is done by classical plastic de-
may be its hardness, provided that the tem- formation by either cold or hot working.
perature to which the sample has been This process, however, may not involve
heated is above the recrystallization tem- any martensite formation. The material
perature. Its shape, however, remains as it must therefore be in a special metallurgical
was after plastic deformation. If the sample condition, which may require additional
is made of a martensitic shape-memory thermal treatments. In Fig. 9-40, for
material and is plastically deformed (bent, example, depicting a temperature-actuated
twisted, etc.) at any temperature below Mf shape-memory switch, two different “re-
and subsequently heated to temperatures membering” shapes are used. The initial
above Af , we observe that the shape that shape may be obtained by hot extrusion or
the specimen had prior to the deformation wire drawing and may or may not receive
starts to recover as soon as the As tempera- an additional cold or hot working in order
ture is reached and that this restoration is to obtain the required shape. The shapes
completed at Af . This behavior is called the formed must receive a heat treatment, con-
“shape-memory effect”, abbreviated to SME. sisting of a high-temperature annealing,
If the SME sample is subsequently followed by water quenching. The speci-
cooled to a temperature below Ms and its men is now martensitic, provided that the
shape remains unchanged on cooling, we composition is such that Mf is above room
talk about the “one-way shape-memory ef- temperature. In order to induce the shape
fect”. If it spontaneously deforms on cool- memory, the martensitic specimens are
ing to temperatures below Ms into a shape bent either to be curved or to be straight
approaching the shape that it had after the and are placed into the actuator at room
initial plastic deformation, the effect is temperature. If the temperature of the actu-
called the “two-way shape-memory ef- ator exceeds the reverse transformation
temperature of the shape-memory material, perature, we do not expect further shape

the specimens recover towards their “re- changes to occur. In order to reuse the
membered” position. The electrical con- specimens after having performed the
tacts are either closed or opened. shape-memory effect, they must be bent to
Special procedures for handling of the be either curved or straight again. Reheat-
shape-memory device are needed if we ing these deformed specimens for a second
want to induce the two-way memory effect. time to temperatures above Af will result in
This can be explained by again taking the shape memory. If this cycle, bending – heat-
temperature-actuated switch as an exam- ing – cooling, is repeated several times,
ple. If the specimen taken in its remem- gradually a two-way memory sets in. Dur-
bered position is cooled back to room tem- ing cooling the specimen reverts spontane-
Figure 9-39. Application of

Nitinol for a shape-memory
spacecraft antenna. From
“Shape Memory Alloys” by
L. McDonald Shetky. Copy-
right (1979) by Scientific
American, Inc. All rights re-
served.
Figure 9-40. Temperature-actuated switch designed so that it opens or closes above a particular temperature.
From “Shape Memory Alloys” by L. McDonald Shetky. Copyright (1979) by Scientific American, Inc. All
rights reserved.
ously to its “deformed” positions, thus in the third deformation step, because it
opening or closing the electrical contacts may not exceed the maximum strain that
on cooling. This repeated cycling, defor- can be recovered by the phase transforma-
mation in martensitic condition followed tion itself. Because these strains are asso-
by a heating – cooling, is called “training”. ciated with the martensitic transformation,
We can thus induce two-way memory by the maximum amount of recoverable strain
using a training procedure. is bound to the crystallography of the trans-
A further comment should be made here formation. Exceeding this amount of de-
concerning the shapes that can be remem- formation in the third fabrication step will
bered. We have to distinguish three shaping automatically result in unrecoverable de-
procedures: the fabrication step from raw formation.
material towards, for example, a coiled Many examples of shapes that can be re-
wire such as for the antenna, the fabrica- membered are possible. A flat SME speci-
tion of the “to be remembered position”, men can elongate or shorten during heat-
such as the additional shaping for the actu- ing, can twist clockwise or counter-clock-
ator, and the final deformation in the mar- wise, and can bend upwards or downwards.
tensitic condition, such as the bending of An SME spring can expand or contract dur-
the actuator. The first two fabrication steps ing heating. All this depends on the second
involve only classical plastic deformation and third fabrication steps.
and, therefore, the type and degree of de- What happens now if, for one reason or
formation are in principle not limited, pro- another, the specimen is restrained to ex-
vided that the material does not fail. The hibit the shape-memory effect? For exam-
degree of deformation, however, is limited ple, what happens if an expanded ring is
fitted as a sleeve over a tube with an outer

diameter slightly smaller than the inner di-
ameter of the expanded ring but larger than
the inner diameter of the ring in the remem-
bered position? On heating, the ring will
start to shrink as soon as the temperature As
is reached. While shrinking it will touch
the tube wall and further shrinking will be
hindered. From this moment, a compres-
sive stress will be built up, clamping the
shrinking ring around the tube. Obviously,
the composition of the alloy should be such
that the Ms temperature is below the value
at which clamping is required; in many
cases, this is below room temperature. For Figure 9-41. A simple shape-memory window
clamping rings it is therefore important that opener made from a copper-based shape-memory al-
loy. From “Shape Memory Alloys” by L. McDonald
on cooling back to room temperature the Shetky. Copyright (1979) by Scientific American,
clamping stress is still present. This means Inc. All rights reserved.
that two-way memory must be avoided,
which is easily achieved by choosing a
shape-memory alloy that exhibits a large
temperature hysteresis.
On heating a shape-memory device,
stresses can thus be built up and mechanical
work can be done. The latter would be the
case if a compressed shape-memory spring
has to lift a weight as in a shape-memory Figure 9-42. A mechanism in which a shape-mem-
actuated window opener (Fig. 9-41). A ory alloy (SMA) spring is used in conjunction with a
very useful device is realized when a bias spring. From “Shape Memory Alloys” by L.
shape-memory device is used, as shown McDonald Shetky. Copyright (1979) by Scientific
American, Inc. All rights reserved.
in Fig. 9-42, in combination with a bias
spring made of a conventional linear elastic
material, both being clamped between two
fixed walls and attached to each other with its original length, and further heating is re-
a plate. At temperatures below Mf , the quired to overcome the force exerted by the
shape-memory spring is closed and com- bias spring. At a certain temperature higher
pressed by the bias spring. The SME spring than Af the shape-memory spring will be
had to be deformed, in this case com- fully recovered. This temperature will de-
pressed, in order to fit into the clamping pend on the strength of the bias spring.
unit. The clamping unit with the two During this temperature excursion, the
springs installed is now heated to tempera- plate that is fixed between the two springs
tures above Af ; as soon as As is reached, the will have moved and can, if an “engine” is
SME spring will start to expand and try to attached to it, deliver work. If the clamping
push back the bias spring. At Af , the shape- unit is now cooled, the bias spring will try
memory spring will not yet have regained to compress the shape-memory spring into
its deformed position. The elastic energy mainly because of economic factors and
that has been stored in the bias spring dur- the reliability of the material. The three al-
ing the heating cycle is now released, alloy systems are Ni – Ti, Cu – Zn – Al and
lowing the plate to perform work also dur- Cu – Al – Ni. Generally, other elements are
ing the cooling cycle. To describe fully added in small amounts (of the order of a
such a working performing cycle, a ther- few weight %) in order to modify the trans-
modynamic treatment is needed (Wollants formation temperatures or to improve the
et al., 1979). The working performing mechanical properties or the phase stabil-
cycle can best be illustrated by taking a ity. In all three cases the martensite is
shape-memory spring that expands or con- thermoelastic. Maki and Tamura (1987)
tracts during heating or cooling and that reviewed the shape-memory effect in fer-
carries a load. The working performing cy- rous alloys, where a non-thermoelastic
cle can then be represented in a displace- Fe – Mn – Si alloy has also been found to
ment – temperature, a stress – temperature, show a shape-memory effect, and commer-
or an entropy – temperature diagram. cialization is being considered. The most
Although the shape-memory effect has important properties of shape-memory al-
been observed in many alloy systems, only loys are summarized in Fig. 9-43, in which
three systems are commercially available, the working temperatures, the width of the
Figure 9-43. Schematic representation of the most relevant shape-memory properties (courtesy Van Hum-
beeck, 1989).
uted here to domain-wall motion, as shown

in Fig. 9-44.
9.10.3 High Damping Capacity

The hysteresis exhibited during a pseu-
doelastic loading and unloading cycle is a
measure of the damping capacity of a vi-
brating device fabricated from a shape-
memory material, which is cycling under
extreme stress conditions exceeding the
Figure 9-44. Schematic illustration of the mecha- critical stress needed to induce martensite
nism for an electronic micro-positioning (Lemons
and Coldren, 1978).
by stress. Vibrating fully martensitic sam-
ples also exhibit high damping. A fully
martensitic sample consists of a large
number of differently crystallographically
hysteresis, and the maximum recoverable oriented domains whose domain boundar-
strain are given. Because of the superior ies are mobile. Under the action of an ap-
mechanical, chemical and shape-memory plied stress these boundaries move but, be-
properties of Ni – Ti alloys, this alloy sys- cause of friction, energy is lost during this
tem has been applied most successfully; movement. If a cyclic stress is applied, this
about 90% of the present applications use foreward and backward boundary move-
these alloys. Owing to the continuous im- ment will lead to damping of the vibration.
provement of the properties of Cu-based Comparing the amount of this damping
alloys, together with their lower price, Cu with the damping that we observe in other
SME alloys have been successfully used in non-SME alloy systems, it is found that the
several applications. damping capacity of martensitic shape-
The commercial applications of shape- memory alloys is one of the highest. The
memory devices can be divided into four shape-memory alloys are said to belong to
groups: the high-damping materials, the so-called
1. motion: by free recovery during heat- HIDAMETS.
ing and/or cooling;
2. stress: by constrained recovery during 9.10.4 TRIP Effect
heating and/or cooling;
3. work: by displacing a force, e.g., in TRIP is the acronym for TRansforma-
actuators; tion-Induced Plasticity and occurs in some
4. energy storage: by pseudoelastic load- high-strength metastable austenitic steels
ing of the specimen. exhibiting enhanced uniform ductility
when plastically deformed. This uniform
Shape-memory effects have also been macroscopic strain, up to 100% elongation,
reported in non-martensitic system, e.g., accompanies the deformation-induced
in ferroelectric ceramics (Kimura et al., martensitic transformation and arises from
1981), and have found applications as a plastic accommodation process around
micro-positioning elements (Lemons and the martensite plates. This macroscopic
Coldren, 1978). The shape change is attrib- strain thus contrasts with that occurring in
Figure 9-45. Idealized stress-assisted and strain-in-

duced regimes for mechanically-induced nucleation
(Olson and Cohen, 1982 a).
shape-memory alloys in being unrecover- Figure 9-46. Transformation-induced plasticity in

able. tensile tests at various temperatures (Fe – 29 wt.%
Ni – 0.26 wt.% C) (Tamura et al., 1969).
TRIP has been extensively studied by
Olson and Cohen (1982 a) and we will fol-
low their approach here. They distinguish
two modes of deformation-induced trans- stresses at which this strain-induced mar-
formation, according to the origin of the tensite is first detected follows the curve
nucleation sites for the martensite plates: BD. At point D, fracture occurs and thus
“stress-assisted” and “strain-induced” trans- determines the highest temperature Md at
formation. The condition under which each which martensite can be mechanically in-
mode can operate is indicated in a tempera- duced.
ture – stress diagram as shown in Fig. 9-45. When the transformation occurs at tem-
At temperatures slightly higher than Mss , peratures below Mss , the plastic strain is
the stress required for stress-assisted nucle- due entirely to transformation plasticity re-
ation on the same nucleation sites follows sulting from the formation of preferential
the line AB. At B, the yield point for slip martensite variants. The volume of the in-
in the parent phase is reached, defining duced martensite is therefore linearly re-
the highest temperature Mss for which the lated to the strain. The existing nucleation
transformation can be induced solely by sites are aided mechanically by the thermo-
elastic stresses. Above this temperature, dynamic contribution of the applied stress,
plastic flow occurs before martensite can reducing the chemical driving force for
be induced by stress. New strain-induced nucleation. Above Mss , the relationship
nucleation sites are formed, contributing to between strain and volume of martensite
the kinetics of the transformation. The becomes more complex, because strain is
9.11 Recent Progress in the Understanding of Martensitic Transformations 649
then a result of plastic deformation of the some new approaches are also explained.
parent phase and of transformation plastic- Most of the information referred to in this
ity. Strain hardening and enhancement of section was presented at the most recent
nucleation of martensite also play an essen- ICOMAT international conference on mar-
tial role. When martensite is formed during tensitic transformations held in 1998 at
tensile deformation, the strain hardening Bariloche (Ahlers et al., 1999).
becomes large. Necking is then expected New directions in martensite theory are
to be suppressed, explaining the enhanced presented by Olson (1999). The nucleation
uniform elongation. Fig. 9-46 shows, as an of martensite, the growth of a single mar-
example, the amount of martensite, the tensite plate, the formation of, for example,
elongation and the ultimate strength meas- self-accommodating groups of martensite
ured after tensile tests of a TRIP steel as a plates, and this within single crystals of the
function of temperature, clearly illustrating parent phase as well as in polycrystalline
the enhanced elongation, especially in the material, and the constraints dictated by the
temperature range between Mss and Md . components where martensitic materials
Such a large elongation (sometimes over are only one (maybe the most important)
200%) can also be produced by subjecting functional element of the component, are
a TRIP steel specimen under constant load all influenced by different interactive lev-
to thermal cycles through the transforma- els of structures (ranging from solute atoms
tion temperature. to components). Nucleation is the first step
in martensite life, and a component whose
functional properties are attributed to those
9.11 Recent Progress in the of martensite, can be considered the final
Understanding of Martensitic step. Olson (1999) constructed a flow-
Transformations block diagram in which the martensitic
transformation is situated in a multilevel
We draw attention here to some recent dynamic system. This new system, shown
papers that illustrate recent progress in the in Fig. 9-47, and the one given in Fig. 9-1,
understanding of martensite, and in which offer powerful tools for a better under-
Figure 9-47. The flow-block diagram

of martensitic transformation as a
multilevel dynamic system (Olson,
1999).
standing of diffusionless phase transforma- As is commonly known, the mechanism

tions. Its use should lead to a better design of bainite transformation is a subject with
of martensitic and bainitic alloys meeting many unresolved issues. Bhadeshia (1999)
specific requirements. Close analysis of the gives an overview of the transformation
papers presented at ICOMAT 98 shows that mechanisms proposed to explain “among
such an approach can already be found in others” the growth of bainite. The develop-
many papers. ment of bainite at both high temperatures
Special attention is given to the influ- (upper bainite) and low temperatures
ence of external constraints, such as hydro- (lower bainite) is discussed and is illustrat-
static pressure, the application of a mag- ed in Fig. 9-48. According to Bhadeshia,
netic field, and to martensite formed in thin the unresolved issues are:
films prepared by either sputtering or rapid
the growth rate of an individual bainite
solidification. Kakeshita et al. (1999) stud-
plate
ied the influence of hydrostatic pressure in-
a theory explaining the kinetics to esti-
stead of uniaxial stress, in order to formu-
mate the volume fraction of bainite in
late a thermodynamic approach for a better
austenite obtained during an isothermal
understanding of the nucleation of marten-
transformation
site. The strengthening mechanisms in
the modeling describing quantitatively
steel due to martensite are reviewed, the
the formation of carbides
diffusion of carbon in the various states
and a number of features associated with
(according to the dynamic system of Fig.
the interaction between plastic deforma-
9-47) of martensite is highlighted. In this
tion and bainite formation.
context, the fracture mechanism is related
to the tempering temperature and the car- The influence of carbon on the bainitic
bon diffusion. transformation is treated in great detail and
is shown to be a controlling factor of the
mechanical properties of different multi-
phase TRIP-assisted steels (Girault et al.,
1999; Jacques et al., 1999).
The martensitic transformation in
Fe – Mn-based alloys is treated in various
papers, showing the increasing interest in
developing ferrous shape-memory alloys.
In these alloys, austenite transforms either
into a h.c.p. e-phase (g Æ̈ e) or/and into
a¢-martensite (g Æ̈ a¢).
New approaches and strategies are dis-
cussed for the application of shape-mem-
ory alloys in non-medical (Van Humbeeck,
1999) as well as in medical applications
(Duerig et al., 1999). Only two examples
are shown here. The first example (Fig.
Figure 9-48. A schematic representation of the 9-49) shows that Ni – Ti superelastic alloys
mechanism explaining the growth and development improve significantly the cavitation ero-
of bainite (Bhadeshia, 1999). sion resistance if compared with marten-
9.12 Acknowledgements 651
Figure 9-49. The weight loss of a martensitic (Ni

Ti – 1) and a pseudoelastic (Ni Ti – 2) Ni – Ti shape-
memory alloy (Richman et al., 1994).
sitic Ni – Ti. But it should be remarked that

this figure is only an enlargement of a fig-
ure giving an overall view of the cavitation
resistance of other common alloys. For ex- Figure 9-50. A shape-memory device for repairing
ample, the weight loss after 10 h is already defects in the heart wall (Duerig et al., 1999).
20 – 30 mg for the “common” alloys in
comparison with the negligible weight loss
of both Ni – Ti shape-memory alloys after
10 h tested under the same conditions. A are currently being developed and mar-
very impressive example of the application keted.
of Ni – Ti shape-memory alloys is given in
Fig. 9-50. This figure shows an atrial septal
occlusion device with Nitinol (Ni – Ti 9.12 Acknowledgements
shape-memory alloy) wires incorporated in
a sheet of polyurethane. This device allows The author would like to thank M. Ah-
holes in the atrial wall of the heart to be lers, J. W. Christian, M. De Graef, R. Gott-
closed without surgery. The two umbrella- hardt, P. Haasen, H. S. Hsu, J. Ortín, K. Ot-
like devices are folded in two catheters, suka, J. Van Humbeeck and P. Wollants for
which are placed on either side of the support and advice while preparing the
hole. Once the two folded umbrellas are manuscript, and M. Van Eylen, M. Nol-
withdrawn from their catheters, they are mans, H. Schmidt and K. Delaey for their
screwed together in such a way that the assistance. The “Nationaal Fonds voor We-
hole is closed. Because of the flexibility of tenschappelijk Onderzoek” of Belgium is
both materials, the heart can again beat acknowledged for financial support (pro-
normally. This device illustrates the con- ject No. 2.00.86.87). The author especially
cept of the elastic development capacity acknowledges the continuing interest and
of shape-memory alloys. Because Ni – Ti encouragement he received from A. De-
shape-memory alloys have proposed to be ruyttere. For help in preparing the revised
biocompatible (see Van Humbeeck, 1977), version, I would like to thank M. Chandra-
many applications of these Ni – Ti alloys sekaran.
9.13 References Delaey, L., Krishnan, R. V., Tas, H., Warlimont, H.

(1974), J. Mat. Sci. 9, p. 1521; 1536; 1545.
Delaey, L., Deruyttere, A., Aermandt, E., Roos, J. R.
Aaronson, H. I., Reynolds, W. T. Jr (1988), in: Phase (1978), MTM Internal. INCRA Report. Report
Transformations ’87: Lorimer, G. W. (Ed.). Lon- 78 R1. Leuven: Dep. Met. & Mat. Eng., KV Leu-
don: The Institute of Metals, p. 301. ven.
Ahlers, M. (1974), Z. Metallkd. 65, 636. Delaey, L., Gobin, P. F., Guénin, G., Warlimont, H.
Ahlers, M. (1982), in: International Summer Course (1979), in: Proc. ICOMAT ’79. Cambridge (MA):
on Martensitic Transformations. Leuven: Dep. MIT, p. 400.
Met. & Mat. Eng., KV Leuven, p. 2.1. Delaey, L., Chandrasekaran, M., Andrade, M., Van
Ahlers, M., Kostorz, G., Sade, M. (1999), ICOMAT Humbeeck, J. (1982 a), in: Proceedings of Solid to
’98, International Conference on Martensitic Solid Transformations. New York: AIME, p. 1429.
Transformations, 7 – 11 December 1998, S. C. de Delaey, L., Mukherjee, K., Chandrasekaran, M.
Bariloche, Argentine Mater. Sci. Eng. A 273 – 275. (1982 b), in: International Summer Course on
Bain, E. C. (1924), Trans. AIME 70, 25. Martensitic Transformations. Leuven: Dep. Met.
Ball, J. M., James, R. D. (1988), in: Phase Transfor- & Mat. Eng., KV Leuven, p. 7.1.
mations ’87: Lorimer, G. W. (Ed.). London: The Duerig, T., Pelton, A., Stöckel, D. (1999), Mater. Sci.
Inst. of Metals, p. 425. Eng. A 273 – 275, 149.
Barsch, G. R., Krumhansl, J. A. (1988), Metall. Falk, F. (1982), in: ICOMAT ’82: Editions de Phy-
Trans. 19 A, 761. sique C 4, p. 3.
Becher, P. F., Rose, L. R. F. (1994), in: Materials Sci- Fujita, F. E. (1982), in: International Summer Course
ence and Technology, Vol. 11: Swain, M. (Ed.). on Martensitic Transformations. Leuven: Dep.
Weinheim: VCH, p. 409. Met. & Mat. Eng., KV Leuven, p. 10.1.
Bevis, M. J., Allan, P. S. (1974), Surface and Defect Funakubo, H. (1987), Shape Memory Alloys. New
Properties of Solids 3, 93. York: Gordon and Breach.
Bhadeshia, H. K. D. H. (1999), Mater. Sci. Eng. Girault, E., Jacques, P., Ratchev, P., Van Humbeeck,
A 273 – 275, 58. J., Verlinden, B., Aernoudt, E. (1999), Mater. Sci.
Bidaux, J. E. (1988), Ph.D. Thesis. Lausanne: EPFL. Eng. A 273 – 275, 471.
Bilby, B. A., Christian, J. W. (1961), J. Iron Steel Greninger, A. B., Troiano, A. R. (1949), Metall.
Inst. 197, 122. Trans. 185, 590.
Bilby, B. A., Frank, F. C. (1960), Acta Metall. 8, Guénin, G., Clapp, P. C. (1986), Proc. ICOMAT ’86.
239. Sendai: Japanese Institute of Metals, p. 171.
Binder, K. (2001), in: Phase Transformations in Ma- Guénin, G., Gobin, P. F. (1982), in: ICOMAT ’82:
terials: Kostorz, G. (Ed.). Weinheim: Wiley-VCH, Editions de Physique C 4, p. 57.
Chap. 4. Hara, T., Ohba, T., Otsuka, K. (1998), Proc. PRIM-3,
Bowles, J. S. (1951), Trans. AIME 191, 44. Hawaii, TMS, p. 1199.
Bowles, J. S., Mackenzie, J. K. (1954), Acta Metall. Hocke, U., Warlimont, H. (1977), J. Phys. F 7, 1145.
2, 158; 224. Horikawa, H., Ichinose, S., Morii, K., Miyazaki, S.,
Buerger, M. J. (1951), in: Phase Transformations in Otsuka, K. (1988), Metall. Trans. 19 A, 915.
Solids: Smoluchowski et al. (Eds.). New York: Wi- Hornbogen, E. (1978), Progr. Colloid Polymer Sci.
ley. 64, 125.
Bullough, R., Bilby, B. A. (1956), Proc. Phys. Soc. B, Hornbogen, E. (1988), Pract. Metall. 25, 157.
LXIX, 1276. Hsu, T. Y. (1980), in: Martensitic Transformation
Christian, J. W. (1965), The Theory of Transforma- and Martensite: Hsu, T. Y. (Xu Zuyao) (Ed.).
tions in Metals and Alloys. Oxford: Pergamon Shanghai: Science Press, p. 10.
Press. Hsu, T. Y. (1985), J. Mat. Sci. 20, 23.
Christian, J. W. (1976), Suppl. Trans. JIM. 17, 21. Hsu, T. Q., Zhou Xiaowang (1989), Acta Metall. 37,
Christian, J. W. (1982), in: International Summer 3091.
Course on Martensitic Transformations. Leuven: Ichinose, S., Funatsu, Y., Otsuka, K. (1985), Acta
University of Leuven, p. 5.1. Metall. 33, 1613.
Christian, J. W., Knowles, K. M. (1982), in: Proceed- James, R. D. (1988), in: Materials Instabilities in
ings of Solid to Solid Transformations. New York: Continuum Mechanics and Related Mathematical
AIME, p. 1185. Problems: Ball, J. M. (Ed.). Oxford: Univ. Press,
Cohen, M. (1982), in: International Summer Course p. 175.
on Martensitic Transformations. Leuven: Dep. Jacques, P., Girault, P., Catlin, T., Geerlofs, N., Kop,
Met. & Mat. Eng., KV Leuven, p. 1.1. T., Van der Zwaag, S., Delannay, F. (1999), Mater.
Cohen, M., Olson, G. B., Clapp, P. C. (1979), in: Sci. Eng. A 273 – 275, 475.
Proc. ICOMAT ’79. Cambridge (MA): MIT, p. 1. Kajiwara, S., Kikuchi, T., Uehara, S. (1986), Proc.
de Fontaine, D. (1973), in: Phase Transitions 1973. ICOMAT ’86. Sendai: Japanese Institute of Metals,
New York: Pergamon Press, p. 169. p. 301.
9.13 References 653
Kakeshita, T., Saburi, T., Shimizu, K. (1999), Mater. lin, D. E., Sekerka, D. F., Wayman, C. M. (Eds.).
Sci. Eng. A 273 – 275, 21. Warrendale: Met. Soc. AIME, p. 1145.
Kaufman, L. (1965), in: Physical Properties of Mar- Olson, G. B., Cohen, M. (1986), in: Dislocations in
tensite and Bainite: Special Report 93. London: Solids: Nabarro, F. R. N. (Ed.). New York: Else-
The Iron & Steel Inst., p. 48. vier Science Publishers, p. 295.
Kaufman, L., Cohen, M. (1958), Progr. Metal Phys. Olson, G. B., Hartman, H. (1982), Proc. ICOMAT
7, 165. ’82, J. de Physique 43, 855.
Krauss, G. (1992), in: Materials Science and Tech- Oota, A., Muller, J. (1987), Proc. ICOMAT ’86. Sen-
nology, Vol. 7: Pickering, F. B. (Ed.). Weinheim: dai: Japanese Institute of Metals, p. 1121.
VCH, p. 1. Ortin, J., Planes, A. (1989), Acta Metall. 37, 1433.
Krauss, G., Marder, A. R. (1971), Metall. Trans. 2, Osmond, M. F. (1895), Arts Chimiques 94, 480.
2343. Otsuka, K. (1986), Proc. ICOMAT ’86. Sendai: Japa-
Kriven, W. M. (1982), in: Proceedings of Solid to nese Institute of Metals, p. 35.
Solid Transformations. New York: AIME, p. 1502; Otsuka, K., Shimizu, K. (1986), Int. Metals. Rev. 31,
and in: International Summer Course on Marten- 93.
sitic Transformations. Leuven: Dep. Met. & Mat. Otsuka, K., Wayman, C. M. (1977), in: Reviews on
Eng., KV Leuven, p. 9.1. the Deformation Behavior of Materials, Feltham,
Kriven, W. M. (1988), J. Am. Ceram. Soc. 71, 1021. P. (Ed.). Intl. Quarterly Scient. Reviews Journal II,
Kurdjurnov, G. V. Sachs, G. (1930), Z. Phys. 64, 325. 81.
Lemons, R. A., Coldren, L. a. (1978), Rev. Sci. In- Otsuka, T., Ren, X. (1999), Mater. Sci. Eng. A 273 –
strum. 49, 1650. 275, 89.
Liao, K. C., Allen, C. W. (1982), in: Solid-State Patel, J. R., Cohen, M. (1953), Acta Metall. 1, 531.
Phase Transformations: Aaronson, H. I., Laugh- Pops, H. (1970), Metall. Trans. 1, 251.
lin, D. E., Sekerka, D. F., Wayman, C. M. (Eds.). Pelton, A. (2001) in: Phase Transformations in Ma-
Warrendale: Met. Soc. AIME, p. 1493. terials: Kostorz, G. (Ed.). Weinheim: Wiley-VCH,
Ling, H. C., Owen, W. s. (1981), Acta Metall. 29, Chap. 1.
1721. Purdy, G. R., Bréchet, Y. J. M. (2001), in: Phase
Liu, Y., Xia, Z., Van Humbeeck, J. (1999 a), Mater. Transformations in Materials: Kostorz, G. (Ed.).
Sci. Eng. A 273 – 275, 673. Weinheim: Wiley-VCH, Chap. 7.
Liu, Y., Xia, Z., Van Humbeeck, J., Delaey, L. Rao, C. N. R., Rao, K. J. (1978), Phase Transitions in
(1999 b), Mater. Sci. Eng. A 273 – 275, 679. Solids. New York: McGraw-Hill.
Maki, T., Tamura, I. (1987), Proc. ICOMAT ’86. Sen- Richman, R. H., Zimmerly, C. A., Inal, O. T., Hodg-
dai: Japanese Institute of Metals, p. 963. son, D. E., Rao, A. S. (1994), in: Proceedings
Massalski, T. B. (1984), Metall. Trans. 15 A, 421. International Conference on Shape Memory Alloys
Muddle, B. C. (1982), in: Proceedings of Solid to and Superplastic Technologies SMST ’94, Pelton,
Solid Transformations. New York: AIME, p. 1347. A., Hodgson, D., Duerig, T. (Eds.). Pacific Grove,
Mukherjee, K. (1982), in: International Summer CA, USA: SMTS Proceedings, p. 175.
Course on Martensitic Transformations. Leuven: Roitburd, A. L. (1988), in: Phase Transformations ’87:
Dep. Met. & Mat. Eng., KV Leuven, p. 4.1. Lorimer, G. W. (Ed.). Institute of Metals, p. 414.
Nakanishi, N. (1979), Progr. Mat. Sci. 24, 143. Rühle, M., Kriven, S. (1982), Proc. Phase Transf.
Nakanishi, N., Nagasawa, A., Murakami, Y. (1982), Confer. AIME, p. 1569.
in: ICOMAT ’82: Editions de Physique C 4, p. 35. Saburi, T., Nenno, S. (1987), Proc. ICOMAT ’86.
Nakanishi, N., Shigematsu, T., Machida, N., Ueda, Sendai: Japanese Institute of Metals, p. 671.
K., Tanaka, H., Inaba, T., Iwatsu, O. (1993), in: Salzbrenner, R. J., Cohen, M. (1979), Acta Metall.
International Conference on Martensitic Transfor- 27, 739.
mation ICOMAT ’92: Wayman, C. M., Perkins, J. Schetky McDonald, L. (1979), Sci. Am. 241, 74.
(Eds.). Carmel, CA, USA: Monterey Institute of Schroeder, T. A., Wayman, C. M. (1977), Acta Met-
Advanced Studies, p. 977. all. 25, 1375.
Nishida, M., Li, S. (2000), Mat. Sci. Forum 327/328, Shapiro, S. M. (1981), Metall. Trans. 12 A, 567.
103. Shibata, M., Ono, K. (1975), Acta Metall. 23, 587.
Nishiyama, Z. (1934), Sci. Rept. Tohoku Univ. 23, p. Shibata, M., Ono, K. (1977), Acta Metall. 25, 35.
325. Shimizu, K., Kakeshita, T. (1989), ISIJ Int. 2, 97.
Nishiyama, Z. (1978), Martensitic Transformations. Sikka, S. K., Vohra, Y. K., Chidambaram, R. (1982),
New York: Academic Press. Progr. Mat. Sci. 27, 245.
Olson, G. B. (1999), Mater. Sci. Eng. A 273 – 275, Sugimoto, K., Nakaniwa, M. (2000), Mat. Sci.
11. Forum 327/328, 363.
Olson, G. B., Cohen, M. (1982 a), Metall. Trans. A 6, Tamura, I., Maki, T., Hato, H., Aburai, K. (1969), J.
791. Jpn Inst. Met. 33, 1383.
Olson, G. B., Cohen, M. (1982 b), in: Solid-State Tas, H., Delaey, L., Deruyttere, A. (1973), Metall.
Phase Transformations: Aaronson, H. I., Laugh- Trans. 4, 2833.
Thadhani, N. N., Meyers, M. A. (1986), Progr. Mat. “International Conference on Martensitic Transfor-
Sci. 30, 1. mations ICOMAT ’79” (1979), Cambridge (MA):
Vallade, M. (1982), in: International Summer Course Dep. of Mat. Science and Techn., M.I.T.
on Martensitic Transformations. Leuven: Dep. “International Conference on Martensitic Transfor-
Met. & Mat. Eng., KV Leuven, p. 3.1. mations ICOMAT ’82” (1982), Delaey, L., Chan-
Van Humbeeck, J. (1977), J. Phys. (Paris) IV, 7-C5, 3. drasekaran, M. (Eds.), Journal de Physique 43,
Van Humbeeck, J. (1989), Dep. Met. & Mat. Eng., Conf. C-4.
University of Leuven, Leuven, unpublished. “International Conference on Shape Memory Al-
Van Humbeeck, J. (1999), Mater. Sci. Eng. A 273 – loys” (1986), Youyi Chu, Hsu, T. Y., Ko. T. (Eds.).
275, 134. Beijing: China Academic Publishers.
Van Tendeloo, G. (1989), University of Antwerp, “International Conference on Martensitic Transfor-
Antwerp, unpublished. mations ICOMAT ’86” (1987), Tamura, I. (Ed.).
Verma, A. R., Krishna, P. (1966), Polymorphism and Sendai: The Japan Inst. of Metals.
Polytypism. New York: Wiley. “The Martensitic Transformation in Science and
Vintaikin, E. Z., Litvin, D. F., Udovenko, V. A., Technology” (1989), Hornbogen, E., Jost, N.
Schcherebinskij, G. V. (1979), in: Proc. of ICO- (Eds.). Oberursel: DGM-Informationsgesellschaft
MAT ’79. Cambridge: MIT-Press, p. 673. Verlag.
Wadhawan, V. K. (1982), Phase Transitions 3, 3. “Shape Memory Materials” (1989), Otsuka, K., Shi-
Warlimont, H. (1989), Vacuumschmelze GmbH, Ha- mizu, K. (Eds.), Proceedings of the MRS Interna-
nau. tional Meeting on Advanced Materials, vol. 9.
Warlimont, H., Delaey, L. (1974), Progr. Mat. Sci. Pittsburgh: Mat. Research Soc.
18, p. 1. “International Conference on Martensitic Transfor-
Wayman, C. M. (1964), Introduction to the Crystal- mations ICOMAT ’89” (1990), Morton, A. J.,
lography of Martensitic Transformations. New Dunne, B., Kelly, P. M., Kennon, N. (Eds.) to be
York: MacMillan Comp. published.
Wayman, C. M. (1983), in: Physical Metallurgy: “International Conference on Martensitic Transfor-
Cahn, R. W., Haasen, P. (Eds.). New York: Else- mations ICOMAT ’92” (1993), Wayman, C. M.,
vier Scientific Publishers, p. 1031. Perkins, J. (Eds.). Carmel, CA, USA: Monterey In-
Wayman, C. M. (1987), Proc. ICOMAT ’86. Sendai: stitute of Advanced Studies.
Japanese Institute of Metals, p. 645. “International Symposium and Exhibition on Shape
Wechsler, M. S., Lieberman, D. S., Renard, T. A. Memory Materials (SME ’99)” (2000). Zurich:
(1953), Trans. AIME 197, 1503. Trans. Tech. Publications.
Williams, J. C., Thompson, A. w. (1981), in: Metal- “International Conference on Shape Memory and
lurgical Treatises: Tien, J. K., Elliott, J. F. (Eds.). Superelastic Technologies (SMST ’97)” (1997),
Warrendale: Met. Soc. AIME, p. 487. Pelton, A., Hodgson, D., Russell, S., Duerig, T.
Wollants, P., De Bonte, M., Roos, J. R. (1979), Z. (Eds.). Santa Clara, CA, USA: SMST Proceedings.
Metallkd. 70, 113. “Displacive Phase Transformations and Their Appli-
Yang, J., Wayman, C. M. (1993), in: International cations in Materials Engineering” (1998), Inoue,
Conference on Martensitic Transformation ICO- K., Mukherjee, K., Otsuka, K., Chen, H. (Eds.).
MAT ’92, Wayman, C. M., Perkins, J. (Eds.). Car- Warrendale, OH, USA: TMS Publications.
mel, CA, USA: Monterey Institute of Advanced “European Symposium on Martensitic Transforma-
Studies, p. 1187. tions ESOMAT ’97” (1997), Beyer, J., Böttger, A.,
Mulder, J. H. (Eds.). EDP-Sciences Press.
“European Symposium on Martensitic Transforma-
tions ESOMAT ’94” (1995), Planes, A., Ortin, J.,
Proceedings Lluis Manosa. Les Ulis, France: Les Editions de
of International Conferences Physique.
“International Conference on Martensitic Transfor-
on Martensitic Transformations mations ICOMAT ’98” (1999), Ahlers, M., Kos-
torz, G., Sade, M. (Eds.). Amsterdam: Elsevier.
“Physical Properties of Martensite and Bainite” “Martensitic Transformation and Martensite” (1980),
(1965), Special Report 93. London: The Iron and Hsu, T. Y. (Xu Zuyao) (Ed.). Shanghai: Science
Steel Institute. Press (in Chinese).
“Shape Memory Effects in Alloys” (1975), Perkins, “Pseudoelasticity and Stress-Induced Martensitic
J. (Ed.). New York: Plenum Press. Transformations” (1977), Otsuka, K., Wayman, C.
“New Aspects of Martensitic Transformations” M. (Eds.), in: “Reviews on the Deformation Be-
(1976), Suppl. to Trans. Jap. Inst. Met. 17. havior of Materials”.
“Martensitic Transformations ICOMAT ’77” (1978),
Kiev: Academy of Science.
10 High Pressure Phase Transformations
Martin Kunz
ETH Zürich, Labor für Kristallographie, Zürich, Switzerland

10.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 659
10.2 Pressure-Driven Phase Transitions . . . . . . . . . . . . . . . . . . . . 660
10.2.1 Framework flexion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 661
10.2.2 Increase in coordination number . . . . . . . . . . . . . . . . . . . . . . . 663
10.2.3 Pressure-induced ordering . . . . . . . . . . . . . . . . . . . . . . . . . . 664
10.3 Generating High Pressure . . . . . . . . . . . . . . . . . . . . . . . . . 666
10.3.1 Dynamic pressure generation . . . . . . . . . . . . . . . . . . . . . . . . 667
10.3.2 Static pressure devices . . . . . . . . . . . . . . . . . . . . . . . . . . . . 668
10.3.2.1 Large-volume presses . . . . . . . . . . . . . . . . . . . . . . . . . . . . 668
10.3.2.2 Diamond anvil cells . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 671
10.4 Probing Phase Transformations in Materials at High Pressure . . . . . 673
10.4.1 Volumetric techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . 674
10.4.2 Spectroscopic techniques . . . . . . . . . . . . . . . . . . . . . . . . . . 674
10.4.2.1 Microscopy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 674
10.4.2.2 Raman and infrared spectroscopy . . . . . . . . . . . . . . . . . . . . . . 674
10.4.2.3 Mössbauer spectroscopy . . . . . . . . . . . . . . . . . . . . . . . . . . . 675
10.4.2.3 X-Ray absorption spectroscopy . . . . . . . . . . . . . . . . . . . . . . . 676
10.4.2.5 NMR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 676
10.4.3 Ultrasonic sound velocity . . . . . . . . . . . . . . . . . . . . . . . . . . 677
10.4.4 Diffraction techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . 678
10.4.4.1 X-ray diffraction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 678
10.4.4.2 Neutron diffraction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 679
10.5 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 679
10.5.1 Zincblende-type semiconductors . . . . . . . . . . . . . . . . . . . . . . 679
10.5.1.1 Si and Ge . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 680
10.5.1.2 GaAs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 681
10.5.1.3 InSb . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 682
10.5.2 Materials in the B-C-N system . . . . . . . . . . . . . . . . . . . . . . . . 683
10.5.2.1 C . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 684
10.5.2.2 B–N . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 685
10.5.2.3 C–N . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 686
10.5.2.4 B–C . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 686
10.5.2.5 B–C–N . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 686
10.5.3 H2O . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 687
10.5.3.1 Ice Ih, XI and Ic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 688

ISBN: 3-527-30256-5
656 10 High Pressure Phase Transformations
10.5.3.2 Ice II . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 689

10.5.3.3 Ice III and Ice IX . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 689
10.5.3.4 Ice V, Ice IV and Ice XII . . . . . . . . . . . . . . . . . . . . . . . . . . . 690
10.5.3.5 Ice VI, Ice VII, Ice VIII, Ice X . . . . . . . . . . . . . . . . . . . . . . . 690
10.5.3.6 Amorphous ice at high pressure . . . . . . . . . . . . . . . . . . . . . . . 691
10.6 Acknowledgment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 692
10.7 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 692

a, b, c lattice parameters
E0 internal energy at zero pressure in shock experiment
EH Hugoniot internal energy in a shock experiment
G Gibbs energy
H enthalpy
l wavelength
n number of moles
P pressure
P0 ambient pressure
PH Hugoniot pressure
Q heat
r0 density at zero pressure
S entropy
T temperature
Tc critical temperature in superconductors
T0 temperature at zero pressure
TH Hugoniot temperature
U internal energy
UP sample velocity
Us shock velocity
V molar volume
V0 molar volume at zero pressure
VH molar volume in the shocked Hugoniot state
DVdis difference in molar volume due to an order/disorder process
Vdisordered molar volume of the disordered state
Vordered molar volume of the ordered state
W work
ADX angle dispersive X-ray diffraction

AX compound semiconductors of 1 : 1 stoichiometry
bcc body centered cubic
ccp cubic closest packing (= fcc)
ct carat
DAC diamond anvil cell
dhcp hexagonal closest packing with doubled c-axis
EDX energy dispersive X-ray diffraction
EXAFS extended X-ray absorption fine struture
fcc face centered cubic (= ccp)
hcp hexagonal closest packing
HDA high density amorphous phase of H2O
HEL Hugoniot elastic limit
HP high pressure
HT high temperature
IR infrared
LDA low-density amorphous phase of H2O
NMR nuclear magnetic resonance
RT room temperature
XAS X-ray absorption spectroscopy
10.1 Introduction the thermodynamics of solid or liquid

material (see also the Chapter by Binder,
The understanding of condensed matter 2001). In a more practical way, materials
requires a knowledge of the relationship scientists are interested in adding pressure
between temperature, pressure and chemi- to the variables temperature and chemical
cal environment on the one hand and vol- composition in order to stabilize new mate-
ume, bonding and electronic and magnetic rials with technically interesting proper-
structure on the other. In gases, pressure, ties. A prominent example of a class of ma-
temperature and volume largely follow a terials whose synthesis and exploration de-
simple and universal relationship, but the pend on the application of high pressure is
situation in condensed matter is much more the family of novel abrasives and super-
complicated. This is mainly due to inter- hard materials. In recent years, high-pres-
atomic interactions, which are much sure experiments are even used in the life
stronger in condensed material than in sciences. A popular key issue in this area
gases and cause structural, electronic and is the question on the origin of life (e.g.,
magnetic correlations. Pelton (2001, Chap. 1 Pedersen, 1997). A more applied interest
of this book) explores the interdependence of biology in high pressure is the search
between chemical composition and tem- for commercially viable alternative ways
perature in great detail. In this chapter, we of food sterilization (e.g., Ondrey and Ka-
will mainly focus on the effects of pressure miya, 2000; Thakur and Nelson, 1998;
on the state of solid material. Mermelstein, 1999).
The exploration of the physics and Forcing a given set of atoms into a small
chemistry of material under high pressure volume by applying high pressure in-
is of a remarkably inter-disciplinary inter- creases the interatomic interaction. This in
est. This is reflected in the variety of scien- turn leads to structural phase transitions
tific problems that are the focus of modern and changes in physical properties. The
high-pressure research. The understanding driving forces for high pressure-induced
of the interdependence between structural phase transitions are obviously linked to
distortions and electronic and magnetic the need to optimize the volume occupied
properties is a typical example of a physi- by the atoms for a given pressure. In Sec.
cal problem that benefits from experiments 10.2, we will discuss the most important
performed at high pressure. Another topic mechanisms, which are the immediate
within physics that relies on high-pressure cause of pressure-driven phase transitions.
experiments is the calibration and refine- Extracting information on solid or liquid
ment of theoretical models describing the material at high pressure (“high” pressure
interaction of atoms. High-pressure experi- in this chapter refers to pressures in the
ments on very simple covalent systems range 0.1 – 100 GPa) is an experimental
such as solid hydrogen or helium provide challenge and was first tackled by Percy
valuable data, which help to formulate fun- William Bridgman in the first half of this
damental concepts on the nature of matter. century (e.g., Bridgman, 1946). He was
A fascinating problem in condensed-matter awarded the Nobel Prize for physics in
chemistry is the structural change of a solid 1946 for his ground-breaking achieve-
(or liquid) and its phase transitions as a re- ments in this important field. Since the
action on an external pressure. Its under- time of Bridgman the technology necessary
standing offers chemists vital insights into to perform experiments at high pressure
has experienced extensive development. interactions between the ensemble of con-

Section 10.3 will briefly summarize the stituting atoms.
current state of the art of high-pressure ex- The absolute value of the internal energy
perimentation. Although third-generation cannot be measured. However, the differ-
synchrotron sources produced a minor rev- ence between two states is independent of
olution in high-pressure crystallography, the path and mechanism of the change of
X-ray diffraction is by no means the only the system (Hess’s law). This makes the
way of investigating material when sub- difference in the internal energy dU a very
jected to pressure. In Sec. 10.4, we will re- important quantity in comparing a system
view the most common probes used for at two different states:
physical and chemical investigations at
dU = T dS – P dV (10-2)
high pressure. Section 10.5 treats in detail
the high-pressure phase transitions for a se- Entropy S and volume V are two indepen-
lected sample of materials, which are of dent variables that are very difficult to con-
special interest within materials science. trol in an experiment. However, two suc-
cessive Legendre transformations of U
transform the internal energy first into the
10.2 Pressure-Driven Phase enthalpy H
Transitions
H = U + PV (10-3)
The physical quantities that define the and then into the Gibbs energy G
thermodynamic state of a system can be
differentiated into intensive quantities (i.e., G = H – TS = U + PV – TS (10-4)
temperature T, pressure P, chemical poten- The total exact differential of G yields
tial µ) and extensive quantities (i.e., en-
tropy S, volume V, number of moles n). dG = – S dT + V dP (10-5)
Conjugate quantities are pairs of an inten- which has the desirable property that the
sive and an extensive variable (i.e., T, S or independent variables T and P are easily
P, V ). Their products have the dimensions modified and controlled in an experiment.
of energy or a volume-normalized energy, This makes G the critical quantity, which
which can be used to describe the relative has to be considered when comparing a
stability of a given system. The thermody- system at different states. While Pelton
namic stability is determined by the mini- (2001) focuses on the variation of G as a
mum of the internal energy U, which is de- function of the chemical composition and
fined as a sum of a heat term Q and a work temperature, we will here investigate the
term W, change in G upon increasing pressure.
U=Q+W (10-1) If a system is subject to changing pres-
sure at constant temperature, Eq. (10-5) re-
where Q is a function of temperature and duces to
entropy and W is a function of pressure and
volume. On an atomistic level, Q can be dG = V dP (10-6)
viewed as the vibrational energy of the at-
The change in the Gibbs energy on chang-
oms oscillating around their equilibrium
ing pressure is thus
position. W can be visualized as the sum of
the potential energies from the inter atomic (∂G/∂P)T = + V (10-7)
10.2 Pressure-Driven Phase Transitions 661
Equation (10-7) states an intuitive trivial- tions (coesite can be quenched to room
ity, i.e., that in order to minimize the Gibbs pressure), both structures exhibit nearly
energy upon increasing pressure, a given ideal Si–O–Si angles of between 143° and
system will reduce its volume. Volume re- 144°. The phase transition is thus structu-
duction is thus the ultimate driving force of rally characterized by building the tetrahe-
structural change and phase transformation dral framework in coesite more densely
as a reaction on increasing pressure. How- than in a-quartz. This can be demonstrated
ever, the way a volume is reduced can be by looking at the oxygen surroundings in
different, depending on the initial structure both structures: In a-quartz, each oxygen
and configurational entropy of the system has three oxygen neighbors at ~ 2.6 Å,
and also on the amount of pressure applied. which represent the tetrahedral oxygen
In the following we will focus on the three neighbors. The closest oxygen atoms from
most important mechanisms responsible any other tetrahedron are found at a dis-
for structural changes at high pressure. tance of ~ 3.5 Å. In coesite on the other
These are framework flexion, increase in hand, there are also three oxygen neighbors
coordination number and pressure-induced at 2.6 Å around each O, indicating the stiff
ordering. behavior of the SiO4 tetrahedrals. The next
nearest oxygen neighbors in coesite, how-
ever, are observed at a distance of only
10.2.1 Framework flexion
3.0 – 3.2 Å (room pressure). Both structure
A number of technologically important types react to applied pressure with a very
oxide materials (e.g., zeolites, perovskites) large decrease in the Si–O–Si angles (Fig.
can be viewed as being built of relatively 10-1), while the Si–O bond lengths remain
rigid corner-linked polyhedra forming a more or less constant. Densification within
rather flexible framework. These materials the stability field of quartz is thus achieved
are known to react on changing tempera- by flexion of the framework through rigid
ture and/or pressure mainly by flexion at rotations of the SiO4 tetrahedra, resulting
the polyhedral joints rather than by polyhe- in a decrease of the Si–O–Si angle from
dral compression, i.e., bond length reduc- 144° to 125°. If this reaches a limit, denser
tion (e.g., Hazen and Finger, 1978; Velde packing of the polyhedra is achieved by
and Besson, 1982; Hemley et al., 1994). framework reconstruction. Further densifi-
Silica (SiO2 ) can be viewed as a typical ex- cation, however, eventually involves an in-
ample of the behavior of framework struc- crease in the coordination number (see Sec.
tures under pressure. It has a rich phase di- 10.2.2). Thus, a pure compression at R T or
agram in P – T space. The ambient condi- a rapid (shock) compression of a-quartz
tion phase is a-quartz, which consists of leads to an amorphization involving both
interconnected spirals of corner-linked framework collapse and an increase in co-
SiO4 tetrahedra. Silica has two high-pres- ordination number (Hemley et al., 1994).
sure polymorphs, of which only coesite can Pressure-induced amorphization seems to
be viewed as a tetrahedral framework be a common feature of many framework
structure. Its framework is characterized by structures, indicating a volume reduction
four-fold rings of tetrahedra connected to by a collapse of the framework. This de-
chains, which in turn are packed via shared stroys the long-range order while still
tetrahedral corners to the densest known largely maintaining the short-range order
tetrahedral framework. At ambient condi- of the primary building blocks. A similar
Figure 10-1. Si–O–Si angles (solid symbols) and average Si–O bond lengths (open symbols) vs. pressure for
quartz (squares) and coesite (diamonds). Note the (within experimental errors) constant ·Si–OÒ distances with in-
creasing pressure, while the Si–O–Si angles show a significant negative correlation. This demonstrates that frame-
work structures (such as tetrahedral SiO2 polymorphs) adapt their volume through framework flexion. Data from
Levien et al. (1980) and Levien and Prewitt (1981).
effect is also thought to be responsible for of the tetrahedral framework into a glassy
the high-density form of amorphization of tetrahedral arrangement.
ice (Mishima et al., 1984; see also Sec. If the channels and holes in a framework
10.5.3.6). are large enough, we can even obtain the
The most open frameworks known in in- seemingly paradox result of a negative
organic chemistry are adopted by the zeolite compressibility. This happens if the molec-
structure family. These structures exhibit ular size of the pressure medium is small
some peculiar effects upon compression. enough to be squeezed into the channels.
Zeolite frameworks are characterized by Because the pressure medium seems to be
large cages and channels, which can be oc- packed more densely within the zeolite
cupied by extra-framework cations or mole- structure than in the fluid, the total volume
cules. As in silica, volume reduction on zeo- of the system (crystal plus pressure me-
lite can be most easily achieved by a reduc- dium) decreases. The volume of the crystal
tion of the framework channels, which is alone, however, increases with applied
achieved through rigid body rotations of the pressure. In a similar way, Hazen (1983)
framework tetrahedra and thus bending of observed compressional anomalies for zeo-
the Si–O–Si angles. A study of the natural lite 4a, which exhibits different phase tran-
zeolite natrolite for example showed that sitions depending on the pressure medium
compression leads to a continuous reduction used. All of the observed high-pressure
of the unit-cell volume without any structu- phases showed higher compressibilities at
ral phase transition up to ~ 7 GPa (Belitsky high pressure.
et al., 1992). Above 7 GPa, the investigated A more frequent but still unusual phe-
samples underwent amorphization, similar nomenon observed upon compressing
to the transition observed for silica. This framework structures is a negative linear or
amorphization again indicates the collapse areal compressibility. Materials with these
properties will expand in one or two di- (i.e., framework flexion) with increasing
mensions upon hydrostatic compression. pressure. An example of this is the increas-
As shown by Baughman et al. (1998), this ing orthorhombic distortion of MgSiO3
also implies a negative Poisson’s ratio, i.e., with increasing pressure (Fiquet et al.,
lateral contraction upon uniaxial compres- 2000).
sion. These authors also demonstrate that
properties such as a negative Poisson’s ra-
10.2.2 Increase in coordination number
tio are in most cases linked to geometric
constraints in hinged framework structures. Many crystal structures are best de-
Because such behavior can result in inscribed by a closest packing of anions with
creasing surface area upon increasing hy- some of the interstitial sites (two tetrahedra
drostatic pressure, it is of potential interest and one octahedron per anion) occupied by
to material scientists. In principle it is con- a cation. In this simple, but in many cases
ceivable to manufacture these compounds very successful picture, the large and soft
into porous composite material with zero anions are in contact with each other and
or negative volume compressibilities. the small, more rigid cations are isolated
Another example of a structure family from each other and in contact only with
whose phase transitions are characterized their surrounding anions. The number of
by framework flexion are the perovskites. anions surrounding any given cation is de-
This material, with the general formula termined by the ratio of the “sizes” of the
ABX3, can be viewed as a stuffed deriva- cations and anions (for the problem of de-
tive of WO3-type structures. WO3 (BX3) is fining the “size” of an atom see e.g., Ross
built up of a three-dimensional network of and Price (1997)). In such a close-packed
corner-linked WO6 octahedra, the W (B) array of anions, rigid polyhedral rotation is
cations occupying the corners of the cube- usually not able to accommodate a volume
shaped unit cell. In perovskite the center of reduction imposed by increasing pressure.
this cube is occupied by the A cations. In The only way to optimize the volume upon
the ideal structure, the B–X–B angle at the increasing pressure is through more effi-
connecting octahedral corners is 180°. As cient packing of the anions involved, thus
shown by Glazer (1972), this octahedral reducing the anion–anion distances within
framework is susceptible to a number of the structure. This reduces the ‘size’ of the
distortions which are characterized by rigid anions without affecting the respective size
octahedral rotations, thus pure framework of the cations, which leads to a higher cat-
flexions. These distortions can even be ob- ion/anion size ratio. As a consequence, the
served at ambient conditions depending on coordination number of the cations in-
the nature of the A and B cations. At ambi- creases. Such a coordination increase is
ent conditions, the geometry is controlled generally accompanied by a lengthening of
by the relative size and thus the bonding re- the cation–anion distances, which at first
quirements of the cations involved. In a glance may appear a surprising effect for a
similar way the effect of compression de- high-pressure phase transition. However,
pends on the relative size at ambient condi- the lengthening of the first coordination
tions and the relative compressibilities of sphere is compensated by a shortening of
the A and B cations. In most cases the A the second shell. It is therefore worthwhile
cation is the more compressible unit, thus looking not only at the nearest neighbors of
leading to increasing polyhedral tilting a cation. When looking at both the first and
Figure 10-2. The exponential rela-

tionship between bond valence and
bond length leads to more regular
geometries at high pressure; High
pressure tends to destabilize struc-
tures through overbonding. For a
given average bond length, the aver-
age bond valence is higher (= more
overbonding) the more the individ-
ual bond lengths are different from
each other (= irregular geometry).
second coordination shells of a cation, a ica (SiO2) from coesite (4-coordinated Si,
coordination increase can be visualized as 2-coordinated O) to stishovite (6-coordi-
making the bonding environment of a cat- nated Si, 3-coordinated O). While the co-
ion less distorted. Keeping this in mind, the esite structure is described by a very dense
increase in coordination number can be packing of corner-linked SiO4 tetrahedra,
quantitatively rationalized by the distortion stishovite adopts the rutile structure, char-
theorem (Brown, 1992). The compression acterized by chains of edge-sharing SiO6
of a close-packed array of anions leads first octahedra. Other prominent examples of
to a reduction in the anion–anion distance increasing coordination numbers at in-
without rearranging the actual packing. creasing pressure are the well-known phase
This reduces all cation–anion distances, transitions from the NaCl (6-fold) type
which leads to a strengthening of the indi- structure to the CsCl (8-fold) structure in
vidual bonds. In the framework of Brown’s alkali chlorides or the high-pressure transi-
bond valence approach, a general strength- tions from zincblende (4-fold) to NaCl or
ening of cation–anion bonds around a b-Sn (6-fold) in many AX semiconductors
given atom induces an ‘overbonding’ of (see Sec. 10.5.1).
the atoms, which in turn destabilizes the
structure. Because of the exponential rela-
10.2.3 Pressure-induced ordering
tionship between bond strength and bond
length (Fig. 10-2), a set of equal bond The potential effect of high pressure on
lengths will have a lower bond valence sum order–disorder phase transitions has only
than the same number of bonds with the recently been fully realized. The interested
same average value, but different individ- reader is referred to the excellent and
ual bond lengths (distorted arrangement). thorough review by Hazen and Navrotsky
An increase in coordination number thus (1996). In this section we only give a brief
reduces the overbonding of the atoms in summary of the main features of pressure-
the crystal by making the bonding environ- induced order–disorder phenomena.
ment around the atoms less distorted. A Atoms on a given crystallographic site
very instructive example of this effect is can order with respect to their chemical
the aforementioned phase transition in sil- species, exact position, magnetic moment
Figure 10-3. Schematic drawing

to illustrate a possible mecha-
nism of pressure-induced order-
ing. Two sets of balls of different
sizes occupy (a) a smaller vol-
ume in an ordered arrangement
than (b) the same number of
balls in a disordered distribution.
or electronic state. In the following, disor- vors an ordered arrangement (Hazen and
der with respect to chemical species is re- Navrotsky, 1996). This can be understood
ferred to as ‘substitutional disorder’, by the fact that an ordered arrangement of
whereas disorder on two neighboring sites two atomic species of different ‘sizes’
is called ‘structural disorder’. Changes in leads to an alternation of ‘small’ and
any of these order parameters across a ‘large’ layers or rods (Fig. 10-3 a). A disor-
phase transition are known to have an ef- dered arrangement, on the other hand,
fect on the molar volume of the material forces each of the disordered sites to have
(e.g., Owen and Liu, 1947). Depending on the apparent size of the largest atom shar-
whether the molar volume is smaller for an ing this site (Fig. 10-3 b). An ordering de-
ordered or disordered state, pressure can pendence of the molar volume (and there-
thus – in principle – promote ordering or fore a pressure dependence of the ordering)
disordering effects. Nevertheless, such ef- can also be observed in flexible framework
fects are very often hampered by reduced structures such as feldspars. At first glance,
diffusion rates at high pressure. It is there- the molar volume of such structures should
fore fair to say that at moderate tempera- not be critically correlated with the size of
tures (Ⰶ1500 K), high pressure does not the cations, because in these structures
necessarily induce order–disorder transi- most of the volume change induced by or-
tions, but certainly supports such reactions. dering or pressure can be accommodated
The fast kinetics above ~ 1500 K, however, by the intra-polyhedral angles (Sec.
allow rapid equilibration with respect to or- 10.2.1). The observed volume changes,
dering and even in many cases inhibit however, can be understood on the basis of
quenching of an ordering pattern stable at the variation of intra-polyhedral angles, de-
high pressure and high temperature (Hazen pending on the cation species occupying
and Navrotsky, 1996). the respective polyhedra (e.g., Geisinger et
Based on present data, the volume al., 1985). Simple geometric considera-
change DVdis = Vdisordered – Vordered of struc- tions show that varying distributions of a
tural or substitutional disorder tends to be given set of angles in space leads to differ-
positive, suggesting that high pressure fa- ent enclosed volumes (Fig. 10-4).
Figure 10-4. Schematic two-

dimensional sketch to demon-
strate how the distribution of a
given set of rigid triangles
(polyhedra in 3-d) affects the
area (volume in 3-d) enclosed.
The area between the triangles
in (a) is 2.450, while the trian-
gles in (b) enclose an area of
2.414 (after Hazen and Navrot-
sky, 1996).
The largest volume changes are ob- the charge-ordering transition to lower
served for substitutional ordering pro- pressures and eventually even induces an
cesses. A4+B22+O4 spinels, for example, iso-structural phase transition where the
show a difference in molar volume of up to mixed valence compound Sm32+Sm3+Bi33–
3.5% between the normal (fully ordered) changes to a purely 3-valent material
and the inverse (disordered on the octahe- Sm43+Bi34– (Ochiai et al., 1985).
dral site) modifications (Hazen and Yang, Another technically interesting phenom-
1999). The pressure behavior of the spinel enon that is connected to the pressure de-
family is especially interesting, because pendence of ordering and correlation of
they show both pressure-induced disorder- charge carriers is the well-documented
ing as well as pressure-induced ordering pressure dependence of Tc in certain ce-
for one and the same structure type de- ramic superconductors (e.g., Acha et al.,
pending on the chemical species involved 1997; Han et al., 1997). Although neither
(Wittlinger et al., 1998; Hazen and Yang, the superconductivity nor its relationship
1999). to high pressure is fully understood in ox-
The rather scarce data on pressure de- ide materials, it is justified to assume that
pendence of charge distribution seems to the observed strong shift in Tc with increas-
indicate that charge ordering tends to be ing pressure is connected to a subtle inter-
suppressed by high pressure. NaV2O5 for play between pressure-induced structural
example shows a charge-ordering transi- distortions, orbital overlaps and charge-
tion around 35 K at ambient pressure. This carrier distribution.
transition seems to be shifted to lower tem-
peratures at higher pressures and disap-
pears completely around 1 GPa (Ohwada et 10.3 Generating High Pressure
al., 1999). This suggests that in the case of
NaV2O5 , high pressure induces a charge- The technology of pressure cells com-
disorder phase transition. In a similar way, patible with in situ experiments for mate-
the application of pressure to Sm4Bi3 shifts rial characterization has experienced tre-
10.3 Generating High Pressure 667
mendous development since the ground- energy rises from E0 to EH. The Rankine–
breaking work of Bridgman. Consequently, Hugoniot relations combine these quanti-
there is a very comprehensive and vast lit- ties to (r0 = density at ambient conditions):
erature on this subject to which the more
VH = V0 (US – UP)/US (10-8)
committed reader is referred (Miletich et
al., 2000; Holzapfel, 1997; Eremets, 1996; P H = r0 U S U P (10-9)
Ahrens, 1987). Only a brief overview will
EH – E0 = (V0 – VH) PH/2 (10-10)
be given in this section.
There is a huge variety of different tech- The measurements of US and UP give the
niques for generating pressures. They can quantities on the left of Eqs. (10-8) to (10-
initially be divided into static methods and 10) for one experiment. Various experi-
dynamic techniques. ments at different strengths of explosion or
different velocities of the projectile pro-
duce different points, forming the Hugon-
10.3.1 Dynamic pressure generation
iot curve describing the Hugoniot equa-
The highest pressures (102 – 103 GPa) tions of state VH (PH). In order to reduce the
can be obtained using dynamic shock-wave Hugoniot equations of state into an isother-
generation. This is achieved by means of mal equation of state, careful thermody-
explosives or by a projectile that is acceler- namic corrections have to be applied
ated toward the target with a gas gun (Ah- (Poirier, 1991). A material passed by a
rens, 1980, 1987). The shock-wave tech- shock wave usually displays various
nique was originally developed in the mid- stages. Up to a pressure of the Hugoniot
1950s at Los Alamos, USA, in the course elastic limit (HEL) (0.2 to 20 GPa), the
of the development of atomic bombs sample behaves elastically, corresponding
(Walsh and Christian, 1955). In its simplest to the propagation of the longitudinal
case, the impact of a projectile on the target shock wave. Above the HEL, plastic defor-
produces a uniaxial shock wave. The shock mation of the material occurs, giving rise
wave passes through the sample at shock to the fluid-like behavior that creates the
velocity Us . The sample itself is acceler- Hugoniot curves. If a material undergoes a
ated to the sample velocity Up, and Us and phase transition, this is readily observed as
Up , together with the temperature, are the changes of slope in the Hugoniot curves,
quantities measured during a shock-wave separated by a mixed-phase regime (Fig.
experiment. The velocities are usually de- 10-5).
termined by measuring entrance and exit In this sense, in a shock-wave experi-
times of the shock wave. For samples of ment the material can only be investigated
a few millimeters in length, the time to by the difference in UP and US, which both
be measured is in the range of 10–1 to depend on the volume reduction of the
~ < 101 µs. Ignoring the yield strength of sample. The accuracy of these measure-
the solid (which is justified at the high ments becomes critical once the volume re-
shock pressures encountered during a ductions are small at very high pressures.
shock-wave experiment), the material be- Third-generation synchrotron radiation
haves as a fluid. In such an experiment, the sources or even free-electron lasers of the
volume decreases from V0 to VH, the tem- future may offer improved possibilities.
perature increases from T0 to TH and the The increase in X-ray flux of these X-ray
pressure from P0 to PH, while the internal sources may allow for stroboscopic X-ray
Figure 10-5. Hugoniot curve of a

hypothetical material undergoing a
phase transformation at high pres-
sure. The low-pressure phase ap-
proaches its hydrostatic behavior af-
ter passing the Hugoniot elastic limit
(HEL). The existence of a shock-in-
duced high-pressure phase is seen
by the onset of the mixed-phase re-
gime. The hydrostat of the high-
pressure phase can be recognized in
the high-pressure phase regime (af-
ter Ahrens, 1980).
diffraction in the microsecond regime. This 20 GPa to be obtained. Diamond anvil

would enable structural information to be cells, in contrast, enclose a volume of
extracted of a material under extreme < 10–3 mm3, i.e., only a few pico-liters.
dynamic pressure. Today, shock-wave ex- These devices are capable of maximal
periments are highly complementary to pressures up to 200 to 500 GPa.
static high-pressure experiments. Undoubt-
edly, one of their most crucial roles is the
103.2.1 Large-volume presses
creation of an equation of state up to ex-
treme pressures without any additional Historically, the first large-volume cells
pressure standard. This provides invaluable were hydraulic presses as built by Bridg-
anchor points for all pressure scales cur- man. He was able to compress fluids up to
rently used in static high-pressure experi- 10 GPa. In modern high-pressure research,
ments (Mao et al., 1978; Jamieson et al., hydraulic devices are limited to maximal
1980). pressures between 1 and 2 GPa and will
therefore not be discussed any further in
this chapter. Presses operating with solid
10.3.2 Static pressure devices
pressure media up to 50 GPa are the most
Devices to generate static pressures that important tools in modern materials sci-
are sustainable for an a priori indefinite ence for the synthesis of materials at simul-
time can be divided into ‘large-volume taneously high pressure and high tempera-
presses’ and ‘diamond anvil cells’. The pri- ture. Increasingly, large-volume devices are
mary difference between these two fami- also used for in situ studies in combination
lies of devices lies in the volume of mate- with synchrotron and neutron radiation.
rial subjected to high pressure and conse- Piston-Cylinder systems are the simplest
quently the maximum pressure attainable. design for compressing material. They rely
For large-volume devices the compressed on a cylinder acting as the sample chamber.
volume lies between 1 mm3 and 1 cm3. The sample is compressed by a piston fit-
This allows maximum pressures of around ted within the cylinder. Closed-end cylin-
Figure 10-6. Schematic drawing of (a) a girdle-type press and (b) a belt apparatus. The shaded area corre-
sponds to gasket material. S shows the sample position. Note how consequent cutting of stress amplifying cor-
ners on the girdle and decrease of length-to-width ratio on the piston lead from a girdle press to a belt design.
ders are closed on one side and the sample most straightforward way to reinforce a
is compressed with one piston from the cylinder is by increasing the ratio of its di-
other side. Open-end cylinders rely on two ameter to length. If this is pursued conse-
opposite pistons. The maximal pressure of quently, we arrive at two conical pistons
5 GPa attained by piston-cylinder devices compressing a sample contained in a girdle
is generally limited by the yield strength of (Fig. 10-6 a). A modification of the girdle
the cylinder. Various techniques exist to in- design is obtained by optimizing the coni-
crease this yield strength. In a technique cal shape of the anvils. A cycloid shape of
called frettaging the cylinder is prepared the pistons, as shown in Fig. 10-6 b, en-
such that in its default state, the outer part sures the optimal compromise ensuring
retains a residual tensile stress and the insufficient cylinder support at low pressure
ner part a compressive stress. Adding the while maintaining a reasonable stroke in
work stress leads to a partial cancellation the high-pressure regime. This ‘belt
of these stresses, thus allowing for the ap- design’ is by far the most widely used high-
plication of higher loads to the sample. pressure apparatus for materials synthesis
Pre-loading the cylinder is another way to and is frequently applied in industry for the
increase the yield strength of the cylinder synthesis of super-hard material. The max-
to higher values. Pre-load can be achieved imum pressure achieved by belt devices is
by using a soft but incompressible material around 10 GPa. Such devices have also
(lead) as the inner cylinder material and been optimized with respect to sample vol-
by initially overstraining it into its flow re- ume; a flat belt apparatus constructed by
gime. Another method relies on the appli- Fukunaga et al. (1987) was capable of
cation of an external load on the cylinder compressing a sample of 125 ml up to pres-
simultaneously with the application of the sures of 5 – 7 GPa.
load on the piston. Alternatively, cylinders The disadvantage of piston-cylinder as-
can be reinforced by winding a strong wire semblies is the intrinsic opacity of both pis-
under tension around the cylinder. The ton and cylinder, thus inhibiting most in
Figure 10-7. The principle of

the toroidal anvil. The cut is
circular symmetric around the
vertical axis of the plot. The
sample (S) is surrounded by a
washer within a toroidal belt
containing gasket material. This
geometry prevents the sample
from being extruded, thus creat-
ing a hydrostatic pressure de-
spite the uniaxial force applied.
situ observations. This problem is allevi- is also translated into an equatorial com-
ated by opposed anvil systems which oper- pression through wedges attached to the
ate with a gasketing system instead of a four equatorial anvils. Although initially
cylinder. The very first example of this designed for synthesis experiments, vari-
type was built by Bridgman (1952). A log- ous models of this type have been used
ical extension of the Bridgman design are in recent years at synchrotron radiation
profiled Bridgman anvils as first proposed sources for in situ X-ray diffraction studies
by Ivanov and coworkers in the 1960s in Japan (Photon factory and Spring-8),
(e.g., Ivanov et al., 1991) leading to tor- Brookhaven (NSLS) and Hamburg
oidal anvils (Fig. 10-7). This principle was (DESY). This is possible because the small
optimized most consequently in the gaps between the individual anvils allow
Paris–Edinburgh cell design (Besson et al., for entry and exit of X-rays, and the cell as-
1992). The advantage of X-ray and neutron sembly containing the sample is made of X-
transparent gasketing has been exploited to ray transparent material (BN, epoxy). With
the maximum in this cell. This is the reason tapered anvils made of sintered diamond,
why this cell design has advanced to the maximal pressures of 20 – 25 GPa at tem-
state-of-the-art model for in situ high-pres- peratures around 1000 °C can be attained.
sure neutron diffraction and is also em- A rather different geometry in multi-an-
ployed for X-ray experiments. A develop- vil technology is the split-sphere design
ment of the Los Alamos neutron group (Liebermann and Wang, 1992) pioneered at
pushes the limits of P and T attainable with Okayama University by Ito. It consists of a
toroidal anvils up to 50 GPa and 3000 K, uniaxial ram applying force on a split
thus enabling in situ neutron studies up to sphere (first stage) that contains a cube-
these conditions. shaped cavity with the body diagonal along
A quite different approach to compress- the axis of the ram. The cavity is occupied
ing large volumes up to pressures of by a cube built of eight WC cubes with
30 GPa was first developed by researchers truncated corners. The truncated corners in
in Japan using multi-anvil designs (Aki- turn form an octahedral cavity that hosts
moto, 1987). A very successful product is the cell assembly. The cell assembly is usu-
the DIA-type cubic anvil press. In this de- ally a MgO octahedron with a Pt capsule
vice, six tungsten carbide (WC) or sintered containing the sample in its center. A vari-
diamond anvils are arranged parallel to the ation of this design is the split-cylinder de-
faces of a cube that encloses the cell as- sign in which the sphere of the first stage is
sembly. A ram exerts an axial force, which replaced by a cylinder. Multi-anvil presses
based on the split-sphere approach are also 10.3.2.2 Diamond anvil cells
used as in situ devices at the ‘Synchrotron
Radiation Source’ in Daresbury (UK) and Experiments at extreme pressures, which
the ‘Advanced Photon Source’ at Argonne are of interest not only to geophysicists and
National Laboratory (USA). planetologists, but also to physicists and
However, the determination of pressure chemists studying, for example, solidifica-
is not straightforward. In principle, pres- tion and metallization of ‘gases’, can be
sure can be calculated by dividing the force achieved using diamond anvil cells (DAC).
by the area on which it is acting. In practice Diamond anvil cells are in principle very
this is not applicable, mainly because of small opposed-anvil devices of Bridgman
the unpredictable friction losses and illde- type. The anvils are made of diamond sin-
fined compressibilities of the cell assem- gle crystals, shaped in the brilliant cut with
bly. Therefore, pressure has to be deter- the bottom tip truncated to form the anvil
mined through a calibration procedure. surface (culet). This simple design allows
When in situ diffraction techniques are im- the special properties of diamonds to be
possible, this can be done by determining used in two ways. First, the extreme hard-
phase transitions revealed through changes ness of diamonds allows very high pressure
in resistivity of metals and semiconductors to be generated. Pressures attained depend
(e.g., Bi, Ba, ZnS, GaAs, GaP). The re- of course on the size of the culet. Maximal
spective high-pressure phase transitions pressures of 500 GPa have been reported
are determined in hydraulic pressure de- (Xu et al., 1986) and pressures between
vices whose pressure can be directly meas- 100 and 200 GPa can be reliably repro-
ured by a pressure gauge. If in situ observa- duced. The second advantage of using
tion is possible, the equations of state of single crystalline diamonds as pressure
materials such as NaCl, Cu, Mo, Ag and anvils is the high transparency of diamond
Pd, which have been determined up to very for almost the entire electromagnetic spec-
high pressures through shock-wave experi- trum. This allows us not only to easily
ments, can be applied to calibrate the press. observe samples under high pressure, but
Because the reproducibility for a given also to probe them with spectroscopic
press and cell assembly is very high, pres- methods as well as X-ray diffraction
sure of subsequent experiments can be de- (Sec. 10.4). The strength of DACs is thus
termined from the force applied on the cell their huge range of pressure combined with
assembly. the ease of performing in situ experiments,
The strength of large-volume devices and this all with a device of the size of a
lies in their potential for material synthesis fist, which is also easy to operate by non-
at simultaneously high temperature and specialists.
high pressure. The relatively large size of The very first DAC was constructed us-
the compressed volume (~ 0.1 – 1 cm3) al- ing a big (8 ct) gem-quality diamond (taken
lows inclusion of a heater (usually cylindri- from smugglers by the US government and
cal graphite or LaCrO3 resistance heaters) donated to the US National Bureau of Stan-
as well as thermocouples, which enable dards) in which a hole was drilled (Jamie-
pressure and temperature to be combined son, 1957; Lawson and Tang, 1950). Pres-
in a very controlled way. Due to the large sure on the sample in the hole was applied
volume, these devices are limited to maxi- via a piston (piano string) pressed on the
mal pressures of 50 GPa at the very best. sample. The limited pressure range obtain-
Figure 10-8. Sketch illustrating the prin-

ciple of the diamond anvil cell. The axial
force applied to the diamonds is partly
translated into a plastic deformation of
the gasket, which results in a circular
symmetric equatorial force. This ensures
quasi-hydrostatic conditions at the sample
(S), which is embedded in a pressure me-
dium.
able with this approach was painfully real- research by Bassett (Merrill and Bassett,
ized when the 8 ct diamond was crushed 1974) and Mao (Mao and Bell, 1975). Fur-
during an experiment. As a consequence, ther details on the technology of DACs and
the opposed-anvil geometry was developed also on its various modifications and devel-
(Weir et al., 1959; Jamieson et al., 1959), opments are given in excellent reviews by
first by simply squeezing a powdered sam- Hazen and Finger (1982) and Miletich et
ple between the culets of two opposed anal. (2000).
vils and later by introducing the gasketing As for the large volume devices, the ac-
technique (Van Valkenburg, 1964). Its prin- curate determination of pressure in a dia-
ciple has remained unchanged since its in- mond anvil cell is a difficult issue. The
vention and is as simple as it is efficient most accurate values are obtained by add-
(Fig. 10-8). A hole about 100 to 200 mm in ing an internal standard to the sample
diameter is drilled in a metal foil (Fe, W, whose equation of state is known with suf-
Re). This hole serves as sample chamber ficient precision to relate its diffraction
and is filled with the sample (powder or pattern to a pressure value (e.g., Angel et
single crystal) and a pressure medium (al- al., 1997). A very convenient and popular
cohol, liquid gas). The pressure medium is alternative, albeit not quite as accurate, is
compressed through an axial force exerted the exploitation of the pressure shift of flu-
by the diamonds on the gasket. The gasket orescence lines. The most frequently used
seals the sample chamber and at the same fluorescence is the R1 line of ruby (Mao et
time transforms part of the axial pressure al., 1978). This method again benefits from
into an equatorial pressure through the the transparency of the diamond high-pres-
plastic deformation of the gasket material. sure windows. A ruby, which is packed to-
In this way the pressure medium is com- gether with the sample into the gasket hole,
pressed isotropically and therefore trans- is illuminated with a green or blue laser
mits a hydrostatic pressure on the sample. that induces a red fluorescence line. The
This design was popularized for materials wavelength of this line depends on pres-
10.4 Probing Phase Transformations in Materials at High Pressure 673
sure (dl/dP = 0.37 nm GPa–1) and can thus is by measuring the black-body radiation
be used to determine the pressure within of the glowing sample and fitting it to
the sample chamber. More recent develop- Planck‘s spectral function (e.g., Bassett
ments of this approach use different fluo- and Weathers, 1987). When doing HP-HT
rescence lines with various pressure and experiments with laser heating we should
temperature dependence in order to simul- bear in mind the possibility of thermal
taneously determine pressure and tempera- pressure, i.e., the increase of pressure in
ture in the sample chamber (e.g., Datchi et the heated area, while the pressure deter-
al., 1997). mined by a ruby chip outside the hot-spot
Because of the very small volumes of the remains constant (Andrault et al., 1996).
sample chamber (~ 0.001 mm3), combining
pressure and temperature in a diamond an-
vil cell is a difficult task. In principle there 10.4 Probing Phase
are two different approaches, namely exter- Transformations in Materials
nal and internal heating. With external at High Pressure
heating, the whole sample chamber, includ-
ing diamonds, is enclosed in a resistance A very important aspect when doing ex-
heater (e.g., Hazen and Finger, 1982; Bas- periments on high-pressure phase transfor-
sett et al., 1993) that heats the entire assem- mations is the possibility of investigating
bly consisting of gasket, diamond anvils the phase transformation in situ at condi-
and sample. Temperature is measured via tions of high pressure and possibly simulta-
thermocouples attached to the outer dia- neously high temperature. As mentioned
mond facets, assuming that the high ther- above, this is much easier to do in static ex-
mal conductivity of the diamonds allows periments than with dynamic shock-wave
for only a very small temperature gradient techniques. This is mainly due to the very
between sample and outer diamond facets. short time that is available in a shock-wave
This technique has consequently been opti- experiment. We will therefore focus on
mized by using the gasket material itself as static experiments in the following. Among
a resistance heater, thus minimizing the the static experiments, the transparency of
heated volume and thermal gradients (Du- the diamond pressure windows in a DAC
brovinsky et al., 1997). An alternative ap- allows for much more versatile experimen-
proach is the use of an infrared laser beam, tal techniques compared with large-volume
which is focused through the (IR-laser presses. For large-volume experiments, di-
transparent) diamonds onto the sample, rect observation of the sample is limited to
where it is absorbed and thus transformed either transport properties (e.g., electric re-
into heat. This technique was again pio- sistance) or, if using some sort of electro-
neered by Bassett (e.g., Bassett and Ming, magnetic radiation as a probe, severe com-
1972) and then further developed and opti- promises in signal-to-background and ac-
mized by Boehler (Boehler and Chopelas, cessible space have to be accepted. Never-
1991) and Fiquet and Andrault (Fiquet et theless, the range of in situ techniques ap-
al., 1994). Measuring temperature with this plicable to both DACs and large-volume
technique is even more difficult because it devices has grown considerably and con-
explicitly assumes that only the sample is tinues to expand. There is an extensive spe-
heated up. The only way to obtain a quanti- cialized literature on each of the various
tative estimate of the sample temperature techniques. We will give here a brief over-
view on and introduction to the available 10.4.2.1 Microscopy

techniques, with references to the more de-
A rather cheap, but in many cases highly
tailed literature. Many techniques can be
efficient ‘spectroscopic’ device is the hu-
used to measure different physical proper-
man eye. Visual observation of a sample
ties and vice versa. In the following we will
under high pressure can be a very useful
differentiate by the technique rather than
and sensitive tool for observing a phase
by the measured property.
transformation and pinpointing it in P – T
space. An example of this has been de-
10.4.1 Volumetric techniques
scribed by Arlt et al. (2000), where the
The first in situ observations in large- pressure dependence of a high-temperature
volume cells were pioneered by Bridgman phase transition in Mn-pyroxenes turned
(e.g., Bridgman, 1940) on piston-cylinder out to be difficult to determine using dif-
devices. He made volumetric measure- fraction techniques, but could be optically
ments by simply measuring the stroke of observed through the discontinuous change
the piston as a function of force (pressure) in birefringence.
applied to the sample. This technique obvi-
ously assumes that deformation of the pis-
ton and cylinder, as well as leakage, can be 10.4.2.2 Raman and infrared
neglected. The accuracy in V/V0 obtained spectroscopy
through this method can be fine-tuned to
The most popular spectroscopic tech-
about 1 in 1000 (Anderson and Swenson,
niques applied in high-pressure studies are
1984). In general, the accuracy is limited to
Raman and infrared (IR) spectroscopies.
lower values by deformation of the piston
They both probe the vibrational properties
and the cylinder. Compressibilities are
of the material under investigation. Be-
therefore most often and much more accu-
cause lattice vibrations strongly depend on
rately determined using diffraction tech-
the topology of the chemical bonds of a
niques (see Sec. 10.4.4). Prior to the avail-
substance, these spectroscopic techniques
ability of in situ diffraction techniques,
are very sensitive to phase transformations.
however, volumetric measurements were
Vibrational modes involving a dipole
the only way of measuring compressibil-
change can be excited by absorbing an in-
ities as a function of pressure, and they pro-
frared photon giving rise to an absorption
vided extremely valuable data.
band in the infrared. The energies of these
lattice modes are thus of the same magni-
10.4.2 Spectroscopic techniques
tude as the energy of the IR photons. If the
There are a variety of spectroscopic photon energy is much higher, the interac-
techniques making use of various segments tion of the photon with the lattice can in-
of the electromagnetic spectrum. Although duce a vibrational mode, where the energy
the transparency of diamonds makes DACs of the lattice mode is transferred from the
a very obvious tool for spectroscopic tech- photon to the lattice. This decreases the en-
niques, they are by no means limited to ergy of the photon and therefore induces a
them, but – due to the advent of brilliant wavelength shift on the scattered photons
synchrotron radiation sources – are in- (Raman spectroscopy). Because energy can
creasingly also applied in combination be transferred in both ways (to and from
with large-volume presses. the lattice), the respective wavelength shift
can be positive or negative. By measuring tural environment (resonant absorption)

the wavelength shifts of transmitted laser while lifting it into an excited state. The
light of a wavelength around 500 nm, a energy of the excited state of a given nu-
characteristic fingerprint of the vibrational cleus is a function of its structural environ-
properties of a given substance is obtained. ment, and therefore varies between indi-
Again, this fingerprint critically depends vidual substances. This energy shift (iso-
on bond strengths and the structural config- mershift) is another characteristic finger-
uration and is therefore ideal for detecting print for a given structural state as well as
phase transformations. While in general for the electronic charge of the nucleus.
Raman shifts are measured for the optical The isomer shift can be measured by alter-
branches of the vibrational spectrum, very ing the relative energy of the g-quanta
carefully designed experiments are able to through the Doppler effect caused by rela-
determine the minute frequency shifts tive movements of sample and source. The
caused by the acoustic branches, too. Such g-quanta of the most popular Mössbauer
Brillouin scattering experiments are very nuclei (57Fe26) are in the range of 14.4 keV
interesting, because they allow the full and are thus strongly absorbed even by di-
elastic tensor to be measured as a function amonds. It is for this reason that the first
of pressure (e.g., Sinogeikin and Bass, high-pressure Mössbauer experiments
1999). The elastic tensor is a material prop- were performed using more exotic
erty that is useful in its own right. Its strong Mössbauer nuclei such as 153Eu, 129I and
170
dependence on structural parameters Yb, which have higher g-energies.
makes it a very sensitive probe for detect- These experiments helped answer some
ing and investigating phase transforma- interesting questions about the physics of
tions (Carpenter and Salje, 1998). A very magnetic materials (e.g., Pasternak et al.,
instructive and comprehensive overview of 1986; Abd-Elmeguid et al., 1980). The de-
high-pressure specific problems and appli- velopment of especially miniaturized dia-
cations of IR and Raman spectroscopy is mond anvil cells (e.g., Pasternak and Tay-
given by Gillet et al. (1998). lor, 1990, 1996) together with advances in
detector technology also allowed success-
ful Mössbauer experiments to be per-
10.4.2.3 Mössbauer spectroscopy
formed on 57Fe nuclei (e.g., McCammon et
Mössbauer spectroscopy has become inal., 1998). While traditional Mössbauer ex-
creasingly popular in materials science to periments make use of the energy structure
probe site-dependent distributions of of the emitted g-rays, modern synchrotron
charge and magnetic moment. Its applica- sources with their time-pulsed radiation
tion to high pressure is limited to DAC ex- also allow us to exploit the time structure of
periments (Pasternak and Taylor, 1996; a Mössbauer event. As mentioned above,
McCammon, 2000). In a classical Möss- a Mössbauer event involves the absorption
bauer experiment, the radiation emitted of a g-quantum by lifting the Mössbauer
by the g-source lifts the sample nuclei into nucleus in an excited state. This state lasts
an excited state through an absorption for a time span in the range of 100 ns be-
event. During re-emission, a fraction of the fore it decays and emits the scattered radia-
g-quanta is emitted without recoil on the tion. Because individual synchrotron X-ray
lattice (recoil-free fraction) and can thus be bursts can be gated into a time interval in
reabsorbed by a nucleus in the same struc- the range of 100 ps with 1 ms–1 repetition
rates, the actual decay event can be sepa- Brown et al. (1988) for a review). If an in-
rated from the primary synchrotron radia- cident X-ray photon hitting an atom in the
tion. This allows the individual decay sample has an energy equal to the energy
events and their interaction with the nu- difference between the ground state and the
clear and magnetic lattice to be investi- excited state of a core electron, it will be
gated directly. In such a ‘nuclear resonant absorbed by the atom while lifting the core
scattering’ experiment, the emitted g-rays electron to an excited state. Photons at this
of the individual nuclei in the material have energy have a high probability of being ab-
different wavelengths owing to the hyper- sorbed by the sample, which leads to char-
fine interaction of the magnetic 57Fe nu- acteristic absorption edges in the transmit-
cleus. The different wavelengths of radia- ted X-ray spectrum. Core electrons in their
tion emitted from different nuclei cause a excited state (photoelectrons) are delocal-
quantum-beat oscillation pattern, which in ized from their parent atom and can thus
turn yields information on the material in- interact with the intermediate surroundings
vestigated (e.g., Smirnov, 1999). The high of the atom as well as with other photoelec-
brilliance of synchrotron radiation sources, trons, leading to an oscillatory contribu-
which are required for such experiments, tion in the X-ray spectrum in the vicinity of
also makes nuclear forward scattering an absorption edge. It is this oscillatory
much easier to apply in combination with part that is extracted from an EXAFS ex-
diamond anvil cells. High-pressure experi- periment. It is dependent on the immediate
ments using nuclear forward scattering as a surroundings of an atom and therefore con-
probe to investigate structural and mag- tains information about the local neighbor-
netic phase transitions have therefore hood of an atom in a solid or a liquid.
quickly become very popular (e.g., Nasu, When photoelectrons are recaptured by an
1996; Lubbers et al., 1999). atom, they will emit characteristic fluores-
cence radiation which by itself can again
be used as a probe for characterizing a ma-
10.4.2.4 X-Ray absorption spectroscopy
terial. The advantage of XAS methods in
The high X-ray brilliance of modern syn- comparison with diffraction methods is
chrotron radiation facilities not only revo- their sensitivity to local and short-range ef-
lutionized high-pressure diffraction (see fects. This makes them very powerful for
Sec. 10.4.4), but also provided the opportu- the investigation of phase transformations,
nity to apply X-ray absorption spectros- not only in crystalline solids but also in
copy (XAS) methods in high-pressure amorphous solids and liquids (e.g., Buon-
research. In particular, developments on tempo et al., 1998).
large-volume presses (Paris–Edinburgh)
which were originally built for neutron dif-
10.4.2.5 NMR
fraction, proved to be very useful for in situ
XAS experiments. Most popular among the A final spectroscopic tool that can be ap-
high-pressure XAS techniques are the ‘ex- plied to samples under high pressure is
tended X-ray absorption fine structure’ NMR. An NMR experiment exploits the
(EXAFS) experiments (i.e., Katayama et interaction between the magnetic moment
al., 1997). The principle of EXAFS is well and the spin of the nucleus on the one hand
known and has been described since the and a static magnetic field disturbed by
early days of quantum mechanics (see pulses of a radio frequency field on the
other. Introductory texts on NMR are avail- 10.4.3 Ultrasonic sound velocity
able, e.g., by Akitt (1983) or Kirkpatrick
(1988). The spin of a nucleus leads to a Measuring sound velocities at high pres-
precession motion of its magnetic moment sure is of great interest, mainly for seismol-
if placed in a static magnetic field. A mag- ogists who use this to correlate seismic
netic perturbation in the radio frequency models with mineralogical models of the
range of the static field causes a tilting of earth. The importance of this method for
the precession axis. The time the nucleus materials science is not quite so obvious al-
needs to revert its precession axis into its though in principle, discontinuities in the
static position depends on the nature of the change in sound velocity at high pressure
nucleus itself as well as its immediate sur- can be used to detect phase transformations
roundings in the crystalline or amorphous and to characterize them through the elastic
host. NMR, similar to XAS techniques, is tensor. A simple pulse-transmission or
therefore a useful probe for investigating pulse-echo method can be combined with a
short-range phenomena. Alternatively, the piston-cylinder apparatus in order to mea-
analysis of NMR spin echoes also allows sure velocities at pressures up to about
information on viscosity and self-diffusion 1 GPa. A more sophisticated approach is
rates in liquids to be extracted. ultrasonic interferometry, which can be
The application of NMR to high pressure combined with a split-sphere apparatus. In
is an experimental challenge because it has this technique, two phase-coherent pulses,
to deal with both the inherently small vol- separated in time by the approximate return
umes encountered in high-pressure experi- travel time in the sample, are applied to a
ments and the magnetic susceptibility of cell assembly under pressure, which is at-
most materials suitable for construction of tached to a buffer rod. When combining
high-pressure devices. Nevertheless, the this technique with a large-volume press,
first high-pressure NMR experiment was the anvils are simultaneously acting as buf-
performed as early as 1954, in combination fer rods. For two consecutive pulses, the
with a Bridgman-type press (Benedek and echo of the first pulse from the far end of
Purcell, 1954). The press was almost en- the sample overlaps with the echo of the
tirely made from a non-magnetic Be–Cu al- second pulse from the sample–buffer inter-
loy. Combining NMR and DAC is even face. The interference caused by this over-
more difficult because of the extremely lap leads to a modification of the amplitude
small sample volumes. However, Bertani et of the resultant signal. This amplitude in
al. (1992) developed an NMR–DAC, again turn can be modified by changing the car-
made of Be–Cu alloy. This DAC is compat- rier frequency, causing a beat pattern as a
ible with a cylindrical cryostat and is thus function of frequency. The difference on
able to combine high pressure and low tem- the frequency scale between two succes-
perature, allowing for the study of the pres- sive interference extremes yields the sound
sure and temperature dependence of the velocity, if the dimensions of the sample
Knight shift in solids (e.g., Kluthe et al., are known. This method has an advantage
1996). An NMR set-up for the investiga- over more traditional pulse-transmission
tion of gases and liquids at pressures up to techniques, in that it involves only one
20 bar has been developed by Woelk and interface, reducing the possible coupling
Bargon (1992). problems for the acoustic signals. Owing to
the applied high pressure, mechanical con-
tact and thus coupling between the materi- acterize even subtle phase transformations
als forming the interface is further en- (Carpenter et al., 1998, Angel, 2000 b). In
hanced. A more thorough introduction to addition, measurements of single-crystal
this technique is given by Rigden et al. diffraction intensities serve to establish an
(1992) and Li et al. (1998). There have accurate structural model for the observed
been recent efforts to combine ultrasonic phase transformations and are therefore of
interferometry with diamond anvil cells. In great importance to the understanding of
this case, however, because of the very phase transformations at high pressure.
small sample volumes, the carrier frequen- X-ray powder diffraction is much more
cies have to be in the GHz range (Shen et difficult to perform at laboratory sources
al., 1998). (e.g., Haines et al., 1998), mainly because
of the inherently small sample volumes.
For this same reason even at synchrotron
10.4.4 Diffraction techniques
sources the dominating diffraction tech-
Probably the most versatile and at the nique for investigating powdered samples
same time powerful tools for characteriz- has been energy-dispersive diffraction (e.g.,
ing phase transformations at high pressure Holzapfel, 1997), making use of the much
are diffraction methods. There is a huge higher flux of a white beam compared
amount of literature on this field. Here we with monochromatic radiation. Energy-dis-
only briefly present the various domains of persive diffraction was initially the only
high-pressure diffraction giving appropriate method of choice for in situ diffraction
references for the more interested reader. studies in combination with large-volume
presses (Yagi et al., 1987, Weidner et al.,
1992, Mezouar et al., 1996). In the energy-
10.4.4.1 X-ray diffraction
dispersive mode, the diffraction signal is
A comprehensive introduction to the ba- recorded on the energy scale for a fixed
sic principles of X-ray diffraction is given Bragg angle using a Ge solid-state detec-
by, e.g., Stout and Jensen (1989). X-ray tor. Modern synchrotron radiation sources,
diffraction at high pressure has initially coupled with new two-dimensional detec-
only been possible in combination with di- tor technology, however, allow monochro-
amond anvil cells. Despite the high trans- matic powder experiments with both dia-
parency of diamonds to X-rays, diffraction mond anvil cells (e.g., Fiquet et al., 2000)
experiments suffer both from the small and large-volume presses (Parise et al.,
sample volume and from shielding and ab- 1998, Mezouar et al., 1999 a). In this tech-
sorption effects from the pressure cell com- nique, the diffraction signal of a monochro-
ponents. This is the reason why high-pres- matic X-ray beam is recorded as a function
sure experiments on sealed-tube sources of the Bragg angle. In contrast to energy-
are mostly restricted to single crystals, be- dispersive data, such angle-dispersive data
cause the diffraction signal of a crystal of a allow for a quantitative interpretation of
size of about 100 mm is sufficiently strong the diffracted intensities in order to estab-
to pass through the diamond anvil (Hazen lish and refine a structural model. Another
and Finger, 1982; Miletich et al., 2000, An- advantage of angle-dispersive diffraction
gel et al., 2000). By measuring accurate over energy-dispersive techniques is the
lattice parameters on single crystals (An- higher resolution achievable with mono-
gel, 2000 a), we are able to detect and char- chromatic radiation.
10.5 Examples 679
Although synchrotron radiation can neutron diffraction again in combination

overcome the intensity problem caused by with the Paris–Edinburgh cell (Klotz et al.,
the small sample volumes in high-pressure 1996; 1997). Such data complement the
powder diffraction experiments, the addi- static picture of a given material as ob-
tional intensity cannot resolve the problem tained through an elastic scattering experi-
of inadequate powder statistics of small ment with the dynamic nature (phonons) of
sample volumes. This problem can be the crystalline material.
partly alleviated using digital two-dimen-
sional detectors such as online image-plate
detectors or CCD cameras (Shimomura et
al., 1992).
10.5 Examples
A further step in the use of the unique
10.5. Zincblende-type semiconductors
properties of synchrotron radiation has re-
cently been made by performing inelastic Due to their pivotal role in modern elec-
X-ray scattering through a diamond anvil tronic technology, semiconductors are
cell in order to determine the acoustic pho- probably the most intensely studied class
non dispersion in materials (Krisch et al., of materials. Experiments at high pressure
1997). The application of this very useful are crucial for the testing of theoretical
technique in high-pressure research is still models, because the application of pressure
very new and under development. is a straightforward way of modifying
the band structure of the material with-
out changing the chemistry. It is therefore
10.4.4.2 Neutron diffraction
not surprising that semiconductors were
A very comprehensive introduction to among the very first materials to be inves-
the general principles of neutron diffrac- tigated at high pressure (e.g., Bridgman,
tion is given by Bacon (1975). The advan- 1935; Miller and Taylor, 1949). Since then,
tage of neutron diffraction over X-ray dif- theoretical as well as experimental work in
fraction is that neutrons are more sensitive this field has increased greatly.
to light elements. Furthermore, neutron Application of high pressure on semi-
diffraction is able to distinguish between a conductors leads to a closing of the inter-
lattice of nuclei and a magnetic lattice. atomic distances and thus of the gaps
Neutron powder diffraction at high pres- between formerly localized molecular orbi-
sure has been carried out since the 1960s. tals. Increasing overlap of the orbitals leads
The development of the Paris–Edinburgh to an increasing delocalization of the elec-
press (see Sec. 10.3.2.1) dramatically in- trons, which in turn enhances electrical
creased the pressure range accessible for in conductivity up to a metallization of the
situ neutron powder diffraction. The quest originally insulating or semiconducting
towards higher pressures is followed even compounds. Narrowing and closing of the
more consequently with revolutionary sap- gap between valence and conduction bands
phire cells (e.g., Goncharenko and Mire- not only has an effect on the electrical con-
beau, 1998) allowing for neutron diffrac- duction but also on the optical absorption.
tion experiments up to pressures of 50 GPa. It is for this reason that optical spectros-
Another remarkable development in high- copy was one of the first methods used to
pressure neutron diffraction is the possibil- study phase transformations in semicon-
ity of performing inelastic single-crystal ductors at high pressure (Slykhouse and
Drickamer, 1958; Paul, 1959; Goñi and NaCl (Cmcm, cinnabar, Imm2, Immm) or,
Syassen, 1998). b-tin (Imma, simple hexagonal) structures.
High-pressure driven phase transforma- An interesting aspect also is that a site-or-
tions have also been extensively studied in dered version of the diatomic b-tin struc-
the past 30 to 40 years using diffraction ture, which previously was believed to be
methods. In the following only an over- of great importance in semiconductor crys-
view of phase transformations in zinc- tal-chemistry, probably does not exist for
blende-type semiconductors will be given, any of the group III–V and II–VI semicon-
without claiming completeness even in this ductors. A further peculiarity, which has
limited segment of semiconductors. A most been found in Si, Ge and GaAs, is the oc-
thorough and comprehensive review of this currence of different phases at a given
subject is given by Nelmes and McMahon pressure, depending whether one is on a
(1998). compressional or decompressional branch
First structural diffraction studies on of the P-path. In a similar way, InSb, HgSe
these materials were performed on pow- and HgTe exhibit intermediate phases dur-
dered samples using energy-dispersive X- ing their sluggish phase transitions from 4-
ray diffraction (EDX) (see Sec. 10.4.4). fold to 6-fold coordination. These interme-
The picture emerging from these studies is diate phases (“hidden” phases) cannot so
a sequence of phase transitions where vol- far be isolated as single phases, but always
ume optimization is achieved by a stepwise occur together with the stable low- or high-
increase in coordination number (see Sec. pressure modification. Most of these pecu-
10.2.2). This increase in coordination num- liarities are believed to be linked to site-
ber was believed to happen through the fol- ordering problems (non-existence of dia-
lowing sequence of phase transformations: tomic b-tin) or to problems of achieving a
4-fold (diamond, zincblende, wurtzite) true equilibrium state when compressing at
Æ 6-fold (NaCl, b-tin) Æ 8-fold (simple room temperature („hidden“ phases).
hexagonal) Æ 8-fold with six close second Systematic experiments at simultaneously
nearest neighbors (bcc, CsCl) or 12-fold high temperature and high pressure are
(ccp, hcp). For a description of the struc- therefore the next experimental step to be
ture types relevant for semiconductor crys- taken to obtain a complete understanding
tal chemistry, see Nelmes and McMahon of semiconductor phase transformations.
(1998). The availability of in situ angle- In the following the phase transforma-
dispersive X-ray powder diffraction tions in Si, Ge, GaAs and InSb will be dis-
(ADX) (see Sec. 10.4.4) has changed this cussed in more detail, because they are in-
simple model dramatically, mainly because structive examples of most of the special
of the much higher resolution of the ADX features encountered in the high-pressure
technique relative to EDX. The overall pic- behavior of group IV, III – V, and II – VI,
ture emerging from these new experiments semiconductors.
is that at least the change from 4-fold to 6-
fold coordination is characterized by a de-
10.5.1.1 Si and Ge
crease in symmetry, in contrast to the gen-
eral crystal-chemical trend of increasing The phase transformations of Si and Ge
symmetry with increasing pressure. The are quite similar. The main differences are
new 6-coordinated phases can be structu- found in large discrepancies in the respec-
rally described as distorted derivatives of tive transformation pressures and in an ad-
10.5 Examples 681
ditional ‘intermediate phase’ in Si at high to transform to a hexagonal closest packed

pressure. Both Si and Ge crystallize in the structure (hcp) (Si-VII) around 41.8 GPa
diamond structure at ambient conditions. (Duclos et al., 1990) and finally to a cubic
The diamond structure can be viewed as closest packed structure (ccp) around
the monatomic version of the zincblende 78 GPa (Duclos et al., 1987). In hcp as well
structure. The zincblende structure is a cu- as ccp, the individual atoms are bonded to
bic closest packing (ccp) of anions with 12 nearest neighbors. The ccp structure is
half of its tetrahedral interstices occupied believed to remain stable until at least 248
by the cations. The hexagonal closest pack- GPa (Duclos et al., 1990).
ing (hcp) analog to the zincblende structure On decompression, five more phases
is the wurtzite structure. At 11 GPa (Si) have been found for Si. Slow pressure re-
and 10.6 GPa (Ge), the diamond-type lease from Si-II (b-tin structure) leads to
structure transforms to the b-tin structure. the Si-III phase (Besson et al., 1987). It
The exact transition pressure is known to adopts a body centered cubic structure
depend on the amount of stress in the sam- (BC8) with 16 atoms per unit cell. The
ple. For silicon, the b-tin structure trans- transition from Si-II to Si-III has recently
forms into an orthorhombically distorted been shown to occur via an intermediate
version of the b-tin structure at 13.2 GPa rhombohedral phase (Crain et al., 1994),
(McMahon and Nelmes, 1993). The corre- which represents a slight rhombohedral
sponding phase transformation also occurs distortion (R8) of the BC8 structure type.
in Ge, but at around 75 GPa (Nelmes et al., Transition pressures from b-tin to R8 and
1996). These orthorhombic phases trans- R8 to BC8 are 10.1 GPa and 2 GPa, respec-
form to the previously detected simple hex- tively. The decompressional behavior of Si
agonal structure at 15.6 GPa (Si) and is different when the rate of pressure re-
~ 85 GPa (Ge) (Ruoff and Li, 1995; Vohra lease is very fast. In this case, Si-II (b-tin)
et al., 1986). A further increase in pressure seems to transform into two tetragonal
causes Si to adopt another orthorhombic phases, whose structures are not yet
(Cmca) structure (Si-VI) around 37.6 GPa known.
(Duclos et al., 1990; Hanfland et al., 1999), Slow decompression of Ge-II (b-tin
where the Si atoms are 10- and 11-coordi- structure) leads to a phase (Ge-III) with a
nated. A remarkable detail of the Si-VI non-centrosymmetric, tetragonal structure
phase is that it is iso-structural to Cs-V (ST12) with space-group P43212. Unex-
(Schwarz et al., 1998). Not only is the to- pectedly, the BC8 structure, which in Si is
pology between these two phases identical, obtained through slow pressure release,
but also the axial ratios and free atomic co- can be grown in Ge by fast decompression
ordinates have almost exactly the same val- from the b-tin structure.
ues. Until now, an analogous phase has not
been found for Ge. Instead, Ge is believed
10.5.1.2 GaAs
to transform to a hexagonal closest packed
structure with a doubled c-axis (dhcp) GaAs is the III – V analogue of Ge. It is
(Vohra et al., 1986) around 100 GPa. The of very great technological importance,
experimental evidence presented in favor which explains the large amount of re-
of such a dhcp structure is, however, scarce search performed on this compound. De-
and further experiments are required to spite the intense scrutiny devoted to GaAs,
confirm this phase. Si, in turn, is observed its high-pressure phase transformations re-
mained a controversial topic for many Two more phases of GaAs have been
years. GaAs has the zincblende structure at found. The first occurs upon decompres-
ambient conditions. A high-pressure phase sion of the GaAs-II phase. It leads to a 4-
transformation to a GaAs-II phase was first coordinated cinnabar structure (McMahon
found through electrical measurements and Nelmes 1997). The other can be grown
(Minomura and Drickamer, 1962). In the by heating the GaAs-II phase to about
1980s, it was recognized that the structure 450 K at a pressure of ~ 14 GPa (McMahon
would most probably be of orthorhombic et al., 1998). Its structure (SC16) is the di-
symmetry. However, space group as well as atomic equivalent of the BC8 structure
structural topology remained ambiguous found in Ge and Si.
(Shimomura et al., 1980; Baublitz and Ru-
off, 1982; Weir et al., 1989) in that a pro-
10.5.1.3 InSb
posed model (space group Pm2m) implied
very close Ga–Ga and As–As contacts. InSb, which at ambient conditions also
ADX experiments again proved to be the adopts the zincblende structure, is probably
key to resolving the puzzle, although the the most impressive example of a phase di-
very similar X-ray scattering power of Ga agram that has been completely rewritten
and As created additional problems. The in the past 5 – 10 years. The problem of
high resolution of the ADX patterns were, very similar scattering powers is even more
however, able to reveal small reflections, severe in InSb than in GaAs. In order to un-
which allowed for an experimental distinc- ravel the crystal chemistry of its high-pres-
tion between Pm2m and Cmcm (which can sure phases, the ADX technique had to be
be described as a distorted NaCl structure). supported by anomalous scattering in order
Ga and As form slightly distorted NaCl- to distinguish between true symmetries and
type planes parallel (001). Neighboring pseudo-symmetries caused by the similar
planes are shifted parallel to [010] by about scattering powers (Nelmes and McMahon,
half a Ga–As distance (Nelmes and McMa- 1998). The peculiar feature of InSb is that
hon, 1998). Upon further compression, two different successions of high-pressure
GaAs has been reported to undergo another phases as a function pressure can be ob-
phase transition to a simple hexagonal served, albeit in a reproducible way. The
structure (Weir et al., 1989). However, a di- basic succession is from the zincblende
atomic compound such as GaAs, cannot type to a superstructure of Cmcm (InSb-
adopt the simple hexagonal structure, al- IV) around 3 GPa. At 9 GPa the structure
though the topology of the atoms does have transforms to an orthorhombic distorted
this symmetry. To adopt this structure, NaCl phase (InSb-III) with space group
GaAs would have to be completely site- Immm. The Immm phase appears to be at-
disordered. It can be speculated that the tained through an intermediate site-disor-
similar scattering power of Ga and As may dered orthorhombic phase with space
make a diffraction pattern (because of lim- group Imma. The transformation from
ited resolution and signal-to-noise ratio) Imma occurs with time (a few hours) or
appear to have simple hexagonal symme- upon slight heating. InSb-III, in turn, trans-
try, when in truth it remains of orthorhom- forms at 17 GPa to a new phase with yet
bic symmetry. This would imply that GaAs unknown structure, and upon further com-
remains in Cmcm structure up to the high- pression over 21 GPa, to a site-disordered
est pressure investigated, i.e., 108 GPa. body centered cubic structure. A new fea-
10.5 Examples 683
ture in this case is that the transition from 10.5.2 Materials in the B–C–N system
the zincblende-type phase to InSb-IV can
also be preceded by an additional phase Diamond (i.e., the high-pressure poly-
transition around 2.1 GPa, which produces morph of carbon) is not only a precious and
a mixture of a site-disordered b-tin phase popular gem stone, it also has great techno-
and the InSb-II phase with orthorhombic logical significance owing to its extraordi-
symmetry (Immm). This mixture recrystal- nary physical properties. The most promi-
lizes to InSb-IV when left for a few hours. nent of these is a Mohs hardness of 10 (di-
Even more peculiar is the fact that if, dur- amond is the hardest material known) mak-
ing the presence of the mixed phase the ing it a most efficient abrasive material.
pressure is increased to about 3 GPa within The reason for the extreme hardness of di-
a few hours, the formation of InSb-IV is amond lies in the combination of high co-
suppressed (Nelmes et al., 1993a). The valency together with the smallness of the
InSb-II phase persists up to 9 GPa where it core electron shell of carbon, distinguish-
does not show any phase transition to InSb- ing it from the isomorphic compounds Si
III, because InSb-III already has the Immm and Ge (see Sec. 10.5.1.1). Also unusual is
structure. the combination of a very high thermal
As outlined in this brief summary of the conductivity coupled with a low electrical
unusual high-pressure behavior of GaAs conductivity, which makes it an interesting
and InSb, there seems to be a number of material as a heat sink in microelectronic
unusual phenomena in the structural be- applications. Graphite (i.e., the low-pres-
havior of zincblende-type semiconductors sure polymorph of C) also has technologi-
at high pressure. These phenomena may cal importance thanks to its electrical (e.g.,
well be linked to kinetic problems as they graphite electrodes) as well as its mechani-
are often encountered during phase trans- cal (steel additive, graphite composite
formations at high pressure. As shown by material, nanotubes) and lubricating prop-
Leinenweber (1993) and Kunz et al. erties. Therefore, (pseudo-)isoelectronic
(1996), such kinetic problems can be alle- systems of C have attracted much attention
viated by adding high temperature to high in the hope of obtaining materials with
pressure. Very often heating a compressed graphite- and diamond-like properties.
sample may eventually lead to a stable There is reason to believe that ternary com-
phase. In a similar way, the nonexistence of pounds in the B–C–N system with graph-
a site-ordered b-tin phase could be linked ite-like structures might be semiconductors
to microstrain and/or kinetics. Mezouar et with high thermal stability or, alternatively,
al. (1999 b) showed that different phases diamond-type phases in the B–C–N field
can be obtained at a given pressure, de- may exhibit abrasive properties exceeding
pending on the microstrain and hydrosta- those of diamond combined again with a
ticity present in a pressure cell. It will thus much higher thermal stability (Kurdyumov
be a challenge in future high-pressure semi- and Solozhenko, 1999).
conductor research to combine high pres- Crystallographic investigations of super-
sure and high temperature while simultane- hard materials in situ at high pressure pose
ously probing the sample. an additional challenge: the extreme hard-
ness of the material causes strong devi-
atoric stresses when compressing a pow-
dered sample (Weidner et al., 1994 a, b).
The deviatoric strain caused by the devi- a pressure of about 8 GPa (5 GPa if a C-so-
atoric stress leads to anisotropic peak lution in a metal melt is used as precursor)
broadening, which in turn makes the accu- usually has to be combined with a tempera-
rate interpretation of powder patterns more ture of ~ 1000 K in order to obtain cubic di-
difficult. amond. The crystal structure of cubic dia-
mond is the monatomic equivalent of the
zincblende structure [see Sec. 10.5.1 and
10.5.2.1 C
Nelmes and McMahon (1998) for detailed
Carbon crystallizes in three different descriptions of these structure types]. Al-
polymorphs (Fig. 10-9 a). Graphite, the though the phase boundary has a positive
stable phase at ambient conditions, is char- slope and thus would extend to much lower
acterized by a stacking of hexagonal pressures at room temperature, graphite
layers, where the individual layers are compressed at room temperature does not
formed by a honeycomb net of 6-mem- readily adopt the diamond structure. In-
bered rings formed by C. The layered stead, pure compression of graphite at
structure is responsible for the very low room temperature up to 12 GPa and subse-
hardness (Mohs 1 – 2) as well as the aniso- quent annealing at about 1000 K leads to
tropic electrical properties of graphite. the formation of a hexagonal form of dia-
Graphite can exist in a hexagonal (2H) and mond (londsdaleite). This transformation
rhombohedral (3R) modification, where is reversible in the temperature range
these polytypes differ in the stacking of the between 1000 and 1300 K. If heated above
layers: ABABAB for 2H and ABCAB- 1400 K, however, londsdaleite converts
CABC for 3R. If graphite is pressurized, it to cubic diamond. Subsequent further
transforms to diamond. For kinetic reasons, changes in the P – T field do not induce any
Figure 10-9. P – T phase diagrams for (a) carbon and (b) BN. Note that the two topologies are identical. (b) is
different from (a) in that it is shifted to lower pressures. This causes the phase boundary between h-BN (graph-
ite-type) and c-BN (diamond-equivalent) to cross the zero pressure line at ~ 1600 K, thus making c-BN the
stable phase at ambient conditions. After Bundy (1989) and Solozhenko (1994, 1999).
10.5 Examples 685
further phase transformations as long as we type cubic phase (c-BN) crosses the 1 bar
remain above the graphite–diamond phase line at a temperature of about 1600 K. This
boundary. This implies that londsdaleite, makes the zincblende-type c-BN the stable
which crystallizes in the wurtzite structure, phase at ambient conditions (Solozhenko,
is metastable with respect to diamond. The 1994; Solozhenko et al., 1999). However,
stability of diamond under high pressure the growth of c-BN at atmospheric pres-
seems to be much higher than that of sili- sure is only possible in the presence of
con or germanium (see Sec. 10.5.1.1). No supercritical fluids and seed crystals. Spon-
phase changes have been observed up to taneous crystallization of c-BN has not
pressures of 550 GPa (Xu et al., 1986). been observed at pressures below 2 GPa
This is attributed to the fact that C, unlike (Solozhenko, 1994). This fact has, of course,
Si and Ge, does not have any core p-elec- important technological consequences, as
trons, which brings the valence electrons in c-BN is the phase with the desired abrasive
C much closer to the nucleus and thus sta- properties. The graphite-like phase (h-BN)
bilizes the cubic high symmetry structure is the stable phase at low pressure and high
(Bundy, 1989). temperatures. The graphite-like phase can
In the B–C–N phase field, there are four occur as hexagonal (2H) or rhombohedral
important ternary phases known, namely (3R) polytype (e.g., Britun et al., 1999).
BCN, BC2N, BC3N and BC4N. In addition Like most binary and ternary phases in the
to these, there are numerous binary phases B–C–N system, h-BN has a turbostratic
such as BN and several B-rich B–N solid structure. This structure can be described
solutions, B4C and BC3 along the B–C line as a graphite structure with good in-plane
and C3N4 as the only binary carbon–ni- ordering but random orientation of the in-
tride. Many of these phases can be synthe- dividual planes around the layer normal.
sized as graphite-like compounds through This leads to the absence of hkl or h0l
sputter deposition or chemical vapor depo- reflections, while hk0 and 00l reflections
sition (CVD), yielding thin films suitable exhibit very irregular peak shapes (e.g.,
for abrasive protection. Because in this Andreev and Lundström, 1994). The phase
chapter we are mainly interested in high- boundary between c-BN and h-BN has a
pressure phase transformations, we will fo- positive slope and meets the liquidus in a
cus on structural transformations relevant triple point at 3480 K and 5.9 GPa (Solo-
to the synthesis of bulk samples. zhenko et al., 1999). Similar to the carbon
phase diagram, the BN also forms a meta-
stable phase (w-BN) at high pressure,
10.5.2.2 B–N
which crystallizes in the wurtzite structure.
The III-V analogue to carbon is BN. Ow- The transitions between the various poly-
ing to its industrial importance as a dia- types are either martensitic or diffusional,
mond substitute in drilling and polishing depending on the synthesis and formation
applications, a large amount of research conditions (shock compression vs catalytic
has been done in this field. The topology of synthesis) as well as on the ordering degree
the phase diagram is very similar to that of of the precursor material (Kurdyumov,
carbon (Fig. 10-9). However, it is shifted 1995). A curious but potentially very inter-
towards lower temperatures so that the esting BN phase has recently been synthe-
phase boundary between the graphite-type sized by laser heating a c-BN sample up to
hexagonal phase (h-BN) and a zincblende- 5000 K at pressures between 5 – 15 GPa
(Golberg et al., 1996). This procedure was pressurized to about 7 GPa and tem-
yielded multi-walled nanotubes, which did peratures between 600 and 1200 K. This
not contain any other inclusions as they are procedure yielded faceted crystals of a-
frequently obtained when producing nano- and b-C3N4.
tubes via a plasma-arc discharge method.
10.5.2.4 B–C
10.5.2.3 C–N
The crystal chemistry along the B–C
A lot of effort has been put into the syn- axis has been well studied (Thévenot,
thesis of sp3-bonded carbon nitrides, again 1990). The most important binary is B4C.
in the hope of creating super-hard materi- Its structure is a stuffed derivative of a-
als. This hope has been supported by theo- boron, where in B4C the individual boron
retical calculations predicting the stability icosahedral chains are linked through a
of C3N4 phases whose bulk moduli were three-atom C–B–C chain. So far no high-
calculated to exceed the value of diamond pressure phases of B4C are known. How-
(Sung and Sung, 1996; Liu and Cohen, ever, owing to the electron-deficient nature
1989, 1990). These theoretical studies pre- of the intra-icosahedral bonding, boron-
dict five different phases of C3N4. Besides rich solids such as B4C are expected to be
a graphite-like phase, an a- and a b-phase inverted-molecular solids, where the com-
are expected. These would be structurally pressibility of the molecules is higher than
identical to the corresponding phases of that of the intermolecular links. This is un-
Si3N4. Their expected bulk moduli are 425 usual because in most cases the compress-
GPa and 451 GPa, respectively. A more re- ibility of a molecular crystal is controlled
cent theoretical study (Teter and Hemley, by the stiffness of the weak van der Waals
1996) predicts two more high-pressure bonding between the molecules, while the
phases. A cubic phase should be structu- bonds within the molecules are rather rigid.
rally identical to the high-pressure Zn2SiO4 The inverted compressibility was experi-
polymorph. The bulk modulus for this mentally confirmed for B4C (Nelmes et al.,
phase is calculated to be higher than that of 1995). Such an inverted behavior is re-
diamond (496 GPa vs. 444 GPa). Nguyen markable for a compound like B4C in view
et al. (1998) managed to synthesize a bulk of the very strong and covalent B bonds
C3N4 phase by laser heating a C60 film with within the B12 molecules.
liquid nitrogen in a diamond anvil cell at
2000 K and 18 GPa. The diffraction pattern
10.5.2.5 B–C–N
of the retrieved phase is very close to the
one predicted theoretically for the cubic Ternary phases in the B–C–N system are
structure by Teter and Hemley (1996). The rather difficult to grow, mainly because of
measured compressibility, however, is the high stability of BN–C mixtures rela-
much higher than predicted (249 GPa vs. tive to true B–C–N phases. Nevertheless,
449 GPa). The phase was found to decom- several ternary phases have been synthe-
pose upon pressure release below 14 GPa. sized (see Solozhenko (1997) for a review).
Successful attempts at bulk synthesis of Graphite-type phases with ternary compo-
C3N4 have also been reported by He et al. sitions are mostly grown using CVD meth-
(1998) and Dymont et al. (1999). In both ods or nitration of B- and N-containing
cases a precursor containing C, N and H compounds in the condensed state. Investi-
10.5 Examples 687
gations of possible diamond-type struc- commonly called ice. Presently, there are
tures in this system are difficult owing to 12 crystalline and two amorphous phases
the difficulty of distinguishing true ternary of ice accepted. Crystalline ice phases are
phases from a mechanical mixture of C and labeled with Roman numerals from I to XII
BN with diffraction methods. In order to in the approximate sequence of their dis-
create diamond-type ternary phases (exhib- covery. The high structural variability can
iting zincblende- or wurtzite-related struc- be explained by a subtle interplay between
tures), the diffusion during the high-pres- enthalpic and entropic contributions. The
sure/high-temperature experiment has to be flexible O–H…O hydrogen bond with a
suppressed, otherwise the ternary phases strong O–H bond and at least one weak
decompose into c-BN and diamond. This O…H hydrogen bond around each hydro-
can be understood on the basis of the sig- gen enables the formation of a large num-
nificant positive deviation of Cx(BN)1–x ber of energetically similar topologies. In
solid solutions from Vegard’s law. It im- addition, the total energy of the system can
plies that ternary high-pressure phases will be significantly altered through the entropy
have a higher volume than a linear combi- of a disordered distribution of the O–H…O
nation of the components involved and will asymmetry in the framework.
thus, provided sufficient atomic diffusion The flexibility of the hydrogen bonds is
occurs, be unstable at high pressure. The intimately linked to their asymmetric
most efficient way of preventing diffusion geometry. With simple bond valence and
and thus segregation is shock compres- ionic radius arguments, it is possible to
sion (e.g., Komatsu et al., 1996). Another show that this asymmetry is a direct conse-
possibility of transforming graphite-type quence of the fact that the closest un-
C–B–N material into a diamond-type phase strained O–O distance is longer than the
is by laser heating a sample in a diamond sum of two relaxed O–H bonds. This would
anvil cell (Knittle et al., 1995). The sam- lead to an underbonding of the hydrogen
ples were grown at pressures between 30 atom. As outlined above (see Sec. 10.2.2),
and 50 GPa combined with temperatures such an underbonding can be compensated
between 2000 and 2500 K. This approach for by an asymmetric coordination geome-
appears so far to be the only published ex- try. This model implies straight O–H..H
ample where a ternary phase could actually geometries for strong hydrogen bonds, but
be grown from both ternary graphitic pre- bent (because under-constrained) O–H…O
cursors and mechanical mixtures. It should configurations for weaker hydrogen bonds
be mentioned, however, that the analytical (Brown, 1976, 1995). Therefore the model
data of Knittel et al., (1995) could equally correctly predicts the counter-intuitive fact
well be interpreted as resulting from me- that high-pressure phases of ice will tend to
chanical mixtures of dispersed diamond have weaker hydrogen bonds. This is due
and c-BN (Solozhenko, 1997). to the requirement for a denser packing of
water molecules in high-pressure phases
(see Sec. 10.2.1), leading to bent and thus
10.5.3 H2O
weakened hydrogen bonds.
Water not only occurs as the life-sustain- The hydrogen bond is not only critical
ing liquid phase, but also has an exception- for H2O with respect to its structural as-
ally rich phase diagram (Fig. 10-10) in its pects, but also with respect to its electronic
solid state. All solid phases of water are properties. The disordering of the hydro-
Figure 10-10. (a) P – T phase diagram for H2O, (b) enlargement of the central section of (a). Note that in (a)
the pressure scale is logarithmic. (After Petrenko and Whitworth, 1999.)
gen bonds strongly increases the polariz- tions, lack of kinetics favors a chaotic
ability and electrical conductivity of H2O. framework collapse leading to an amor-
There exists a vast literature on the struc- phous phase (Sec. 10.5.3.6).
ture, physics and chemistry of H2O, includ-
ing a very comprehensive and thorough re-
10.5.3.1 Ice Ih, XI and Ic
view by Petrenko and Whitworth (1999).
A common feature of all ice phases is The ice phase occurring at ambient pres-
that the H2O molecules form a three-di- sure (ice Ih) has a hexagonal structure,
mensional tetrahedral framework. This, where the oxygen atoms occupy the posi-
combined with the versatility of the hydro- tions of the monatomic version of the wurt-
gen bonds, makes H2O an interesting study zite structure (see Sec. 10.5.1.1). The hy-
case for understanding the possible mecha- drogen atoms are placed asymmetrically
nisms of a framework to reduce its volume. between neighboring oxygen atoms. Each
At low pressures, H2O adopts rather open oxygen thus forms two O–H bonds and two
frameworks (see Sec. 10.5.3.1). At pres- O…H hydrogen bonds. The distribution of
sures above ca. 0.6 GPa, however, it is able the O–H bonds and O…H hydrogen bonds
to reduce the molar volume significantly is disordered, resulting in space group
by forming two interpenetrating, but un- P63/mmc (hcp space group). The local dis-
connected frameworks (see Sec. 10.5.3.5). tribution of the H atoms, however, has to
In the intermediate range between 0.2 and follow the two ice rules (Bernal and
0.6 GPa, it adjusts to the imposed volume Fowler, 1933), which are valid for all other
constraint mainly by straining the individ- ice phases; (1) there are always two O–H
ual bonds (Sec. 10.5.3.3 and Sec. 10.5.3.4), bonds for each oxygen, (2) there is only
which can lead to a weakening of O…H one H per O–O contact. These rules follow
hydrogen bonds with increasing pressure, directly out of bond-valence rules and cat-
as mentioned above. Under certain condi- ion–cation repulsion considerations.
10.5 Examples 689
The cubic analog to Ih exists as a metastability-field, its structural topology is

stable polymorph designated Ic (ccp space characterized by puckered 6-membered
group Fd3m). It is structurally character- rings of H2O, which are stacked along the
ized by a diamond-type sub-lattice of the hexagonal c-axis to form open channels,
oxygen atoms. Ice Ic can be formed by re- similar to the ones formed by Ih. The space
–
covering any of the high-pressure phases in group of ice II is R3. The higher density of
liquid nitrogen and subsequently heating it ice II relative to ice Ih is mainly achieved
to 120 – 170 K. There is no way of produc- by the puckering of the 6-membered rings
ing Ic directly from Ih, heating Ic to about and the denser stacking of the rings by
200 K on the other hand transforms it to means of bending and thus weakening of
Ih. The energetic difference between Ic the O–H…H bonds. Furthermore, the chan-
and Ih is only between 13 and 50 Jmol–1 nels of ice II can be occupied by He atoms,
(Handa et al., 1988). Hexagonal ice can in creating a He hydrate similar to clathrate
principle be transformed into an ordered hydrates (Londono et al., 1993). Incorpora-
polymorph around 72 K (ice XI). However, tion of He actually extends the stability
the process of collectively rearranging field of ice II up to the liquidus. Heating ice
O-H…H bonds can only take place if there II in the absence of He, however, trans-
are a sufficient number of protonic point forms ice II to ice III.
defects present, which enable the propaga-
tion of the bond reorientations. Such de-
10.5.3.3 Ice III and Ice IX
fects can easily be incorporated through
doping with alkali hydroxides. The phase Ice III is structurally characterized by
transition Ih Æ XI was first observed puckered 5-membered rings that are mutu-
through dielectric permittivity (Kawada, ally linked in three dimensions to give a
1972). The exact structure is now agreed to complicated framework of tetragonal sym-
have the orthorhombic space group Cmc21, metry, space group P41212 (Kamb and Pra-
with all OH-dipoles parallel c oriented in kash, 1968). The hydrogen distribution
a ferroelectric way (Line and Whitworth, within the stability field of ice III is highly
1996; Jackson et al. 1997). disordered (Whalley et al., 1968). How-
A special and well known common prop- ever, it can be ordered by cooling at con-
erty of all these low-pressure phases of stant pressure. If the temperature is re-
H2O is that their density (0.92 – 0.934 g duced into the stability field of ice II, ice
cm–3) is somewhat lower than that of liquid III remains metastable. Between – 65 and
water. – 108°C, the ordering increases gradually,
inducing the phase transformation to ice IX
(the ordered variety of ice III) as revealed
10.5.3.2 Ice II
by measurements of electrical permittivity.
The first true high-pressure phase was Neutron diffraction, however, reveals a sig-
discovered in 1900 by Tammann. It can be nificant amount of residual ordering also in
formed through compression of Ih at tem- ice III (Londono et al., 1993, Kuhs et al.,
peratures between – 35° and – 70 °C. Its 1998). Ice III is the high-pressure poly-
structure is quite peculiar in two ways. morph of H2O with the lowest density
First, it has a perfectly ordered arrange- (1.165 g cm–3). However, unlike Ice Ih, its
ment of the hydrogen bond pattern. Sec- density at the liquidus is higher than the
ondly, despite the high pressure of its corresponding density of the liquid phase,
resulting in a positive slope of the liquidus within the stability field of ice V. The
in P – T space. new phase can reproducibly be grown at
0.55 GPa and 260 K with slow cooling
rates. Its structure consists of two symmet-
10.5.3.4 Ice V, Ice IV and Ice XII
rically independent sets of oxygen atoms.
Further compression of ice III or ice II The first one forms zig-zag chains parallel
leads to ice V. The structure of ice V is a c. Neighboring chains are rotated by 90°.
complex framework built by 4- and 5- These chains are connected by the sec-
membered oxygen rings (Kamb et al., ond type of oxygen atoms thus forming
1967). Each oxygen atom is tetrahedrally a tetrahedral framework, whose density
surrounded by four neighboring oxygen at- (1.437 g cm–3) is very close to that of ice
oms. The symmetry is monoclinic with IV. While in ice IV, the increase in density
space group A2/a. The hydrogen atoms are relative to ice V is achieved by self-pene-
distributed in a disordered arrangement tration, ice XII compresses by additional
between the oxygen links. In fact, the two- bending. In both metastable phases, the hy-
fold axis of this space group does not allow drogen positions are disordered between
for any proton ordering. Proton ordering, if the O–O links.
it occurs, would thus imply a noncentro-
symmetric space group Aa.
10.5.3.5 Ice VI, Ice VII, Ice VIII, Ice X
A special feature of ice V is that in its
stability field, at least two ice phases with- Compressing ice V beyond 0.6 GPa in-
out their own stability fields have been de- duces a phase transformation to ice VI.
scribed. Ice IV has been observed within This is the first example of an ice phase,
the stability field of ice II, ice V and ice VI. where the structure is characterized by two
Even a liquidus for ice IV, which lies interpenetrating frameworks which are not
within the field of these solid phases has linked together (‘self-clathrates’, Kamb,
been established (Chou and Haselton, (1965)). In ice VI, the topologies of two
1998). The structure of ice IV is quite re- frameworks are identical. They are sym-
markable. Only slightly puckered 6-mem- metrically related by an inversion and
bered rings are linked together within a shifted relative to each other by a/2 + b/2 +
plane by strongly elongated 6-membered c/2. The individual frameworks can be de-
rings forming a puckered plane with scribed as chains of tetrahedra, extended
trigonal symmetry. The planes are stacked along c. All tetrahedra of one framework
in an ABABAB sequence where A- and B- have the same orientation. The chains of
planes are shifted relative to each other by each framework are connected parallel a
the value of the hexagonal a-axis. Like- and b with each other, but unconnected
type planes (A–A and B–B, respectively) with the chains of the other sublattice.
are linked together by hydrogen bonds Ice VII has, together with ice VIII and
through the regular 6-membered rings of ice X, the densest H2O topology. Its pack-
the neighboring planes. This results in a ing is as efficient as it is simple and can be
single, self-penetrating framework with a described by two interpenetrating frame-
higher density (1.436 g cm–3) than ice V works of the cubic ice Ic. The hydrogen po-
(1.402 g cm–3). sitions in ice VII are completely disordered
Only recently, another new phase of ice (Walrafen et al., 1982; Kuhs et al., 1984).
has been described by Lobban et al. (1998) The interesting aspect of this structure is
10.5 Examples 691
that the oxygen atoms form a body cen- In ice X, the O–O contacts are squeezed
tered cubic lattice. This implies that each sufficiently close together that the hydro-
oxygen atom has eight nearest neighbors. gen can sit unstrained in the center of the
However, it forms hydrogen bonds only to O–O-link. Such a symmetric O–H–O ar-
four of them in tetrahedral orientation and rangement thus has no potential for disor-
in this way forms two interpenetrating but der in the first place. Such a phase is differ-
independent frameworks. This also implies ent from a double-well phase with a dy-
that the O–O distances to the hydrogen- namically disordered proton, which from a
bonded oxygen neighbors are equivalent to physical point of view should still be called
the non-bonded O–O contacts. ice VII. Ice X has been predicted by Kamb
The disorder–order transition from ice and Davis (1964). Although experimental
VII to ice VIII was first observed through evidence for a symmetric phase was ob-
dielectric measurements (Whalley et al., served from IR spectroscopic studies by
1966). Although the structure crystallizes the appearance of a new mode characteris-
in the tetragonal space group I41/amd, the tic of a symmetric O–H–O configuration
positions of the oxygen atoms deviate only (Goncharov et al., 1996; Aoki et al., 1996,
little from a body centered cubic lattice, Pruzan et al., 1997), it should be pointed
confirming the order–disorder nature of the out that up to now there is no direct experi-
phase transformation of ice VII to ice VIII mental evidence for a truly proton-ordered
(Kuhs et al., 1984). The ordering pattern of single-well phase of ice X. This lack of ex-
the O–H bonds is ferroelectric within one perimental evidence is due to the difficulty
sub-lattice, but antiferroelectric between the of performing experiments that are able to
two frameworks. In ice VIII, the O–D dis- localize hydrogen or deuterium positions in
tances remain almost constant with increas- the pressure range of 60 GPa and more.
ing pressure (Nelmes et al., 1993 b), al-
though the O–O distances decrease signifi-
10.5.3.6 Amorphous ice at high
cantly. This leads to a decrease in the dis-
pressure
tance between the two potential wells of the
hydrogen position. Therefore, the entropy As mentioned above, there are two
difference between the ordered and disor- amorphous high-pressure phases of H2O in
dered state decreases gradually with in- the solid state. A first phase is formed by
creasing pressure, leading to a progressively compressing ice Ih at 77 K (where slow ki-
steeper phase boundary VII–VIII, which netics inhibit the recrystallization of ice II)
eventually ends as a vertical slope at 62 GPa to the extrapolated liquidus of Ih. Its den-
(Pruzan et al., 1993). At this point, the en- sity in the quenched state is 1.17 g cm–3.
ergy barrier between the two proton posi- Upon heating, this high-density amorphous
tions is sufficiently small to be crossed with (HDA) phase transforms to a second amor-
the thermal energy of the protons, thus lead- phous phase with lower density (0.94 g
ing to a dynamic disorder. This phase thus cm–3) at 125 K (LDA). LDA, in turn can be
has the average position of the hydrogen at- transformed into HDA at 77 K by compres-
oms in the middle between the two oxygen sion to 0.6 GPa.
atoms. An identical structural configuration The research on the high-pressure phases
can be expected for a structure with a single of ice is far from completed. The under-
potential minimum as is predicted for ice X standing of the various amorphous phases
at pressures exceeding 62 GPa. and the theoretical and experimental quest
for ice X and beyond are just the most Angel, R. J., Downs, R. T., Finger, L. W. (2000), in:
Comparative Crystal Chemistry, Reviews in Mine-
prominent examples of a lively topic with ralogy, Vol. 40: Hazen, R. M. (Ed.). Washington
relevance to physical chemistry and mate- DC: Mineralogical Society of America, Chapter 5,
rials science. in press.
Aoki, K., Yamawaki, H., Sakashita, M., Fujihisa, H.
(1996), Phys. Rev. B 54, 15673.
10.6 Acknowledgements Arlt, T., Kunz, M., Stolz, J., Armbruster, T., Angel,
R. J. (2000), Contrib. Mineral. Petr. 138, 35.
Bacon, G. E. (1975), Neutron Diffraction, 3rd edn.
The author wishes to thank Vladimir Oxford: Clarendon Press.
Solozhenko, Richard Nelmes, Mohammed Bassett, J. M., Ming, L.-C. (1972), Phys. Earth
Planet. Mater. 6, 154.
Mezouar, Guillaume Fiquet, and Denis An- Bassett, W. A., Weathers, M. S. (1987), in: AGU
drault for valuable discussions. This work Geophysical Monograph, Vol. 39, Mineral Phys-
received support from the ‘Schweizerische ics 2, Manghnanni, M. H., Syono, Y. (Eds.).
Washington DC: American Geophysical Union,
Nationalfonds’ (Swiss National Science p. 129.
Foundation) through grant # 21-052682.97. Bassett, W. A., Shen, M. J., Bucknum, M. J., Ming
Chou, I. (1993), Rev. Sci. Instrum. 64, 2340.
Baublitz, M., Ruoff, A. L. (1982), J. Appl. Phys. 53,
6179.
10.7 References Baughman, R. H., Stafström, S., Cui, C., Dantas, S.
O. (1998), Science 279, 1522.
Abd-Elmeguid, M. M., Micklitz, H., Buschow, K. J. Belitsky, I. A., Fursenko, B. A., Gabuda, S. B., Khol-
H. (1980), J. Phys. IV, Colloque, 41, C1. dev, O. V., Seryotkin, Y. V. (1992), Phys. Chem.
Acha, C., Loureiro, S. M., Chaillout, C., Tholence, J. Minerals 18, 497.
L., Capponi, J. J., Marezio, M., Nunez-Regueiro, Benedek, G. B., Purcell, E. M. (1954), J. Chem.
M. (1997), Physica C 282–287, 1167. Phys. 22, 2003.
Ahrens, T. J. (1980), Science 207, 1035. Bernal, J. D., Fowler, R. H. (1933), J. Chem. Phys. 1,
Ahrens, T. J. (1987), in: Methods of Experimental 515.
Physics, Vol. 24A: Sammis, C. G. Henyey, T. L. Bertani, R., Mali, M., Roos, J., Brinkmann, D.
(Eds.). New York: Academic Press, p. 185. (1992), Rev. Sci. Instrum. 63, 3303.
Akimoto, S. (1987), in: AGU Geophysical Mono- Besson, J. M., Mokhtari, E. H., Gonzales, J., Weill,
graph, Vol. 39, Mineral Physics 2: Manghnanni, G. (1987), Phys. Rev. Lett. 59, 473.
M. H., Syono, Y. (Eds.). Washington DC: Ameri- Besson, J. M., Hamel, G., Grima, T., Nelmes, R. J.,
can Geophysical Union, p. 1. Loveday, J. S., Hull, S., Häusermann, D. (1992),
Akitt, J. W. (1983), NMR and Chemistry. New York: High Pressure Res. 8, 625.
Chapman and Hall. Binder, K. (2001), in: Phase Transformations in Ma-
Anderson, M. S., Swenson C. A. (1984), J. Appl. terials: Kostorz, G. (Ed.). Weinheim: Wiley-VCH,
Phys. 56, 2697. Chap. 4.
Andrault, D., Fiquet, G., Peyronneau, J., Itié, J. P., Boehler, R., Chopelas, A. (1991), Geophys. Res. Lett.
Richet, P., Gillet, Ph. (1996), in: Proceedings of 18, 1147.
the 3rd NIRIM International Symposium on Ad- Bridgman, P. W. (1935), Proc. Am. Acad. Arts Sci.
vanced Materials (ISAM ’96), National Institute 70, 285.
for Research in Inorganic Materials, Tsukuba, Ja- Bridgman, P. W. (1940), Proc. Am. Acad. Arts Sci.
pan, p. 297. 74, 11.
Andreev, Y. G., Lundström, T. (1994), J. Appl. Crys- Bridgman, P. W. (1946), Collected Experimental Pa-
tallogr. 27, 767. pers of P. W. Bridgman, Vol. VI, p. 164.
Angel, R. J. (2000a), in: Comparative Crystal Chem- Bridgman, P. W. (1952), Proc. Am. Acad. Arts Sci.
istry, MSA Reviews in Mineralogy, Vol. 40: Hazen, 81, 167.
R. M. (Ed.). Washington DC: Mineralogical So- Britun, V. F., Kurdyumov, A. V., Petrusha, I. A.
ciety of America, Chapter 8, in press. (1999), Mater. Lett. 41, 83.
Angel, R. J. (2000b), in: Transformation Processes in Brown, G. E. Jr, Calas, G., Waychunas, G. A., Petiau,
Minerals, MSA Reviews in Mineralogy, Vol. 39: J. (1988), in: Spectroscopy Methods, MSA Reviews
Redfern, S., Carpenter, M., Salje E., Seifert F. in Mineralogy, Vol. 18: Hawthorne F. C. (Ed.).
(Eds.). Washington DC: Mineralogical Society of Washington DC: Mineralogical Society of Amer-
America, p. 38. ica, p. 431.
Angel, R. J., Allan, D. R., Miletich, R., Finger, L. W. Brown, I. D. (1976), Acta Crystallogr. A 32, 24.
(1997), J. Appl. Crystallogr. 30, 461. Brown, I. D. (1992), Acta Crystallogr. B 48, 553.
10.7 References 693
Brown, I. D. (1995), Can. J. Phys. 73, 676. Hanfland, M., Schwarz, U., Syassen, K., Takemura,
Bundy, F. P. (1989), Physica A 156, 169. K. (1999), Phys. Rev. Lett. 82, 1197.
Buontempo, U., Filipponi, A., Farcia-Martinez, D., Hazen, R. M. (1983), Science 219, 1065.
Postorino, P., Mezouar, M., Itié, J. P. (1998), Phys. Hazen, R. M, Finger, L. W. (1978), Geological Soci-
Rev. Lett. 80, 1912. ety of America, Annual Meeting 1978, Toronto;
Carpenter, M. A., Salje, E. K. H. (1998), Eur. J. Min- Abstracts, 10, 418.
eral. 10, 693. Hazen, R. M., Finger, L. W. (1981), Rev. Sci. In-
Carpenter, M. A, Salje, E. K. H, Graeme-Barber, A. strum. 52, 75.
(1998), Eur. J. Mineral. 10, 621. Hazen, R. M., Finger L. W. (1982), Comparative
Chou, J. M., Haselton, H. T. (1998), Rev. High Pres- Crystal Chemistry. New York: John Wiley & Sons.
sure Sci. Technol. 7, 1132. Hazen, R. M., Navrotsky, A. (1996), Am. Mineral.
Crain, J., Ackland, G. J., Maclean, J. R., Piltz, R. O., 81, 1021.
Hatton, P. D., Pawley, G. S. (1994), Phys. Rev. B Hazen, R. M., Yang, H. (1999), Am. Mineral. 84,
50, 13043. 1956.
Datchi, F., LeToullec, R., Loubeyre, P. (1997), J. He, D. W., Zhang, F. X., Zhang, X. Y., Qin, Z. C.,
Appl. Phys. 81, 3333. Zhang, M., Liu, R. P., Xu, Y. F., Wang, W. K.
Dubrovinsky, L. S., Saxena, S. K., Lazor, P. (1997), (1998), J. Mater. Res. 13(12), 3458.
Geophys. Res. Lett. 24, 1835. Hemley, R. J., Prewitt, C. T., Kingma, K. J. (1994) in:
Duclos, S. J., Vohra, Y. K., Ruoff, A. L. (1987), Phys. Silica, MSA Reviews in Mineralogy, Vol. 29:
Rev. Lett. 58, 775. Heany, P. J., Prewitt, C. T. Gibbs, G. V. (Eds.).
Duclos, S. J., Vohra, Y. K., Ruoff, A. L. (1990), Phys. Washington DC: Mineralogical Society of Amer-
Rev. B 41, 12021. ica, p. 41.
Dymont, V. P., Nekrashevich, E. M., Starchenko, I. Holzapfel, W. B. (1997), in: High-Pressure Tech-
M. (1999), Solid State Commun. 111, 443. niques in Chemistry and Physics: Holzapfel, W.
Fiquet, G., Andrault, D., Itié J. P., Gillet, P., Richet, B., Isaacs, N. S. (Eds.). Oxford: Oxford University
P. (1994), EGS XIX General Assembly, Annales Press, p. 64.
Geophysicae, C6, Supplement I, Vol. 12, 24. Ivanov, A. N., Nikolaev, N. A., Pashkin, N. V., Sa-
Fiquet, G., Dewaele, A., Andrault, D., Kunz, M., venko, B. N., Smirnov, L. S., Taran, Y. V. (1991),
LeBihan, T. (2000), Geophys. Res. Lett. 27, 1; 21. Institute of Theoretical and Experimental Physics.
Fukunaga, O., Yamaoka, S., Akaishi, M., Kanda, H., Preprint No. 91 - 76. Dubna.
Osawa, T., Shimomura, Nagashima, T., Yoshi- Jackson, S. M., Nield, V. M., Whitworth, R. W.,
kawa, M. (1987), in: AGU Geophysical Mono- Oguro, M., Wilson, C. C. (1997), J. Phys. Chem. B
graph, Vol. 39, Mineral Physics 2: Manghnanni, 101, 6142.
M. H., Syono, Y. (Eds.). Washington DC: Ameri- Jamieson, J. (1957), J. Geol. 65, 334.
can Geophysical Union, p. 17. Jamieson, J. C., Lawson, A. W., Nachtrieb N. D.
Geisinger, K. L., Gibbs, G. V., Navrotsky, A. (1985), (1959), Rev. Sci. Instrum. 30, 1016.
Phys. Chem. Minerals 11, 266. Kamb, B., Davis, B. L. (1964), Proc. Natl Acad. Sci.
Gillet, P., Hemley, R. J., McMillan, P. F. (1998), in: USA 52, 1433.
Ultra-High Pressure Mineralogy, MSA Reviews in Kamb, B. (1965), Science 150, 205.
Mineralogy, Vol. 37: Hemley, R. J. (Ed.). Wash- Kamb, B., Prakash, A., Knobler, C. (1967), Acta
ington DC: Mineralogical Society of America, Crystallogr. 22, 706.
p. 525. Kamb, B., Prakash, A. (1968), Acta Crystallogr. B
Glazer, A. M. (1972), Acta Crystallogr. B 28, 2284. 24, 1317.
Golberg, D., Bando, Y., Eremets, M., Takemura, K., Katayama, Y., Mezouar, M., Itié, J. P., Besson, J. M.,
Kurashime, K., Yusa, H. (1996), Appl. Phys. Lett. Syfosse, G., LeFevre, P., DiCicco, A. (1997), J.
69, 2045. Phys. IV, Colloque 7, C2.
Goncharenko, I. N., Mirebeau, I. (1998), Rev. High Kawada, S. (1972), J. Phys. Soc. Jpn. 32, 1442.
Pressure Sci. Technol. 7, 475. Kirkpatrick, R. J. (1988), in: Spectroscopy Methods,
Goncharov, A. F., Struzhkin, V. V., Somayazulu, M. MSA Reviews in Mineralogy, Vol. 18: Hawthorne
S., Hemley, R. J., Mao, H. K. (1996), Science 273, F. C. (Ed.). Washington DC: Mineralogical Society
218. of America, p. 341.
Goñi, A. R., Syassen, K. (1998), Semiconductors and Klotz, S., Besson, J. M., Hamel, G., Nelmes, R. J.,
Semimetals 54. New York: Academic Press, p. 247. Loveday, J. S., Marshall, W. G. (1996), High Pres-
Haines, J., Leger, J. M., Chateau, C., Bini, R., Ulivi, sure Res. 14, 294.
L. (1998), Phys. Rev. B 58, R2909. Klotz, S., Besson, J. M., Braden, M., Karch, K., Pa-
Han, C. Y., Lin, W., Wu, Y. S., Yin, B., Tang, D. S. vone, P., Strauch, D., Marshall, W. G. (1997),
(1999), Physica C 282–287, 1189. Phys. Rev. Lett. 79, 1313.
Handa, Y. P., Klug, D. D., Whalley, E. (1988), Can. J. Kluthe, S., Markendorf, R., Mali, M., Roos, J.,
Chem. 66, 919. Brinkmann, D. (1996), Phys. Rev. B 53, 11369.
Knittle, E., Kaner, R. B., Jeanloz, R., Cohen, M. L. McMahon, M. I., Nelmes, R. J. (1993), Phys. Rev. B
(1995), Phys. Rev. B 51, 12149. 47, 8337.
Komatsu, T., Nomura, M., Kakudate, Y., Fujiwara, S. McMahon, M. I., Nelmes, R. J. (1997), Phys. Rev.
(1996), J. Mater. Chem. 6, 1799. Lett. 78, 3697.
Krisch, M., Mermet, A., San Miguel, A., Sette, F., McMahon, M. I., Nelmes, R. J., Allan, D. R., Bel-
Masciovecchio, C., Ruocco, G., Verbeni, R. monte, S. A., Bovornratanaraks, T. (1998), Phys.
(1997), Phys. Rev. B 56, 8691. Rev. Lett. 80, 5564.
Kuhs, W. F., Finney, J. L., Vettier, C., Bliss, D. V. Mermelstein, N. H. (1999), Food Technol. 53, 86.
(1984), J. Chem. Phys. 81, 3612. Merrill, L., Bassett, W. A. (1974), Rev. Sci. Instrum.
Kuhs, W. F., Lobban, C., Finney, J. L. (1998), Rev. 45, 290.
High Pressure Sci. Technol. 7, 1141. Mezouar, M., Besson, J. M., Syfosse, G., Itié, J. P.,
Kunz, M., Leinenweber, K., Parise, J. B., Wu, T. C., Häusermann, D., Hanfland, M. (1996), Phys.
Bassett, W. A., Brister, K., Weidner, D. J., Vau- Status Solidi B 198, 403.
ghan, M. T., Wang, Y. (1996), High Pressure Res. Mezouar, M., LeBihan, T., Libotte, H., LeGodec, Y.,
14, 311. Häusermann, D. (1999a), J. Synchrotron Rad. 6,
Kurdyumov, A. V. (1995), Powder Metall. Metal 1115.
Ceram. 34, 409. Mezouar, M., Libotte, H., Députier, S., LeBihan, T.,
Kurdyumov, A. V., Solozhenko, V. L. (1999), Sverkh- Häusermann, D. (1999b), Phys. Status Solidi B
terdye Materialy (J. Superhard Mater.) 21, 3. 211, 395.
Lawson, A. W., Tang, T. Y. (1950), Rev. Sci. Instrum. Miletich, R., Allen, D. R., Kuhs, W. F. (2000), in:
21, 815. Comparative Crystal Chemistry, MSA Reviews in
Leinenweber, K. (1993), National Synchrotron Light Mineralogy, Vol. 39: Hazen, R. M. (Ed.). Wash-
Source, Annual Report. ington DC: Mineralogical Society of America,
Levien, L., Prewitt, C. T., Weidner, D. J. (1980), Am. Chap. 2, in press.
Mineral. 65, 920. Miller, P. H., Taylor, J. H. (1949), Phys. Rev. 76, 179.
Levien, L., Prewitt, C. T. (1981), Am. Mineral. 66, Minomura, S., Drickamer, H. G. (1962), J. Phys.
324. Chem. Solids 23, 451.
Li, B., Ganglin, C., Gwanmesia, G. D., Liebermann, Mishima, O. H., Calvert, L. D., Whalley, E. (1984),
R. C. (1998), in: Properties of Earth and Planetary Nature 310, 393.
Materials at High Pressure and Temperature. Nasu, S. (1996), High Pressure Res. 14, 405.
AGU Geophysical Monograph, Vol. 101: Mangh- Nelmes, R. J., McMahon, M. I., Hatton, P. D., Piltz,
nani, M. H., Yagi, T. (Eds.). Washington DC: R. O., Crain, J. (1993a), Phys. Rev. B 47, 35; B 48,
American Geophysical Union, p. 41. 9949 (Corrigendum).
Liebermann, R. C., Wang, Y. (1992), in: High Pres- Nelmes, R. J., Loveday, J. S., Wilson, R. M., Besson,
sure Research: Application to Earth and Planetary J. M., Pruzan, Ph., Klotz, S., Hamel, G., Hull, S.
Sciences. AGU Geophysical Monograph, Vol. 67: (1993b), Phys. Rev. Lett. 71, 1192.
Syono, Y., Manghnani, M. H. (Eds.). Washington Nelmes, R. J., Loveday, J. S., Wilson, R. M., Mar-
DC, American Geophysical Union, p. 19. shall, W. G., Besson, J. M., Klotz, S., Hamel, G.,
Line, C. M. B., Whitworth, R. W. (1996), J. Chem. Aselage, T. L., Hull, S. (1995), Phys. Rev. Lett. 74,
Phys. 104, 10008. 2268.
Liu, A. Y., Cohen, M. L. (1989), Science 245, 841. Nelmes, R. J., Liu, H., Belmonte, S. A., Loveday, J.
Liu, A. Y., Cohen, M. L. (1990), Phys. Rev. 41, S., McMahon, M. I., Allan, D. R., Häusermann,
10727. D., Hanfland, M. (1996), Phys. Rev. B 53, R2907.
Lobban, C., Finney, J. L., Kuhs, W. F. (1998), Nature Nelmes, R. J., McMahon, M. I. (1998), Semiconduc-
391, 268. tors and Semimetals 54. New York: Academic
Londono, J. D., Kuhs, W. F., Finney, J. L. (1993), J. Press, p. 145.
Chem. Phys. 98, 4878. Nguyen, J. H., Caldwell, W. A., Benedetti, L. R.,
Lubbers, R., Wortmann, G., Grünsteudel, H. F. Kruger, M. B., Jeanloz, R. (1998), Materials Re-
(1999), Hyperfine Interact. 123, 529. search Society Symposium Proceedings, Vol. 499,
Mao, H. K., Bell, P. M. (1975) Carnegie Institution p. 303.
Washington Year Book, 74, p. 402. Ochiai, A., Suzuki, T., Kasuya, T. (1985), J. Magn.
Mao, H. K., Bell, P. M., Shaner, J. W., Steinberg, D. Magn. Mater. 52, 13.
J. (1978), J. Appl. Phys. 49(6), 3276. Ohwada, K., Nakao, N., Takesu, N., Isobe, M., Ueda,
McCammon, C. (2000), in: Transformation Pro- Y., Fujii, Y. (1999), IUCr XVII Glasgow ’99. Con-
cesses in Minerals, Vol. 40: Redfern, S., Carpenter, gress and General Assembly. Abstract Book,
M., Salje E., Seifert F. (Eds.). Washington DC: P09.OC.002, 114.
Mineralogical Society of America, p. 241. Ondrey, G., Kamiya, T. (2000), Chem. Eng. 107,
McCammon, C., Peyronneau, J., Poirier, J. P. (1998), 26.
Geophys. Res. Lett. 25, 1589. Owen, E. A., Liu Y. H. (1947), Philos. Mag. 38, 354.
10.7 References 695
Parise, J. B., Weidner, D. J., Chen, J., Liebermann, Solozhenko, V. L. (1994), Diamond Related Mater. 4, 1.
R. C., Chen G. (1998), Annu. Rev. Mat. Sci. 28, Solozhenko, V. L. (1997), Eur. J. Solid State Inorg.
349. Chem. 34, 797.
Pasternak, M. P., Taylor, R. D. (1990), Hyperfine Solozhenko, V. L., Turkevich, V. Z., Holzapfel, W. B.
Interact. 62, 89. (1999), J. Phys. Chem. B 103, 2903.
Pasternak, M. P., Taylor, R. D. (1996), in: Mössbauer Solozhenko, V. L. (1999), J. Phys. Chem. B 103,
Spectroscopy applied to Magnetism and Materials 2903.
Science, Vol. 2: Long, G. J., Grandjean F. (Eds.). Stout, G. H., Jensen, L. H. (1989), X-ray Structure
New York: Plenum Press, p. 167. Determination. New York, John Wiley & Sons.
Pasternak, M. P., Farrell, J. N., Taylor, R. D. (1986), Sung, C. M., Sung, M. (1996), Mater. Chem. Phys.
Hyperfine Interact. 28, 837. 43, 1.
Paul, W. (1959), J. Phys. Chem. Solids 8, 196. Tammann, G. (1900), Annalen der Physik, Series 4,
Pedersen, K. (1997), FEMS Microbiol. Rev. 20, 399. 2, 1.
Pelton, A. D. (2001), in: Phase Transformations in Teter, D. M., Hemley, R. J. (1996), Science 271, 53.
Materials: Kostorz, G. (Ed.). Weinheim-VCH, Thévenot, F. (1980), J. Europ. Ceramic Soc. 6, 205.
Chap. 1. Van Valkenburg, A. (1964), Diamond Res. 17.
Petrenko, V. F., Whitworth, R. W. (1999), The Phys- Velde B., Besson J. M. (1982), in: High Pressure in
ics of Ice, Oxford: Oxford University Press. Research and Industry, 8th AIRAPT Conference
Poirier, J. P. (1991), Introduction to the Physics of the Proceedings, Vol. 2, p. 704.
Earth’s Interior, Cambridge: Cambridge Univer- Vohra, Y. K. Brister, K. E., Desgreniers, S., Ruoff, A.
sity Press, Chap. 4.7. L., Chang, K. J., Cohen, M. L. (1986), Phys. Rev.
Pruzan, Ph., Chervin, J. C., Canny, B. (1993), J. Lett. 56, 1944.
Chem. Phys. 99, 9842. Walrafen G. E., Abebe, M., Mauer, F. A., Block, S.,
Pruzan, Ph., Wolanin, E., Gauthier, M., Chervin, J. Piermarini, G. J., Munro, R. (1982), J. Chem.
C., Canny, B., Häusermann, D., Hanfland, M. Phys. 77, 2166.
(1997), J. Phys. Chem. B 101, 6230. Walsh, J. M., Christian, R. H. (1955), Phys. Rev. 97,
Rigden, S. M., Gwanmesia, G., Jackson, I., Lieber- 1544.
mann, R. C. (1992), in: High Pressure Research: Weidner, D. J., Vaughan, M. T., Ko, J., Wang, Y., Liu,
Application to Earth and Planetary Sciences. X., Yeganeh-Haeri, A., Pacalo, R. E., Zhao, Y.
AGU Geophysical Monograph, Vol. 67: Mangh- (1992), in: High Pressure Research: Application to
nani, M., Syono, Y. (Eds.). Washington DC, Amer- Earth and Planetary Sciences. AGU Geophysical
ican Geophysical Union, p. 167. Monograph, Vol. 67: Syono, Y., Manghnani, M. H.
Ross, N. L., Price G. D. (1992) (Eds.), The Stability (Eds.). Washington DC, American Geophysical
of Minerals, London: Chapman & Hall, Chap. 1. Union, p. 13.
Ruoff, A. L., Li, T. (1995), Annu. Rev. Mater. Sci. 25, Weidner, D. J., Wang, Y., Vaughan, M. T. (1994a),
249. Geophys. Res. Lett. 21, 753.
Schwarz, U., Takemura, K., Hanfland, M., Syassen, Weidner, D. J., Wang, Y., Vaughan, M. T. (1994b),
K. (1998), Phys. Rev. Lett. 81, 2711. Science 266, 419.
Shen, A. H., Reichmann, H. J., Chen, G., Angel, R. Weir, C. E., Lippincott, E. R., Van Valkenburg, A.,
J., Bassett, W. A., Spetzler, H. (1998), in: Prop- Bunting, E. N. (1959), J. Res. Natl Bur. Standards
erties of Earth and Planetary Materials at High USA 63A, 55.
Pressure and Temperature. AGU Geophysical Weir, S. T., Vohra, Y. K., Vanderborgh, C. A., Ruoff,
Monograph, Vol. 101: Manghnani, M. H., Yagi, T. A. L. (1989), Phys. Rev. B 39, 1280.
(Eds.). Washington DC, American Geophysical Whalley, E., Davidson, D. W., Heath, J. B. R. (1966),
Union, p. 71. J. Chem. Phys. 45, 3976.
Shimomura, O., Kawamura, T., Fukamachi, T., Ho- Whalley, E., Heath, J. B. R., Davidson, D. W. (1968),
soya, S., Hunter, S., Bienestock, A. (1980), in: J. Chem. Phys. 45, 3976.
High Pressure Science and Technology, Vol. 1: Wittlinger, J., Werner, S., Schulz, H. (1998), Acta
Vodar, B., Marteau, P. (Eds.). Oxford: Pergamon Crystallogr. B 54, 714.
Press, p. 534. Woelk, K., Bargon, J. (1992), Rev. Sci. Instrum. 63,
Shimomura, O., Takemura, K., Fujihisa, H., Fuji, Y., 3307.
Ohishi, Y., Kikegawa, T., Amemiya, Y., Matsu- Xu, J. A, Mao, H. K, Bell, P. M (1986), Science 232,
shita, T. (1992), Rev. Sci. Instrum. 63, 967. 1404.
Sinogekin, S. V., Bass, J. D. (1999), Phys. Rev. B 59, Yagi, T., Akaogi, M., Shimomura, O., Tamai, H., Ak-
R14141. imoto, S. (1987), in: High Pressure Research in
Slykhouse, T. E., Drickamer, H. G. (1958), J. Phys. Mineral Physics. AGU Geophysical Monograph,
Chem. Solids 7, 207. Vol. 39: Manghnani, M. H., Syono, Y. (Eds.).
Smirnov, G. V. (1999), Hyperfine Interact. 123/124, Washington DC, American Geophysical Union,
31. p. 141.
Index
Aaronson–Aaron model 506 – martensites 634

absolute stability limit, directional dendritic alloys
growth 148 – atomic ordering 523 ff
acetone, dendritic growth 115 – dendritic growth 119
acoustic phonons 258 – diffusion 199, 221 ff
activation energy, diffusion 201, 491 – diffusionless transformations 583 ff
activity coefficient – eutectic growth 158
– decomposition 351 – order–disorder phenomena 283 ff
– Henrian 30 – precipitation 314 ff
– phase diagrams 14, 70 – spinodal decomposition 451
adjoining phase regions law 47 f, 56 f – spinodal 363 ff
Ag–Au system 564 amorphization, pressure-induced 661
Ag-based alloys 598 f amorphous alloys 366, 454
Ag–Cd system 632 amplification factor, decomposition 357
Ag–Cu system 150 analytical electron microscopy (AEM) 327
Ag–Mg system 30 analytical field ion microscopy (AFIM) 328, 360, 366
age hardening alloys, microstructure 401 ff angle dispersive X-ray powder diffraction
aggregation (ADX) 678, 680
– clusters 469 anharmonic terms, phase transitions 292
– diffusion 191 anion–anion distances, high-pressure
aging parameters 314 ff, 321 ff, 325 ff transformations 663
Akaiwa–Voorhees model 373 anisotropic attachment kinetics, dendritic growth 115
Al–Ag–Zn system 398 anisotropic interactions, atomic ordering 558
Al–Ag system 488 anisotropy, phase transitions 268, 280
Al–Cu system annealed defects 468
– directional dendritic growth 150 annealed disorder 279
– precipitation 325 annealing, shape memory 643
Al–CuAl2 alloys 161 antiferromagnets 273, 297
Al–Fe system 150 antiphase boundaries 487
Al–Li alloys, coarsening 375 anvils 670 ff
Al–Mg–Si alloys 325 Arrhenius behavior, diffusion 191, 199 ff, 216 ff
Al–Ni–Co system 377 associative jumps, vacancies 197
Al–Zn–Ag system 329 athermal transformations 583–654
Al–Zn–Mg system atom–vacancy exchange 218
– phase diagram 45 atomic jumps 189
– precipitation 397 atomic ordering 519–581
Al–Zn system attachment kinetics, anisotropic 115
– dynamic fluctuations 277 Au–Cd alloys 590
– interfacial diffusion 510 Au–Cu system, atomic ordering 561, 564
– precipitation 397, 505, 510 Au–Ni system
– spinodal decomposition 451 ff – atomic ordering 563 f
alkali alloys 634 – phase diagram 21
alkali halides 639 Au-based alloys, martensite 599 f
Allen–Cahn concept 486 f austenite 588, 600, 617
allotropic transformations austenite stabilizing, martensitic transformations 625
– interfacial diffusion 499 autocatalytic growth, martensite 618 f

ISBN: 3-527-30256-5
698 Index
average jump frequencies 199 Born approximation, electrical resistivity 469

averaging, phase transitions 279 boron nitride, displacive transformation 639
Avrami equation 401, 465 boundary conditions
– dendritic growth 102 f, 108
– diffusion 176 ff
B–C–N system, high pressure 683 ff – interfacial diffusion 484 ff, 503
B–C system, high pressure 686 – LBM theory 423
B–N system, high pressure 685 – solidification 91 ff, 122
b.c.c. structures 537, 558 f, 577 boundary motion 495 f
bacterial flagellae, martensitic transformation 640 Bragg–Williams approximation 255, 287
Bain strain 594 f, 601 f, 620 Bragg peak
bainite 650 – ordering 463 ff
band–dendrites transitions 150 – phase transitions 250, 258, 293
banded structure, diffusionless transformations 611 f – solidification 87
Be–Cu alloys 677 Bridgman solidification 96 f, 668 ff
Becker–Döring theory Brillouin scattering 675
– decomposition 342 ff, 352 f Brown’s bond valence approach 664
– precipitation 320 ff, 395 Buerger deformations 611
belite, diffusionless transformations 639 bursts, martensitic transformations 633, 636
belt apparatus, high-pressure transformations 669
bending–heating–cooling cycles 644
beta alumina, diffusion 209 C–N system, high pressure 686
Bethe approximation Cahn–Hilliard–Cook approximation
– atomic ordering 524 358 f, 420 ff,438 ff, 448
– phase transitions 287 Cahn–Hilliard model
Bi–Sn–Cd system 40, 44 ff – decomposition 346 ff
bicritical points, phase transitions 275 – interfacial diffusion 486 ff
bicrystals see also: continuum model
– discontinuous precipitation 506 Cahn treatment
– interfacial diffusion 490 – interface migration 496 ff, 507
bifurcation – spinodal decomposition 437
– directional solidification 127 f calcite–argonite transition 259
– precipitation 337 calculation models, ternary from binary data 72 ff
bilinear coupling 280 see also: Monte Carlo
binary alloys CALPHAD method
– diffusion 200 – coarsening 373
– phase transitions 285, 296 – concomitant processes 394
– point variables 527 – phase separation 323 ff
– precipitation 314, 317 f, 397 CaO–Fe2O3 system, phase diagram 53
– substitutional, diffusion 226 CaO–MnO system, phase diagram 14
binary fluids, spinodal decomposition 455 CaO–SiO2 system, phase diagram 27
binary mixtures capillarity
– Ising model 443 – decomposition 340
– spinodal decomposition 414 ff – directional solidification 124
binary monotectic reaction 26 capillary-corrected diffusion 158
binary systems capillary forces, interface migration 492
– atomic ordering 523 ff capillary length, dendritic growth 108, 125, 131
– diffusion 208 f carbides
– eutectic 23 – atomic ordering 523
– liquid 246 – diffusionless transformations 650
– metallic 503 carbon
– phase diagrams 14 ff, 42, 314 – diffusionless transformations 639
Binder model see: cluster dynamics/model – high-pressure transformations 684
biological materials, martensitic transformations 640 carbon nitrides, high pressure 686
Bitter–Crum theorem 448 CASCADE computer code 219
bivariant phase regions 31 cast duplex steels, phase separation 366
block copolymer melts 259, 271 casting
body-centered cubic see: b.c.c. – eutectic growth 158
Boltzmann–Matano analysis 177 ff, 223 – solidification 96, 120
bond model, diffusion 200, 227 catastrophic nucleation 344
bonding energy, phase diagrams 12 catatectic invariants 32
Index 699
cation diffusion 231 – precipitation 315 ff

Cd–Al–Mn system 562 – spinodal decomposition 415 ff, 444
Cd–Mg system 563, 573 Co–Al system, ordering 570
Cd–Na system 36 f Co–Ni–O system 51
cell spacing, directional solidification 130 ff coagulation
cellular growth – clusters 352
– dendrites 140 ff, 146 – coarsening 377
– solidification 85, 126 f – spinodal decomposition 433, 446 f, 452
cements, displacive transformation 639 coarse graining
ceramic solutions, phase diagrams 64 – decomposition 347
ceramics – phase transitions 255, 263, 300
– diffusionless transformations 639 – spinodal decomposition 414 f, 418 ff, 438 ff
– spinodal decomposition 454 f coarsening
chain polymers, diffusionless transformations 639 – concomitant processes 381 ff
charge carrier distribution, high-pressure – decomposition 326, 365
transformations 666 – precipitation 315, 321 ff, 336, 370 ff
charge diffusion coefficient 212 – spinodal decomposition 416, 445
chemical composition gradient 178 coarsening stages, experimental identification 387
chemical diffusion 184 ff, 232 cobalt alloys, diffusionless transformations
– binary system 209 608, 617, 634 ff
– dendritic growth 108 coexistence curve 414
– directional solidification 124 coherency strains 377
chemical equilibrum 6 coherent elastic misfit 447
chemical forces coherent interfacial energies 390
– interface migration 493, 503 coherent miscibility gaps 571
– phase separation 322 coherent phases, precipitation 324
chemical interdiffusion 178 f coherent potential approximation (CPA), effective
chemical potential 10 cluster interactions 550
chemically induced grain boundary migration coherent solvus lines 316
(CIGM) 503 ff collective correlation factors, diffusion 206
Chemla experiment 181 collective diffusion coefficients 179
chill plate directional solidification 98 colloid crystallization 466
classification, displacive transformations 587 f, 634 columnar growth 95
Clausius–Clapeyron equation composition triangle, ternary 39
– diffusionless transformations 607 compositions
– directional solidification 126 – atomic ordering 544, 566
– martensitic transformations 629 – eutectic 23
cluster–cluster aggregation 469 – phase transitions 246
cluster–diffusion–coagulation mechanism see also: stochiometric compositions
– coarsening 377, 399 compounds
– spinodal decomposition 452, 456 f – phase diagrams 9, 65
cluster correlation functions 529 ff – stoichiometric 29
cluster dynamics approach compressibility 662
– concomitant processes 394 f computer simulation techniques, phase transitions
– nucleation 352 283, 293
– spinodal decomposition 432, 472 concentrated alloy systems 199, 226
cluster interactions, atomic ordering 548 concentration–depth profile 231
cluster kinetics 340 f concentration fluctuations 339
cluster probabilities concentration gradients 15, 175
– atomic ordering 529 f concentration profile
– phase transitions 289 – diffusion coefficients 235
cluster site approximation (CSA) 560 – interfacial diffusion 502
cluster variation method (CVM) concentration waves
– atomic ordering 551, 556 f, 573 – interfacial diffusion 486
– decomposition 345 – spinodal decomposition 415 f, 434
– phase transitions 288 f concomitant nucleation, spinodal decomposition 433
clusters concomitant processes, precipitation 381 ff, 401
– atomic ordering 524 ff, 529 ff configuration polyhedron, atomic ordering 531 ff
– concomitant processes 381, 391 configurations, atomic 523 ff
– decomposition 340 ff congruent compounds 29
– ionic crystals 230 conjugate phases 336
700 Index
conjugate thermodynamic variables 60, 247 Cu–Ag–Au alloys, diffusionless transformations

conode 15 607
conservation laws, phase transitions 277 Cu–Ag system, diffusionless transformations 564
consolute temperature 21 Cu–Al–Mn system, diffusionless transformations 562
constant composition sections, phase diagrams 46 Cu–Al–Ni system
constrained growth 95 – diffusionless transformations 602, 621 f
continuity equation, solidification 92 ff – shape memory effect 645
continuous casting 96 Cu-based alloys, diffusionless transformations 598 f
continuous cooling 501 Cu–Bi–Fe system, decomposition 397
continuum model Cu–Cd alloys, discontinuous precipitation 513
– decomposition 346 Cu–Co system
– diffusion 175 – decomposition 345, 350
– interfacial diffusion 486 f – discontinuous precipitation 505
– spinodal decomposition 356 f, 421 Cu–Mn system, phase separation 397
conventional transmission electron microscopy Cu–Ni–Fe system, decomposition 329, 334, 365
(CTEM), decomposition 326, 393 Cu–S–O system 7
cooling Cu–Sn system, martensitic transformations 617, 622
– diffusionless transformations 587 Cu–Ti system
– interfacial diffusion 501 – decomposition 317, 325, 331, 350, 388
– phase diagrams 15 – interfacial energies 390, 512
– precipitation 314 f, 401 Cu–Zn–Al system
– shape memory effect 644 – diffusionless transformations 598 ff
coordination number – shape memory effect 645
– atomic ordering 524, 539 Cu–Zn system, martensitic transformations 617
– high-pressure transformations 663 Curie–Weiss law, phase transitions 261
copper, alloying elements, diffusionless cycle procedures, shape memory effect 644
transformations 611, 634 cycling behavior, martensitic transformations 633
correlation factor, diffusion 178, 193 cyclohexanol, dendrites 117
correlation functions
– atomic ordering 526 ff, 567
– phase transitions 262 damping capacity, diffusionless transformations
– solidification 87 593, 636, 647
– spinodal decomposition 430 Darken–Manning relation 185 ff, 209
correlation length Datye–Langer theory 155, 157
– intrinsic 85 Dauphiné twins 601, 606
– phase transitions 249 decomposition
corresponding pairs, phase diagrams 60 – experimental techniques 326 ff
Cottrell solute drag 496 – precipitation 309–408
Coulomb potentials – spinodal 409–480
– colloidal suspensions 24 defects
– pair interactions 294 – diffusion 193, 201, 213
coumarin-succinonitrile, solidification 141 – eutectic growth 161
coupling mechanisms, defects–order parameter – interfacial diffusion 484 f
280 – ionic crystals 227, 231
Cowley–Warren short-range order parameters 531 – phase transitions 280 f
cristobalite, diffusionless transformation 639 – precipitation 401
critical amplitude, spinodal decomposition 438 – spinodal decomposition 415, 468
critical clusters 434 deformation, diffusionless transformations 592, 615
critical exponents 260 f degeneracy, solidification 92
critical nucleus 341 dendrites 97
critical phenomena, phase transitions 260 f dendritic growth 85 ff, 100 ff
critical radius, nucleation 88, 381, 392 density differences, phase transitions 246
critical temperature deviatoric components, diffusionless
– phase diagrams 21 transformations 588
– spinodal decomposition 416 DEVIL computer code 219
cross-boundary diffusion 497 f devil's staircase 274
crystal growth, solidification 91 Devonshire–Ginzburg–Landau theory 603 f
crystallization diamond, high-pressure transformations 683
– colloid 466 diamond anvil cells (DAC) 668 f, 671, 678
– primary 41 diblock copolymers, phase separation 272
crystallographic relations 593 f, 614 f, 620 dielectric solid systems, phase transitions 246
Index 701
differential scanning calorimetry, diffusionless drift terms, diffusion 175, 183

transformations 590 driving forces
diffraction techniques – interface migration 492
– decomposition 334 – martensitic transformations 624
– diffusionless transformations 587, 611 – phase separation 322
– high-pressure transformations 671, 678 ff droplets
diffuse interface model 347 f – decomposition 340
diffusion–coagulation mechanism 446 – phase transitions 300
diffusion – spinodal decomposition 415, 418, 442 f
– coarsening 377 Dufour effect 181
– dendritic growth 108 dumbell split configuration, diffusion 190
– directional solidification 124 duplex stainless steel, phase separation 364
– in crystals 171–238 dynamic pressure generation 667
– interfacial 481–518 dynamic scaling
– solidification 91 – phase transitions 277 f
diffusion constants – precipitation 395 ff
– definitions 175 ff see also: scaling
– effective, short circuits 187 dynamics, phase changes 414 f
– spinodal decomposition 422
diffusion-controlled growth
– precipitation 350, 370 f, 374, 381 E–pH diagrams 9
– solidification 94 early stage decomposition
– supersaturated matrix 350 – concomitant processes 381 f
diffusion couples 568 – kinetics 339 ff
diffusion kinetics Eckhaus band instability, solidification 128 f, 135 f
– Cahn–Hilliard 420 ff effective cluster interactions (ECIs) 548
– solids 171–238 effective diffusion coefficients 187
diffusion length Einstein equation 185, 192
– directional solidification 126 elastic constants
– eutectic growth 153 – diffusionless transformations 608
diffusion-limited aggregation (DLA) 101 – precipitation 336
diffusionless transformations 583–654 elastic interaction, decomposition 447 f
diffusivity 175, 485 f elastic phase transitions 257
dilatation dominant transformations 587 f, 610 elastic strain
dilute alloys, diffusion 221, 226 – diffusionless transformations 606 f, 636
dimensionality – interface migration 493
– critical 268 – precipitation 324
– order parameters 250 – spinodal decomposition 447 f
direct imaging techniques 326 f electric fields, phase transitions 246
directional dendritic growth 131 ff electrical resistivity
directional solidification 86, 95, 98, 120 ff – diffusionless transformations 587
discontinuous precipitation 484 f, 504 f – spinodal decomposition 468
discrete lattice point model 346 f electron back-scattered diffraction (EBSD),
dislocations solidification 144
– diffusion 186 electron energy loss spectroscopy (EELS), phase
– diffusionless transformations 620 separation 327
– interfacial diffusion 485, 495 electronic defects, ionic crystals 231
– spinodal decomposition 415 electronic properties, high-pressure
displacement temperature, shape memory effect 645 transformations 659
displacive transformations 292, 587–651 Ellingham diagrams 8 f
disruptive phase transitions 259 energy dispersive X-ray diffraction (EDX)
dissipation, spinodal decomposition 420 – decomposition 327
dissociative mechanism, diffusion 191, 197 – high-pressure transformations 680
distortions enstatite, diffusionless transformations 639
– high-pressure transformations 659 enthalpy
– phase transitions 283 – diffusion 217
divacancies, diffusion 219 – diffusionless transformations 636
domain structures – phase diagrams 5, 11, 34
– diffusionless transformations 600 – precipitation 318
– phase transitions 259, 274 entropy
doublons, solidification 114, 144 – diffusionless transformations 607, 636
702 Index
– directional dendritic growth 149 Fe–Cr–Ni system, martensitic transformations 617

– directional solidification 121 Fe–Cr–O system, phase diagram 52
– high-pressure transformations 660 Fe–Cr–V–C system, phase diagram 48
– phase diagrams 5, 11, 34 Fe–Cr system, phase separation 397
– phase transitions 247 Fe–Cu system, phase separation 350, 390
entropy–temperature diagram, shape memory effect Fe–Mn–Si system, shape memory effect 644
645 Fe–Mn system
equiaxed growth, solidification 95 – martensitic transformation 617
equilibrum cooling 15 – shape memory effect 650
equilibrum properties, solid–solid interfaces 484 ff Fe–Ni system, martensite 590, 617
Eshelby theory, displacive transformations 606 Fe–Ni–C alloys 619
Euler–Lagrange equation, second order transitions Fe–Ni–Cr system, precipitation 364
261 Fe–Pd alloys, diffusionless transformation 608, 613
eutectic growth 86, 99, 151 ff Fe–Pt alloys, diffusionless transformation 607
eutectic invariants 31 f, 41 Fe–Ru system, martensitic transformation 627
eutectic systems 22 Fe–S–O system, phase diagram 8
eutectic temperature 23 Fe–Ti–Al system, ordering 562, 569
eutectoid invariants 32 feldspar frameworks, high pressure 665
eutectoid reactions, interfacial diffusion 505 ferroelectric transitions 246
eutectoids 159 ferroic transformations 608
eutecular invariants 52 ferromagnetic systems, atomic ordering 570
evaporation–condensation mechanism, spinodal ferrous martensite 366
decomposition 456 ferrous systems, diffusionless transformation
evaporation, phase transitions 246 607, 635
excess properties fibers, solidification 152
– phase diagrams 13 ff fibrous systems, eutectic growth 158
– polynomial representation 34 Fick's laws of diffusion 175–235
exchange mechanism field ion microscopy (FIM), phase separation 326
– atomic ordering 539, 569 finite precipitate volume fractions 373
– diffusion 191, 217 f first neighbor interactions 532, 540, 555 f
– phase transitions 284 first-order phase transitions 249, 271
existence domains, atomic ordering 531 ff – martensites 599
extended X-ray absorption fine structure – solidification 86 ff
(EXAFS) Fisher renormalization 269, 279
– atomic ordering 563 five-frequency model, impurity diffusion
– high-pressure transformations 676 197, 207, 215
extension rules, phase diagrams 32 flat interfaces, crystal growth 91
extrinsic defects, diffusion 230 Flory–Huggins approximation 426 f
fluctuations
– dendritic growth 113
f.c.c–b.c.c. transformations 594 – heterophases 302
f.c.c. structures, atomic ordering 532 ff, 540 ff, 555 f – phase transitions 261, 271, 277
faceted systems, eutectic growth 158 – precipitation 319
faceting transitions 489 – solidification 87
FACT (Facility for the Analysis of Chemical – spinodal decomposition 414 f
Thermodynamics), expert system 75 fluid–suprafluid transition 246
fast growth rates, solidification 98 fluid magnet analogy, phase transformations 247
fast-mode theory, spinodal decomposition 428 fluids, spinodal decomposition 437 f, 455 f
faulting, diffusionless transformations 622 fluorescence line, high pressure 672
faults, eutectic growth 161 flux balance, eutectic growth 153
Fe–Al system force–velocity relationship, interface migration
– atomic ordering 562, 565 492, 496
– order parameters 250 formation energies, phase diagrams 29
– phase diagram 276 fractal structure 266
Fe–C–Ni system, martensitic transformations 617 framework flexion, high-pressure
Fe–C system, martensitic transformations 617 transformations 661
Fe–Co–Al system, martensitic transformations free growth
562, 569 f – dendritic 100 ff, 118
Fe–Cr–Co system – solidification 96
– coarsening 377 freezing 91 ff, 120
– decomposition 329, 332 Frenkel defects 213, 220, 228
Index 703
frictional forces, interface migration 496 – spinodal decomposition 415

frozen-in clusters 444 Gruzin method, diffusion 232
frozen-in disorder 468 Guggenheim model 68
frozen-in impurities 279 Guinier–Preston zones 325, 360
frozen phonons 259 Guinier approximation, small-angle scattering 391
Furukawa function 447, 453 gyration radius 426
fusion temperature, phase diagrams 15
habit plane 595 ff, 615 f, 622, 637

gallium arsenide, high pressure 681 halides, diffusionless transformations 639
gaseous compounds, phase diagrams 9 Hamiltonians, phase transitions 247 ff, 255 ff, 262 f,
gasketing system, high pressure 670 271, 281 ff, 419
Gaussian noise, phase separation 420 hardening
Ge–Si system, phase diagram 17 – diffusionless transformations 648
gel–gel transitions 456 – steel 641
germanium, high pressure 680 Hart equation 187
Gibbs–Duhem equation 11 ff, 30, 61 Haven ratio 183 ff, 190, 196, 211 f
Gibbs–Thomson equation heat flow, solidification 181
– decomposition 351, 370 f, 382 ff, 394 heat sinks, solidification 120
– eutectic growth 154 heating–cooling cycles, shape memory effect 644
– interfacial diffusion 492, 510 heating, diffusionless transformations 590, 607
Gibbs absorption isotherm 490 heating rate, decomposition 330
Gibbs energy 5 ff Heisenberg antiferromagnet 297
– minimization 73, 6 f Heisenberg model 267
Gibbs phase rule 7 f Hele–Shaw flow 106
Ginzburg–Landau equation helimagnetitc structures 273
– diffusionless transformations 603 Helmholtz energy
– interfacial diffusion 487 – decomposition 347
– solidification 89 – minimization 560
Ginzburg criterion – precipitation 318 ff, 337 ff
– phase transitions 263 ff, 300 ff – spinodal decomposition 417
– spinodal decomposition 423, 431, 462, 471 – vacancies 218
girdle-type press 669 Henry's law 21, 29 ff
glasses, phase separation 361, 367, 454 f Hess' law 660
glissile interfaces, martensitic transformations 622 heterodiffusion 180, 235
gold alloys, displacive transformations 635 heterogeneous nucleation, precipitation 315, 319,
Gorsky effect, diffusion 233 468
grain boundaries heterogeneous regions, interfacial diffusion 484
– diffusion 186 heterophase fluctuations 414 f, 431 f
– migration 503 ff heterophases 302
– phase transitions 282 high-angle grain boundaries 485
– precipitation 317, 325 high damping materials (HIDAMETS) 647
– spinodal decomposition 415 high-pressure generation 666 ff
grain refinement 119 high-pressure phase transformations 655–695
graphical integration analysis, diffusion 233 high-resolution microanalysis 511
graphite, transformations 6, 683 high-resolution transmission electron microscopy
gravity effects, spinodal decomposition 456 (HRTEM) 326, 620
Greninger–Troiano relations 616, 637 f Hillert–Sundman treatment 496 ff
ground states, atomic ordering 538 ff hole hopping, diffusion 205
group formation, diffusionless transformations 619 homogeneity range, phase diagrams 28
group theoretical techniques 255 homogeneous lattice distortive strain 590
growth homogeneous second phase precipitation 309–407
– bainite 650 homophase fluctuations, spinodal decomposition
– decomposition kinetics 332 414 f, 431 f
– dendritic 116, 148 Hopf solution 92
– diffusion-controlled 350 Hugoniot elastic limit (HEL) 667
– diffusionless transformations 636 Hume–Rotherey alloys 619 f, 634
– interface migration 495 hydraulic presses 668 ff
– martensites 599, 632 hydrodynamics, spinodal decomposition 446
– precipitation 319 ff, 370 hydrogen bonds, high-pressure transformations 687 f
– solidification 93 ff hydrostatic pressure 649
704 Index
hyperscaling relation, phase transitions 267 interstitial alloys

hysteresis behavior, martensitic transformations – atomic ordering 523
633 f – diffusion 224
interstitial ferrous martensites 636
interstitial jumps 213
ice amorphization 662, 687 ff interstitial mechanism, diffusion 189 f, 216
ideal solutions interstitial solid solutions, diffusion 205
– Henrian 29 interstitial solutions, phase diagrams 64
– Raoultian 12 f interstitials, metal diffusion 231
immiscibility 26 interstitial-substitutional mechanism, diffusion 191
impurities intrinsic correlation length, solidification 85
– diffusion 176 ff, 213 intrinsic defects, diffusion 227
– directional solidification 120 intrinsic diffusion 179 f, 184 f
– interfacial diffusion 500 invariants, binary phase diagrams 31
– ionic crystals 231 invariant plane strain (IPS) 595, 613 ff
– metals diffusion 220 invariant reactions, nomenclature 49
– phase transitions 279 inverse cluster variation method 550
– spinodal decomposition 415, 468 f inverse coarsening 380
impurity–vacancy interaction 198, 221 inverse Monte Carlo technique 550
impurity correlation factors 196, 208 ionic conductivity, diffusion 183, 189, 204 f, 208 ff
incoherent miscibility gaps 570 ionic crystals, diffusion 227
incommensurate superstructures 273 ionic solutions, phase diagrams 62
incubation time, precipitation 319, 343 ionic valence, diffusion 181
indirect interstitial mechanism, diffusion 190 iron
indium-based alloys, diffusionless transformations – alloying elements 611
607, 634 – massive transformation 500
inertial hydrodynamics, spinodal decomposition 446 see also: Fe
infrared spectroscopy, high pressure 674 iron aluminides, ordering 565
inoculants, solidification 95 iron-based alloys, martensitic transformations 635
inorganic compounds, diffusionless transformations iron whiskers, martensitic transformations 601
639 irregular structures, eutectic growth 161
InSb, high-pressure transformations 682 irregular tetrahedron approximation, atomic
instabilities ordering 553
– dendritic growth 106, 136 ff irreversible thermodynamics, diffusion 181 ff
– eutectic growth 157 Ising model
– solidification 93 – antiferromagnet 200
integral molar enthalpy 12 – atomic ordering 524
interatomic interactions 659 – coarsening 380
intercalation compounds 205 – decomposition 367 ff
intercluster spacing 336 – ferromagnets 303
interconnected microstructures 377 – phase transitions 264, 275, 285, 293
interconnected structures 364 – spinodal decomposition 419 ff, 435 ff, 462
interdiffusion isobaric temperature–composition diagrams 51
– chemical 178 ff isopleths, ternary 46
– metals 227 isopropylacrylamide gel 456
– spinodal decomposition 424, 472 isothermal precipitation 315 f
interdiffusion coefficients 233, 357 isothermal sections, ternary systems 43, 569 ff
interfaces isotope effect, diffusion 196, 213
– decomposition 340, 347 isotropic pair probabilities, atomic ordering 527
– diffusionless transformations 622 Ivantsov–marginal stability, solidification 118
– solidification 85 ff, 89 f, 99 Ivantsov solution 102 ff, 111, 131
interfacial diffusion 481–518
interfacial energies 389 ff
interlamellar spacing 511 Jänecke coordinates 51
intermediate phases 26 Johnson–Mehl equation 465
– silicon 681 jump dicontinuity, interface 89
intermetallic phases 26 jump frequency
– ordering 523 – decomposition 343
– precipitation 314 f – diffusion 194, 216 f
internal friction, diffusionless transformations 587 – metal diffusion 226
interphase boundaries 484 ff jump singularities 247, 262
Index 705
Kampmann–Wagner approach lattice wave modulations 592, 597

– precipitation 382 ff ledges, structural 488
– spinodal decomposition 433 Legendre polynomials, thermodynamics 34
Kawasaki model 367 Legendre transformation 538, 660
kickout mechanism, diffusion 191 length scales, solidification 85
Kikuchi natural iteration 525, 554 Lennard–Jones interaction 294
kinetic coefficients lens-shaped two phase regions 20
– dendritic growth 115, 134 lenticular martensite 618
– eutectic growth 155 lever rule
– solidification 87 – binary systems 15
kinetics – phase diagrams 41, 248
– diffusionless transformations 588, 603 – spinodal decomposition 417
– martensitic transformations 624 f, 632 f LiF–NaF system, phase diagram 35, 67
– precipitation 314 f Lifshitz–Slyozov–Wagner model 370 ff, 382 f, 452,
– spinodal decomposition 416 f 456, 463, 471
Kirkendall effect 179 f, 233 Lifshitz invariants 273 ff
Kittler–Falikov theory 288 Lifshitz point 275
Koch–Cohen cluster 230 light scattering, precipitation 328, 361
Kosterlitz–Thoules phase transitions 268 limiting slopes, phase boundaries 66
Kröger–Vink defect notation 228 f linearized theory, spinodal decomposition 421 f
Kurdjumov–Sachs relations 616, 620 f, 637 f Liouville equation 440
liquid–gas transitions 246 ff
liquid–liquid miscibility 26
La–Ni–Al system, phase separation 454 liquid–solid phase transitions 81–170, 246
lamellar modulations 361 ff liquid compounds, phase diagrams 9
lamellar systems 152, 158 liquidus 39 f, 74, 279
Landau coefficient 127 lithium alloys, diffusionless transformations
Landau symmetry classification 246 f 617, 635
Landau theory local relaxation, atomic ordering 550
– diffusionless transformations 603 localized nucleation models, martensitic
– omega transformations 609 transformations 629
– phase transitions 250 ff, 262 f, 269, 290 long-range order (lro) configurations
– spinodal decomposition 419, 438, 463 – atomic ordering 529 ff, 539, 546, 566
Langer–Baron–Miller approximation 359, 423 ff, – phase transitions 71, 288
440 ff, 471 low-angle tilt boundaries 485
Langer–Schwartz model lutidine–water system, decomposition 455
– concomitant processes 382 ff
– spinodal decomposition 433, 445
Langer theory 321 ff macroscopic diffusion 175 ff
large-amplitude fluctuations 418 magnetic exchange energy, atomic ordering 570
large-volume presses 668 magnetic ordering, spinodal decomposition 460
laser trace, dendritic growth 101 magnetic properties, high-pressure
laser treatment, solidification 96 transformations 659
lasers, high-pressure transformations 667 magnetic solid systems, phase transitions 246
laths, diffusionless transformations 617 magnetization, precipitates 339
lattice anisotropy, decomposition 420 magnetostrictive couplings 271
lattice deformations, diffusionless transformations magnon modes 277
620, 639 manganese-based alloys, diffusionless transformations
lattice diffusion length 187 608, 612, 635
lattice distortive displacements 588 Manning model 185 f, 199, 209, 215 f, 226
lattice-gas model marginal stability hypothesis, dendritic growth
– diffusion 201 110, 118
– spinodal decomposition 428, 444 f Marqusee–Ross model 373
lattice relaxation, ordering 523, 573 Marsh–Glicksman model 373
lattice softening, diffusionless transformations 597 martensitic transformations 260, 587–651
lattice structures mass action law, ionic crystals 230
– atomic ordering 529 ff, 558 mass transfer, diffusion 180
– diffusion 194, 223 massive transformations, interfacial diffusion 498 ff
– order parameters 251 f mean field approximation (MFA)
– phase transitions 271 – decomposition 359
see also: structures – phase transitions 255, 259 f, 287 ff
706 Index
– precipitation 318 ff – decomposition 333, 339 ff, 390, 416 f, 460

– spinodal decomposition 417 ff, 431 f – phase diagrams 14
mechanical forces, interface migration 493 – precipitation 314 ff, 401
mechanical properties, diffusionless misfit strain, precipitates 447
transformations 590 mixed alkali effect 209
melting mixing energies, phase diagrams 9 ff
– congruent 29 mobilities
– phase transitions 283 – diffusion 183, 205, 211
melting entropy 149 – interface migration 485 f, 496, 508
melting point – solidification 89
– atomic ordering 571 – spinodal decomposition 421
– phase diagrams 8, 18 modified Langer–Schwartz model 382 ff
melting temperature modulated phases 273
– discontinuous precipitation 505 modulated structures, decomposition 360, 364
– solidification 91 ff molar enthalpy, phase diagrams 12
mesophases, block-copolymer melts 259, 271 molecular crystal systems 246
metallic glasses, decomposition 455 molecular dynamics
metallic substitutional alloys 523 ff – interfacial diffusion 484
metallic systems – phase transitions 293
– decomposition 360, 366 molecular field approximation 288
– diffusionless transformations 634 molecular field theory 288 ff
– interfacial diffusion 491, 503 monoclinic transformations 600
– phase separation 322 f monodisperse polymers, spinodal decomposition 427
– spinodal decomposition 451 f monolayers, absorbed, order-disorder 288
– substitutional ordering 523 ff monotectic invariants, phase diagrams 26, 32
metals Monte Carlo simulations
– dendritic growth 119 – atomic ordering 551, 556 f
– diffusion 220 ff – decomposition 367
metastability – diffusion 200, 215
– interfacial diffusion 486 – interfacial diffusion 484 ff
– phase transitions 298 ff – phase transitions 287
– precipitation 316 – precipitation 321, 396 f
– spinodal decomposition 417 – spinodal decomposition 428 ff, 437 f
metastable phase boundaries 39 morphological instabilities, directional dendritic
methanol–cyclohexane mixture, decomposition 457 growth 139 f
MgO–CaO system, phase diagram 22 morphology
Mg–Zn system 19, 45 – decomposition 360 ff
microanalytical techniques 326 ff, 511 – diffusionless transformations 588, 616, 636
microclusters 340 – interfacial diffusion 488
microscopic diffusion 189 ff – precipitation 334 ff
microscopy, high-pressure transformations 674 – spinodal decomposition 415, 470
microsegregation 131 Mössbauer spectroscopy (MBS)
microstructures – diffusion coefficients 234
– coarsening 377 – diffusionless transformations 587
– diffusionless transformations 618 ff – high-pressure transformations 675
– polydispersed 382 mottled microstructure 364
– precipitation 314 f moving boundaries 503
midribs, diffusionless transformations 618 f Mullins–Sekerka instability
migration – dendritic growth 106, 136 ff
– atomic 175 – solidification 93, 120, 126 ff
– coarsening 379 Mullins equation 491 f, 510
– diffusion 186 mullite, formation 5
– interfacial diffusion 484 f, 503 ff, 514 multicomponent phase diagrams 14, 39 ff, 47
– ionic crystals 228 multicomponent systems, atomic ordering 523
minerals, diffusionless transformations 639 multicritical phenomena, phase transitions 269
minimization algorithm multilayers 514
– atomic ordering 525, 554 multiple interface transformations, martensitic 633
– Gibbs energy 73 multisoliton lattice 274
miscibility gap mutual solubility, limited 29 f
– atomic ordering 558, 564, 572 f
– binary systems 21
Index 707
Na–K–F–Cl salt system, phase diagram 49 optimization techniques, phase diagrams 34 f

natrolite frameworks, high pressure 662 orbital overlaps, high pressure 666
nearest neighbors model order–disorder transitions
– atomic ordering 532, 558 – atomic ordering 567
– decomposition 367 – high-pressure transformations 664
– high-pressure transformations 663 – kinetics 462
– phase diagrams 63, 268, 288, 295 – phase diagrams 71
– spinodal decomposition 444 f, 450 – solidification 87
needle crystal solution, solidification 101, 113 – spinodal decomposition 415 f, 460
neighbor interactions 531 f – symmetry breaking 247 f
nematic liquid systems, phase transitions 246 order parameters
neodynum pentaphosphate, displacive – interfacial diffusion 486 f
transformations 600 – phase transitions 71, 246 f
Nernst–Einstein equation 183 f, 211 – solidification 87
neutral stability, directional solidification 126 ordering
neutron diffraction, phase transformations 587, 678 – atomic 519–581
Newton–Raphson technique 554 – pressure-induced 664
Ni–Al–Mo system, phase separation 381, 450 – short/long range 68 ff, 529 ff
Ni–Al system – spinodal 415
– decomposition 346, 354 f organics, diffusionless transformations 639
– interfacial energies 390 orientation relationship, diffusionless
Ni–Cu–Al system, precipitation 315 f, 332, 390 transformations 616
Ni–Si system, precipitation 397 orientational glasses 281 f
Ni–Ti system, displacive transformations 645, 651 oriented clusters, atomic ordering 524
nickel-based alloys, displacive transformations 635 Ornstein–Zernike type susceptibility 264
Nishiyama–Wassermann relations 616, 620 f, 637 f orthorhombic transformations 600
nitinol, shape memory effect 642 f oscillations, eutectic growth 162
nitrates, displacive transformations 639 Oseen tensor 440
nitrides, ordering 523 Ostwald ripening 370
noble metal-based alloys see also: coarsening
– atomic ordering 564 outgassing 224
– diffusionless transformations 599 f oxidation resistance 565
noble metals, diffusion 221 oxides
nonfaceted systems, eutectic growth 158 – atomic ordering 523
nonferrous alloys, diffusionless transformations 637 – decomposition 361, 367
nonisothermal precipitation 401 oxygen partial pressure, ionic crystals 230
nonlinear theories, spinodal decomposition 437 f oxygen self-diffusion 178
nonmetals, diffusionless transformations 638 f
nonstoichiometric compounds, phase diagrams 65
nuclear magnetic resonance (NMR) pair interactions, atomic ordering 527 f, 538
– diffusion coefficients 234 pair variables 527 ff
– high-pressure transformations 676 parabolic directional dentrites 134
nuclear methods, diffusion 231, 234 parabolic growth, precipitates 351
nucleation parabolic Ivantsov solution 104 f, 113
– concomitant processes 381 ff paraboloid symmetry, dendritic growth 115
– decomposition kinetics 332, 340 ff paramagnetic states, ordered phases 565
– eutectic growth 155 Paris–Edinburgh cells 670, 676
– interface migration 495 partial diffusion coefficients 179 f
– martensitic transformations 598, 630 partial properties, phase diagrams 11 f
– phase transitions 300 particle (precipitate) coherency 392
– precipitation 315 ff particle (precipitate) pinning 498
– solidification 87 f particle (precipitate) splitting 377
– spinodal decomposition 415, 468 path probability method (PPM) 200, 209
numerical modeling, precipitation 381 ff, 399 f Pb–Sn bicrystals, boundary displacement 506
Péclet number 103 ff, 110, 132
percolation
octahedron cluster 535, 573 – diffusion 209
Olsen–Cohen model 629 – phase transitions 281
omega transformation 609 – spinodal decomposition 416, 433, 442 ff
Onsager coefficients 428 peritectic invariants, phase diagrams 26 f, 31 f
Onsager theorem 181 f, 184, 206 peritectics 163
708 Index
peritectoid precipitation 484 polytetrafluoroethylene (PTFE), diffusionless

perovskites, high pressure 663 transformations 640 f
pertubations polythermal projections, liquidus surfaces 41
– dendritic growth 111, 139 polytypic transitions, shear transformations 613
– directional solidification 127 f Porod law 447, 460
phase boundaries, phase diagrams 66 positron annihilation diffusionless
phase diagrams 1–80 transformations 587
– CVM/MC methods 554 potassium, diffusion 232
phase equilibra, atomic ordering 551 ff potential–composition diagrams 5 ff
phase field models potential axes, phase diagrams 51
– directional dendritic growth 151 f Potts model 255, 259 f, 270 f, 285, 293
– interfacial diffusion 487 Pourbaix diagrams 9 f
phase rule, Gibbs 7 f power law approximation, precipitation 395 ff
phase separation precipitate-growth stages, experimental identification
– precipitation 314 f, 320 f 387
– spinodal decomposition 416 precipitation
phase transformations – discontinuous 504 f
– high pressure 655–695 – homogeneous 309–407
– solidification 85 – interfacial diffusion 484 ff
phase transitions, statistical theories 239–308 – spinodal decomposition 442
phase/microstructure selection, eutectic growth 164 precursor effects
phenomenological theories – diffusionless transformations 598 f
– diffusionless transformations 620 – phase transitions 282
– phase transitions 246 ff predominance diagrams 5 ff
– precipitation 398 preexponential factor, diffusion 220
– spinodal decomposition 416 f preloading, high-pressure transformations 669
– interfacial diffusion 487 premartensitic phase 639
phonons pressure–composition phase diagrams 19
– diffusionless transformations 597 f pressure, phase diagrams 6
– phase transitions 258, 277, 282, 291 pressure–temperature diagram, water 54
physical correlation factor, diffusion 184, 204 pressure-driven phase transitions 246, 655–695
pinning, interfaces 498 pretransformation states, diffusionless 595 f
piston–cylinder systems 668 primary cell spacing, directional dendritic growth
pivalic acid (PVA), dendrites 115 135 f
plastic crystals primary cristallization, phase diagrams 41
– dendritic growth 116 primary order parameters, phase transitions 252
– phase transitions 258 pseudo time averages, phase transitions 294
– solidification 96 pseudobinary alloys 332
plastic deformation, diffusionless transformations pseudocomponents, phase diagrams 64
606 pulse transmission/echo method, high-pressure phase
plastic response, interface migration 494 transformations 677
plate-shaped martensitic regions 617 pulsed field gradient NMR, diffusion 234
point clusters, atomic ordering 524 Purdy–Lange model 506
point defects purity, dendritic growth 116
– diffusion 193 PVA–ethanol, solidification 115
– phase transitions 280 pyroxene chain silicates, diffusionless transformation
– spinodal decomposition 415 639
point variables, atomic ordering 525 f
polarization, phase transitions 246, 258
polydispersity, polymers 427 quadrupolar glasses, ordering 282
polygons, atomic ordering 534, 541 quadrupole moments 252, 258
polyhedral joints, high-pressure transformations 661 quantum liquid systems, order parameters 246
polyhedron, atomic ordering 531 ff quartz, high-pressure phase transformations 661
polymer alloys, decomposition 416 quasibinary phase diagrams 47
polymer mixtures, decomposition 361, 426 f, 455 ff quasichemical model 68
polymer solutions, phase diagrams 66 quasieleastic neutron scattering (QNS), diffusion
polymers, diffusionless transformations 639 f 234
polymorphic states 611, 640 quasiharmonic approximation 292
polynominal representation, excess properties 34 quasimartensitic transformations 588, 611 ff
polystyrene–poly(vinyl methyl ether) system, quasistationary approximation
decomposition 328, 455 ff – directional dendritic growth 144
Index 709
– eutectic growth 157 rotational symmetry, dendritic growth 115

– solidification 92, 125 roughening transition 268
quenched disorder 279 f Rouse model 428
quenched impurities 468 f
quenched-in nuclei 95
quenching Saffmann–Taylor limit 130
– decomposition 330 f, 339 salt system, reciprocal ternary 49 f
– diffusionless transformations 587 ff, 643 salts, transparent, solidification 115
– precipitation 314 ff scaling
– predecomposition 391 – dendritic growth 110 f, 114, 135
– spinodal decomposition 414 ff, 427 – directional solidification 124
– interfacial diffusion 502
– solidification 94 f
R-phase 639 – spinodal decomposition 425, 430, 445 f, 453 f
radioisotopes, diffusion coefficients 232 scaling laws
Raman spectroscopy, high pressure 674 – eutectic growth 156
random alloy model 185 ff, 199, 226 – phase transitions 260 f, 266, 277
random bond, phase transitions 280 – precipitation 395 ff
random phase approximation 427 scanning transmission electron microscope (STEM),
random walk theory 189 microanalysis 511
Rankine–Hugoniot relations 667 scattering data, phase transitions 250, 260, 287
Raoultian solutions, ideal 12 scattering techniques, precipitates 328 f, 423
rapid diffusion 191 Schottky defects 227
rapid solidification 499, 649 Schreinemaker rule 46, 57
rare-earth alloys, diffusionless transformations second-neighbor interactions 535, 543 ff, 557 f
608, 635 second-order transitions 249, 252, 258, 260 f
rate-determining mechanism, directional second-phase precipitation, homogeneous 309–407
solidification 120 secondary ion mass spectroscopy (SIMS), diffusion
Rayleigh–Benard systems 127 232
Rayleigh number 124 segregation, interfacial 484, 489 f
reaction stoichiometry 5 ff segregation coefficients 125
real-system applications, atomic ordering 561 self-accomodation, diffusionless transformations
reciprocal ternary phase diagrams 49 619, 636
reconstructive phase transitions 260 self-averaging, phase transitions 280
recrystallization 495, 505 self-diffusion 176 ff
Redlich–Kister form, phase diagrams 34 – grain boundaries 491
reduced geometry, phase separation 467 – isotope effect 215
regular solution theory 23 – metals 202
regular structures, eutectic growth 158 self-similarity, precipitation 395 ff
relative partial properties, phase diagrams 11 f semiconductors
relaxation – atomic ordering 523
– atomic ordering 523, 550, 573 – dendritic growth 119
– diffusionless transformations 635 – zincblende-type, high pressure 679
– interface migration 495 shallow quenching 427
– spinodal decomposition 424, 467 shape deformations, diffusionless
relaxation methods transformations 600 f, 615
– diffusion coefficients 233 shape-memory alloys (SMA) 590 ff, 633 ff, 638,
– interfacial diffusion 484 641 f
– metal diffusion 224 sharp interface model 340
renormalization group methods shear transformations 606, 613, 636
– phase transitions 266 ff, 271, 279 shock wave techniques, high-pressure transformations
– spinodal decomposition 419 f, 429 f 667
reptation, spinodal decomposition 428 Shoji–Buerger lattice deformations 594
residual acitivity, diffusion coefficients 232 short-circuit diffusion 186 ff
resistivity, electrical, diffusionless transformations short-range order (sro) configurations
587 – atomic ordering 529 ff, 539
retarding forces, interface migration 492 – phase transitions 68, 288
Reynolds number 124 short-wavelength distortions, phase transitions 283
Rigsbee–Aaronson model 488 shuffles, diffusionless transformations
Rikvold–Gunton model 398 588 ff, 596 ff, 606 ff
rotational jumps 197 Si–O–Si angles 661
710 Index
SiC–AlN ceramics, decomposition 454 space model, ternary 39

side branching, dendrites 108 ff, 143, 151 spacing
silica, phase transformations 639, 661 – dendritic growth 111, 117
silicate slags 69 – directional solidification 130 ff
silicon, high-pressure phase transformations 680 – eutectic growth 159
silicon-based alloys, rapid solidification 499 – interfacial diffusion 485, 509
silver alloys, diffusionless transformations 634 – solidification 99
simplex, atomic ordering 534 – spinodal decomposition 426, 450
simulation techniques, phase transitions 293 see also: lattice spacing, interface spacing,
single chain dynamics 428 cell spacing
single-phase solid solutions 15 spectroscopic techniques, high-temperature
single sites, phase transitions 289 transformations 674
singularities, critical 260 sphere crystal growth 92
SiO2–MgO–MnO system, phase diagram 74 spin glass 281
site cluster approximation 560 spin representation, Ising 286
site occupation operator 526 spin variables 524, 570
slip, diffusionless transformations 590, 605, 622 spinodal alloys 363 ff
slow growth rates, solidification 96 spinodal decomposition 332, 354 ff, 409–480
slow mode theory 428 – precipitation 319 ff
slowing down, phase transitions 277 spinodal point, phase transitions 22, 299
small-amplitude fluctuations, spinodal decomposition spins, phase transitions 255, 259 f
418 split sphere anvil technology 670
small-angle neutron scattering (SANS), precipitates splitting
328, 391 – clusters 352 f
small-angle X-ray scattering (SAXS), precipitates – coarsening 377 f
328, 360 – spinodal decomposition 433
Snoek effect, diffusion 233 sponge-like microstructure 364
snow flakes, dendritic growth 100 spontaneous growth, ordered domains 462
soft phonons 277, 282, 291 spontaneous magnetization 246
softmode phase transitions stability
– diffusionless 596 – dendritic growth 118
– martensitic transformations 631 – nucleation 350
solid compounds, phase diagrams 9 – spinodal decomposition 417
solid binary mixture systems 246 stability domains, phase diagrams 8
solid–liquid equilibra 18 stability length
solid–liquid interface, directional solidification 120 – dendritic growth 106, 111
solid–liquid miscibility, phase diagrams 14 – solidification 93
solid–solid transformations 484, 490 stability limit, phase transitions 254
solid solutions stabilizing, diffusionless transformations 611, 625
– decomposition 314 f, 385 stable pair, exchange reaction 63
– diffusionless transformations 635 stacking faults, diffusionless transformations 601, 634
– phase diagrams 15 stacking sequences, diffusionless transformations
solid-state diffusion 189 595, 613
solidification 81–170 staggered fields, ordering 248
solidus line 279 standard model, directional solidification 125
soliton staircase 274 standard molar Gibbs energy 6
solubility standard states, phase diagrams 11 f
– mutual 29 static pressure devices 668
– precipitation 316, 370, 393 statistical theories, phase transitions 239–308
solutal dendrites 117 steady cooperative growth 507
solute–solute binding 225 steady-state nucleation rate 342
solute–vacancy binding 222 steels, precipitation 366
solute drag, interfacial motion 496 steepest descent technique, atomic ordering 554
solute enhancements factors 221 stiffness tensor 447
solution models, phase diagrams 62 ff Stirling approximation
solvability theory, dendritic growth 106, 116 – atomic ordering 552
solvent enhancement factors, diffusion 221 – diffusion 219
solvus lines 23, 314 f – phase diagrams 13
Soret effect 181 stoichiometric compositions 566
sound velocity, high-temperature transformations 677 stoichiometric crystals 227
space groups, atomic ordering 526 f575 stoichiometric product phase 515
Index 711
stoichiometry, phase diagrams 5 ff, 28 – diffusionless transformations 635

Stokes' law, cluster diffusion 456 – interfacial diffusion 499
strain – precipitation 314 f, 323 ff, 396 ff
– diffusionless transformations 588, 606, 647 f – spinodal decomposition 414 f
– interface migration 493 superstructures
– precipitation 324, 337 – atomic ordering 529, 546
strain spinodals 631 – b.c.c. lattice 577
strengthening, martensite 649 – f.c.c. lattice 575
stress–temperature diagramm 645 – phase transitions 271 ff
stress induced transformations surface diffusion 180 f, 192
– martensitic 633 – coefficient measurements 234
– diffusionless 647 surface effects
stress relaxation, diffusionless transformations 635 – disorder 283
strontium titanate, displacive transformations – phase transitions 279 f
591, 607 – spinodal decomposition 467 f
structural distortions, high-pressure transformations surface melting 283
659, 665 f surface tension 88, 124 ff
structural models 5 surfactant micellar solutions 456
structure functions, spinodal decomposition 442 symmetry, dendritic growth 115
structure modulation 360, 364 symmetry breaking, order parameters 247
structures symmetry lowering, ferroic phase transitions 608
– diffusionless transformations 593 f symmetry properties, phase transitions 256, 270
– eutectic growth 158 f synchroton radiation, high-pressure studies 667, 670
– grain boundaries 484 f syntectic invariants, phase diagrams 32
– ordered 523 f
– precipitation 314 f
styrene–butadiene mixture, phase separation 457 tail instability, solidification 137
subcluster probabilities, atomic ordering 529 tangent constructions 11 f, 16
subgrains, stored energy 495 temperature–composition phase diagrams 5 ff, 14
sublattice magnetization 246 temperature–concentration plane, spinodal
sublattices decomposition 435
– atomic ordering 529 temperature–stress diagram, diffusionless
– interfacial diffusion 484 transformations 647
– phase transitions 62, 251 temperature gradients
– spinodal decomposition 415 – atomic ordering 551 ff
substitutional alloys – diffusion 181
– atomic ordering 523 ff – spinodal decomposition 470
– diffusion 221, 226 temperature ranges
substitutional ferrous martensite 636 – decomposition 340
substitutional impurities, spinodal decomposition – diffusionless transformations 587, 590 f, 607
415 – interfacial diffusion 501
substitutional solution model, ideal 12 ternary alloys
substructures, diffusionless transformations 616, 637 – pair variables 528
succinonitrile (SCN) – phase transitions 285
– dendritic growth 109, 114 f, 139 f – precipitation 334, 397
– solidification 96 ff ternary mixtures, spinodal decomposition 415
sulfides, diffusionless transformations 639 ternary systems 5, 39 ff
superconductivity, pressure dependence 666 – atomic ordering 569 f
supercooling tetrahedral framework, high-pressure
– dendritic growth 109, 131, 135 transformations 661
– eutectic growth 154 tetrahedron, atomic ordering 573
– solidification 88 ff, 92 tetrahedron approximation, phase transitions 288
supercritical clusters 434 tetrahedron clusters, atomic ordering 536 f, 544 f
superelastic alloys 650 tetrahedron–octahedron approximation 553 ff, 560
superhard materials, high-pressure transformations thallium alloys, diffusionless transformations 634
659 thermal dendrites 116
superheated melts 95 thermal diffusion
supersaturated matrix, precipitation 350 – dendritic growth 108
supersaturation – solidification 91
– coarsening 377 thermal fluctuations
– decomposition 330, 345, 359, 414 f – dendritic growth 113
712 Index
– solidification 87 – phase transitions 269, 272

– spinodal decomposition 414 f – spinodal decomposition 460
thermal gradients, directional solidification 126 triple points, phase diagrams 8, 55
thermally induced martensitic transformations 633 true phase diagram sections 56 f
thermodynamic compilations, phase diagrams 77 trunk spacing, dendrites 144 ff
thermodynamic factor twinning 595, 600, 605, 615, 618, 622, 634
– atomic ordering 565 twist boundaries 491, 495
– diffusion 205 two-component systems, directional solidification 121
thermodynamic variables, conjugate 247 two-phase stability 9
thermodynamics 1–80 two-phase regions 20
– diffusionless transformations 603 ff two sided model, dendritic growth 102
– irreversible 181 f two-way shape memory effect, diffusionless
– martensitic transformations 624 f transformations 644
– phase separation 322
– precipitation 317 f
– spinodal decomposition 416 f ultrasonic sound velocity, high-pressure
– two-component systems 121 transformations 677
thermoelastic behavior, martensitic undercooling
transformations 622, 632 f – dendritic growth 118, 146
thin layer methods, diffusion coefficients 231 – diffusionless transformations 612
three phase stability 9 – martensitic transformations 625
Ti–Al–X systems, atomic ordering 562 – precipitation 325
Ti–Ni–Cu alloys, martensite 631 – solidification 95 ff
tie lines 15, 41, 60 undistorted lines, diffusionless transformations 592
tilt grain boundaries 485 ff, 495 f univariant lines 41
tilted domains 156 univariant phase regions 31
tilted lamellar arrays 151 universality principle
time-dependent nucleation rate 343 – phase transitions 266, 293
time residence algorithms, Monte Carlo simulations – spinodal decomposition 431 f
485 unstable states, spinodal decomposition 417
tip radius uphill diffusion 357, 422
– dendritic growth 110 ff, 116 f, 120
– directional dendritic growth 133, 143
titanium alloys, diffusionless transformations vacancies
607, 617, 623, 634 ff – interfacial diffusion 485
tomographic atom probe (TAP), precipitates 328 – spinodal decomposition 468 f
topology vacancy–impurity complexes, diffusion 223
– phase diagrams 45 vacancy concentration
– phase transitions 25, 296 – diffusion 218
torques – ionic crystals 230
– discontinuous precipitation 506 vacancy mechanism, diffusion 191
– interfacial diffusion 499 vacancy-wind effect 182 f, 185 f, 208 ff
trace impurities, interfacial diffusion 500 van der Waals forces 24, 255
tracer correlation factor vapor–liquid equilibrium 18
– defective systems 204 variance, phase diagrams 31
– microscopic diffusion 193 vector model, order parameters 251
tracer diffusion 178 ff, 182 ff vibrational displacements 588, 597
– interfacial 490 vibrational energy, atomic ordering 573
– metals 226 vibrational modes, high-pressure transformations 674
– methods 231 viscoelastic phase separation 470
training, shape memory effect 643 viscous hydrodynamics, phase separation 446
transformation hysteresis, martensitic 633 f Volmer–Weber theory 342 ff
transformation-induced plasticity (TRIP) 593, 647 volume fractions
transformations, interfacial diffusion 481–518 – diffusionless transformations 587
transmission electron micrographs – eutectic growth 158
– decomposition 360, 366 – precipitation 373 f, 395
– diffusionless transformations 600 – spinodal decomposition 427
– directional dendritic growth 149 volume relaxation, atomic ordering 550
transparent salts, dendritic growth 115 volumetric pressure techniques, phase
tricritical point transformations 674
– atomic ordering 565
Index 713
water X-ray diffraction, phase transformations 587, 612,

– high-pressure transformations 687 ff 678
– phase diagram 54, 688 X-ray microanalysis 326, 511
wave modulations, diffusionless transformations 592
weak fluctuations, spinodal decomposition 418 ff
wetting phase transitions 283 Young's modulus, diffusionless transformations 590
whiskers, martensitic transformations 601
Widmanstätten precipitates 507
Willis cluster 230 Zeldovich factor 341 ff
wollastonite, diffusionless transformations 639 zeolite frameworks, high-pressure transformations
Wulff construction 662
– decomposition 344 f zero-phase fraction (ZPF) lines 58
– interfacial diffusion 488 zincblende type semiconductors, high-pressure
– solidification 90 transformations 679
wurtzite zirconium, diffusionless transformations 598 ff, 639
– diffusionless transformations 639 zirconium alloys, diffusionless transformations
– high-pressure transformations 680, 688 607, 617, 634
zone melting 98
Zr–Ti–Cu–Ni–Be glass, phase separation 455
X-ray absorption spectroscopy (XAS), transformations
676

Gernot Kostorz-Phase Transformations in Materials-Wiley-VCH (2001) PDF

Încărcat de

Informații document

Titlu original

Drepturi de autor

Formate disponibile

Partajați acest document

Partajați sau inserați document

Opțiuni de partajare

Vi se pare util acest document?

Este necorespunzător acest conținut?

Drepturi de autor:

Formate disponibile

Gernot Kostorz-Phase Transformations in Materials-Wiley-VCH (2001) PDF

Încărcat de

Drepturi de autor:

Formate disponibile

Gernot Kostorz (Editor)

Phase Transformations in Materials

Phase Transformations in Materials. Edited by Gernot Kostorz

Eric Lifshin (Ed.)

Weinheim · New York · Chichester · Brisbane · Singapore · Toronto

Library of Congress Card No. applied for

British Library Cataloguing-in-Publication Data:

Die Deutsche Bibliothek – Cataloguing-in-Publication Data:

© WILEY-VCH Verlag GmbH, D-69469 Weinheim (Federal Republic of Germany), 2001

Printed in the Federal Republic of Germany

Indexing: Borkowski & Borkowski, Schauernheim

Phase Transformations in Materials. Edited by Gernot Kostorz

Phase Transformations in Materials. Edited by Gernot Kostorz

ond Phase Precipitation” by R. Wagner, R. Kampmann and P. W. Voorhees) and Chapter 6

Gernot Kostorz, Zürich

Prof. Kurt Binder Prof. Gerhard Inden

Prof. Yves J. M. Bréchet Dr. Reinhard Kampmann

Phase Transformations in Materials. Edited by Gernot Kostorz

Prof. Heiner Müller-Krumbhaar Prof. Gary R. Purdy

Prof. Arthur D. Pelton Prof. Richard Wagner

1 Thermodynamics and Phase Diagrams of Materials . . . . . . . . . . . . 1

3 Diffusion in Crystalline Solids . . . . . . . . . . . . . . . . . . . . . . . 171

4 Statistical Theories of Phase Transitions . . . . . . . . . . . . . . . . . . 239

5 Homogeneous Second Phase Precipitation . . . . . . . . . . . . . . . . . 309

6 Spinodal Decomposition . . . . . . . . . . . . . . . . . . . . . . . . . . . 409

7 Transformations Involving Interfacial Diffusion . . . . . . . . . . . . . . . 481

8 Atomic Ordering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 519

9 Diffusionless Transformations . . . . . . . . . . . . . . . . . . . . . . . . 583

10 High Pressure Phase Transformations . . . . . . . . . . . . . . . . . . . . 655

Phase Transformations in Materials. Edited by Gernot Kostorz

Centre de Recherche en Calcul Thermochimique, École Polytechnique,

List of Symbols and Abbreviations . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

Phase Transformations in Materials. Edited by Gernot Kostorz

1.6 Application of Thermodynamics to Phase Diagram Analysis . . . . . . 34

List of Symbols and Abbreviations

b.c.c. body-centered cubic

1.1 Introduction ergy. General computer programs are avail-

Predominance diagrams are a particu- Hence, along this line:

Figure 1-1. Predomi-

1.3.2 Ellingham Diagrams

Figure 1-3. Predominance diagram

conditions of two- and three-phase equilib- Predominance diagrams have found

nent i, and the derivative is taken with all

1.4.3 Tangent Construction 1.4.4 Gibbs–Duhem Equation

giE = R T ln g i (1-58) The temperature – composition (T – X )

Figure 1-5. Phase dia-

Figure 1-6. Ge – Si phase diagram at

For example, the enthalpy of vaporization

A similar expression can be derived for the

1.5.5 Miscibility Gaps

If gE > 0, then the solution is thermody-

Figure 1-12. Phase diagram

This reaction occurs isothermally at 744 K

Such a description is justifiable only if g solvent ≈ 1

tectic reaction is l + s2 Æ s1 ; (ii) syntectics

1.6 Application of important step in the development of data-

following expansions for the partial excess

Figure 1-20. Sodium ac-

Eq. (1-106) reproduces the calorimetric

1.6.4 Calculation of Metastable Phase and Cd – Bi are found. Lines of constant

Figure 1-23. Perspective view of ternary space

Figure 1-24. Projection of

of a liquid of composition q on the liquidus is somewhat exaggerated in Fig. 1-25 for