Diffusion in Polymers PDF

DIFFUSION
POLYMERS
NEOGI
University of Missouri-Rolla
Rolla, Missouri
Marcel
Dekker,
Inc. New York. Basel Hong
Kong
Library o f Congress Cataloging-in-Publication Data
Diffusion in polymers/ edited by P. Neogi.

p. cm. -(Plastics engineering ;32)
Includes bibliographical references and index.
ISBN 0-8247-9530-X (alk. paper)
1. Polymers-Permeability. I. Neogi, P. (Partho).
11. Series: Plastics engineering (Marcel Dekker, Inc.)
;32.
QD381.9.P45D53 1996
668.94~20 95-51156
CIP
The publisher offers discounts on this book when ordered in bulk quantities. For more
information, write to Special SalesProfessional Marketing at the address below.
This book is printed on acid-free paper.
Copyright 1996 by Marcel Dekker, Inc. All Rights Reserved.
Neither this book nor any part may be reproduced or transmitted in any form or by any
means, electronic or mechanical, including photocopying, microfilming, and recording,
or by any information storage and retrieval system, without permission in writing from
the publisher.
Marcel Dekker, Inc.

270 Madison Avenue, New York, New York 10016
Current printing (last digit):

l0987654321
PRINTED IN THE UNITED STATES OF AMERICA

Preface
To most researchers in the area of diffusionin polymers, the 1968 book Diffusion
in Polymers by J. Crank and G. S. Park is a very familiar and most appreciated
one. important reason for its success, and one that will never revisit this area
again, is that the book appeared when research activity was about to explode
withtheadvent of membrane separations, barriermembranes,newneeds to
study polymer devolatilization, and on. It is now both out of print and out of
date, as is one updateof Polymer Permeabilityedited by J. Comyn. The books,
MembraneHandbook edited by W. S. H0 and K K. Sirkar and Polymer Gel
Separation Membranesedited by D.R. Paul andY. P. Yampol’skii, stress diffusion
only as a precursor to studying separations. Another, D i m i o n in and Through
Polymers by W. R. Vieth remains in the mainstream of diffusion in polymers.
This book began with the realization that fundamental changes have taken
place in this area. Diffusivity is no longer a phenomenological coefficient and
very firm validation from moleculartheoriesnow exists Fick’s law. High-
speed computers have become available that, in principle, can be used to cal-
culate these difhsivities. In practice the results are few, but presentavery
important view of the shape of things to come. The key results, however, are
provided by real-world phenomenology, whether it concerns understanding the
matrix of the solid polymers or predicting and correlating the diffusivities of
small molecules. These are presented to complement the more abstract concepts.
The molecular interpretations are not foregone, but at the same time numerical
accuracy is the more important criterion.
iii
iv PREFACE
Another development lies in the area of transport phenomena. It is no longer

possible to be content withmechanisms-in-words, because mathematical restric-
tions now exist to quantify constraints rising out of thermodynamics,mass,
momentum, and energy and species balances, and their methods of solutions
have become more transparent. In particular, conventional transport phenomena
used to address fluids had three important assumptions: homogeneity, isotropy,
and local equilibrium. None of these applies to solid polymers uniformly. Some
progress has been made in addressing these special effects.
A third development lies in advances in understanding the polymer matrix,
covering the physical chemistry of solid state and architectures at the molecular
level or at the scale of the membrane. Even in “structureless” melts, the study
of molecular conformations has proved to be critical.
This book examinesthese aspects and will serve chemical engineers whoare
involved in separations, controlled release, development of barrier membranes,
and transport phenomena in general; chemists, both physical chemists for some
of the same reasons and those who synthesize and evaluate new materials; and
finally physicists, to whom we owe the development of the molecular theories.
l? Neogi
Contents
Preface iii
Contributors ix
Chapter Diffusion in Homogeneous Media
J. M. D. MacElroy
I. Introduction 1
II.Diffusion Fundamentals 3
III. Simulation and Modeling of Diffusion in Fluid/Solid
Systems 13
IV. Concluding Comments 62
References 63
Chapter 2. Molecular Simulations of Sorption and Diffusion
in Amorphous Polymers
Doros N. Theodorou
I. Introduction 67
II.Characterization of Structure and Molecular Motion in
Amorphous Polymers 72
111. Prediction of Sorption Thermodynamics 91
IV. Prediction of Diffusivity 104
V. Conclusions and Future Directions 137
References 139
V
vi CONTENTS
3.
J. L. Duda and John M. Zielimki
I. Introduction
11. Free-Volume Concepts
III. Diffusion Above the Glass Transition Temperature
Iv. The Influence of the Glass Transition
V. More Complex Systems
References
173
l? Neogi
I.Introduction
11.Mathematical Methods
111.Non-Fickian Diffusion
Iv. Change Phase
V. Multiphase, Multicomponent, and Inhomogeneous
Systems
VI. Conclusions
References
5.
21 1
Sei-ichi Manabe
I.Introduction
11.Structural Characteristics of Polymer Solids
111.Thermal Motion of Polymer Chains in a Solid
IV. Correlation Between Chemical Structure, Composition,
and Penetrant Transport
V. Effects of Fine Structure (Crystallinity, Orientation, etc.)
of a Polymer Solid on Permeation Properties
VI. Concluding Remarks
References
251
Peter E Green
I. Introduction
11. Translational Dynamics in Homopolymer Melts
111. Diffusion of Chains of Differing Architectures
Iv. Interdiffusion
v. Diffusion in Block Copolymers
vii
CONTENTS
295
VI. Concluding Remarks
References
303
Index
This Page Intentionally Left Blank
Contributors
The Pennsylvania State University, University Park,

Pennsylvania
SandiaNationalLaboratories,Albuquerque, NewMexico
University College Dublin, Dublin, Ireland
Fukuoka Women’s University, Fukuoka, Japan
University of Missouri-Rolla, Rolla, Missouri
University of Patras,Patras,Greece
M. Air Products & Chemicals, Inc., Allentown,

Pennsylvania
I
Diffusion in Homogeneous Media
J. M. D. MacElroy
University College Dublin
Dublin, Ireland
I. INTRODUCTION
Recent advances in separation science and technology and in reaction engineer-
ing owe their origin, in part, to the development of specialized solid materials
that interact kinetically as well as thermodynamically in a unique and controlled
manner with multicomponent fluid mixtures. This ongoing technological growth
has taken place in parallel with an improvement in our understanding of the
fundamental properties of fluids in contact with solids. Noteworthy examples in
the chemical and biochemical process industries include energy-efficient and
nondestructive separation of molecular and macromolecular solutions by sorp-
tion onto solid substrates (Ruthven, 1984; Chase, 1984a,b; Norde, 1986; Yang,
1987), membrane separation of gases and liquids (Turbak, 1981; Drioli and
Nakagaki, 1986; White and Pintauro, 1986; Sirkar and Lloyd, 1988), and chro-
matographic separation of multicomponent mixtures (Yau et al., 1979; Chase,
1984a,b; Belter et al., 1988; Brown and Hartwick, 1989). The fundamental
mechanisms that govern the behavior of fluid/solid systems are also central to
research and development in such diverse areas as enhanced oil recovery, toxic
waste treatment, textile manufacturing, food technology, and biomedical engi-
neering, and although significant progress has been achieved much still remains
to be done.
The characterization of fluid/solid systems is particularly difficult when the
dispersion of the components of the fluid within the solid medium is determined
1
2 MAcELROY
solely by intimate details of the molecular structure of both the fluid and the
solid. For example, the very high sorptive specificity of some rigid microporous
materials is directly related to the geometrical and topological constraints
posed by the pore structure on the components of the adsorbing fluid. Solids
that fall into this class include the zeolites (Weisz, 1973; Satterfield, 1980; Ruth-
ven, 1984), which are cyrstalline media possessing pore apertures in the range
of 0.3-1.0 nm, with the actual aperture size depending on the origin and/or
method of manufacture of a given zeolite. Another example of a rigid medium
that exhibits a high degree of selectivity is molecular sieving carbon, which
contains local pore bottlenecks smaller than 0.5 nm (Juntgen et al., 1981). The
specificity of this material is most clearly demonstrated by its ability to separate
nitrogen from air. The mechanism for the separation process is kinetic in origin
in that the diffusion rates of oxygen and nitrogen within the pores of molecular
sieving carbon usually differ by a factor of 10 or more even though the sizes
of the molecules of these two species differ by only a few percent.
When the “solid” material is also nonrigid, the analysis of diffusion is much
more complicated. At a given temperature one is confronted with the need for
detailed information onthe time evolutionof the size, shape, and number of the
microvoids or cavities locally within the medium as well as the required char-
acterization of the fluid-solid intermolecular interactions. The temperature de-
pendence of the translational, rotational, and intramolecular motion of the mem-
brane atoms and particles and the concomitant existence of phase transitions
(glassy amorphous states to rubbery or liquid crystalline states and vice versa)
further complicates the description. In view ofthe rapidly growing technological
importance of materials of this type, particularly polymers, much effort has been
expended in elucidating the numerous subtle effects associated with these intra-
membrane characteristics (Crank and Park, 1968; Stem and Frisch, 1981; Vieth,
1991; Roe, 1991).
The material presented and discussed in this chapter is primarily of an intro-
ductory nature, and later chapters in the book should be consulted for details of
more specific methods of analysis and applications. The general framework of
the presentation provided here takes the following form. In Section I1 the flux
equations for homogeneous fluids are initially considered with reference to for-
mulations basedonbothnonequilibriumthermodynamics(phenomenological
description) andnonequilibrium statistical mechanics (moleculardescription).
These equations generally form the basis for the development of the flux rela-
tions for porous media and membranes, and in closing Section II the diffusion
equations for such systemsare presented and their limitations discussed. In Sec-
tion I11 the novel methodology of molecular simulation, particularly molecular
dynamics, and its application to diffusion in fluidholid systems are of primary
concern. Examples and applications are described for three different methods of
modeling the internal structure of permeable media: (1)idealized pore shapes in
MOGENEOUS
INDIFFUSION MEDLA 3
rigid media, (2) random bicontinuous media with a stationary solid phase, and
random media with a mobile solid phase (polymers). Finally, in Section IV
concluding comments are provided.
for
The conservation equationfor component i within an infinitesimally small vol-
ume element of a nonuniform, homogeneous system centered at r at time t in
the absence of chemical reactions is given by
aPi -
"- v piui
at
where and ui 2re the local mass density and velocity of component i at r and
at time The groupof terms piui is the flux of i relative a stationarylaboratory
frame of reference, i.e.,
JI = piui (2)
and is generally considered to be composed of two terms: (1)a convective con-
tribution arising from the local bulk motion of the fluid and (2) the residual
microscopic thermal motion of the molecules of component i relative this
convective flow. The definition one employs for the velocity of the bulk con-
vective motion is largely a matter of convenience, andone of the most common
frames of reference is based on the center-of-mass velocity of the fluid at r and
at time t which, for a v-component fluid, is given by
Thus, defining the mass diffusion flux of component i relative to the center of
mass of the fluid as
JIb = - U) (4)
where the superscript b refers to the barycentric frame of reference, Eq. (1) may
be written as
4 MAcELROY
In this equation is the local fluiddensity xi is the mass fraction of

component i, and D/Dt is the substantive derivative
~a -
"_ + U - v
Dt at
Historically, the mass fluxes J[ and J,!b and the corresponding molar fluxes
= niui (6)
or
JP = ni(ui - U) (7)
where ni is the local molar concentration of i, have been investigated either
phenomenologically via nonequilibriumthermodynamics [see, e.g., Prigogine
(1961),deGrootandMazur(1963),andHanley(1969)l or theoretically via
kinetic theory [e.g., Hirschfelder et al. (1954), Hanley (1969), and Chapman and
Cowling (1970)l. Although kinetic theory can, in principle, provide exactresults
for the fluxes in terms of the transport parameters and driving forces for mass
transfer, tractable expressions may be obtained in only a limited number of cases
(e.g., low-density gases). Nonequilibrium thermodynamics, on the other hand,
provides a general framework within which a consistent set of forces may be
prescribed for fluxes defined for various frames of reference. This approach,
however, is not without its own limitations, and the phenomenology of the
original theory of irreversible processes was a matter controversy for many
years. However, developments in nonequilibrium statistical mechanics stemming
from the early work of Green (1952,1954), Kubo(1957), and Kubo etal. (1985)
using linear response theory and, later, the projection operator formalism of Mori
(1965) have to some extent alleviated suspicions regarding the applicability of
irreversible thermodynamics. In the following the general tenets of nonequili-
brium thermodynamics are briefly summarized and a number of exact results
provided by nonequilibrium statistical mechanics are cited.
Nonequilibrium thermodynamics is founded on two postulates in conjunction
with the Onsager reciprocal relations [for details the reader is referred to Pri-
gogine (1961), deGroot and Mazur (1963), and Hanley (1969)l:
Postulate I: Microscopically large though macroscopically smallvolume

elements of a nonequilibrium system are themselves in local equilibrium, and
therefore the fundamental relations of equilibrium thermodynamics are locally
applicable. The rate of entropy production under these conditions is given by
where Ji and Xi are the conjugate f l u e s and forces.

MEDIA
MOGENEOUS
IN DIFFUSION 5
Postulate Ik For systems not too far removed from equilibrium, the fluxes
heat, mass, and momentum are linear homogeneous functions of the ther-
modynamic driving forces arising from the gradients temperature, chemical
potential, and the components of velocity
J=LX
where L is the matrix phenomenological (kinetic) coefficientsL,.
Onsager's reciprocal relations: with a suitable choice conjugate forces and

fluxes, the matrix phenomenological coefficients in Eq. is symmetric, i.e.,
L, = Lji.
For isotropic mediaandunder conditions of mechanical equilibrium, two

additional theorems may also be invoked in the general definition of the fluxes
given in Eq. (9). The first is Curie's theorem, which states that in an isotropic
medium the matrixL is a scalar and theforces andfluxes in Eq.(9)are therefore
of the same tensorial rank. This theorem is assumed to apply for the homoge-
neous systems under consideration in this chapter. The second theorem is that
proposed by Prigogine, which states that at mechanical equilibrium, the flow-
frame reference velocity in the definition of the diffusion jlu may be selected
arbitrarily without affecting the rate of entropy production. The condition of
mechanical equilibrium considerablysimplifies the analysis of diffusion, and
since this condition is usually involved in experimental diffusion measurements
it is appropriate to employ it here. Furthermore, in light of Prigogine's theorem
it is also particularly convenient to consider the diffusion fluxes in a laboratory
fixed frame of reference, in which case Eq. (9) provides the following expression
for the molar diffusion flux of component i:
where V T p j- Fjand T"VT are the thermodynamic driving forces for mass and
heat transfer as prescribed by theform of the entropy production equation arising
from Postulate I. [ A s an aside it is noted that the momentum driving force does
not appear directly in Eq. (10) because of its tensorial rank. This, however, does
not preclude an indirect influence of inertial and viscous effects on the flux of
material in systems that are notatmechanicalequilibrium.] p j and Fj are
the chemical potential of component j and the external force acting on compo-
nent j @er mole), respectively, and V, is the gradient operator at constant tem-
perature T.
Equation is the principal result of nonequilibrium thermodynamics that
will be employed in this chapter. The implications this expression from the
point of view of nonequilibrium statistical mechanics wereinvestigated in detail
6 MAcELROY
by Altenberger et al. (1987) and Kim et al. (1992), and it is worthwhile at this
point to outline a number of the salient features of these studies and prior the-
oretical developments in statistical mechanics. Altenberger and coworkers sum-
marized much of the earlier work of Mori (1965) on the projection operator
formalismin transport processes,andthey also extended Mori’smethod to
frames of reference other than the laboratory frame and the fluid center-of-mass
frame. The essence of Mori’s theory is that the molecular fluxes of the species
in a multicomponent system (and therefore the macroscopic fluxes Ji defined
above) are determined by a random component that is orthogonal to the space
spanned by the density and temperature fluctuations within the medium and an
induced or systematic contribution arising from the decay of these fluctuations.
In the interests of brevity, only the results for isothermal conditions are consid-
ered here, in which case the molecular diffusion flux of component i is given
by
ji& c) = c
NI
I= 1
vli exp(-ik r,i)
where vli and rli are the center-of-mass velocity and position of particle I of
component i at time t, X,& t ) is the wavevector-dependent diffusional ther-
modynamic force for component j given earlier in Eq. (lo), and L;@, t ) is an
“after-effect” function for diffusion that is related to but not equal to the phe-
nomenologicalcoefficient L, definedearlier. The distinction betweenthe
Green-Kubo linear response theory (Green 1952, 1954; Kubo, 1957; Kubo et
al., 1985) (which leads to a comparatively simple expressionfor L , and which
we specialize to below) and Mori’s projection operator formalism lies in the
random component j@, t) of the flux that appears explicitly on the right-hand
side of (llb) and also implicitly in the coefficients L # , t), i.e.,
1
Li(k, t ) = -k (j#,
t) jf(-k, 0)) k
k2kBT
where the angular brackets represent averaging overthe unperturbed equilibrium

canonical ensemble. The term inside the angular bracketsin this equation is the
time-correlation function of the random components of the microscopic fluxes
of components i and j , and the second term in Eq. (llb) characterizes the decay
of microscopic fluctuations in the medium. Mori (1965) showed that in thelimit
k + 0 (the subsystem volume,V, is much larger than the scale of local molecular
density inhomogeneities) the frequency-dependent form of Eq. (12) simplifies
MEDIA
MOGENEOUS
IN DIFFUSION 7
to the corresponding expressionprovidedby linear response theory (Green,

1952, 1954; Kubo, 1957; Kubo et al., 1985),
i.e., a kinetic coefficient determined by the mechanical properties j , in contrast

to the random components of these fluxes that are involved in the generalized
expression for the coefficients Lk(k, t ) in Eq. (12). Equation (13) is much easier
to evaluate and is of primary concern in this chapter. Fortunately this expression
is essentially exact except when variations in composition over lengthscales on
the order of the molecular dimensions within the fluid are of interest, in which
case nonlinear contributions in the thermodynamic forces that appear explicitly
in the general frequency-dependent form of L& W ) = J: exp(-iwt)L:,(k, t)dt
(Mori, 1965) need to be taken into consideration.
Mori also showed that in the low-frequency limit W + 0 the random contri-
bution j f in Eq. (llb) may be neglectedand under these conditions the diffusion
flux within a microscopically large but macroscopically small volume element
of the nonuniform system may be expressed in the form given earlier in Eq.
(10) with the phenomenological coefficients given by
where qij(t)is the velocity correlation function (VCF)
The zero frequency limit corresponds to time scales that are significantly longer
than the decay times of the VCFs appearing in Eq. (14a),and therefore the
principal restrictions involved in the application of the linear flux relations given
in Eq. (10) with Lij given by Eq. (14a) are that (1) the local thermodynamic
properties in the nonequilibriumsystem should not vary significantly over length
scales on the order of molecular dimensions and (2) the time scales of interest
should belonger than the characteristic relaxation times for molecular processes.
In many situations of interest these restrictions are not crucial and the linear
relations coupled with the Green-Kubo integrals for L, provide an accurate
description of diffusion in nonuniform homogeneous fluids.
Another mathematical form for the kinetic coefficients originally proposed
by Einstein is obtained by carrying out the integration indicated in Eq. (14a).
One finds
8 MAcELROY
Although this equation is frequently cited and employed in the literature, in this
chapter the Green-Kubo form provided in (14a) is favored in view of its
significant theoretical and experimental interest.
As noted above, Altenberger et al. (1987) also provide expressions for the
multicomponent diffusion fluxes in a variety of frames of reference, and one of
the most importantconclusions of their work is that whilethe kinetic coefficients
in different flow-frames maybe determined from a knowledge of the coefficients
L, [the laboratory fixed frame parameters provided in Eq. (14)], the reverse is
not true. Of particular interest are the kinetic coefficients for the center-of-mass
frame of reference, which may be determined from the fixed frame coefficients
using the relationship
where Mkis the molecular weight of component k. The molar diffusion fluxes
J: in this frame are again given by (10) with L, replaced by R, and with
L , replaced by a similar expression for Riq [see Altenberger et al. (1987) for
details]. Further comment onthe coefficientsL, is postponed until later, although
at this point it is worthwile considering an additional result [originally derived
by Mason and Viehland (1978)l stemming from Eq. (10) and the concomitant
condition of mechanical equilibrium.
Dividing Eq.(8)by ni and subtracting from this resultthe corresopnding
expression for component k, one finds
Now consider the condition of mechanical equilibriumas expressedby the

Gibbs-Duhem equation
with
where P is the local pressure in the system. TreatingV+,, - F, as the dependent

driving force in Eq. (17), then Eq. (16) may be written as
MOGENEOUS
EDIA IN DIFFUSION 9
where
Now multiplying J2q. (19) across by ni?Ik/n2Dik,summing over all species, and
defining
where Dikis the mutual dif€usion coefficient for the pair of species i and k, then
the flux equations may be written in the Stefan-Maxwell form
(V+, - Fi) - kTiV In T (i = 1, . . . , v)
where kTiis the thermal diffusion ratio for component i and is given by
kTi = c
k=l
d'iq - n&kq
n2Dik
From the definition given in (21)and the condition that the dependent
driving force is V+, - F,, it is easily shown that the mutual diffusion coeffi-
cients and the kinetic coefficients are interrelated by the expression
where 6, is the Kronecker delta.

The Stefan-Maxwell equation [M. (22)] is equivalent to Eq. (10) and is a
very useful way of expressing the diffusion fluxes in multicomponent systems.
In particular, this equation has been widely used in the development of models
for diffusion in porous media, most notably the Dusty Gas model [Mason and
Malinauskas (1983)], and in the next section the underlying principles of the
latter approach are employed in the formulation of general expressions for the
diffusioncoefficients of amulticomponentfluid in homogeneous fluidholid
media.
for
If the volume element of the fluid/solid system in which the concept of local
equilibrium can be considered applicable includes both the fluid components
and particles of the solid material, then all of the equations discussed in the
previous section can be used to describe diffusion within the solid with the solid
10 MAcELROY
medium itself also treated as one ofthe components in the multicomponent

system. For the solids of interest here, i.e., microporous media and polymeric
materials, local inhomogeneities usually exist only over length scales on the
order of atomic or molecular dimensions, and therefore it is assumed here as a
working premise that the concept of local equilibrium applies. For fluidholid
systems not far removed from equilibrium, the flux equations are therefore
(i = 1, .. ., V, m) (25)
or
(i = 1, . . . , U, m) (26)
with L, and D , given by Eqs. (14) and (24), respectively. For convenience the
contribution arising from the solid (component m) has been separated out.
In treating the solid as a component of the mixture it is assumed implicitly
in Eqs. (25) and (26) that thematrix of the material making up the solid medium
conforms to the condition of isotropicity. If this is nottrue,thenthe above
equations may still be considered applicable locally within the medium as long
as the anisotropic character of the local solid structure is taken into account. For
example, a very simple model that is frequently employed in the analysis of
diffusion in porous membranes is the cylindrical pore model; i.e., the solid
structure of the medium is assumed to form long cylindrical cavities along the
axial (z) direction of which the components of the fluid are allowed to diffuse.
Equations (25) and (26) may be employed under these conditions to predict the
axial diffusion fluxes in a given pore using
(i = 1, . . . , v, m) (27)
or
d In T
- kn - (i = 1, . . . , v, m) (28)
dz
DIFFUSION IN HOMOGENEOUS
MEDIA l1
These results can then be incorporated into a network model for the pores of
the medium to estimate the macroscopic mass transfer rates through the mem-
brane. It is not the intention here to review the various network models that
haveappearedin the literature over the last few decades,andthe reader is
referred to a number of articles dealing withrecent developments in this subject
(Reyes andJensen, 1985; Nicholson et al., 1988;Sahimi,1988; Zhangand
Seaton, 1992). However, there are a few points worth noting with regard to the
application of the above equations to anisotropic pore structures.
1. The influence of the solid arises explicitly through the terms L? and D?.
At first glance this type of formulation might appear to becounterintuitive,
i.e., the pore fluid should be considered separately and a boundary value
problem should be solved. For microporous systems, however, it is much
more convenient to includethe solid phase as a component in the di€€usion
equations, as permitted by Postulate I, even if the atoms of the solid assume
a geometrically ordered configuration [for further comments on this aspect
of transport in model pores, see Mason et al. (1963)l.
2. A fundamentalproblemdoes arise,however,in applying linear response
theory to porous media that are locally anisotropic due to the limiting con-
dition of zero wavenumber that is implicit in Eqs. (27) and (28) [see the
discussion following Eq. (12)]. Consider the following question: For very
fine pores, how can one obtain a sensible measure of the macroscopic dif-
fusion parameters if, owing to the very dimensions of the micropores, the
linear response kinetic coefficients in the limit k + 0 and -+ 0 are, under
certain conditions, nonexistent? The difficulties posed by this question were
clearly illustrated in the work of Vertenstein and Ronis (1986, 1987) and
by Schoen et al. (1988). In particular, Schoen et al. demonstrated that al-
though linear response theory does indeed providemeaningfuldiffusion
coefficients for diffusion parallel to the surface of thesolid material on either
side of a slit-shaped pore regardless of the width of the slit, this is not
generally true for diffusion normal to the pore walls. Later in this chapter
results obtained from an application of linear response theory to diffusion
in cylindrical pores (SuhandMacElroy, 1986) arediscussed to further
illustrate this point.
The above comments, coupled with the earlier remarks on the theoretical
results provided by Mori usingprojection operator theory, lead to the following
summary of limitations associated with Eqs. (25)-(28):
1. The fluidholid system must be isotropic ifEqs.(25)and(26) are to be
considered applicable. For solid structures that are strongly anisotropic lo-
cally, equations similar to Eqs. (27) and (28) maybe used along axeswithin
the medium that are translationally invariant.
12 MAcELROY
2. Density and temperature variations within a microscopic volumeelement of

the medium should be negligible, and the time scales for molecular relax-
ation processes should bemuch shorter than the macroscopic time scales of
interest in the diffusion measurements.
In many situations of interest these limitations do not significantly influence the
prediction of transport rates in homogeneous media. The application of Eqs.
(25)-(28) does, however,require consideration of the mobilityof the solid phase
both at the macroscopic level (i.e., U,) and at the microscopic level [v& in
Eq. (14b)l. Furthermore, in view of the inverse dependence of D , on the number
density of thesystem, n, as predicted generally by kinetic theory, it is convenient
to redefine the pair diffusion coefficients for the system as follows (Mason and
Viehland, 1978; Mason and Malinauskas, 1983):
where = n,,i.e., the numberdensityof the fluid within the medium(fluid

particles per unit volume of the fluid/solid system as a whole). With these def-
initions and specializing to isothermal conditions, then Eqs. (25) and (26) may
be rewritten for the fluid species as
and
(i = 1, ..., V) (32)
The Gibbs-Duhem equation [Eq. (l?] has been used to simplify Eq. (31). If
the particles or atoms of the solid phase are assumed to be stationary (“rigid”
media), then Eq. (31) simplifies further because J, = 0 and Lh is also zero as
indicated by Eq. (14).
Equations (31) and (32) have been widely used in the literature in the de-
velopment of correlative models for diffusion in porous media and polymeric
materials (frequently subject to the assumption L, = 0 and in many cases for
single-component difision only), and the reader is referred to these sources for
full details of the modeling techniques in current use [see, e.g., Crank and Park
(1968), Stem and Frisch (1981), Vieth (1991), andMasonandMalinauskas
(1983)l. In Section 111, a methodology that has been developed over the last
decade is reviewed. This approach involves direct molecular dynamics simula-
tion of confined fluids in model systems to computethe VCFs appearing in Eq.
MEDM
MOGENEOUS
INDIFFUSION 13
(14b)and hence the kinetic coefficients L,. The rapid advances in computer
technology over the last 15-20 years, and particularly the advent of supercom-
puters and more recently dedicated desktop workstations, now permit “exact”
determination of the transport (as well as equilibrium) properties for a wide
variety of systems. The advantages of computer simulation as a means for in-
vestigating the behavior of fluids and solids are clearly demonstrated in each of
the works that have appeared in the last decade [see, e.g., Roe (1991), Nicholson
and Parsonage (1982), and Allen and Tildesley (19831, and the most important
of these advantages fromthe point of view of transport in homogeneous media
are summarized as follows:
1. Microscopic properties, which cannot bereadilymeasuredexperimentally

but which are nonetheless central to a physical understanding of the under-
lying mechanisms for fluid transport within solid media, are accessible via
molecular simulation.
2. Although simulation can neverreplace actual experimental measurements,
it can, in conjunction with theoretical modeling and limited experimental
data, serve as an accurate andpowerful predictive tool for extrapolating
beyond the range of possible laboratory measurements. This aspect of mo-
lecular simulation is particularly important for microporous media and mem-
branes in view of the complexity and, in many cases, theexpense associated
with detailed experimental measurements of transport in such systems.
The discussion in Section III is presented in three parts. As an introduction

to those whoare not familiar with molecular dynamics simulation, computations
based an idealized pore geometry are considered first. The pore geometry
employed here is the cylindrical pore model that has been widely used in the-
oretical studies of boththetransportand equilibrium properties offluids in
membranes. This is followed by a summary of work on random media in which
the particles and atoms of the solid phase are held stationary, and finally recent
developments in the simulation and prediction of diffusion in amorphous poly-
mers are discussed.
SIMULATION AND MODELING OF DIFFUSION IN

FLUID/SOLID SYSTEMS
A. Diffusion in an Idealized Pore Geometry-
The Cylindrical Pore
For a cylindrical pore of length >> R,, where R, is the pore radius, and
assuming that the solid phase is immobile, Eqs. (31) and (32) with L, = 0 are
14 MAcELROY
applicable for diffusion in the axial (z) direction along the pore,
or
(i = 1, . . . , v) (34)
limiting cases of these equations are considered here: a single-component
pore fluid and a binary mixture.For simplicity it isalso assumed that noexternal
forces act on the fluid components.
1. Single-ComponentPoreFluid
The diffusion flux of a pure fluid f is given by
or
Jr)= - -- ","IT
(36)
For a bulk external phase at a chemical potential p , ,in local equilibrium with
the pore fluid at z, = +
= p,O(T) k,T In tB, where tBis the activity of the
bulk fluid that satisfies the limiting condition CB + n, as n, + 0. Substituting
this expression into Eq. (36) gives
where Kf is the partition coefficient for the fluid f defined by
Recall that n, is the local number density of the pore fluid, and for single-pore
analyses of the type under consideration here this number density traditionally
has units of fluid particles per unit pore volume (i.e., the pore wallis the bound-
DIFFUSION IN HOMOGENEOUS MEDIA 15
ary of the control volume under investigation). Equation (37b) corresponds to

the Darken equation forthe diffusion flux,while Eq. (37c) takes on a particularly
simple form in the ideal gas limit nm + 0,
If the gas does not adsorb on the pore walls, the diffusion coefficient appearing
in this equation is equivalent to D X , the Knudsen or free-particle diffusion co-
efficient for the gas f. If the diffusing gas particles are treated as points, then
the functional form of Dm is simply (Kennard, 1938)
where is molecular mean speed (8kT/mn)'" andf is the fraction of gas particles
that are reflected from the pore walls according to the cosine law for diffuse
scattering. Also note that for point gas particles the partition coefficient K' = 1.
When the s u e of the gas moleculesis taken into consideration, these expressions
require modification. For example, for spherical nonadsorbing gas moleculesof
diameter uf,E q . (40a) is rewritten as (Suh and MacElroy, 1986)
Dm=- (-
2 ')
VRp(l - A)
where A = uf/2Rp. The partition coefficient under these conditions also depends
on A in accordance with (Ferry, 1936)
Note that the net effect of these modifications is to introduce the factor (1 -
A)3 into the Knudsen permeability DX& and therefore, even for the smallest
gas particles (e.g., helium), steric effects cannot be neglected for diffusion in
pores of diameter less than approximately nm.
For adsorbing gases and vapors and for liquids, there are, unfortunately, no
general tractable expressions for either D Z or Kf (or for the bulk phase activity
CB), and approximate theoretical results exist in only a limited number of cases.
Theoretical developments in this area may be guided, however, by direct com-
puter simulation of the configurational and dynamical properties of the pore
fluid, and it is this method of approach to which we now turn.
In general the kinetic coefficient LE) and hence the diffusion coefficient DZ
for the pure micropore fluid may be expressed as follows using Eq. (14):
16 MAcELROY
and
where $(t) is the instantaneous center-of-mass velocity of the pore fluid as a

whole in the axial direction of the pore at time t, i.e.,
Also recall that the angular brackets in the above expressions represent aver-
aging over an equilibrium ensemble and therefore n, is the local equilibrium
number density the pore fluid. Furthermore, N is the number of fluid particles
in the locally equilibrated pore fluid, and it is of interest to observe that the
diffusion coefficient is very simply related to fluctuations in the velocity
field of these particles. Since this diffusion coefficient is also an intensive prop-
erty, the integral J: (uf)(t)uf)(O)) dt in the aboveequations is an extensive
property and disappears in the limit N
A variety of molecular simulation techniquesare in current use that are based
on sampling fromeither equilibrium ornonequilibrium ensembles, andthe com-
putational procedures involved in these methods have been described in detail
in a number of very readable texts [see, e.g., Nicholson and Parsonage (1982),
Allen and Tildesley (1987), and Roe (1991)l. In this chapter, only the method
most frequently employed in the computation of the VCFs appearing in Eqs.
(14), (42),and(43)-namely, equilibrium molecular dynamics (MD) in the
microcanonical ensemble (fixed particle number N,fixed volume V, and fixed
energy E)-is considered. Furthermore, since the primary objective here is to
illustrate the principles involvedin the application of the equilibrium MD
method to fluidholid systems, the diffusing fluid molecules are assumed to be
simple structureless spherical particles. models for the interparticle inter-
actions in the fluid are considered: (1)the hard-sphere interaction potential
and (2) the Lennard-Jones (12-6) interaction potential
+ij(rij) = k i jZl
[);( - ($1
where, in both cases, rij is the relative separation of particles i and j , uijis the
relative separation of the particles when the potential energy becomes positive
(repulsion), and q j in (46) is the potential energy minimum for attraction
(pairwise additivity of the interactions is also assumed here).
MEDM
MOGENEOUS
INDIFFUSION 17
The principal objective in the (classical) MD method is to solve Newton’s

equations of motion for the center-of-mass of each of the fluid particles in the
system subject to the interparticle forces derivedfromEq.(45) or Eq. (46)
and the forces associated with interactions between the fluid and the pore walls.
The latter interactions will also involve a hard-core potential and a London-
van der Waals potentialsimilar to the above expressionsfor the respective fluids;
however, before discussing the details of these interactions it is worthwhile to
briefly outline the method of solution of Newton’s equations of motion for each
particle i:
dr,
- = v;
dt
For the hard-sphereinteractiongivenbyEq. (45) the force, F,,acting on

particle i is zero between collisions and is impulsive during any given collision.
Noting that a collision takes place when the relative separation of the colliding
pair of particles is rij = uij,then, it is readily shown that the time to collision,
given that the initial positions of the particles are rioand rjo and their precolli-
sional velocities are vio and vjo, is given by
(vijo r,, < (48a)

where vijo = vjo - vio and r,, = rjo - r,. The condition vijo rijo- simply
refers to the fact that the particles must be approaching one another for the
collision to occur. Also note that a collision is not predicted if the group of
terms under the square root is less than zero (the particles bypass each other).
When the particles collide, their momenta and energies are changed in ac-
cordance with the momentum and energy conservation laws of physics subject
to possible constraints associated with the external forces acting on the system.
For a system of particles that are not subject to external forces and that obey
the normal rules of specular scattering (smooth hard spheres), conservation of
total linear momentum and energy for the colliding pair provides
m,mi
m,Av, = - 2 -(V;,n k) k
mi + mj
where AV, is the change in the center-of-mass velocity of particle i, mi is the
particle mass, and k is the unit vector along the line of centers at collision. A
similar equation of opposite sign is obtained for AV,. [Other hard-core particle
18 MAcELROY
models are discussed by Allen and Tildesley (1987), and the reader is referred
to this source for details.]
For Lennard-Jones fluids, the forces Fi in Eq. (47b) are generally nonzero at
all times andvary continuously with time as the particles move within the
potential field exerted by their respective neighbors. For pairwise additive in-
teractions in homogeneous Lennard-Jones fluids, the force on a given particle i
is determined by
jti
with $ij given by Eq. (46). At a given time t the position of each of the particles
in the system is known, and therefore the forces (or particle accelerations) can
be determined using the above expression. The positions and velocities of the
particles a short time later may then be calculated by expressing Eqs. (47a) and
(4%) in finite-difference form. A number of finite-difference algorithms have
been proposed [see Allen andTildesley(1987)for details], one of themost
popular of which is known as the “velocity Verlet” algorithm (Swope et al.,
1982). This procedure relates the particle positions and velocities at time I +
At to the corresponding values and the accelerations at time t according to
ri(t + At) = ri(t) + At v,(t) + -21 At”,@) (50a)
and
vi(t + At) = v,(t) + -21 At [a,(t) + ai(t + At)]

For atomic fluids the time step to be used in the above equations typically lies
in the range lo-” At S. This rangeof values usually ensures reliable
conservation of energyduring agiven simulation run(AllenandTildesley,
1987).
For hard-sphere or Lennard-Jones fluids confined within cylindrical pores,
the only additional contribution that needs to be included in the equations of
motion for the particles is the fluid particle-pore wall interaction. In the follow-
ing the conditions appropriate for the hard-core system are considered in detail
first and later we return to the simulation conditions for the Lennard-Jones pore
fluid.
Hard-core Interactions. For particle-pore wall hard-core collisions, the time
to collision is given by
MEDIA 19
where vioand rioare the two-dimensional vector contributions to the initial ve-
locity and the initial position of particle i in the xy plane of the pore cross
section and the origin of the coordinate frame lies on the pore axis. The eval-
uation of the velocity change on collision depends on the mode of scattering
assumed,andthe results reportedby Suh andMacElroy (1986) for the
limiting cases of specular reflection and cosine law diffuse scattering [f= and
1, respectively, in Eq. are considered here. For specular reflection, only
thevelocity components inthe plane the pore cross section are changed
during the collision:
Avi = - 2 7(vio riw)
with riw= R, - For cosine law diffuse scattering, however, all three com-
ponents of thevelocity are altered.Here only elastic diffuse scattering is
concern, in which case the kinetic energy of the colliding particle is conserved
during the collision and, as shown by Suh and MacElroy (1986), the postcol-
lisional components of the particle velocity in cylindrical coordinates are
where 5, and tz are random numbers that are uniformly distributed on the in-
terval (0,l).
The computation of the individual particle trajectories using the above equa-
tions is generallysupplemented with one or more simplifying computational
devices to alleviate the burden of the calculations [a number of “tricks of the
trade” are described in detail by Allen and Tildesley (1987)l. The most impor-
tant of these devices is common to nearly all molecular simulatins and involves
periodic imaging of a fundamental cell containing a finite number particles.
It was clearly recognized in the 1950s and 1960s by the pioneers of molecular
simulation methods that no computer in existence at that time or indeed at any
time in the foreseeable future could determine the trajectories for a system of
macroscopic volume containing billions of particles. Since computationscan be
carried out for only a finite numberof particles, the major difficulty to overcome
in simulations of homogeneousmedia is the condition associatedwith the
boundary of the simulation cell. A simple impenetrable wall is clearly out of
the question due to the severe distortion such a boundary would induce on the
particle phase coordinates, and a straightforward method for mimicking the be-
havior of the pseudo-homogeneoussystem is to consider the cell to be sur-
rounded on all sides by images of itself. Surprisingly, with fundamental cells
containing as few astens or hundreds of fluid particles this approach can provide
20 MACELROY
particle trajectory data that are sufficiently accurate for evaluation of the ther-
modynamic and transport properties of bulk macroscopic fluids. (For a number
of interfacial or near-critical states, this, unfortunately, is not the case due to the
long-range correlations involved in such systems, and care must be exercised in
the selection of the size of the simulation cell.)
The imaging procedureis illustrated schematically in Fig. 1for the cylindrical
pore under discussion here, and for the hard-core system a given simulation run
would typically proceed as follows. The number of particles, N,to be employed
in the simulation are placed within the fixed volume defined by the radius R,
and half-length L of the fundamental unit of the pore in either an orderly [see,
e.g.,HeinbuchandFischer(1987)l randommanner(SuhandMacElroy,
1986), the procedure selected generally depending on the required density of
the fluid. The initial velocities ofthe particles are thenusuallyassigned by
randomly selecting components from the Maxwell-Boltzmann velocity distri-
bution function [seeAllenandTildesley(1987) for details] subject to fixed
energy and zero net fluid momentum. The particle trajectories are then traced
in a sequence of steps in which
1. The minimum collision time predicted by either Eq. (48a) or Eq. (51) de-
termines the next collision.
2. All of the particles in the fundamental cell are moved through this minimum
collision timeandthe collision takes place. (If, during this process, any
particle moves out of the fundamental cell across the boundaries at ?L,
then it reappears in the cell at the opposite boundary.)
The momenta of the colliding particles are changed in accordance with Eqs.
(48b), (52), or
z=o
of the fundamental cell (below

-
particles shown inside the cell.
2
= -L and above
L j
Fundamental unit of the cylindrical pore. The filled spheres on either side
= +L) are images of the shaded
MEDIA
MOGENEOUS
INDIFFUSION 21
4. The collision times for the particles involved in step 3 are reevaluated.
5. Theabove steps are repeated.
After an initial equilibration period lasting approximately 500-1000 colli-
sions per particle, this sequence of computations continues until a sufficiently
large statistical sample of the (equilibrium) particle positions and velocities has
been recorded. Suh and MacElroy (1986) found that fundamental cells contain-
ing 200 particles were representative of the macroscopic thermodynamic system,
and equilibrium trajectories in the microcanonical ensemble approximately lo4
collisions per particle inlength were found to provide statistics of sufficient
accuracy for subsequent evaluation of the diffusion coefficients using Eqs.
(42)-(44). The equilibriumtime-correlation function (uf)(t) uf)(O)) appearing in
these equations is readily evaluated by sorting the stored trajectory data into
equally spaced time intervals 6t, and, using the ergodic hypothesis, the ensem-
ble-averaged VCF is given by
where M is the number of independent time origins employed in the averaging

process for a given value of j . For small j (short times), M will generally be
very large, and very accurate evaluation of the time-correlation function can be
achieved. For large j , i.e., times approachingthe length of the trajectory, M will
necessarily be a small number, and for this reason the statistical error in the
computation of the time-correlation function at long times will be large. Fre-
quently, this limitation does not play a significant role in the evaluation of the
kinetic coefficient L$) or the diffusion coefficient DC because the VCF usually
decays rapidly to zero. Under these conditions the upper limit of infinity in Eqs.
(42) and (43) may be replaced by a time t = tMnx(which, in many cases of
interest, is significantly shorter than the total length of the trajectory) with little
or no loss in accuracy inthe numerical integration involved in these expressions.
If the VCF does not decay rapidly to zero, it is still possible to obtain reliable
estimates for the transport coefficients by suitable extrapolation of the long-time
tail of the VCF, although this does require some prior knowledge or information
concerning the expected time-dependent behavior of the long-time tail [e.g., via
scaling theories (Havlin and Ben-Avraham, 1987)l. For the moment it will be
assumed that the VCF is zero at or beyond t-; we return to the problem of
long-time tails in Section IILB.
A number of typical center-of-mass VCFs for the pure hard-sphere porefluid
both in the axial direction and in the plane of the pore cross section are illus-
trated in Figs. 2 and 3 (Suh and MacElroy, 1986). (Also shown in these figures
are the VCFs for tracer particles, which are discussed in Section II.A.2). The
VCFs in the plane of the pore cross section are simply obtained from the
22 MAcELROY
velocity components of the center-of-mass of the fluid using the expression
The results shown in Figs. 2 and are normalized to 1.0 at zero time, and the
dimensionless time, T*, in these figures is in units of 2RP(1- A)/ij.
The VCFs illustrated in Fig. 2 correspond to the dilute gas (i.e., Knudsen)
limit, in which case, by definition, the diffusing particles never collide with one
another, i.e., the center-of-mass VCF simplifies to
where k = xy or Due to the absence of interparticle collisions, cross-

correlations for the individual particle velocities are nonexistent as implied by
2 Normalizedvelocityautocorrelationfunctionsforfree-molecule(Knudsen)
diffusion versus reduced timeT*. (-) Theoretical VCF in the axial direction for diffuse
scattering; moleculardynamics axial VCFfordiffusescattering; (- - -) and
(- - -) molecular dynamics VCFs in the plane of the pore cross section for diffuse and
specularscattering,respectively.[ReproducedfromSuhandMacElroy with
permission.]
DIFFUSION IN HOMOGENEOUSMEDU
1 I I
(a)
-
-
"
-
"
_.-
/
-0.2 - \ / -
- -
-0.60 1 1 I
0.25 0.50 0.75
T*
-
-
-
-0.4 - -
I 1 I
0.25 0.50 0.75
T*
3 Normalizedvelocityautocorrelationfunctions vs. reducedtime for X =
0.21 and n& (= n,&) = 0.6. (a) Specular reflection; @) diffuse reflection. Curve 1, VCF
corresponding to L?!,.; curve 2, VCF corresponding to L$); short- and long-dash curves,
VCFs in the plane the pore cross section for L!?/. and L$'), respectively. [Reproduced
from Suh and MacElroy (1986), with permission.]
24 MAcELROY
Eq. (55a), and, as indicated by Eq. (55b), the VCF under these conditions is
equivalent to the VCF for a single, isolated particle (N= 1) diffusing within the
pore. Note that the axial component of the VCF for particle/pore wall spec-
ular reflection[Es. (52)] is not explicitly shown in Fig. 2. The normalized axial
VCF in this case must be equal to 1.0 at all times because v(*)(t) is unchanged
during a collision with the pore wall. Also note that for these specular reflection
conditions the diffusion coefficient predicted by Eq. (43) is infinite, in accord
with the result predicted by Eq. (40) when = 0.
For cosine law diffuse scattering, the axial component of the momentum of
the particle is not conserved during pore wall collisions (one can hypothesize
the existence of an external clamping force on the solid that holds the pore wall
stationary during any given collision, and it is this force that would be required
in the balance equations to reinstate conservation of momentum). In this case
the axial momentum of any given particle is subject to “memory’’ loss during
collision with the pore wall, and for this reason the axial VCF shown in Fig. 2
for diffuse scattering decays to zero with increasing time. The solid line shown
in this figure is the theoretical (as opposed to simulation) prediction of the VCF
for difhse scattering, and its integral over time provides the Knudsen diffusion
coefficient given in Eq. (40) with = 1. It is of interest to note that although an
exponentially decaying VCF is frequently assumed in approximate theories of
diffusion, the decay in the axial VCF shownin Fig. 2 is not a simple exponential
as shownby Suh and MacElroy (1986). Indeed, a simple exponential decay
rarely describes the true temporal behavior of the VCF for a wide variety of
systems [even for homogeneous dilute gases (Alder and Wainwright, 1970)],
and care must be exercised when interpreting relaxation time constants obtained
assuming pseudo-exponential decay. A particularly important example of non-
exponential behavior that is believed to be of direct relevance to rigid glassy
polymers is considered in Section 1II.B.
section shown in Fig. 2 (nf - -

The fluidcenter-of-massVCFs for motion in the plane of the pore cross
0 and hence n, 0) and in Fig. at a higher
(liquidlike) reduced bulk density demonstrate oneof the shortcomings discussed
earlier with regard to the prediction by linear response theory of local diffusion
coefficients in anisotropic systems. It is clear from Figs. 2 and that the oscil-
latory behavior of the VCF will lead to diffusion coefficients that are com-
pletely different from the axial results, and in fact it is readily shown that in-
tegration of the dashed curves shown in Fig. 2 and the long dashed curves in
Fig. 3 over the time range from 0 to provide zero valuesfor 0%’. Such values
for 02)should be viewed as questionable in light ofthevery strong local
inhomogeneities involved in the fluid density, and due consideration should be
given to a more in-depth analysis based on the wavenumber dependenceof the
kinetic coefficientspredictedbythe projection operator formalism of Mori
(1965). Unfortunately, there is at present no simple way of evaluating the pro-
MEDIA 25
jected random fluxes appearing in this theory, and until further work in this area
is undertaken it will be necessary, as noted earlier, to restrict the application
the results of linear transport theory to diffusion in translationally invariant sys-
tems (in the present case, in the axial direction of the cylindrical pore).
The data represented by the filled circles illustrated in Fig. 4 (MacElroy and
Suh, are selected results for D$ for a pure fluid f as a function Xf =
uf/2R, at a fixed bulk phase reduced density n& = n,u: = 0.4054. (The open
circle and open square results are for the individual species in a binary mixture
at the same bulk density with h,/A, = These results are in the dimensionless
form D$/D,, where DK is given by Eq. (40) (with f = In the limit At +
(the hard spherical fluid particles approach the size the pore), diffusion within
the pore is described solely by free-particle motion (a result that is independent
of density). In the opposite limit, h, + 0, the diffusion mechanism is usually
referred to as viscous slip, and the coefficient D$ under these conditions is a
function of fluid density, decreasing with increasing density (Suh and MacElroy,
MacElroy and Suh, It was also shown by MacElroy and Suh
x,
Reduced axial di€fusion coefficient relative the membrane as a function
of particle reduced radius. Single-component system (a= R.(0)and (0)Results for
the solvent (a= and the solute (a= 2), respectively, in the binary system. [Reproduced
from MacElroy and Suh with permission.]
26 MAcELROY
that for pore sizes less than approximately one-tenth the diameter of the fluid
particles (typically R, nm)thetransportofa dense fluid or gas ina
pressure gradient is primarily determined by slip flow and not by shear flow;
i.e., the Hagen-Poiseuille equation or Darcy’s equation is not applicable in very
fine pores. This has long been known for dilute gases (Kennard, and, as
illustrated by MacElroy and Suh it is now possible to quantify the range
of validity of continuum formulations such as the Hagen-Poiseuille equation
for dense fluids and liquids using molecular simulation techniques.
London-van der Waals Interactions. For the cylindrical pore model a num-
ber of different particle-pore wall interaction potentials have been investigated,
primarily with the equilibrium properties of the pore fluid in mind (Peterson and
Gubbins, Peterson et al., although the transport properties have re-
ceived attention in afew studies (Heinbuch and Fischer, MacElroy and Suh,
[Transport characteristics have also been investigated via MD simulation
for London-van der Waals fluids confined within slit-shaped pores (Schoen et
al., Magda et al., Usually the particle-pore wall potential function
is represented by a two-body interaction in’whichthe solid is treated as a smeared
continuum of Lennard-Jones interaction sites. Heinbuch and Fischer em-
ployed a layered structure of concentric cylindrical shells of smeared solid atoms
in MD simulations of an adsorbing Lennard-Jones vapor. However, the most com-
mon representation is that of a continuum solid that is devoid of any internal or
surface structure, and for a pore fluid characterized by the potential given in Eq.
it has been shown that the potential energy for interaction between a fluid
particle i and the pore wall in this case is given by (Nicholson,
where ri is the radial positionofthefluid particle within the pore, eiWis

the potential minimum for interaction between the fluid particle and a single
Lennard-Jones site in the solid, urnis the corresponding Lennard-Jones size
parameter, and nw is the number of Lennard-Jones sites per unit volume in the
and in the limit R, -

solid phase. The two functions f9’(ri)and f”’(ri)are polynomials in ri (Nicholson,
Eq. simplifies to the potential function
frequently used in the modeling of sorption on flat surfaces (Steele, It is
also important to note that the definition of R, in Eq. differs from that
involved in the hard-core interactions discussed above. This difference is readily
seen by comparing Figs. and 5a.
In molecular dynamics simulations of a Lennard-Jones pore fluid whose in-
teractions with the pore wall are described by Fq. the total force experi-
enced by a given fluid particle i is obtained by including the force field exerted
DIFFUSION IN HOMOGENEOUSMEDH 27
5 Model cylindricalpore structures. (a) Particle-pore wall continuum interactions.

The hatched region r R, represents the inner repulsioncore of the solid surface atoms.@)
Structured pore wall. &a1 positions 1 and 2 are referred as the pore window and polygonal
cage, respectively. (Reproduced from MacElroy and Suh (1989), with permission.)
by the solid in Eq. (49) to give

N
ri&,(R, - r,)
= mia,(t) = - -- -
j=l rij drij ri dri
jfi
where riis the two-dimensional vector coordinateof the particle in the plane of
the pore cross section; i.e., for the smooth pore wall there is no axial force
component on the fluid arising from interactions with the solid phase. Therefore,
as in the case of the hard, specularly reflecting pore wall discussed earlier, the
axial momentum of the fluid is also conserved here and D E is predicted to be
infinitely large. Only if one introduces amechanism for axial backscattering
during interaction with the solid phase will a finite diffusion coefficient be ob-
served, and the simplest way to achieve this is to incorporate a discrete atomic
or molecular structure in the pore wall. Such a structure was introducedby
MacElroy and Suh (1989) [and in the slit-pore studies reported by Schoen et al.
(1988)] that is represented by a single periodic layer of surface atoms {S}whose
coordinates are given by
rj(j S) = R, cos
(k = 1, .. . , NR; Z = . . ., + (58)
28 MAcELROY
where NRis the number of surface atoms in a polygonal ring and U, is the axial
spacing of the rings (NRis 12 for the diagram shown in Fig. 5b). In the work
reported by MacElroy and Suh (1989) the interactions between these surface
atoms andthefluid particles in the pore were described by a Lennard-Jones
potential function similar to Eq. (46) with eij = eiwand uij= ai,. Furthermore,
as implied by the diagram in Fig. 5b, Eq. (56) was employed to characterize
the interaction of the pore fluid with the solid beyond the radial position r =
+
R, uiw/2 - ri, and the total force on a given fluid particle i for this structured
system is therefore
j#i jES
In the molecular dynamics simulation of a Lennard-Jones pore fluid subject

to forces of the type described by J3q. or Eq. (59), it is again quite clear
that only a relatively small number of particles (= 102-103) may be considered
in the finite-difference solution ofNewton’s equations of motion [e.g., using
Eqs. (50)].Periodic boundaries at z = +L are therefore employed to minimize
the influence of edge effects onthe properties determined from the particle
trajectories. For van der Waals interactions of this kind, an additional problem
arises that is not encountered in purely repulsive hard-core systems, namely, the
long-ranged nature of the interaction itself. In principal, this would imply that
a very large simulation cell should be employed, and as this is generally not
feasible it is necessary truncate the range of any given interaction at a point
that is at least as small as half a characteristic dimension of the fundamental
simulation cell [this limit arises from the minimum image convention; see Nich-
olson and Parsonage (1982) and Allen and Tildesley (1987) for details]. Tra-
ditionally the cutoff point or radius R,, for a spherically symmetric interaction
between particles i and j in a condensed phase is taken to lie between 2.50,
and 3.50,. The larger the value of Rcjj,the more closely will the simulated fluid
approach the physical behavior of the model fluid [e.g., the Lennard-Jones (12-
6) fluid characterized by Eq. (46)]; however, the upper limit in R,, is usually
governed by the CPU time required to compute all of the force contributions
within the spherical volume rij Rcij.This CPU requirement increases as N:,
where N, is the number of particles within the cutoff volume. Fortunately, for
London-van der Waals interactions of the type given by Eq. (46), the interaction
approaches zero rapidly with increasing rij and the computations are not seri-
ously influenced by the truncation at R,, [e.g., at a relative separation of 2.50,~
Eq. (46) provides r$ij(Rcij)= and at 3.50, the potential is -0.002eij].
Additional tricks of the trade such as shifted potentials (to ensure energy con-
servation in the microcanonical ensembleMD simulations), neighborhood lists,
cell linked lists, etc. [details of which may be found inAllenandTildesley
DIFFUSION IN HOMOGENEOUS MEDM 29
(1987)] should also be incorporated in the simulation code to improvethe com-

putational efficiency during program execution.
typical simulation run for a Lennard-Jones fluid confined within either of
&e model poresillustrated in Fig, 5 would proceed as follows. As described
earlier for the hard-core system,the N fluid particles are placed in the pore either
randomly or in an ordered manner, and their initial velocities are assigned from
the Maxwell-Boltzmann velocity distribution function at the desired tempera-
ture of the simulation. The total energy botential + kinetic) is again a fixed
quantity, and therefore during the initial stages of the simulation the temperature
(which is determined by the kinetic energy of the particles) will vary as the
fluid relaxes toward equilibrium. This necessitates rescaling the individual par-
ticle velocities during the equilibration period to return the system to the desired
temperature. The number of time steps involved in the finite-difference calcu-
lations during this equilibration period is typically =lo4, where, asnoted earlier,
At usually lies intherange 10”’ c: At < S. After equilibrium has been
achieved, rescaIing is terminated, and during the subsequent computations the
particle trajectories evolve at fixed energy. the equilibrium system no drift in
2
the average kinetic temperature T = (1/3NkB (Xrnp?) will be observed, although
fluctuations in Cuj?lNon the order of l/ N should be present. During the equi-
librium trajectory (usually sampled for approximately lo5 time steps), the par-
ticle positions and velocities are stored at equispaced intervals st for subsequent
evaluation of the VCFs using Eq. (54) and of the diffusion coefficients using
Results for the axial diffusion coefficient D% for the structured pore shown
in Fig. 5b for a range of values of N R [see (58)] and at a fixed bulk liquid
density n, = 0.61~;~ and temperature T = l.l5(~ff/kB) are reported by MacElroy
and Suh (1989). These data are reproduced in Fig. 6, where the reduced form
DEID, is plotted as a function of X = uf/2RFE.The Knudsen diffusion coeffi-
cient involved here corresponds to Eq. (40) with f = 1, and the pore radius is
defined as the effective quantity RFR. This effective pore radius is itself deter-
mined by using the definition of a dividing surface at the pore wall, which is
consistent with the definition of the dividing surface for a smooth, hard wall,
and therefore the magnitude of X obtained here hasa one-to-one correspondence
with the definition given earlier for hard-core interactions [see MacElroy and
Suh (1989) for details].
confirm the existence of viscous slip (nonzero D% in the limit h -

One of the most important aspects of the results shown in Fig. 6 is that they
for a
realistic liquidlsolid interface. It may therefore be concluded that, in general, in
addition to shear flow, slip flow should not be neglected as a viable mechanism
for transport in micropores. In the range of reduced radii h 2 0.5, the fluid
particles within the pore diffuse in single file and the continuum concepts of
shear and slip are no longer tenable. The trends in D E observed in Fig. 6 under
30 MAcELROY
0.10 -
0.08-
0.06-
0.04-
T 1
/
t
III
Figure 6 Reduced axial diffusion coefficient relative to the membrane vs. X for the
structured pore wall (Fig. 5b). Dn(is the free-molecule (Knudsen) diffusion coefficient,
and the upper abscissa, d, is the diametric distance between pore wall surface atoms in
units of Ur. [Reproduced from MacElroy and Suh (1989), with permission.]
these conditions are notably similar to those for the hard-sphere pore fluid
illustrated in Fig. 4.
2. Binary Mixtures
For a micropore fluid in local equilibrium with a bulk fluid mixture, the ther-
modynamic forces appearing in Eqs. (33) and (34) may be replaced by equiv-
alent bulk-phase thermodynamic forces, i.e.,
VTFJ - F, = (VTF, - FJ)B (60)

Furthermore, using the Gibbs-Duhem equation [Eq. (17)] for a binary bulk-
phase mixture of components 1 and 2, it is readily shown that the micropore
driving forces are interrelated by
and vice versa. Also noting that in general
where K, is the partition (or distribution) coefficient for component i, the flux
equations for the two species in the axial direction of the cylindrical pore are
given by Eq. (33) as
where D':)and DF) are the Fickian diffusion coefficients for the two species and
are related to the microscopic properties of the pore fluid as
and
with L:;) given by
The Stefan-Maxwell form of the flux equations may also be employed to

express the Fickian diffusion coefficients defined above in terms of the mutual
diffusion coefficients D\$, D:$,and 01'1(MacElroy and Suh, 1987):
1 1
IM
or
with similar expressions for DF). x1 and x2 are the mole fractions of the two
species within the pore.
32 MAcELROY
The mutualdiffusioncoefficientsmay also be expressedin terms of the

kinetic coefficients L$):
and
D:: = k,T~(')l/n,Ly~ (65c)
where IL"I = Ly:Lfi - L??.

The above results simplify when the special case of self-diffusion is under
consideration.In this case component 1 may be taken as the"solvent"and
component 2 is defined as the tracer, which, in principle, is at infinite dilution
and has the same molecular properties as the solvent particles. For clarity the
tracer is defined here as component l * , and in view of the equivalence of mo-
lecular properties we have Kl. = Kl and D$M = DYA. For this model binary
system, Eq. (61) for the tracer simplifies to
and the expressions for the Fickian and mutual diffusion coefficients reduce to
the following, in which it is assumed that Nl. = 1 and thus Nl = N - 1 (Suh
and MacElroy, 1986):
and
MEDIA 33
where Llf' is given by (42) and

"m
Note that in the thermodynamic limit N

D:? = D:) = v ~ , T L ~ ? , .
- Eq. (67a) simplifies to
(69)
It is this parameter (or more specifically its directionally averaged value for a
randomly oriented pore network) that is measured via nuclear magnetic reso-
nance spectroscopy of radioactive tracer studies, and when these are comple-
mented with molecular simulation results it should be possible to accurately
predict L$) and hence the diffusion coefficientD$M= D$. The latter coefficient,
which is of particular importance in the engineering design of adsorbers, mem-
brane separations modules, etc., may also be measured via gravimetric or vol-
umetric sorption experiments, which in turn may be used as corroborative evi-
dence for the validity of aproposedmolecularmodel for diffusionin
microporous media and membranes. We return to the distinction between DI'!
and D$ later in Section III.B, and for the moment we examinethe characteristic
behavior of D?? alone for tracer diffusion subject to specular or diffuse scattering
interactions with the cylindrical pore wall.
The determination of the tracer diffusion coefficient from the MD simulation
data for a pure micropore fluid involves a straightforward application of Eq.
(54) to a single particle in the system, and as each individual particle may be
independently considered tobe the tracer, a secondary averaging is permitted as
shown in Eq. (68b). This secondary averaging can lead to very accurate results
for DI".',in contrast to the membrane/fluidmutualdiffusioncoefficient D$,
which involvesa single measure the influence of the collective motion of the
pore fluid as a whole. Accurate determination of DE usually requires MD tra-
jectories that are at least an order of magnitude longer than those needed to
obtain tracer diffusivities of similar accuracy.
Sample results for tracer diffusion in a Lennard-Jonesliquid at a fixed chem-
ical potential confined within the smooth-walled and structured cylindrical pores
illustrated in Fig. 5 are provided in Figs. 7 and 8, respectively (MacElroy and
Suh, 1989). These results demonstrate theverysignificanteffect axial back-
scattering has on the diffusion mechanism and the need for a reliable atomistic
model of the solid phase (or, at the very least, some provision for axial back-
scattering) when conducting computer simulations of micropore fluids. As the
pore size decreases (h increases), the diffusion coefficient for the tracer in the
atomically structured pore drops rapidly in agreement with the general trend
34 MAcELROY
IO 3 2
3.5 I 1 I l
I
2.5
7 Reducedaxialdiffusioncoefficientforthetracer as afunction in pores

with smooth walls. MD simulation; (0)Davis-Enskog kinetic theory [Davis (19871
andFischer-Methfessel(1980)approximation; (0)Davis-Enskogkinetictheoryand
bulk fluid approximation; Lower dashed curve, The empirical correlation of Sattertield
etal.(1973). d asinFigure 6. [ReproducedfromMacElroyandSuh(1989),with
permission.]
expected in physically realistic situations. The straight solid line shown in Fig.
8 is, in fact, an empirical correlation obtained by Satterfield et al. (1973) from
a regression analysis on the diffusion coefficients for a variety of dilute aqueous
and nonaqueous solutions in microporous alumina [it is of interest to note that
a similar correlation has also been suggested to describe steric effects poly-
mers; see, e.g., Pace and Datyner (1979a,b,c)]. Somewhat similar results were
reported by Suh and MacElroy (1986) for tracer diffusion in a hard-sphere pore
fluid subject to either specular reflection at the pore wall[i.e., Eq. (52)] or cosine
law diffuse scattering [Eqs. (53a-c)].
DIFFUSION IN HOMOGENEOUSMEDU 35
8 Reduced axial diffusion coefficient for the traceras a function of A in pores

with structured walls. All symbols are as in Fig. 7. The numbers next to the simulation
points refer to the valueof N Rin Eq. (58). [Reproduced from MacElroy and Suh
with permission.]
For diffusion in binary pore fluid mixtures molecularly disparate species,

one the most important questions that frequently arise concerns the relative
importance of cross-coupling effects; i.e., can the cross-kinetic coefficients Lyi
and L';1 appearing in Eqs. (62a,b) be neglected? supplementary question is
then usually posed If the cross-coefficients are neglected, can one assume that
the coefficients L?; and L$; are simply related to their pure component values?
(The simplest approximation here is to assume that these coefficients are equal
to the pure component parameters.) Both ofthese questions are readily answered
MAcELROY
for low-density gas mixtures in macroporous media because reliable molecular

predictions can be made in such cases (Chapman and Cowling, 1970; Hirsch-
felder et al., 1954; MasonandMalinauskas, 1983); however, for micropore
fluids and particularly dense fluids or liquids, answers to these questions are not
easily obtained. a rule (particularly in view of the negative answers usually
implied for dilute gases) one should not neglect cross-effects unless independent
evidence exists to support the assumption that these terms are negligible. Even
for dilute solutions care must be exercised as illustrated by the MD simulation
results for a dilute binary hard-sphere dense fluid mixture reported by MacElroy
simplifies to
-
and Suh (1987). Taking components 1 and 2 as the solvent and solute, respec-
tively, for dilute solutions 0) the solute diffusion coefficient (62b)l
k*T
D , =-L!: (704
I
n2
= (v!)(t)v!)(O))dt (70b)
It has been assumed here that the contribution (KJK,) L$ in Eq. (62b) remains
-
finite in this limit. A similar simplification does not result for the solvent dif-
fusion coefficient [Eq.(62a)l unless K2/K, 0 and/or Lfi +. 0, and neither of
these conditions will be generally true. Thesimulation results reported by
MacElroyand Suh (1987) for both D? and D:) inhard-sphereporefluid
mixtures are shown in Figs. 9 and 10, where &,B in the mutual diffusion co-
efficient of the pair 1-2 in the bulk homogeneous fluid. W Oimportant trends
are observed in Fig. 9: (1)For large pores (A2 0), the rate of diffusion of the
solvent within the pore is much larger than correspondingrates in the bulk phase
[as discussed by MacElroy and Suh (1987), this phenomenon is generally
sociated with diffusive slip, which in turn is intimately related to osmotic trans-
port pphenomena]; and (2) for very small pores the solvent Fickian diffusion
coefficent is negative, and this is due solely to the influence of the cross-coupling
term in Eq. (62a). The latter effect arises from configurational constraints within
the pore fluid associated with both the magnitude of X , and the magnitude of
hl/A2(which equals0.6 in the present case). For therange of pore sizes in which
D? is observed to be negative, it was also shown by MacElroy and Suh (1987)
that KJK, > 1, and since the mutualdiffusioncoefficients DfA and D?; are
always positive it is seen from Eq. (64b) that it is this ratio that governs the
sign on D?).
Finally, from Fig. it is of interest to note that for large pores (and “non-
adsorbing” solutes) the solute diffusion coefficient reduced by the bulk-phase
solutelsolvent mutual diffusion coefficient is a universal function of the solute
reduced radius h,; i.e., the ratio D$)/D,%B, where is the solute, is independent
DIFFUSION IN HOMOGENEOUS MEDM 37
Reduced axial porediffusioncoefficientforthesolventasafunctionof

soluteparticlereducedradius.[ReproducedfromMasonandChapman, with
permission.]
of the size of the solvent particles, an observation that is in agreement with

experimental measurements (Satterfield et al., 1973). This is further confirmed
by the solid line shown in Fig. 10, which corresponds to the theoretical predic-
tions for diffusion of a solute of finite size in a continuum solvent [see, e.g.,
Anderson and Quinn (1974) and Brenner and Gaydos (1977)l. For small pores
+
(h, h2 > l),however, the pseudocontinuum assumption for the solvent breaks
down, and the intrinsic particulate or molecular structure of each component in
the pore fluid needs to be taken into consideration.
B. Diffusion in RigidRandom Media

Single-pore analyses of the type discussed above can provide valuable insight
into a variety of properties of fluids confined within cavities of molecular di-
mensions. However,one very important aspect that is not covered insuch studies
is the manner in which interpore connectivity in a macroscopic random medium
can influence the overall transport process. Two independent though comple-
38 MAcELROY
0.8
0.6 -
0.2 -
Reducedaxialporediffusioncoefficient for thesoluteasafunction of

solute reduced radius. Simulation results for the single-component system (a= l*)
with n& = 0.4054; (0)simulation results for the solute (a= 2) in the binary system with
n;, = 0.4, n& = 0.01; (--) continuum-mechanical theory [58,591; (- - -) Eq.
(40b) (a = l * and f = 1.0) reducedby D,l..8; (- - -) Eq. (40b) (a = 2) reducedby
D,ze. [Reproduced from Mason and Chapman, (1962), with permission.]
mentary approaches have been employed in the last 20 years to investigate the
effects pore space topology on both the equilibrium and transport properties
of fluids in random media, the first approach being based on lattice models of
the pore network [e.g., Shante and Kirkpatrick (1971), Kirkpatrick (1973), Reyes
and Jensen (1985), Nicholson et al. (1988), Sahimi (1988), and Zhang and Sea-
ton (1992)], while the second views the random medium (pore space and solid
phase) as a continuum [see, e.g., Haan and Zwanzig (1977), Nakano and Evans
(1983), Abassi etal.(1983),Chiew et al. (1985), Torquato (1986), Park and
MacElroy(1989),MacElroyandRaghavan(1990, 1991), andRaghavanand
MacElroy (1991)l. In the lattice models, a “bond” joining any two nodes or
“sites” in the lattice usually represents an individual pore channel (frequently
assumedto be cylindrical in shape) that is connected to other pores inthe
networkatthe respective nodes. These models are very versatile (individual
bonds may be assumed to represent pores different sizes or shapes, the nodes
may be assumed to have zero or nonzero volume, the site or bond coordination
number may be varied locally or globally, etc.) and are particularly popular in
MEDIA 39
studies relating to percolation phenomena in disordered systems. A primary dis-

tinction between lattice models and continuum models is the assumption in the
former that the topology and structure of the pore network can be treated in-
dependently of the underlying physics of the diffusion mechanism (the mecha-
nistic problem itself is usually considered for each bond individually using equa-
tions similar to those provided in Section IILA). In a continuum description of
a random medium the structure and topology of the system and the diffusion
(or percolation) process are usually modeled simultaneously. The solid phase
may, for example, be represented by inclusions randomly distributed in space,
with the interinclusion void space correspondingto the volume accessible to the
diffusing fluid components, or the inverse description may be employed in which
the void region corresponds to the space occupied by the (interconnected) in-
clusions. If the inclusions are spherical, the first model is sometimes called a
anno non ball^^ solid, whereas the secondmodel is referred to as a “Swiss
cheese” medium. The flux equations for diffusion in this case are essentially
the same as those described in Section III.A, with the following exceptions:
If it is assumedthat the medium is isotropic, then the flux expressions are
obtained using Eqs. and rather than Eqs. and and each of
the kinetic coefficients (and hence the diffusion coefficients) is obtained from
averages over all three Cartesian coordinates [as, for example, in Eq. and
the influence of the structure, shape, and topology of the pore space is im-
plicitly taken into consideration in the determination of the transport parameters
themselves.
For a wide variety of permeable media, and particularly for gels and polymer
films, it is believed that continuum rather than lattice models better represent
the structure of the material, and for this reason only continuum models are
considered in this section. Furthermore, since the microscopic structures of gels
and polymers frequently have the appearance of a network of beaded particles
(e.g., interconnected colloidal particles or monomer units), it is reasonable to
model the backbone of the medium as solid inclusions randomly distributed in
space. Early theoretical studies in this area were primarily based on the work
of Maxwell on the dielectric permeability of particle suspensions [for a
review of later improvements on Maxwell’s original formulation, see Barrer,
Chapter in Crank andPark andmorerecently statistical methods
(Prager, Weissberg, have been employed with some success. These
models, however, are applicable primarily to systemsin which the diffusing fluid
within the void space can be treated as a continuum (i.e., the size of the fluid
particles relative to the size of a typical cavity in the void region is inconse-
quential) and are therefore restricted to macroporous media or to heterogeneous
composites. As may be clearly inferred from the results discussed earlier for
idealized microcapillaries, in microporous media or in the amorphous regions
of dense polymer films the molecularproperties of the components in the system
40 MAcELROY
must be taken into consideration. Another limitation of these models and also
of recent variational formulations of gas diffusion in random media (Ho and
Strieder, 1980) is their inability to predict the properties of the permeating fluid
at or nearapercolationtransition. As we will see below, there is reason to
believe that the temporal evolution of the microscopic properties of severely
hindered diffusing species has a direct bearing on case 11 diffusion in polymers.
In a number of recent articles, both Monte Carlo (MC) (Nakano and Evans,
1983; Abassi et al., 1983) and molecular dynamics (MD) (Park and MacElroy,
1989; MacElroyandRaghavan, 1990, 1991;RaghavanandMacElroy, 1991;
MacElroy and Tomlin, 1992) simulationresults have been reported for diffusion
of gases in random media. The random media modeled in these studies were
composed of assemblies of solid spheres randomly distributed in space, and the
gas phase was usually considered to be nonadsorbing, although MD studies of
adsorbing Lennard-Jones vapors were also reported by MacElroy and Raghavan
(1990, 1991). In the following only the results obtained via molecular dynamics
are discussed, both for reasons of brevity and in view of the real-time analysis
involved in such computations,and the reader is referred to the work of Nakano
and Evans (1983) and Abassi et al. (1983) for details on the MC technique.
In the studies reported by Park and MacElroy (1989), the four model solid-
sphere assemblies illustrated inFig. 11 were investigated, and the pore fluid
simulated was a single nonadsorbing diffusing hard-sphere particle. These sim-
ulations therefore correspond to Knudsen diffusion, and to further simplify the
computations the solid spheres in each of the models shown in Fig. 11 were
chosen to be of uniform radius Extensive computations over a wide range
of porosities and for ensembles containingvery long trajectories were conducted
in these studies in order to clearly determine the long-time behaviorof the VCF
appearing in the free-particle diffusion coefficient
where v(t) is the velocity of the diffusing particle at time t. By definition the
(stationary-state) diffusion coefficient is independent of time; however, it is con-
venient in a number of specific cases to consider a time-dependent diffusion
coefficient defined by the integral to the right of the limit in Eq. (71), and we
return to this below.
The MDmethod employed byPark and MacElroy (1989) involved astraight-
forward application of Eq. (48a) to predict the time to collision for the diffusing
particle and the immobile solid spheres illustrated in Fig. 11. For simplicity the
postcollisional velocities were evaluated using (48b) (with the mass of the
solid spheres, m,,set equal to infinity) instead of the more realistic cosine law
reflection condition represented by Eq. (53). It should be emphasized here, how-
ever, that this does not have a significant influence on the outcome of the sim-
DIFFUSION IN HOMOGENEOUSMEDL4 41
11 Schematicrepresentations of theoverlappingandnonoverlappingspheres
models. The solid phase is indicated by shaded regions, and the full circle corresponds
to a fluid particle. [From Park and MacElroy with permission.]
ulations, and indeed only a numerical factor is involved in the final evaluation
of the diffusion coefficient (Mason and Chapman, 1962). The general form of
the VCF and in particular its long-time behavior as determined by storing the
particle velocity as a function of time is unaffected by the collision dynamics.
Early work by Alley (1979) on a two-dimensional overlapping disk analog of
Fig. l l a demonstrated this, and this was confirmedrecently for thethree-
dimensional systems of interest here (MacElroy, 1992, unpublished).
Selected results for the diffusion coefficient of the low-pressure nonad-
sorbing gas diffusing within the overlapping spheres models are shown in Fig.
12 (for the random overlapping spheres medium illustrated in Fig. l l a ) and in
Fig. 13 (the data represented by the full circles in this figure correspond to free-
particle diffusion within theconnected overlappingsystem illustrated in
Fig. llb). Results for the nonoverlappingsystems may befound in Park
andMacElroy(1989). The Boltzmanndiffusioncoefficient
-
cited inFig. 12
corresponds to the low-density limit for the solid (i.e., n: = n p ; 0) and is
42 MAcELROY
\cI
1.00 0.50 0.25 0.05 0.025
The free-particle diffusion coefficient reduced by its Boltzmann value (left

ordinate axis) and the open porosity relative to $ (right ordinate axis) as a function of
the reduced density (or porosity) for the random overlapping system (see Fig.lla).
MD simulation; 1, kinetic theory for moderately dense gases (van Leeuwen and Weijland
(1967) and Weijland and van Leeuwen (1968)); 2, memory function/mode coupling the-
(Gotze et al. (1981a,b)); repeated ring kinetic theory (Masters and Keyes (1982)).
[Reproduced from Park and MacElroy (1989), with permission.]
given by
where U is the diameter of the diffusing hard-sphere particle. Furthermore, since

the mean pore radius and the void space accessible to this particle in the random
overlapping spheres medium are also determined byboth U and n: accord-
ing to
(RANDOM)
0 0.2 0.4 0.6 0.8 1.2
I ' I 1
I
-
-
a0.644
-
0.200-
-
- -
0.080 - -
0.060,-
-
0.020.
-
0.008.
'
-
0.006
2.0
13 The diffusion coefficient reduced by its value for = 0 as a function of

fluid particle size (lower abscissa) or particle reduced size = UDR, (upper abscissa) in
overlapping systems with = (0)Random overlapping system (Fig. lla);
connected overlapping system (Fig. llb). [Reproduced from Park and MacElroy (1989),
with permission.]
and
44 MAcELROY
where S is the surface area per unit volume of the medium and JIT is the total
porosity available to a point particle, then Eq. (72) may also be written as
The simple relationship between this expression and the point particle Knudsen
diffision coefficient given in (40) is quite apparent. Recall, however, that
Eq. (75) corresponds to specular collision dynamics, and to obtain the expression
for cosine law elastic scattering the right-hand side of Eq. (75) is simply divided
by 1 + (4/9)f(MasonandChapman,1962), where f is the fraction of gas
particles undergoing diffuse scattering during collision with the solid spheres.
This modification may be used in conjunction with Fig. 12 to obtain estimates
of the free-molecule diffusion coefficient for nonadsorbing gases in amorphous
porous materials including rigid glassy polymers. The upper abscissa in Fig. 12
corresponds to Eq. (74), and therefore for a gas particle of a specific molecular
size and for a membrane of a known total porosity $T it is possible to imme-
diately read off the ratio D,/D, and hence obtain the “tortuo~ity’~-~~rrected
diffusion coefficient Dm. Note that the lower abscissa in Fig. 12 is to be used
only when the diffusing particle has zero volume, i.e., U = 0.
example of the application of the procedure described aboveis illustrated
in Fig. 13 for a medium with a comparatively high total porosity JIT = 0.644.
Also shown in this figure are simulation results for the more realistic model
depicted in Fig. l l b , which takes into consideration the effect of solid particle
connectedness in real media [it should be noted that for JIT < 0.5 the random
and connected models in Figs. l l a and l l b are equivalent because the random
overlapping spheresmedium satisfies the criterion of complete solid connectivity
under these conditions [Shante and Kirkpatrick, 1971; Kirkpatrick, 1973; Haan
and Zwanzig, 1977)l. At the porosity JIT = 0.644 the differences between the
two systems are significant only when the size of the diffusing gas particle is
greater than one-third the average pore size (the point particle coefficients DC
differ by only 10%). For small particle sizes the dependence of D , on U is
approximately exponential, in agreement with experimental observations for gas
diffusion in a variety of polymeric materials [see Pace and Datyner (1979a,b,c)].
However, for large gas particles (gas particles that are commensurate in size
with typical monomer units in polymer macromolecules), the diffusion coeffi-
cient is extremely small and is predicted to go to zero at a specific value of U.
This transition is generally referred to as the percolation threshold, and for the
three-dimensional models shown in Fig. 11its position depends not only on the
size of the diffusing particle but also on the porosity (or bulk density as illus-
tratedinFig. 12) of the rigid medium. The word rigid is italicized here to
emphasize the fact that in dense polymeric media the effects of a percolation
MEDM
MOGENEOUS
INDIFFUSION 45
transition will not be observed abovethe glass transition. Under these conditions
the intrinsic mobility of the polymer chains will preclude the existence of per-
colation phenomena and the diffusion coefficients for comparatively large non-
adsorbing penetrant molecules will be significantly greater than the predicted
values for the rigid systems shown in Figs. 12 and 13. However, at or below
the glass transition the characterization of the diffusive properties of polymer/
penetrant systems will, at least in part, be assisted by studies of diffusion in
rigid media. Of particular interest in this respect is the temporal behavior of the
VCF, and it is this to which we now turn.
The VCFs for a number of the diffusion coefficients reported in Fig. 12 are
provided in Fig. 14. These correlation functions are plotted here in the normal-
ized form defined by
W ) = (v@) v(O))/(v2(0))
One of the most important points to note concerning the long-time behavior
the VCFs is the increasing depth of the negative tail and its persistence to very
long times as the porosity of the medium decreases and/or as the size of the
diffusing particle increases. The minimum in theVCF results from particle back-
scattering after an initial positively correlated forward momentum, and the neg-
ative correlation persists to long times dueto a greater than average probability
for the particle to return to the origin of its trajectory. This is particularly true
at low porosities and/or for large diffusing particle sizes where partial or even
complete caging of the particle amplifies this effect. When it is further noted
that the diffusion coefficient is obtained from the integral of the VCF over time,
the relative importance of the long-time tails shown in Fig. 14 becomes abun-
dantly clear. The simulation results for these tails also confirm the power law
time dependence predicted by approximate kinetic or mode-coupling theories
(Gotze et al., 1981a,b; Masters and Keyes, 1982; Ernst et al., 1984; Machta et
al., 1984) for diffusion in random media, i.e.,
a
lim (v(t) v(0)) = - - (76)
"I re
where a and are density-dependent parameters. Inserting this expression into
Eq. (71) gives
DfM(t)= DfM + - >'(

1) tP-1
(77)
in which the time t is significantly larger than the time required for the VCF to
pass through its minimum (for theresults shown in Fig. 14, the power law decay
of the VCF is seen to characterize the long-time behavior for times greater than
approximately 15 average collision times). It has been assumed in this integra-
46 MAcELROY
9 = 0.500
o.21'\L
0
-0.21
0 1
2
-0.035
IC3
8-
6-
2-
U
10-4-
' 8-
9=0.500
t"
Normalized velocity autocorrelation function for the random overlapping

system. (a) Short-time behavior for several porosities. (b) The long-time tail for$ = 0.5.
MD simulation; (-) fit a t*-" power law decay [the reduced timet* is in units
of the mean free time T~ = 3D&', where D,, is given by Eq. (75)]. (- - -) Mode-
coupling theory Emst et al. (1984) and Machta et al. (1984).
47
DIFFUSION IN HOMOGENEOUS MEDU
I I I l
1 18 X)
14 Continued (c) As in (b), but for = 0.367. (d) The long-time tail for
= 0.035. MD simulation. (-) leastsquares fit to a powerlawdecay (p =
1.57). [Reproduced from Park and MacElroy (1989), with permission.]
48 MAcELROY
tion that p # 1, and this is indeed the case for the three-dimensional media under
consideration here. [A novel perspective of the influence of the power law ex-
ponent on the long-time tail of the VCF on diffusion in random media is con-
sidered by Muralidhar et al. (1990)l. For high-porosity media and small gas
particles, the power law exponent f3 is 2.5 (this result is generally true at low
densities for a class of random media known as Lorentz gases [see Ernst et al.
(1984) and Machta et al. (1984)], which includes each of the models shown in
Fig. ll), while for conditions at or near the percolation threshold the value of
p is 1.57 according to the simulation results given in Fig. 14d. Although the
latter result does not have the same generality as the high-porosity value of 2.5,
it can be related to a set of “universal” constants (critical exponents) via scaling
theory. In the vicinity of the percolation threshold the time-dependent diffusion
coefficient predicted by scaling theory at long times takes the general form
(Havlin and Ben-Avraham, 1987)
where
6 = cI./(2v + cI. - y) (79)
and $c is the critical voidage below which the penetrant is localized ($c = 0.035
for the random overlapping system as shown in Fig. 12). The parameters k, v,
and y are the critical exponents appearing in the expressions
-
and
P($) ($ - $JY ($ $c+) (82)
where 5 is the correlation length and P($) is the percolation probability, which,
in the present case, is equivalent to the relative magnitude of the percolating
void fraction, $”, to the void fraction $ (results for this quantity are provided
in Fig. 12 for the random overlapping spheres model). The function +(x) in Eq.
(78) has the following properties:
The limiting condition in Eq. (83a) applies above the percolation threshold and
leads to Eq. (80) on substitution into Eq. (78). Below the percolation threshold,
Eq. (83b) provides DM(t) = l/t, while at the transition [Eq. (83c)], Eq. (78)
DIFFUSION IN HOMOGENEOUS MEDLA 49
simplifies to Eq. (77) with the stationary-state diffusion coefficient DTM= 0 and
p-1=s.
At high porosities and/or for small diffusing particles, the comparatively large
value of p results in a rapid disappearance of the long-time tail on the VCF,
and hence this tail does not seriously interfere with the assumed validity of the
stationary form of the Fickian diffusion flux. This is not true, however, when
the diffusing particle is severely hindered in its motion and its time-dependent
diffusion coefficient is described by the scaling form in Eq. (78). Under these
conditions the power law exponent on the VCF is generally found to be much
smaller than 2.5, and the very slow decay of the VCF raises serious questions
concerning the applicability of simple constitutive forms of the type provided
in Eq. (38) [or indeed the more general case of Eqs. (31) and (32)]. The con-
clusion to be drawn from these observations is that for conditions that give rise
to anomalous diffusion in homogeneous media (as this non-Fickian behavior is
generally called) it is necessary to generalize Eq. (31) to a time-dependent con-
volution form as implied by the zero wavevector limit of Eq. ( l l ) , i.e.,
which simplifies to
for a rigid random medium. The q,(t) are the VCFs defined in Eq. (14b).
It is noteworthy that modified flux equations similar to Eq. (84) have been
proposed for rigid media (Lorentz gases) of the type shown in Fig. 11 (Alley,
1979) and for glassy polymers [see Neogi (1983a,b) and Chapter 5 of this text].
The analysis for Lorentz gases described by Alley (1979) is of interest from the
molecular point of view in that it employs the concept of a waiting time distri-
bution (Montroll and Weiss, 1965) in a continuous-time random walk theory for
the free-particle diffusion process [see also Klafter et al. (1986)l. At each step
in the random walk the diffusing particle is assumed to be subject to a waiting
time that results from partial entrapment in holes and dead-end pores within the
medium, and this mechanism is not unrelated to the view that penetrant diffusion
in glassy polymers is governed primarily by a sequence of activated jumps in
which slow segmental motion and relaxation of the polymer chains can retard
the jump frequency. The modified form of Fick's second law derived by Alley
(1979) is, in one dimension,
50 MAcELROY
where q(t) is the single-particle VCF (v'"(t) v'"'(O)), which is related to the
waiting time distribution function via the Laplace transform
where ( P ) is mean square jump distance. The coefficient D is equal to Dm,and

D4and higher order terms are known as the Burnett coefficients. In the zero
wavenumber limit, the Burnett terms may be neglected.
Equation (85) was verified by Alley (1979) via molecular dynamics simu-
lation for a two-dimensional random overlapping hard disk Lorentz gas, and
recently (MacElroy and Tomlin, 1992; MacElroy, 1992, unpublished), its valid-
ity was further confirmed for the three-dimensional analog this system (Fig.
lla). The time-dependent behavior suggested by Eq. (77) was also observed in
molecular dynamics studies adsorbing Lennard-Jones fluids [Eq. (46)] con-
fined within the micropores of a model silica medium (MacElroy and Raghavan,
1990, 1991),and oneexample of this is providedinFig. 15 for the tracer
coefficient D,.(t),
The results shown in Fig. 15 are for a liquid-filled void space in a model non-
overlapping connected spheres medium similar to that shown in Fig. l l d . In
this case each of the solid (silica) spheres is atomistically modeled as illustrated
in Fig. 16 [for details see MacElroy and Raghavan (1990)], with the fluid par-
ticles interacting with the individual atoms the solid via a Lennard-Jones (12-
6) potential. The relative size, A, of the fluid particles and the pores within the
medium in these studies is approximately 0.2, and therefore anomalous effects
associated with the percolation threshold are not evident here.There is, however,
a significant long-time tail on the VCF as illustrated in the inset Fig. 15.
Finally, to return a point madeearlier with regardto the relative importance
of the cross-kinetic coefficients L,(i # j ) , results also provided by MacElroy and
Raghavan (1991) for sorbed vapor diffusion over the entire range of pore-filling
conditions are reproduced in Fig. 17. The filled circles in this figure correspond
to the tracer diffusion coefficient
D,.= 3I (v&) - v,.(O)) dt

D**@*)
"
0.2 2
Tiie-dependent tracer diffusion coefficient for a model porous silica me-

dium saturated with a Lennard-Jones liquid.).( Results obtained from integration of the
VCF; (0)results obtained from the long-time slope of the mean-square displacement [Eq.
(14c) at finite times]. The inset shows a power law fit of the normalized VCF for the
tracer, and the solid lines in both the main figure and the inset are for = 1.8 2 0.1 t*
is in units of &G, and Dl.(t*) is in units of I s i o G f , where is the mass of
the Lennard-Jones particles, ernis the potential well depth for interactions between the
fluid particles and the oxygen atoms of the silica medium, and lsiois the Si0 bond length
insilica (0.162 nm). [Reproduced from
MacElroyand
Raghavan (1991), with
permission.]
52 MAcELROY
16 simulated silica microsphere. Each of the solid spheres shown in Fig.

l l d is modeled as one of these. The open circles are nonbridging surface oxygens, and
the shaded region illustrates the exposed interior bridging oxygens. [Reproduced from
MacElroy and Raghavan with permission.]
and the open circles correspond to the results for the total sorbate diffusion
coefficient
The inset in Fig. 17 also provides estimates of the relative magnitude of the
cross-coupling effects in terms of the ratio (D, - Dl.)/Dm.The numerator in
this ratio is simply
Note that Dl. and D , are equal only in the limit -D 0 and that at any nonzero
concentration cross-effects are always present. The results plotted in the inset
of Fig. 17 demonstrate that these cross-effects can contribute as much as 70%
0.8
? 0.6
0.002 D
0 0.2 0.2
17 Reduceddiffusioncoefficients of theadsorbingvaporas a function of

fractional pore loading. Tracer diffusion coefficient; (0)total adsorbate dif€usion co-
efficient; (-) EnskogtheoryMacElroyandKelly (1985). The diffusion coefficients
are in units of Z s i , , G f . The inset illustrates the relative importance cross-effects
inthemicroporefluid.[ReproducedfromMacElroyandRaghavan with
permission.]
54 MAcELROY
to the overall diffusion coefficient at saturation for the system considered by

MacElroy and Raghavan (1991).
If the microcavities in an amorphous polymeric medium are significantly larger

than the penetrant molecules, then the assumption of solid-phase rigidity should
not be an unreasonable approximation. However, for large solute species and/
or high polymer densities, the diffusion mechanism is widely considered to be
governed by the formation of holes via local motion of the macromolecular
chains, and it is clear that under these conditions the dynamics of the chain
segments should be included in the overall model. The flux equations now in-
clude the kinetic coefficients L,, and the diffusion velocity of the polymer, J,,
is not zero (swelling or shrinkage may be observed) unless external constraints
are imposed on the system. For a fluidholid system that is not subject to external
forces, conservation of total linear momentum within the medium at any given
instant requires
where m, and v,(t) may be considered to be the mass andcenter-of-mass

velocity, respectively, of a polymer chain, a monomer unit, or the individual
atoms of the macromolecule [in which case an additional sum over atomicspe-
cies is implied onthe left-hand side of (90)]. The kinetic coefficients defined
in Eq. (14) are then interrelated by
and Eq. (31) may be rewritten as
Modeling of single-component diffusion has been of primary concern in the

literature, and for this reason the discussion below is restricted to such systems.
In this case (92) simplifies to
with Dm again given by Eq. (89). This coefficient quantifies the collective dif-
fusion of the penetrant within the medium, and as noted earlier it is collective
properties of this type that need to be addressed in the engineering design of
DIFFUSION IN HOMOGENEOUS MEDLA 55
sorption or membrane separation processes. Another diffusion coefficient that

may be defined for single penetrant systems is the tracer or se!f-diffusion co-
efficient, and it is this parameter that is most frequently reported in MD simu-
lation studies (and measured experimentallyvia NMR) of diffusion in polymers.
Using Eq. (92), it is readily shown that the tracer diffusion flux is given by the
vector form of Eq. (66), i.e.,
J1. = - D,.Vn,. (94)
with the tracer diffusion coefficient, Dl., given by (88).
One of the major difficulties involved in the simulation of polymers is the
availability (or lack a reliable model for the potential interactions arising not
only between the penetrant and the polymer macromoleculesbut also, and most
important, the intramolecular interactions within the polymer chains themselves.
To date essentially all of the molecular dynamics simulation studies conducted
on diffusion in polymers have modeled the polymer chains as alkane structures
[polyethylene (Trohalaki et al., 1989, 1991; Sonnenburg et al., 1990; Takeuchi
andOkazaki, 1990; Takeuchi, 1990a,b; Takeuchi et al.,1990),polypropylene
(Muller-Plathe, 1992), and polyisobutylene (Muller-Platheet al., 1992)] withthe
exception of the recent work of Sok et al. (1992). The simplifying features of
the alkane systemlie in the absence of polar or electrostatic interactions and the
observation that the interactions between nonbonded atoms (-C- or -H) or
sites (“CH, or - C H r)on the chains are adequately represented by the (short-
range) Lennard-Jones potential[Eq.(46)]. The potential functions associated
withintrachain vibrations and rotations are also frequently described by the
comparatively simple forms
kb
v
b = - (b - bo)*
2
and
<
V, = k, 2
n=O
a, COS” (+)
where b is the bond length between two neighboring atoms or sites (with an
equilibrium value of b,), 8 is the angle between two adjacent bonds (with its
equilibrium value equal to and +is the dihedral angle definedby three
successive bond vectors in the polymer chain. [A concise description of these
potential functions andthe equations for the forces onagivenatom or site
arising from these interactions may be found in Appendix C of the text by Allen
and Tildesley (1987).]Othermathematical expressions reproducing the same
56 MAcELROY
general functional behavior have been employed for the torsional potential V,
primarily as a matter of computational convenience.
Molecular dynamics simulationsof polymer/penetrant systems, in which the
penetrants have also usually been modeled as Lennard-Jones fluids [Eq. (46)],
havebeenconductedmost frequently in one of possible ways: (1) The
trajectories of each atom and/orsite in the system are computed using Newton’s
equations as written in Eqs. (47a) and (47b) for unconstrained motion with the
forces acting on particle or site i determined by the pairwiseLennard-Jones
interaction potentials and the potential functions (or equivalent forms) provided
in Eqs. (95), or (2) the lengths of the bonds within the polymer chains are fixed
[thus eliminating Eq. (95a) and reducing the number of degrees of freedom by
the number ofbonds constrained] by incorporating constraint forces in Newton’s
equations of motion that act along the bonds.
Conceptually and computationally the unconstrained formulationis the easier
ofthe two approaches, although difficulties arise owing tothemuchhigher
frequencies usually involved in the bond vibrations in contrast to the compar-
atively low frequency of the torsional rotations. overcome this, realistic es-
timates of the spring constant kb are reduced by a factor of 7 (Rigby and Roe,
1987) and the equations of motion are solved using finite-difference methods of
thekindgiveninEq.(50) with comparatively large timesteps. The relative
importance of the assumption involved in the lower kb value for diffusion of
penetrants in polymers has not been quantitatively assessed as yet, and this has
raised questions concerningthe reliability of the activation energies for diffusion
determined via unconstrained molecular dynamics (Trohalaki et al., 1991).
The second method, involving constraint dynamics, makes the not unreason-
able assumption that, in view of the very high frequency of the bond vibrations,
the bond lengths over moderate time scales are effectively constant (ina number
of versions of this method the bond angles, 8, are also fixed). The method was
first proposed in the 1970s (Ryckaert et al., 1977), and improvements and var-
iants on the original technique weresummarized by Allen and Tildesley (1987).
In effect, at the beginning of a given time step the atoms or sites of the polymer
chain are assumed to move along free, unconstrained trajectories subject only
to intermolecular forces [derived from Eq. (46)] and intramolecular forces [de-
rived from Eqs. (95b) and (95c)l. At the end of the time step, the system is
“shaken” by imposing the constraint forces in an iterative manner,and the
instantaneous configurational and dynamical phasecoordinates of the chains are
obtained to within a given tolerance. This algorithm (the original form of which
is called SHAKE) ormodifications of it have been successfullyusedinthe
simulation of both small molecular structures and complex macromolecules.
Although molecular simulations have been undertaken over the last 15 years
or for a variety of polymer systems [see Roe (1991) and references cited
therein], it is only very recently that the firstMD simulation results for diffusion
MEDIA 57
in polymers have been reported (Trohalaki et al., 1989, 1991; Sonnenburg et

al., 1990; Takeuchi and Okazaki, 1990;Takeuchi, 1990a,b; Takeuchiet al., 1990;
Muller-Plathe, 1992; Muller-Plathe et al., 1992; Sok et al., 1992). In the first
few of these studies (Trohalaki et al., 1989,1991;Sonnenburget al., 1990;
Takeuchiand Okazaki,1990; Takeuchi,1990a,b;Takeuchi et al., 1990), the
unconstrained MD technique described above (method 1) was employed, and
the polymer investigated was amorphouspolyethylene. The simulations reported
by Trohalaki et al. (1989, 1991), Sonnenburg et al. (1990), Takeuchi and Oka-
zaki (1990), Takeuchi (1990b), and Takeuchi et al. (1990) were carried out at
conditions well above the glass transition temperature, Tg,and it was found that
even for a system of “polyethylene” chains containing as few as 20 methylene
units, the computed self-diffusion coefficients for a number of rare gases, 02,
and CO2 were in fair agreement with experiment. Furthermore, it was shown
that the degree of chain flexibility and, in particular, intramolecular torsional
rotations was a controlling factor in the rate of penetrant diffusion and the ap-
plicability of free-volume theory(Cohenand Turnbull, 1959) andmolecular
models (Pace and Datyner, 1979a,b,c) to both diffusion and polymer relaxation
processes was confirmed. l h o systematic effects were observed, however: (1)
The diffusion coefficients determined via simulation were approximately three
to four times larger than those observed experimentally and (2) in a number of
cases the activation energies were a factor of 2 lower than the experimental
values. In one study (Trohalaki et al., 1991) the latter effect was associated with
density differences between the simulated chain system and the experimental
density of amorphous polyethylene; however, in the work reported by Takeuchi
and Okazaki (1990), densities consistentwith experimental values were em-
ployed, and it was shown that the source of error in the estimation of the acti-
vation energies lay with the simulation method itself in that constant-pressure
simulations (the NpEensemble)ratherthanconstant-volume conditions (the
usual NVE ensemble) are required. In an effort to rationalize the systematically
higher diffusivities obtained irrespective of the ensemble considered, Takeuchi
(1990b) investigated the effects of the chain length of the “polyethylene” mol-
ecules and found that for chains of infinite length (in contrast to the (&chains
considered in prior studies) only a marginal improvement in the simulation re-
sults could be obtained. He suggested that the experimental diffusivities may be
influenced by the degree of crystallinity of polyethylene, i.e., crystalline inclu-
sions embedded in the amorphous region in a manner analogous to the models
illustrated in Fig. 11 (with the solid spheres corresponding to the crystallites)
couldlead to asignificantreductionintheself-diffusioncoefficientof the
penetrant.
To clarify the possible effects of crystallinity as well as a number of other
issues, Muller-Plathe (1992) conducted MD simulations for a number of pene-
trants (H2, 02,and C&) diffusing in atactic polypropylene above the glass tran-
58 MAcELROY
sition temperature. This polymer is known to be 100% amorphous, and therefore

any questions concerning crystallinity areavoided.Furthermore, the polymer
model investigated by Muller-Plathe (1992) was semiatomistic (improving on
the structureless site model for the methyl and methylene groups employed in
the earlier polyethylene simulations), and the fixed bond length constraint dy-
namics method described above was employed (thus eliminating the problem
associated with the weak spring constant kb employed in the unconstrained MD
method). As additional safeguards against the introduction of extraneous errors,
Muller-Plathe simulated a very long polymer chain at the appropriate bulk den-
sity of amorphous polypropylene and employed carefully equilibrated polymer
configurations to initiate the trajectories [see also Theodorou and Suter (1985)l.
Although no direct comparison could be made with experiment because of
the absence of data, the self-diffusion coefficients obtained by Muller-Plathefor
O2 and CH, were in good agreement with experimental results for a number of
other polymeric materials. This was particularly true for C&, for which the
simulated and experimental diffusivities were in very close agreement. However,
for the H,/polypropylene system the simulation diffusion coefficient was an or-
der of magnitude higher than all available experimental data, and although sta-
tistical sampling in the MD simulation work may be a contributing factor the
reason for this disparity remains an open question.
One possible reason for the higher than expected diffusion coefficients some-
times observed in MD simulations is inferred from the very recent results re-
ported by Muller-Plathe et al. (1992) for oxygen diffusion in rubbery polyiso-
butylene (PIB).In this work it was clearlydemonstratedthat a united-atom
representation of the methyl groups in PIB [(-CH2-C(CH,)2-),] gave rise to
diffusion rates that were almost two orders of magnitude larger than the corre-
sponding rates in an all-atom model of the polymer. Although differences of
this order are generally larger than those usually encountered in comparative
studies of united-atom (smeared) and all-atom simulations, they are consistent
with documented results for other systems [e.g., compare Figs. 7 and 8; see also
MacElroy and Raghavan (1990) for comparative work on microporous silica].
Another significant outcome of additional simulations conducted by Muller-
Plathe et al. (1992) for helium tracer diffusion in PIB is illustrated in Fig. 18.
In this figure the ensemble-averaged mean square displacementof the He atoms
diffusing in an all-atom model of PIB is plotted as a function of time, and when
it is noted that the time-dependent diffusion coefficient defined in Eq. (87) may
also be expressed, using Eq. (14c), as
where rl.(t) is the position of the tracer particle at time t, then it is clear from
the short-time behavior shown in Fig.18 that the heliumparticles are undergoing
MEDIA
MOGENEOUS
INDIFFUSION 59
s
L
.t;
*
L
-
W
-2.0 t
-1 .o 0.0 1.o 2.0 3.0 4.0
m,,WPs)
Mean-square displacement of helium atoms in amorphous polyisobutylene

obtained via MD with an all-atom force field for the polymer. [Reproduced from Muller-
Plathe et al. with permission.]
anomalous diffusion. It is also of interest to note that the exponent in the

anomalous regime [see Eqs. (76) and (77)] is found to be 1.50, which compares
very favorably with the exponents reported by Park and MacElroy (1989) for
diffusion in rigid random media [a similar observation was made by Muller-
Plathe etal. with reference to the stochastic MC simulations carried outby
Gusev and Suter (1992) for helium diffusion in static polycarbonate structures].
At long times the results for the mean square displacement shown in Fig. 18
assume a linear (Fickian) relationship with time.
In addition to the diffusion coefficients themselves, the MD simulation data
provide a wealth of information on the microscopic properties of polymer/pen-
etrant systems, and an interesting perspective onthe diffusion mechanism is
provided by the results reported by Muller-Plathe (1992) for the time-dependent
behavior of the magnitude of the relative displacement of a given penetrant
molecule fromits initial position Ir(t) - r(0)l. His published results for the three
penetrants H,, O,, and CH., are reproduced in Fig. 19, and the jumps observed
for methane (and to a lesser extent oxygen) confirm the hopping mechanism
postulated for diffusion in rubbery polymers. The small size of the hydrogen
60 MAcELROY
t @S)
Relativedisplacementsofrepresentativemolecules of H,, O,, and CH,
diffusing in atactic polystyrene. For clarity, the ordinate axes for 0, and H, are offset by
1 nm and 2 nm, respectively. [Reproduced from Muller-Plathe (1992), with permission.]
molecule results in a fluidlike behavior with a comparatively short time involved .

between jumps from onemicrocavity to neighboring holes. Similar observations
were reported by Sok et al. (1992) for He (fluidlike tracer diffusion) and CH.,
(latticelike tracer diffusion) in rubbery polydimethylsiloxane(PDMS), and it is
also worth noting that in the case both penetrants very good agreement was
found between the simulation and experimental results for the tracer diffusion
coefficients.
Each of the above-cited studies has addressed self-diffusion in polymer sys-
tems above the glass transition temperature, and it is believed that comparable
MD studies below T, are not possible at the present time. In only one case has
an attemptbeenmade to elucidate the transport mechanism below the glass
transition temperature (Takeuchi, 1990a) where it was shown that although rigid
cage structures are a predominant feature of polymeric materials below T, it is
possiblethat the (slow)thermalmotion local chain segments can induce
changes in the shape of the microcavities in the system without changing the
free volume. This is illustrated in Fig. 20, where the sequence events taking
place during the jump of an oxygen molecule in a model polyethlene medium
is clearly demonstrated (note that the time origin indicated in this figure is not
DIFFUSION IN HOMOGENEOUS MEDU 61
, I
1=6.0 PS 1112.1 PS 143.3 1=16.1 ps
1=10.1 PS 1112.5 PS 143.7 PS
1=11 .l PS 1~14.1PS
Figure 20 The temporal evolution of thc contours of the potential energy surface for
an oxygen molecule as it jumps from one microcavity to a neighboring one in glassy
polyethylene. [Reproduced from Takeuchi (IgSOa), with permission.]
MAcELROY
the origin of the trajectory itself; the true origin was 70 ps earlier). Under fa-
vorable conditions the shapes of neighboring cavities are distorted sufficiently
to form an interconnecting channel throughwhichthe oxygen molecule can
easily diffuse. The jump actually occurs between steps (d) and (g), and it is
important to observe that the potential energy of the oxygen molecule is un-
changed as it moves from one cavity to the next, i.e., the oxygen molecule itself
is not subject to anysignificantenergybarrier.Unfortunately, a preliminary
analysis of the conformational dynamics of the local chain motion during the
jump failed to reveal the underlying cause for the formation of the channel, and
further work is needed to clarify the mechanism involved.
W. CONCLUDING COMMENTS
It is now possible to formally express the diffusion equations for fluids in ho-
mogeneous media in terms of the microscopic properties of the fluid/solid sys-
tem. Difficulties arise, however, when one attempts to extract mathematically
tractable results for use in engineering design. Approximate theories based on
free-volume and molecular concepts are currently employed for this purpose
although it is recognized that future developments in this area will rely heavily
on an improved understanding of the molecular phenomena involved.
In this chapter one technique,molecular dynamics simulation,which has
contributed significantly to our understanding liquid and solid behavior over
the last four decades, has been described in some detail, and a number of ap-
plications for simple fluids confinedwithin idealized micropores, random media,
and polymers have been discussed. The MD method can serve as an alternative
and complementary means for investigating diffusional behavior in fluid/solid
systems, and in view of its intrinsic microscopic character it can provide valu-
able information concerning molecular properties that cannot be readily mea-
sured experimentally.
Although current computational facilities would appear to limit the applica-
tion of the MD technique to dif€usion in rubbery polymers, in the next chapter
of this book complementary methods employing both MD and stochastic sim-
ulation techniques that may extend the scope of computational nonequilibrium
statistical mechanics to the glassy state are discussed. One of the most chal-
lenging aspects of transport in polymers that still needs to be addressed is the
molecular mechanism@) thatgive rise to anomalous and case II diffusion below
the glass transition temperature. MD simulation of diffusion in homogeneous
random media at or near a percolation threshold demonstrates the importance
of low-frequency molecular relaxation phenomena that can influence the con-
stitutive form of the flux equations over macroscopic time scales. This may have
a direct bearing on the non-Fickian character the flux equations in glassy
polymers.
DIFFUSION IN HOMOGENEOUS MEDU 63
Abassi, M. H., J. W. Evans, and I. S. Abramson (1983). AZChE J., 29, 617.
Alder, B. J., and T. E. Wainwright (1970). Phys. Rev., Al, 18.
Allen, M.P.,andD. J.Tildesley(1987). ComputerSimulation of Liquids, Clarendon
Press, Oxford.
Alley, W. E. (1979). Studies in molecular dynamics of the friction coefficient and the
Lorentz gas, Ph.D. Thesis, Univ. California, Davis.
Altenberger, A. R., J. S. Dahler, and M. Tirrell (1987). J. Chem. Phys., 86, 2909.
Belter, P.A., E. L. Cussler,andW.-S.Hu(1988). Bioseparations-Downstream Pro-
cessing for Biotechnology, Wiley, New York.
Brown, P. R., and R. A. Hartwick, Eds. (1989). High Performance Liquid Chromatog-
raphy, Wiley, New York.
Chapman, S., and T. G.Cowling(1970). TheMathematicalTheoryofNon-Uniform
Gases, 3rd ed., Cambridge Univ. Press, London.
Chase, H. A. (1984a). Chem. Eng. Sci., 39, 1099.
Chase, H. A. (1984b). J. Chromatogr, 297, 179.
Chiew, Y. C., G. Stell, and E. D. Glandt (1985). J. Chem. Phys., 83, 761.
Cohen, M. H., and D. 'hrnbull(l959). J. Chem. Phys., 31, 1164.
Crank, J., and G. S. Park, Eds. (1968). Diffusion in Polymers, Academic, New York.
Davis, H. T. (1987). J. Chem. Phys., 86, 1474.
deGroot, S. R., and P. M m r (1963). Non-Equilibrium Thermodynamics,North-Holland,
Amsterdam.
Drioli, E., and M. Nakagaki, Eds. (1986).Membranes and Membrane Processes,Plenum,
New York.
Emst, M.H., J. Machta, J. R. Dorfman, and H. van Beijeren (1984). J. Stat. Phys., 34,
477.
Ferry, J. D. (1936). J. Gen. Physiol., 20, 95.
Fischer, J., and M. Methfessel (1980). Phys. Rev., A22, 2836.
Gotze, W., E. Leutheusser, and S. Yip (1981a). Phys. Rev., A23, 2634.
Gotze, W., E. Leutheusser, and S. Yip (1981b). Phys. Rev., A24, 1008.
Green, M. S. (1952). J. Chem. Phys., 20, 1281.
Green, M. S. (1954). J. Chem. Phys., 22, 398.
Gusev, A., and U. W. Suter (1992). Polym. Prepr, ACS Division of Polymer Chemistry,
Washington, DC, 33, 631.
Haan, S. W., and R. Zwanzig (1977). J. Phys. A: Math. Gen., 1547.
Hanley, H. J. M., Ed. (1969). Trunsport Phenomena in Fluids,Marcel Dekker, New York.
Havlin, S., and D. Ben-Avraham (1987). Adv. Phys., 36, 695.
Heinbuch, U.,and J. Fischer (1987). Mol. Simul., 1, 109.
Hirschfelder, J. O., C. F. Curtiss, and R. B. Bird (1954). The Molecular Theory of Gases
and Liquids, Wiley, New York.
Ho, F. G.,andW. Strieder (1980). J. Chem. Phys., 73, 6296.
Juntgen, H., K. Knoblauch, and K Harder (1981). Fuel, 817.
Kennard, E. H. (1938). Kinetic Theory of Gases, McGraw-Hill, New York.
Kim, W.-T., A. R. Altenberger, and J. S. Dahler (1992). J. Chem. Phys., 97, 8653.
Kirkpatrick, S. (1973). Rev. Mod. Phys., 45, 574.
64 MAcELROY
Klafter, J., R. J. Rubin, and M. F. Shlesinger, Eds. Transport and Relaxation in

Random Materials, World Scientific, Singapore.
Kubo, R. J. Phys. Soc. Jpn. 12,
Kubo,R.,M.Toda,andN.Hashitsume Statistical Physics 11. Nonequilibrium
Statistical Mechanics,Springer-Verlag, Berlin.
MacElroy, J. M. D., and J. J. Kelly AIChE J., 31,
MacElroy, J. M. D., and K. Raghavan J. Chem. Phys., 93,
MacElroy, J. M. D., and K.Raghavan J. Chem. Soc. Faraday Trans., 87,
MacElroy, J. M. D., and S.-H. Suh Mol. Phys., 60,
MacElroy, J. M. D., and S.-H. Suh Mol. Simul., 2,
MacElroy, J. M. D., and B. Tomlin Proc. 3rd ZChemE Res. Symp., Dublin,
Machta, J., M. H. Emst, H. van Beijeren, and J. R. Dorfman J. Stat. Phys., 35,
Magda, J. J., M. Tirrell, and H. T. Davis J. Chem. Phys., 83,

Mason, E. A., and S. Chapman J. Chem. Phys., 36,
Mason, E. A., and A. P. Malinauskas Gas Transport in PorousMedia: The Dusty
Gas Model, Elsevier, Amsterdam.
Mason, E. A., and L. A. Wehland J. Chem. Phys., 68,
Mason, E. A., R. B. Evans 111, and G. M. Watson J. Chem. Phys., 38,
Masters, A., and T. Keyes Phys. Rev., A26,
Maxwell, C. Treatise on Electricity and Magnetism, Vol. I, Oxford Univ. Press,
London.
Montroll, E. W., and G. H. Weiss J. Math. Phys., 6,
Mori, H. Prog. Theor. Phys., 33,
Muller-Plathe, F. J. Chem. Phys., 96,
Muller-Plathe, F., S. C. Rogers, and W. F.van Gunsteren Chem. Phys.Lett., 199,
Muralidhar,R.,D.Ramkrishna,H.Nakanishi,and D. Jacobs Physica A , 167,
Nakano, Y., and J. W.Evans J. Chem. Phys., 78,

Neogi, P. AZChE J., 29,
Neogi, P. AZChE J., 29,
Nicholson, D. J. Chem. Soc. Faraday Trans. Z, 71,
Nicholson, D., and N.G. Parsonage Computer Simulation and the Statistical
Mechanics of Adsorption, Academic, New York.
Nicholson, D., J. K Petrou, and J. H. Petropoulos Chem. Eng. Sci., 43,
Norde, W. Adv. Colloid Znterface Sci., 25,
Pace, R. J., and A. Datyner J. Polym. Sci., Polym. Phys. Ed., 17,
Park, 1.-A., and J. M. D. MacElroy Mol. Simul., 2,
Peterson, B. K., and K E. Gubbins Mol. Phys., 62,
Peterson, B. K., K. E. Gubbins, G. S. Heffelfinger, U. Marini Bettolo Marconi, and F.
van Swol(l988). J. Chem. Phys., 88,
Prager, S. Physica, 29,
MEDIA
MOGENEOUS
INDIFFUSION 65
Prigogine, I. Introduction to the Thermodynamics of Irreversible Processes,2nd

ed., Wiley-Interscience, New York.
Raghavan, K., and J. M. D. MacElroy Mol. Simul., 8,
Reyes, S., and K F. Jensen Chem. Eng. Sci, 40,
Rigby, D., and R. J. Roe J. Chem. Phys.,
Roe,R.J.,Ed. Computer Simulation of Polymers, Prentice-Hall,Englewood
Cliffs, NJ.
Ruthven,D.M. Principles of Adsorption and Adsorption Processes, Wiley-
Interscience, New York.
Ryckaert, J. P., G. Ciccotti, and H. J. C. Berendsen J. Comp. Phys.,
Sahimi, M. Chem. Eng. Sci., 43,
Satterfield, C. N. Heterogeneous Catalysis in Practice, McGraw-Hill, New York.
Satterfield, C. N., C. K. Colton, and W. H. Pitcher, Jr. AIChEJ., 19,
Schoen, M., J. H. Cushman, D. J. Diestler, and C. L. Rhykerd, Jr. J. Chem. Phys.,
Shante, V. K. S., and S. Kirkpatrick Adv. Phys.,

Sirkar, K. K, and D. R. Lloyd, Eds. New Membrane Materials and Processes
for Separation, AIChE Symp. Ser., No.
Sok,R.M.,H. C. Berendsen, and W.F. van Gunsteren J. Chem. Phys., 96,
Sonnenburg, J., J. Gao, and J. H. Weiner Macromolecules, 23,

Steele, W. A. The Interaction of Gases with Solid Surfaces, Pergamon, Oxford.
Stem, S. A., and H.L. Frisch Annu. Rev. Mater. Sci., 11,
Suh, S.-H. and J. M. D. MacElroy Mol. Phys., 58, 445.
Swope, W.C., H. C. Andersen, P. H. Berens, and K.R. Wilson J. Chem. Phys.,
Takeuchi, H. J. Chem. Phys., 93,

Takeuchi, H. J. Chem. Phys., 93,
Takeuchi, H.,and K Okazaki J. Chem. Phys., 92,
Takeuchi, H.,R. Roe, and J. E. Mark J. Chem. Phys., 93,
Theodorou, D.N.,and U.W. Suter Macromolecules,
Torquato, S. J. Stat. Phys., 45,
Trohalaki, S., D. Rigby, A. Kloczkowski, J. E. Mark, and R. J. Roe Polym. Prepr.,
Trohalaki, S., A. Kloczkowski, J. E. Mark, D. Rigby, and R. J. Roe In Computer

Simulation of Polymers, R. J. Roe, Ed., Prentice-Hall, Englewood Cliffs, NJ, Chap.
Turbak, A. F.,Ed. Synthetic Membranes, Vols. I and 11, AmericanChemical

Society, Washington, DC.
van Leeuwen, M. J., and A. Weijland Physica, 36,
Vertenstein, M., and D. Ronis J. Chem. Phys., 85,
Vertenstein, M., and D. Ronis J. Chem Phys.,
Vieth, W. R. Diffusion In and Through Polymers, Hanser, Munich.
Weijland, A., and J. M. J. van Leeuwen Physica, 38,
Weissberg, H. L. J. Appl. Phys., 34,
MAcELROY
Weisz, P. B. (1973). Chem. Tech., 498.

White, R.E., andP. N. Pintauro, Eds. (1986). Indush’al Membrane Processes, AIChE
Symp. Ser., No. 248, 82.
Yang, R. T. (1987). Gas Separation by Adsorption Processes, Butterworths, Boston.
Yau,Y. Y., J. J. Kirkland,and D.D. Bly (1979). Modern Size-Exclusion Liquid Chro-
matography, Wiley, New York.
Zhang, L., and N. Seaton (1992). AIChE J., 38, 1816.
2
Molecular Simulations of Sorption and
Diffusion in Amorphous Polymers
Doros N. Theodorou
University of Patras
Patras, Greece
INTRODUCTION
A. Scope of Molecular Simulations
Many technologically importantprocesses rely upon the design of polymers with
tailored characteristics of permeability and selectivity toward fluid molecules.
Examples include gas separation with polymeric membranes, food packaging
using plastic films, and encapsulation of electronic components in polymers that
act as bamers to atmosphericgases. To solve the polymer design problem suc-
cessfully, one needsto relate the chemical constitution of the polymer (setduring
synthesis) and its morphology (set during processing) to the sorption isotherms
and diffusivities of fluid molecules within it. There are several ways of estab-
lishing quantitative relations between polymer structure (chemical constitution
andmorphology)and sorption andpermeation properties. One wayis direct
experimental measurement; a second way is the formulation and use of phe-
nomenological correlations based on experimental evidence froma wide variety
of systems. third way to structure-property relations, which constitutes the
main focus of this chapter, is the development and application of theories and
simulation techniquesthat rely directly upon fundamental molecular science.
Theories start from a more or less detailed model of the polymer/penetrant
system and proceed to derive functional expressions for sorption isotherms or
diffusivities through statistical mechanical analysis aided by the introduction of
judicious approximations. Examples of theories are the lattice fluid theory of
68 THEODOROU
gas solubility inpolymermelts, the dual-modesorptiontheory for polymer

glasses, and activation energy and free-volume approaches to the diffusivity of
pure or mixed gases in polymers. excellent review of theoretical work on
sorption and diffusion in polymers has been given by Petropoulos (1994). A
theory is typically cast in closed form as a set of algebraic, integral, or differ-
ential equations into which the chemical “personality” of the polymer and pene-
trant enter through a handful of parameters. To afford such a closed formulation,
one typically has to introduce simplifying approximations in the mathematical
treatment or invoke relatively crude models for the microscopic representation
of the system at hand. Computer simulations, on the other hand, can be thought
of as numerical solutions of the full statistical mechanics given a model for the
molecular geometry and interaction energetics for all molecular species present.
Simulations, or “computer experiments,” involve the generation of configura-
tions representing a particular material system, from which structural, thermo-
dynamic, and transport properties are estimated. In principle, simulations can
provide “exact” results for a given model representation of the polymer/pene-
trant system. In practice, compute time considerations necessitate the introduc-
tion approximations in simulation work as well, but typically these approx-
imations are less severe than the ones invoked in theories.
Molecular simulations can improve our ability to describe and predict sorp-
tion and diffusion phenomena in polymers in several ways. First, simulations
can elucidate the molecular mechanisms underlying the macroscopic behavior
of polymer/penetrant systems and thereby serve as a guide for the development
of better theories. Direct comparisons between the results of simulations and
theories based on exactly the same molecular representation are an excellent
way to assess the legitimacy and implications of the approximations invoked in
the theories. Second, simulations based on models thatcorrectly capture the
salient features of molecular geometry and energetics can helpidentify expected
changes in performance with changes in chemical constitution. This is valuable
for addressing ‘‘what if ” questions of a materials design nature without going
throughcostlytrial-and-error laboratory synthesis and testing of properties.
Third, simulations that rely on sufficiently detailed molecular models can ac-
tually be used for the quantitative prediction of solubility and diffusivity values
directly from chemical constitution. Although the field of polymer molecular
simulations is still at a stage ofmethoddevelopmentandvalidation against
experimental evidence from existing systems, there is every expectation that it
will develop into a powerful framework for the design of new materials.
In this chapter we review recent molecular simulation work relating to the
sorption and diffusion of gases in polymer melts and amorphous glassy poly-
mers. As practically all published simulation work to date has been performed
on relatively simple amorphous systems in which permeation is found macro-
scopically to proceed by a solution-Fickian diffusion mechanism, we focus on
MOLECULAR SI-TIONS OF SORPTION AND DIFFUSION 69
such systems. We briefly discuss the statistical mechanical principles that can
be used to construct simulation techniquesfor sorption and diffusion and sum-
marize what has been learned from the application of such techniques to date.
In the remainder of this introduction we discuss qualitatively some molecular
aspects of sorption and diffusion in polymers. Simulation techniques for char-
acterizing the internal structure and mobility of amorphous polymers are pre-
sented in Section II. Section I11 deals with the prediction sorption thermo-
dynamics; theproblem of calculating the sorption isotherminapolymer is
formulated in a general way, and existing applications are reviewed. The prob-
lem of predicting diffusivities is discussed in Section IV, both molecular dy-
namics simulations and transition state theory-based techniques for species that
move slowly through the polymer matrix are presented. Finally, Section V sum-
marizes some conclusionsand future directions.
B. Some Molecular Aspects of Sorption and Diffusion in

Amorphous Polymers
The solubility of a penetrant in a polymer is expected to depend on the nature
and magnitude of polymer-penetrant interactions in relation polymer-poly-
mer and penetrant-penetrant interactions, as well as on the distribution of shapes
and sizes of open spaces formed among chains within the polymer, where pene-
trant moelcules can reside. Thus, a prerequisite for predicting the solubility is
that one be able to describe the distribution of molecular configurations taken
on by the penetrant/polymer system. For melt systems in thermodynamic equi-
librium, this distribution of configurations is well established by equilibrium
statistical mechanics. Consider, for example, a pure melt of Nppolymer chains
occupying a volume V. The potential energy of this system is a highly convo-
luted function 'Tp(rp)of the coordinates of all atoms constituting the chains,
which we symbolizecollectively as a vector r,. For macroscopic specimens, the
dimensionality of the configuration space from which r, takes values is on the
order of Avogadro's number. In the schematic of Fig. 1 we display the entire
configuration space asa single axis, for graphical simplicity; the potential energy
hypersurface is thus plotted as a curve. When the temperature T is suf-
ficiently high for the polymer to be an equilibrium melt, every point in config-
uration space is visited with a frequency proportional to the Boltzmann factor
exp(-pV,), where = l/k,T. When NA molecules of a penetrant are present in
the polymer melt, the dimensionality of the configuration space is augmented
and the potential energy hypersurface 'T(rp, is modified. The different con-
figurations (r,, however, are again Boltzmann-distributed. From the point
of view of simulation, the solubility prediction problem in a gas/melt system is
thus conceptually analogous to the problem of predicting vapor-liquid equilib-
rium of low molecular weight mixtures, theonly difference being the involatility
70 THEODOROU
N. V Const.
Schematicrepresentationofthepotentialenergy V, asafunction of the

microscopic degrees of freedom r, for an amorphous polymer region of volume V con-
taining N p macromolecules. In the glassy state, the polymer configuration is confined to
fluctuate in the vicinity of local minima of the potential energy, as shown by the bold
arrows.
of the polymer and the formidable computational challenges associated with the
large-length-scale structure andlong-time-scalemotion of macromolecular
systems.
Glassy polymers, on the other hand, are not in thermodynamic equilibrium.
It is conceptually useful to think of the configuration in a pure polymer glass
as arrested, that it can fluctuate within a small region of configuration space
in the vicinity of a local minimum or of a relatively small set of local minima
of the potential energy. Such regions are displayed with bold arrows in Fig. 1.
Transitions of the configuration from one such region into another (constituting
the phenomenon of structural relaxation) are severely inhibited by high energy
barriers, which cannot be overcome over ordinary time scales at the prevailing
temperature. [Volume relaxation times for physical aging below TBare on
the order of years (Eisenberg, 1984).] These regions of configuration space in
which a glassy system may be trapped are thus effectively disjoint over ordinary
experimental time scales. How likely it is to find the local configuration of a
macroscopic polymer sample residing in each of these regions is no longer
dictated by the Boltzmann distribution but rather is strongly influenced by the
formation history of the glass (e.g., the cooling rate).
The assumption of Boltzmann-distributed configurations locally within each
region is legitimate (in fact, it defines the regions) but breaks down if extended
across barriers. When a gaseous penetrant is sorbed in the glass at low concen-
tration, this picture remains valid; an estimate of the solubility can be obtained
MOLECULAR SIMULATIONS OF SORPTION AND DIFFUSION 71
by formulating the phaseequilibrium problemseparatelyineach disjoint r,

region of configuration space and then averaging the results according to the
weight withwhich each region contributes to the overall glass thermodynamics.
At high concentrations, however, it is expected that a strongly interacting pene-
trant may modify the potential energy hypersurface sufficiently reduce barrier
heights between r, regions that were disjoint in the pure polymer glass and start
allowing transitions between these regions over ordinary time scales. The prob-
ability distribution in configuration space will start shifting toward the global
equilibrium distribution characteristic of a melt and eventually lead to devitri-
fication of the polymer. This is a qualitative molecular interpretation Of plasti-
cization effects often observed at high penetrant activities. From the point Of
view of molecular simulation, the problem of sampling the relevant fluctuations
that shape sorption equilibria in an intrinsically nonequilibrium glassy System is
a serious conceptual and computational challenge.
The diffusivity of a dilute penetrant in an amorphous polymer matrix is gOv-
emed by the penetrant sue and interactions with the polymer as well as by the
shape, she, connectivity, and time scales of thermal rearrangement of Unoccu-
pied space within the polymer. In a high-temperature melt (T >> Td, openings
among chains that are capable of accommodating the penetrant undergo rapid
redistribution in space. One can envision that the penetrant is “carried along”
by density fluctuations caused by the thermal motion of surrounding chains.
in the free-volume picture liquid-state diffusion, one could envision that a
penetrant molecule resides in a certain position of the polymer matrix until the
motion of surrounding chains, modified by thepenetrant’s presence, leads to the
formation a cavity, at a distance commensurate with the penetrant’s diameter,
into which the penetrant can move. After the move, the cavity in which the
penetrant was originally accommodated is closed. A succession of such small
random moves of the penetrant constitutes diff3sion. In this high-temperature
melt limit, formation of a cavity of sufficient size to accommodate the penetrant
can be viewed as the rate-controlling step for a move. The time scale between
moves is thus set by the relaxation time of density fluctuations on the length
scale of the penetrant diameter within the polymer matrix.
At temperatures below T,, on the otherhand, onewould expect the molecular
picture of diffusive jumps to be substantially different. The distribution of open
spaces within the configurationally arrested glassy matrix is more or less per-
manent (long-lived). One can envision a network of preexisting cavities, the
magnitude and shape of which fluctuates somewhat with thermal motion and is
modulated by the possible presence of a penetrant molecule. A dilute penetrant
spends most of its time rattling within a cavity and occasionally jumps from
cavity to cavity through a “window” that opens instantaneously via fluctuations
in “soft” regions (i.e., regions of lower density or enhanced molecular mobility)
between the cavities. The overall diffusivity in the glass would thus depend on
72 THEODOROU
the distribution of distances and connectivities among the cavities as well as on

the magnitude and distribution of rate constants governing the infrequent jumps
of the penetrant between adjacent cavities.
In the following sections we shall see how the simple conceptual picture
outlined above is substantiated by molecular simulations.
II. CHARACTERIZATION OF STRUCTURE AND MOLECULAR

MOTION IN AMORPHOUS POLYMERS
A. Methods of Generating Amorphous Polymer
Configurations
As pointed out in the qualitative discussion of Section I, the ability to represent
the molecular level structure and mobility of amorphous polymers is a prereq-
uisite for simulating sorption and diffusion in them. In this section we discuss
existing computational methodologies for generating model amorphous poly-
mers and analyzing the distribution of unoccupied space and the short-time
molecular motion of chains. We confine our discussion to relatively detailed
models in continuous space, in which the bonded geometry (bond lengths, bond
angles) and interaction energetics of chains are represented realistically; only
such models are conducive to a quantitative study of diffusion. In these models
chains are represented as groups of interaction sites, i.e., atomic nuclei and
centers of charge associated with polar groups. The potential energy Tpis ex-
pressed as a sum of bond and bond angle distortion terms, torsional potentials,
as well as intermolecular and intramolecular exclusion, dispersion, electrostatic,
and charge transfer interactions that depend on the coordinates rp of the inter-
action sites. Groups of atoms may be lumped into single interaction sites, as in
the “united-atom” model of linear polyethylene, consisting of a sequence of
methylenes capped by methyls at the two ends. The parameterization of the
potential energy is based on fitting experimental structural and thermodynamic
properties of low molecular weight analogs of the polymer or on ab initio cal-
culations of partial charges and specific interactions between polar groups (Allen
and Tildesley, 1987; Ludovice and Suter, 1989).
1. The “Amorphous Cell“
As in simulations of small molecular weight liquids (Allen and Tildesley, 1987),
the model system considered in amorphous polymer simulations is a box, or
cell, filled with chains and characterized by three-dimensional periodic boundary
conditions (see Fig. 2a). Currently, amorphous cells of edge length 20-50 A
containing 500-5000 interaction sites are studied with conventional computa-
tional resources. According to Flory’s “random coil hypothesis,” for which
ample experimental evidence has been collected through neutron scattering, the
conformation of chains in an equilibrium melt or amorphous glass remains es-
Figure 2 Schematic representation of amorphous cell and periodic boundary condi-

tions. Wo-dimensional projections of two three-dimensional periodic model configura-
tions are shown. The primary simulation cell lies at the center and is surrounded by eight
images of itself. ( a ) Finite molecular weight system; the contents of the amorphous cell
are formed from four “parent” chains. (b) Infinite molecular weight system; the contents
of the amorphous cell are formed from a single chain with no ends; three images of this
chain go through the primary box.
74 THEODOROU
sentially unperturbed by interactions between topologically distant sites along

the chain. a consequence, the mean square radius of gyration (S’) of a long
chain in the bulk is related to the number of skeletal bonds n and to the skeletal
bond length 4 as (S’) = (1/6)nCJ2, where the characteristic ratio, is a chain
length-independent constant that can be predicted by conformational analysis
Of short sections of the chain (Flory, 1969). Strictly, the edge length L of the
periodic box must fulfill the condition L > 2(s2)’”, that different images of
the Same chain do not interact significantly with each other, as such interaction
may distort the long-range conformationalcharacteristics of chains; this imposes
an upper limit on the molecular weight that can be simulated with a cell of
given dimensions. In glassy polymer simulations, however, where chainsdo not
have the opportunity to relax their long-range conformation, meaningfulresults
have been obtained with as few as one “parent” chain per box, provided care
is taken in the initial preparation of the cell so that the conformation of the
chain is close to unperturbed (Theodorou and Suter, 1985). Infinite chain length
models have also been constructed to examine properties in the absence of chain
end effects (Weber and Helfand, 1979). The model system of Fig. 2b, for ex-
ample, is entirely constructed from images of a single noncyclic chain with no
ends. In the preparation of such models, care is taken so that the basic periodic
elements of which the system is formed (e.g., the chain section between A and
A‘ in Fig. 2b) are close to unperturbed; clearly, however, this cannot be said
about the entire infinite chain, which is strongly oriented in the AA’ direction
when examined at length scales substantially larger than L .
2. MolecularMechanics
The objective of molecular mechanics (MM) is to generate static minimum en-
ergy configurations at prescribed density, corresponding to the local minima of
V , shown in Fig. 1. The technique was developed to model glassy amorphous
polymers (Theodorou and Suter, 1985). One method for creating a minimum
energy configuration starts by generating an initial guess Configuration through
bond-by-bondgrowthofthe parent chains in the amorphous Cell, observing
periodic boundary conditions. This growing procedurecan be based On FloY’S
rotational isomeric state model for unperturbed chains, modified SO as to avoid
intra- and intermolecular excluded volume interactions. The initial guess con-
figuration is then subjected to minimization of the total potential energy VPWith
respect to all microscopic degrees of freedom. The minimization is most effi-
ciently accomplished in a stagewise fashion,wherein one Starts with Purely
repulsive interatomic potentials of reduced range to relax the most Severe ex-
cluded volume overlaps, gradually enlarges the atomic radii to their actual Size,
and finally introduces attractive interactions aswell (Theodorouand Suter,
1985). advantage of the MM technique is its speed; a model glassy Config-
uration can be arrived at within a few minutes to an hour of CPU time On a
Cray-Y/MP vector machine, i.e., with two orders of magnitude less computer
time than is required by methods that incorporate thermal motion. A disadvan-
tage of molecular mechanics is the fact that thermal fluctuations are not explic-
itly accounted for and that the system density has to be set a priori; thus, only
properties of aglassysystemthat is locallyarrestedin the neighborhoodof
potential energy minima can be examined. Also, the procedure followed in gen-
erating minimum energy configurations does not correspond to a well-defined
history of vitrification. Nevertheless, arithmetic averaging of the properties of
minimum energy configurations (which is based on the assumption that such
configurations are generated by molecular mechanics with a probability distri-
bution comparable to that encountered in a real-life glassy polymer) has led to
encouraging predictions of structure, elastic constants, and surface thermody-
namic properties (Theodorou and Suter, Mansfield and Theodorou,
Onthe other hand, static minimumenergy structures generatedby MM are
satisfactory as starting configurations for molecular dynamics and Monte Carlo
simulations that incorporate thermal motion.
3. Molecular Dynamics
Molecular dynamics (MD) tracks the temporal evolution of a microscopic model
system throughnumerical integration of the equations ofmotion for all the
degrees of freedom. Originally developed in the microcanonical (constant num-
ber of molecules, volume, and total energy, or ensemble, the MD method
can now be applied straightforwardly in the canonical (m and isothermal-
isobaric (NPT) ensembles as well. Constraints on the microscopic degrees of
freedom, such as constancy of bond lengths and bond angles, can also be han-
dled (Allen and Tildesley, The MD method has been applied widely to
united-atom polyethylene-like systems by Rigby and Roe
To create a model melt system, an MD simulation of monomeric segments was
conducted, in the course of which bonded forces were gradually “turned on” to
form chains. Glassy configurations were obtained from the melt through a series
of MD runs at progressively lower temperatures (the cooling rates were too fast
to allow crystallization). asset of molecular dynamics is that it provides directly
a wealthof detailed informationon short-time dynamical processes in the polymer.
Its major limitation is that, since it faithfully mimics molecular motion in actual
amorphous polymer systems, it is fully subject to the bottlenecks that limit this
motion. Hundreds of CPU hours on a vector supercomputer are required to sim-
ulate a nanosecond of actual motion with atomistic MD. In view of the much
longer relaxation timesof actual long-chain polymer melts, the mere equilibration
of a model melt by MD becomes problematic, and results are not trustworthy if
one starts from an improbable initial configuration. Model glassy configurations
formed by MD cooling of the melt have a well-defined history, which, however,
is very far from that used in glass formation experiments; the cooling rates em-
76 THEODOROU
ployed in MD vitrification are on the order of 10'l-lO'o K/s. Some problems

associated with the generation of amorphous model structures through MD have
been addressed by McKechnie et al. (1992).
Carlo
The objective of a Monte Carlo (MC) simulation is to generate a large number
of configurations of the microscopic model system under study that conform to
the probability distribution dictated by the macroscopic constraints imposed on
the system. For example, a Monte Carlo simulation of a melt of N chains in
volume V at temperature T generates a large sequence of configurations,in
which each configuration r, occurs with
frequency proportional
exp["Vp(rp)], as dictated by the canonical ensemble. At each step of the MC
simulation one attempts to generate a new configuration from the current con-
figuration through an elementary move. The attempted move is either accepted
or rejected according to selection criteria designed that the resulting sequence
(or Markov chain) of configurations asymptotically samples the probability dis-
tribution of the ensemble of interest. Thermodynamic properties are calculated
as averages over the sampled configurations (Allen and Tildesley, 1987).
initial configuration for starting the MCsimulation can be obtained
through one of several strategies. Vacatello et al. (1980) started their pioneering
simulations of liquid triacontane by placingchains onatetrahedrallattice,
whereas Boyd (1989) employed a crystal, which was melted during theMC
equilibration stage, as an initial configuration for his simulations of liquid tetra-
cosane. Minimum energy configurations from molecular mechanics have also
been used (Dodd et al., 1993). To accelerate convergence of an MC simulation
with elaborate atomistic models, Vacatello recently proposed using equilibrated
configurations of a simple model of tangent sphere chains with the same end-
to-end distance as the actual chains as a starting point (Vacatello, 1992). De-
signing elementary moves for a polymer MC simulation is more of a problem
than for a small-molecule Simulation,becauseofthe constraints imposed by
bonds and bond angles. Rotations around bonds near the chain ends have been
used along with "reptation" moves, wherein a terminal segment of a chain is
deleted and a new terminal segment is appended on the other end at a random
torsion angle. In the recently developed continuum configurational bias (CCB)
algorithm (de Pablo et al., 1992), an end section of random length is cut
from a randomly chosenchain and regrown bondby bond by selectively placing
segments in regions where they are not likely to experience excluded volume
interactions with other chains; the bias associated with this selective growth
procedure is removed with appropriate selection criteria. MC simulation with
concerted rotations (Dodd et al., 1993), on the other hand, employs coordinated
torsions around seven adjacent skeletal bonds to change the conformation of
chains withoutaffecting bond lengths and bond angles. Moves that involve cut-
MOLECULAR
SIMULATIONS OF SORPTION AND DIFFUSION 77
ting chains and rejoining the resulting segments inadifferentway,thereby

drastically altering the system configuration, are currently being explored as an
extension of the concerted rotation move. An advantage of the MC method is
that, by judicious choice of the elementary moves, one can circumventthe bot-
tlenecks in configuration space that inhibit molecular relaxation in real polymer
systems and in MD simulations and thus effect rapid equilibration of a multi-
chain model system. There are strong indications that MC schemes under de-
velopment today afford order-of-magnitude savings in CPU time relative to mo-
lecular dynamics as means of preparing equilibrated melt systems. In addition,
the MC method can readily be appliedinavariety of ensembles, including
ensembles in which the total number of particles of some species is allowed to
fluctuate; this is of strategic significance in simulations of sorption (see below).
A disadvantage of MC simulation is that it does not provide true dynamical
information. When the moves employed mimic dynamical processes that can
actually take place in the polymer, one can establish a loose correspondence
between attempted moves per interaction site and elapsed time that allows con-
clusions about long-time dynamical processes to be extracted (‘‘dynamic Monte
Carlo” approaches). Furthermore, in using MC simulations to predict the ther-
modynamics of configurationally arrested glassy systems, one has to rely upon
appropriately chosen “quasi-dynamic” moves to stay confined within the rele-
vant regions of configuration space and avoid tunneling through physically in-
surmountable energy barriers (compare Fig. 1).
A variety of properties can be predicted from an amorphouspolymer model and

tested against experimental evidence to confirm that the model provides a rea-
sonable representation of reality. The pair distribution functions g(.) for all site
pairs in the system can be obtained readily from the configurations generated in
the course of a simulation (Allen and Tildesley, 1987). Note that the location
of the first peak in the intermolecular part of the pair distribution function be-
tween skeletal atoms gives a direct measure of the average distance between
chain backbones in the amorphous bulk. Through Fourier transformation of the
pair distribution functions one can predict X-ray, neutron,or electron diffraction
patterns that can be compared to experimental wide-angle patterns from the bulk
polymer. Similarly,from the intramolecular pair density function onecan predict
small-angle neutron diffraction patterns obtained experimentally by selective la-
beling of single chains and thus check long-range conformationalcharacteristics
of the chains in the model sample. The ability to predict PVT behavior in the
melt is a rather stringent test of the model representation used in a simulation
and is a prerequisite for reliable studies of sorption and diffusion. Dynamical
properties, such as self-difisivity of chains and viscosity in the melt and fre-
quency and activation energy of relaxational motions, can be compared to scat-
78 THEODOROU
2.:
1.t
-
4
I
0.7
2z
CA
v
A -0.2
-1.1
c20
-2.0
4 8 12 1 )
(4 Magnitude of wave vector (i-')

Figure 3 Structural and thermodynamic predictions from isothermal-isobaric MC
simulations of linear alkane liquids. (a) Diffraction pattern compared against experiment;
(b) equation-of-state properties compared against experiment. See text for details.
tering, rheological, and spectroscopic measurements. Such comparisons are cur-

rently practiced only to a limited extent, owing to the difficulties in accessing
long-time dynamics with atomistic simulations. Smith and Boyd (1992) com-
pared activation energies for side-group rotations in their model glassy polymers
against experimental evidence, with favorable results. More recent work by
Smith and Yoon (1994) demonstrated that a well-calibrated explicit atom poten-
tial, when used in equilibrated MD simulations of a high-temperature tridecane
melt, can reproduce chain self-diffusivities and correlation times for C-H bond
reorientational motion in excellent agreement with experiment. A united-atom
potential is less successful in capturing the local segmental dynamics, although
it can reproduce equation-of-state behavior, scattering patterns, and chain self-
diffusivities quite well. As an indication of the agreement that can be achieved
between simulation estimates on the one hand and structural and thermodynamic
measurements on the other, we display some results from MC simulations of
long-chain linear alkane liquids in Fig. 3 and Table 1. A consistent united-atom
representation was used for the simulations (Dodd and Theodorou, 1994). Figure
MOLECULAR SIMULATIONS OF SORPTlON AND DIFFUSION 79
Figure 3 Continued.
Table 1 Conformational Characteristics of Tetracosane Chains

Chain Bulk NPT Continuous
property Monte Carlo" unperturbedh
~~
350 354
43 44
0.659 0.666
0.394 0.406
0.532 0.524
0.110 0.104
"Sampled in the course of an isothermal-isobaric Monte Carlo simulation of the bulk liquid.
hUnperturbed chains governed by the same intermolecular potentials but not experiencing nonlocal
interactions.
80 THEODOROU
3a compares the k-weighted structure factor for simulated C, at 315 K and 1

bar (curve) against accurate neutron diffraction measurements (points) (Haben-
schuss and Narten, 1990). Table 1 displays a comparison of several C, single-
chainconformational characteristics, namely the rootmean square radiusof
gyration ( s ’ ) ’ ~ ,the rms end-to-end distance (r’)’’’, the mean fraction p I of skel-
etal bonds in a trans conformation, the mean fractions p f fand p s of pairs of
adjacent bonds in trans-trans and trans-gauche conformations, and the mean
fraction pfg,of triplets of successive bonds in a trans-gauche-trans conforma-
tionagainstthe corresponding predictions for unperturbedchains. Figure 3b
shows simulation predictions for the specific volume of C& and C& at 450 K
as a function of pressure (triangles) compared with experimental values(squares
and dotted lines) (Dee et al., 1992). The comparisonsindicate that the simulation
is well-equilibrated and free of model system size effects and that the model
representation employed is reasonable.
B. Accessible Volume and Its Distribution

Atomistic model configurations obtained through the simulation techniques de-
scribed in Section 1I.A can serve as a starting point for characterizing the free
spaces within the polymer where a penetrant molecule can reside. Such geom-
etry-based (as opposed to energy-based) analyses of theinternal structure of
amorphous polymers can be conducted with little computational expense once
the model configurations are available. A main objective of these analyses is to
relate the magnitude and distribution of “free volume,” which has played a
central role in theories of sorption and diffusion, to chain chemical constitution
and architecture. Experimental efforts to determine the distribution of unoccu-
pied volume in polymermatrices through positron annihilationlifetime measure-
ments have appeared recently (Malhotra and Pethrick, 1983; Kluin et al., 1993;
Deng and Jean, 1993); the geometrical analysis of model structures is helpful
in the interpretation of such measurements.
Several definitions have been used for “free volume” in theoretical work.
For the purpose ofcharacterizing void space in model structures, it is meaningful
to considereach interaction site (atom) on the polymer chains or on the penetrant
molecule as a hard sphere of diameter equal to its van der Waals radius ro. We
use the term unoccupied volume to refer to the volume of the three-dimensional
domain composed of points within a configuration that lie outside the van der
Waals spheres of all polymer atoms. The term accessible volume refers to the
volume of the domain composed of points that can be occupied by the center
of mass of the penetrant molecule without any overlap between the van der
Waals spheres of the penetrant and those of the polymer atoms. Our discussion
here is confined to spherical penetrant molecules represented as single interac-
tion sites. It is emphasized that the analytical techniques we discuss are applied
to the pure polymer matrix. The penetrant molecule is used as a geometrical

probe of the internal structure of the polymer, which does not respond to the
penetrant in any way.
The determination of the volume, within a given configuration, that is ac-
cessible to a given penetrant can be conveniently conducted as follows. The
radii of all polymer atoms are augmented by a length equal to the penetrant
radius r:, and the unoccupied volume of the resulting model system, in which
each atom i is represented through its “excluded volume sphere” of radius r:
+ r:, is calculated @oddand Theodorou, 1991; Greenfieldand Theodorou,
1993). Conversely, the unoccupied volume of a polymer configuration coincides
with its accessible volume in the limit 1-2 + 0. Figure 4 depicts the three-
dimensional domains within a glassy atactic polypropylene configuration that
are accessible to helium (r: = 1.28 A) and to argon (r: = 1.91 A) (Greenfield
and Theodorou, 1993). In both cases the accessible volume consists of disjoint
clusters. As the penetrant radius is reduced, the accessible clusters grow in size;
tentacle-like protrusions on the periphery of different clusters come together,
causing pairs of clusters to merge into a larger cluster. At the same time, new
clusters becomeavailable. At some critical penetrantradius, r,, an infinitely
extended cluster appears that spans the entire periodic array of boxes repre-
senting the polymer; that is,percolation accessible volume occurs throughout
the model polymer. The percolation threshold value r, varies somewhat from
configuration to configuration; it also depends on the edge length L of the pri-
mary box, smaller boxes being easier to percolate. A systematic study of r, in
primary boxes different L can be used to deduce the percolation threshold in
the limit L-?, For glassy atactic polypropylene, the average r, in the infinite
box limit is around 0.9 i.e., smaller than the radius of any gaseous penetrant
that might permeate the polymer (Greenfield and Theodorou, 1993).
The calculation of unoccupied volume is complicated by the fact that bond
lengths are typically small relative to van der Waals radii, and thus the van der
Waals spheres of atoms along a chain interpenetrate profusely. This interpene-
tration is even more pronounced in the case of excluded volume spheres used
for the calculation of accessible volume. Shah et al. (1989) introduced a Monte
Carlo integration technique for the determination of accessible volume. The
technique consists in choosing a large number [S(106)] of points randomly in
the simulation box and determining what fraction of these points lie outside the
excluded volume spheres of all polymer atoms. computationally more efficient
approximate technique was introduced in the pioneering work of Arizzi et al.
(1992). In this technique, the model configuration is partitioned into tetrahedra
of nearest-neighbor atoms, and the accessible volume is computed separately in
each tetrahedron by an analytical procedure that accounts for twofold overlaps
between excluded volume spheres; tetrahedra in which threefoldor higher over-
laps are observed are considered fully occupied. exact analytical solution of
82 THEODOROU
4 Three-dimensional depiction of the volume within a model configuration of

glassy atactic polypropylene that accessible to helium (top) and argon (bottom). The
edge length of the model box is approximately 23 A. Different clusters free volume
are displayed in different colors (shown here in shades of gray). Periodic boundary con-
ditions are evident.
this problem, based on an efficient algorithm that accounts for sphere overlaps
of any order (Dodd and Theodorou, 1991), was presented recently (Greenfield
and Theodorou, 1993).
Unoccupiedand accessible volume calculations have beenconductedon
modelglassy atactic polypropylene(&PP),poly(viny1 chloride) (PVC),and
bisphenol A polycarbonate (PC), as well as on model melts of united-atom
polyethylene(PE)and &-PP. The unoccupied volume fraction in the glassy
polymers is found to be around 0.35, the exact value depending on the nature
of the polymer. For example, Arizzi et al. report 0.354 2 0.001 for &-PP and
0.39 2 for PC. The accessible volume fraction is a monotonically de-
creasing convex functionof penetrant radius (see Fig. 5). In glassy &PP, Arizzi
et al. report accessible volume fractions of 0.16,0.08, and for He, 02,and
NZ,respectively; the corresponding values in PC are 0.19 for He, 0.12 for Oz,
and 0.11 for N2. It is interesting to compare theavailabilityof free volume
between a glassy polymer and an atomic glass; random close-packed (rcp) con-
figurations of spheres constitute a reasonable modelfor the latter. Figure 5 shows
such a comparison between&PP and an rcp structure consisting of 2 diameter
spheres. This diameter is commensurate with that of methylene, methine, and
methyl segments in the propylene model; the atomic “gran~larity~~ of the rcp
and polymer structures is thus comparable. Although the unoccupied volume of
the rcp structure is somewhat higher, its accessible volume falls more rapidly
with increasing penetrant diameter than thatof the polymer. The macromolecular
constitution of the polymer gives rise to a broader distribution of void sues,
which can accommodate larger penetrants. This is also reflected in the perco-
lation characteristics of accessible volume in the two types of glassy structures:
r, for the rcp structure is around 0.55 which is significantly smaller than the
value of 0.9 for the polymer glass. For both the rcp and polymer glass
tures, the accessible volume fraction at the percolation threshold is in the range
0.02-0.04, close to the value observed for the “Swisscheese” model(see
Chapter 1 of this book).
For given penetrant radius, the accessible volume is distributed spatially into
clusters (see Fig. 4). Several techniqueshave been used to quantify the size and
shape distribution of these clusters. Boyd and Pant (1991a) chose to examine
the distribution of the radii of the largest spheres that can be inscribed within
tetrahedral interstitial sites formed by polymer atoms. Arizzi et al. (1992) intro-
duced a rigorous procedure for defining clusters of accessible volume that relies
on Delaunay tessellation of the model polymer configurations. In a three-di-
mensional Delaunay tessellation, an arbitrary collection of points (atomic cen-
ters) is partitioned completely into tetrahedra, each tetrahedron having four near-
est-neighbor points as its apices; the circumsphere of a Delaunay tetrahedron
does not contain any other points inside it. Fast algorithms for performing the
Delaunay tessellation and its dual Voronoi tessellation are available (Tanemura
84 THEODOROU
0.4
1u5 I....I...,I....I....I....I
Ob 1.0 1.5 2.0 2.5
0
0 0.5 1.o 1.5 2.0 2.5
penetrant radius
5 Accessible volume fraction as a function penetrant size in atactic poly-
propylene as obtainedfrom Monte Carlo simulations the polymer in the glassy and
melt (0) states (Greenfield and Theodorou, 1993). The accessible volume fraction a
random close-packed (rcp) structure 2 A diameter spheres, representative an atomic
glass, is also shown The diameter the rcp spheres is roughly equal to the van der
Waals diameter methyl, methylene, and methine units constituting the polymer.
et al., 1983). Delaunay tetrahedra are an excellent means for identifying inter-
stices of accessible volume within a polymer configuration. For a given r:, if
the interior of a tetrahedron is completely filled by the excluded volumespheres
of polymer atoms, then the tetrahedron is inaccessible; otherwise, the tetrahedron
has a pocket of accessible volume in its interior. W Oaccessible tetrahedra are
said to be connected when they share a face (triangle) that is not completely
blocked by the excluded volumespheres polymer atoms. Uninhibited passage
of the penetrant from the interior one tetrahedron into the other through the
shared face is thus possible. Accessible tetrahedra can be grouped into sets of
connectedtetrahedrausinga simple connectivityalgorithm(Greenfieldand
Theodorou, 1993). A cluster of accessible volume is simply the union the
accessible volumes of such a set of connected tetrahedra. Although fast and
accurate, Delaunay tessellation is not the only tessellation whereby clusters of
unoccupied volume can be analyzed. Takeuchi and Okazaki (1993a), for ex-
ample, partitioned their MD configurations into cubic elements for the same
purpose.
For given penetrant radius r i , one can define a volume-weighted probability

density distribution of accessible cluster volumes, pv(v; r:), such that pv(v; r:)
dv equals the fraction of the total accessible volume foundin clusters of volume
+
to v dv. py(v;r i ) is generally a decreasing function of v; its reliable deter-
mination requires analysis of a large number of configurations. Computed plots
of versus v seem to exhibit some fine structure (local extrema) related to
specific intermolecular packing patterns that depend on the detailed geometry,
conformationalpreferences, and interactionforces among chains (Greenfield and
Theodorou, 1993); their general appearance is inreasonableagreementwith
positron lifetime spectroscopy results (Deng and Jean, 1993). The average clus-
ter size can be characterized through the average cluster volume (v), (first mo-
ment of or through the root mean square cluster radius of gyration (Green-
field and Theodorou, 1993). In the range of penetrant radii r: > r,, the average
size of accessible clusters is a smoothly decreasing function of r: (compare Fig.
The shape of accessible clusters can be characterized by comparing the
principal axes of their radius of gyration tensor; in the case of &PP, the as-
phericity of the clusters was found to be comparable to that of an ellipsoid of
revolution with principal axis lengths in the ratio 2:l:l.
picture such as Fig. 4a suggests that a penetrant molecule of sufficiently
large r l , sorbed at low concentration within a glassy polymer, would spend most
of its time confined in the interior of small disjoint clusters of accessible volume.
The dif€usion of the penetrant would proceed by infrequent jumps from cluster
to cluster through short-lived passages opening momentarily between the clusters
(see also Section IV.D). The likely location of such passages that could open up
through thermal fluctuations and thus act as difhsion pathways for the penetrant
can be identified by examining the accessible volume distribution at a value of
the penetrant radius equal to, e.g., r,, for which long-range connectivity is bound
to be established.Suchanexaminationrevealsthe coordinationnumber of a
cluster, i.e., the most probable number of clusters with which a given cluster is
connected through difhsion pathways. For Ar and He in glassy at-PP (Greenfield
and Theodorou, 1993), the coordination number of a cluster is found to follow
rather broad distributions with most probable values of 4 and 2, respectively.
moreelaborateandcomputationallymuchmoretime-consumingapproach for
identifying clusters and passages between themis to analyze the potential energy
field experiencedby a penetrant at every point in a glassy polymer configuration.
Such an energetic analysis of He in PC, conducted by Gusev et al. (1993), re-
vealed accessible clusters of diameter 5-10 & connected by bottleneck regions
ca. 5-10 long and 1-2 in diameter, in agreement with the geometric analysis
(seeFig. 4). It should be emphasized that the rate constants for passage from
cluster to cluster through a diffusion pathway follow a broad distribution, and
therefore the connectivityof the network of clusters is a functionof the time scale
over which the network is examined. We return to this point in Section lYD.3.
86 THEODOROU
It is informative to track thermal fluctuations in the distribution of accessible

volume. Results from such a study, conducted in the course of long MC simu-
lations of an &PP glass and melt, are shown in Fig. 6. In the glass (top), one
clearly sees that the distribution of accessible volume changes very little. The
clusters present at the beginning of the MC run are surviving at the end of the
run; the configuration is locked in, andthe void distribution can be characterized
permanent over the effective time scale spanned by the simulation. In the
Figure Evolution of thedistribution of volumeaccessibletoHewithinatactic

polypropylene, as obtained from long MC simulations of the polymer. (a) Polymer glass
configuration at K. (b) Configuration obtained from (a) after million attempted
MC moves at1 bar and K. (c) Polymer melt configuration 400 at K (d) Configuration
obtained from (c) after 10 million attempted moves at 1bar and 400 K
melt (bottom), and over the same number of MC moves, one sees a dramatic
change in the accessible volume distribution. The void space has reorganized to
such an extent that it is impossible to recognize clusters in the bottom right
snapshot as having evolved from clusters in the bottom left snapshot. Melt clus-
ters are transitory. This difference in the rate of accessible volume redistribution
in the high-temperature versus low-temperature amorphous polymers has im-
portant implications for the mechanism of diffusion.
Although not capable of providing predictions for sorption isotherms and
diffusion coefficients, the simple geometricconsiderations discussed in this sec-
tion are useful as a prelude andguide to the more elaborate (and computationally
much more demanding) energy-based approaches discussed in Section 1V.D.
C. Characteristic limes of Molecular Motion

Molecular motion in amorphous polymers is governed by a wide spectrum of
characteristic times. Bond and bond angle vibrations occur on time scales on
the order of 10”3-10”4 S, which are relatively insensitive to molecular packing.
The rates of rapid rotations of pendant groups exhibit an Arrhenius temperature
dependence; they are associated with p-relaxation phenomena observed in low-
temperature glasses. Librations of skeletal bonds in their energy wells and con-
formational transitions across torsional energy barriers are accomplished through
localized distortions of the chain backbones involving on the order 10 bonds
along a chain. The rate of such “segmental” motions is very sensitive to density.
In a high-temperature melt they are are relatively uninhibited by surrounding
chains, occurring over time scales of 10”2-10”o S. As temperature drops they
slow downdramaticallyandbecomeincreasingly cooperative, theassociated
a-relaxation functions usually being fit to a stretched exponential Kohlrausch-
Williams-Watts (KWW) form and the temperature dependence of the observed
relaxation time following a manifestly non-Arrhenius Williams-Landel-Ferry
(WLF) equation(PlazekandNgai,1991).Finally,large-scale conformational
rearrangement, reorientation, and self-diffusionof chains are frozen-in in a glass
but present in a melt. The characteristic times for such large-scale conforma-
tional rearrangements lie in the “terminal” region of the relaxation spectrum;
they are very sensitive to chainlength, their chain length dependence being
described rather satisfactorily by the Rouse and reptation models in unentangled
and entangled polymer melts, respectively. (See Chapter 6 by P. F. Green.)
Here we discuss briefly some findings about local segmental motions in an
amorphous polymer matrix, as obtained from MD simulations. These motions
are particularly relevant to diffusion of small penetrants through the matrix, as
they dictate the thermal fluctuation of accessible volume clusters and diffusion
pathways (compare Section ILB).
Takeuchi and Roe (1991a,b) quantified the rate at which individual torsion
angles lose memory of their initial values by defining anautocorrelation function
88 THEODOROU
for dihedral angles, R,(t) as
where the ensemble averagesare taken over all skeletal bonds and Over all time
origins along the MD simulation. Rm(t) starts at 1 and decays to zero in a
stretched exponential fashion. In the melt, the time 7, at which R, has decayed
to l / e is studied as a function of temperature. In an infinite molecular weight
PE melt at 300 K, Takeuchi and Roe found T+ = 24 PS; in the high-temperature
melt, decreased with increasing T with an activation energy of roughly 3.75
kcal/mol, which is comparable to the trans-gauche torsional barrier. As tern-
perature was reducedtoward T, (which, as definedthrough the break inthe
volume versus temperaturecurve, is 201 K for this polymer at a cooling rate of
1.67 10” Ws), was found to increase dramatically. A KWW fit to Eq. (1)
at T, gave a relaxation time = 1.39 ns. Within the glass the rate of well-to-
well conformational transitions was found to be on the order of 1 ns“ down to
a temperature of 148 K. Comparable time scales for conformational transitions
have been found in an MD simulation of a short-chain atactic polypropylene
glass near its T, (Mansfield and Theodorou, 1991). The latter simulation indi-
cated a large degree of spatial heterogeneity in terms of the ability of bonds to
isomerizeconformationally. The glass wasfound to contain isolated “soft
spots,” where conformational transitions occur at rates comparable to those seen
in polymer melts, surrounded by a ‘‘stiff ” continuum, where no transitions are
observed over hundreds of picoseconds.Torsionalmobility is enhancednear
chain ends, although “soft spots” are likely to be found away from ends as
well. One canenvision that with increasing temperature the size and connectivity
of “soft spots” increases at the expense of the stiff region until the spots per-
colate through the glassy bulk, signaling the devitrification of the polymer.
Another way of studying segmental mobility in MD simulations is to track
the orientational decorrelation of characteristic unit vectors rigidly embedded in
the chains. The correlation times for such motion are measurable with NMR,
dielectric relaxation, electron spin resonance, photon correlation, and fluores-
cence spectroscopy. Let ii be such a unit vector. As a result of thermal motion,
the vector’s orientation at time t, ii(t), will be different from its original orien-
tation ii(0). One can form the two time correlation functions
= (ii(t) * ii(0)) (2)
and
1
M$) = - (3[ii(t) ii(O)]’
m - 1)
2
The time decay of these is not exponential; it is well described by a KWW

expression
Me(t) = exp [-(t/+~e)’] (4)
For a given choice of Q the ratio T,/T~is always larger than 1, its exact magnitude
depending on the mechanism of the orientational relaxation. Smith andYoon
(1994) chose Q as unit vectors along the pendant C-H bonds of a tridecane
liquid. For the C-H bonds attached to the central carbon chains, their explicit
atom MD simulations gave T~ = 10 PS at 312 K with an activation energy of 4
kcal/mol. The value of + T ~was found to besignificantly shorter for the five
carbons near each chain end, dropping to 2.5 PS at the terminal methyls. All
these observations are in excellent agreement with I3C N M R spin-lattice relax-
ationexperiments. The behavior T~ and T~ for pendant bonds tracked the
behavior of T,, all three times being of the same order magnitude. Takeuchi
and Roe (1991a,b)studied three Q vectors with their united-atom modelof poly-
ethylene. Vector a is directed along the bisector of a skeletal bond angle in the
plane of two adjacent skeletal bonds; vector c is normal to a in the plane of the
bond angle and thus points along the direction of the chain backbone; finally,
the “out-of-plane” vector b is the cross product of c and a. In the melt, the
vectors a and b that are directed normal to the backbone were found to relax
with comparable rates, b being somewhat faster. Their behavior was strongly
correlated with bond angle relaxation, T~ being approximately equal to T+ and
being 0.3-0.5 T+ at all temperatures studied. In contrast to a and b, vector c
(oriented along the backbone) was found to relax dramatically slower, its + T ~
being roughly 3807, at 300 K. This anisotropy of orientational relaxation is
comparable to but stronger than that observed in Brownian dynamics simulations
of isolated polymer chains in solution (Ediger and Adolf, 1994). In the glass,
the KWW apparentrelaxation times for a and b were on the order of 1 ns
(Takeuchi and Roe, 1991b; Mansfield and Theodorou, 1991). Bondreorientation
angle distributions in the glassy polymer revealed two mechanisms as respon-
sible for the decorrelation of a and b. One is rotational diffusion, the other a
jumplike process wherein the bond direction changes abruptly by an angle com-
parable to the distance between conformational energy wells.
The low values of the stretching exponent < 1 obtained from fitting
(4) to the relaxation functions M&) and M&) of a, b, or pendant bond vectors
indicate considerable cooperativity of reorientational motion in the melt =
0.45-0.5) and a very high degree of cooperativity in the glass = 0.2). The
limited duration of the MD simulations (= 1 ns), however, does not permit the
reliable prediction of correlation times (seconds to hundreds of seconds near T,)
obtained from dynamic light scattering (Fytas and Ngai, 1988) and two-dimen-
sional NMR (Schaefer et al., 1990) measurements. The latter times are com-
parable to the ones obtained from mechanical measurements (Fytas and Ngai,
90 THEODOROU
1988) of the relaxation. The question of how to predict the time scales of
relaxation in the rubbery and glassystates reliably through molecular simulation
is extremely important but still unresolved.
Perhaps a more complete way to characterize density fluctuations in the amor-
phous polymer bulk is to accumulate the intermediate scattering function F(k,
t), i.e., the Fourier transform of the density-density correlation function of the
amorphous polymer. Consider a polymer consisting of segments of one type
(e.g., polyethylene in a united-atom representation). Let r,Q) be the position of
segment j at time t. The Fourier component of the instantaneous density cor-
responding to wavevector k at time t is
The time evolution of F($ t) reveals how density fluctuations occurring over
length scale 2dlkl disappear through thermal motion. The Fourier transform of
F(k, t ) is the dynamic structure factor S(k, W); it is measurable through coherent
inelastic neutron scattering. Perhaps more interesting than F(k, t ) itself is its
self-part F&, t ) defined as
F& t) = L
Ns
(2
,=l
exp { -i k [rj(t) - rj(0)]}
The time evolution of F&, t ) reveals how individual segments lose memory of
their original position through motions occurring at a length scale 21~44.Its
Fourier transform with respect to time, S& W), is measurable through inco-
herent neutron scattering. The structural “slowing down” occurring as a liquid
is cooled toward the glass temperature can be detected in the time decay of
F& t), and the characteristic frequencies associated with the relaxation at a
given length scale can be extracted from S&, t). This approach has been useful
in studying the kinetic glass transition in colloidal suspensions of spherical par-
ticles, where MD simulation results have been compared with the predictions
of mode-coupling theory (Barrat et al., 1990). Similar investigations were con-
ducted by Takeuchi and Okazaki for a model polymer (Takeuchi and Okazaki,
1993b); the decay of F& t) can be observed only at length scales commen-
surate with the distance between neighboring carbons on different chains (k =
1.5 A”); at this length scale, and at the glass temperature determined from the
V(T) behavior of the model polymer at a cooling rate of 1.67 10” K/s, the
MOLECULAR
relaxation time determined from a KWW fit to F& t) was around 0.75 ns.
The relaxation time was strongly dependent on temperature. The investigation
of structural relaxation over larger length scales is limited by the heavy com-
putational requirements of MD.
PREDICTION OF SORPTIONTHERMODYNAMICS
A. Statistical Mechanics of Sorption in a Compressible,
Involatile Medium
The phase equilibrium between a polymer and a multicomponent fluid mixture
can be predictedfrommolecularlevel structure and interactions basedon
straightforward principles of statistical mechanics. formulate this phase equi-
librium problem in a general way, consider the system shown in Fig. Phase
p (the fluid phase) and phase (the polymer phase) together constitute a closed
systematconstanttemperature T andpressure P . The system consists of c
components. Component 1 is the polymer chains. Components 2, . . . , c are
the fluid molecules whose sorption in the polymer we wish to describe. The
total number of molecules of each component in the two-phase system, Ni =
+
N4 NP (i = 1, 2, . . . , is fixed. Each of the components can be exchanged
freely between the and phases, however. Also, the phase is free to expand
(swell) against the phase.
Phase
p
Can exchange energy and mass with phase
Can expandlcontract against phasep
7 Thermodynamic system considered in the phase equilibrium formulation of
Section IILA.
92 THEODOROU
By the Gibbs rule, the intensive properties of the two phases are fully spec-
ified if one fixes c independently variable intensive properties in the system. As
such it is convenient to choose P, T, and c - 2 mole fractions describing the
composition the fluid phase on apolymer-free basis. If c = 2, T and P suffice
for specifying the intensive state of the two phases.
Let denote collectively the vector of microscopic degrees freedom
C n U
phase a for given NY, N;, . . . ,N:. f l u = (V", ru)encompasses the volume
phase a and the Cartesian coordinatesfor all atoms all molecules constituting
that phase. Similarly, will denote the vector of microscopic degrees of free-
dom of phase for given NB, N!, . . . ,N f . The unnormalized joint Probability
density governing the configurations of the two phases is
pN1. . . N c P T (x,
...,N:, NB, .. . ,N!; flu, a@)
C
-- PuN r . . .N:PT (am) .NFPF

S (Ng + NB - Ni)
i=l
where
1
(W)=
x! - nall
alom
A:,o,,,
exp{-p["ir(x, .. . ,NZ; S ) + P T ] }
stands for the unnormalized configuration-space probability density of phase a
inthe isothermal-isobaric ensemble (McQuarrie, 1976) with "ir the potential
energy function. The product of thermal wavelengthsA raised to the third power
arises from integration over all momentum space; it contains as many terms as
there are atoms in phase a intheconfigurationconsidered. The term
.NFPT similarly stands for the unnormalized configuration-space probability
density phase
It is useful to consider the projected probability density for phase a alone,
integrated over all possible configurations that phase can assume. This is
. (x,. . . , am)
I
,
. .N,-N:PT
= Pu
N P . . N:PT
(n?
-
= Pu
N Y . . .NzPT
Qp(N1 - K 7 - 7 Nc - K,P7 T ) (10)
where Q pis the isothermal-isobaric partition function phase at the indicated
numbers molecules, pressure, and temperature.
We now specialize to the case where component1 is present in phase only
in minute quantities. This would be expected, as the vapor pressure of a pure
MOLECULAR
long-chain polymer is extremely low:

NY = N I and N! =
Furthermore, we will assume that phase is very much larger in extent than a;
therefore, for all components other than 1,
NP c< Ni and NY = Ni (12)
In view of Eqs. (11) and (12), the logarithm of the quantity QP(Nl- NY, ...,
N, - N:, P, T ) appearing in (10) can be written as
In QP(O, NZ - N;, . . . ,N, - N:, P, T )
= -p@(O, NZ,. . . ,N,, P, + x

i=2
NY (T, P, )'X (13)
where stands for the Gibbs energy of the p phase and p! for the chemical
potential of component i in that phase. The composition vectorxP at which the
chemical potentials are evaluated has components
Substituting Eq. (13) into Eq. (lo), we can write the normalized configuration-
space distribution of phase a as
.. ( q , N;, . ..,NZ, W )
(a") n exp Y!(T, p, x"]
(15)
Letting the extent of phase become indefinitely large [Eq. (12)], we can sub-
stitute the upper limits of the summations over numbers of molecules in
(15) byinfinity.In this limit, the composition of phase p becomes constant,
given byEq. (14). This composition enters Eq. (15) exclusivelythrough the
chemical potentials p!. The probability density for phase a defined by Eq. (15)
is reminiscent of the grand canonical ensemble, the only difference being that
94 THEODOROU
the amount of polymer NI remains fixed atall times while the volume is allowed
to fluctuate. In applications it is more convenient to use the fugacities f!(T, P ,
)'X in place of the chemical potentials FB(T, P, xB) (Prausnitz et al., 1986).
Substituting N F m fromEq.
' ' ' we canrewrite (15) as
,,N~fg.. .lk
U (NZ7 * 7 K7 v", rU)
where
(17)
In Eqs. (16) and (17), ni is the number of atoms of which a molecule of species
i is composed. For a given combination of {Nq} values, the vector of atomic
positions r" has dimensionality Nqn,. is an intramolecular configu-
rational integral for a molecule of type i, calculated as
Zi"'" E
I exp
Note that the integral of

(al, . .. , ri ni-l)]d3rild3ri2 -
(18) is taken over only ni - 1 atomic positions;
d3rini-l (18)
the three degrees of freedom corresponding to overall translation of the molecule

are notintegratedover. The term is theintramolecularpotentialenergy of
a molecule of type in the ideal gas state the total potential energy of phase
is merelyasum of such terms. For a monatomic species, is simply 1. The
symbol beneath the integral over all atomic positions in Eq. (17) indicates that
the limits of integration for each atom coincide with the boundaries of phase a.
One should note that the c - 1 fugacities appearing in Eq. (17) are all functions
of the c - 2 independent mole fractions specifying the composition of phase
and of temperature and pressure. By the requirements of phase equilibrium, the
fugacities of all species are the same in phases a and p. In the following we
drop the symbol (Y with the understanding that we focus on the polymer phase.
Some condensation of notation can be achieved by introducing the quantity
has dimensions of (volume)"''. Rewriting the vector of atomic positions r as

(r,, r,, . . . , r,), where ri is the 3Nini-dimensional vector of atomic positions of
all molecules of type i, we recast Eq. (16) as
. ...,N,
(N2, V, rl, . .. , d
pNlf2 .' has dimensions of (volume)"-'Nci~lN'"', as expected from the fact that
it is a probability density in the variables {N2, ..
. . . , N,, V, rl, . , rc}. All
thermodynamic properties pertaining to the equilibrium at T and P between the
polymer phase, originally consisting of component 1, and the fluid phase, con-
sisting of components 2, 3, . . . ,N, at a fixed composition xp, can be extracted
from the probability density of (20).
In the following discussion we specialize to the case of a pure sorbate (c =
2). For greater clarity we use the subscripts P and in place of 1 (polymer)
and 2 (sorbate), respectively. The probability density function for the polymer
phase, Eq. (20), reduces to
V, r ~ r,A )
- exp(-PPv) exp [-Pv(rp,
The potential energy function of the polymer/sorbate system can be written

in general as
where the 3n,-dimensional vector rpiencompasses the position vectors of all

atoms constituting the macromolecule i and the 3n,-dimensional vector r, en-
96 THEODOROU
compasses the position vectors of all atoms in penetrant molecule k. A classical

flexible model (G6 and Scheraga, 1976) is assumed for the description of the
configuration of all molecules.
Of particular interest is the prediction of the sorption isotherm of A in the
polymer phase. This could be accomplished through a series of Monte Carlo
simulations in the NpfAPT ensemble, all carried out at the same amount of
polymer Npand temperature T. Each simulation would be at a different P value.
Given P and T, the fugacity f A is known through the equation of state of the
pure fluid sorbate A. The MC simulation would employ the following elemen-
tary moves: translation, rotation, and conformational rearrangement of a polymer
chain (bringing about changes in rpiand carried out as in a pure polymer sim-
ulation); translations, rotations, and conformational rearrangements of the sor-
bate molecules (bringing about changes in rh); insertion of an A molecule (in-
creasing N A by 1); deletion of an A molecule (decreasing by 1);and dilations/
contractions of the simulation box (changing The acceptance criteria to be
used with each of these moves can be extracted directly (Allen and Tildesley,
1987) from the probability density, Eq. (21).In fact, such an NpfAPT simulation
can be viewed as a hybrid between isothermal-isobaric and grand canonical MC
methods (Allen and Tildesley, 1987). Observables would include the average
volume of the polymer phase (providing a direct measure of swelling phe-
nomena) and the average number of sorbate molecules (N,,) present in the poly-
mer at eachpressure, providing the sorption isotherm.Results fromsuch a
Monte Carlo simulation approach for the direct prediction of the sorption iso-
therm have not yet been reported, but the approach hasbeen developed (Boone,
1995).
The slopeof the sorption isotherm in thelimit of very low pressures (Henry’s
law region) can be obtained through a simpler calculation. To derive an ex-
pression for the Henry’s constant, one can think as follows. From Eq. (21) the
average number of A particles present in the polymer phaseat equilibrium under
given Np,P, T is
r r
( N ~=) N~ dV d3Npprp d3NAnArA

p N ” f l ( ~ , , V, r,, rA)
NA=O
where
d3r, - - d3r,, exp[-pT(rp, rA1,. .., rmJ1

is the configurational integral in the isothermal-isobaric ensemble for Npmole-

cules of polymer and molecules ofpenetrant, withdimensions of
(VOIUme)NPnP+NWA+l
Consider now the polymer phase in the limit P -+ 0. In this limit, -,0,
andby Eq. + Equation (23) thenreduces to
1 dVexp (-pPV) I I
d3N*prp d3”ArAl
exp - pV:”(ril) - p‘VpA(rp,
I
dVexp(-pPV)d3N*prp
I
exp [-pVp(rp)] exp dxmA-’)r:,
The term V:h in Eq. (25) is the intramolecular potential energy of a single
sorbate molecule it is a function of only - 1 atomic coordinates, since it
is invariant to rigid translations of the molecule [compare Eq. (18) and following
discussion]. We have substituted the shorthand notation ril for the 3(n, -
dimensional vector (mll, rA,z,. . . , It follows that rAl = @Al,
The term ‘ V p encompasses all intramolecular and intermolecular interactions of
the macromolecules constituting the polymer matrix, while “ri, is the sum of
all intermolecular interactions between the penetrant and the polymer chains.
[Compare more explicit notation of Eq. (22).]
The average volumeof the pure polymer phase at T and P is obtained
the isothermal-isobaric ensemble as
-
-
1 dV exp (-PPI‘)
I exp [-
d3N’”prp pvp(rp)]
98 THEODOROU
where we have rewritten the volume V as an integral over the three-dimensional

domain spanned by the position vector rAlnA. Combining Eqs. (25) and (26), and
behaves as an ideal gas in the considered limit P -

recognizing that P/kB T is the molecular density c! in the fluid phase, which
we obtain
As defined in Eq. (27), S,, is a dimensionless partition coefficient equal to the

ratio of molecular concentrations of penetrant in the polymer phase and in the
pure sorbate phase in the limit P + it is a direct measure of the low-pressure
solubility of the penetrant in the polymer. Equation (27) expresses this partition
coefficient as a Widom “test particle insertion” average (Allen and Tildesley,
1987). The averaged quantity is the Boltzmann factor of the potential energy of
interaction between a single penetrant molecule and the polymer matrix. The
average is taken over all polymer configurations, weightedaccording to the NPT
ensemble of the pure polymer; over all internal configurations of the penetrant,
weighted by the Boltzmann factor of the corresponding intramolecular energy;
and over all translational degrees of freedom (positions of insertion of the pene-
trant in the polymer). The latter average is purely spatial; i.e., positions of in-
sertion are chosen randomly from within the pure polymer phase without the
polymer “feeling” the presence of the penetrant. Note that for a model system
in which the penetrant is spherical and interacts with all polymer atoms through
hard-sphere repulsive forces only, the partition coefficient of Eq. (27) would
reduce to the accessible volume fraction discussed in Section 1I.B.
The low-pressure solubility can be expressed per unit mass rather than per
unit volume ofpolymer. Invoking the definition of the Henry’s law constant
HhP(Prausnitz et al., 1986),
Equations (27) and (28) are useful for calculating the low-pressure solubility of
the penetrant through Monte Carlo or molecular dynamics simulations of the
pure polymer with Widom insertions of a “test” penetrant molecule.
Heat of mixing effects between the polymer and the penetrant can readilybe
analyzed in the NpfAPT ensemble. We define the differential heat of sorption of
the penetrant at given composition of the polymer phase as the negative of the
partial molar heat mixing:
Qs(T7 p7 XA) h!( T7 p ) - z~(T7p7
where h2 and SA are the molar enthalpy A in the (pure) fluid phase and the
partial molar enthalpy of A in the polymer phase, respectively. Straightforward
thermodynamic analysis leads to the Clausius-Clapeyron equation
where ze(T, P) = PvB/RTis the compressibility factor of pure A in the fluid

phase, is the partial molarvolume of A in the polymer phase, andthe notation
"eq" indicates that the derivative is taken under conditions of thermodynamic
equilibrium.
It is possible to calculate Qs from molecular-level information as
Qs = (h! - h&um) RT + + NAvo ('V!p)ig
where and
molar
the
areenthalpy
and average potential
energy
per molecule of pure A in the ideal gas state, NAvo is Avogadro's number, and
the brackets denote averagesin the NpfAPT ensemble.It is thus possible to
calculate the differential heat of sorption by monitoring correlations between the
number of penetrant molecules and the microscopic enthalpy in the course of a
simulation of the polymer phase carried out in this ensemble.
In the Henry's law region Eq. (31) reduces to
Q*,O
"
- k J + ('V.Y)ig+ W P+ PV)NpR
NAVO
- +"
Ir
+,
" I r Z + PV] exp(- PsrAp))widom
(exP(-P"IrM))wihm
In the above discussion of the statistical mechanics of sorption it was im-
plicitly assumed that the polymer matrix is in thermodynamic equilibrium,
that its configurations are distributed with a probability density proportional to
exp PP(rp)+ PV)]. If the matrix is glassy, this density distribution holds
separately within each a large number of disjoint regions in (rp,V) space, in
which the glass islocally"lockedin" (compare Fig. 1). pointedoutin
Section I.B, the relative weight with which each such region contributes to the
thermodynamics of sorption at low sorbate concentration dependson the history
of formation of the glass. In a simulation, for example, regions around local
l00 THEODOROU
minima generated via molecular mechanics through quenching melt configura-

tions are typically assigned equal weights. As the sorbate concentration rises,
more and more of these originally disjoint regions becomemutually accessible,
theconfiguration-space distribution equilibrating betweenthem.Whenthis
equilibration percolates through the configuration space of the polymer/penetrant
system, plasticization of the glass occurs. A well-designed Monte Carlo simu-
lation should capture this gradual unlocking of the glassy structure with the
moves introduced to sample configuration space.
B. A Theoretical Treatment of Sorption Thermodynamics

Based Static Model Configurations
In this and the following subsection we review existing theoretical and simu-
lation work on the prediction of sorption thermodynamics in polymer glasses
and melts, using the formalism of the previous subsection as a guide.
Gusev and Suter (1991) presented an early theoretical treatment of the sorp-
tion thermodynamics of small spherical gas molecules in glassy polymers that
employs atomistically detailed configurations generated by molecular mechanics.
The following assumptions are introduced:
The glassy polymer configurations are static; any effects of thermal motion
and of the presence of the penetrant on r, are neglected.
Penetrant molecules are assumed to reside within sites. A site is defined as
a region in the three-dimensional positionspace that encompasses a local
minimum of the potential energy olr(rAl;rp)under the considered fixed r,.
A given site can be occupied by at most one penetrant molecule; this ex-
clusion principle is the origin for the term “spatial Fermi gas” attributed
to the penetrant phase. Exchange of a given molecule between different
sites is possible.
All interactions between penetrant molecules occupying different sites in
the polymer are neglected; i.e., all terms olry of the potential energy func-
tion of Eq. are set to zero.
With these assumptions, the configurational integrals in Eq. can be substi-
tuted by
where K is the total number of sites in the considered polymer matrix and is
the configurational integral for site j , calculated as
MOLECULAR
SIMULATIONS OF SORPTION AND DIFFUSION l01
The primes on the summation symbols in Eq. (32) indicate that the sites (jl,j2’
. . . jNJ must all be different from each other.
Upon substitution in Eq. Eq. gives
(34)
j=
For the structureless spherical sorbates considered here,
Thus, the sorption isotherm emergesfrom the Gusev-Suter theoretical approach

as a sum of Langmuir terms, each associated with a particular site in the static
polymermatrix.Different sites are characterized by different bp The overall
Henry’s constant is inverselyproportionalto Xzlbp As long as bj << 1, the
Langmuir term contributed by site j can be approximated by a straight line. The
predictedisothermthen is expected to look like a sum of Henry’s lawand
Langmuir terms, as proposed by the “dual-mode sorption” model (Petropoulos,
1994) at not too high pressures. Thus, the energetic and entropic heterogeneity
of the sites presented by the polymer to the penetrant is proposed as the origin
of dual-mode sorption behavior. Gusev and Suter applied this formalism to the
system methane/bisphenolA polycarbonate. Sites j were identified bycomputing
T(rAl; rp) at the nodes of a three-dimensional grid of lo7points running through
each static structure. Theborders of sites were determinedthroughthree-
dimensional steepest descent constructions initiated at each of the grid points.
(See also SectionN.D.)The site configurational integrals of Eq. were
obtainednumerically from thepotentialenergy values at the grid points.In
describing the dispersive part of the interaction between penetrant and polymer
atoms, Gusevand Suter introduced an adjustable parameter kM that characterizes
deviations from the Lorentz-Berthelot rule. Reasonable agreement with exper-
iment at 300 K and pressures up to 60 bar was obtained with kM = 0.5. The
total sorbed amountwas broken up into contributions from families of sites with
different l/bj, according to Eq. (34). Sites with llbj in the range 50-500 bar
were found to be most important; this range lies significantly higher than the
102 THEODOROU
value l/bj == 10 bar obtained from the dual-mode sorption model. Clearly, the
Gusev-Sutermodelcannotpredict swelling andplasticizationphenomenaat
high penetrant activities, because it postulates a static polymer matrix. The quan-
tity kM is little more than a fitting parameter that absorbs inadequacies of the
approximations invoked. Nevertheless, the model constitutes an interesting and
computationally inexpensive first attempt to capture the physics of high-occu-
pancy sorption in glassy polymers through molecular modeling.
C. Molecular Dynamics and Monte Carlo Simulations of

Sorption Equilibria
We now discuss attempts to predict the solubility ofsmall penetrants in polymer
glasses and melts through atomistic simulations that provide a full numerical
solution to the statistical mechanical formalism of phase equilibrium discussed
above. Simulation work on solubility has been far less extensive than on dif-
fusivity; this is somewhat surprising, as a good understanding of thermodynam-
ics is usually a prerequisite for dynamical studies.
Muller-Plathe (1991b) was the first to apply the Widom insertion technique
to estimate the solubility So through the canonical ensemble counterpart of
(27). He studied H,, NZ, and Oz, represented explicitly as diatomic dumbbells,
within atactic polypropylene at 300 K, which is roughly 45 K above the ex-
perimental glass temperature. Samples of the amorphous polymer were gener-
ated by molecular mechanics, relaxed with NVT MD for 200 PS, and then sub-
jected to a 2 ns longNVT MD production run; configurations fromthis run were
used for the Widom insertions. Results were reported in theform of free energies
per= -k,T In S,. Predicted values of pa were found to decrease in the correct
order He > H, > NZ > 0, > CH, (order of increasing condensability). The
absolute values of pa, however, were lower (more favorable for sorption) by
5-10 kJ/molthan values obtainedfromexperimental data fromsorptionin
ethylene-propylene copolymer and in the amorphous fraction of isotactic poly-
propylene. In fact, for NZ and the more condensable gases it is predicted that
So > 1, contrary to the experimental finding S,, 1. Differences in pa between
different penetrants were in more reasonable agreement with experiment. All
results were based on a single starting configuration for the polymer. A more
reliable approach would be to average over many different starting configura-
tions; it seems unlikely, however, that suchaveraging would bring the computed
estimates much closer to experimental values. The disparity between prediction
and experiment could result from the use of inaccurate potential parameters. In
fact, more recent work (Boone, 1995) indicates that the parameter set used by
Muller-Plathe (1991b) for polypropylene is too attractive. Whether the potential
PVT behavior
parameterization is satisfactory could be judged from the predicted
ofthepurepolymer;unfortunately,however,pressureresultsarenotreported
MOLECULAR
along with the sorption study.The computational effort needed for estimating the
solubility increases significantly with increasing penetrant size, as expected from
the much smaller volumes accessible to larger penetrants (compare Section 1I.B).
Miiller-Plathe finds that decreases linearly with the Lennard-Jones well depth
of penetrant-penetrant interactions. This correlation is reminiscent of the linear
relation between the logarithm of low-pressure solubility and the penetrant’s heat
of vaporization suggested by regular solution theory (Petropoulos, 1994).
Sok et al. (1992) also report computations of the solubility gaseous He
and CH, in a polydimethylsiloxane (PDMS) liquid in the Henry’s law region.
S, and were computed through Widom insertions [see Eq. (27)] in a model
PDMS liquid of degree of polymerization 30 undergoing NPT MD simulation.
To enhance the efficiency Widom insertions, a map of the accessible volume
was maintained using a 100 100 100cubic grid running throughthe
polymer. As in the work of Miiller-Plathe, predicted values were too low
(attractive) by 7.8 kJ/mol for CH, and by 3.6 kJ/mol for He. In contrast to the
solubilities, diffusivities were predicted in excellent agreement with experiment.
(Seealso Section NB.) Qualitatively,the simulation reproduced the experi-
mental finding that the diffusivity He in PDMS exceeds that of CH,, but the
overall permeability of CH, is higher owing to its higher solubility.
remarkable set Monte Carlo computations of alkane solubilities in a
polyethylene melt was conducted by de Pablo et al. (1993). In these studies the
linear polyethylene matrix was represented as a liquid of GI chains undergoing
NPT MC simulation, and penetrants in the range were examined. The
solubilityinthe Henry’s lawregion was computed by theWidominsertion
technique, Eq. (28). For the penetrant sizes examined (de Pablo et al., 1993), a
random insertion will almost certainly lead to severely repulsive overlaps with
the polymer chains and thus contribute negligibly to the ensemble averages of
(28); an extremely large number of random insertions would have to be
carried out to obtain a good estimate of the solubility. remedy this problem,
de Pablo et al. employed biased insertions based on the CCB scheme. Rather
than being thrown randomly into the matrix, the articulated penetrant molecule
is “threaded,” bond by bond, through accessible regions of the matrix; Eq. (28)
can be straightforwardly recastinawaythat takes into account thisbiased
sampling and thus serve as a basis for an efficient computation of the solubility.
Figure 8 displays the weight fraction Henry’s constant H , = HkpMp/MA(where
MPand M,, are the molecular weights of the polymer and the penetrant, respec-
tively) estimated by de Pablo et al. (1993) at bar and temperatures of 423 and
513 K as a function of the penetrant chain length. The simulation estimates,
based on a potential parameterization that reproduces thePVT properties of long
alkanes, are inexcellentagreementwithexperiment. The same authors con-
ducted calculations solubility at higher pressures, outside the Henry’s law
regime. In these calculations they used Panagiotopoulos’s Gibbs ensemble, de-
104 THEODOROU
Data of Maloney Prausnitz(1976)

A CCB ghost molccule. NIT ensemble
I I I
4 6 8 10
Number of Carbon Atoms
Henry’s constants for alkanes in molten polyethylene as predicted through
continuum-configurational-bias Widom insertions within a polymer matrix undergoing
NPT MC simulation (de Pablo et al., 1993) and as measured experimentally by Maloney
and Prausnitz. The abscissa is the number of carbon atoms in the penetrant molecule.
signed to sample the configuration-space probability density of (8) for two

phases coexisting at equilibrium at specifiedpressure,temperature,and total
amounts of all species present. In order to make the particle exchange moves
between the two phases that are required by this method computationally fea-
sible, de Pablo et al. used the CCB scheme [CCBG method (Laso et al., 1992)l.
The bias introduced by the use of CCB-based exchanges is removed by ap-
propriatedesign of the Monte Carlo selection criteria. Computations the
solubility of pentane in polyethylene by CCBG clearly show deviations from
Henry’s law at pressures above 10 bar.
PREDICTION
Statistical Mechanics of Diffusion
A thorough review of the statistical mechanical formulation diffusion is given
by MacElroy (Chapter 1, this volume). Inthis section we summarize somebasic
relations that are useful in extracting the binary diffusivity and the self-diffusiv-
ity from molecular simulations.
Consider a binary system consisting of a polymer P and a penetrant A. The
entire system is at a uniform temperatureand pressure and free of any external
force fields. The system as a whole does not move; i.e., macroscopically, the
center of mass the system remains fixed in space. There is, however, a mac-
roscopic spatial variation in composition that keeps the system away from ther-
modynamic equilibrium and sets up macroscopically observable fluxes of the
two components. We use the symbol (i = P) to denote the macroscopic
flux of component i in a coordinate framethat remains fixed with respect to the
center of mass of the system; this coordinate frame is usedthroughout our
analysis. is measured in molecules per square meter per second. We use the
symbol to denote the microscopic (molecular) current of component i, that is,
the mechanical quantity
where vei is the center-of-mass velocity vector of molecule of species i and

Ni is the total number of molecules of species i at a given time within a mac-
roscopic volume element of the system on which we focus ouranalysis. Linear
response theory leads to the following equation for the thermodynamic flux
~ ( V F B ) ~ where
in terms of the thermodynamic forces ( V P A ) ~and ~, stands
for the chemical potential of species i, in joules per molecule, and V is the
volume of the considered macroscopic elementof the system:
The angular brackets in E q . denote ensemble averages in an equilibrium

system that finds itself at the same temperature, pressure, and average compo-
sition as the considered volume element of the nonequilibrium system at the
considered time. Equation (37) follows directly from combining Eqs. (10) and
(14a) of MacElroy (Chapter 1) for the case considered here.
It is useful to cast in terms of ( V F A ) ~only.
, ~ The Gibbs-Duhem equation,
applied to the considered macroscopic element of the system (assumed to be in
local thermodynamic equilibrium) leads to
(vFA)T,P + XP (vFP)T.P =
where is the mole fraction of species i. Combining Eqs. and (38) yields
-JA = -
1
V ksTxp [XP 1( W ) .i,(t)) dt - xA 1 dt] (VCLA)T,.P
(j.40) .jp(i)>
An entirely analogous equation holds for component P.

106 THEODOROU
Since there is no macroscopic displacement of mass (flow) in the considered

volume element, the fluxes J A and JPmust satisfy the condition
with m, and mp the masses of a penetrant molecule and a polymer molecule,

respectively. Using Eqs. (39) and (40), one can translate the macroscopic con-
dition, Eq. (41), into the microscopic form
The binary d i h i v i t y or transport diffusivity D in our system is defined

through Fick's first law (de Groot and Mazur, 1984; Bird et al., 1960):
where is the mass density and wi is the mass fraction of species i inthe
considered macroscopic volume element.
Comparing Eqs. (39) and (43) and substituting the chemical potential p,A in
terms of the fugacity f A (Prausnitz et al., 1986), we obtain a microscopic ex-
pression for the binary diffusivity:
The binary diffusivity emerges as a product of a term that depends on the ther-
modynamics of sorption and the system composition and a second term that
incorporates the time integrals of time correlation functions of the microscopic
currents of the two components.
A more symmetric expression for D can be obtained in terms of the micro-
scopic interdifjizsion current
This expression is
+
where N = NA Np. A form of Eq. (46) was presented by Hansen and McDonald
(1986). (Note that a factor of 1/N is missing from the equation given in that
reference.) The equivalence of Eqs. (44) and (46) can be readily shown on the
MOLECULAR SIMULATIONS
OF
SORPTION AND DIFFUSION 107
-
basis of Eq. (42), taking into account that the equilibrium average(i,(O) j,(t))
is an even function of time.
Equation (46) for the binary diffusivity can be converted from the Green-
Kubo form to the corresponding Einstein form (Hansen and McDonald, 1986)
by invoking the general equilibrium equality
where d is any scalar dynamical quantity and the dots denote time derivatives.
The Einstein form of Eq. (46) is
with rm,&) the center of mass of all molecules of species i at time t. The Einstein
forms are generally preferable to the Green-Kubo forms for use in equilibrium
MD simulations, as they circumvent the need to integrate the time correlation
functions, whose long-time tails suffer from noise due to limited sample size
(Allen and Tidesley, 1987).
The self-dimivity of species i provides a measure for the displacement of
individual molecules of i as a result of random thermal motion. In a system
exhibiting diffusive behavior, the meansquare displacement of a molecule
linearly with time at long times. The self-diffusivity is extracted from the pro-
portionality constant in that relationship:
In extracting from an equilibrium MD simulation, one usually averages Eq.

(49) over all molecules 4 of species i. Thus, one has a much larger sample size
than is available for the calculation of the collective property D through Eq.
(48). Self-difhsivities cantherefore be obtainedwith much better accuracy than
binary diffusivities from equilibrium dynamic simulations. The Einstein form,
Eq. (49), is equivalent to the Green-Kubo form [compare Eq. (47)]:
The velocity autocorrelation function appearing within the time integral of Eq.
(50) expresses how the penetrant molecule “loses memory” of its original mo-
tional state through interactions (collisions) with the polymermatrix and (at high
108 THEODOROU
concentrations) with other penetrant molecules.A plot of the normalized velocity

autocorrelation function (veA(0) VeA(t))/(lVeJ), as accumulated from equilib-
rium MD simulations of methane in polyethylene (PE) (Pant and Boyd, 1993),
is shown in Figure 9.
Since self-diffusivities can be obtained much more readily from simulation,
there is incentive for expressing D in terms of D,, and D,,p.Substituting the
microscopic currents into Eq. (44) from their definition, (36), we obtain
Consider the system in the limit where A is infinitely dilute in P * 0).

In this limit, the cross-correlation terms between the two species in Eq. (51)
drop. Also, different molecules species A are very far apart; thus, one would
expect the velocity correlation terms between different molecules of species A
in Eq. (51) to reduce to zero. Furthermore, the thermodynamics of the polymer/
penetrant system in this limit conforms to Henry's law (see Section 111), and
thus the thermodynamic derivative in (51) reduces to unity. Invoking Eq.
(50), we conclude that
lim D = lim D,,

wn-0
Equation (52) is a rigorous statistical mechanical result. approximate relation

between the binary and the self-diffusivities at any composition may be arrived
at by assuming that all velocity correlations between different molecules in Eq.
(51) (be they of the same different species) are negligible. This assumption
leads to
Under the same approximation, (53) holds with the subscripts P and A
reversed. Combining this counterpart of Eq. (53) with (53), and recognizing
OF
- 300K
-0.5 c
0 0.4 0.8
t (PS)
1.2 1.6 2
Normalizedvelocityautocorrelationfunctions for amethanepenetrantin

amorphous polyethylene, as accumulated from PS long trajectories at tempera-
tures. The sign reversal at short times results from the first collision of the penetrant with
the cage of polymer atoms surrounding it; it is much sharper at the lower temperature,
where the cageis less mobile. The undulations at long times make clear the computational
difficulties in calculating the self-diffusivity as a time integral of the velocity autocor-
relationfunctionthroughtheGreen-Kuhorelation.[ReproducedfromPantandBoyd
with permission.]
that the thermodynamic correction term (a In $/a In xi)T,pis the same for both
species (Gibbs-Duhem equation), one readily arrives at
Equation (54) is used frequently to express the composition dependenceof the

binary diffusivity in liquid mixtures.
In a polymer/penetrant system, the self-diffusivity Ds,pof polymer chains is
typically much smaller than that of penetrant molecules; its contribution to
(54) can thus be neglected. Equation (53) can be rewritten in a more practically
useful form as
110 THEODOROU
The self-diffusivity of the penetrant, D,, plays a dominant role in shaping the
binary diffusivity.
EquilibriumMolecularDynamicsSimulations
SimplifiedPolyethylene-LikeModels
Over the past few years there has been much interest in conducting equilibrium
MD simulations of amorphous polymer/penetrant systems for the purpose of
predicting the self-diffusivity Ds,* and elucidating microscopic mechanistic as-
pects of diffusion. For small molecular weight penetrants in melts or rubbery
polymers, diffusivities are often high enough to be captured withinthe maximum
time spans that can be simulated with conventional molecular dynamics (tens
of nanoseconds) with present-day computational resources.
Early MD studies of gas diffusion employed united-atom alkane- or PE-like
models for the polymer. They did not lay much emphasis on the calibration of
potential parameters; in fact, the PVT behavior that these simulations yielded
was in poor agreement with experiment (density too low for given pressure and
temperature), and predicted values of diffusion coefficients were quite far from
experimental values. Nevertheless, they revealed interesting relations between
self-diffusivity and polymer structure. Takeuchi and Okazaki carried out
microcanonical MD simulations of 20 small spherical moleculesresembling
oxygen in 30 united-atom 20-mer molecules of “polymethylene.” The model
representation of RigbyandRoe was used for the chain
molecules. The predicted self-difisivity was studied as a function of tempera-
ture, unoccupied volume, and chain stiffness. The activation energy for self-
diffusion under constant density was found to be approximately twice as high
as the activation energyunderconstantpressure, indicating that the thermal
expansion of the polymer (increase inaccessible volume with temperature) dras-
tically affects diffusion. The activation energy underconstant pressure was found
to correlate linearly with the preexponential factor in the Arrhenius expression
for the self-diffusivity, as experimentally observed for many polymers. Frac-
tional unoccupied volumes, v , were accumulated by subtracting an estimate for
the hard-core volume of chains from the total volume of the model system. The
self-diffusivity was foundto correlate well with v, according to the simple Fujita
equation
D,, = Ad exp (- &/vf) (56)
(see Fig. The relaxation frequency of internal rotations, which is on
the order of S, also exhibits a dependence on v, of the form of Eq. (56);
this dependence,however, is considerablyweakerthanthat of the self-
diffusivity.
Under the high-temperature melt conditions corresponding tothe simulations
of Takeuchi and Okazaki, T+ seems dominated by the height of intramolecular
MOLECULAR SIMULATIONS OF
t
2.4 2.8 3.2
l / Vf
Dependence on theself-diffusioncoefficientofsmall,oxygen-likemole-
cules in a polyethylene-like liquid on the fractional free volume of the liquid, as obtained
from equilibrium molecular dynamics simulations. All data are at the same temperature
andare given inreducedunits.[ReproducedfromTakeuchiandOkazaki with
permission.]
torsional barriers. To study the influence of the dynamic flexibility of chains on

self-diffusion, Takeuchi and Okazaki performed simulations in a model chain
liquid the same density, where the torsional barriers were artificially removed.
T,,,inthismodified chain liquid was found to be shorter by more than three
orders magnitude and to exhibit an activation energy one fifth of that exhib-
ited in the presence of the barriers. Upon removal of the torsional barriers, the
activation energy for self-diffusion under constant density dropped by a factor
of 2, although the overall fractional unoccupied volumes v, in the original system
and in the barrier-less system were almost identical. similar decrease in the
activation energy for diffusion was observed under conditions of constant pres-
sure (see Fig. 11). The diffusivity in the high-temperature melt was thus found
to depend on the dynamics of redistribution of unoccupied volume and not only
on the amount of unoccupied volume. interpretation of these results in the
light of the Brandt theory of diffusion was proposed by Petropoulos (1994).
The effects of matrix chain lengthwere briefly examinedby Takeuchi (1990b)
by comparing the results the above simulations with MD trajectories in a PE-
112 THEODOROU
0.19 0.25
1/T
Temperaturedependence of the self-diffusivity of oxygen-like molecules

in a polyethylene-like liquid subject to full torsional barriers (open symbols), and in a
similar liquid of freely rotating chains torsional barriers, filled symbols). All data are
atthesamepressureandare giveninreducedunits.[ReproducedfromTakeuchiand
Okazaki (1990), with permission.]
like matrix of infinite chain length at the same composition, number density,
and temperature. After correcting for the difference in unoccupied volume be-
tween the two systems, D,, and T+ were found to be lower by a factor of 2 in
the long-chain system. The changein was thought to stem in part from the
difference in chain packing between the two systems.
To elucidate the effects of chain packing and the concomitant distribution of
unoccupied volume on penetrant self-difhsivity, Takeuchi et al. (1990) under-
took MD simulations in a series of infinite-chain matrices in which the equilib-
rium value of the bond angle 8 was varied systematically between 100' and
150". All these "mutated" PE-like systems and the original system (equilibrium
0 = 109.5') were studied under the same overall fractional unoccupied volume
v, and temperature. Changes in 8 induced significant changes in the packing of
chains. Increasing brought about a decrease in the most probable intermolec-
ular distance r,,, between mers andsharpened the structure of the polymer
liquid, as seen from the intermolecular pair distribution functions. The tendency
OF
for parallel alignment of neighboring chains also increased, as seen from the
intermolecular orientation correlation function for skeletal chords. In other
words, opening up the bond angles resulted in a tendency for chains to bundle
together more tightly. As pointed out by Takeuchi et al. (1990), however, the
conformational distribution of the chains in these simulations was nottruly
equilibrated; it was strongly affected by the artificial boundary condition used
to create the infinite-chain models (see Fig. 2b). A local unoccupiedvolume
fraction &(r) was accumulated as a function of distance r from the center of a
reference mer; to define fv(r), the average hard-core volume of mers within a
shell centered at the reference mer of inner radius equal to the mer radius and
outer radius r, was subtracted from the volume of the shell. The value v:
fv(rm) was used as a basis for comparing the unoccupied volume distributions
of the different polymer models. v: was found to be a decreasing function of
8, reflecting the tight, parallel alignment of chains at large 8 values. The pen-
etrant self-diffusion coefficient was found to increase roughly linearly with the
chain spacing r,,, and to obey a Fujita-type equation of the form of (56)
as a function of v:. Given that segmental mobility, as quantified through was
comparable among all model polymers, this effect was viewed as a consequence
of differences in the spatial distribution of unoccupied volume, as opposed to
differences in its dynamics.
attempt to correlate with the accessible volume distribution of the
polymer matrix at a probe radius corresponding to percolation was presented by
Takeuchi and Okazaki (1993a). (See Section 1I.B.) MD simulations of diffusion
of two different spherical penetrants (roughly corresponding to 0, and He) were
performed in three different infinite-chain polymer matrices (characterized by
different 8 values) over a variety of densities (therefore, of free-volume frac-
tions). In parallel, the accessible volume distributions at a penetrant radius r,
corresponding to percolation in each polymer matrix were analyzed based on
MD simulations of the pure matrix. For each penetrant, the self-diffusivity was
found to correlate with the total number of clusters n, present at percolation as
D,, = A, exp (-B,, n,), where the constants A,, and B, depend on the penetrant
but not on the matrix (i.e., are universal for all matrices examined). The number-
weighted density distribution of cluster volumes v at percolation was found to
depend on v asThis result is suspect, as it indicates that the volume-
weighted density distribution of clusters pv(v;r,)should be anincreasing function
of v, contradicting recent extensive studies in a similar polymer (Greenfield and
Theodorou, 1993). Subject to this caveat, the quantity log (Ds, min) was found
vary linearly with the fraction a of accessible volume at percolation that is
contained in clusters larger than the hard-core volume of the penetrant. Data
from both penetrants in all polymers collapsed onto a single line when plotted
in this way. This result was interpreted as lending support to the free-volume
theory, according to which D,, is proportional to both the kinetic velocity of
114 THEODOROU
the penetrant and the probability of finding an accessible hole of sufficient vol-
ume for the penetrant to hop into.
Molecular dynamics simulations of diffusion of spherical CO,-like molecules
in a PE-like matrix corresponding to the model of Rigby and Roe were also
carried out by Trohalaki et al. (1989, 1991, 1992). The formulationfor obtaining
the binary diffusivity from the self-diffusivities of penetrant and polymer was
applied in this work.
Sonnenburg et al. (1990) simulated the diffusion of spherical penetrants in
an idealized polymer network in the rubbery regime. The temperature depen-
dence of D exhibited Arrhenius behavior, while its dependence on the penetrant
diameter agreed qualitatively with experimental observations. The diffusion ex-
hibited stronger temperaturedependence in the network than in thesimple liquid
formed by eliminating all covalent bonds. At low density values, diffusion was
faster in the simple liquid than in the network, as one would expect from the
reduction in atomic mobility caused by the covalent bonds. At a certain value
of the density, however, a crossoveroccurred. Beyond that value, diffusion pro-
ceeds at a measurable rate in the network, while it is practically absent in the
simple liquid. This somewhat unintuitive result seems to originate in the intrin-
sically more inhomogeneous distribution of accessible volume in a polymer as
compared to a monomeric substance (compare Fig. 5 and associated discussion
in Section 1I.B.)
2. More Realistic Melt Models

The latest equilibrium MD work on gas diffusion in polymer melts and rubbery
polymers has been characterized bya trend to use more elaborate representations
the polymer matrix and penetrant species, in an effort to predict self-diffusion
coefficients quantitatively. In parallel, the mechanismof motion of the penetrant
has been explored through detailed analysis of MD trajectories.
Muller-Plathe (1991a) simulated 20 methane molecules in a system of 20
united-atom pentacontane (Cso)chains at 300 K using the parameterization pro-
videdby a commercial simulation package. The configurations of the chain
liquid were relaxed through energy minimization and MD. Diffusion was found
to proceed through a “hopping” mechanism, the displacement of the penetrant
occurring mainlythroughinfrequent jumps oflength less thanthepenetrant
diameter. The duration of a jump event (a few picoseconds) was very short
relative to the time spent between jumps. Fast recrossings of a penetrant back
into its original position following a jump were also observed. The infrequent
nature of jump events makes the estimation of the self-diffusivity problematic,
as very long simulations would be required to accumulate sufficient statistics.
To address this problem, Muller-Platheconsidered ascaling factor A multiplying
the penetrant/polymer potential The logarithm of D,, was found to decrease
linearly with A in the range 0.1 A 1.0, and the associated correlation is
OF
proposed as a means for estimating diffusional characteristics efficiently through

simulation. A more recent investigation (Muller-Plathe, 1992) focused on H,,
O,, and CH., in atactic polypropylene at 300 K, generated by an MM method
('I%eodorouand Suter, 1985) and subsequently relaxed by molecular dynamics.
In comparison with PE, this polymer has the advantage being noncrystalliz-
able, which makes the validation the simulation results straightforward, sub-
ject to the availability of experimental data. A rather detailed model was used,
with united-atom methyl groups but with all other atoms treated explicitly, and
production runs as long as 2 ns were conducted. Of the three penetrants studied,
the largest (CH,) exhibited the most expressed hopping motion; the motion of
H, was more akin to a continuous sequence of small random displacements, as
one finds in a liquid. (See Fig. 16 of MacElroy, Chapter 1, this volume.) The
logarithm the predicted self-diffusion coefficient was found to drop almost
linearly with increasing penetrant diameter, in agreement with some empirical
correlations. Comparison with available experimental data from atactic polypro-
pylene and similar polymers led to the conclusion that the diffusivities of H,
and 0, were overpredictedby factors roughly 8 and 3, respectively, whereas
that CH, was very close to experimental. This trend improved agreement
with experiment with increasing penetrant radius led Muller-Plathe et al. (1992a)
to investigate the influence of the potential representation of the polymer on the
estimated diffusion coefficients. Predictions for the self-diffusivity of 0, were
compared in amorphouspolyisobutyleneusing a fully explicit representation and
a representation wherein methylsare lumped into united atoms. Itwas concluded
that the use of the united-atom approximation leads to an overestimationof the
self-diffusivity and that a fully explicit polymer model is necessary for estimat-
ing thediffusivitycorrectly. The interpretationproposed is thatunited-atom
groups give rise to artificially large interstitial sites within the polymer, thereby
accelerating diffusion. This sensitivity to the polymer representation should be
greatest for small penetrants that can explore small interstitial cavities and should
away for larger penetrants; this seems supported by the fact that the diffu-
sivity of CH,was correctly captured in a polypropylene model with united-atom
methyls. No information is given, however, on whetherthe models invoked were
capable of reproducing the experimentalPVT thermodynamicsof the pure poly-
mer, which should be a prerequisite for any simulation that aims at capturing
sorption and diffusion phenomena quantitatively.
Sok et al. (1992) simulated the transport of He and CH., in polydimethylsi-
loxane (PDMS) at 300 K. A rather detailed model with united-atom methyls but
with all other atoms explicit, incorporating both Lennard-Jones and Coulombic
interactions among partial charges, was employed for the chains. In the flexible,
mobile PDMS, significant diffusion can be observed over relatively short MD
trajectories (= 250 PS). The penetrants (especially methane)exhibited a jumplike
process analogous to the one observed by Muller-Plathe. Through analysis and
116 THEODOROU
visualization of trajectories, it was confirmed that the penetrant spends most of

its time within fluctuating cavities in the polymer liquid. A jump occurs when
fluctuations in the chains momentarily open a channel between two cavities.
Eliminating the motion of the chains (i.e., freezing the polymer configuration)
led to trapping of both penetrants, which were only able to “rattle” within their
cavities. The importance of fluctuations of the polymer matrix for diffusion was
thus confirmed. The predicted self-diffbsivities were in excellent agreement with
experiment (especially for CH,), although, as mentioned in Section III.B, the
predicted solubilities were much less satisfactory.
The motion of penetrant molecules within cavities and the jumps between
cavities have been probed in detail for a two-site O2 model in polyisobutylene
(PIB) by Miiller-Plathe and van Gunsteren (1992). Cavities were found to be
wide enough to allow relatively uninhibited motion of the 0, molecule while at
the same time firmly opposing its long-range translational motion. Penetrant
molecules participating in a “jump” (defined as a translation by more than 5
A over 1 ps) were found to be translationally “hotter” than average by 350 ?
100 K, which is taken as an indication that the molecule has to overcome an
energy barrier to perform the jump. This interpretation is not conclusive, how-
ever, as jumping molecules fulfil a minimum velocity requirement and are there-
fore hotter by definition. Both end-on hops (with the molecular axis parallel to
the direction of translational motion) and edge-on hops were observed.
Boyd and Pant (1991) studied the diffusion of methane in PIB and PE using
a united-atom representation. Model matrices consisting of 24 skeletal unit long
PIB and PE chains were created starting from a crystalline arrangement by
reptation-based NPT MC (Boyd, 1989) followed by NVT MD. The simulations
allowed for interesting structural comparisons between the two polymers.
Whereas the intermolecular pair distribution functions in PE reveal a distance
-
of 5.5 A between the backbones of neighboring chains, the corresponding
distance in PIB exceeds 7 A. Chains are compared to cylinders with a dense
core and less dense surface region; the larger diameter of the core in PIB ex-
plains the better packing and higher mass density in that molecule. The predicted
self-diffusion coefficients were found to be very sensitive to the matrix density.
Diffusion was predicted to be slower in PIB than in PE, as seen experimentally.
The absolute values of D, however, were overpredicted and the corresponding
activation energies underpredicted by roughly a factor of 4. To investigate this
disparity between predicted and observed values of the diffusivity in recent MD
work, Pant and Boyd (1992) conducted a study of both the diffusivity and the
polymer melt’s equation-of-state behavior as a function of the potential param-
eterization used for the chains. Chains of infinite molecular weight were used
in this investigation. The conclusion was that past united-atom simulations of
PE had been conducted with potentials leading to incorrect equation-of-state
behavior (e.g., highly negative pressures at experimental densities); the use of
an anisotropic united-atom (AUA) potential that can capture the PVT behavior
of the matrix was found to also give very reasonable values for the self-diffu-
sivity. Thus, the ability of a potential to describe the thermodynamic properties
of the pure polymer was established as a prerequisite for the quantitative pre-
diction of penetrant diffusivity. A thorough investigation of methane diffusion
in PE and PIB at a variety of temperatures, using an AUA potential represen-
tation that can correctly capture PVT properties, was presented by Pant and Boyd
(1993). The predicted diffusivities and activation energies in both polymers are
in excellent agreement with experiment. (In the case of low-temperature PE, an
analysis based on effective medium theory was conducted to extract the diffu-
sivity in amorphous regions from experimental data using the crystalline volume
fraction of the polymer.) Diffusivities in amorphous PE were found to follow a
distinctly non-Arrhenius temperature dependence that can be described by a
WLF-type expression (see Fig. 12). Diffusivities in PIB, on the other hand,
exhibited an Arrhenius temperature dependence. The mean square displacement
(r') as a function of time was found to exhibit three regimes. At very short
times (tenths of a picosecond), fast ballistic motion of the penetrant is observed
between collisions with polymer atoms ( ( r 2 )a t'), which causes the mean square
displacement to rise very quickly to a finite offset. This is followed by a rela-
L
n
0
Q)
-r2 -9.0 7
m
W
E - 0 BPE (MB)
-9.5 0 BPE (LWH)
x BPE (KH)
-10.0 : LPE (MB)
+ LPE (L)
-10.5 0 simulation
WLF f i t
- 1 1.0
-11.5
1.5 2 2.5 3 3.5 4
1000/T
Figure 12 Diffusion constants for methane in amorphous PE from both experiment
and MD simulation. Filled circles: simulation results. Other symbols: experimental values
from various sources, corrected for crystallinity wherever necessa-ry. The temperature
dependence of both simulation and experimental points is of a WLF type. [Reproduced
from Pant and Boyd (1993), with permission.]
118 THEODOROU
tively extended regime, wherein the mean square displacement growsless than
linearly with time (i.e., (r’) f with v < l),but where the rate of growth d(r2)/
dt exceeds that observed in the long-time diffusive regime (6D,,,). The origin
of this subdiffusive region, which Pant and Boyd refer to as the “cage effect”
region, is discussed in more detail in Section W.D. At low temperatures (e.g.,
280 K inPE), this subdiffusiveregion extends out to times on the order of
200-300 ps ((?)ln = 3-4 A), beyond which a clearly diffusive region (d(?)/
dt = const. = 6D,$ sets in. Pant and Boyd (1993) undertake a very informative
analysis of the penetrant’s motion as a function of temperature. To free the data
of the noise associated with “rattling” within clusters of accessible volume,
they average the penetrant position r,, over successivetime intervals of duration
They show that if is chosen long enough, successive changesin the averaged
position of the penetrant are uncorrelated, i.e., the averaged trajectory behaves
like a random walk; 7 values of 16 ps at 300 K and of 7.5 ps at 400 K are
sufficient for this to happen. Plots of the squared displacementof the penetrant
between successive steps along such an averaged trajectory appear strikingly
different at different temperatures. In the low-temperature melt (Fig. 13a), one
sees infrequent large jumps (as long as 6 in the figure) separated by long
periods of quiescence. Jumps longer than the radius of the penetrant account for
76% of the total diffusive progress of methane in PE at 280 K In the high-
temperature melt, on the other hand (Fig. 13b), jumplike displacements of the
penetrant become very frequent. The penetrant is no longer trapped in clusters
of accessible volume for long periods of time but rather seems to be carried
along by rapid fluctuations of the accessible volume. (Compare also Fig. 6 and
associated discussion.) This change in mechanism of the diffusion process with
changing temperature in an amorphous matrix, identified by Pant and Boyd, is
one of the most important results obtained through atomistic MD simulations
of amorphous polymers. Incontrast to PE, PIB displayeda behavior comparable
to that illustrated in Fig. 13a throughout the temperature range studied (350-
600 K). The rate of redistribution of accessible volume in that polymer was very
slow, and the onset of the diffusive regime took longer to be established than
in PE at comparable temperatures.
C. NonequilibriumMolecularDynamics
The simulation method of nonequilibrium molecular dynamics (NEMD) was
applied for the first time only very recently to polymer/penetrant systems by
Muller-Plathe et al. (1993). In contrast to equilibrium MD, NEMD imposes an
external driving force that keeps the model system away from equilibrium and
measures directly the flux elicited by this force. If the imposed force is small
enough for the system to remain in the linear response regime, the associated
transport coefficient can be obtained directly from the ratio of flux to driving
G 20.0
15.0
0.0
100 200 400 700 800
70.0
60.0
50.0
0.0
Displacement history for a single methane penetrant molecule in polyeth-

ylene. (a) T = 300 K. The squared displacement is computed between successive 16
PS positionalaverages of thepenetrant. @) T = 400 K. Thesquareddisplacementis
computed between successive 7.5 PS positional averages of the penetrant. [Reproduced
from Pant and Boyd with permission.]
force. In the case of diffusion, the most commondriving force used is a spatially
uniform force field F acting directly on each molecule. In an equimolar binary
mixture, for example, equal and opposite forces may be applied on molecules
of the two species, as if the molecules were bearing equal and opposite charges
or “colors” in a uniform electric or “color” field; the system center of mass
remains fixed, and the steady-state fluxes of the two species provide an estimate
120 THEODOROU
of the transport diffusivity D . In liquids (Evans and Morris, and micro-

porous sorbents (Maginn et al., NEMD routes to D have proved more
effective than equilibrium routes to the same property based on accumulating
equilibrium autocorrelation functions of molecular currents or species center-of-
mass displacements[compare Eqs. Owing to their collective nature,
these equilibrium quantities can be obtained with only limited accuracy.
To understand the basis of NEMD methods, it is useful to rewrite Fick's law,
as a relationship between mass flux and chemical potential:
The externally imposed force F in NEMD plays the role of the negative gradient
to
-
in chemical potential. The flux J, is calculated as the steady-state nonequili-
brium ensemble-average molecular flux (jA(t Equation then leads
In Eq. it is assumed that the driving force is applied in the a direction and
that the resulting molecular flux is in the same direction. In the case of aniso-
tropic diffusion, all elements DUBof the diffusion tensor can be computed by
applying the driving force in the direction and measuring the steady-state flux
in the a direction.
In thelimit of infinitely dilute penetrant in thepolymer, simpli-
fies to
Miiller-Plathe et al. employed Eq. to estimate the diffusivity of H,,

He, and 0, in PIB through NEMD simulations. They applied no force on the
polymer molecules but rather chose to force half the penetrant molecules with
F and the other half with -F; the velocities of the latter half were inverted in
computing j,. Results from the NEMD simulation were not very encouraging.
The value of (jA(t * m))neq, as computed by averaging over PS long pieces
of the simulation, was found to fluctuatewidely, especially at large field
strengths. The linear response regime broke down at relatively low values of F,
and the kinetic energy of penetrant molecules was foundto rise artificially as a
result of theimposition of F. It is likely that future applications of NEMD
employing many configurations of the polymer and averaging over long times
will lead to better results. Of particular interest for future NEMD investigations
would be the problem of predicting the transportdifhsivity at high occupancies
via Eq.
OF
D. Transition State Theory-Based Prediction of Low

Concentration Diffusivity
As temperature is reduced toward and below Tg,penetrant diffusion through an

amorphous polymer matrix becomes too slow to be predictable by MD simu-
lations. A simple “back of the envelope” calculation can convince of this.
Consider the diffusion of COz gas in glassy (unplasticized) PVC at low concen-
tration and a temperatureof 25°C. To obtain a reliable diffusivity from MD, one
should let the penetrant molecules move for sufficienttime to sample the frozen-
in inhomogeneitiesof the glassy polymer structure.For this to happen, the trans-
lational displacement of molecules must be at least equal to the mean distance
between neighboring clusters of accessible volume, which we can most conser-
vatively estimate as 5 A (compare Fig. 6). Experimentally, the diffusivity in the
consideredsystem has beenmeasured as D , = 2.4 cm% (Berensand
Huvard, 1987). This means that in order to safely estimate the diffusivity, one
would need an equilibrium MD simulation of duration at least
t=-=
(5 A), = 1.7 S = 170 ns
6 Dr,A (2.4 cm2/s)
which exceeds by an order of magnitude the duration of the longest atomistic
polymer MD simulations that have been conducted. Actually,the MD simulation
should be conducted over a variety of starting positions of the penetrant and
over a variety of configurations of the matrix, as to sample the wide distri-
bution of local environments experienced by the penetrant. This would multiply
the time required to estimate D by a couple of orders of magnitude relative to
(60) unless all these different runs were conducted on a massively parallel
machine. A molecule larger than CO, would require even more time for the
first-principles estimation of its self-diffusivity through equilibrium MD.
As already pointed out inSection IV.B, the reason that diffusion through such
a low-temperature amorphous polymer matrix is so slow is that the penetrant
spends most of its time “trapped” within clusters of accessible volume in the
matrix and only infrequently jumps from cluster to cluster. A brute-force MD
simulation exhaustsitself in tracking the relatively uninteresting “rattling” mo-
tions of the penetrant within a cluster but is much too short to accumulate
sufficient statistics on the jumps that govern diffusion. This was shown char-
acteristically in the work of Takeuchi (1990a), which was the first to explore
the mechanism of the jump process in a glassy polymer. Oxygen-like penetrant
molecules in a polyethylene-like, infinite molecular weight matrix were simu-
lated through atomistic MD 30°C below the simulated glass temperature. Only
a single penetrant molecule was seen toexecute jumps between two sites in the
polymer. The unoccupied volume in the immediate vicinity of this molecule did
122 THEODOROU
not show any significant deviation from the unoccupied volume around mole-
cules thatremained “trappedy7 during the entire duration of thesimulation.
Furthermore, the potential energy felt by the jumping molecule [i.e., its contri-
bution to v A p ( r A , rp)]did not show signs of overcoming a barrier. Motions of
the polymer matrix were very important in effecting a jump. Takeuchi has given
a vivid graphical representation of a jump process in terms of the evolution of
the contours of potential energy experienced by the penetrant. [See Fig. 17 of
Chapter 1, this volume).] The process starts with the generation of a channel
connecting two clusters of accessible volume through thermal fluctuations. The
channel is quickly traversed by the molecule and subsequently disappears. The
jump length is on the order of 5 A, and the overall duration of the process is
around 10 PS. We emphasizeherethat this jump process is notconfined to
diffusion in polymers below T,. As shown by the subsequent simulations of
Muller-Plathe (1992), Muller-Plathe and van Gunsteren (1992), Pant and Boyd
(1993), and Sok et al. (1992), it also occurs in low-temperature melts and rub-
bery polymers. When the temperature is sufficiently low for the distribution of
accessible volume clusters to remain relatively unchanged over the time scale
required for the penetrant to move by the mean distance between clusters, a
jumplike situation should ensue (compare Fig. 6).
The above observations on the mechanism of low-temperature diffusion
of small molecules in glassy polymers suggest a transition state theory de-
scription of diffusion as a sequence of infrequent jump events. Each jump
event involves a relatively small number of degrees of freedom in the con-
figuration space (rp, r A ) of the polymer/penetrant system, as shown sche-
matically in Fig. 14a. To facilitate our discussion, we consider a structureless
spherical penetrant in a model glassy polymer matrix with three-dimensional
periodic boundary conditions, represented by the flexible chain model (GG
and Scheraga, 1976). Let v = NPnp + 3 - 3 = 3 Npnp,the total number
of degrees of freedom of this system. (The 3 is subtracted because the total
potential energy is invariant to rigid translations of the whole system. The
three degrees of freedom of the penetrant are defined relative to a reference
atom in the polymer matrix.)
We will use the term state to refer to the region around a local minimum of
the potential energy T(rp,rA)of the polymer/penetrant system. The projections
of the state minima onto the subspace rpof degrees of freedom of the config-
urationally arrested polymer will be close to, but not identical with, the minima
about which the pure polymer configuration fluctuates in the absence of pene-
trant (compare Fig. 1).In other words, the likely polymer/penetrant configura-
tions will differ somewhat from the pure polymer configurations because poly-
mer degrees of freedom around the penetrant are locally modified to
accommodate the penetrant; the larger the penetrant, the more extensive this
modification (see Fig. 14a).
Macrostate i Macrostate
State
Transition j
(a)
(a) Schematic of a penetrant jump in a low-temperature amorphous model

polymer matrix. two-dimensional representation is given, for clarity. Shaded regions
are occupied by chain segments, and periodic boundary conditions are indicated. The
system polymerlpenetrant, initially in macrostate i, passes into macrostate j by crossing
adividinghypersurfacethatseparatesthetwomacrostatesandcontainsafirst-order
saddle point (transition state) of the potential energy. Both penetrant (translational, orien-
tational) and polymer degrees freedom change along the reaction coordinate. Note the
changes in the clusters unoccupiedvolumebeforeandafterthejump. (b) Simple
electrical analog the Poisson process jumping between macrostates. If the reduced
probability pi of beinginamacrostate is associatedwithanelectricalpotential,the
network of macrostatesbehaves as anelectricalnetworkwhereineachmacrostate is
associatedwithacapacitance p:' andeachconnectionbetweenmacrostateswitha
resistance llk, .
124 THEODOROU
We will use the term macrostate to refer to a collection of neighboring states

separated by energy barriers that are low relative to k,T. The system spends
most of its time confined within macrostates. The infrequent diffusive jumps of
the penetrant constitute transitions from one macrostateto another across a bot-
tleneck in "V(rp,r,) separating the macrostates. The projection of a macrostate
onto the subspace r, of the penetrant degrees of freedom will be a confined
three-dimensional domain. The clusters of accessible volume determined by car-
rying out the geometrical construction of Section II.B on the pure polymer pro-
vide areasonable approximation of these three-dimensional projections of
macrostates. With each macrostate i we associate a position (site) vector ri in
three-dimensional space that is representative of the position of the penetrant in
macrostate i. A convenient choice for ri is the Boltzmann-weighted average of
r, over all configurations belonging to the macrostate.
The evolution of the polymer/penetrant systemin time is viewed as a Poisson
process(Feller, 1957) consisting of successive uncorrelated jumps between
neighboring macrostates. With each jump i -P j is associated a first-order rate
constant ki+ Consider an (in general nonequilibrium) ensemble of penetrant/
polymer systems that at time t = 0 conform to a specified butotherwise arbitrary
distribution among macrostates. As time elapses, this distribution evolves
through transitions between macrostates occurring in the individual systems of
the ensemble. Let p i ( t ) be the probability finding a system the ensemble
in macrostate i at time t. The quantities pi(t) evolve according to the master
equations
At very long times, the ensemble reaches its equilibrium distribution, wherein
the probability of each macrostate is pTq. The equilibrium probabilities of each
macrostate obey the condition of microscopic reversibility (detailed balance)
In view of Eq. (62) and the normalization condition X#:q = 1, it is clear that in
a system with a total of m macrostates only (m + 2)(m - 1)/2 of the quantities
{kidj},@?} are independent. (Inpractice most of the independent rate constants
ki,j are zero, as they correspond to pairs of macrostates that are not connected.)
The average residence timein macrostate i at equilibrium is
MOLECULAR
SIMULATIONS O F SORPTIONAND DIFFUSION 125
Combining Eqs. (61) and (62), we find that the reduced probabilities p$) =
pi(t)/ppqobey the evolution equations
Equation (64) suggests an interesting electrical analog (see Fig. 14b). The net-
work of macrostates can bemapped onto athree-dimensionalnetwork with
nodes at the representative points {ri}.With each node i is associated a capac-
itance p?, and with each pair of connected nodes a resistance l/kjj, with k,
defined in (62). The reduced probability distribution {pi(t)}evolves exactly
as the electrostatic potential in the electrical network.
The transient solution to the master equation, (61), can be used to extract
thepenetrantself-diffusivity.Forexample, the evolutionof (pj(t)} could be
tracked under the initial condition pj(0) = Sij by integrating the master equation
forward in time. If the behavior of the system is diffusive, then the profile ( p j
(t)}, after appropriate smoothing to eliminate the consequences of differences in
p;' among different macrostates, should evolve asexp[-(rj - ri)'/(6DSAt)], from
which D, can be extracted. A kinetic Monte Carlo method for extracting the
self-diffusivity from the mean square displacement of the penetrant under con-
ditions of equilibrium wasused by June et al. (1991) in the context of diffusion
in zeolites. This method directly simulates the continuous time-discrete space
Markov process described by Eq. (61) on an ensemble of model systems. For
the problem discussed here, it would proceed as follows:
Consider a three-dimensional network of a large number m of sites placed

at positions ri, i = 1, . . . ,m,with connectivity defined by the rate constants
{kidj}.If the macrostates have been determined throughanalysis of a model
polymer/penetrant system with periodic boundary conditions (see Section
IV.D.2), then the network can be formed by periodic continuation the
macrostates within the primary box. (As an example, consider the network
formed by periodic continuation of the box in Fig. 14b.)
Distribute a large number NE>> m of random walkers on the sites of the
network according to the equilibrium probability distribution (ppq}. Multiple
occupancy of sites is allowed in this deployment of the random walkers.
The walkers will be allowed to hop between sites without interacting with
each other (i.e.,they will behave as ghost particles toward each other). Each
random walker summarily represents a system in the ensemble of polymer/
penetrant systems whose temporal evolution we want to track with kinetic
MC simulation. Let Ni(t)be the number of random walkers that find them-
selves in site i at time t.
126 THEODOROU
For each site i that is occupied at the current time t, calculate the expected
fluxes Ri,(t) = Ni(t)ki, to all sites with which it is connected. Also,
compute the overall flux R(t) = XiXj Ridj(t)and the probabilities qi-j(t) =
Ri+(t)/R(t).
4. Select a random number 5 [0, 1).Choose the time for occurrence of the
next elementary hop event in the network as At = ln(1 - Q/R(t). Choose
the type of the next elementary hop event by picking one of the possible
transitions i j according to the probabilities qi-j(f).
5. Of the Ni(t) walkers present in site i, pick one with probability l/N,(t) and
move it to site j .
6. Advance the simulation time by At. Update the array, keeping track of the
current positions of all walkers to reflect the implemented hop. Update the
occupancy numbers Ni(t + At) = Ni(t) - 1 and Nj(t + At) = Nj(t) + 1.
7. Return to step 3 to implement the next hop event.
The outcome fromperforming this stochastic simulation overa large number

of steps is a set of trajectories r&) for all NErandom wakers. The self-diffusivity
Ds,A = D is estimatedfromthemean square displacement ([rk(t)- r,(O)]')
computed over all NEwalkers through the Einstein equation, Eq. (49). Note that
simulation schemes advancingthe time in equal intervals are also possible (Ter-
monia and Smith, 1987).
If the rate constants ki,j are small, then the time steps At taken by the sim-
ulation are long, and thus the simulation permits accessing timesand displace-
ments that may be several orders of magnitude larger than the ones accessible
through atomistic MD. Thus, the long-time problem of MD is solved. For im-
plementing this model of diffusion as a sequence of infrequent events, however,
it is necessary that one have a good idea of (1) the macrostates and the repre-
sentative nodal points {ri};(2) the equilibrium probability distribution Mq} of
themacrostates;and (3) theconnectivityand rate constants {ki,} governing
transitions between the macrostates. In the next section we discuss how to obtain
these quantities from the detailed potential energy hypersurface "cr(rp,rA)based
on multidimensional transition-state theory.
2. Multidimensional Transition-State Theory Formulation of a

Diffusive Jump
Consider a model polymer matrix containing one penetrant molecule. As dis-
cussed in Section D.1, the system can be described in terms of v microscopic
degrees of freedom specifying the position vectors (rp,rA)of all polymer atoms
and of the penetrant molecule. For convenience in the subsequent analysis, we
will use the mass-weighted coordinates (Vineyard 1957)
In rpij,
xpij = mpi, X, = min (65)
MOLECULAR
where the subscript i j denotes the jth atom of the ith polymer molecule.We use
the notation x = (xp, x*) for the v-dimensional vector of mass-weighted coor-
dinates needed to describe the microscopic configuration of the system. By def-
inition, a macrostate is a region in x-space surrounded by (v - 1)-dimensional
hypersurfaces of high potential energy relative to k,T. A macrostate i contains
one or more states, each of which is constructed around a local minimum of
"Ir(x). The states within a macrostate are mutually accessible over barriers that
are low relative to k,T. We denote the minima in macrostate i by xh, xf2, . . .
(see Fig. 15). At each such minimum,
and
positive definite
The system can move readily among the states within a given macrostate. Tran-
sitions between different macrostates (e.g., macrostates i and j ) , however, can
occur only infrequently along relatively few high-energy paths, each such path
connecting a state in macrostate i to a nearest neighbor state in macrostate j.
Let and $e be two nearest-neighbor minima in the two macrostates between
which such atransitioncanoccur.By the nearest-neighborproperty of
$k and $e7 there will be at least one first-order saddle point x: between them,
at which the gradient of "Ir vanishes and the Hessian matrix H of second deriv-
atives has one negative and v - 1 positive eigenvalues:
g(x;) = 0 (67a)
H(xi) has one negative eigenvalue with associated
eigenvector nu (6%)
The saddle point or transition state x: is the highest energy point on the lowest
energy passage betweenxfk and xis,. construct this passage, or transition path,
which is a line in v-dimensional space, one can initiate two steepest descent
constructions at x,: one with direction +nu and the other with direction -nu.
Each such construction can be carried out in small steps 6x. For example, start-
ing at x: one can displace the configuration toward x: by a small vector 6x =
nu From point x: + 6x one can trace the steepest descent path leading to
xfk as a series of successive steps ax = - (g/lg))6s. A similar construction can
be carried out toward x;~. The resulting transition path is represented in Fig. 15
as a dot-dashed line.
128 THEODOROU
The dividing surface separating states x i and xis,, and therefore macrostates
i and j , is a (U - 1)-dimensional hypersurface with equation C(x) = 0, that has
the following properties (Sevick et al., 1993):
1. It passes through the saddle point x;:
c (x;) = 0 (684
2. At x;, it is normal to the eigenvector correspondingto the negative eigen-

value of the Hessian:
As a consequence, the dividing surface and the transition path are normal
to each other.
3. At all points other than x;, the dividing surface is tangent to the gradient
vector:
V,C(X) - = x # x; (684
The lowest energy region of the dividing surface in the vicinity of x; con-
tributes mostly to transitions between the two macrostates (see below). We
expect that this region will be well approximated by a hyperplane S, through
x; drawn normal to the direction nu. In the following weuse this approximation;
i.e., we represent the dividing surface by
-
C(x) = nu (x - x;) = 0 (69)
In locating the saddle points x: and the associated reaction paths and dividing
surfaces, the geometric analysis of accessible volume clusters described in Sec-
tion I1.B serves as a useful starting point (Greenfield, 1995). Envision clus-
ters, i and j , that have been identified by the geometrical analysis of the pure
polymer matrix using as a probe the penetrant molecule of interest. By carrying
out the geometrical analysis for progressively smaller probe radii, it is possible
to identify a "neck" connecting clusters i and j (compare Fig. 4). Using this
neck point as an initial guess, one can locate the closest saddle point of "V with
respect to the penetrant degrees of freedom, x, keeping all polymer degrees of
freedom x, fixed and equal to those of the bulk polymer. Well-established al-
gorithms for the numerical determinationof the closest saddle point are available
(Baker, 1986). Using this three-dimensional saddle point as an initial guess, one
can progressively augment the set of degrees of freedom with respect to which
the saddle point is calculated (e.g., including more and more of the x, that lie
in concentric spheres of progressively increasing radius around the penetrant),
130 THEODOROU
until further expansion of the set of degrees of freedom leaves the estimates of
x: and 'T(xi) unchanged. Once x i has been identified in this way, the transition
path and the states x; and x; can be located through a steepest descent construc-
tion in the entire x space, as described above. Note that this steepest descent
construction fully accounts for changes in the polymer matrix induced by the
presence of the penetrant.
Having identified the transition path between macrostates i and j and the
associated dividing surface S, we can proceed to estimate the rate constants
kF7 and k , y through v-dimensional transition-state theory In the follow-
ing, it is assumed that only one transition path contributes significantly to the
flux between macrostates i and j . The analysis can be readily generalized for
multiple diffusion paths.
According to the TST approximation(VoterandDoll, 1985; June et al.,
1991), whenever the polymer/penetrant system finds itself on the dividing hy-
persurface between i and j with net momentum directedfrom i to j , a successful
transition between i and j will occur. Under this assumption (Sevicket al., 1993;
Voter and Doll, 1985) the rate constant can be expressed as
d"p [n(x) P1 8 [C(X)l PXC(X)l P- (x7 P) (70)

np>o
where p is the vector of mass-weighted momenta conjugate to x, n(x) the unit

vector normal to the dividing surface at position x, (x, p) the canonical
ensemble probability density in phase space, and the Dirac delta function selects
configurations on the dividing surface. Upon performing all momentum space
integrations, one obtains from Eq. (70)
Equation (71) expresses k z as aratio of two configurational integrals:one taken

over the dividing surface, the other over the entire macrostate {i}. For systems
with small dimensionality v, the configurational integrals of Eq. (71) can be
computed directly by MC integration (June et al., 1991). For the highly dimen-
sional problem of the diffusive jump in all relevant degrees freedom of the
penetrant and the polymer, the direct sampling of configuration space points
uniformly distributed in S, or in {i} is not straightforward, and a free-energy
perturbation technique along the transition path (Elber, 1990; Czerminsky and
Elber, 1990) is more appropriate. To implement this technique, we consider a
set of closely spaced (v - 1)-dimensional hyperplanes S1, S,, . .., S,

..
. ,S" normal to the transition path starting at xfk and ending at x: (see Fig. 16).
We also consider a thin v-dimensional slice of macrostate {i} around the hy-
perplane S1, of thickness Ax&; we use the symbolism A{i}, to denote this thin
domain. Equation (71) can be written as
(2PT)'R kZ =
J1,
d"'x exp [-PV(x)l
l,d"'x exp d"'x exp[-PV(x)]
The ratio of v-dimensional configurational integrals in Eq. (72) equals the prob-
abilitythat a system that samples macrostate i according to theequilibrium
probability density of the canonical ensemble visits the narrow strip A{i}k. It
can be determined through anNVT MD or a Metropolis NVT MC simulationof
the polymer/penetrant system confinedwithin macrostate i. The fraction of con-
figurations sampled by such a simulation that lie in A{i}k provides an estimate
of this ratio. compute each one of the ratios of (v - 1)-dimensional integrals
over successive hyperplanes S,-l and S, appearing in Eq. (72), we consider the
vector Sx, along the transition path between these hyperplanes. If S,.-l and S,
are spaced closely enough, they are practically parallel to each other and normal
to Sx,. Foreachconfiguration x lying onthere is a configuration x + Sx,
lying on S, and vice versa. The ratio of configurational integrals can then be
132 THEODOROU
C(x) =
16 Sequence of (v - 1)-dimensionalhyperplanes S,, S,, . .. and v-dimen-
sional domain A{i}k used for the free-energy perturbation calculation of the jump rate
constant k?; based on multidimensional transition state theory.
written as
d""Ix exp [- exp { + -

L ,
exp
= +=P + -
The ratio of (U - 1)-dimensional integrals over hyperplanes S,,,-, and S,,,
be computed as an ensemble average over a MD Metropolis MC simulation
OF
of the system confined to hyperplane S,,,-.l.The calculation of kiAj through

(72) is thus possible through a v-dimensional and aseries of (v - 1)-dimensional
MD or Metropolis MC simulations.
Transition-state theory is not exact. In reality, not all dynamical trajectories
penetrating Su in the direction from i to j effect a successful transition from i
to j . Recrossing events, wherein the system passes fromi to j but subsequently
crosses back and ultimately thermalizes in i, are well possible. The actual tran-
sition rate constant ki-j can thus be expressed as
ki+ = k z 6-j (74)
where k z is the transition-state theory estimate, obtained as described above,
while Jdj is a dynamical correction factor accounting for recrossing events.
$-j can be estimated by undertaking short MD runs of the system initiated on
the bottleneck surface Su. The fraction of such trajectories that ultimately ther-
malize in the macrostate toward which they were moving when initiated pro-
vides the dynamical correction factor (Voter and Doll, June et al.,
NotethattheMD simulations needed to estimate are much less compute
intensive than “brute force” MD; once placed on the bottleneck hypersurface
Su, the system will quickly move toward a macrostate and thermalize in it.
The equilibrium probability of each macrostate i is obtained as
p;q =
I], d’x exP t-B~(x)l
-
li} d’x exP I-PWol
zli.1
d’x exp [-pT(x)]
-1 d’x exp [-p”lr(x)]
(75)
It can be accumulated in thecourse of a Widom testparticle insertion calculation

[compare Section IKA, Eq. (27)] in which contributions to the solubility from
configurations in which the penetrant resides in different clusters of accessible
volume (corresponding to different macrostates) are accumulated separately and
finally divided by the total solubility. Alternatively, aMetropolis MC simulation
of the penetranVpolymer system can be employed that provides for exchange of
the penetrant among different clusters; the fraction of configurations sampled
by such a Metropolis calculation that belong to macrostate i is p:q.
pioneering TST-based study of the diffusion of argon in amorphous polyeth-

ylene was presented by Jagodic et al. Computational limitations pre-
vented the generation of realistic model polymer structures, the results from
this study were not very conclusive.Recently,a set of thoroughTST-based
calculations of high predictive value was carried out by Gusev et al.
134 THEODOROU
These investigators calculated the infinite-dilution diffusivity of He, HZ,

and NZ in rubbery polyisobutylene and glassy polycarbonate at 300 K, repre-
senting the polymer matrices by rigid minimum energy configurations obtained
through molecular mechanics. With thespherical representation used for all pen-
etrants, the X space is three-dimensional, spanned by the three translational de-
grees of freedom r, of the penetrant, all of which are associated with the same
mass. This greatly simplifies the TST calculation of the rate constants k z . No
distinction was made between states and macrostates; i.e., all local minima of
V(rP) were identified exhaustively andused to definemacrostates,between
which the rate constants were computed. The number of local minima identified
was quite large (approximately 50 for the largest penetrants in a model structure
-
of 30 A edge length). The tessellation of space into states and the identi-
fication of dividing surfaces were carried out by constructing regular grids of
small cubic elementsor voxels through the polymer configurations. Each divid-
ing surface was thus represented as a collection of contiguous square facets
oriented normal to one of the three coordinate axes (see also June et al., 1991).
The three-dimensional and two-dimensional configurational integrals needed for
the evaluation of p:q and k z via Eq. (71) were obtained through direct numer-
ical integration over the values of T(r$ at the node points of the grid, taking
into account the increase in area of the dividing surfaces due to the finite size
of the voxels. The results of the TST analysis were used to accumulate residence
time distributions of the penetrants in the states and to estimate D,, D through
the generation of kinetic Monte Carlo trajectories for the jump process. The
computational requirementsof the approach were very modest relative to “brute
force’’ MD, permitting the generation of millisecond-long stochastic trajectories.
For He, the diffusivities predicted were within a factor of of the experimental
values for both polymers. For the larger penetrants, however, the diffusivities
were underestimated by three to five orders of magnitude as a result of the rigid
representation of the matrix.
improved approximate TST formulation that takes into account thermal
motion in the polymer matrix was presented by Gusev and Suter (1993). To
circumvent the complexity of the full-blown TST formulation outlined in Sec-
tion IV.D.2,Gusev and Suter assumed a quadratic dependence of the polymer-
polymer contribution to the potential energy Vp(rp)of the form
where rPkis the position vector of the kth atom of the polymer and rpk,o the
equilibrium position of the same atom in the static minimum energy configu-
ration. Note that in this Debye-type approximation, atoms are assumed to move
isotropically and independently of each other. The mean square deviation (Ai)
shoulddependontemperatureandatomtype. (The latter dependence is ne-

glected in this implementation; i.e., a mean square positional deviation over all
atom types is used.) The implicit assumption is introduced that “elastic7’ fluc-
tuations the polymer matrix are fast in comparison to the time between dif-
fusive jumps, whereas more drastic motional processesassociated with structural
relaxation in the polymer occur over times much longer than the time between
jumps. On the assumption of this double time scale separation, the transitions
between states aregoverned by athree-dimensional potential mean force
A(rA), obtained by integrating out all polymer degrees of freedom in the con-
figuration-spaceprobability density pNw (rp, rJ. By virtue of Eq. (76), this
potential of mean force is pairwise additive, consisting penetrant-polymer
atom contributions the form - rpks0() that can be computed directly
from the penetrant-polymer atom Lennard-Jones potentials once (Ai) is known.
The TST calculation then proceeds as in the static polymer case (Gusev et al.,
1993), usingA(rA) in place of Critical to this fluctuating matrix approach
are the values (Ai). These are estimated from short (5 PS duration) atomistic
MD simulations of the pure polymer matrix. The (A:) values obtained from
these simulations are found to increase with time, following the approximate
scaling log ((Ai)”) log In view this, it is decided to use in the TST
calculations the value of (Ai) that corresponds to the most probable residence
time T of the penetrant in a state. This introduces a self-consistent character in
the calculation: A value for (Ai) is initially postulated and used to compute the
rate constants k z ; the residence time distribution is extractedfrom the rate
constants, and the time at which it goes through a maximum is identified as
anew value of (Ai) is determinedfrom this andthe whole procedure is
repeated. The self-consistent calculation is found to convergein practically one
iteration. In the case of He and H,, was approximately 0.15 PS and the cor-
responding (Ai)’” was taken as 0.22 A. In the case of 0, and NZ, was ap-
proximately 1 PS and (A:)’” was 0.46 The diffusivities extracted from this
fluctuating matrix approach are within a factor 2 from experimental values
for all penetrant molecules examined, indicating that the “trapping” experienced
by the larger penetrants in static model matrices was eliminated.
interesting observation from the Gusev and Suter simulations is that the
motion of the penetrant molecule is not diffusive up to times on the order
tenths nanoseconds to tenths of microseconds. “anomalous diffusion”
regime, wherein the mean square displacement of the penetrant scales roughly
as (r’) m to.’,is observed at short times, and it is only in the longest time portion
of the stochastic trajectories that the diffusive scaling (r’) t sets in. A strikingly
similar conclusion was reached through atomistic MD the same model sys-
tems (He and in polyisobutylene) at room temperature carried out to very
long times (Muller-Plathe et al., 1992b) and also in the MD simulations of
Pant and Boyd (1993). The crossoverbetween anomalous and diffusive regimes
136 THEODOROU
-1 -
-2 I I I I I I l l
-14 -12 -11 -10 -9 -8 -7 -6 -5 -4
Meansquaredisplacementversustimecurves for heliumandoxygenin

polyisobutylene at 300K, as computed from the transition-state theory approach of Gusev
and Suter (1993) using a Gaussian model for the polymer fluctuations. The data, dis-
played in logarithmic coordinates, are averages over 1500 independent kinetic Monte
Carlo simulation paths. “Anomalous diffusion” behavior is seen out to time scales of
tenths of nanoseconds to tenths of microseconds. [Reproduced from Gusev and Suter
(1993), with permission.]
is seen characteristically in Fig. 17. Clearly, the time span of the anomalous
diffusionregime in low-temperature polymers is many orders of magnitude
longerthanthe “ballistic” regime seen in small molecularweight liquid
simulations.
A qualitative interpretation of the anomalous diffusion regime based on the
jump picture of diffusion can be provided bythe following argument: As pointed
out in Section II.B., the connectivity of the network of accessible volume clus-
ters or macrostates seenby the penetrantmolecule is a function of the time scale
of observation, owing to the wide distribution of kidj values. At very short times,
the network consists of isolated clusters in which the penetrant rattles. At some-
what longer times, the penetrant can execute short length scale (5-10 A) jump
motions withinisolated groups of a few clusters that are connected via diffusion
pathways of high ki+ It is an interesting question whether these clusters be-
tween which transitions are most facile tend to align along the backbones of
chains, as suggested by the Pace and Datyner theory (Dr. J. Petropoulos, per-
sonal communication). As the time scale of observation increases, the clusters
that are mutually accessible become more and more connected until, at acritical
time scale, percolation of the network of clusters occurs. A random walk on
isolated groups sites or on a percolating collection sites is known to be
subdiffusive (Stauffer, 1985). It is only for times that are long compared to the
time scale for percolation that the network is well connected and that diffusive
behavior is observed. By this argument, the time span of the anomalous regime
should belonger the slower the motionof the penetrant (i.e., the lower the values
of ki+j); this is confirmed by the simulaton findings (see Fig. 17).
V. CONCLUSIONS AND FUTUREDIRECTIONS

In recent years there has beenintense computer modeling andsimulation activity
in the areas of sorption and diffusion of small molecular weight gases in amor-
phous polymers. Calculations carried out include the geometrical characteriza-
tion of accessible volume in a pure polymer matrix and its distribution and
rearrangement with thermal motion; the calculation of Henry’s constants char-
acterizing the sorption thermodynamics in melts at low pressures through the
Widom test particle insertion method; an approximate theoretical treatment of
sorption in glassy polymers based on static, energy-minimized model-polymer
structures; numerous equilibriummolecular dynamics and a nonequilibrium MD
investigation of the self-diffusion of gases in melts and rubbery polymers; and
the estimation of the low-concentration diffusivity of gases in glassy and low-
temperature rubbery polymers through approachesbased on transition-state the-
ory. Most of this work has focused on chemically simple penetrants (He, H2,
02,N,, CH,) and polymers (polyethylene, polypropylene, polyisobutylene), al-
though some more complicated matrices(polycarbonate, polydimethylsiloxane)
have also been examined.
The atomistic MD simulation work has revealed a wealth of mechanistic
information on how diffusion takes place at various temperatures; this infor-
mation has been useful in testing theoretical ideas and phenomenological cor-
relations proposed in the past. In systems wherediffusion is not too slow (D 1
cm%), MD simulation provided good quantitative estimates of the self-
diffusivity whenever sufficientcare was taken in the model representation of the
polymer and its interactions with the penetrant. The estimation of Henry’s con-
stants has been generally less successful, with some notable exceptions(alkanes
in polyethylene). Approximate TST-based formulations based on the picture of
diffusion as a succession of jumps in the matrix have proved very promising in
dealingwith diffusion at low temperatures, where atomistic MD becomes
impractical.
138 THEODOROU
There is stilt much to be done before a general framework of theoretical

and simulation methodologies becomes available for use in the efficient first-
principles estimation of sorption and diffusion phenomena and therefore in the
“molecular engineering design”of materials with tailored separation and barrier
properties. The very generation of well-equilibrated model polymer melts or of
glassy configurations that correspond to realistic formation histories is still a
challenge, owing to the very long relaxation times present in these systems.
New Monte Carlo algorithms for the bold exploration of configuration space,
fast multipole algorithms for the rapid summation of interactions, and the use
of parallel machines are promising for alleviating this problem. The need to
validate simulation approaches interms of their ability to predict scattering,
spectroscopic, equation-of-state, sorption, and diffusion measurementsis becom-
ing increasinglyobvious. ability to predictallthese properties in known
systems using a consistent potential representation is a prerequisite for using
molecular modeling as a design tool. The source of the difficulty in predicting
low-pressure sorption thermodynamics accurately has to be clarified. Further-
more, the problem of sorption equilibria at high penetrant activities, including
dual-mode sorption behavior,swelling,andplasticization effects in polymer
glasses, has hardly been addressed from a fist-principles prediction point of
view.
Real-life applications often have to do with large, complex, or strongly in-
teracting solvent or plasticizer molecules whose thermodynamic and transport
behavior hasnot been investigated sufficientlywith molecular modeling. In view
of the very long relaxation times characterizing such systems, it is unlikely that
“brute force” MD will be very useful, and coarse-grained approaches become
imperative. The low-concentration diffusion of large rigid penetrants (e.g., ben-
zene)in an amorphous matrix couldbetreatedwithmultidimensional TST;
explicit incorporation of the polymer degrees of freedom in calculating diffusion
pathways would be imperative in this case (see Section IVD), as the time scale
separation postulated by simpler treatments does not hold. For large, flexible
penetrants with many torsional degrees of freedom, a picture of diffusion as
Brownian motionin a mediumcapable of exerting Langevin andfrictional forces
on the penetrant would seem more appropriate. A quantitative connection be-
tween the “friction factor” invoked in such a coarse-grained description and
the detailed potential energy hypersurface of the system must be established.
All atomistic modeling work conductedto date on diffusion in polymers has
to do with systems that, macroscopically, exhibit Fickian behavior. In the long
run, the quantitative prediction of anomalous and case I1 diffusion from molec-
ular level information should be addressed. This can be accomplished only if
methods for predicting the time spectrum for structural relaxation in glassy poly-
mer matrices, pure or swollen withsolvent, become available. Dealing with the
long-time relaxationbehavior of polymers in a way thatcombines computational
MOLECULAR
tractability with quantitative accuracy is today a grand challenge for polymer

molecular modeling.
The wide spectra of length and time scales present in polymer systems call
for hierarchical modeling approaches consisting of several interfacing “mod-
ules,’, each module being a theoretical formulation or simulation algorithm that
can capture phenomena over a limited range of scales. Modules should receive
input from lower level (smaller scale) modules and provide input to higher level
ones. The transition-state theory-based approach to low-temperature diffusion,
combining geometrical and energetic analysis of configurations of the polymer/
penetrant systems to identify macrostates, calculation of equilibrium occupancy
probabilities for and transition rate constants between the macrostates through
MC or MD techniques, and extraction of the diffusivity from long stochastic
trajectories, provides agood example of suchahierarchy. The conventional
atomistic MD and MC simulation techniques will always be useful for estab-
lishing the ultimate connection to chemical constitution at the base of this hi-
erarchy but are unlikely to solve the property prediction problem alone. The
need for coupling these techniques to more coarse-grained theoretical and sim-
ulation approaches cannot be overemphasized.
I wish to thank my collaborators, Dr. Krishna Pant, Mike Greenfield, and Travis
Boone, for stimulating discussions and for providing some of the research results
discussed in this chapter. am grateful to Dr. John H. Petropoulos for cultivating
my interest in diffusion in polymers and sharing his deep insights on the subject
with me. Randy Snurr is thanked for his thorough reading of the manuscript
and his suggestions. appreciate the permission given to me by Professors Ulrich
W. Suter and Juan J. de Pablo and by Drs. Hisao Takeuchi and Krishna Pant to
include copies of their figures in the chapter. also appreciate the patience of
our Editor, Professor Partho Neogi, with my repeated delays in finishing the
manuscript.
Allen, M. P., and D. J. Tildesley Computer Simulation of Liquids, Clarendon

Press, Oxford.
Arizzi, S., P. H. Mott, and U.W. Suter J. Polym. Sci., Polym. Phys. Ed., 30,415.
Baker, J. J. Compuf. Chem., 7,
Barrat, J. L., J.-N., Roux, and J.-P. Hansen Chem. Phys., 149,
Berens, R., and G.S. Huvard Interaction of polymers with near-critical carbon
dioxide, Presented at the Annual Meeting of the AIChE, New York, Nov.
Bird,R.B., W. E. Stewart, and E. N. Lightfoot Transport Phenomena, Wiley,
New York.
140 THEODOROU
Boone, T. D. Ph.D. Thesis, Univ. California, Berkeley.

Boyd, R. H. Macromolecules, 22,
Boyd, R. H., and K V. Pant In Computer Simulation of Polymers, R. J. Roe,
Ed., Prentice-Hall, Englewood Cliffs, NJ, p.
Boyd, R. H., and P. V. K Pant Macromolecules,
Czerminsky, R., and Elber, R. J. Chem. Phys., 92,5580.
Dee, G. T., T. Ougizawa, and D. J. Walsh Polymer, 33,
de Groot, S. R.,andP.Mazur Non-Equilibrium Thermodynamics, Dover, New
York.
Deng, Q., and Y. Jean Macromolecules, 26,
de Pablo, J. J.,M. Laso, andU.W. Suter J. Chem. Phys., 96,
de Pablo, J. J., M. Laso, U.W. Suter, and H. D. Cochran Fluid Phase Equilibria,
83,
Dodd, L. R. and D. N. Theodorou Mol. Phys., 72,
Dodd, L. R., and D. N. Theodorou Adv. Polym. Sci., 116,
Dodd, L. R., T. D. Boone, and D. N. Theodorou Mol. Phys., 78,
Ediger, M.D. and D. B. Adolf Adv. Polym. Sci., 116,
Eisenberg, A. In Physical Properties of Polymers, J. E.Mark,A.Eisenberg,
W.W. Graessley, L. Mandelkern, and J. L. Koenig, Eds., American Chemical
ciety, Washington, DC, Chap. p.
Elber, R. J. Chem. Phys., 93,
Evans, D. J., and G. P. Morriss Statistical Mechanics ofNonequilibrium Liquids,
Academic, London.
Feller, W. An Introduction to Probability Theory and Its Applications, Vol. II,
Wiley, New York.
Flory, P. J. Statistical Mechanics of Chain Molecules, Wiley, New York.
Fytas, G., and K. L. Ngai Macromolecules, 21,
Go, N.,and H. Scheraga (1976). Macromolecules, 9,534.
Greenfield, M. L. Ph.D. Thesis, in progress, Univ. California, Berkeley.
Greenfield, M. L., and D. N. Theodorou Macromolecules, 26,
Gusev, A. A., and U. W. Suter Phys. Rev. A , 43,
Gusev, A. A., and U. W. Suter J. Chem. Phys., 99,
Gusev, A. A., S. Arizzi, and U. W. Suter J. Chem. Phys., 99,
Habenschuss, A., and A. H. Narten J. Chem. Phys., 92,
Hansen, J. P.,and I. R. McDonald Theory of Simple Liquids, 2nd ed., Academic,
New York, p.
Jagodic, F., B. BorStnik, and A. Aiman Macrornol. Chem., 173,
June, R. J., A. T. Bell, and D. N. Theodorou J. Phys. Chem., 95,
Kluin, J.-E., Yu, S. Vleeshouwers, J. D. McGervey, A. M. Jamieson, R. Simha, and
K. Sommer Macromolecules, 26,
Laso, M., J. J. de Pablo, and U. W. Suter J. Chem. Phys., 97,
Lijwen, H., J.-P. Hansen, and J.-N. Roux Phys. Rev. A, 44,
Ludovice, P.J.,andW. W. Suter In Encyclopedia of Polymer Science and En-
gineering, 2nd ed., J. I. Kroschwitz, Ed., Wiley, New York, Suppl. Vol. p.
McKechnie, J. I., D. Brown, and J. H. R. Clarke Macromolecules, 25,
MOLECULAR
McQuarrie, D.A. (1976). Statistical Mechanics, Harper and Row, New York, p. 55.
Maginn, E. J., A. T. Bell, and D. N. Theodorou (1993). J. Phys. Chem., 97, 4173.
Malhotra, B. D., and R. A. Pethrick (1983). Macromolecules, 16, 1175.
Mansfield, K. F., and D. N. Theodorou (1990). Macromolecules, 23, 4430.
Mansfield, K F., and D. N. Theodorou (1991). Macromolecules, 6283.
Muller-Plathe, F. (1991a). J. Chem. Phys., 94, 3192.
Muller-Plathe, F. (1991b). Macromolecules, 6475.
Muller-Plathe, F. (1992). J. Chem. Phys., 96, 3200.
Muller-Plathe, F., and W. F. van Gunsteren (1992). Polym. Prepr. (Am. Chem. Soc., Div.
Polym. Chem.), 33(1), 633.
Muller-Plathe, F., S. C. Rogers, and W. F. van Gunsteren (1992a). Macromolecules, 25,
6722.
Muller-Plathe, F., S. C. Rogers, and W. F. van Gunsteren (1992b). Chem. Phys. Lett.,
199, 237.
Muller-Plathe, F., S. R.Rogers,andW. F. van Gunsteren (1993). J. Chem. Phys.,
9895.
Pant, P. V. K., and R. H. Boyd (1992). Macromolecules, 25, 494.
Pant, P. V. K., and R. H. Boyd (1993). Macromolecules, 26, 679.
Petropoulos, J. H.(1994).In Polymeric Gas SeparationMembranes, D. R.Pauland
Yu. P. Yampolski (eds.), CRC Press, Boca Raton.
Plazek, D. J., and K. L. Ngai (1991). Macromolecules, 1222.
Prausnitz, J. M., R. Lichtenthaler, and E. G. Azevedo (1986).
Molecular Thermodynamics
Fluid Phase Equilibria, Prentice-Hall, Englewood Cliffs, NJ.
Rigby, D., and R.-J. Roe (1987). J. Chem. Phys., 87, 7285.
Rigby, D.,and R.-J. Roe (1988). J. Chem. Phys. 89, 5280.
Rigby, D.,and R.-J. Roe (1989). Macromolecules, 22, 2259.
Rigby, D.,and R. J. Roe (1991). In Computer Simulation of Polymers, R.-J. Roe, Ed.,
Prentice-Hall, Englewood Cliffs, NJ, p. 79.
Schaefer, D., H. W. Spiess, U. W. Suter, and W.W. Fleming (1990). Macromolecules,
23, 3431.
Sevick, E. M., T. Bell, and D. N. Theodorou (1993). J. Chem. Phys. 98, 3196.
Shah, V. M., S. A. Stem, and P. J. Ludovice (1989). Macromolecules, 22, 4660.
Smith, G. D., and R. H. Boyd (1992). Macromolecules, 25, 1326.
Smith, G. D., and Y. D. Yoon (1994). J. Chem. Phys., 100, 649.
Sok, R. M., H. J. C. Berendsen, and W. F. Gunsteren (1992). J. Chem. Phys. 97, 4699.
Sonnenburg, J., J. Gao, and J. H. Weiner (1990). Macromolecules, 23, 4653.
Stauffer D.(1985). Introduction to Percolation Theory, Taylor and Francis, London.
Takeuchi, H. (1990a). J. Chem. Phys., 93, 2062.
Takeuchi, H. (1990b). J. Chem. Phys., 93, 4490.
Takeuchi, H., and K Okazaki (1990). J. Chem. Phys. 92, 5643.
Takeuchi, H., and K. Okazaki (1993a). Makromol. Chem. Mucromol. Symp., 65, 81.
Takeuchi,H.,and K. Okazaki(1993b). On glasstransitionobtainedfrommolecular
dynamicssimulation ofpolymers:lowenergyexcitationinglassypolymers,
submitted.
Takeuchi, H., and R.-J. Roe (1991a). J. Chem. Phys., 94, 7445.
142 THEODOROU
Takeuchi, H., and R.-J. Roe (1991b). J. Chem. Phys., 94, 7458.
Takeuchi, H., R.-J. Roe, and J. E. Mark (1990). J. Chem. Phys., 9042.
Tanemura, M., T. Ogawa, and N.Ogita (1983). J. Comput. Phys., 191.
Termonia, Y., and P. Smith (1987). Macromolecules, 20, 835.
Theodorou, D. N., and U. W. Suter (1985). Macromolecules, 18, 1467.
Theodorou, D.N.,and U. W. Suter (1986). Macromolecules, 139.
Trohalaki, S., D. Rigby, A. Kloczowski, J. E. Mark, and R. J. Roe (1989). Polym. Prepr.
(Am. Chem. Div. Polym. Chem.), 30(2), 23.
Trohalaki, S., A. KJonowski, and J. E. Mark (1991). In Computer Simulation of Poly-
mers, R.-J. Roe, Ed., Prentice-Hall, Englewood Cliffs, NJ, Chap. 17, p. 220.
Trohalaki, S., D. Rigby, A. Kloczowski, J. E. Mark, and R. J. Roe (1992). Polym. Prep.
(Am. Chem. Div. Polym. Chem.), 33(1), 629.
Vacatello, M.(1992). Polym. Prepr. (Am. Chem. Soc., Div. Polym. Chem.), 33(1), 529.
Vacatello, M., G. Avitabile, P. Corradini, and A. J. Tuzi (1980). J. Chem. Phys., 73, 543
(1980).
Vineyard, G. H. (1957). J. Phys. Chem. Solids, 3, 121.
Voter, A., and J. Doll J. Chem Phys., 82,
Weber, T. A., and E. Helfand (1979). J. Chem. Phys., 71, 4760.
3
Free-Volume Theory
J. Duda
The Pennsylvania State University, University Park, Pennsylvania
John M. Zielinski
Air Products & Chemicals, Inc., Allentown, Pennsylvania
INTRODUCTION
The phenomenon of small-molecule mobility in macromolecular materials dic-
tates the effectiveness of polymerization reactors as well as the physical and
chemical characteristics of the polymer produced. The molecular weight distri-
butionandaveragemolecular weights, for example, are among the physical
properties influenced by the diffusion-controlled termination step of free-radical
polymerization reactions. Other polymer processing operations affected by mo-
lecular transport include devolatilization, mixing of plasticizers (or other addi-
tives),and formation offilms, coatings, andfoams.Inaddition, distinctive
molecular diffusion behavior is essential for miscellaneous polymer products
such as bamer materials, controlled drug delivery systems, and membranes for
separation processes. The fundamental physical property required to design and
optimize the processing operations is the mutual diffusion coefficient, D.
Numerous techniques have been developed to correlate and predict mutual
diffusioncoefficients for systems composed of two or more low molecular
weight materials (Reid et al., 1977). These techniquesare not, however, suitable
for systems in which one of the species possesses chainlike characteristics such
as a synthetic or natural polymer. The failure to extend theoretical developments
that accurately describe mass transfer in systems of low molecular weight spe-
cies to macromolecular systems stems from characteristics unique to polymeric
materials.
143
144 DUDA AND ZIELINSKI
In mixtures of low molecular weight molecules, for example, molecular mi-

gration is typically a weak function of temperature and concentration, whereas
in polymericsystems diffusional transport canbe significantlyenhanced(or
diminished) by varying either condition. The most pronounced effects are ob-
served near the glass transition temperature (Ta, where a 1% increase of solvent
weight fraction in a polymer solution can increase D by three orders of mag-
nitude.Diffusion
coefficients ranging between
and cm’ls havebeen
reported, and apparent activation energies (ED)greater than 50 kcal/mol have
been determined (Vrentasand Duda, 1986).Consequently,significantexperi-
mentation is required to quantify and optimize a processing operation governed
by molecular transport.
In addition to temperature and composition, diffusion in polymers is con-
trolled by morphological features such as crystallinity and cross-linking, both
of which tend to reduce molecular mobility. Although these complexities inhibit
theoretical analysis of mass transfer, experimental measurements of D are, in
general, not any more laborious than those of amorphous materials. Under cer-
tain conditions, however, transport in even thermoplastic amorphous polymers
does not follow the laws of classical molecular diffusion. This class of transport
is often designated as anomalous (or non-Fickian) and is not discussed further
in the context of this chapter. This topic, however, is addressed thoroughly in
Chapter 5.
Depending on experimental conditions, polymeric materials may exhibit a
variety of mechanical properties ranging anywhere from brittle (glassy) to de-
formable (rubbery). Thus, they may be viewed aseither highly viscous, molten
liquids or relatively low viscosity solutions. Several theoretical formalisms
have been developed that attempt to describe and predict mutual diffusion in
both the rubbery and glassy states. example of a successful transport theory
in this vein is given by Vrentas and Duda (1977a,b). The model accurately
describes diffusion both above and below Tg, through the presumption that
transport is controlled by the availability of free volume within a system. In
this chapter, the fundamental concepts of free-volume theory are introduced in
the context of the Vrentas-Duda formalisms. These models are subsequently
employed to correlate (and predict from material characteristics) mutual binary
diffusion coefficients for solutions of amorphous polymersand small-molecule
solvents.
In Sections II and 111, diffusion within concentratedpolymer solutions above
TBof the mixture is emphasized, whereas SectionIV focuses on transport in the
glassy state. Section V addresses the ways in which diffusional behavior is
affected by additional considerations such as chemical cross-linking, block co-
polymers, antiplasticization, and multicomponents. Throughout the chapter, we
implicitly assume that transport processes occurat conditions that are described
by classical (Fickian) diffusion.
FREE-VOLUME THEORY 145
FREE-VOLUMECONCEPTS
The presumption that molecular transport is regulated by free volume was first
introduced by Cohen andTurnbull (1959). Although their conceptualization was
initially believed to be suitable only for liquids that could be envisioned as an
ensemble of uniform hard spheres, the greatest impact of their development,
ironically, has been on describing mass transfer in solutions consisting of long
polymer chains mixed with small solvent molecules.
FromtheCohenand Turnbull perspective,thehard-sphere molecules that
constitute an idealizedliquid exist in cavities (or cages) formedby nearest neigh-
bors. Thus, the total volume the liquid could bedivided into two components:
occupied volume and free volume. Despite possessing the innate ability to vi-
brate within its cage, each sphere was presumed incapable of migration until
natural thermal fluctuations caused a hole (or vacancy) to form adjacent to its
cage. The hole had to be sufficiently large to permit a significant displacement
of a spherical molecule. A single step the diffusional transport mechanism
was successfully completed when the cavity a molecule left behindwas occupied
by a neighboring molecule. Translational motion, according toCohen and Turn-
bull, did not require a molecule toattain a prerequisite energy level to overcome
an activation energy barrier. Rather than creating holes by physically displacing
nearest neighbors, as suggested by the activation energy approach (DiBenedetto,
1963; Brandt, 1955; Pace and Datyner, 1979), molecular transport was presumed
to rely on the continuous redistribution of free-volumeelementswithin the
liquid.
If the volume of each hard sphere is denoted V*, then the occupied volume
in a liquid containing N spheres is Additionally, if the average free volume
per sphere is given by V', then the total volume of the liquid, V,, can be ex-
pressed as V, = + Nv'. The migration (or self-diffusion) rate of a hard
sphere is therefore proportional to the probability of finding a hole of volume
V* or larger adjacent to the sphere.
When free-volume voids form adjacent to spherical molecules by means of
natural thermal fluctuations within a liquid, the molecules undergo a single step
in the diffusion process. By describing the dispersion of free-volume elements
within a liquid in mathematical terms, Cohen and Turnbull developed a distri-
bution function that provides the probability of finding a free-volume hole of a
specific size. The diffusion coefficient, considered proportional to the probability
of finding a hole of volume V* or larger, may be written as
D = exp(-yV*/v') (1)
In this expression, D is the self-diffusion coefficient the molecules, V* is the

minimum volume hole size into which a molecule can jump, and y is a nu-
merical factor between 0.5 and 1.0 that is introduced to account for the overlap
Table 1 Free-Volume
Theory
Parameters
for ToluenePoly(viny1 acetate) System
Parameter
D,, lo4
4.82'
(cm2/s)
X 0.393
K,,/Y lo3 K) 1.45
K& lo4 (cm3/g K) 4.33
KZ - T ~ (K)I -86.32
- T82 (K) -258.2
y; (cm3/g)
V ; (cm3/g) 0.728
5 0.82
E (cal/mol) 0.0
'Do = 4.82 X cm'/s is estimatedusingphysical
properties pure toluene(Zielinski and Duda,
1992a). Analysis of toluenePVAc mutual diffusion
data at 40°C yields a Do value 10.36 lo" cm*/
S. The effect of Do on diffusion coefficient predic-
tions is illustrated in Fig. 1.
between free-volume elements (i.e., free volume shared by neighboring mole-

cules). The proportionality constant A was considered by Cohen and Turnbull
to be related to the gas kinetic velocity.
This formalism indicates that the self-diffusion coefficient is an exponential
function of the ratio of the size of the diffusion molecule tothe free volume per
molecule in the liquid. While relatively simplistic, this framework constitutes
the forerunner of the free-volume models presented in the following sections
for polymer/solvent solutions. Numerous investigators have extended the free-
volume concepts outlined here, and several of these developments have been
incorporated into the theory of Vrentas,Duda,and coworkers (Vrentasand
Duda, 1977a,b; Vrentas et al., 1993; Duda et al.,1982; Zielinski andDuda,
1992a,b). The Vrentas-Duda approach, generally accepted as one of the most
successful theories for describing molecular diffusion in concentrated polymer
solutions, is the main focus of this chapter.
DIFFUSION ABOVETHE GLASS TRANSITION

TEMPERATURE
The Cohen and Turnbull development continues to serve as a valuable foun-
dation for modem theoretical frameworks that successfully describe difisional
behavior in concentrated polymer solutions. The formalism, as originally de-
FmE-VOLUME THEORY 147
rived, provides a relationship betweenthesystem free volume and the self-

di€€usion coefficient D, for a one-component liquid. This relation can be readily
extended to describe self-diffusion of a single species in a binary mixture:
-
D l= D,, exp(-rVl/Vm)
-. (2)
v:
where is the critical molar free volume required for a jumping unit of species
of 1 to migrate v, is the free volume per mole of all individual jumping units
in the solution, and Dol is a temperature-independent constant.
In the original Cohen and Turnbull representation, a jumping unit was des-
ignated as a single hard-sphere molecule undergoing diffusion. In solutions of
real molecules, particularly in mixtures of macromolecules, however, an indi-
vidual molecule can be composed of multiple jumping units covalently bonded
together. Free-volume holes that readily accommodateentire polymer molecules
simply do not form. Instead, polymer chain migration is envisioned to result
from numerous jumps of small segments alongthe polymer chain. To complicate
matters further, low molecular weight molecules of sufficient size and flexibility
are also capable of migrating by a mode, reminiscent of polymers, that involves
coordinated motion between several parts of the molecule (Amould and Laur-
ence, 1992; Vrentas et al., 1985a). To generalize the Cohen and Turnbull theory
to describe motion in binary liquids, Vrentas and Duda used the relationship
- %H
-
v, =
(moles of jumping units)/g
-
ol/Mlj + wz/Mzi
where , p is the specific hole free volume of a liquid with a weight fraction mi
of species i, and with jumping unit molecular weights of M,(i = 1 or 2).
In a binary solution, the hole free volume is distributed equally among all
the jumping units. In cases for which species 1 is a simple molecule (such as
benzene), the jumping unit molecular weight,M,, is equal to the total molecular
weight of the solvent. In contrast, if species 2 is a polymer chain, then Ma is a
relatively small fraction of the total molecular weight. Combining Eqs. (2) and
(3) results in an expression for solvent self-diffusioninapolymer solution,
namely,
c; -.
where is the specific hole free volume of component i requiredfor a diffusive
-*
step and = Vli/V2+
A critical step in’ implementing free-volume concepts to describe transport in
polymer solutions involves quantifyingthe specific free volume,. ,p As a first
approximation, onemight assume that the total volume of a liquid is composed
of two parts, the occupied volume and the free volume (as suggested by Cohen
148 DUDA AND ZIELINSslyl
and Turnbull). The specific occupied volume of a liquid is generally defined as

the specific volume of the equilibrium liquid at 0 K [Q"(O)]. Hence, the specific
free volume of a species as a function of temperature is given by
cm= Q(T) - P(0)

where is the specific volume of an equilibrium liquid at any temperature
T. Although not directly measurable, can be estimated by the group con-
tribution methods discussed by Haward (1970), provided the chemical compo-
sition of the species of interest is known. Even though Eq. (5) can be used to
approximate the total free volume in a liquid, some question still exists as to
whether this free volume is the same as the one considered in the Cohen and
Turnbull framework.
It is possible that the total free volume cannot be continuously redistributed
without overcominga precise activation energy barrier. incorporate this con-
dition, Vrentas and Duda followed the concepts of Kaelble (1969) and divided
the free volume into two types. One portion, denoted the interstitialfree volume,
requires a large redistribution energy and is thus not implicated in facilitating
transport through the mixture. The remaining free-volume allotment, which is
presumed to dictate molecular transport, is termed to the hole free volume and
is redistributed effortlessly.
The Vrentas-Duda theory, therefore, regards vFH as the specific hole free
volume and precludesdetermination of frommeasurements of the specific
volume and estimates of the occupied volume, as suggested by Eq. (5). To
circumvent this problem, Vrentas and Duda adopted the ideology espoused by
Berry and Fox(1968) and developed arelationship between the hole free volume
and well-defined volumetric characteristics of the pure components in solution:
For polymer solutions, K,, and K2,denote free-volume parameters for the sol-
vent,while K,,and are free-volumeparameters for thepolymer. The glass
transition temperature of species i is given by Tgi.Throughout the ensuing de-
velopment, subscript l refers to the solvent and 2 to the polymer. These free-
volume parameters can be determined from pure-component viscosity data, as
a function of temperature, for the individual components in the solution. Fur-
thermore, these free-volume parametersare directly related to the parameters in
the Williams-Landel-Ferry (WLF) equation, which has been used extensively
for correlating polymer viscosity data (Williams et al., 1955).
In the original Cohen-Turnbull framework, a molecular jump does not in-
volve any activation energy but is solely related to the probability of locating a
sufficiently large free-volume void.Macedoand Litovitz (1965) andChung
(1966) challenged this conceptual notion and suggested that a jumping unit must
FREE-VOLUME 149
overcome the attractive forces with adjoining moleculesprior to a diffusive step.

According totheir viewpoint, therefore, theappropriate expression for the preex-
ponential term in Eq. (4) is
Dol = Do exp(-EIRT)
In this expression, E is the activation energy required for a jumping unit to

break free from its neighbors before it can moveinto a contiguous free-volume
void. This quantity should not be confused with the activation energy for dif-
fusion (ED),which is defined as
ED = R T a2 In
( D
y )
OlP
In concentrated polymer solutions where the availability of hole free volume

limits translational mobility, the preexponential factor Dol is often less dependent
on temperature than the exponential term related to free volume. In many cases
Dol can be taken as a constant.
When all of these proposed modifications to the original Cohen and lhmbull
theory are incorporated, the following expression emergesfor the self-diffusion
coefficient a solvent in a polymer solution:
analogous expressionfor the polymer self-diffusion coefficient, D2,was de-

veloped by Vrentas and Duda (1977b) based on the entanglement theory for
polymer solutions proposed by Bueche (1962). Although self-diffusion coeffi-
cients are intrinsic properties of a given solution and can be measured directly
by techniques such as nuclear magnetic resonance (NMR) and radioactive la-
beling, the mutual binary diffusion coefficient, D, reflects the mass transfer rate
required for the design and development of polymer processes and products.
Although self-diffusion coefficients constitute a measure of the mobility of the
various species in a homogeneous solution, the mutual binary diffusion coeffi-
cient provides the rate at which concentration gradients within a mixture dissi-
pate and is the critical quantityrequiredindefiningmosttechnological
applications.
Typically, molecular theories present expressions for the self-diffusion coef-
ficients (Dl and D2) rather than D . Thus, it is highly desirable to express D in
terms of D, and D2 indisputable relationship between the friction coefficients
that relate these diffusion coefficients does not, however, presently exist. Incor-
porating the work of Bearman (1961), Duda et al. (1979) proposed an approx-
150 AND DUDA ZIELINKI
imation for low solvent concentrations that couples D to the self-diffusion co-
efficients for polymedsolvent systems:
Here, p,, is the chemical potential of the solvent. Although many thermodynamic
theories are available to determine the concentration dependence of the solvent
chemical potential, the Rory-Huggins theory has provided an adequate repre-
sentation of polymer solution thermodynamicsin many cases 1942; Hug-
gins, 1942a,b). Consequently, Eq. (10) is often rewritten as
D = Dl(1 - W2(1- 2x44 (11)
where is the solvent volume fraction in the solution and x is the polymer-
solvent interaction parameter.
Combining Eqs. (g), and (11) results inanexpression for the mutual
binary diffusion coefficient in a polymer/solvent system. While this relationship
contains numerous parameters, each one possessesphysical significance. In other
words, these parameters are not added to simply facilitate empirical fits to data.
The total number of parameters needed to define a polymer/solvent system can
be reduced to 10 by grouping some together. The key resulting variables are
K&, K,, - T,,, Kl2Iy,K2, - Tgz, e;, x, Do,E, and 5, each ofwhich can
be estimated a priori without the use of any diffusivity data. Information that is
requiredhere to determine all of the parameters to estimate D includes the
following:
1. Chemical structure of both the solvent and the polymer
2. Viscosity versus temperaturedata for both pure components, or comparable
variable-temperaturerelaxation data (Zielinski et al., 1992; Spiess, 1990;
Bidstrup and Simpson, 1989; Blum et al., 1986; Dekmezian et al., 1985)
Density data for the pure solvent
4. Critical volume of the solvent
5. The Flory-Huggins x parameter or some other information characterizing
thermodynamic compatibility between the polymer and solvent such as sol-
ubility parameters
6. Glass transition temperature of the polymer, T,.
Comparison of theoretical predictions with experimental data suggests that
some parameters, particularly D, and 5, are critical for accurate predictions
whereas other parameters such as E and x are of secondary importance. For
example, reasonably good predictions have been attained by assuming that E
equals zero as a first approximation (Vrentas et al., 1989).
The solid lines in Fig. 1 correspond to predictions the mutual diffusion
coefficient for the toluene/poly(vinylacetate) system based on techniques of
FREE-VOLUMETHEORY 151
Zielinski and Duda (1992a). This system is representative of many in that al-
though the predictions are reasonable they may be nonetheless inadequate for
the design of a polymer processing operation. As one might expect, the avail-
ability of even a limited number of diffusion coefficient data points significantly
enhances the predictive capabilities of the theory. To illustrate this point, the
data at were used to obtain a better estimate of Do. The resulting semi-
predictive dashed lines shown in Fig. 1indicate that the theory does anexcellent
job of correlating diffusivity data and extrapolating to other experimental con-
ditions (e.g., 80°C and 110°C in Fig. 1). In contrast, the prediction of mutual
binary diffusion coefficientswithout the use of any experimental diffusivity data
is, at best, precarious.
"""----""_ -"
0.1 0.2 0.3 0.5
Comparison of mutualdiffusioncoefficientspredictedfromfree-volume
theory with experimental data for toluene/poly(vinyl acetate) system40,80,
at and 110°C.
Solid lines are based on the purely predictive techniques of Zielinski and Duda (1992a).
Dashed lines use data at 40°C to determine Do.
152 DUDA AND ZIELINSH
The primary hurdle in applying free-volume models has been the accurate
estimation of the jumping unit molar volumes, vlj vzj,
and required to calculate
6. survey of numerous studies indicates that solvents can be divided in two
categories. Spherical, rigid molecules belong to the first (and most elemental)
class and diffuse as single units. Larger solvent molecules canassume the shapes
of rods or chains and constitute the second category. In this case, Mij is less
than the total molecular weight of the solvent. vlj
a first approximation, can
be estimated as the molar volume of the solvent at 0 K [denoted here as R(O)]
if the entire molecule jumps as a unit. Experimental evidence reveals that mo-
lecular shape also influences the effective size of the jumping unit. and
coworkers (1992) developed an empirical relation between the effective molar
volume of solvent jumping units and weighted averages of the principal dimen-
sions of the solvent molecule.
Methodologies sought for estimating the size of the jumping unit of the poly-
mer chain, vzj, have been much more elusive. However, vzj can be estimated
from diffusivity data for any solvent in the polymer of interest because the size
of the polymer jumping unit is a characteristic feature of the polymer and is
therefore independent of the solvent. The only known protocol available to es-
timate vZj without the use of any diffusivity data is based on the correlation of
the jumping unit molar volume with polymer stiffness as a given by the glass
transition temperature (Zielinski and Duda, 1992a). correlation of this type is
anticipated to yield poor values ofvzj for polymers of high stiffness that none-
theless exhibit low glass transition temperatures, such as polybutadiene. Thus,
we emphasize once again that the theory’s ability to predict temperature and
concentrationdependence of diffusioncoefficients is significantlyenhanced
through the use of a few data points.
The free-volume theory for diffusion in polymer systems can be applied on
several levels. It is generallyacceptedthat the theory providesan excellent
framework for the correlation ofdiffusivity data andthatwithlimited data,
diffusion coefficients can be predicted at temperatures and solution concentra-
tions where data are not available.At the present state of development, however,
the predictive capabilities of the theory are inadequate for some applications.
Nonetheless, this is the only theory currently available that can predict mutual
binary diffusion coefficients in concentrated polymer solutions to within an order
of magnitude. The theory can also provide a basis for a qualitative understanding
of the influence of such variables as temperature, concentration, and solvent
molecular size on diffusional behavior. The followingrules of thumb are a direct
result of the free-volume formalism:
1. The apparent activation energy of the mutual binary diffusion process is a

strong function of temperature and concentration and increases as the
lution approaches Tgof the system [see Eq. (S)]. In contrast, changes in the
FREE-VOLUME 153
state of the system that remove it from T, (e.g., increases in solvent con-
centration or system temperature) result in a decrease in the apparent acti-
vation energy. Similarly, the concentration dependenceof the mutual binary
diffusion coefficient is greatest near T, and decreases with increasing tem-
perature and concentration. All of these trends are related to the fact that
increasing the temperature of the system or adding smallmolecules, which
possess larger free volumes than their neighboring polymers, will increase
the overall free volume available for molecular migration.
2. In general, the temperature and concentration dependence diffusion co-
efficients for polymer/solvent systems increases with solvent size. This is
particularly true for large, spherical solid molecules, which are expected to
move as single jumping units. The probability of a large molecule locating
sufficient free volume to take adiffusive step is relatively low; consequently,
small changes in the available free volume caused by increases in temper-
ature or solvent concentration can have a substantial impact on the diffusion
coefficient. In contrast, the diffusivities for systems of small molecules such
as fked gases in polymer systems are very weak functions concentration
and can often be viewed as independentof concentration. Similar arguments
suggest that the apparent activation energy for such systems is relatively
low and independent of temperature and concentration.
Elastomers or rubbers at conditions far above their T, possess a large frac-
tional hole free volume. Thus, diffusion coefficients under these circum-
stances are not only large in magnitude but are also relatively weak func-
tions of concentration and temperature.
4. In contrast to observed behaviorin dilute polymer solutions, themutual
binary diffusion in concentrated polymer solutions is a very weak function
of the molecular weight of the polymer. Although the free volume associated
with the chain segments at the end of each chain is greater than freethe
volume
associated with segments in the interior sections of the chain, the increase in
the number end units does not significantly impact the overall free volume
until relatively low molecular weight polymers are involved. In general, vari-
ations in molecular weight distribution and in the average molecular weight
most commercial polymers do not significantly influence diffusion in concen-
trated polymer systems in the regime processing interest.
5. The influence additives on diffusion characteristics is directly related to
the contribution the additives to the free volume of the system. For
example, plasticizers significantly enhance the available free volume, which
leads to a decrease in TB and anenhancement inthe rate molecular
migration. The addition of impermeable or solid additives (such as fillers)
maynotsignificantlymodifythe available free volume but will cause a
significant decrease in the diffusivity due to tortuosity effects, i.e., an in-
crease in the path length along which a molecular must travel.
154 DUDA AND ZIELINKI
6. In most cases, the free volume associated with the solvent is significantly
greater than the free volumeof the pure polymer, and thus the mutual binary
diffusion coefficient increases with the addition of solvent. As mentioned
above, this rate of change of the diffusivity with solvent concentration will
be greatest when the system is close to its glass transition and when large
solvents diffuse as single units. Thermodynamic forces, however, serve to
decrease the mutual diffusion coefficientwith increasing solvent levels. This
is particularlytruewhen the solvent is a poor solvent for the polymer.
Consequently, we have two forces acting in opposition. The net result is a
maximum in the mutual binary diffusion coefficient as a function of solvent
concentration that shifts to lower solvent concentrations as temperature is
increased. In rare cases when a polymer/solvent system at tempera-
tures far above the system glass transition, the thermodynamic forces dom-
inate and the mutual binary diffusion coefficient will actually decrease with
increasing solvent concentration over the entire concentration range (Iwai
et al., 1989). In contrast, the solvent self-diffusion coefficient, which does
not contain the thermodynamic term, will generally increase with solvent
concentration.
One concern often expressed regarding free-volumetheory is that molecular

interactions between the polymer chain and the solvent molecules are not ex-
plicitly considered. Specific polymer-solvent interactions are, however, intro-
ducedintwoways: (1) Equation (10) accounts for molecular interactions
through the chemical potential of the solvent in the solution, and (2) the acti-
vation energy, E , in the most general formulationreflects the interaction between
the solvent jumping unit and its neighbor. Clearly, the energy required for a
solvent molecule to break the interactions with its neighbors may depend upon
whether the neighbors are other solvent molecules or segments of the polymer
chain, and the possibility that E is a function of solvent concentration hasbeen
discussed by several investigators (Macedo and Litovitz, 1965; Vrentas et al.,
1980; VrentasandVrentas, 1993; Zielinski andDuda,1992b). Although the
incorporation of a concentration-dependent E will improve the correlation of
diffusivity data, it can be argued that this is strictly an empirical procedure that
merely introduces more adjustable parameters. At the present time, the activation
energy and its concentration dependence have not been related to fundamental
molecular interactions.
From all available experimental data, for concentrated polymer solutions in
the vicinity of the glass transition (T < T,, + 100°C), the diffusion process is
dominated by the scarcity of free volume, and specific molecular interactions
are of secondary importance. This conclusionmay be biased, though, due to the
fact that the great majority of accessible diffusivity data correspond to systems
that do not exhibit strong interactions such as hydrogen bonding.
FREE-VOLUME 155
Although many recent studies are based on the Vrentas-Duda model, this
discipline was dominated for several decades by the Fujita theory (Fujita, 1961).
Both the FujitaandVrentas-Duda models havetheir origin in the older
Cohen-Turnbull formalism. However, several significant differences exist be-
tween the two models.
From a practical point view, the principal distinguishing feature is that the
Fujita theory is purely correlative and is incapable of extrapolating results be-
yond the range of available data. Consequently, it has no predictive capabilities.
Perhaps the greatest difference conceptually is that the Fujita theory is based on
the free volume per unit volume of solution, whereas the Vrentas-Duda theory
is based on the average free volume per jumping unit (Vrentas et al., 1993). It
could be argued on conceptual groundsthat this difference renders the Vrentas-
Duda approach more consistent with the original Cohen-nmbull concept. A
more important question to be answered, though, is whether the Fujita theory
can be related directly to the Vrentas-Duda theory.
For the sake comparison, the Vrentas-Duda model can be rearranged to
the form reminiscent of the Fujita model (Vrentas et al., 1993):
where &(O) denotes the solvent self-difhsion coefficient at zero solvent con-
centration, R is the fractional hole free volume of pure component i, and eis
the specific volume of pure component i at the temperature of interest. This
relationship is based on the assumption that the partial specific volumes of the
solvent and polymer are independent of composition that no volume changes
arise upon mixing. Similarly, the following equation can be derived from the
Fujita theory:
where the free-volume parameters of the theory are &,f;, and In using the
Fujita expression, it is common practice to correlate experimental data to deter-
mine f& and f a d . A physical interpretation these parameters is not nec-
essary in employing Eq. (14) as a correlative model. However, the similarity of
the two theories becomes evident if the following relationships are assumed:
fl = fi,f2 = f2, and Bd = -&@. When related in this fashion, Eqs. (12) and
(14) reveal that the two theories are identical when
"-
This analysis demonstrates that the Vrentas-Duda expression reduces to the

Fujita model when certain restrictions are imposed. For example, therelationship
given by Eq. (15) is satisfied when the jumping units of the polymer and solvent
are identical and the specific volumes of the two species areapproximately
equal. Under these restricted conditions, the two theories are identical in form
and have the same number of parameters.
Although both theories are excellent correlative tools, the utilityof the
Vrentas-Duda theory arises from its ability to serve as a predictive model with-
out the use of diffusivity data from the polymer/solvent systemof interest. Un-
like the Fujita formalism, the Vrentas-Duda theory is also capable of describing
diffusional behavior over broad ranges of temperature and concentration from
limited diffusivity measurements. Thefundamental strength of the Vrentas-
Duda formalism is based on the determination of the free volume associated
with each jumping unit as an integral part of the generalization of the original
Cohen-Turnbull theory to describe solvent self diffusion.
W. THE INFLUENCE OF THE GLASS TRANSITION

As amorphousrubbers are cooled, the motionof individual polymer chains
becomes constrained that the cooling rate becomes faster than the rate at
which the polymer sample can volumetrically relax. The resulting nonequili-
brium condition is referred to as the glassy state. Thus, the passage from the
rubbery to glassy states is denoted the glass transition. The glass transition is a
dynamic phenomenon and occurs over a temperature increment whose range is
influenced by the thermal and mechanical history of the polymer as well as the
rate of cooling and/or frequencyof the experiment. A schematic representation
of the idealized volumetric behavior of a polymer above and below its Tg is
illustrated in Fig. 2.
At T > Tg, polymer chains are capable of achieving equilibrium configura-
tions, whereas polymer segments in the glassy state do not have sufficient mo-
bility to attain equilibriumconformationswithin commonlyreferencedtime
scales. Consequently, extra hole free volume becomes trapped withinthe poly-
mer as it is cooled through the glass transition. Although the rate of molecular
motion within glassy polymers prevents volumerelaxation from reaching equi-
librium, molecular motionsare not completely eliminated within a glass. Density
fluctuations persist and consequently necessitate redistribution of the hole free
volume.
LIQUID VOLU
J
>
z
3
a
Tg
TEMPERATURE
2 Characteristics thevolume of apolymeraboveandbelowtheglass

transition temperature (TA.
The formalisms presented in the preceding section would be appropriate for

elucidating and modeling moleculartransport in a glassy polymerif the polymer
were given sufficient time to relax to its equilibrium state. Due to the long time
scale required for glassy polymers to relax fully, however, diffusion in glassy
polymers typically occurs under nonequilibrium conditions wherein a polymer
sample possesses more hole free volume than it would at equilibrium. Conse-
quently, predictions of the availability of hole free volume in glassy polymers
based on transport measurements made in an equilibrium rubbery state are al-
ways appreciably low and consequently lead to predictions of diffusion coeffi-
cients in the glassy state that are lower than those measured experimentally.
We emphasize that although additional hole free volume is effectively trapped
in a glassy polymer, this volume continues to be redistributed throughout the
sample. Within the time frame of interest, the rate of volumetric collapse due
to relaxation is negligible. The development presented below for diffusion below
TB is basedon the concepts previously discussed (VrentasandDuda,
Vrentas and Vrentas, Vrentas et al.,
Accurate prediction of diffusion in the glassy state requires quantification of
the extra holefree volume trapped within the glassy state. With this accomplished,
the free-volume formalism developed for difhsion in the rubbery state can be
158 DUDA AND ZIELINKI
modified to encompass mutualdiffusion below the glass transition of the solvent/

polymer mixture, Tgm.A general relationship between the mutual diffusion co-
efficient, hole free volume, and the polymer-solvent thermodynamic interaction
can be written as
D = Do e x p ( g ) exp( + o,~Q;)
pm
)Q
The symbols in this equation retain their meanings from the last section, al-
though @m is used to denote the hole free volume of the mixture, rather than
vm7 to emphasize the glassy state. In addition, Q accounts for the concentration
dependence of the solvent chemical potential [see Eqs. (10) and (ll)]. Only the
magnitudes ofand Q areexpected to be affected by the glass transition.
The polymer/solvent mixture below T, is considered as a combination of
the glassy polymer at the temperature of interest with a liquid solvent at equi-
librium. Additivity of hole free volume yields
i",(T .c Tgm)= ",Pm1(T) + Wz*m(T) (17)
Here pm constitutes the hole free volume of the equilibrium polymer plus the
excess hole free volume trapped due to the glass transition:
where the excess hole free volume of the polymer, be estimated in

terms of the difference between the thermal expansion coefficients of the equi-
librium rubber and the glassy polymer (a,and aZg, respectively). Thus,
where @ @ ), is the specific volume of the equilibrium polymer at TBm'

The corresponding expression for the hole free volume in the glassy state,
therefore, becomes
%(T < T,) = d t m ( T ) + (20)

+ wz%(Tgm)(azg - a)(T - T,)
where and vm, are the hole free volumes of the pure species in the equi-
librium state and are related to the free-volume parametersKl,, Kzl, K12,and KZ
[Eq (6)], which can be estimated from pure-component viscosity data.
In summary, the hole free volume, for a glassy polymer/solvent system
can beestimated from the equilibrium hole free volumes of the pure components
in the system, and information concerning - aZ,T,, and @(TV). In ad-
dition, the overlap factor, y, has to be determinedindependently. Techniques are
available for estimating the TV dependence on solvent concentration (Chow,

1980). Similarly, e ( T ) is available for most commercial polymers from either
direct measurements or correlations. Consequently, - becomes the critical
parameter for quantifying the extra hole free volume in a glassy state, which
can depend strongly on the processing history of the system.
The secondparameter required to estimate dBusion coefficients in the glassy
state from free-volume concepts is Q. Despite the usefulness of the Flory-
Huggins theory in reflecting thermodynamic interactions between polymer and
solvent molecules withinpolymer solutions above the glass transition, the equi-
librium theory does not adequately describe thermodynamicsin nonequilibrium
glassy systems.
If thermodynamic data such as vapor-liquid equilibrium measurements are
available, Q can be calculated explicitly. Alternatively,correlative and predictive
formalisms that accurately portray polymer-solvent thermodynamic interactions
can be used to estimate Q. Numerous sorption models have been proposed in
the literature, and a few representative ones are listed below:
1. The dual-mode theory of gas sorption (Barrer et al., 1958; Michaels et al.,
1963; Vieth and Sladek, 1965) has been used extensively to describe the
solubility of gases in glassy polymers. The fundamental assumptionof the
theory is the existence of two distinct solute populations withinthe polymer
matrix (i.e., gas molecules that are either sorbed by an ordinary dissolution
mechanism or reside in preexisting voids trapped within the glassy polymer
upon cooling below TA. Local equilibrium between these two solute pop-
ulations is presumed to be maintained throughout the polymer matrix.
2. The Vrentas sorption model (Vrentas and Vrentas,1989, 1991a) differs from
the dual-mode sorption model in that it considers the existence of a single
sorption environment. In addition, the model accounts for the effect a pene-
trant moleculehas onpolymer structural rearrangement and subsequent
volumetric change.
Starting with the free-volume concepts introducedhere,Ganeshand co-
workers (1992) proposedafree-volume-based lattice modelthat predicts
both gas sorption and transport in glassy polymers.
4. Lipscomb (1990) described penetrant sorption as the combination of solid
deformation followed by mixing with the penetrant. The sorption isotherms
predicted from this theory are consistent with experimental data, and the
model parameters can be related to those in the dual model at the low-
sorptionbigh-bulk modulus limit.
5. Weiss et al. (1992) modified the dual-mode sorption model to be a multisite
model, based on a single continuous distribution of Langmuir isotherms.
Their model correlates data nearly as well asthe dual-mode sorption model
with one fewer adjustable parameter.
160 DUDA AND ZIELINSKl
6. The gas-polymer-matrix model (Raucher and Sefcik, 1983) assumesa sin-

gle sorption environment and correlates data equally as well as the dual-
mode sorption model. Correlations of dual-mode parameters with material
properties, however, seem to provide morephysical insight into the sorption
process (Barbari et al., 1988).
Astarita et al. (1989) incorporate an internal state variable, the free volume,
into a lattice model of polymer solutions. Their theory represents the sorp-
tion of carbon dioxide in poly(methy1 methacrylate) well.
Since many alternatives exist for estimating the polymer-solvent thermo-
dynamic interactions, Q is often readily available when mutual binary diffusion
coefficients in glassy polymer systems are to be estimated from the free-volume
theory for molecular diffusion. apparent paradox arises herein: Polymer
lutions are inevitably not at equilibrium below T,,, and yet equilibrium ther-
modynamicconcepts are suggested to estimate Q in the glassy state. The
assumption often introduced to address this seeming contradiction is that once
the nonequilibrium glassy structure forms (regardless of the temperature and
solvent concentration), the structure remains unchanged during the time frame
of interest. Relaxation of the polymer structure toward the equilibrium liquid
structure will unquestionably influence not only the thermodynamic relationship
of the system but the availability of hole free volume. Consequently, the
developments presented here are limited to systems that exhibit classical Fickian
diffusion.
The applicability of the free-volume approach topenetrant diffusion in glassy
polymers has often been questioned. Primarily, the debate addresses the consis-
tency of the theory with the physical concept associated with dual-mode sorp-
tion. Of the sorption models alluded to earlier, the dual sorption model is the
most widely implemented for gas sorption in glassy polymers. For molecular
transport in rubbers, however, the free-volume approach described here is more
applicable. Since the Vrentas sorption model uses the same principles exercised
in developing the diffusion modelsto address both rubbery and glassy polymers,
inthis section wecompare the fundamental differences between thetwo
approaches.
In dual-mode sorption @MS), certain regions of the polymer matrix do not
relax, owing to extensive chain entanglements that prohibit molecular reorien-
tation. Some segmentsof the free volume are subsequently trapped in the matrix
and are not continuously redistributed by random thermal fluctuations. These
locally trapped free-volume pockets donot contribute appreciably to the vacant
space available for transport, in which case not all of the excess free volume
associated with the glassy state is available for molecular transport.
From this point of view, the excess free volume associated with the glassy
state could be divided into two parts consisting of (1)excess hole free volume
and (2) excess interstitial free volume. The principal difference in physical states
conceptualized by the DMS and free-volume approaches is reflected in the pa-
rameter - ag.In free-volume theory, all the excess free volume in the glassy
state contributes to the hole free volume. This viewpoint is contrary to some
variations of the DMS approach that suggest that all of the excess free volume
is immobile and is trapped as interstitial free volume. At the present time, suf-
ficient experimental data are lacking to ascertain with any certainty if some of
the excess free volume in the glassy state should be identified as interstitial free
volume.
Vrentas and Vrentas (1992) demonstrated that the predictions of Eqs. (16)
and(20) are qualitatively consistent with observations of diffusion in glassy
polymer/solvent systems. Unlike diffusion above Tg,the free-volume theory has
not been extensively evaluated for glassy systems. At its present state of devel-
opment, the theory is semipredictive for glassy polymers, because parameters
such as azgdepend on the specific system history and cannot be related to pure-
component properties alone.
The most extensive data available for the evaluation of this theory at T Tg
exist for diffusion of trace amounts of solvent in a glassy polymer. In this limit
(as + 0), Eqs. (16) and (20) reduce to
e)
and
= [Ku+ X(T - T@)]

Y
where
h = 1 - (012 - a z g ) / K l z (23)
In this limit, by definition, the mutual binary diffusion coefficient, D , is equal
to the self-diffusion coefficient of the solvent in the polymer, Dl. These rela-
tionships suggest that diffusion of a trace amount of solvent relies upon X , which
reflects the extra hole free volume trapped in the glassy state. Aside from X , the
remaining parametersare identical to those used in modeling diffusion in poly-
mer melts.
Numerous experimental investigations using various analytical techniques re-
veal that the temperature behaviorof the diffusion coefficient is consistent with
predictions of Eqs. (21)-(23) when the glass transition temperature is traversed
(Coulandin et al., 1985; Amould, 1989; Hadj Romdhane, 1994). These studies
confirm that a single value of X can be used with success to characterize the
diffusion coefficients of different solvents in a single glassy polymer. This be-
havior is expected from Eq. (23), which specifies that is a characteristic quan-
tity of the pure polymer and therefore must be independent of the penetrant.
Representative data correlations are provided in Fig. 3 for the diffusion of tol-
uene in polystyrene above and below Tgz.
Although free-volume transport theory has not been extensively evaluatedfor
diffusion at finite solvent concentrations at temperatures belowT,, it continues
to lend itself to qualitative interpretations consistent with experimental obser-
vations. for the case of diffusion above the glass transition, we delineate the
qualitative behavior of diffusion in glassy systems suggestedby the free-volume
formalisms:
1. Contrary to intuitive beliefs, the rate of diffusion (or the diffusion coeffi-
cient) does notabruptly decrease as a system cools to the glassystate.
Rather, free-volume theory predicts that a fixed solvent concentration the
diffusioncoefficient is a continuousfunction of temperatureas T, is
traversed.
2. The apparent activation energy for diffusion, ED [see Eq. undergoes a
step change at T,. For a given polymer, the change in ED will be greater
for larger penetrant molecules. Consequently, it is possible that measure-
1Oa/(K,,-T,,+T) (K")
3 Free-volume theory correlation of data for the diffusion of toluene in pol-
ystyrene above and below the glass transition temperature. Data are from Pawlisch
(1985)
and Hadj Romdhane and Danner (1993).
FREE-VOLUME 163
ments of the diffusion of small molecules, such as permanent gases, will

not register any E,, change at Tp, whereas larger molecules, such as solvent
vapors, will exhibit significant increases in E,, as the temperature is lowered
through Tm In addition, the change in ED at T p will decrease as solvent
concentration increases, i.e., the influence of theglass transition on diffusion
is most apparent for trace amounts of large solvent molecules in the polymer
matrix.
3. Equations (16) and (20) relate solvent concentration to diffusivity. If, as is
usually the case, the hole free volume of the solvent is greater than that of
the polymer (vml vm),
> then the addition of solvent increases the hole
free volume andcorrespondingly increases thediffusivity.However,the
addition of solvent also lowers Tp and consequently reduces the excess
hole free volume associated with the glassy state [see Eq. (19)]. The addition
of hole free volume from the solvent often outweighs the loss of excess
hole free volume associated with loweringTp. This, however, is not always
the case, andthus the mutual binary coefficient candecrease withincreasing
solvent concentration(Vrentas et al., 1988; VrentasandVrentas,19921.
Solvents or penetrants that significantly relax the glassy matrix structure
withoutenhancing the free volume of the mixture are referredto as
antiplasticizers.
4. The rate of diffusion in a glassy polymer system is strongly influenced by
the processing history of the polymer. Aging or annealing cause a densifi-
cation of the polymer and a reduction in excess hole free volume. The free-
volume framework suggests that as a glassy polymer ages and relaxes to-
ward its equilibrium state, diffusional rates become slower.
V. MORECOMPLEXSYSTEMS
A. Lightly Cross-linked Systems
Equation (9) can be considered a general form of the free-volume theory. This
relationship can be modified to describe diffusion for specific classes of systems
through development of appropriate relationships for the parameter vm. Vrentas
and Vrentas (1991b) proposed modifying this theory to address solvent diffusion
in cross-linked polymers.In their development, systemsmust possess a relatively
low degree of cross-linking, X , such that 50 or more monomers separate neigh-
boring cross-link points.In this limit, as afirst approximation, manyof the
parameters in (16) be considered to be independent of the degree of
cross-linking. For example, the solvent properties vml,v,*l, and p: are clearly
independent of X . Furthermore, Vrentas and Vrentas (1991b) suggest that the
distribution of free volume in the polymer and the size of the jumping unit of
- polymer chaindo
the notdependon X, inwhich case neither does
V,;, nor p;. The above approximations leadto the result that 5 is also independent
164 AND DUDA ZIELINKI
of X. When the cross-link points are sufficiently spaced, the jumping rates of
the solvent molecules and the interactions between these molecules and their
neighbors (which are represented by D,, and E ) will not be influenced by the
presence of the cross-links. Following this line of reasoning, one is led to con-
clude that cross-linking a polymerinfluencesonly the diffusioncoefficient
through the thermodynamics of the polymer-solvent interactions and the spe-
cific hole free volume of the polymer, pm2.The addition of chemical cross-links
to a polymer will obviously inhibit the segmental and molecular motion of the
chains that arise due to thermal fluctuations; consequently, the free volume of a
cross-linked polymer is expected to be less than that of a non-cross-linked poly-
mer. This anticipatedbehavior is substantiated by the fact that cross-linking
increases the polymer density. Furthermore, it seems reasonable to assume that
the loss of free volume associated with the formation of cross-links reduces the
hole free volume, which dictates solvent transport, and that the occupied and
interstitial free volumes of the polymer are both independent of X . Thus, the
theory must be appropriately modified to incorporate the influence of X as well
as temperature on pm2for cross-linked polymers. One way to do presumes
that the hole free volume of a cross-linked material at a given temperature,
pm2(T,X),is a fractional portion of the hole free volume of the polymer in the
absence of cross-links, i.e.,
Vrentas and Vrentas (1991b) presented both theoretical and experimental evi-
dence to indicate that is virtually independent of temperature but is related to
the specific volumes of the pure cross-linkedanduncross-linkedpolymer,
@(T,X) and @(T,O), respectively:
Consequently, the influence of chemical cross-linking on the polymer free vol-

ume can be characterized by a single parameter, which is determined directly
from volumetric data on both the cross-linked and uncross-linked polymer.
Following the Vrentas and Vrentas (1991b) development in the limit of a
trace amount of solvent in a polymer (col+ 0), Eq. (9) takes the form
When = 1, this relationship correctly reduces to the expression for solvent

self-diffusioninanuncross-linkedpolymer, whichhas a hole free volume
pm2(T,0). Experimental measurementsprobing the influenceof temperature,
concentration, degree of cross-linking, and solvent size on Dl (in the limit of
FREE-VOLUME 165
o1 - 0) are all relatively consistent with the predictions of Eq. (26). Conse-
quently, the incorporation of only one new parameter, is required to modify
the conventional free-volumetheory to address diffusive transport in amorphous,
lightly cross-linked systems. Although the theory for cross-linked systems has
not been extensively evaluated, the following qualitative behavior is suggested
by this free-volume formalism:
1. The diffusion coefficient decreases with increasing cross-link density, and
this increase can be quite significant. Furthermore, the decrease in the dif-
fusivity due to the cross-linking is more pronounced for larger penetrants.
2. The activation energy for diffusion, ED,increases with increasing degree of
cross-linking. Furthermore, the influence of cross-linking on ED is enhanced
for larger penetrants. Consequently, the activation energy for diffusion in a
polymer with a particular cross-link density increases with increasing pen-
etrant size.
The diffusion coefficient in a cross-linked polymer increases as the solvent
concentration is increased, since low molecular weight solvents or polymers
bring more hole free volume to the system. The influence of concentration
pure polymer limit (ol -

on the diffusivity is more pronounced in cross-linked systems, and in the
0) the dependence of the diffusivity on penetrant
concentration increases as the cross-link density increases.
Numerous polymer processes and applications involve multicomponent diffusion

such as in the formation of many coatings, devolatilization of solvent mixtures,
and membrane separation of two or more species. Vrentas et al. (1984, 1985b)
considered the free-volume framework for the case of two solvents diffusing
through a polymer during devolatilization. From basic diffusion theories, four
independent diffusion coefficients are required to describe fully the molecular
fluxes of all the species in a ternary system. For the case of mutual diffusion in
a ternary system, the mass diffusive fluxes relative to a volume-average velocity
can be related to the concentration gradients in the solution by four diffusion
coefficients:
aP1
j l = -Dll - - D,2 -
ax ax
+ Pj2+ =0 (29)
where j i is the mass diffusive fluxof species i relative to volume average velocity,
is the mass density of species i, and plis partial specific volume of species i.
166 AND DUDA ZIELINSKI
The free-volume theory, like most fundamental theories of diffusion, results

in expressions that describe the self-diffusion coefficient of a species in solution.
Relating self-diffusion coefficients to the diffusivities, Dij, that describe mutual
diffusioninamulticomponent system is nontrivial.In fact, Bearman (1961)
showed that no unique relationship exists between self-diffusion coefficients and
mutual diffusion coefficients. Vrentas et al. (1985b) showed that for the case of
two solvents (species 1 and 2) in a polymer (species in some concentration
interval near the pure polymer limit (w3 l), the following expressions can be
-+
obtained:
Dl1
Dl*
D22
--
+ D1
0
D2
1)
(30)
(3
(32)
D21 + 0 (33)
where D l and D2 are the self-diffusion coefficients for solvents 1 and 2, respec-
tively. In close proximity to the pure polymer limit, the principal diffusion co-
efficients (Dll and DZ2)are significantly larger than the cross-diffusion coeffi-
cients (Dl2and DZl)and are approximately equal to the self-diffusion coefficients
of the two solvents. Consequently, multicomponent diffusion taking place in a
process such as devolatilization that involves low concentrations of the constit-
uent solvents can be conveniently analyzed with only the self-diffusion coeffi-
cients of the two solvents.
The basic free-volume expression forthe solvent self-diffusion coefficient
[Eq. can be readily modified for a ternary system. First, the distribution of
the available hole free volume amongall the jumping units of solvent 1, solvent
2, and the polymermust be considered. In addition, the available hole free
volume must include contributions from the two solvents as well as from the
polymer. The resulting free-volume expressions for the solvent self-diffusion
coefficients in a ternary system of two solvents and a polymer are
where D , is the preexponential factor for component i, KIi and KZ are free-
volume parameters for component i, and vm
is the average hole free volume
per gram of mixture.
FREE-VOLUME 167
As for binary polymer/solvent systems, the parameters in the free-volume

expressions for D , and D2 can be experimentally determined from volumetric,
viscosity, and diffusivity data for single-component or binary systems. Recall
that for each of the three components can be estimated fromgroup contri-
bution techniques for estimating the equilibrium liquid volume of a component
at 0 K The parameters K l i / y and KZi- Tgi for each component can also be
discerned from viscosity data. Finally, Do,, &,Dm,and can be attained from
diffusivity data for the binary polymer/solvent systems. Ferguson and von Meer-
wall (1980) were the first investigators to modifythefree-volumetheory to
describe a ternary system and correlate the self diffusion of two solvents in a
ternary polymer solution.
In conclusion of this section, a fairly straightforward modification of the free-
volume approach leads to formalisms for the self-diffusion coefficients of two
solvents in a two-solvent/polymer ternary system. These self-diffusion coeffi-
cients can be used to describe mutual ternary diffusionunder conditions of
relatively low solvent concentrations. Extension of the present theory to describe
mutual diffusion for the entire concentration range in a ternary system would
require a suitable approximation regarding the relationship between friction co-
efficients as discussed by Bearman (1961) and Vrentas et al. (1985b).
C. Block
Chemical modification of polymer matrices provides enhanced opportunities
to regulate and tailor materials for diverse technologies such as packaging,
coatings, and separations. Within the last 20 years oneparticular class of chem-
ically altered polymers has received significant experimental and theoretical
attention due to the unique characteristics of the chemical species; they are
block copolymers.
As the name implies,block copolymersare chains composed of two (or more)
monomer species inwhich long linear sequences of like monomer units are
covalently bonded together. This physical attribute leads to interesting morpho-
logical repercussions because phase separation is restricted by covalent bonds.
Microphase separation produces a variety of ordered morphologies and occurs
in block copolymers when (1) sufficient thermodynamic incompatibility exists
between the blocks and (2) the blocks are long enough to self-assemble into
microdomain structures. The macroscopic physical attributes of block copoly-
mers can be finely tuned through specific tailoring of the microstructure.
Several studies have focused measuring (and modeling) the rate of dif-
fusion in microphase-separated block copolymers (Rein et al., 1992; Ferdinand
andSpringer, 1989; Csernica et al.,1987). As far as we know,however, no
attempt has addressed transport in homogeneous, or disordered, block copoly-
mers at conditions well removed from the order-disorder transition (OD”).This
168 DUDA AND ZIELINSKl
section, therefore, focuses on solvent self-diffusion within atwo-component,

homogeneous block copolymer melt. In the following analysis, 1 and 2 refer to
the solvent and polymer, respectively, while 2a and 2b correspond to blocks A
and B of the copolymer.
If pm denotes the specific hole free volume in a block copolymerlsolvent
mixture, then the molar free volume available for molecular transport can be
written as
Here, oiis the weight fraction of component i (i = 1 or 2), and and 0 2 b are
the weight fractions of blocks A and B within the copolymer, respectively. The
molecular weightsof the jumping units for the solvent, polymer A, and polymer
B are given by M,, MZja,and M2,%,respectively. o1and o2sum to unity, as do
02a and 0 2 b .
Substitution of Eq. (37) into (2) and introduction of the overlap factor
(7) and the energy to break free from neighbors ( E ) yields the transport ex-
pression for solvent diffusioninahomogeneous block copolymer melt. The
resultant expression can becast into a form comparable to that provided earlier
if
and
Then
where (k = a or b) is the specific volume of block k in the copolymer at 0

K In the limit that the copolymer becomes a single-component homopolymer
(i.e., 02a= 0 2 b = 1, or = B), then Eq. (40) correctly collapses to the original
expression for solvent self-diffusion in a homopolymer [see Eq. This re-
lationship atthe very least suggests self-consistency with the methodology
adopted.
Equation (40) is expected to describe solvent self-diffusion inany
monomer block copolymer irrespective of molecular architecture [e.g., AB
diblocks, ABA triblocks, or (AB),, multiblocks], provided the copolymers are
homogeneous melts. The difference in transport characteristics for the various
architectures is expected to arise in the vmly term.
Previous researchers have investigated the influence of microstructural mor-

phologyand chemical composition on transport within microphase-separated
block copolymersand blends (e.g., Kinning et al., 1987; Sax and Ottino, 1985).
Although not elaborated upon here,we speculate that the mechanism of diffusion
in microphase-separated block copolymers may be reminiscent of that in semi-
crystalline polymers. This analogy is particularly appropriate when one of the
phases is a glassy polymer with low diffusion characteristics and the other is a
continuous rubberyphasethrough which solvent molecules can diffuse with
relative ease. Therefore, perhapsa simple modification of Eq. (40) to reflect the
reduction in transport rate as a result of tortuosity may be adequate to describe
the effect of morphology on solvent transport in ordered systems.
Amould, D., and R. L. Laurence (1992). Znd. Eng. Chem. Res., 31, 218.
Amould, D. D. (1989). Capillary column inverse gas chromatography for the study of
diffusion in polymer-solvent systems, Ph.D. Thesis, Univ. Massachusetts, Amherst,
MA.
Astarita, G., M. E. Paulaitis, andR. G. Wissinger (1989). J. Polym. Sci., B: Polym Phys.,
27, 2105.
Barbari, T. W. J. Koros, and D. R. Paul (1988). J. Polym. Sci., B: Polym. Phys., 26,
729.
Barrer, R. M.,J. Bame, and J. Slater (1958). J. Polym. Sci., 27, 177.
Bearman, R. J. (1961). J. Phys. Chem., 65, 1961.
Berry, G. C., and T. G. Fox (1968). Adv. Polym. Sci., 5, 261.
Bidstrup, S. and J. Simpson (1989). Proc. 18th N.Am. ThermalAnal. Soc., 1,366.
Blum, F. D., B. Durairaj, and S. Padmanabhan (1986). J. Polym. Sci. B: Polym. Phys.,
24,493.
Brandt, W. (1955). Phys. Rev., 98, 243.
Bueche, F. (1962). Physical Properties of Polymers, Interscience, New York.
Chow, T. S. (1980). Macromolecules, 24, 2404.
Chung, H. S. (1966). J. Chem. Phys., 44, 1362.
Cohen, M. H., and D. 'hrnbull (1959). J. Chem. Phys., 31, 1164.
Coulandin, J., D. Ehlich, H. Sillescu, and C. H. Wang (1985). Macromolecules, 18, 587.
Csernica, J., R. F. Baddour, and R. E. Cohen (1987). Macromolecules, 20, 2468.
Dekmezian, D. E. Axelson, J. J. Dechter, B. Borah, and L. Mandelkern (1985). J.
Polym. Sci. B: Polym. Phys., 23, 367-385.
DiBenedetto, T. (1963). J. Polym. Sci., A , 1; 3477.
Duda, J. L.,Y. C. Ni, and J. S. Vrentas (1979). Macromolecules, 12, 459.
Duda, J. L., J. S. Vrentas, S. T. Ju, and H. T.Liu (1982). AIChE J. 28, 297.
Ferdinand, and J. Springer (1989). Colloid Polym. Sci., 267, 1057.
Ferguson, R. D., and E. von Meerwall (1980). J. Polym. Sci., B., Polym. Phys., 18.
Flory, P. J. (1942). J. Chem Phys., 10, 51.
Fujita, H. (1961). Forsfschr.Hochpolym.-Forsch., 3, 1.
170 DUDA AND ZIELINSSKI
Ganesh, K., R. Nagarajan, and J. L. Duda (1992). Ind. Eng. Chem Res., 31, 746.
GUO, C.J., D. DeKee, and B. Harrison (1992). Chem. Eng. Sci., 47, 1525.
Hadj-Romdhane,I. (1994). Polymer-solvent diffusion and equilibrium parameters by
inverse gas-liquid chromatography, Ph.D. Thesis, Pennsylvania State Univ.
Hadj Romdhane, I. and R. P. Danner (1993). AIChE J., 39, 625.
Haward, R. N. (1970). J. Macromol. Sci. Rev. Mucromol. Chem., C4, 191.
Huggins, M. L. (1942a). J. Am. Chem. Soc., 1712.
Huggins, M. L. (1942b). J. Phys. Chem., 46, 151.
Iwai, Y., S. Maruyama,M.Fujimoto, Miyamoto,and Y. Arai (1989). Polym. Eng.
Sci., 29(12), 773-776.
Kaelble, D. H. (1969). In Rheology, Vol. 5, F. R. Eirich, Ed., Academic, New York, p.
223.
Kinning, D. J., E. L. Thomas, and J. M. Ottino (1987). Macromolecules, 20, 1129.
Lipscomb, G. G. (1990). AIChEJ., 36(10), 1505.
Macedo, P. B. and T. A. Litovitz (1965). J. Chem Phys., 42, 245.
Michaels, A. W. R. Vieth, and H. Bixler (1963). J. Polym. Lett., 1, 19.
Pace, R. J. and A. Datyner (1979). J. Polym. Sci., B., Polym. Phys., 17, 437-451.
Pawlisch, C.A. (1985). Measurement of the diffusive and thermodynamic interaction
parameters of a solute in a polymer melt using capillary column inverse gas chro-
matography, Ph.D. Thesis, Univ. Massachusetts, Amherst, MA.
Raucher, D., and M. D. Sefcik (1983). ACS Symp. Ser., 223, 111.
Reid, R. C., J. M. Prausnitz, and T. K Sherwood (1977). The Properties of Gases and
Liquids, 3rd ed., McGraw-Hill, New York.
Rein, D. H., R. F. Baddour, and R. E. Cohen (1992). J. Appl. Polym. Sci., 45, 1223.
J. E., and J. M. Ottino (1985). Polymer, 26, 1073.
Spiess, K.W. (1990). Polym. Prepr., 31, 103.
Vieth, W. R., and K. J. Sladek (1965). J. Colloid Sci., 20, 1014.
Vrentas, J. and J. L. Duda (1977a). J. Polym. Sci., 15, 403.
Vrentas, J. S. and J. L. Duda (1977b). J. Polym. Sci., 15, 417.
Vrentas, J. S. and J. L. Duda (1978). J. Appl. Polym Sci., 22, 2325.
Vrentas, J. S., and J. L. Duda (1986). DifFusion, in Encyclopedia of Polymer Science and
Engineering, Vol. 5, H. F. Mark, N. M. Bikales, C. G. Overberger, and G. Menges,
Eds., Wiley, New York.
Vrentas, J. S. and C. M. Vrentas (1989). Macromolecules, 22, 2264.
Vrentas, J. and C. M. Vrentas (1991a). Macromolecules, 24, 2404.
Vrentas, J. and C. M. Vrentas (1991b). J. Appl. Polym. Sci., 42, 1931.
Vrentas, J. S. and C. M. Vrentas (1992). J. Polym. Sci.: B Polym. Phys., 30, 1005.
Vrentas, J. and C. M. Vrentas (1993). Macromolecules, 26, 1277.
Vrentas, J. H. T. Liu, and J. L. Duda (1980). J. Appl. Polym. Sci., 25, 1297.
Vrentas, J. J. L. Duda, and H. C. Ling (1984). J. Polym. Sci: B Polym. Phys., 22,
459.
Vrentas, J. J. L. Duda, and A. C. Hou (1985a). J. Appl. Polym. Sci., 31, 739.
Vrentas, J. J. L. Duda, and H. C. Ling (1985b). J. Appl. Polym. Sci., 30, 4499.
Vrentas, J. S., J. L. Duda, and H. C. Ling (1988). Macromolecules, 21, 1470.
Vrentas, J. C. W. Chu, M. C. Drake, and E. von Meerwall (1989). J. Polym. Sci., B:
Polym. Phys., 27, 1179.
Vrentas, J. S., C. M. Vrentas, and J. L. Duda (1993). Polym. J., 25(1), 99-101.
Weiss, G. H.,J. T. Bendler, and M.F. Shlesinger (1992). Macromolecules, 25(2), 990.
Williams, M.L., R. F. Landel, and L. D. Ferry J. Am. Chem. Soc., 77, 3701.
Zielinski, J. M.and J. L. Duda (1992a). AIChE J., 38, 405.
Zielinski, J. M., and J. L. Duda (1992b). J. Polym. Sci., B: Polym. Phys., 30, 1081.
Zielinski, J. M., J. Benesi, and J. L. Duda (1992). Znd. Eng. Chem. Res., 31, 2146.
Transport Phenomena in Polymer
Membranes
Neogi
University of Missouri-Rolla
Rolla, Missouri
INTRODUCTION
The literature on diffusion is vast and is mainly mathematical. The two books
by Crank (1975, 1984), for instance, have become a part of the standard reading
material on the subject. Other authors, like Cussler (1976), feel that much on
diffusion can be learned without resorting to such endless mathematics. In a
very specialized area such as polymers, the conventional mathematicalmodeling
ends a little too quickly. In this chapter the origins of the key equations, their
representations, and methods solution are analyzed first. The unusualvariants
of the equations importance to this area are discussed next. analysis of
the role of mass transfer in bringing about morphological changes follows. One
of the well-known examples where such changes are engineered for applications
is the reverse osmosis membrane, discussed in Section IV. Finally, in Section
V, systems that the investigators have put together with applications in mind are
reviewed. The special features of these systems are that they contain more than
one component or phase and their transport phenomenaare quite differently
organized. There are pressing reasons to understand them quantitatively. The
very new area of measurement of diffusivity with NMR is one example.
The conservation of species equation in one dimension (as appropriate in
membranes) is given by
174 NEOGI
where c is the concentration of the diffusing species, t is the time, and j x is the
flux in the direction, which is along the thickness of a membrane and the only
direction in which mass transfer is taking place. If is the area of the face of
the membrane, then Ala >> L, the membrane width, which makesthe dynamics
one-dimensional. Equation (1) assumes that there is no convection. Actually, a
convective term arises due to a change in the volume as the polymer swells in
the presence of the solute. However, this term is not considered as Duda and
Vrentas (1965) showed that it has no effect unless the excess volumeof mixing
is nonzero,and the excess volume of mixing is almost always neglectedin
polymers, with somejustification.
The entire problem needs to be converted to a boundary value problem in
concentration c, which requires that the flux be related to the concentration field.
The simplest way to do this is to employ Fick’s law,
This form is valid when the solute is dilute (Bird et al., 1960) or when the
volume-averaged reference velocity is being used or assumed (Cussler, 1976).
Combining Eqs. (1) and (2), one has
!?=a
at ax
(.E)
which is essentially the equation that has to be solved subject to appropriate
initial and boundary conditions. The object then becomes to describe the mea-
sured quantities in terms of diffusivity D, following which the theory and ex-
periments can be compared to back out a number for D .
In the simple form of sorption experiments, a membrane is suspended in
vacuum. A vapor or gas is then introduced and maintainedat a constant pressure.
The solute dissolves and diffuses into the membrane, and the weight gain is
measured gravimetrically. The data are reported as the fractional mass uptake
(with respect to the equilibrium value) as a function of time. For constant D ,
the solution to Eq. leads to
” 4 D ( h L ; l)’nit]
(4)
M,
where M, and M, are the mass uptakes at time t and at infinite time and L is
the membrane width. Obviously, the solubility can be calculated from M,. It is
seen in Eq. (4) that the exponential terms decreasedrastically with m. If a half-
time is defined as t = t,, where MIIM, = 112, then Eq. (4) can be approximated
as
1 8 4Dr2
2 ,IT
TRANSPORT
PHENOMENA IN POLPNER MEMBRANES 175
or
It has been assumed that tIn is sufficiently large that all terms other than the
first in the series can be neglected. Equation (4) does not quite show what the
solution is like. This is given by another solution in the form
3
M, = 8 @ ) I R + 2%
m=O
ierfc
(-l)m
"I
where ierfc is the integral of the error function (Crank, 1975, p. 375). At short
times, Eq. (6) approximates to
1R
M,
that is, only the first term is important. If one insists on calculating the half-
time from the approximate equation, Eq. one has
Equations (5) and (8) are respectively

tln = 0.01224L2/D (94
and
tln= 0.01227L2/D
J
Obviously,Eq. ( gives a good description of the solution, which is that M,/
M, is linear in t/L past the half-time but away from equilibrium as shownin
Fig. la. The experimental data are plotted against g t / L , and Eq. (9) is used to
calculate D.
In the permeation experiments the two sides of the membrane, which are
initially under vacuum, are sealed from one another. Then the gas is intro-
duced on the upstream side and kept at a constant pressure p l . On the down-
stream side the pressure p 2 slowly rises as the permeant is being stored. How-
ever, the magnitudes of the pressures are such that p , >> p2(t) = Under those
conditions the total amount that has permeated through the membranes Qt is
given by
"""_
Ql Dt
LC, - L'
1 1
6 ,=,
NEOGI
1 (a)Thefractionalmassuptakeversus t”/L inasorptionexperiment.The

diffusivityiscalculatedfromthehalf-timeortheinitialslope. (b) Theresultsfrom
permeationexperimentsare in a form such that the slope at steady state gives
permeability and the intercept on the t axis leads to diffusivity.
where c, is the upstream concentration, and, if Henry’s law holds,
Cl = HP, (11)
where H is the Henry’s law coefficient. If V, is the volume of the container

downstream, then under the ideal gas law
~ 2 V 2= (&APT (12)
TRANSPORT
PHENOMENA
IN POLKVER MEMBRANES 177
where A is the area of the membrane. That is, p 2 can be monitored to get Q, as
a function of time. At large times Eq. (10) becomes
et_"_
Dt 1
LHp,
L2 6
or
This is exactly at steady state, and P is called the permeability. The units of
permeability are cubic centimeters of gas that passes through the membrane at
STP per second per atmosphere pressure drop per square centimeter membrane
area, times the total membrane thickness in centimeters (Pauly, 1989). The per-
meability is seen to be a property of the solute-polymer interaction only, and
its importance lies in the fact that when two solutes have permeabilities suffi-
ciently removed from one another they can be separated by using a membrane.
Equation (13a) can be rewritten as
Equations (10) and (13c) are shown in Fig. lb. The asymptote given by Eq.
(13c) makes an intercept L216D on the t axis and has a slope of DH, which
allows one to find both the solubility and the diffusivity.
Permeabilities are so low that large pressure differences and membranes of
small thicknesses have to be employed. These restraints tend to exclude con-
densable vapors. In contrast, only vapors of sufficientlyhigh solubilities are
suitable for the gravimetric measurementsusedinthe sorption experiments.
Consequently there are only a very few systems for which both sorption and
permeation results have been reported. few examples have been given by
Stem et al. (1983), Kulkami and Stem (1983), and Subramamian et al. (1989).
compilation of conventionally measured values was made available by Pauly
(1989).
The experimental apparatus has changed very little in its basic outline, a
feature that is made clear by reviews (CrankandPark, 1968; Rogers, 1985;
Vieth, 1991). Corrections that need to be consideredwhen the reservoir pressure
changes due to dissolution have been quantified. In improving the scope of the
experiments, the emphasis appears to have beenlaid on measuring mass or
pressure more accurately, on measuring changesin the dimensions of the mem-
brane, and on the ability of the system to go tohigher pressures or temperatures.
[See in particular the systems developed by Stem and coworkers and Koros and
coworkers as cited by Vieth. See also Costello and Koros (1992).] very dif-
178 NEOGI
ferent approach was reported by Vrentas et al. (1984b, 1986) where the input
was oscillatory. The results show that the method enjoys additional advantages
over the step-changehorption experiments, whichare discussed later.
II.
When the difhsivity D is a constant, Eq. (2) becomes
Equation (14) is similar to the heat conduction problem where the temperature
replaces concentration and the thermal diffusivity replaces D. The book onheat
conduction by Carslaw and Jaeger (1959) and one on difhsion by Crank (1975)
cover a great range of solutions to equations such as m. (14). These include
various geometries, initial and boundary conditions, concentrated sources and
sinks, simple composites, etc. The solutions are all analytical, or exact, as they
are generally called. It becomes a little difficult to solve for the case where the
diffusivity is a function of concentration. In that case Eq. becomes
ac aD ac
-=-(-).+D- a2c
at ac ax ax2
Crank (1975) provides a general discussion on the subject. Some useful sim-
plifications can be made. One of them is that the diffusivity is an increasing
function of concentration, and a simplistic representation is
D = Doemc (16)
where Doand are constants. The impact of concentration dependence ondif-
fusivities is discussed next.
Concentration-DependentDiffusivities
A critical theorem in this area is one due to Boltzmann (1894), who showed
that even for this case (7) holds away from equilibrium. example is
shown in Fig. 2. The apparent diffusivity, obtained using an equation such as
Eq. (9, is obviously an average value over the concentration range employed.
It is also suggested that in this range there exists one concentration value that
corresponds to this diffusivity. However, it has been difficult to find that con-
centration. The important observation to make here is that as long as the dif-
fusion is Fickian, an effective constant diffusivity can be used to mimic the
sorption results. It is shown later that this holds even under more severe con-
ditions (see Figs. 6 and 8).
TRANSPORT
PHENOMENA
POLYMER
IN MEMBRANES 179
.T~//L
XI m in 1Wcm
2 The data of Kishimoto for methyl acetate in poly(methy1 acrylate) at 35"C,

as reproduced by Fujita (1968).
180 NEOGI
Diffusivities of solutes in solid polymers are almost always givenbyan

increasing function of concentration as shown in Eq. (16). Specifically, for dif-
fusivities given by Eq. (16) and for concentration-dependent diffusivities in gen-
eral, Crank (1975) put together a detailed compilation of the approximate
lutions. These solutions tend to become more approximate asthe concentration
dependence becomessteeper and to become more exact as dfisivities become
less concentration-dependent. Neogi (1988) obtained an approximate solution
for the case when in Eq. (16) is very large. Many of the characteristics ob-
tainedthere are well known frombefore.Boththeoldandnew results are
summarized below.
1. Boltzmann's scaling is upheld.It is seen that the fractional massuptake

when plotted against t is almost linear. However, the telltale sign that it is
actually Fickian is observed in thefact that thefractional mass uptake curves
make a slope of 90" at the origin. Grayson et al. (1987) and Korsmeyer et
al. (1986) appear to have observed such responses experimentally.
2. When diffusivity is a constant, the concentration profiles are seen to change
smoothly and gently within the polymer. If the diffusivity is an increasing
function of concentration, then a concentration sharpening of the profile
takes place and a sharp front is obtained. This happensbecause for the same
concentration gradient, the flux is greater at the greater concentration. As a
result, the movement of the concentration front is controlled by the region
oflowest concentration. Calculated concentrationprofiles showing self-
sharpening are illustrated in Fig. 3. It should be noted that (15) is clas-
sified as a parabolic equation (Ames, 1965) and can in no case lead to a
sharpening to the extent of showing a discontinuous shock, that is, where
the profile drops suddenly and discontinuously.
3. In desorption, the fractiondesorbedshouldbeequaltothefractional
mass uptake if the diffusivity is a constant. If the diffusivity increases
with concentration, then the desorption is slower than sorption, as seen
in Fig. 2.
Eventually the problem reduces to the fact that it is difficult to obtain the
concentration dependence of diffusivity without using a model for diffusivity
such as Eq. (16). At least experimentally this problem can be handled partially.
In the sorption experiments a step change is given from the initial concentration
of zero to c,. This is the integral sorption. If the step change is made from ci to
c,, where ci is very close to c, then it could be assumed that the diffusivity can
be approximated to be a constant and evaluated at a mean concentration, say
(1/2) (c, + ci).This is the differential sorption and provides informationon
diffusivity as a function of concentration.
TRANSPORT PHENOMENA IN POLWER MEMBRANES 181
~/(4w,l)~
3 Numericalsolutions of concentrationprofiles(dimensionless) in asemi-
infinite system where the diffusivityis concentration-dependent,are shown. The numbers
on plot are the different values dimensionless a defined in Eq. (16). [Reproduced
from Crank (1975), with permission.]
Solutions
The ability to solve these equations numerically has greatly improved our un-
derstanding of the diffusion process. The solutions obtained are approximate. In
finite-difference schemes the partial differential equations are converted to al-
gebraic equations (more accurately, difference equations). Instead of the contin-
variable one has a discrete set of points {xj}, and instead of time t one
has the set {ti}. Likewise, the concentrations become c, and the derivatives
become
and
Inserting @S. (17) and (18) into (14) and rearranging, one has
where the increments At = ti+l - ti and = xicl - xi are constants. If the

initial concentration profiles ckj are known, then Eq. (19) provides the means
182 NEOGI
for moving up in time to ci+lj. Such iterations can be continued until the nu-
merical approximation to equilibrium or steadystate is achieved.
It is obvious from Eqs. (17) and (18) that the basic representation (forward
time central space, FTCS) in this finite-difference scheme is an approximate one
that improves asthe size of the increments is decreased. However, the price that
is paid in this process is that the computationsbecomeimpossibly lengthy.
Besides accuracy, stability provides an important criterion for constructing an
effective algorithm. For a number of reasons, errors accumulate at nodes, such
as at the (i,j) node. The question that arises now is that of whether this error
would propagate. If in the next iteration, at (i + lJ),this error increases, then
the algorithm is unstable. Linear stability analysis for the system given by Eq.
(19) shows that when the stability ratio R = D At/(Ax)’ 1/2 (Carnahan et al.,
1969), the system is stable. This causes considerable difficulties. If Eq. (19) is
cast into dimensionless form where goes from 0 to 1, and there are 20 sub-
divisions, then = 0.05. The dimensionless diffusivity is 1, and the stability
criterion allows At = 0.00125 as the largest step size in dimensionless time. The
-
diffusion process is best observed around t 1, which would take 800 iterations
in time! To circumvent this problem, various techniques are used, of which the
semi-implicit form of algorithm is a common one. Here the procedure requires
that one or more terms onthe right-hand side in Eq. (19) have a term in i + 1.
Among these semi-implicit schemes, the one known as the Crank-Nicholson
technique is frequently used.
There is a more direct reason for requiring semi-implicit or stable schemes
in the case of diffusion in a solid polymer membrane, which is that the diffu-
sivity D changes by a few orders of magnitude in the range of concentration
changes. Thus, the value of At that keeps the algorithm stable in the explicit
scheme when D is moderate no longer works when D increases by a few orders
of magnitude; much lower values of At are required there. However, another
problem arises in that often in semi-implicit schemes an algebraic equation has
to be solvedor a matrix has to be inverted. If one of the terms in these becomes
very small, which will happen in the present problem, the roundoff error can
become exceedingly large. Consequently, in such problems, which are called
stiff, some independent means of checking the accuracy is desirable. Such mis-
chief does not happen in the explicit schemes, although in any systemthe errors
in the steps shown in (17)and (18) mayaddup. There is no wayof
predicting the cumulative error. Finally, stability can break down dueto bound-
ary conditions, a feature that has not been explored in the literature in detail
(Roache, 1982).
In recent yearsanother method hasbecome popular. The finite-element
method seeks an approximate solution that is exact at selected points. When
these points are brought close together, the accuracy increases. The importance
ofthismethod lies inthe fact that its programming is generalizedandnot
TRANSPORT
PHENOMENA
POLYMER
IN MEMBRANES 183
tailored to a specific problem, and it is about the only method that can handle
complex geometries:three-dimensionalandasymmetric.However, it has not
seen significant applications in mass transfer. On the other hand, orthogonal
collocation has been used extensively to seek the descriptions of many mass
transfer operations (Holland and Liapis, 1983).
Boundary
One importantclass of problems involve systems where the position of a bound-
is not known beforehand and has to be obtained as part of the solution. In
particular, the boundarycan move as a function of time that also has be
determined. For simplicity it is assumed below that the boundary lies on the
plane, that is, perpendicular to the direction being studied here. The key feature
is the jump boundary conditions at the interface, where many quantities become
discontinuous (Slattery, 1981; Miller and Neogi, 1985). These conditions are on
the total material balance,
pyv: - v ) = pyv: - v) (20)
species balance,
and energy balance,
~ " = n~ nc (n v-,v ) - P -aT'1

~1 c1 (I v , - v ) - aT1 +HR
ax ax
The superscripts I and II denote the two phases; the above conditions apply at
the interface between the two, which moves with a velocity v; and v, is the
velocity inside a phase. As constructed here, all velocities have only compo-
nents. Further, c is the concentration, T is the temperature, and c', k', and D' are
the specific heat, thermal conductivity, and diffusivity in phase I, etc. R is the
rate of generation of the species at the interface, and H is the heat liberated per
unit generation. The symbol p denotes density.
What these equations represent is the fact that total mass, mass of a species,
or energy transfers fully across an interface,providedthat the observer also
moves with the interface and provided that there is no accumulation (reactions,
adsorption,etc.) at the interface. In solids there is no convection, and the v,
terms can be ignored.
Moving boundary problems are encountered in the drying of polymer films
and have been analyzed by Shah and Porter (1973), who provide a complete
but nevertheless simplified analysis. integral heat balance is used [which
incorporates the jump heat balance of Eq. (22)], an integral balance of polymer
184 NEOGI
is used[which incorporates (21) for the polymerandthe fact that it is

nonvolatile], and the jump balancefor the solute is avoided using the results for
the polymer. Actual problems in devolatilization (Biesenberger, 1983) can in-
volve vacuum conditions and lead to nucleation (Albalak et al., 1990).
The basic scheme for solving these problems requires a well-defined initial
state. Assumingthis initial position, the firstupdating of the concentrationprofile
in time is done. One boundary condition remains, which is used together with
the new concentration profile to update the position of the interface. The issue
of numerical stability of the algorithm at the moving interface is a complex one,
and Crank (1984)provides a few details.
Some of the basic assumptions made about a medium (particularly a fluid me-
dium) break down in solid polymers. These are conditions of isotropy, homo-
geneity, and local equilibrium. Glassy polymers are not at equilibrium, but they
relax slowly toward it (Rehage and Borchard, 1973). Even crystallites in semi-
crystalline polymers are not at equilibrium (Hoffman et al., 1975). Stretching
beyond the elastic limit gives rise to anisotropy in polymers, which is not re-
stored on heating then cooling, for instance, indicating nonequilibrium config-
uration. Besides semicrystalline polymers, glassy polymers are also inhomoge-
neous although over much smaller length scales as determined from solubility
studies (Hopfenberg and Stannett, 1973). The net result is that the Boltzmann
scalling in sorption is no longer satisfied. There are two basic types ofnon-
Fickian sorption. The first of these was identified by Alfrey (1965); the mass
uptake in the glassy polymer studied was found to be proportional to time in-
stead of to the expected square foot of time. The concentration front was seen
to be very sharp, appearing to be discontinuous, and to move with a constant
velocity.Concentration-sharpened fronts ofFickian origin, showninFig. 3,
move at a speed proportional to
One can generalize these types of behavior as
4 = k"t"
-
M,
where
n>l supercase I1 (Jacques et al., 1974)
n=l case II (Alfrey, 1965)
112 n 1 anomalous (Haga, 1982)
n = 112 classicalJFickian
(Boltzmann, 1894)
n < 112 pseudo-Fickian(Liu
and
Neogi, 1991)
Only a few references have been cited here. Windle (1985) has reviewed such
data.
The second kind of non-Fickian behavior was reported by Bagley and Long
(1955) and by Long and Richman (1960). The fractional mass uptake curve was
seen to be sigmoidal but with somewhat classical scaling. Figure 4 explains the
different types of responses including onecalled a two-step response. The phys-
ical mechanisms behind the non-Fickian behavior are only now beginning to
emerge and are notcomplete. One mechanism is thatofcoupledmassand
momentum transport, coupled through swelling. The other is based on the fact
that glassy polymers are inhomogeneous. This, as a mechanism for non-Fickian
effects, has seen much less success. Finally, in all the references cited here (and
more), the polymer is glassy. Although exceptions are very few, they do exist
(Kishimoto and Matsumoto, 1964; Odani, 1968), including even polyethylene
(Rogers et al., 1959), which has a low glass transition temperature of
t
case I1
4 Thedifferentclasses of non-Fickian sorption.

186 NEOGI
A. Memory-Dependent Sorption
To explain case 11, it becomes apparent that the parabolic form of Eq. will
not be sufficient. From the characteristics reported by Alfrey (1965), which are
discontinuity in the profile and front moving at constant speed, one concludes
that the governing equation would have to be hyperbolic (Ames, 1965). The
simplest model would then be a memory-dependent diffusion, where this form
of memory is encountered in viscoelasticity (Frederickson, 1964). At short times
the governing equationis hyperbolic (corresponding to elastic response), and at
large times it is parabolic (corresponding to viscous flow). Neogi (1983b) pro-
posed that the diffusion be expressed as
ac
=- p(t - t') - t') dt'
ax
where a simple form of relaxation function would be
p(t) = DiS(t) + ~
where Di and D, are the initial and final diffusivities, S(t) is the Dirac delta
function, and T is the relaxation time. In Eq. (24) the concentration gradient at
a previous time t' also drives the flux at the present time t. Thus it is memory-
dependent. The reach of the recall is t - t', and realistically the proportionality
factor should decay with t - t'; this is called the fading memory. It is possible
to see that the relaxation time is
= tp(t)dt (26)
and that
This last result can be used to show that (24) tends to the classical form of
(2) at large times. At short times, and for D, much larger than D i , one has
the hyperbolic wave equation
which predicts adiscontinuity where the front travels with aconstantspeed

W).
Neogi (1983b) justified the model by asserting that the glassy polymer is not
at equilibrium but relaxes slowly toward it. The nonequilibrium parts and their
TRANSPORT
PHENOMENA IN POLYUER MEMBRANES 187
relaxation are modeled using classical irreversible thermodynamics, and there-

fore the thermodynamic constraints behind Eq. (24) are satisfied. One important
feature of transport phenomena is that although the entire system may not be at
equilibrium (global equilibrium), every point is at equilibrium and satisfies the
equation of state there. It is this lack of local equilibrium that is modeled using
irreversible thermodynamics. We do not know howto apply laws whenthe local
equilibrium does not apply. Consequently, it is assumed that the departure from
local equilibrium is sufficiently small that the system of equations can be lin-
earized about the local equilibrium, and there the detailed constraints are known.
The model still needs an additional equation describing relaxation, which is
generally done phenomenologically. The alternative to memory is to model
changes in the matrix as an aging process, which makes the diffusivity an ex-
plicit function of time, D(t) (in contrast, memory dependence is often called
heredity). This has also shown good results (Petropoulos, 1984; Crank, 1953).
However, we have no knowledge of any constraints on the model for the aging
phenomenon, and hence it has not been considered further.
One example of case II is shown in Fig. 5. Generally speaking, case I1 is
observed to blendinFickiandiffusionwithchanges in theconditions of the
experiments (Baird et al., 1971; Hopfenberg et al., 1969,1970).Vrentasetal.
(1975) suggest that if diffusion is coupled with elastic effects, then a Deborah
=
=
too
Time [. hrsl
5 ThedataofHopfenbergetal. (1969) forn-pentaneinbiaxiallydrawn
polystyrene. Some crazing also takes place, but Hopfenberg et al. showed later (1970)
that crazing and case I1 were not directly related. (Reproduced with permission.)
188 NEOGI
number, which is the ratio between the relaxation time and an overall time, can
be defined. They choose the relaxation time from the elastic modulus (actually,
their formula applies only to viscoelastic liquids), and the overall time scale as
L’ID, where L is the membrane thickness. At small Deborah numbers, the relax-
ation times are small, the material becomes viscous quickly, and the diffusion is
Fickian. At large Deborah numbers, the material is a frozen solid, and diffusion
is again Fickian. It is in between that case 11 is seen. In Neogi’s (1983b) model
as well, it is possible to define a Deborah number, which is d(L’/D). Case 11 is
observed in the vicinity of a Deborah number of 1 (Adib and Neogi, 1987, and
diffusion becomes Fickian as we move away from there in either direction.In the
Fickian limit, the governing equation, (14), is analogous to the equation for
viscous flow, and in the limit in which case II is predicted, one has Eq. (28),
which governs wave propagation in anelastic medium, and in that Neogi’s model
is also “viscoelastic.”
Using the linear form, many successful comparisons between theory and ex-
periments have been made (Neogi, 1983b; Adib and Neogi, 1987). These are
the sorption results of Vrentas et al. (1984a), which, though Fickian, contain
small but decaying oscillations. The data of Odani (1968) and Kishimoto and
Matsumoto (1964) on fractional mass uptakes for differentmembranethick-
nesses L do not collapse into a single curve when plotted against V t I L . These
could be quantified, as well as case I1 itself. However, the index n in Eq. (23)
inthe linear model can be shown to varybetween 112 and 1 only,andnot
beyond this range. This form also does not predict sigmoidal uptake.
One important feature of this model is that the solubility (or skin concentra-
tion) changes with time. Mehdizadeh and Durning (1990) have shown, using a
model derived earlier (Durning and Tabor, 1986) that memory-dependent dif-
fusion can also predict the two-stage sorption. The linear version of their con-
servation equation is mathematically the same as that presented here, but the
expression for the changing solubilities is quite different. Their calculations are
done at very low Deborah numbers, and thus the diffusion response is Fickian,
except at times too short to have a direct impact. However, it has a significant
effect on the solubilities. There is a quick change in solubilities over a short
period followed by a Fickian uptake; hence a two-stage response results.
It was stressed earlier that diffusion in polymers is strongly concentration-
dependent. However, Eqs. (24) and (25) are not derivable unless most of the
coefficients are assumed to be independent of concentration. A more complete
model can be made based on intuition (Jackle and Frisch, 1985) or through a
detailed analysis (Durning and Tabor, 1986). However, problems arise if such a
model is cast as p(c, t). For, when we insert into Eq. (24) the function c(x, t),
which is clearly understandable asthe concentration atpoint at time t, becomes
c(& t - t ’ ) . We know that c(x, t ’ ) is the concentration at point at a previous
time t‘, but we do not know what c(x, t - t ’ ) is. Therefore, no number can be
TRANSPORT
PHENOMENA
POLYMER
IN MEMBRANES 189
supplied for that quantity. The remedy is to use an expansion

ac a2c
-+q.-+...=-
at at2
where 7 and D can be functions of concentration. Substituting Eq. (24)into Eq.
(1) and inverting by using Laplace transform leads to Eq. (29). Camera-Roda
and Sarti (1990) solved such an equation using only the first two terms on the
left in Eq. (29). This truncation limits the solution large times.
The sorption experiments of Vrentas et al. (1984b, 1986) based on oscillatory
response become quite important in memory-dependent diffusion. Imagine that
the relaxation function in Eq. (24) has no concentration effect, an assumption
that will hold in differential sorption in all cases. Then a sinusoidal response
leads one to a Fourier transform of the relaxation function as a function of the
frequency. Thus a frequency response with small amplitude provides with a
direct means for measuring the relaxation function, where the small-amplitude
part plays the same role as in differential sorption, that is, inactivating the con-
centration dependence.
Finally, the difficulties of numerically solving equations that are nearly hy-
perbolic are enormous. The worst occurs at a Deborah number of 1. The present
methods introduce what is called artificial viscosity (Roache, 1982). Thismoves
the Deborah numbers away from 1, and obviously one needs to ascertain what
these effective Deborah numbers become during the computations.
Sorption
The first model in this class that could be comparedwith experiments wasgiven
by Kim and Neogi (1984). They followed an earlier formalism put forward by
Larch6 and Cahn (1982), in which swelling is seen as a strain, and as strain
causes stress and stress is the generalization of pressure, the chemical potential
is affected. This in turn affects the flux, which can be written as
DC a p
="- RT ax
Because the swelling at equilibrium causes no stress, the stresses generated are
transient. They occur primarily due to the condition of impenetrability; that is,
when swelling occurs the region in the membrane tries to take up space that it
is not allowed, and this act of exclusion is effected by a force or stress.
Now the stresses have to satisfy force balances, which brings into play the
shape and the boundary conditions of the overall system. This brings into the
expression for the flux not just the properties of the point but those of the
whole system. As a result, Larch6 and Cahn (1982) point out, their flux ex-
pression is not local anymore, which goes againstthe usual assumptionthat
190 NEOGI
fluxes can be described by the properties at a point (Truesdell and Noll, 1965).
Carbonell and Sarti (1990) pointed out that some models, such as that of Dum-
ing and Tabor (1986), do not satisfy a force balance. Whereas this is desirable,
it should also be noted that if the local equilibrium is not satisfied, forces do
not always have to balance; thus the practical problem ofhow to define, quantify,
or measure “internal stresses” that occur inglassypolymersremainsunre-
solved. However, if the model is being defined to the point that even the con-
stitutive equations are being specified, then forces ought to be balanced to be
at least consistent.
The approximate solution of Kim and Neogi (1984) to the sorption problem
shows that the diffusion remains Fickian but the solubility changes with time.
The chemical potential is made up of the concentration effect and the strain
energies and is expressed as
= - qv1 tr(4 (31)
where U is the stress, tr denotes trace, v1 is the specific volume of the solute,
and q is the change in the volume due to concentration changes. The chemical
potential in the absence of stresses is As the strain energy changes with
time, the skin concentration adjusts accordingly such that the overall chemical
potential at the surface remains constant and equal to the value of that in the
reservoir during sorption. This changing skin concentration has been measured
experimentally (Bagley and Long, 1955; Long and Richman, 1960). At large
elastic modulus of the material, it appears that when the diffusion process is
long over,the solubility can still continueto increase and the concentration
profile becomes spatially constant. Such profiles have been measured for water
vapor diffusing in epoxy (Wolf, personal communication, 1987). In general, all
non-Fickian effects vanish at low concentrations in the swelling model. Neogi
et al. (1986) later solved numerically a more detailed model, which showed that
the uptake was sigmoidal. This allowed them to quantify the data of Bagley and
Long (1955) and Long and Richman (1960), as showninFig. 6. They also
calculated that large compressive stresses would arise, sometimes large enough
to give rise to plastic deformation. Downes and MacKay (1958) observed that
in their experiments a Fickian responseoccurred when the step changes in sorp-
tion were small. These were also reproducible. When large step changes were
given, sigmoidal uptakes were observed,but once they occurred the older Fick-
ian responses couldnot be reproduced.Presumably, plastic deformation had
changed the matrix. In a similar vein, Tamura et al. (1963) correlated sigmoidal
uptakes to yield stresses.
In semicrystalline polymers, a polymer molecule can run through more than
one crystallite, and the effect is that the crystals are chained to one another. This
prevents the amorphous regions from swelling, which gives rise to forces that
pull at the anchors. In extreme cases the crystallites unravel, which constitutes
TRANSPORT
PHENOMENA
IN POLWER MEMBRANES 191
0.75
0.50
0.25
0.2 0.5 0.6 0.7 0.8 0.9
Thetheoreticalresults of Neogi et al. onswelling-inducedeffects

fitted to the data of Long and Richman showing how the skin concentration and
mass uptake change with time. The system is methyl iodide in cellulose acetate. Here S
is a dimensionless elastic modulus and is dimensionless time. effective diffusivity
has been used insteadof a concentration-dependent one. [Reproduced from Neogi
with permission of the American Institute of Chemical Engineers. AIChE. All
rights reserved.]
192 NEOGI
ductile failure in these materials. Rogers reported such effects in poly-

ethylene where the amorphous regions are rubbery, and Durning and Rebenfeld
reported similar effects in poly(ethy1ene terephthalate) where the amor-
phous regions are glassy.
It becomes evidentthat more detailed stress-strain constitutive equations can
be worked into the theory, and the theory itself can be given a broader basis
thanthat given by Larch6andCahn Lustig et al. provided a
generalized framework for diffusion including relaxation and stress effects to
which all constitutive equations must conform.
C. Heterogeneities
One key feature observed by Neogi et al. and one that can be generalized
to other systems, is that the effects of swelling die out as the concentrations are
lowered. That is probably the reason that in the work by Downes and MacKay
Fickian diffusion was observed at lower concentrations, and sigmoidal
at high concentrations. However, more detailed differential sorption data given
by Odani et al. andKishimoto et al. show thatroughly with
decreasing concentrations, sigmoidal (swelling plus elastic effects), then Fickian
(disappearance of swelling effects?),two-step(swelling plus viscoelastic ef-
fects?), Fickian (disappearance of swelling at low concentrations), and finally a
sigmoidal response at very low concentrations are observed. Although there is
some confusion as to what happens in the intermediate ranges, there is never-
theless some information. In contrast, no mechanism has been described far
that can explain the sigmoidal behavior at the very lowest concentrations.
At suchlow concentrations one peculiarity of glassy polymersis summarized
by its dual-mode solubility isotherm
where the first term on the right represents dissolution in the solid polymer and
the second term is due to the adsorption on the walls of the microvoids (Hop-
fenberg and Stannet, That is, the solid matrix has small holes, called
microvoids, and the adsorption is described by a Langmuir adsorption isotherm.
Very detailed experimental measurements are now available in support of this
solubility model (Chan et al., Patton et al.,
The question now is, do these holes have an effect on transport? Michaels et
al. found that if adsorption equilibrium was assumed to exist at the
interfaces, then the holes had no effect on fluxes. Of course, it was also assumed
that the diffusivity in the holes was much greater thanthatin the polymer.
Petropolous suggested that holes would contribute to the flux whatever
the detailed mechanisms were, and therefore the flux would reduce to
where D , is the diffusivity in the polymer andDHis that of the adsorbedspecies,

and c, and C , are concentrations there. D,c,,/RT and DHcHIRT are the two mo-
bilities, and the last term is the driving force, the same for both. Paul and Koros
showed how to combine Eqs. and (32) to get a complete expression
for flux. Subramanian et al. used that expression to comparethe experi-
mental data withtheory to get a value for DH, which was found tobe less
than D,.
In another study,TshudyandvonFrankenberg used the holes and
adsorption as a sourcelsink term in the conservation equation. The adsorption-
desorption kinetics are also included. The model was notlikedbecausethe
contribution of the holes to the flux was made even more obscure.
Vieth gave a very detailed account of the use made of the above
models to interpret experimental data on glassy polymers. No absurdities are
encountered, but neither are any insights gained into the nature of mobility of
the diffusantin the microvoids. Barrer and Fredrickson andHelfand
put together molecular models to justify and generalize a flux expression
such as that given in (32). Their results, however, leave one with no clear
connection with the more easily understood features of a continuum approach.
Further,when Sada et al. compareone set of experiments with this
generalized model, they find thatthe direct effect of the microvoids on transport
is negligible and that an indirect effect in terms of a coupling between holes
and the surroundings dominates: DHin Eq. (33) is negligible, and a cross term,
not given there, is important! The observation appears at present to be general,
and the progress at this point (in view of Petropoulos’s original premise) begins
to resemble a snake swallowing its tail.
Some aspects of the presence of holes on transport phenomena were ignored
until the work of Sangani namely, the matters of space invoked in laying
out the holes. Both key observations that result when this is recognized-that
near a hole the concentration profiles change drastically and that holes are ran-
domlyplaced-areaccounted for in transport phenomenathroughaveraging
methods. Obviously, in a randomly heterogeneous systemthe detailed profile is
complicated and can never be measured. Instead, one measures some kind of
average response. We know only the detailed equations, and the averaging pro-
cesses lead to the equations for the average responses. Sangani attempted to
average the dual sorption model together with Petropoulos’smodel for the flux.
The method used is valid over a large range of volume fractions of the holes,
which is a very difficult problem to solve. The final expression, though incom-
plete, does show that quite a few changes occur on averaging.
194 NEOGI
Neogi attemptedtoformulate theproblem using a more detailed

model. There are four resistances to mass transfer to a hole: diffusion in the
polymer,adsorption-desorption resistance the
inpolymer,
adsorption-
desorption resistance at the interface on the side of the hole, and diffusion in
the hole. The adsorbed layer can move by surface diffusion as well. (Thus, this
approach is a more detailed version of the model of Tshudy and von Franken-
berg.) The averaging technique is local volume averaging(Hinch, which
was originallyconstructed by Maxwelland applies to systemswith a small
volume fraction of holes. l k o kinds of results can be obtained. In the first, the
constitutive equation is averaged to yield an effective diffusivity. It shows that
to a good approximation the diffusion is Fickian even at unsteady state. This
formulation is what one needs to describe steady-state permeability exactly. In
the second, the conservation equation is averaged, which is needed to describe
the total solute in a polymer,that is, the properdescriptionfor the sorption
experiments. The results show sigmoidal sorption. The main parameter there is
theratio of the rate of adsorption to the rate of diffusion. The comparison
between the theory and the experiments is shown in Fig. 7.
This model also uses a linear adsorption relation. If a nonlinear relation is
used that has a saturation (the present state of averaging principles does not
allow this), such as a Langmuir adsorption isotherm, thenthis non-Fickian effect
will disappear atsufficientlyhigh concentrations whensaturation is reached.
The anomalous effect is due to hole filling and disappears when the holes
are filled. All of this occurs atlowconcentrations.In contrast, the swelling
phenomenon takes place at high concentrations and disappears at low
concentrations.
When the holes are very small, it becomes difficult to define adsorption. A
model for this case, by Kasargod et al. predicts pseudo-Fickian sorption
(see Equation
A very different kind ofaveraging wasintroduced by Di Marzio and Sanchez
In many mechanisms one can have a forward step and a backward step,
each with an activation energy. In a heterogeneous system, there will be a dis-
tribution of such energies, which can be used to obtain an average rate process.
In dielectric relaxation, Di Marzio and Sanchez were able to show that
the simple exponentialrelaxation becomes “stretched exponential”exp[ - ( t / ~ ) ’ ]
on averaging. These ideas were used by Adib and Neogi to average the
relaxation discussed earlier. Whereas case I1 itself was always exhibited, a great
many features were seen in the calculated fractional mass uptake curves at large
times. Such features (some of which can get to be a little exotic) have been
documented in experiments. Comparison with experiments indicates that more
often than not the relaxation time as defined in Eq. cannot exist, that is,
will give negative or infinite values!
TRANSPORT
PHENOMENA
7 The theoretical results of Neogi on hole-filling effects at very small

concentrations as fitted to the experimental data of Odani et al. and Kishimoto et
al. The systems are benzene in atactic and isotactic (2) polystyrene at
Here is dimensionless time. (Reprinted with the permission of John Wiley and Sons,
Inc.)
Whereas there has been a concerted effort to bring together memory-

dependent diffusion with the effects of swelling on diffusion, there has been
relatively little work on incorporating the effects of the heterogeneities that un-
doubtedly exist in these systems. Some beginnings have been made.
W.
Phase changes take place during mass transfer operations in a variety of situa-
tions. The production or collapse of polymer foams, devolatilization, etc., offer
instances in polymer processing where vaporization and nucleation of the vapor
phase are of considerable importance (Han and Han, 1990; Albalak et al., 1990).
Diffusion of gases through a lamellar composite, where one material has a high
NEOGZ
permeability and the other has a low permeability, can give rise to local super-
saturation and nucleation (Graves et al., 1973). However, the most often studied
phenomenon is precipitation as it applies to membrane formation. A quite dif-
ferent class of phase transition is polymer crystallization, which can also be
promoted by a solute, diffusionprocess, etc. Only these two are discussed below.
A. ReverseOsmosis Membranes
Loeb and Sourirajan (1963)made the first reverse osmosis (RO) membrane
capable of withstanding the rigors of industrial use. The membrane was made
of cellulose acetate (about 22% polymer) in a solution of acetone (68%) and
water (10%). Water is a nonsolvent in this case, and acetone is a good solvent.
inorganic salt such asmagnesium perchlorate was also added. The procedure
used was as follows. A thin film of about 250 pm thickness was layered onto
a surface kept at -10 to 0°Cand after a few minutes was plunged into hot
waterandkept there for over an hour. The surface that was exposedtothe
atmosphere developed a dense skin -0.1 pm thick (Riley et al., 1964, 1966)
and contained pores as small as 0.3-5 nm (Kotoh and Suzuki, 1981). The rest
of the membrane was a spongy mass and formed the backing, whereas the dense
skin was needed for desalination (Yasuda and Lamaze, 1970).
In water,a chargedevelopson the surface of the solid polymer,and an
electrostatic field results that extends into the aqueous phase. However, such a
field is confined to a very thin layer next to the wall, and only if the pores are
very fine, such as in the dense film, will the electrostatic field cover the entire
pore space. The membrane itself can separate a reservoir with some salt (say
NaC1) from another with no salt. Now if these pores are negatively charged,
thenthe “solubility” (andhence the permeability) of the Na.’ioninsucha
pore will be veryhighandthatof the Cl- ion will be verylow.But in the
absence of an externally applied voltage difference there will be no current, and
the fluxes ofthe two ions will be equal. The overall effect is that there is virtually
no net flow of NaCl across the membrane (Cl- ions holding back the Na’ ions)
even though there can be a pressure-induced flow of water from the salt side to
the pure water side. This is the phenomenon of reverse osmosis (Jacazio et al.,
1972), which is also known as hyperfiltration.
It is also possible to do regular filtration, but with particles that are very fine,
even molecular in dimension, such as proteins (Michaels, 1968). This is called
ultrafiltration. There are many books and reviews on the subject of R 0 mem-
branes that can be consultedfor additional details (Sourirajan, 1970, 1977;Mer-
ten, 1966; TurbackJ981; Lonsdale and Podall, 1972; Bungay et al., 1986; Prob-
stein, 1989).
The key feature of interest here is the way in which the mass transfer gives
rise to the structure of the R 0 membranes. Kesting (1971) discussed qualita-
TRANSPORT
PHENOMENA
POLYMER
IN MEMBRANES 197
tively the nature of precipitation/aggregation in the evaporating solution. The

most important partof his discussion concerns the relationship between swelling
and the structure of the precipitate. Acetone, being a goodsolvent for cellulose
acetate, swells it; water, being a nonsolvent, decreases the swelling. Inorganic
salts also increase the swelling. Now, the precipitation from a good solvent takes
place rather late, that is, at a higher degree of supersaturation, and the precipitate
has large irregular pores. The precipitate from a poor solvent has very small,
well-marked holes. Thus, it can be concluded that the acetone-water-salt mix
was arrived at as one where the resulting solvent gave the right pore structure.
Nothing was suggested regardingthe origin of the asymmetry. In fact, this ques-
tion was not studied until Sirkar et al. (1978) showed experimentally thatit took
less than 0.01 S of evaporation time for the skin to form. Neogi (1983a) gave
different mechanisms for the formation of the skin and the backing. At very
short times during the evaporation step the fluxes are very high, which gives
rise to a sharp changein the densities,which in turn causes interfacial instability.
The shape of the interface becomes bumpy, and some of the pits develop into
pores. This chain of events was shown to take place virtually instantaneously.
The slower mechanism of backing formation was ascribed to the growth of
nuclei. Some effort was made to connect this step to swelling.
Ray et al. (1985) also gave a mechanism based on interfacial instability and
density effects. This is confined to the effect of density on colloidal forces and
that of the latter on flow. The effects of density variation on other phenomena
are ignored. Tsay and McHugh (1991) provided detailed calculations that show
that the skin formation is over in at most 20 S.
B. Solvent-InducedCrystallization
Polymers with simple molecular structure, such as polyethylene, never fail to
crystallize when cooled from melt. In contrast, atactic polystyrene is ungainly
enough that packing problems are enormous, and it never crystallizes. This is,
of course, an observation within the cooling rates that are currently available.
Poly(ethy1ene terephthalate) (PET), on the other hand, can be semicrystalline or
fully amorphous when it is quenched from melt.
One key observation is that when a crystallizable, but presently amorphous,
polymer is exposed to solute that forms a solvent for the polymer, crystallization
sets in. This phenomenon is called solvent-induced crystallization (SINC).
The common polymeric crystal is a lamella. On the edge of the lamella, a
polymer molecule attaches itself and weaves up and down onto the face of the
lamella. The rate of growth of the sheet is usually denoted by the symbol G. In
turn, lamellae are gathered into sheaves, which are pinched in the middle to
form spindles. Finally, the empty parts of the spindle are filled to form a spher-
ulite. The outer surface of the spherulite is made up the edges of the lamellae,
198 NEOGI
and hence if r is the radius of the spherulite, then

dr
-=G (34)
dt
The term G can be expressed as
G = vzGoexp(- AEDIRT) exp(-AF*/RT) (35)
where v2 is the volume fraction of the polymer, Go is a rate constant, AEDis the
transport activation energy (that needed to bring the polymer to the growing
interface), and AF* is the free energy for the attachment of the next polymer
chain to the growing edge. The first activation energy is identified as the WLF
energy
ED = C,T/(C2 -I- T - Tg) (36)
where C, and C, are universal constants. The other activation energy, AF*,is a
more complicated function of different constants of the system plus the melting
point T,. The nature of crystals and that of their growth rates are covered in
more detail by Hoffman et al. (1975). Avrami (1939, 1940) showed how the
rate of growth G can be related to the conversion into crystallites with time:
X = 1 -exp(-kt") (37)
where n = 3 is a typical value.
In SINC it is seen that as the concentration of the solute advances, crystals
are formed. If the front advances by a diffusion-controlled mechanism, its rate
is seen to be proportional to <t. The first modeling effort by Zachrnann and
Konrad (1968) identified that it was necessary to combine crystallization growth
with diffusion. Their methods of calculation were revisited by Makarewicz and
Wilkes (1978), who among other things came to an important conclusion re-
garding the mechanism behind SINC: The presence of the solute depresses the
glass transition temperature Tg, which increases the value of G in Eq. (35) by
orders of magnitude. The plasticizing effect can be quantified using the Kelly-
Bueche equation. The proper transport phenomena problem was solved by Durn-
ing and Russell (1985), who solved separate equations for the conservation of
solute and the amorphous polymer. Generally there are two limiting cases. In
one the diffusion is rate-controlling and crystallization is very rapid; in this case
the formation of crystals will increase as V?. At the other extreme, when the
diffusion is rapid and crystallization is slow, the crystallization rate would be
given by an Avrami-type equation. The solution by Durning and Russell shows
that a great many things can happen in between. One interesting case is when
diffusion is rapid and the membrane in a sorption experiment equilibrates first.
Then Crystallization starts, but as the crystallites are impermeable to the solute,
some solute is driven off the membrane. That is, the mass uptake passes through
TRANSPORT PHENOMENA IN POLYMER MEMBRANES 199
a maximum. The comparison between Durning-Russell theory and experiment

is shown in Fig. 8.
A note of discord has been struck by Mandelkern (1955) and Yoon and Flory
(1979). They contend that hE, should be the activation energy of the self-
diffusion coefficient of the polymer, and not the WLF energy. It is likely that
the main issues raised here will remain intact although the quantification may
see some changes in the future.
V. MULTIPHASE, MULTICOMPONENT, AND

INHOMOGENEOUS SYSTEMS
In this section some systems with more than one component or phase are re-
viewed. A notable exception is the case of coupled fluxes. None have been
detected, nor has there been any attempt to search seriously for such effects.
The antisymmetric permeability in a membrane with a gradient in the polymer
composition has been discussed briefly by Cussler (1976) and will not be con-
sidered here. A few of the more active areas are described below.
A. Chromatography
There are endless types of chromatographies, and the only ones of concern at
present are those that lead to the diffusion in polymers of the type considered
here. The simplest is gas-solid chromatography, which employs a column con-
taining a solid packing material. An inert gas, such as N2, is the carrier and
flows through the packed bed at a steady rate. The sample/species under con-
sideration is introduced as a pulse at the inlet and would normally exit the
column at elution time L/(v), where L is the length of the column and (v) is the
average velocity of the carrier gas.
ac ac
- + (v) - = - j*
at ax
The species under consideration is at very dilute concentrations. However, the

material adsorbs on the surface of the packing. In Eq. (38), the mass transferred
from the gas phase to the adsorbed phase is given by j * . The axial dispersion
has been neglected. In the adsorbed phase the conservation equation is
ar
-= j* (39)
at
where r is the surface concentration. If local equilibrium holds, then I' = Kc,
where K is a Henry's law constant. Substituting into Eq. (39) and summing Eqs.
200 NEOGI
c/?.
Ap x 10- (s''Z/m)
8 The theoretical observations of Durning and Russel (1985) compared to the
data of Overbergh et al. (1975) for methylene chloride diffusing in isotactic polystyrene
films at 30°C and different film thicknesses. Effective diffusivities were used. (Repro-
duced with permission.)
(38) and one
The solution, when the input is a pulse, is

TRANSPORT PHENOMENA IN POLYMER MEMBRANES 201
where 6 is the Dirac delta function and M is the total mass in the pulse. The
elution time at = L is L(l + K)/(v)(Littlewood, 1970). As K differs from
species to species, the separation of species is made possible. More detailed
calculations are discussed by Holland and Liapis (1983).
In gas-liquidchromatography (GLC), a viscous liquid is used to coatthe
solid packing. If the layer is very thin, the above concepts apply.
. For separation and identification, a mixture is injected at the inlet. Inside the
chromatograph the species are separated and elute at the outlet at different times
depending on their K values. The they take is that of a sharp pulse. How-
ever, in real systems, axial dispersion plays the role of smearing out the pulse
into a bell-shaped curve. If these curves overlapin the eluant, then identification
(by their retention times) becomes difficult, and quantitative analysis (area under
the curve) becomes impossible.
Golay (1958a,b) developed intuitively many features of the dispersion coef-
ficient D, without knowledge of Taylor's (1953) work inthis area: that it caused
widening of the peak, that it was inversely proportional to the molecular dif-
fusivity, and, most important, that it decreased strongly with decreasing tube
radius.Golayarguedthat one should perform chromatographyin a capillary
column with the walls coated with the absorbent. The methods for fabrication,
experimental verification of sharpening of the peaks, and a simple model were
given. Capillary chromatographs are now available commercially and give the
best resolutions. The investigations in the period following this were reviewed
by Vieth (1991).
Pawlisch and Lawrence (1987, 1988) took this up and provided a more thor-
ough analysis. Much improved equations were used, and capillary chromatog-
raphy was shown to be an excellent way of obtaining the diffusivities at infinite
dilution in the absorbent, a polymer in their case. A comparison between their
theoretical calculations and experimentally obtained peakshape is shown in Fig.
9.ArnouldandLawrence(1992)reported the vast amounts of data on such
diffusivities thathavebeen thus obtained anddiscussed their impact on the
current theories of diffusivities.
We are entering a yet newer era in the use capillary chromatography. As
the layer of polymer is thin, the orientation of the top layer affects the response
of the polymer, and the orientation of this layer itself is affected by the fluid on
top of it. important features emerge, according to Andrade (1988). These
are that the surface effects (interfacial phenomena) can be made to govern the
overall dynamics and that these vary with the environment.
The problemin transport phenomena that polyblends give rise to is that the
medium becomes inhomogeneous in a discontinuous manner. Well-defined do-
202 NEOGI
I l I
0 1 2
9 Comparison between the theoretical (a) and experimental (b) elution curves.
The polymer is polystyrene, and the vapor is benzene; the system at
is 130°C. [Reprinted
with permission from Pawlisch and Lawrence (1988). Copyright 1988 American Oem-
ical Society.]
mains of the two phases are evident when they are examined locally, but they
are distributed in a random fashion. Some methods of analyzing such media,
particularly when one the phases is dilute, were discussed briefly in Section
1II.C These originate from the work of Maxwell and have been used by Robeson
et al. (1973) to compare permeation data with theoretical predictions. The prob-
lem where the two phases are of comparable amounts but one phase is clearly
dispersed in another was reviewed by Barrer (1968). The dispersed phase is
given a geometric shape such as a cube. The entire system is divided into cells,
and each cell has a cube located right in the center.
As far as polyblends are concerned, all the above models ignore two key
features. The first is that they all require some degree of geometric niceties.
However, the shapes and distribution of the phases are actually random, and it
is even difficult to say which phase is continuous. In fact, in polyblends, a good
blend is described as having an “interprenetrating network.” That is, one phase
TRANSPORT
PHENOMENA
IN POLYUER MEMBRANES 203
Randomly cut polygons (Voronoi tessellations) distributed in phases.

[Reproduced with permission from Kuan et al.
passes through another many times in an intimate way. Thus, this brings up the
second point, thatif one hastwo different samples, bothhaving the same volume
fractions of the two phases, their performances will be different because the
detailed layout of the two phases will be different. Consequently, only a dis-
cussion of the average response is meaningful.
Rogers discussed some attempts at quantifying diffusion in such sys-
tems. The new methods of analyzing such systems were usedby Ottino and
coworkers (Sax and Ottino, Ottino and Shah, Shah et al.,
Without going into detail, some basic ideas of random systems are analyzed
below. In Fig. 10 is shown a two-dimensional system made out of two phases
marked with white and black in the figure. Their domains are random: the sys-
tem was cut up into random polygons, each of which was assigned randomly
to be black or white. When the fraction of the black phase is low it becomes
the dispersed phase, and at highfractions it is the continuous phase. In between,
both phases are continuous, that is, the system becomesbicontinuous. The point
of inversion is the percolation threshold. The question now is, what will the
overall diffusivity be if the two phases have two different diffusivities? This is
the quantity that is calculated in a variety of ways. The above references carry
the mathematical details.
The phenomenon itself has many applications. Spherical polymer particles
are coated with silver and compressed into a film. Since the silver on the surface
forms a continuous phase (percolates), the film has a very high conductivity
comparable to that of pure silver, even though the amount of silver can be as
low as v01 % [according to one model (Park and MacElroy, If we
have a pure silver film and we carve out spheres and replace them with polymer
randomly, and the spheres can overlap, we can keep doing this until of
the silver has been removed. After that thesilver stops being a continuous phase
and the conductivity falls dramatically. (In this connection it should be noted
204 NEOGI
that in two dimensions one could go down to l%.) The reverse case has been
achieved by putting mica flakes in a film. The flakes tend to orient parallel to
the film,and no diffusant passes through the flakes.If the flakestouch one
another at theedges,thena continuous impermeable surface is formed that
prevents any diffusant from passing through, that is, a perfect insulation to per-
meation is formed. Actuallythere is no guarantee that the flakes will touch, and
hence this system is not perfect.
Finally, we note that these methods of calculating averages are confined to
linear systems and constant conductivities. They also ignore the possibility of
the morphology itself affecting the diffusivity. We know that it could do even
in a single-component system that is semicrystalline through the free-volume
effect; this has been shown to be the case in polyethylene (Liu and Neogi,1988).
C. NMR Self-DiffusionCoefficients
The measurement of NMR self-diffusion coefficients in polymer systems now
is fairly routine [the pulsed gradient spin echo (PSGE) technique in particular],
and yet there are still some uncertainties regarding the nature and utility of the
quantities that are being measured. The followingis a brief review of how these
measurements are made and an attempt to explain them based on the d f i s i o n
coefficients encountered here.
Some nuclei have magnetic dipoles: 'H, 'H, I3C, etc. They orient in a mag-
netic field and absorb energy from an rf (radio-frequency) souce provided that
the frequency correspondsto the characteristic Larmor frequency of the nucleus.
That is, with the rf burst it is possible to affect a particular species, whereas in
a steady field there is no selectivity of that kind.
Very briefly, a steady field is applied, after which an rf burst is sed to flip
the magnetization by 90". A brief gap follows, after which it is flipped by an
additional 180"by a second rf burst. This last maneuver is equivalent to flipping
the steady magnetic field by 180". The periodoverwhichafirst rf burst is
applied is called the dephasing, and the second rf burst constitutes refocusing.
Essentially, dephasing induces a phase lag in the spinning magnetic dipole and
refocusing restores it, hence the "echo." However, these shifts are also depen-
dent on the steady field strengths. If the steady field strength varies spatially
and the nucleus wanders (due to diffusion) in between the application of the
radio frequencies, then the echo strength will be attenuated. It has been
possible to relate the attenuation diffusion. The reviews by von Meerwall
(1985), Stilbs (1986), and Urger et al. (1988) cover experimentaland theoretical
aspects quite extensively. The one by von Meerwall is exclusively on polymers.
In polymer solutions it has been shown that the NMR self-diffusion coeffi-
cient of the solute measures the self-diffusion coefficient, as given by Vrentas
and Duda (1979), excellently (Pickup and Blum, 1989). Blum et al. (1990) also
TRANSPORT
PHENOMENA
POLYMER
IN MEMBRANES 205
showed that in their system the NMR self-diffusion coefficient was the same as
the mutual diffusion coefficient measured by Duda et al. (1979), only at infinite
dilution of the solute. For the other case of diffusion of polymer molecules,
Gibbs et al. (1991) showed that for ovalbumin, a globular protein, the N M R
self-diffusion coefficient was the same as the mutual diffusion coefficient mea-
sured with an ultracentrifuge. The last observation is exciting but needs more
investigation.
VI. CONCLUSIONS
Without doubt this area of physicochemical effects involving diffusion in poly-
mers and their quantification has been a rich one. Particular emphasis should be
placed on quantification. In the area of non-Fickian diffusion, one would have
no direction or even a coherent way of looking atthephenomenon without
models. Other than rationalizing existing data, one could also find new directions
to take for making meaningful inventions, as seen with polyblends. In general,
there can be very little meaning to investigating the dynamics of diffusion or
its applications without including transport models and their solutions. This has
proven to be the key feature in the past and will continue to be in the future.
REFERENCES
Adib, F., and P. Neogi (1987).AfChE J., 33, 164.
Adib, F., and P. Neogi (1993). J. Phys. ZI, 3, 1109.
Albalak, R. J., Z. Tadmor, and Y. Talmon (1990). MChEJ., 36, 1313.
Alfrey, T. (1965). Chem. Eng. News 43(41),
Ames, W. F. (1965). Nonlinear Partial Differential Equations in Engineering,Academic,
New York.
Andrade, J. D., Ed. (1988). Polymer Surface Dynamics, Plenum, New York.
Amould, D., and R. L. Lawrence (1992). ZEC Res., 31, 218.
Avrami, M. (1939). J. Chem. Phys., 7, 1103.
Bagley, E., and F. Long (1955). J. Am. Chem. Soc., 77, 2172.
Baird, B. R., H. B. Hopfenberg, and V. T. Stannett (1971). Polym. Eng. Sci., 274.
Barrer, R.M. (1968). In DifFcsion in Polymers, J. Crank and G. S. Park, Eds., Academic,
New York, p. 165.
Barrer, R. M.(1984). J. Membrane Sci., 18, 25.
Biesenberger, J. A. (1983). Devolatilization of Polymers, Hanser, Munich.
Bird, R. B., W. E. Stewart,andE. N. Lightfoot (1960). Transport Phenomena, Wiley,
New York, p. 557.
Blum, F. D., S. Pickup, and R. A. Waggoner (1990). Polym. Prepr., 30, 125.
Boltzmann, L. (1894). Ann. Phys. Chem. 53, 959.
206 NEOGI
Bungay, P. M., H. K. Lonsdale, and M. N. de Pinto, Eds. Synthetic Membranes,

NATO AS1 Series, Reidel, Boston.
Camera-Roda, G., and G. C. Sarti AIChE J., 36,
Carbonell, R. C., and G. C. Sarti ZEC Res., 29,
Carnahan, B., H. Luther, and J. Wilkes Applied Numerical Methods, Wiley,
New York, p.
Carslaw, H. S., and J. C. Jaeger Conduction ofHeat in Solids, 2nd ed., Clarendon
Press, Oxford.
Chan, A. H., W. J. Koros, and D. R. Paul J. Membrane Sci., 3,
Costello, L. M.,and W. J. Koros Z&EC Res., 31,
Crank, J. J. Polym. Sci., 11,
Crank, J. The Mathematics of Diffusion, 2nd ed., Clarendon Press, Oxford.
Crank, J. Free and Moving BoundaryProblems, Oxford/Clarendon Press, Oxford.
Crank, J., and G. S. Park In Diffusion in Polymers, J. Crank and G. S. Park,
Eds., Academic, New York, p.
Cussler, E. L. Multicomponent Diffusion, Elsevier, New York, p.
Di Marzio, E. A.,and I. C. Sanchez In Transportand Relaxation inRandom
Materials, J.Klafter,R. J. Rubin,and M. F. Shlesinger,Eds.,WorldScientific,
Singapore, p.
Downes, J. G., and B. H. MacKay J. Polym. Sci., 28,
Duda, J. L., and J. S. Vrentas J. Phys. Chem 69,
Duda, J. L., Y.C. Ni, and J. S. Vrentas J. Appl. Polym. Sci., 23,
Durning, C. J., and L. Rebenfeld J. Appl. Polym. Sci., 29,
Durning, C. J., and W. B. Russel Polymer, 26,
Durning, C. J., and M. Tabor Macromolecules, 19,
Frederickson,A. E. Principles and Applications of Rheology, Prentice-Hall,
Englewood Cliffs, New Jersey.
Fredrickson, G. H., and E. Helfand Macromolecules, 18,
Fujita, H. In D i m i o n in Polymers, J. Crank and G. S. Park, Eds., Academic,
New York, p.
Gibbs, S. J., A. A. Chu, E. N. Lightfoot, and T.W. Root J. Phys. Chem.,
Golay, M.J. E. In Gas Chromatography, V. J. Coates, H. J. Noebels, andI. S.

Fagerson, Eds., Academic, New York, p.
Golay, M. J. E. In Gas Chromatography 1958, D. H. Desty, Ed., Academic,
New York, p.
Graves, D. J., J. Idicula, C. J. Lambertsen, and J. A. Quinn Science, 179,
Grayson, M. A., P. S. Pao, and C. J. Wolf J. Polym. Sci. Polym. Phys. Ed., 25,
Haga, T. J. Appl. Polym Sci., 27,

Han, J. H., and C. D. Han J. Polym. Sci. Polym. Phys. Ed., 28,
Hinch, E. J. J. Fluid Mech., 83,
Hoffman, J. D., G. T. Davis, and J. I. Lauritzen, Jr. In Treatise on Solid State
Chemistry, Vol. N. B. Hannay, ed., Plenum, New York.
Holland, C. D., and A. I. Liapis Computer Methods for Solving Dynamic Sepa-
ration Problems, McGraw-Hill, New York.
TRANSPORT PHENOMENA
Hopfenberg, H. B., and V. Stannett In The Physics of Glassy Polymers, R. N.

Haward, Ed., Applied Science, London.
Hopfenberg, H. B., R. H. Holley, and V. T. Stannett Polym. Eng. Sci., 9,
Hopfenberg, H. B., R. H. Holley, and V.T. Stannett Polym. Eng. Sci., 10,
Jacazio, G., R. F. Probstein, A. A. Sonin, and D. Yung J. Phys. Chem.,
Jackle, J., and H. L. Frisch J. Polym. Sci Polym. Phys. Ed.,
Jacques,C.H.M.,H. B. Hopfenberg,and V. T. Stannett In Permeability of
Plastic Films and Coatings to Vapors and Liquids,H. B. Hopfenberg, Ed., Plenum,
New York.
Kasargod, S. S., F. Adib, and P. Neogi J. Chem Phys., 103,
Kesting, R. N. In Cellulose and Cellulose Derivatives, Vol. N. M. Bikales and
L. Segal, Eds., Wiley-Interscience, New York, p.
Kim, M., and P. Neogi J. Appl. Polym. Sci., 29,
Kishimoto, A., and K. Matsumoto J. Polym. Sci., A-2,
Kishimoto, A., H. Fujita, H. Odani, M. Kutara, and M. Tamura J. Phys. Chem.,
Korsmeyer, R. W.,E.vonMuerwall,and N. A. Peppas J. Polym. Sci. Polym.

Phys. Ed.,
Kotoh, M., and S. Suzuki In Synthetic Membranes, Vol. A. F. Turback, Ed.,
ACS Symp. Ser. ACS, Washington, DC, p.
Krager, J., H. Pfeifer, and W. Heink Adv. Magnetic Resonance, 12,
Kuan, D.-Y., R. and H. T.Davis Chem. Eng. Sci., 38,
Kulkami, S. S., and Stem, S. A. J. Polym. Sci. Polym. Phys. Ed., 21,
LarchC, C. F., and J. W.Cahn Acta Metall., 30,
Littlewood, A. B. Gas Chromatography, 2nd ed., Academic, New York.
Liu, C.-P. A., and P. Neogi J. Membrane Sci., 35,
Liu, C.-P. A., and P. Neogi J. Macromol. Sci. Phys., B31,
Loeb, S., and S. Sourirajan Adv. Chem. Ses, 38,
Long, F. A., and D. Richman J. Am. Chem. Soc., 82,
Lonsdale, H. K., and H. E. Podall, Eds. Reverse Osmosis Membrane Research,
Plenum, New York.
Lustig, S. R., J. M. Caruthers, and N. A. Peppas Chem. Eng. Sci.,
Makarewicz, P. J., and G . L.Wilkes J. Polym. Sci. Polym.Phys.Ed., 16,
Mandelkem, L. J. Appl. Phys.,

Mehdizadeh, C., and C. J. Duming AIChE J., 36,
Merten, U.,Ed. Desalination by Reverse Osmosis, MlT Press, Cambridge, MA.
Michaels, A. S. Chem. Eng. Prog.,
Michaels, A., W. Vieth, and J. Barie J. Appl. Phys., 34,
Miller, C. A., and P. Neogi Interfacial Phenomena, Marcel Dekker, New York.
Neogi, P. AIChE J., 29,
Neogi, P. AIChE J., 29,
Neogi, P. Chem. Eng. Commun., 68,
Neogi, P. J. Polym. Sci. Polym. Phys. Ed., 31,
Neogi, P., M. Kim, and Y. Yang AIChE J., 32,
208 NEOGZ
Odani, G. S. J. Polym. Sci., A-2,

Odani, H., S. Kida, and M. Tamura Bull. Chem. Soc. Jpn., 39,
Ottino, J. M., and N. Shah Polym. Eng. Sci.,
Overbergh, N., H. Berghmans, and G. Smets Polymer, 16,
Park, 1.-A., and J. M. D. MacElroy Mol. Simul., 2,
Patton, C. J., R. M. Felder, and W. J. Koros J. Appl. Polym. Sci., 35,
Paul, D.R., and W. J. Koros J. Polym. Sci., 14,
Pauly, S. In Polymer Hundbook,3rd ed., Vol. J. Brandup and E. H. Immergut,
Eds., Wiley-Interscience, New York, p.
Pawlisch, C. A., and R. L. Lawrence Macromolecules, 20,
Pawlisch, C. A., and R. L. Lawrence Macromolecules, 21,
Petropoulos, J. J. Polym. Sci., 8,
Petropoulos, J. J. Polym Sci. Polym. Phys. Ed., 22,
Pickup, S., and F. D.Blum Macromolecules, 22,
Probstein,R.F. PhysicochemicalHydrodynamics, Butterworth-Heinemann,
Boston.
Ray, R. J., W. B. Krantz, and R. L. Sani J. Membrane Sci., 23,
Rehage, G., andW. Borchard In The Physics of Classy Polymers, R. N. Haward,
Ed., Applied Science, London, p. 54.
Riley, R., J. 0. Gardner, and U. Merten Science, 143,
Riley, R., U. Merten, and J. Gardner Desalination, 1,
Roache, P. J. ComputationalFluidDynamics, Hermosa,Albuquerque,NM,
p.
Robeson, L. M.,A.Noshay, M. Matzner, and C. N. Merriam Die Angew Mak-
romol. Chem., 29/30,
Rogers, C. E. Polym. Prepz ACS Div. Polym. Chem., 3(2),
Rogers,C.E. In PolymerPermeability, J. Comyn,Ed.,Elsevier,NewYork,
p.
Rogers, C. E., V Stannet, and M. Szwarc J. Phys. Chem., 63,
Sada, E., H. Kumazawa, and P. Xu J. Membrane Sci., 39,
Sangani, A. S. J. Polym. Sci. Polym. Phys. Ed., 24,
J., and J. M. Ottino Polym. Eng. Sci., 23,
Shah, N., J. E. Sax, and J. M. Ottino Polymer, 26,
Shah, Y. T.,and J. H.Porter J. Appl. Polym. Sci., 17,
Shlesinger, M.F.,and J. T. Bendler In Phase Transitions in Soft Condensed
Matter, T. Riste and D. Sherrington, Eds., Plenum, New York, p.
Sirkar, K. K., N. K. Agarwal, and G. Pandurangaiah J. Appl. Polym. Sci., 22,
Slattery, J. C. Momentum, Energyand Mass Transfer, 2nd ed., Robert E. Krieger

Pub. Co., Huntington, New York, p.
Sourirajan, S. Reverse Osmosis, Logos/Academic, New York.
Sourirajan,S., Ed. Reverse Osmosisand Synthetic Membranes,Natl. Res. Council
of Canada, Ottawa.
Stem, S. S. S. Kulkami, and H. L. Frisch J. Polym. Polym. Phys. Ed.,
21,
TRANSPORT PHENOMENA INPOLMMER MEMBRANES 209
Stilbs, P. (1986). Prog. NMR Spectrosc., 19,

Subramanian, S., J. C.Heydweiller,and S. A. Stem(1989). J. Polym. Phys. Ed., 27,
1209.
Tamura, M., K Yamada, and H. Odani (1963). Rep. ProF. Polym. Phys. Jpn., 6, 163.
Taylor, G. I. (1953). Proc. Royal (London),A219, 186.
Truesdell, C., andW.No11(1965). In Handbook of Physics, Vol. IW3, S. Fliigge, Ed.,
Springer-Verlag, Berlin.
Tsay, C. S., and A. J. McHugh (1991). J. Polym. Sci. Polym. Phys. Ed., 29, 1261.
Tshudy, J. A., and C. von Frankenberg (1973). J. Polym. Sci. A-2, 11, 2027.
Turback, A. F., Ed. (1981). Synthetic Membranes, Vols. I and 11, ACS Symp. Set, ACS,
Washington, DC, pp. 153-154.
Vieth, W. R. (1991). Diffiion In and Through Polymers, Hanser, New York.
von Meerwall, E. D. (1985). Rubber Chem. Tech., 58, 527.
Vrentas, J. S., and J. L. Duda (1979). AIChE J., 25, 1.
Vrentas, J. S., C. M. Jarzebski, and J. L. Duda (1975). AZChEJ., 24, 894.
Vrentas, J. S., J. L. Duda, and A.-C. Hou (1984a). J. Appl. Polymer Sci., 29, 399.
Vrentas, J. S., J. L. Duda, S. T. Ju, and L.-W. Ni (1984b). J. Membrane 18, 161.
Vrentas, J. S., J. L. Duda, and W. J. Huang (1986). Macromolecules, 19, 1718.
Windle, A. H. (1985). In Polymer Permeability, J. Comyn,Ed.,Elsevier,NewYork,
1985, p. 75.
Yasuda,H.,andC.E.Lamaze(1970). In Membranes from Cellulose and Cellulose
Derivatives, A. F. Turback, Ed., Interscience, New York, p. 157.
Yoon, D. Y., and P. J. (1979). Disc. Faraday 68, 288.
Zachmann, H. G., and G. Konrad (1968). Macromol. Chem., 118, 189.
5
Supermolecular Structure of Polymer
Solids and Its Effects on Penetrant
Transport
Sei-ichi Manabe
Fukuoka Women’s University
Fukuoka, Japan
I. INTRODUCTION
A. Scope and Context the Presentation
The penetration of small molecules into a polymer solid or liquid frequently
plays an important role inindustrial processes such asthe spinning and finishing
processes in fiber manufacturing (T.akeda and Nukushina, 1963) and the casting
and finishing processes in membrane manufacturing (Manabe et al., 1987). The
kinds of solvent molecules used in solidifying a polymer through coagulation
influences the supermolecular structure of the finished polymer solids. For ex-
ample, acetone, which is employed for the coagulation of a solution of poly-
paraphenylene terephthalamide film (Haraguchi et al., 1976) and cuprammon-
ium-regenerated cellulose hollow fiber (Fujioka and Manabe, 1995), works
as to orient the hydroxy group in the direction parallel to the surface plane of
a film or a fiber, whereas waterwould orient it in the direction perpendicular to
the plane.
When one solvent component of a polymer solution remains in the polymeric
material during the solidification stage, it is found that even though it is finally
removed from the polymer solid, this component can penetrate easily through
the solid in question as if the solid has retained the memory of that solvent
(Iijima and Manabe, 1983).
Although suchsmall-molecule effects contribute to the completion of the fine
structure of the polymer solid andthe specific permeation of the molecules
21l
212 UANABE
through the solid, this chapter deals only with the supermolecular structure of
the polymer solid (see Section 11) relating to the penetrant transport (Sections
IV and V), in addition to the thermal motion of the polymer chain in the solid
(Section III) and the molecular interaction (Section IV) between the polymer
solid and the small molecule.
6. Methodology for Development of Novel Materials in the

Field of Penetrant Transport
Studies of the correlation of penetrant transport andthe conditions of preparation
of the solid polymer have been carried out rigorously, andsome empirical equa-
tions for the ideal manufacturing procedure have been proposed in the field of
membrane technology (Kesting, 1985) and in that of fibers (Mark et al., 1967).
These studies can be classified intoa blackbox that connects phenomena without
causal sequence. When the demandfor penetrant transport becomes complicated
as seen in the case of virus removal filters (Manabe, this methodology
gives only a low level of efficiency in attaining the final goal.
Figure 1 summarizes the flowchart that relates social needs and demands and
manufacturing conditions through the knowledge supermolecular structure
and physical properties. When we suceed in determining the quantitative rela-
tionship between structure and properties-the causal sequence, to speak-
then we can design the fine structure of the polymer solid that will satisfy the
demand. The key research area for this methodology is the investigation of
supermolecular structure. The research and development for novel materials ori-
ented to the market is carried out along the lines of the flowchart shown in Fig.
) ~ e m b n eManufac (Forming Mechanism)

Correlation betweenpreparation
condition and pore chsract.
FineStructure)
!;rope
;tis I(Transport M e c k i )
Version technical t e n i n d o g y
ISocial Needs I
Figure Flowchart for relating social needs and membrane manufacturing: The num-
bers 1, 2, and indicate the stream development a novelmembranethat fits a
social need.
SUPERMOLECULAR STRUCTURE OF POLYMER
SOLIDS 213
1. Consumer demand may be transformed into physical properties, and then the
supermolecular structure that is ideal for getting these properties is designed
using the structure-property correlation. This supermolecular structure can be
realized by the leading principle governing the mechanism of solidification of
polymer liquids or melts. One example research and development following
the chart is demonstrated in the virus removal filter (Manabe, 1992).
C. General Description of Supermolecular Structure of

Polymer Solids
We define supermolecular structure and fine structure in the following section
although these definitions are stricter than those that have been accepted so far.
We classify supermolecular structure into many structural factors according to
a numbered scheme as the second structural factor, the third structural factor,
and ondependingonthe size ofthedomainnecessary for characterizing
experimentallythe structure inquestion. The supermolecular structure repre-
sented by the lower structural order factor is influenced by the chemical structure
(this structure is defined as the first structural factor), and the supermolecular
with the higher order factor can be controlled by appropriate preparation con-
ditions independent of the chemical structure. Fine structure constitutes the third
structural factor in a narrow sense. The structural characteristics belonging to
fine structure are, for example, the orientation of chain molecules alonga crystal
plane, the crystallinity, the crystal size including lamellar thickness, the disper-
sion state of the crystal and amorphous regions, and the intersurface character-
istics. All these are concerned with sizes between 5 and lo3 nm.
Figure 2 shows a schematic representation of the relationship between su-
permolecular structure and the mechanisms of penetrant transport through the
polymer solid. The arrow shown by a full line indicates the flow of a penetrant
witha single flowmechanism as denoted,and the broken line indicates the
actual general penetrant transport. Because the transport the penetrants occurs
mainly from the environmental liquid or gas to the inside the polymer solid,
a higher order structural factor always contributes to the transport. When the
penetrant molecule is small, the rate-determined region transport is related to
the lower order structural factor. For example, dissolution/diffusional flow is
affected much more by the first-order structural factor (i.e., chemical structure)
than by a higher order factor. In the case viscous flow, the coagulation struc-
ture (the fourth structural factor) dominates penetrant transport and is almost
independent of other factors. In the actual case of penetrant transport, the com-
plex situation shown by the broken line in Fig. 2 may occur, and the contribution
of each factor the transport may change from case to case.
Table 1 summarizes the classification supermolecular structure and lists
the characterization methods for the various structural orders. A first-order struc-
214 “E
Generalpenetranttransport
DissolutionMiffusion flow -Free volume

model
Surfacediffusional flw
2 Schematicrepresentationofrelationshipbetweensupermolecularstructure
and penetrant transport through a polymer solid. The numbers indicate the order of the
structural factor. The area crossed by the full-line arrow denoted by the flow mechanism
indicates the contribution of that structural factor to the flow mechanism.
tural factor can be mainly evaluated through resonance-type spectrometries such

as infrared or nuclear magnetic resonance spectrometry. The second can be eval-
uated through difhaction methods such as X-ray and/or electron diffraction and
through resonance spectroscopy. The third structural factor can be characterized
mainly through electron microscopy and relaxation absorption, and the fourth
through optical microscopy and/or the methods of interfacial phenomena such
as mercury intrusion porosimetry. The probe-type microscopy developed re-
cently gives information on the second and third structural factors, especially
on interfacial structure.
When we use the additivity principle we can estimate the properties of the
whole system from those of the basic units. For example, the refractive index,
optical absorbance, degree of swelling and dissolution, specific heat capacity,
latent heat of the first-order transition, thermal expansion coefficient, etc., can
be estimated from the values of the components through the additivity law based
on weight or volume fractions. On the other hand, the dynamical properties can
be calculated through a series or parallel type of combination of components or,
more generally, by applying Takayanagi’s (Takayanagi, 1967; Manabe and Tak-
ayanagi, 1970b) model. In the case of penetrant transport, three types of com-
bination methods have been employed: the additivity law of the component units
based on weight fraction, the series connection, and Takayamagi’s model type
of connection. The type of combination that should be employed is determined
Definition of Supermolecular Structure and Methods of Evaluating It
arget Structural
Structure
description
Evaluation
methods.
(nm) sizeorder
First 0.1-1.0 Chemical structure, moleculal Spectrometry (NMR,
(chemical weight and its distribution, IR, vis, W, MS),
structure of configuration and tacticity, chromatography
monomer) composition of copolymer W , GPC,"LC),
and blend and its light scattering,
distribution. viscometry
Second 1-10 Conformation, local chain NMR, IR, LS,
(conformation) orientation, crystal diffractometry (X-
structure, amorphous ray, electron), CD,
structure ORD, EM, AFM
Third 10-10' Orientation (chain axis, NMR (broad-line),
(fine structure) crystal plane), interfacial IR, LS (small-
structure, crystal size, angle), X-ray
crystallinity, crystal diffraction, EM
perfection, molecular @EM,
packing (density, optical
regularity), lateral order microscopy,
thermal analysis
@TA, DSC),
viscoelastometry
(dynamic, static),
refractometry
Fourth lo2-lo4 Aggregation structure Electron microscopy,
(aggregation (particle size, porosity, optical
structure) degree of amalgamation), microscopy,
fibril and microfibril (size, viscoelastometry,
length, orientation), permeability,
microvoid, pore structure porosimetry,
(pore size and its adsorption
distribution, pore shape, isothermometry
porosity)
Fifth 104-106 Shape of membrane (plane, Optical microscopy,
(morphological hollow fiber), symmetrical light scattering
structure) and asymmetrical
membrane, complex
membrane, dynamic
membrane, liquid
membrane
'NMR = Nuclear magnetic resonance, IR = infrared, vis = visible, UV = ultraviolet, MS = mass
spectrometry,LC = liquidchromatography,GPC = gel permeationchromatography, TLC = thin
layerchromatography, LS = light scattering, CD = circular dichroism, ORD = optical rotary dis-
persion, EM = electronmicroscopy, AFM = atomicforcemicroscopy,SEM = scanning EM,
TEM = transmissionEM,DTA = differentialthermalanalysis, DSC = differentialscanning
calorimetry.
215
216 W X B E
through information on the supermolecular structure of the higher order struc-

tural factor.
II. STRUCTURALCHARACTERISTICS OF POLYMER SOLIDS

mentioned in other chapters (see Chapter 3), the penetrants permeate into a
polymer solid through the pores (capillary model) and/or through the free vol-
ume (free-volume model). We must note that the difference between models
exists because of the assumption that pores may exist in the solid. The pores
play the main role in transport in the capillary model and are neglected in the
case of the free-volume model. In the latter model the fine structure of the
amorphous region, where the polymer segments frequently jump from their equi-
librium positions to adjacent holes that generate by the random thermal motion
of the segments at temperatures above the glass transition temperature Tg,de-
termines the transport.
Figure 3 illustrates the structural features of both models and their peculiar-
ities. In the capillary model, the contribution of the chemical structure is taken
into account by the adsorption and/or solubility coefficient of the penetrant in
the polymer solid. How to characterize the pores is discussed in Section ILC.
On the other hand, information about fine structure of the noncrystalline region
is necessary for the description of the free-volume model because the penetrant
molecule can diffuse into the domain with a large free-volume fraction, and this
domain is limited only by the amorphous region where the molecular chains are
activated by the micro-Brownian motion or the local twisting motion. A more
detailed interpretation is given in Section II.B.
Capillary model Free volume model
3 Comparison between capillary model and free-volume model from the view-
pont of dye uptake. The small dots stand for water molecules, the open circles are dye
molecules, and the full lines stand for polymer chains. In the capillary model, the water
molecules in a pore make a water channel for a dye molecule to diffuse through. In the
free-volume model, the diffusion region is the amorphous region with water molecules
where the polymer segments move actively under segmental micro-Brownian motion.
SUPERMOLECULAR
STRUCTURE OF POLXMER
SOLIDS 21 7
A. Supermolecular Structure of Crystalline Polymer Solids

When the chemical structure of a polymer is given, we can distinguish whether
or not the polymer is inherently crystallizable. For example, a polymer that has
atactic side chains cannotcrystallize under normal conditions. On the other hand,
even if a polymer is potentially crystallizable judging from its chemical struc-
ture, it may have very low crystallinity at room temperature becauseof its low
melting temperature and/or the preparation conditions, prohibiting its crystalli-
zation when quenched from the melt, for instance. Then we define a crystalline
polymer solid as a polymer solid whose crystallinity is more than at
A noncrystalline polymer solid has crystallinity ofnot more than This
definition is based on actual crystallinity andnotonthe polymer’s chemical
structure. Since crystallinity varies depending onthe methods of evaluationsuch
as X-raydiffraction,density,andinfrared absorption, the X-raydiffraction
(wide-angle) method is employed here as a standard.
In a general way, polyethylene (low density, high density,linear low density),
isotactic poly-a-olefins such as isotactic polypropylene and isotactic polybutene-
1 and on, except polymers with side chains whose number of carbon atoms
ranges between four (polyhexene-l) and eight (polydecene-l), isotactic poly-4-
methylpentene-l, polytetrafluoroethylene, Nylon 6 and Nylon 66, polyethylene
terephthalate (PET), polyacetal, cellulose, and poly-pura-phenylene terephath-
alamide are crystalline solids. We can also easily prepare noncrystalline solids
from PET and from cellulose. The polymers that tend to solidify into noncrys-
talline solids are, for example, atactic polymerssuch as atactic polystyrene,
atactic polymethyl methacrylate, polyacrylonitrile, polyvinyl acetate, and poly-
vinyl chloride; polymers whosechemical compositionis complicated with bulky
side chains such as polycarbonateand many kinds of copolymers; and polymers
with low meltingpoints such as cis-174-polybutadieneand cis-1,4-polyisoprene.
The crystalline polymer solids have very complex supermolecular structures
because of the coexistence of crystalline and noncrystalline regions. Although
the noncrystalline regin is the principal area for diffusion of penetrants, both the
dispersion state of the crystals and its content (crystallinity) directly influence
penetrant transport, and the domain boundary of a crystal may affect the trans-
port indirectly.
There is a tendency in a crystal to retain its molecular conformation even in
thenoncrystallineregion. For example, thenoncrystalline region of isotactic
poly-a-olefin polymers shows two peaks in the curve of X-ray intensity of wide-
angle diffraction versus diffraction angle. The first indicates a distance of ca.
0.4 nm, which corresponds to the distance between the nearest-neighbor side
chains, and a longer one that varies depending on the length of the side chains
and corresponds to the distance between adjacent main chains. These two are
also observed in the crystalline region (Manabe and Takayanagi, 1970~).
218 MANABE
The size of a crystal along the direction of the molecular chain is several
tens of manometers, and this value is far less than that of the length of a mo-
lecular chain. This fact supports the fringed micelle structure shown in Fig. 4a.
Many polymer single crystals have been observed. This observation supports
the folded-chain crystal model (see Fig. 4b) and has been regarded as a model
more widely applicable than the fringed micelle model. molecular chain is
folded within ca. nm thickness in the folded-chain crystal, and this type of
crystal is referred to as a lamellar crystal in a polymer solid. The lamellae are
observed in a spherulite generated from a polymer melt.
When polyethylene melt is crystallized under a high pressure (e.g.,
atm), extended chain crystals are generated. In such crystals, chain molecules
are disposed parallel to each other with extended form and the crystal thickness
corresponds to the length of the chain (see Fig. 4c).
Underhigh shear rates in dilute polyethylene solutions (ca. 0.1 wt % in
xylene), folded-chain crystals are first generated and then they are stretched and
deformed by shear stress. Finally, the center of the aggregates formed is con-
structed of an extendedchain crystal, and folded crystals that grow perpendicular
to the direction of the extended chain crystal are formed at the periphery of the
extended chain crystal. This structure is called a shish kebab structure because
of its similarity in appearance to the Turkish food of that name.
The dispersion state of crystalline andnoncrystalline regions is given by
Takayanagi's model shown in Fig. 5 (Takayanagi, 1967). The black area indi-
cates the noncrystalline region and the white area the crystalline region, and the
--
.
A.
?.
.
"
...
n .b..
""
C
..-
4 mica1 molecular arrangement in a crystal. (a) Fringed micelle;

chain crystal; (c) extended chain crystal.
folded-
SUPERMOLECULAR STRUCTURE OF POLYMER SOLIDS 219
5 Takayanagi’smodel(Takayanagi, 1967) forrepresentingfinestructure of

crystallinepolymersolid.Blackarea:noncrystallineregion.Whitearea:crystalline
region.
two regions are connected in parallel and/or in series. Although this model has
been proposed to interpret the viscoelastic response, we must take into account
this type of connection even in the case of penetrant transport because the pen-
etrant may pass through both regions in the case of a polymer blend.
B. Supermolecular Structure of Noncrystalline Polymer

Solids
When we define the noncrystalline region as the part not belonging to the crys-
talline region, then its content will vary depending on the evaluation method
employed for the crystalline region. Table 2 shows some examples of the extent
of the noncrystalline region evaluated onthe assumption that thecontent is equal
to 1 - crystallinity. The content cannot be determined definitely. The noncrys-
talline region is classified into three states according to the packing regularity
of the molecular chains: the smectic state of two-dimensional regularity, the
nematic state of one-dimensional regularity, and the amorphous state with no
dimensional regularity. Additionally, the molecular chains in the noncrystalline
region can be characterized by thermal motion (e.g., segmental microBrownian
motion) and are represented by the packing density of the molecular chain with
the same regularity (Manabe and Kamide, 1984).
220 MANME
Table 2 Content of Noncrystalline Region' with Three Methods of Evaluating

Crystallinity
Polymer IF Densityd
0 terephthalate
Polyethylene 0.25
0.27 Polyethylene 0.28 0.26
Regenerated cellulose - 0.25
'Given by 1 - crystallinity.
bWide-angle X-ray diffraction method.
lnfrared ray method.
dApparent density method.
In Fig. 6 is shown a schematic representation of the noncrystalline region for

three types of polymer solids-a typical amorphous polymer solid, a typical
noncrystalline polymer solid, and a typical crystalline polymer solid-observed
from two different standpoints of regularity and packing density, the latter being
the representation of the segmental thermal motion.
Schematic representation of frequency distribution of molecularchainsin

the region characterized by a given regularity and a given packing density. Curves1, 2,
and show the values of a perfect amorphous polymer, a noncrystalline polymer, and a
crystallinepolymer,respectively. N, S, andContheregularityordinateindicate
amorphous, nematic, smectic, and crystalline regions, respectively.
SUPERMOLECULAR
STRUCTURE OF POLYMER
SOLIDS 221
The fine structure of the noncrystalline region has been investigated through
electron microscopy for direct observation, electron diffraction, and dark-field
imaging. Nodules 2-4 nm in size and grain boundaries of 1-2 nm were ob-
served andfound to possess a little regularity,andthe remaining intergrain
region was composed of random coil molecules. On the basis of these results,
the model shown in Fig. 7 was proposed (Yeh, 1972).
The radius of gyration of a molecular chain in a melt was foundto coincide
with that of achain in a noncrystalline solid and to be proportional to the square
root of its molecular weight. This result supports the conclusion that the mo-
lecular chains in a noncrystalline region have the random coil conformation.
The supermolecular structure of the noncrystalline region of actual polymer
solids such as a fiber and/or a membrane depends on their preparation condi-
tions. Sections II.C and 1II.A will deal with the structures achieved under dif-
ferent preparation conditions.
Fibers have a fiber axis as a symmetric axis of the supermolecular structure,

and the direction of penetrant transport is perpendicular to the fiber axis. On the
other hand, the symmetric of the supermolecular structure of a film is, in
principle, perpendicular to the film surface, and the transport direction coincides
with that of the symmetric axis. Consequently, the contribution of supermolec-
ular structure to penetrant transport is different between a fiber and a film, a
fact that should be taken into consideration in the series mode and parallel mode,
respectively, in the additivity models.
Modelrepresentingthestructure a noncrystallineregion(Yeh, 1972).

OD, GB, and IG denoteordereddomain,grainboundary,andintergrainregions,
respectively.
222 MANABE
of
For a fiber, the most important characteristics is the molecular orientation along
the fiber axis. The orientation is caused by the spinning process and influences
the transport of penetrants. For the removal of viruses using hollow fibers, this
orientation should be minimized as it prohibits the generation of void-type pores
penetrating the hollow fiber walls (Tsurumi et al., 1990).
Figure 8 shows two typical structural models representing the noncrystalline
region in a fiber. The fringed micelle model (Fig. Sa) requires that each molec-
ular chain penetrate both crystalline and noncrystalline regions, resulting in a
part of the noncrystalline region having the lateral order in molecular chain
packing. Here, the lateral order indicates that the molecular chains pack regularly
side by side in some order that makes the distance between homogeneous chains
and may even have semihexagonal packing. This model is believed to be ap-
plicable to the fibers manufactured by wet and/or dry spinning such as regen-
erated cellulose, polyacrylonitrile, and aramide fibers. Another model, shown in
Fig. 8b, originates in the folded-chain crystal model. The feature of molecular
chain packing is divided into six types: tight loop, loose loop, tie molecule, cilia,
floating chain, and extended chain. Tight loops are the chains located on a
lamellar crystal that are stretched back to the same lamella. The loose loop also
W Otypical structural structural models representing the noncrystalline re-

gion in a fiber. (a) Fringed micelle model. C and NC stand for the crystalline and
crystalline regions, respectively. The area denoted by 1.0. indicates the NC region with
lateral order. @) Noncrystalline model based the folded-chain crystal. 1, tight loop
chain; 2, loose loop chain; 3, tie molecule; 4, cilia; 5, floating chain; 6, extended chain.
starts on a lamella and returns to the same lamella, making a long loose loop.
The tie molecule connects two lamellae, and the cilia are molecular chains that
arise from one lamella and have the other end free. The floating chain is an
isolated amorphous random coil. The extended chainkeeps the shape under the
aggregation state stretched through multiple lamellae. Fibers prepared by melt
spinning have the noncrystalline region shown in Fig. 8b.
Figure 9 represents thefibrilandmicrofibrilmodel (Peterlin, 1969). The
diameter of a filament ranges from 10 to pm. The filament is considered
to be composed of fibrils having sizes between 100 and nm, and these
fibrils are composed of microfibrils about 10 nm in size. The supermolecular
structure of the microfibril is given by the model shown in Fig. 8.
'\
\
\
\
\
l
I
I
I
/
/
/
Schematicrepresentation of the fibril and microfibril model proposed for

melt-spun crystalline polymer (Peterlin, 1969). A filament pm in diameter is
composed of many fibrils having diameters between and 1000 nm. A fibril is com-
posed of many microfibrils with diameters of = 10 nm. The supermolecular structure of
a microfibril is shown at the right in this figure under an appropriate enlargement and is
similar to that shown in Fig. 8b.
224 UANABE
The change in supermolecular structure along the radial direction is also a

very important factor for the permeation molecules (i.e., dyeing) from the
surface to the inside of a fiber.
Figure 10 shows an example of the change in the supermolecular structure
of Kevlar 49 alongthe radial direction (Manabe et al., 1980). The microfibril is
subdivided into fine tips having widths of 5-10 nm and lengths 0.3-0.5 pm.
The chain orientation (c axis orientation) denoted by the length the arrows c
and the plane orientation denoted by the length the arrows b decrease toward
the center of the fiber.Fibers prepared by thewet-spinning process show changes
similar to that of Kevlar 49. For example, the cuprammonium-regenerated cel-
lulose fiber Bemberg showsmaximum crystallinity at ca rlR (where R is the
radius of the fiber and r is the radial coordinate measured from the center of
the fiber), and the maximum of the molecular chain orientation (b axis orien-
tation) is located at ca. rlR (Manabe et al., 1980). The change in porosity
for various kinds hollow fibers has been clarified (Kamideet al., 1989). Melt-
spun fibers such as polyethylene terephthalate also show a similar change, es-
pecially in the case the fiber manufactured through high-speed take-up(Kuriki
et al., 1985).
5-10 nm
a block
‘a fiber \
a microfibril
Fibril-like block structural model for Kevlar 49. The original model pre-
pared by Manabe et al. (1980) is modified a little in this figure. The figure on the left
shows the aggregation stateof the microfibrils between5 and nm in width. The degree
of orientation of the b axis along the radial direction of a fiber is shown by the length
of the arrows labeled b. The figure on the right shows the supermolecular structure in a
microfibril cut to a length of less than 0.3-0.5 pm. Here, the length corresponds to the
pleated sheets of the c axis. The molecular chain exists in a microfibril in the extended
state. The chains in a noncrystalline region shown by wavy lines are also extended along
the fiber axis.
For a film, the orientation of the molecular chains can be classified into three
types: undrawn, uniaxially drawn, and multiaxially drawn. It is difficult to dis-
tinguishbetweenundrawnandmultiaxiallydrawnfilms from the chain axis
orientation because the latter is generally drawn as to be isotropic. The me-
chanical properties of drawn films are superior to those of an undrawn film.
film frequently shows the orientation of a specific plane near the film
surface. Thistype of orientation may influence the transport of penetrants
because the adsorption/dissolution properties may be governed by the orien-
tation of the active groups, and this orientation near the film surface accom-
panies the plane orientation. Ultrathin films with a thickness of less than 20
nm show characteristic peculiarities. For example, the direction of the inter-
molecular hydrogen bonds at the film surface varies depending environ-
mental conditions (Haraguchi et al., 1976; Fujioka and Manabe, 1995), and
the distribution of the component in the polymer blend is also influenced by
the film thickness and the materials in contact with thefilm surface (Kajiyama
et al., 1995). The molecular conformation and molecular aggregationin a thin
membrane such as a biomembrane play the main role in molecular transport
through the membrane.
In artificial membranes, although the structure of homogeneous membranes that

have no observable pores under an electron microscopecan be regarded as
having the same type of supermolecular structure as described above for films,
the aggregation of polymer chains in the porous membrane (the fourth structural
factor in Table 1) should be characterized in advance, as the contribution of a
pore to the transport as a whole increases strongly with pore size.
Figure 11 illustrates three typical commercially available porous polymeric
membranes: a membrane with straight-through cylindrical pores, an asymmet-
rical membrane, and a membrane with a multilayer structure (Manabe et al.,
1987). All of these have observable pores, and when the membrane is sliced
parallel to the surface in ultrathin slices, the section shows the straight-through
pore structure. Consequently, in order to characterize a porous membrane, we
must first take into account the pore structure in a plane. Here “porous mem-
brane” includes both porous plane-type membranes and porous hollow fiber-
type membranes.
We define the pore radius distribution function N(r) (Kamide and Manabe,
1980) as the total amount of the pore, whose radius ranges between r and r +
dr, is given by N(r)dr in a unit surface area. The ith-order mean pore radius ri,
the ith order moment Xi, the porosity P,, and pore density N are given by the
226
- ,
. .
.
252
Three typical porous polymeric membranes. (a) Amembrane with straight-

through cylindrical pores. In general, the pore density is about loRpores/cmz surface
area,andthethickness d islessthan 30 pm. asymmetricalmembrane.This
membrane is composed two types of pores: very fine pores on the top surface and
very large pores (or finger-shaped pores) inside the membrane. The top surface is com-
posed of very fine particles-10 nm in size (these particles are called primary particles).
This thin layer with a thickness -70 nm including the top surface is considered to
be the active layer for the particle separation. (c)The multilayer structural membrane.
The wholeof the membrane is composedof particles with a size of ranging between
50 and nm (these are termed secondary particles). Because of the anisotropic feature
of the membrane-manufacturing process, the particles are deformed and amalgamated,
resulting in the layer structure shown in cross-sectional view.
SUPERMOLECULAR
SOLIDS 227
following equations:
The scanning electron microscope (e.g., field emission type, FE-SEM) shows
the detailed pore structure and at high-resolution power pores more than 5 nm
in size can be observed directly. A novel technique for observing the connection
between pores has been proposed (Tsurumi et al., 1990a), and we can easily
obtain information of pore structure as a function of radial coordinates (Kamide
et al., 1989).
Many evaluation methods (Kamide and Manabe, 1980) for N(r) have been
proposed far: (1)mercury intrusion (MI) method, (2) bubble pressure and
pressure dependence of the filtration rate offluid(liquid and/or gas) (BP
method), (3) permselectivity of various particles (UP), (4) gas permeation (GP)
method, (5) electron microscopy (SEM and TEM methods), and others (ther-
modynamical methods such as melting point of an immersingliquid, evaporation
pressure of adsorbed gas using Kelvin's equation).
Figure 12 shows a comparison for N(r) distributions obtained using various
methods of measurement (Manabe et al., 1985). The values of N(r) depend on
the method used. Consequently, the features of the evaluation method should be
known in advance, and the marginal limit of the method must be taken into
account.
Some formsof the mean radius ri can be evaluated directly without knowing
N(r). The filtration rate per unit filtration area for a liquid, J , under a transmem-
brane pressure of gives the mean pore radius r, using the following equation
derived through the Hagen-Poiseuille equation,
where P, is the porosityof the membrane, d is its thickness, and-q is the viscosity
of the liquid. The value of rf corresponds to (r3r.,)lRfor the membrane with
straight-through cylindrical pores. When the surface area of the membrane in-
cluding pore and membrane surface S, is observed, we can calculate the mean
228
.-
0 0.5
Pore radius /pm Pore radius /pm
Comparison of poresizedistributionfunctions N(r) for the porous cu-

prammonium-regenerated cellulose membrane evaluated through various methods. (a)
N(r) obtained through electron microscopy (EM). The broken line was obtained under
the assumption that all pores have ellipsoidal cross sections, and the full line represents
values obtained in the caseof spherical pores. Curves1 and 2 are the valuesof the front
and back surfaces of the membrane. @)N(r) values obtained through EM, MI, and BP
methods. Pores are assumed to be straight-through cylindrical pores in all cases.
pore radius r, (corresponding to r2)as
The permeability coefficient a gas gives the specific mean pore radii corre-
sponding to r,, r,, and (r3r4)1R
(Nohmi et al., 1978).
The porosity P, can also be evaluated directly through the apparent density
using the equation
p, = 1 - PalPp 0
where the subscript attached to P, indicates that the P, value is evaluated
through the experimental value and pp is the density the polymer solid
SUPERMOLECULAR
SOLIDS 229
employed in the membrane. The porosityobtained through the immersion

method, P,, is calculated using the equation
where W, and W,are the weights of themembraneand the liquid used for
immersion, and ps is the density of the liquid. Table shows a comparison of
the values obtained by various methods mentioned above for the two types of
porous membranes shown in Fig. 5.11a and llc. When the pore structure be-
comes complicated as in the case of aregenerated cellulose membrane, the
difference in the characteristic values obtained through several methods becomes
more obvious.
The maximum pore radius r,, is evaluated through the bubbling method.
The bubbling pressure D b is the pressure at which a small amount of a fluid
(liquid and/or gas) passes through the membrane initially in the course of in-
creasing the transmembrane pressure. The value of r,, is calculated as
r,, = 2 u J h p b (9)
where u2 is the interfacial tension between the two fluids at the interface. The
value of r,, is often employed as the quality control parameter for the mem-
brane and as the integritytest value for the sterilization filterandthe virus
removal filter.
D. How to Actualize the Intended Supermolecular Structure

Although the supermolecular structure of a polymer solid is determined in ad-
vance as discussed in the previous section, it is only when we know the coag-
ulation mechanism of a polymer chain that a more critical understanding of the
structure becomes possible.Fromthe manufacturer’s aspect, the preparation
method and preparation conditions for the solidification of a polymer fluid are
essential for effective development of novel polymer solids.
Table 4 summarizes the molding method and the solidification mechanism
employedin the molding.Here,wet,dry,andmelt moldings aredefined as
follows:
Wet molding. A polymer solution contacts another solution, and this contact
gives rise to the transport of good solvents from the polymer solution to the
other solution and/or of poor solvents in the opposite direction, and finally
the polymer solidifies.
Dry molding. A polymer solution contacts gases, and evaporation of the good
solvents changes the solution into a solid.
' I
**m
y-9
000
+ m +
99q
000
O + d
0 0 0
230
SUPERMOLECULAR
SOLIDS 231
Melt molding. A polymer melt is cooled to a temperature below the melting

temperature and/or the glass transition temperature. During the cooling,
lidification by crystallization and/or vitrification occurs.
In the actual molding process, a complex molding method is often employed.
For example, the polymer solution is first cast or spun into a gas environment
and is then immersed ina solution. There are four important solidification mech-
anisms: crystallization, fine particle generation,vitrification,andpolymer-
polymer two-phase separation.
As for crystallization, there are two mainsteps, nucleation and crystal growth.
Nucleation occurs in a sporadic and/or predetermined manner. The number
nuclei generated by nucleation decreases withan elevation the crystallization
temperature, resulting in an increase in crystal size. In contrast, the number of
nuclei generated through predetermined nucleation remains constant, indepen-
dent of the crystallization temperature and time, but varies with the coexistence
Table 4 Preparation Method and Solidification Mechanism for Crystalline and

Noncrystalline Polymer Solids
Molding Porous
hod solid Polymer
ary Wet Crystalline Primary and
particleparticle" secondary
particles
Dry ? ? Primary particles
Crystal-Crystal- Melt Phase separa-
lization lization tionb, crystal-
lization and
drawing
mary
Primary Wet
Noncrystalline Primary and
particles
particles secondary
particles
Amalgamation
Primary Dry Primary and
gelation
particles
or secondary
particles
Melt Vltlifl- Vitrification Phase separa-
cation tion,
dissolution
'Fine particle generation through microphase separation.
bPolymer solid/polyrner solid two-phase separation.
232 M A B E
C i rcular pore 1
I
Noncircularpore
l nside ..
13 Schematicrepresentation of thepore-formingprocessstartingfromcel-
luloseacetatesolution(Kamideetal., 1977). Thefilledcirclesanddomainsarethe
polymer-rich phase and/or polymer-concentrated gelatinous phase. The circular and non-
circular pores denote that the shape of the pore is a smooth circular arc or an irregular
shape,respectively."Interfacialsurf"and"Inside"indicatethecross-sectionalview
near the interfacial surface of the feed solution and the view at the inside of the mem-
brane, respectively. Prim = primary particle, secon. = secondary particle, m.p. = micro-
phase separation, aggre.= aggregation of the particles, phase inv.= phase inversion, and
solidif. = solidification.
of extraneous substances such as remaining catalysts andoligomers. Crystal

growth proceeds along onedimension (fiberlike growth) or two dimensions (la-
mellar growth) or, more likely, in three dimensions (spherulitic growth). When
a constant temperature gradient is imposed, growth tends to occur in one di-
mension along the direction of the temperature gradient as in the case of b-axis-
oriented crystals inpolypropylene.When we carry out the crystallization of
polymer melts on a foreign material, many nuclei are generated on the surface
and crystals are forced to grow in one dimension directed as a whole perpen-
dicular to the surface (this type of crystals is called a trans crystal).
The generationof fine particles has been observed in cases of both crystalline
and noncrystalline polymer solids. In the wet-molding process in particular, the
supermolecularstructure of the polymer solid be interpreted as the manifold
aggregations of the particles. The polymer solid is constructed of kinds of
fine particles, the primary and secondary particles described below. The typical
example applied for this particle generation mechanism in the solidification of
polymer solids is the case of a porous polymeric membrane. Figure 13 shows
a schematic representation of the process of pore formation in cellulose acetate
SUPERMOLECULAR STRUCTURE OF POLXMER SOLIDS 233
membranes (Kamide et al., 1977). This processis an exampleof the microphase

separation method (Manabe et al., 1973) for preparing porous polymericmate-
rials. Here, the microphase separation method consists of the following steps:
In the wet and/or dry molding process, the homogeneous one-phase solution
separates into two phases, a polymer-rich phase and a polymer-lean phase. This
separation begins with the formation nuclei with a high concentrationof
polymer, and these nuclei grow into primary particles having sizes ranging from
10 to 50 nm. The primary particles collide with each other due totheir Brownian
motion, and after the collision the particles amalgamate, resulting in the for-
mation of secondary particles. Secondary particles have a size between 50 and
several hundred nanometers. Since the particles are rather stable and the change
in size appears small, this phase separation is referred to as microphase sepa-
ration in the membrane-forming process. Similar fine particles have been ob-
served in many kinds of membranes and hollow fibers and in wet-spun textile
fibers (Manabe, 1993). The secondary particles do not correspond to the droplet
model proposed by Kesting and Manefee (1969). The concept of phase sepa-
ration with droplets indicates that the interfacial domain of the fine particles
(droplets) generated from the polymer solution is composed of the polymer-rich
layer and the inside the particle is a solution with a highercontent of the swelling
agents and poor solvents than that the outer solution. According to the the-
oretical approaches of Kamide and Manabe (1985) for the microphase separation
method, five fctors govern the final pore structure a porous polymeric mem-
brane: (1)the number of nuclei, (2) the polymer concentration, (3) the interfacial
tension between two phases after phase separation, (4) the rate of diffusion of
particles (viscosity, temperature, etc.), and (5) the holding time after reaching
the composition that causes phase separation (i.e., the phase separation time).
Tsurumi et al. (1990b) used this model to study regenerated cellulose hollow
fibers.
THERMAL MOTION OF POLYMER CHAINS IN A SOLID

Although two distinct models for interpreting molecular diffusion in a polymer
solid, the free-volume model and the capillary model, are based solidly on
permolecular structure, we note that the significant feature of this structure is
that it is not static but dynamic. The polymer molecules do not keep stationary
but exhibit micro-Brownian motion or other thermal motion. The movement of
polymer segments gives rise to increases in the free volume orvacant space into
which a penetrant can diffuse. Conversely, the penetrant may affect the seg-
mental thermal motion by giving its own free volume to the segment.
Figure 14 shows a comparison of the thermal motion of a crystalline and an
amorphous polymer solid. The dynamic shearmodulus G’ and the dynamic loss
tangent, tan S, are typical representatives of the dynamic viscoelasticity. The
234 W A B E
Tg
Temperature
Comparison of thethermalmotionofpolymersolids of crystalline and
amorphous polymer solids. Full line: a crystalline polymer solid. Broken line: an amor-
phous polymer solid. a,,local twisting motion of a main chain in a crystalline region;
cxa, segmental micro-Brownian motion in an amorphous region; pa,local twisting motion
of a main chain in an amorphous region; %, free rotation or local twisting motion

of a side chain in an amorphous region; ymc,free rotational motion of an end methyl
group. T, and T,,, aretheglasstransitiontemperatureandthemeltingtemperature,
respectively.
existence of the melting point and the dynamic absorption (peak of tan a,
originated by the thermal motion of the molecular chains in a crystal distin-
guishes a crystalline polymersolid from an amorphous one.Many dynamic
absorptions are common in crystalline and amorphous polymers.These are, start-
ing from the higher temperature side, the absorption a, that originates in the
micro-Brownian motion of polymer segments located in the neighborhood of
the glass transition temperature Tg, the absorption Pa due to the local twisting
motion of a main chain, the absorption or due to the free rotational and/
or local twisting motion of a side chain,and the absorption ymcdue to the
rotational and/or other types of motion of a methyl group. The generalempirical
rules obtained far are as follows:
1. The larger the moving unit is, the higher the temperature of the correspond-
ing absorption.
2. When the apparent activation energy of the movement increases, the tem-
perature shifts to a higher value even if the moving unit is the same in size.
SUPERMOLECULAR
SOLIDS 235
As the intermolecular interaction increases, the absorption from the inter-

acting molecules moves to the higher temperature side.
The peak value of the absorption increases with the size of the moving unit
and the number of moving units.
A. First-Order Thermodynamic Transitions (Melting, Crystal

Transformation, Liquid Crystal Phase Transition)
From the viewpoint of thermodynamics, many kinds transitions can be de-
fined. The first-order transition is defined as a transition during which latent heat
is generated or absorbed. In the case of melting, the equilibrium melting point
T: is given by
T: = AH/M
where AH and M are the differences in enthalpy and entropy between the melt
and the crystalline state and the contribution the surface energy to the Gibbs
free energy is neglected. When the crystal size is not as large, the contribution
of the surface energy is not negligible, and the melting point T,(Z) of a crystal
thickness l is given by
where and are the surface energy and the heat of fusion per cubic cen-
timeter of crystal. A polymer solid with a high melting temperature has strong
intermolecular interaction and high molecular rigidity. Above the melting point,
a crystalline polymer behaves very like a liquid, that is, a polymer melt.
Other examples of transitions belonging to the first order are crystal trans-
formation andliquid crystal phase transition. The transition temperatures are
given by equations similar to In this case the differences in enthalpy
and entropy arethe differences between the values before and after the transition.
The membrane formed by a mixture of a polymer with a liquid crystal shows
changes in the permeation characteristics of the membrane when the permeation
temperature crosses the liquid crystal transition temperature(Washizuetal.,
The first-order transition characteristics of a polymer liquid crystal have
been discussed in detail for various polyphosphazens (Schneideret al.,
B. Glass Transition and Segmental Micro-Brownian Motion

Part of a polymer melt of a crystalline polymer or most of the melt a non-
crystalline polymer can be frozen into the glassy state without crystallization.
This transition is the glassy transition. At this point, the second derivatives of
thermodynamic properties such as the specific heat capacity and the coefficient
of thermal expansion show a stepwise change. Although this step change is
236 " B E
apparentlysimilar to the second-order phase transition, the glass transition

should be regarded as arelaxationphenomenoninwhich segmental micro-
Brownian motion plays the principal part. This transition is the most important
one for a noncrystalline polymer solid.
The glass transition temperature T, changes depending on the time scale of
measurement and the relaxation time of the segmental motion. Here, the me-
chanical absorption originated by this motion is called a,absorption. The tem-
perature dependence of the relaxation time T is represented by theWLF equation
given by Eq. (10) as a function of the temperature difference T - TB.
-17.44(T - T,)
log):( = 51.6 + T - T,
where T~ and,.T are the relaxation times at temperatures T and T,, respectively.
Equation is known to fit the data down to about TB.Equation indicates
that when the temperature approachesTB,the relaxation time increases abruptly.
Under certain limiting conditions T, can beregarded as the constant representing
the material. Table 5 lists the values of TBfor representative polymer solids.
The size of the chain segments that initiate their micro-Brownian motion at
T, is estimated to be less than carbon atoms (Nakayama et al., 1977) and
more than 20 (Manabe et al., 1969), evaluated from theapparent activation
energyof a, absorptionand from 2 to nm inlengthevaluatedfrom the
relationship between the viscoelasticity of a polymer blend and its dispersed
state observed through electron microscopy (Manabe et al., 1969). When the
size of the penetrant is similar to that of a segment, the transport may be dom-
inated by the segmental motion,and below T, penetrant transport becomes very
difficult. The diffusion of a dye molecule correspondsto this case. Even in the
case of a penetrant smaller than a segment, the diffusional flow is influenced by
the segmental movement. The free-volume model for diffusion is based on this
segmental micro-Brownian motion. For a gas, since the size of the penetrant is
far smaller than that of the segment, the temperature dependence of the diffusion
coefficient shows the linear relation in the Arrhenius plot indicating no abrupt
change at T,.
Many physical properties other than transport properties change drastically
at T,. The dynamic properties including dynamic modulus G' and tan 6, the
dielectric properties, and gas adsorption isothermsare examples.
The apparent activation energy of a,absorption ranges between 150 and 850
kJ/mol, and the activation energy of the dye diffusion into a polymer solid falls
in this range. When we plot the diffusion coefficient against the reciprocal of
the absolute temperature (Arrhenius plot), we obtain the hypothetical diffusion
coefficient D,, extrapolated to infinite temperature. The value of Do is known to
closely depend on T,, decreasing with increases in T,.
Table 5 Glass Transition Temperature TEfor Various Polymer Solids

aluation Polymer TE (K)
Polybutene-l Dilatometry 249
Polychlorotrifluoroethylene Refractive index 318
Polyethylene Dilatometry 148, 243
Poly-4-methylpentene-1 Dilatometry 302
Polypentene-l Dilatometry 233
Polypropylene
Atactic Dilatometry 253
Isotactic Dilatometry 263
Polytetrafluorethylene Dilatometry 160,400
Polymethyl acrylate Dilatometry 279
Polyethyl acrylate Dilatometry 249
Polymethyl methacrylate
Syndiotactic Dilatometry 388
Isotactic Dilatometry 318
Polystyrene Dilatometry 373
Polyacrylonitrile Dilatometry 378,433
Polyvinyl acetate Dilatometry 301
Polyvinyl alcohol Dilatometry 358
Polyvinyl chloride Dilatometry 354
Nylon 6 Dilatometry 348
Nylon 6, 6 Differential thermal analysis 330
Polyethylene terephthalate Dilatometry 342
Cellulose triacetate Dilatometry 378,430
C. Local Twisting Motion of Polymer Chains in an

Amorphous Region
phous region has been named (3.

The dynamic absorption caused by local twisting of polymer chains in an amor-
sorption is usually located between

absorption. The peak temperature of Pa ab-
and below TB'The apparent ac-
tivation energy generally ranges between 40 and kJ/mol. Although the
contribution of this type of polymer chain motion to penetrant transport has not
yet been clarified, the antiplasticizer effect that causes a decrease in thediffusion
coefficient due to the addition of small molecules may be related to the disap-
pearance of the absorption caused by this addition (Robeson,
Whena liquid with small molecules penetrates into a polymer solid, the
noncrystalline region where the molecule can diffuse is dependent on the inter-
molecular interaction between the small molecules and a polymer chain. For
example, benzene and other hydrophobic small molecules can diffuse into the
limited domain in an amorphous region where the intermolecular hydrogen bond
238 WAL3E
is poor and the van der Waals interaction is dominant (Fujioka and Manabe,
1995). Some of these moleculesprevent the absorptionlocatedatabout
-60°C as in the case of a regenerated cellulose solid (Fujioka and Manabe,
1995). This indicates that smaller molecular penetrants can diffuse only into that
restricted amorphousregionthat is related to thedevelopmentofthe
absorption.
Table 6 summarizes the peak temperature T,,, of a, and Pa absorptions and
the apparent activation energy A H, for various polymer solids including other
dynamic absorptions described below. Since the values of T,, and A H , also
depend on the supermolecular structure, we need to recognize that the values in
this table represent only typical cases.
D. Other Molecular Motions of a Polymer Chain

When the chemical structure becomes complex as when there are long side
chains, then one or more additional dynamic absorptions are generated. Figure15
shows the temperature dependenceof the dynamic tensile modulus E' and loss
modulus E" at a measuring frequency of 110 Hz for various polymer solids
(Manabe and Takayanagi, 1970c; Manabe et al., 1970b). The polymer employed
is isotactic poly-a-olefin with unbranched side chains. If we define the number
of carbon atomsin the side chain asN, then if N 6 (see Fig. Ea), the dynamic
absorptions a,, Pa(mc),and a,(,)can be observed in the temperature range
between -180 and +lOO"C. Here the subscripts c and a indicate that the ab-
sorptions are caused by the motion of the molecules in the crystalline or amor-
phous region, respectively. The mc and sc in parentheses indicate main chain
and side chain, respectively. When N > 6 (Fig. 15a), the absorptions a:,
a:(,cl,a:(+ and Pa(%)are observed.
Figure 16 shows a plot of the peak temperature against 1/(N + 2), where N +
2 corresponds to the number of carbon atoms in a monomer unit. Most of the
characteristic temperatures show a linear dependence on 1/(N + 2). When we
extrapolate these values into zero or 1/3 at the values of 1/(N + 2), we obtain
those of polyethylene or polypropylene, respectively.
The dynamic absorptions caused by the motion of the side chains have ac-
tivation energies ranging between 40 and 100 kJ/mol depending onthe value of
N. The absorption Pa(sc), whose mechanism is a local twisting motion in a side
chain, can be observed when N > and becomes more pronounced with an
increase in N. These features of Pa(,) are quite similar to those of in the case
of polyamide (Kawaguchi, 1962).
The polymer chain end group is easy to move compared to groups inside the
chain. The methyl group at the endof the main chain or side chain shows
mechanical absorption between -220 and -120"C, and the activation energy
Table 6 Peak Temperatures of Various Viscoelastic Absorptions Under an Isochronal Measurement at 110 Hz and Apparent Activation
Energy AHa for Some 'Qpical Polymer Solids
~~
Yrne aa(sc)y Pa~sc, Pa a.

Tmax ma Trnm AH* Tmm m a Tm, m a
Polymer 6 (k.J/mol) 6 (kJ/mol) 00 (kJ/mol) (m orJ/mol)
Polybutene-1 70 10 230 62 280 146
Polyethylene 150 54 260 154
Polypropylene 60-160 10 220 58 290 191
Polymethylacrylate <60 2 260 40 300 179
Polyethylacrylate 100 7 190 79 270 162
Polymethyl 70-100 1-20 370 87 410 516
methacrylate
Polystyrene 210 37 320 125 400 537
Poly acrylonitrile 340 62 430 395
Polyvinyl acetate 230 46 310 171
Polyvinyl alcohol 320 229
Polyvinyl chloride 270 46 370 283
Nylon 6
Nylon 6, 6
150
160
37
42
350-370
350-390
266-374
229-279
E
B
Polyethylene 220 75 340-380 341-279 h
terephthlate
Cellulose triacetate 32 320 420-450 320-333
Cellulose 200 208 570 1206
240 92 525 287
385 208
240 "E
. -
.
I
-100 0
Temperature I'C
' I
-100
.
0
Temperature I"C
Temperaturedependence of dynamic tensile modulus E' and loss tensile
modulus E" for poly-a-olefins with unbranched side chains. (a) The number of carbon
atoms in a side chain is more than 6. IPO 1 = isotactic polyoctene-l (N = 6), IPD
1 = isotactic polydecene-l (N= 8); IPDD 1 = isotactic polydodecene-l (N = 10); IP'I'D
1 = isotactic polytetradecene-l (N= 12); IPHD 1 = isotactic polyhexadecene-l (N= 14);
P O D 1 = isotactic polyoctadecene-l (N= 16); PE = polyethylene. (b) The value of N
is less than 6. PP 1 = isotactic polypentene-l (N = PB 1 = isotactic polybutene-l
(N= 2); PP = isotactic polypropylene (N = 1).
Plots of melting temperature and peak temperature of various viscoelastic

absorptions versus reciprocal of number of carbon atoms of a monomer, 1/(N + 2), for
isotactic poly-a-olefins with unbranched side chains. mp, melting point; 4,absorption
caused by segmental microBrownian motion; local twisting motion. The subscripts
a and c indicate that the absorption is from the molecules in an amorphous (a) region or
crystalline (c) region. The symbols mc and sc in parentheses stand for main chain and
side chain, respectively.
ranges between and 20 kJ/mol. As the moving unit is very small, this ab-
sorption may influence a penetrant gas molecule.
We have few reports of studies on the relationship between penetrant trans-
port and the molecular motion emphasizing the side chains and the methyl end
groups. This topic is open future research.
IV. CORRELATIONBETWEENCHEMICALSTRUCTURE,
COMPOSITION, AND PENETRANT TRANSPORT
A. MolecularInteraction
Molecular conformation may be decided by inter- and intramolecular interac-
tions. Intramolecular interaction determines the angle of rotation around a single
bond. For example, in the of n-butane, the trans form is most stable and
two gauche forms come in second. Even in the same chain, two atoms in ques-
tion could be far apart along the chain but close to each other in space due to
242 " B E
the folding back of the chain. This case ofmolecular interaction shouldbe
regarded as intermolecular interaction.
Interactions betweennonbonded atoms can be classified into three kinds:
hydrogen bond interaction, electrostatic interaction, and van der Waals interac-
tion. The van der Waals interaction is caused by three mechanisms: dipole-
dipole interaction, dipole-induced dipole interaction, and dispersion forces (in-
stantaneousdipole-instantaneous dipole interaction). The potentialenergy U
caused by the van der Waals interaction is expressed in Eq. (11) as a function
of the distance r between atoms:
U = - A/r6 -k B/r", n= ., . ,
where A and B are constants determined bythe combination of atoms interacting.
The first term in Eq. (11) represents the effect of the attractive force, and the
second that of the repulsive force.
Molecular interaction contributes to penetrant transport through both equilib-
rium and the dynamic rate process. Examples of the former are the molecular
chain conformation and the regularityof the molecular packing in a crystal. The
second stmctural factor in Table 1 is a reflection of intermolecular interaction.
The molecular interaction between a penetrant molecule and a polymer chain
molecule (or segment) governs the solubility of the penetrant molecule in the
polymer solid and is given a more detailed description in the next section using
the theory of affinity. In a rate process, such as diffusion, the rate of sorption
and desorption, the molecular interaction demonstrates its effect through control
of the glass transition temperature Tg and the activation energy A H a of the rate
process. Both T, and A H a increase with an increase in the interactions.
B. Affinity and Its Role in Transport Phenomena

In this section the term affinity is defined as the chemical affinity that corre-
sponds to - A b . Here, A b is the difference in the chemical potential between
the product (final goal) system and the initial (starting) system. Consequently,
affinity is a thermodynamic quantity and is independent of the path from the
start to the product.
Suppose that a penetrant in a solution is diffusing into a polymer solid. The
affinity -Ape is given by
- A b = RT In([P],/[P],)
where R is the gas constant, T the temperature in kelvins, and [P], and [P], are
the concentrations of the penetrant in the polymer solid andin the solution,
respectively. Equation (12) is derived under the assumption that the penetrant is
soluble in both phases and the solutions are ideal solutions. This equation is
SUPERMOLECULAR
STRUCTURE OF POLYMER SOLIDS 243
transformed into
[P],/[P], = K = exp(-ApdRT) (13)
where the constant K is the partition coefficient and is taken to be independent
of the concentration of the penetrant. In thespecial case when the concentration
of the penetrants is located on the surface of the solid, the relationship between
[P], and [P], becomes the adsorption isotherm. On the other hand, for the case
when the penetrant molecule is in the gas phase, [P], should be changed to the
concentration of the penetrant in the gas phase (i.e., partial pressure, denoted as
[PIg). Then, in the case of penetrant transport from the gas phase to a polymer
solid, Eq. holds if K is replaced by k defined by
[PlF./[Pl, = k (14)
The constant k is the solubility coefficient. The value of k increases with the
boiling point of the penetrant.
Suppose a penetrant in a solution diffuses into a polymer solid through a
capillary (the capillary model) and dissolves into the polymer solid with the
partition coefficient K. Then the apparent diffusion coefficientD , is theoretically
given by
D, = ( 4 K ) D o + D, (15)
where D,, is the diffusion coefficient of the penetrant in the medium without the
polymer solid, is the weight of the medium in the capillaries per unit weight
of the polymer solid, and D,, is the diffusion coefficient of the penetrant in the
solid. The value of D , decreases with anincrease in K.
If a penetrant in a solution diffuses into a polymer solid through the free-
volume model, the value of K contributes to the permeation coefficient P as
follows:
P=DK (16)
In this case, the affinity works only on K , resulting in a strong contribution to
the penetrant transport. Here, the driving force of the transport is regarded to
be the transmembrane pressure originating from the difference in concentration
of the penetrant between the two sides of a membrane (i.e., the osmotic pres-
sure). This type of permeation is called the dissolution/diffusion mechanism.
A gas moleculealso moves through a polymersolid by mechanisms of either
the capillary model or the free-volume model depending on the size of the
molecule and that of the pore in the solid. When the boiling point of a molecule
increases andwhenthe pore size is less than 20 nmand larger than4 nm,
surface diffusional transport dominates (Kamide et al., 1983). In this type of
transport, the adsorptionandtwo-dimensionaldiffusioncoefficientshould be
taken into account. Affinity works mainly on the adsorption. When thepore size
244 " B E
is less than 2 MI, the dissolution/diffusion mechanism becomes the dominant

factor for transport and the permeability coefficient is expressed by
P = Dk (17)
To increase the permeability coefficient of a gas molecule, a bulky side chain
is induced into aromatic polymer solids such as polyimide(Okamoto et al.,
1993), polyester (Sheu and Chern, 1989), polycarbonate (Hellums et al., 1992),
polysulfone (Aitkekn et al., 1990), and polyphenylene ether (Aguilar-Vega and
Paul, 1993), whose glass transition temperatures are high. The side chain in-
creases the free volume, providing additional space for thediffusion of gas
molecules.
C. Effect of Chain Conformation and Chain Configuration

on Penetrant Transport
The chain configuration indicates the steric regularity of the chemical bond, i.e.,
tacticity here. This structural characteristic belongs to the first-order structural
factor of Table 1. Chain conformation is the shape of the chain molecule, for
example, a random coil, a helix, thep-structure in polypeptide, andplane zigzag,
etc. In general, the configuration governs the conformation. The regular form of
chainconformationandconfiguration results inacrystal or aliquidcrystal.
Penetrant transport is influenced by the existence of these types of crystals as
mentioned in Section 11, and an increase in the regularity of the chain confor-
mation and the chain configuration gives rise to a decrease in penetrant transport.
When the conformationofthechain molecule constituting amembrane
changes, then the transport properties of the membrane change. For example, a
polypeptide membrane shows electrical pulses under the concentration gradient
of a salt. This is due to a conformational change between a helix and a coil,
which depends on the concentration of the salt (Minoura et al.,1993). The
conformational changeof a protein molecule may play a very important role in
ion transport across a biomembrane.
V. EFFECTS OFFINESTRUCTURE(CRYSTALLINITY,
ORIENTATION, ETC.) OF A POLYMER SOLID ON
PERMEATION PROPERTIES
The fine structure treated in this section belongs to the third-order structural
factor in Table 1. The structural factor includes the characteristics of a crystal,
such a crystal size, crystallinity, crystal orientation, and crystal shape, the dis-
persion state of crystalline and noncrystalline regions, and the characteristics
representing the interfacial aggregation structure.
The adsorption inside a crystal region and diffusion through a crystal are
considered to be negligible. The effect of fine structure on the permeation prop-
SUPERMOLECULAR
SOLIDS 245
erties can be described through the noncrystalline region including crystal sur-
faces. Since the contribution of the monolayer on the surface to the total volume
of the noncrystalline region is usually small, the solubility coefficients k and K
are expressed as
k = (1 - x)k, K = (1 - )OK, (18)
where x is crystallinity and k, and K,are the values of k and K,respectively, in
the noncrystalline regions. We note that the solubility coefficient is defined at
thermodynamic equilibrium. In contrast, we apply this value to rate processes
such as permeation under the assumption that Eq. (18) may hold approximately
only on the surface of the polymer solid where the diffusion and permeability
coefficients are small. The value of x should be the value at the surface. In the
case of the capillary model, the equilibrium is assumed to hold over the entire
surface of the capillary. As for the diffusion coefficient, the existence of crystals
affects the transport in a complicated manner. For example, crystals in a non-
crystalline region give the molecular chains in the region some kind of strain
that results in immobilizing the chains. This effect is called the immobilization
x
factor g. Both the crystallinity and the size of a crystal hinder the transport
of a penetrant, and the path of the penetrant becomes tortuous. The degree of
tortuosity varies with the change in the dispersion state of a crystalline region
even in the same valuesof x and the crystal size. This tortuosity factor y should
be used to correct the diffusion coefficient, The diffusion coefficient D for a
crystalline polymer solid is represented by (Michaels and Parker, 1959)
D = D,/Py (19)
where D, is the diffusion coefficient of a penetrant in a noncrystalline region.
A. Effect of Draw Ratio on Permeability

There are two types of molecular chain orientations, those in acrystalline region
and those in a noncrystalline region. The orientation of the crystalline part may
contribute to penetrant transport through the variation in the value of y and can
give rise to an anisotropic diffusion coefficient.On the other hand, the molecular
orientation in a noncrystalline region causes not only the anisotropic feature in
the penetrant transport but also the change in the region of diffusion for the
penetrant. The increase in the degree of orientation gives rise to an increase in
the regularity in a noncrystalline region and also to an increase in the number
the extended chain molecules through which most molecules do not diffuse.
These changes may reflect the change in the value in (19).
The drawing of a polymer solid produces both increases in the crystallinity
and increases in the degree of the orientation of a molecular chain. According
to the results reported by Michaels et al. (1964), the permeability coefficient P
246 MANABE
of a gas molecule decreases with an increase in the draw ratio, but when the
drawing approaches the breaking point, P increases abruptly.
B. Effects of Heat Treatment on Permeability

The effect of heat treatment results in a change in the supermolecular structure
and a change in the permeability of a penetrant. Heat treatment is usually per-
formed under three typical conditions:ina system free of tension,constant
length, and constant tension. In the crystalline region, the treatment results in
(1) an increase in crystallinity, (2) an increase in crystal size, an increase in
the perfection of the crystal, and a decrease in crystal orientation. The treat-
ment condition of the free of tension preferentially brings about change 4. The
condition of constant tension promotes changes and a minimum change in
crystal orientation. Constant length in one and/or two dimensions is often em-
ployed, and heat treatment then produces changes all four types of changes.
In the noncrystalline region, the treatment gives rise to relaxation of the
remaining strain, (2) a decrease in the size of the region, change in the mo-
lecular chain aggregation into a closer packing, and a decrease inorientation.
The treatment free of tension may show all four types of changes most effec-
tively, and treatment under constant tension may work on changes 2 and
In general, heat treatment changes the supermolecular structure as to de-
crease the transport rate of a penetrant. For example, even in the case when the
temperature of the heat treatment is far below T,, the permeability decreases and
the permselectivity increases in polyimide thinfilms (Pfromm and Kovos,
Figure 17 shows the dependence of the dye exhaustion by polyethylene ter-
ephthalate fibers spun at 5, 6, 7, and km/min on the temperature of the
heat treatment T, for 1 S at constant length (Kuriki et al., The dye mol-
ecule employed as a disperse dye of 1,5-diamino-2-bromo-4,8-dihydroxyanthra-
quinone with a molecular weight of The dye exhaustion shows a slight
minimum at ca. 150°C and then increases with the elevation of T,. The heat
treatment results in the relaxation of the remaining strain and a decrease in the
orientation of the noncrystalline region, and these changes produce loosely and
relaxed chain molecules in the diffusible amorphous region of the dye molecule,
resulting in a change opposite to the closer packing mentioned for the noncrys-
talline region.
C. Effects of Solvent Treatment on Permeability

The prerequisite of the treatment is a solvent molecule that can diffuse readily
into a polymer solid. The diffusional region can be evaluated through measure-
ment of the dynamic viscoelasticities of the solid under the coexisting solvent
(Fujioka and Manabe, The solvent molecule accelerates the thermal mo-
tion of the polymer segment, resulting in effects similar to those of heat treat-
Or 150 200 250

Annealing temperaturelOC
17 Change in exhaustion of adispersedyewithannealingtemperature for
polyethylene terephthalate fibers spun at various speeds ranging between 4 and 9 km/
min (Kuriki et al. 1985). The numbers in this figure are the take-up speed in kilometers
per minute. The dye employed was Resolin Blue FBL,and the annealing time was 1 S.
ment. We can expect the following effects from solvent treatment: (1)an in-
crease in crystallinity, (2) an increase in crystal size, (3) an increase in crystal
perfection, (4) a decrease in the orientation of the crystalline and noncrystalline
regions, (5) the relaxation of the remaining strain in a polymer chain, (6)closer
packing of molecular chain aggregation, (7) the dissolving out of a part of the
component, (8) swelling by the remaining solvent, and (9) the generation of
interfacial molecular orientation. Although these effects influence the penetrant
transport, very few workers have succeeded in making a quantitative evaluation
of these effects separately.
Treatment of an acetate film with boiling water causes a loss in luster. This
phenomenon is the result of effect 7, and the dissolving component is a low
molecular weight component with small amountsof acetyl groups (Kamide and
Manabe, 1978). The water molecule develops the local heterogeneity of the
packing density of chain molecules. This type of change accelerates the rate of
transport of the dispersed dye molecule. The increase in heterogeneity in the
molecular chain aggregation brought about by solvent treatment has been ob-
served in polymerblends (Manabe et al., 1969). Figure 18 shows the temperature
dependence of the dynamic viscoelasticity of a polymer blend of nitrile buta-
diene rubberandpolyvinyl chloride after treatmentwith a solvent that acts
selectively for one component in the polymer blend. The increase in the half-
width value of the loss modulus E" versus temperature curve indicates an in-
crease in the heterogeneity in the aggregation of component polymers. Solvent
248 UANABE
-21 '
-50 0 50
Temperature I0C
-50 0 50
Tempetat we l OC
Temperature dependence of dynamic modulus E' and loss modulus E" for
polymer blend of polyvinyl chloride/nitrile butadiene rubber(PVCN3R) after treatment
withsolvent(Manabeetal.,1969).The PVC/NBR ratio = 77/23byweight. (a)The
solventisdimethylformamide,anditscontentis 6 and 14 wt %. @)The solventis
chloroform, and its content is 7 and 20 %.
treatment against a porous membrane can impart a composition gradient to the

membrane (Aptel and Cabasso, 1980).
When polypropylene film is treated with p-xylene at various temperatures,
the permeability coefficients for toluene, methylcyclohexane, and isooctane
through the film increase with the treatment temperature (Michaels et al., 1969).
The permeability coefficients of toluene, methylcyclohexane, and isooctane be-
fore treatment in one case were 15.3 lo-',14.3 and 1.4 lo-' g
cm/@ cm*),respectively. After the treatment at 100°C, and coefficients were
59.4 lo-', 46.2 lo-', and 21.2 g cm/(h cm2).
VI. CONCLUDINGREMARKS
The supermolecular structure of a polymer solid is determined in two ways by
the chemical structure (the first-order structural factor) and the preparation con-
ditions. Penetrant transport is governed by three functional factors of the super-
molecular structure, the thermal motion of polymer chains, and the intermolec-
ular interaction between the penetrant and the polymer chain. Because of the
structural complexity and heterogeneity of the solid, it is very difficult to obtain
a complete view of the structure. As for the relation to penetrant transport, the
evaluation of the second- and third-order structural factors, that is, the charac-
terization of the supermolecular structure on a nanometer scale, is important.
The methods of novel microscopies such as scanning probemicroscopy (SPM),
atomic force microscopy, friction force microscopy, and magnetic force micros-
copy are very powerful tools for investigating supermolecular structure.
correlate supermolecular structure and penetrant transport, we must em-
ployan adequate transportmodel. Although the free-volume modelandthe
capillary model and their combinations have been proposed far, some mod-
ifications or the use of new models that take supermolecular structure into ac-
count are needed. Combined with advanced theory, a novel preparation method
for apolymersolidguidedbytheoryshouldbedevelopedinthe future for
preparing solids with the desired penetrant transport properties.
REFERENCES
Aguilar-Vega, W., and D.R. Paul (1993). J. Polymer Sci., Polym. Phys. Ed., 31, 1577.
Aitkekn, C. L., D. R. Paul,andW. J. Kovos (1990). ICOM '90, Preprint, p. 821.
Aptel, P., and I. Cabasso (1980). J. Appl. Polymer Sci., 25, 1969.
Fujioka, R., and Manabe (1995). Polym. Prep. Jpn., 44(5), 759.
Haraguchi, H., T. Kajiyama, and M. Takayanagi (1976). Rep. Prog. Phys. Jpn., 16,303.
Hellums, K. W.,W. J. Kovos, and J. C. Schmidhauser (1992). J. Membrane Sci., 67, 75.
Iijima, H.,and Manabe (1983). Japanese Patent, Tokkai Sho 58-183907.
Kajiyama, T., K Tanaka, I.Ohki, S. R. Ge, S. Yoon, and A. Takahara (1995), Mac-
romolecules, 27, 7932.
Kamide, K., and Manabe (1978). Sen-i Gakkaishi, 34, T53.
Kamide, K., and S. Manabe (1980). In Ultrafiltration Membranesand Applications, A. R.
Cooper, Ed., Plenum, New York, p. 173.
Kamide, K.,and Manabe (1985). In Materials Science ofsynthetic Membranes, D.R.
Lloyd, Ed., American Chemical Society, Washington,DC, p. 197.
Kamide, K.,and S. Manabe (1986). Polymer J., 167.
Kamide, K., S. Manabe, T. Matsui,T.Sakamoto,and Kajita (1977). Kobunshi Ron-
bunshu, 34, 205.
Kamide, K., Manabe, T. Nohmi, H. Makino, H. Marita, and T. Kawai (1983). Polymer
J., 15, 179.
Kamide, K, Nakamura, T. Akedo, and Manabe (1989). Polymer J., 21, 241.
250 UANABE
Kawaguchi, T. (1962). InKoubunshi no Bussei, T. Kawai, T. Kawaguchi, and A. Miyake,

Eds., Kagakudoujin, Tokyo, p. 141.
Kesting, R. E. (1985). Synthetic Polymeric Membranes, Wiley, New York.
Kesting, R. E., and A. Manefee (1969). Kolloid 230, 241.
Kuriki, T., K. Kamide, S. Manabe, and M. Iwata (1985). Sen-i Gakkaishi, 38, T85.
Manabe, S. (1992). InAnimal Cell Technology: BasicandApplied Aspects,H. Murakami,
S. Shirahata, and H. Tachibana, Eds., Kluwer, Boston, p. 87.
Manabe, S. (1993). Sen-i Gakkaishi, 49, 195.
Manabe, S., and K. Kamide. (1984). Polymer J., 16, 375.
Manabe, S., and M. Takayanagi (1970a). Kogyo Kagaku Zasshi, 73, 1572.
Manabe, S., and M. Takayanagi (1970b). Kogyo Kagaku Zasshi, 73, 1581.
Manabe, S., and M. Takayanagi (1970~).Kogyo Kagaku Zasshi, 73, 1595.
Manabe, S., R. Murakami, and M. Takayanagi (1969). Mem. Fac. Eng. Kyushu Univ.,
28, 295.
Manabe, S., S. Minami, and M. Takayanagi (1970a). Kogyo Kagaku Zasshi, 73, 1576.
Manabe, S., H. Nakamura, S. Uemura, and M. Takayanagi (1970b). Kogyo Kagaku Zas-
shi, 73, 1587.
Manabe, S., E. Osafune, and K. Kamide (1973). Japanese Patent Sho 48-25055.
Manabe, S., S. Kajita, and K. Kamide (1980). Sen-i Kikai Gakkai, 33, T93.
Manabe, S., Y. Kamata, and K. Kamide (1985). Polymer J., 17, 775.
Manabe, S., Y. Kamata, H. Iijima, and K. Kamide (1987). Polymer J., 19, 391.
Mark, H. F., S. M. Atlas, and E. Cernia, Eds. (1967). Man Made Fibers, Wiley,New
York.
Michaels, A. S., and R. B. Parker, Jr. (1959). J. Polymer Sci., 41, 53.
Michaels, A. S., W. R. Vieth, and H. J. Bixter (1964). J. Appl. Polym. Sci., 8, 2735.
Michaels, A. S., W. R. Vieth, A. S. Hoffman, and H. H. Alcalay (1969).J. Appl. Polym.
Sci., 13, 577.
Minoura, N., S. Aiba, and Y. Fujiwara (1993). J. Am. Chem. Soc., 115, 5902.
Nakayama,C.,K.Kamide, S. Manabe, and T. Sakamoto (1977). Sen-i Gakkaishi, 33,
T139.
Nohmi, T., S. Manabe, K. Kamide, and T. Kawai (1978). Kobunshi Ronbunshu,
Okamoto, K., N. Umeo, S. Okamyo, K. Tanaka, and H. Kita (1993). Chem. Lett., 225.
Peterlin, A. (1969). J. Polym. Sci., A-2, 7, 1151.
Pfromm, P. H., and W. J. Kovos (1993). ICOM '93, p. 213.
Robeson, L.M. (1969). Polym. Eng. Sci., 9, 277.
Schneider, N. S., C. R. Desper, and J. J. Beres (1978). In Liquid Crystalline Order Poly-
mers, A. Blumstein, Ed., Academic, New York, p. 299.
Sheu, F. R., and R. T. Chern (1989). J. Polym. ScL Polym. Phys. Ed., 27, 1121.
Takayanagi, M. (1967). Pure Appl. Chem., 65, 555.
Takeda, H.,and Y. Nukushina (1963). J. Chem. Soc. Jpn. Ind. Chem. Sect., 68, 1241.
T S U N ~ ~ T.,
, N.Osawa,H.Hitaka, T. Hirasaki,K.Yamaguchi, S. Manabe,and T.
Yamashiki (1990a). Polym. J., 22, 751.
Tsurumi,T., T. Sato, N. Osawa, H. Hiraka, T. Hirasaki, K. Yamaguchi, Y. Hamamoto,
S. Manabe, T. Yamashiki, and N. Yamamoto (1990b). Polym. J., 22, 1085.
Washizu, S., I. Terada, T. Kajiyama, and M. Takayanagi (1984). Polymer J., 16, 307.
Yeh, G. S. Y. (1972). J. Macromol. Sci., B6, 465.
Translational Dynamics of
Macromolecules in Melts
Peter F. Green
Sandia National Laboratories
Albuquerque, New Mexico
INTRODUCTION
A flexible polymer chain is capable of assuming an enormous number of pos-
sible configurations by rotations of chemical bonds. Such local motions, which
occur on time scales on the order of lo-" S, facilitate the translational motions
of single chains in melts. Their effect on the overall dynamics of a macromol-
ecule can be accounted for by a mean monomeric frictional coefficient, The
time scale its translational diffusivity is determined primarily by N,the num-
ber of monomer segments that compose a single chain, and 5. The intent this
chapter is to discuss the translational center-of-mass motions that chains in poly-
mer melts undergo in response to thermal agitation.
A remarkable feature of the dynamics of long polymer chains is an important
interaction that arises from the fact that these one-dimensional objects cannot
cross each other. The topological constraints, known as entanglements, have a
profound impact on chain dynamics. Because of the entanglement effect, poly-
mer liquids exhibit ratherremarkable behavior that distinguishes themfrom
other materials (Ferry, 1980). One important feature can be seen inthetime
dependence of the stress relaxation of a melt that has been subjected to a strain.
A sudden imposition of a shear strain, y, that displaces the chains in a melt
of linear chains from their equilibrium configurations results in a time-dependent
relaxation of the system as the chains return toequilibrium through some
Brownian process. It is convenient to describe the time dependence of such a
251
252 GREEN
process by defining a shear-stress relaxation modulus, G(t) = u(t)/y, where u(t)

is the time-dependent shear stress thatreflects the structure and interactions
between the chains in the system (Ferry, Graessley, Figure
depicts the time dependence of the shear modulus. At short times, the reduc-
tion of G(t) results from the relaxation of stresses along portions of the individ-
ual chains (note that such a process is characterized by a distribution of relax-
ation times). At intermediate times, for chains that are longer than an effective
number of monomers, Ne, where Ne is the number of monomers betweenentan-
glements, G(t) becomesindependent of t, the plateauregion. This so-called
plateau region, characterized by a modulus G:, can be rationalized only by the
presence of “entanglements.” For the most part, G$) is relatively insensitive to
temperature. It is a property of the melt and is determined by structure. While
G:) is independent of N , the onset of the terminal region is a strong function
of N. This will be addressed later. At long times, in the terminal region, G(t)
rapidly relaxes tozero. Here the chains assume equilibriumconformations
through a translational (Brownian) process.
Properties such as the viscosity are also affectedby entanglements. For chains
shorter than a critical number of segments, N,, where N, is a material-dependent
property(empirically N, W,),the viscosity in thelimit of zero shear, qo -
-
N , while for N > N,, qo ~ 3 . 4 ’ 0 . 2 (Berry and Fox, Figure 2 shows a plot
of the molecular weight dependence of the viscosity of polyethylene. It is im-
portant to note that for N N,, qois observed to vary as N after corrections are
made to achieve a constant free volume state. The triangles represent the un-
I /
Log t
Schematic thetimedependence of thestressrelaxationmodulus, G(t),
a melt. C: is the plateau modulus, and M and M* are the molecular weights of the
melt.
TRANSLATIONALD W M I C S IN MELTS 253
14
12
10
-2
2 3 4 6 8
log
2 Molecularweightdependenceof the zeroshearrateviscosityofpolyeth-
ylene. For N > N,,the viscosity varies as N3.4.
For N < N,, it varies as N (0) after the
raw data (A)are adjusted for chain end effects. This behavior is typical. [Data
of Colby
et al. (1987).]
corrected data. The N3.4dependence remains unchanged for N > N,. This be-
havior is typical.
The foregoingdiscussion provides the framework withinwhich we can begin
to understand translational dynamics in polymer melts or concentrated systems.
Because the large number of configurations that a polymer chain canassume,
the shape of a single polymer chain can be described statistically 1969).
Consequently, a description the dynamical process necessarily exploits this
simplification. In an attempt to understand the dynamics of chains in solution,
254 GREEN
Rouse(1953)arguedthat a chain could be considered as a series of beads

connected by springs. The dynamics of such a chain could be modeled by the
Brownian motion of beads whose positions are described by a position vector
RN.Each bead is assumed to experience an average frictional force, 5. As shown
later, what ultimately emerges from such a picture is that the translational dif-
-
fusion coefficient DRo 5 8 ” and that the longest relaxation time of the chain
T~~- The Rouse model failed miserably at describing translational dynam-
ics in dilute solution primarily because it neglected the effects of intramolecular
hydrodynamic and excluded volume interactions. Interestingly, as shown later,
it provides the basis for understanding translational dynamics in melts where
long-range hydrodynamic and excluded volume interactions are assumed to be
screened and where each chain can be described by ideal Gaussian statistics.
Before deGennes introduced reptation (deGennes, 1971, 1979) just over 20
years ago, a number of unsuccessful attempts (Bueche, 1968) had been made
to understand the dynamics of entangled melts. In reptation, a polymer chain is
imagined to undergo one-dimensional translational motion in the presence of a
fixed network of obstacles. The motion is facilitated by the flow of a series of
“defects” (kinks) along the contour of thechain. The chain translates in a
manner that simulates the motion of a snake. Reptation was later introduced by
Doi and Edwards (1978a,b, 1986) that a constitutive equation could be de-
veloped to understand the behavior of true melts. They introduced the concept
of a tube in order to account for the effect of entanglements.
It was noted that although permanent cross-links exist in elastomers, they are
not present in linear polymer systems. Therefore, although one might introduce
the concept of the tube in linear polymer melts, it has to be acknowledged that
the topology of such a tube is time-dependent. We show later that there is a
regime of molecular weights over which predictions based on the chain under-
going Brownian motion in a fixed “tube” are in agreement with experiment.
Complications that result from the time-dependent topology of the tube can be
dealt with mathematically, andexperimental data on diffusion and viscoelasticity
support the new predictions. One might emphasize at this point that calculations
the viscoelastic properties of polymers based on this picture, while qualita-
tively correct, do have shortcomings.
In this chapter we begin with a detaileddescriptionof the Rouse model
because it provides the basis for subsequent discussions. This is followed with
a discussion of reptation of linear chains in a fixed tube and a subsequent dis-
cussion of the effects of the time dependence of the topology of a single chain
as it diffuses. This provides the context within which we can begin to describe
the Brownian motion of chains of different architectures (rings, stars). dis-
cussion of experimental tests and the extent to which the results support the
predictions follow. Finally, a discussion of interdifhsion in two-component mis-
cible blends is presented. Here the concept of the Flory interaction parameter
TRANSLATIONAL DWAMICS
MELTS IN 255
x, a measure of the thermodynamic interactions between the components in the

blend, is introduced. It is shown that the time scales of interdiffusion are deter-
minednot only by N and &,, but also by x. When x < 0, the interdiffusion
coefficient, D , is enhanced overthe value of the tracer diffusion coefficient,D*,
of a single chain. This is known as a thermodynamic “acceleration” of D.
Similarly, when x > 0, a thermodynamic “slowing down” of D is realized. We
conclude with a brief discussion of new developments on diffusion in block
copolymer systems and some recent theoretical advances.
II. TRANSLATIONAL DYNAMICS IN HOMOPOLYMERMELTS

A. Dynamics of Unentangled Chains: Rouse Model
The Rouse model is expected to provide a reasonable description of the trans-
lational dynamics of unentangled chains in melts p o i and Edwards, 1986). This
is primarily because itneglects excluded volumeand hydrodynamic interactions,
both of which are assumed to be screened in melts. The central assumption in
the Rouse model is thatthe dynamics of a chain are governed by localized
interactions along the chain. Each chain is imagined to be divided into a series
of submolecules, each composed of a number of monomers. Each submolecule
is represented by a bead, and thebeads are connected by Hookeansprings, where
each spring constant is 3kBT/b2;b is the separation between beads. The position
of each bead is determined by coordinates {Rl, R2, R3,. .. ,RN},and each bead
experiences an average frictional drag 5. The motion of the beads can be de-
scribed by a linearized Langevin equation. In essence, therefore, we now have
an equation thatdescribes the Brownian motion ofa series of coupled oscillators.
A transformation is subsequently made that the system can now be described
in terms of normal mode coordinates where each mode behaves independently
of the other. The equation is solved to yield a spectrum of N relaxation times,
each relaxation time, T,.,corresponding to one normal mode. The essential pre-
dictions follow.
The diffusion coefficient of the center of mass is
DRo= k,T(N{)” (1)
One might recognize that q.(1) is essentially the Einstein relation if the pa-
rameter N( is identified as the effective friction coefficient of the chain in its
environment.
The time correlation function of the end-to-end vector is
256 GREEN
where P(t) = R,.,(f)-RN(0) and r1is the relaxation time of the first mode (note
that there are p modes). The relaxation time of the first mode is the longest
relaxation time, r l , and is given by
=
rRo = (b2/3~2k,T)
yV2 (3)
The viscosity in the limit of zero shear is predicted to be
rlo - N5 (4)
As mentioned earlier, this model is expected to provide a reasonable description
of the dynamics of unentangled chains in the regime N < N,.
Shown in Fig. is a schematic of a labeled chain composed of N segments in

a meltsurroundedby other chains, where each ofthe surrounding chains is
composed of P segments. For the initial purposes of discussion, the length of
the P-mer chains is considered to be infinite. Therefore theN-mer chain is
assumed to undergo Brownian motion effectively in a fixed “tube.” The motion
of the chain occurs by the propagation of stored length, or kinks, as depicted
in Fig. 4a (de Gennes, 1979). Curvilinear motion of the chain occurs along an
average trajectory that Edwards called the primitive path (Fig. 4b). Edwards
defined the primitive path as the shortest distance connecting the ends of the
chain that has the same topology as the chain p o i and Edwards, 1980).
There are effectively two time scales in this problem. At short time scales,
the dynamics are controlled by rapid motions (propagation of kinks) around the
3 (a) Schematic of a labeled chain of degree of polymerization N in the

presence of a host of chains of degree of polymerization P. (b) Schematic of the labeled
chain restricted to a tube-like region.
TRANSLATIONALDW'ICS IN MELTS 257
ii)
--- """""-- ---*
b)
+
\
PRIMITIVE
PATH
(a) A schematicdemonstratingthemechanism of diffusionfor a linear
flexible chain. (b) A depiction of the chain moving along its primitive path.
primitive path, and at longer time scales they are controlled by the translational
motion the chain, which results in changes in the primitive path. the chain
diffuses in random directions it creates and destroys the ends of the primitive
path. The translational motion of the chain is characterized by a time scale q,
over which the chain loses complete memory of its original tube, or, equiva-
lently, the original primitive path. In the pure reptation calculation where the
chain is imagined to diffuse in a fixed environment (tube), fluctuations in the
primitive path are ignored; i.e., it is considered to be a constant contour length
L. One can define the number steps on the primitive path as Z; hence L =
Zu, where a is the length of a step along the path. The diffusive motions along
258 GREEN
the path are assumed to be well described by the Rouse model, and therefore
Eq. (1) provides an adequate description of the motion of the chain in the tube.
It is further assumed that the conformation of the primitive path remains Gauss-
ian, at least at large length scales. The mean square end-to-end distance of the
primitive chain (the dynamical equivalent of the primitive path) should be the
same as that of the Rouse chain, it follows that
L = Nb2/a (5)
One can define the end-to-end vector of the chain as P(t) = R(L, t) - R(0, t).
The time-dependent correlation function of the end-to-end vector is given by
(P(t) P(0)). It is proportional to +(t), the probability that a segment of the tube
that was originally occupied at time t = 0 remains occupied at time t. Here (Doi
and Edwards, 1986),
(W)- W)) = (6)
and
The longest relaxation time, rd, is calculated as

?d = (b4/~2kBTa2)
In contrast to the longest Rouse relaxation time, the reptation relaxation time,
?d, varies as N3 as opposed to NZ. It is interesting to note that because of the
effect of entanglements, the actual relaxation time for a very long chain is re-
duced by a factor of 32.
It is interesting to note that the tube disengagement time, ?d, could be cal-
culated from a simple scaling relation by noting that
L2/& (9)
which, using Eq. (5), yields
7d (b4/kBTa2) (10)
The diffusion coefficient of a chain moving by a reptation process could be
derived by noting that during the time interval ?d the center of mass of the chain
should have moved a distance equivalent to the radius of gyration of the chain,
N1”u,
DRcp= NU2/7d (11)
Therefore,
D , = (kBTu2/3b2()
NZ (12)
LTS INTRANSLATIONAL
DI%M."IC 259
equivalent form of this equation may be realized by recognizing that the

tube diameter U can be written as (Graessley, 1980; Graessley and Edwards,
1981)
= (4/5)Nbz(M/Me) (13)
where M = MOA' (Mo is the molecular weight of a monomer on the chain and
M, is the molecular weight between entanglements). It follows that
D , = Do"' (14)
where D, = (4/15)M0M,kBT/5.A rigorous derivation of this result may be found
elsewhere (Doi and Edwards, 1986). Equation (14) is a very important result
that predicts that the center-of-mass diffusion coefficient of a flexible polymer
chain diffusing in a highly entangled melt varies as M-', independent of the
length of the P-mer chains in the host environment(recall the initial assumption
that P >>
We now use the above results to calculate qo that we can show howDRep
can beexpressed in terms of viscoelastic parameters (Graessley, 1980). The
viscosity can be obtained from the relationship
= G: W) dt
Upon integrating this expression one arrives at
This result shows that the viscosity is determined by the longest relaxation time
[To GN(O)'T~].
It fOllOWS that
We show later that upon substitution of the appropriate viscoelastic parameters,

this result yields very reasonable predictions for the magnitude of DRepfor dif-
ferent polymers.
C. Measurement of Diffusion in Polymer Melts
At this point it is important to differentiate between tracer diffusion, D*, and

self-diffusion, D,; in a later section we address mutual diffusion. The tracer or
self-diffusion coefficient of a species A can be defined in terms of the mean
260 GREEN
square displacementof the center of mass

1
D = lim - ([R(r) - R(O)]*) (18)
6t
where R is thepositionof the center of mass ofthe diffusant and r is the
diffusion time. In reality, the tracer diffusion coefficient is realized when the
molecule has translated a distance that is on the order of a few Rg; at smaller
length scales D may have contributions from segmental displacements. In the
case of tracer diffusion, one monitors the displacement of a single labeled mol-
ecule executing Brownian motion in a given environment. Self-diffusion cor-
responds to the situation in which the motion of a molecule is monitored in an
environment of identical molecules. The driving force for tracer diffusion and
self-diffusion is entropic in origin, unlike mutual diffusion, which is driven by
gradients in chemical potential. Spica1 diffusivities in polymer melts range from
to cm2/s,which means that one needs experimentaltechniqueswith
sufficientlygood depth resolution to measuresmall diffusion coefficientsin
reasonable time scales. As an illustration of this point, one might consider at-
tempting to measureadiffusioncoefficient of a molecule thatdiffused for
one day. If the diffusion coefficient is lo-’ cm2/s, then the mean displacement
is 2.3 cm.On the other hand, if D* = cm2/s,then the mean displacement
is only 207 A.
It is reasonable to state that any technique that can be used to determine the
concentration profile of a diffusant after some specified time interval can be
used to determine a diffusion coefficient. This requires labeling the diffusant in
such a manner that the label does not affect the mobility of the diffusant. An
alternative approach might involve monitoring a time-dependent process that
reflects the motion of the diffusant. Techniques that have been used to measure
diffusion in polymer melts include radiolabeling techniques (Crank, 1975), in-
frared microdensitometry (IRM) (Klein and Briscoe, 1979), ion beam analysis
techniques [Rutherford backscattering spectrometry(RBS)(Chu et al., 1978;
GreenandDoyle, 1990; Greenetal.,1985),forward recoil spectrometry
(FRES), and elastic recoil detection (ERD)] (Mills et al., 1984; Green et al.,
1986; Green and Doyle; 1990), nuclear reaction analysis (NRA) (Chaturvedi,
1990),fluorescencerecovery after patternphotobleaching (FRAPP) (Smith,
1982; Smith et al., 1984), forced Rayleigh scattering (FRS) (Antonietti et al.,
1984), small-angle neutron scattering ( S A N S ) (Bartels et al., 1984, 1986), and
nuclear magnetic resonance(NMR) (Fleisher, 1983, 1984, 1992;von Meerwall,
1991; McCall et al.,1986). The range of techniques is notlimited to those
mentioned, but those are among the most commonly used. We now briefly dis-
cuss the essential features of these techniques and point out some of their more
important limitations. The interested reader is referred to the original references
for further details.
TRANSLATIONAL D W M I C S IN MELTS 261
The review that follows is necessarily brief. Of these techniques, the radio-
tracer labeling technique, which involves determining the concentration profile
of a radioactively labeled species, is the oldest. At best its depth resolution is
on the order of micrometers, and it would not be well suited for measuring the
slower dmsivities. The ion beam analysis techniques, andIRM, are
techniques that rely on labeling the diffusant in order to determine D. Further,
they rely on the time evolution of concentration gradients.
These RBS, ERD, NRA, and IRM techniques yield concentration profiles
that can then be compared with the theoretically predicted profiles extract D*
or D,. In RBS, the energy of an incident monoenergetic beam of ions (most
diffusion experiments in polymers have been performed using helium ions) is
measured after the ions are backscattered from a sample. The energy and yield
of the backscattered particles can be analyzed to obtain the concentration versus
depth profile of the appropriate species in the target. In ERD, an incident beam
of particles causes nuclei to recoil from the target. analysis of the yield and
energy of the appropriate recoils gives the desired concentration versus depth
profile. In NRA, the natureandenergyof the incident particles is carefully
chosen that when these particles impinge on the appropriate nucleus a nuclear
reaction will occur. The yield and energy of the nuclear products can be con-
verted to the concentration versus depth profiles. The ion beam analysis tech-
niques, in general, have excellent depth resolution, 200-800 & and are ideally
suited for measuring midrange to very slow diffusion coefficients (10-10-10”6
Cm”S).
In IRM, a sample that is initially composed of a bilayer of two chemically
distinct species (one could be deuterated) that are allowed to interdiffuse at a
chosen temperature is microtomed into layers perpendicular to the diffusion
direction and analyzed using infrared spectroscopy. The concentration profile
from which the diffusion coefficient is extracted is then constructed with the
information. The depth resolution of this technique is on the order of micro-
meters. Hence the IRM is not well suited to study slowly diffusing species.
(D*s > lo-’ cm’/s.)
exploits the strong coherent scattering contrast between deuterated and
hydrogenated polymers. The samples are multilayered, and the coherent scat-
tering intensity is measured only when the layers have interdiffused. The time
dependence of the coherent scattering intensity yields information about the
diffusion process. is capable of measuring D$ in the range of 10-8-10”6
cm2/s.
While FRAPP and FRS require labeling of the appropriate species, they do
not yield directly the concentration profiles of the d&sant. In FRAPP, fluoro-
phores in a sample are bleached in a spatially periodic pattern by an intense
laser beam. The recovery of the fluorescence in the bleached region is monitored
by a less intense beam as unbleached molecules diffuse in that region. The time
GREEN
dependence of this process yields D. The concept of FRS is similar to FR4PP

except that a photochromatic dye is used in place of the fluorophore. In this
technique two intersecting laser beams producespatially periodic regions of high
and low intensity to create a grid in the sample, a result of the photoisomeri-
zation (a reversible process). The time dependence of the decay intensity of
an incident laser beam diffracted from the grating yields D.
NMR, unlike the other techniques, does not depend on labeling or on con-
centration gradients. It depends on the local environment of a nucleus in the
presence of an inhomogeneous magnetic field. Pulsed field gradient NMR has
been used for the past 30 years to measure diffusion coefficients. Here a se-
quence of radio-frequency pulses results in a spin echo a given amplitude.
If the nuclei in the region of the sample into which the magnetic field is applied
are undergoing adiffusion process, then the amplitudeis attenuated. The approach
for extracting the diffusivities exploits this process. Traditionally, this technique
has been capable of measuring very fast diffusivities, 10-5-10-9 cm2/s in poly-
mers. Recent refinements of the technique may enable one to determine diffusiv-
ities as slow as cmz/s(Fleisher, 1992).
There are a number of other techniques that have been used to study diffusion
in polymers that we will not discuss here. These include conventionaltransmis-
sion Fourier transform infrared spectroscopy (Highet al., 1992), scanning elec-
tron microscopy X-ray fluorescence (Gilmore et al., 1980), infrared attenuated
total reflectance (IRATR) spectroscopy (Van Alsten and Lustig, 1992), dynamic
light scattering (Murschall et al., 1986), and a modified optical schlieren tech-
nique (MOST) (Composto et al., 1990).
It is very clear from the foregoing discussion that the choice of technique
will depend on the range of diffusion coefficients to be determined, the avail-
ability of suitable labels, the depth resolution of the experiment, and the avail-
ability equipment. Other considerations include the ease with which samples
can be made. One very important concern is the effect of polydispersity on the
result of the actual measurement. The importance of this effect varies among
these techniques. For example, the effect of polydispersity on the diffusivity in
the NMR experiments may result in an increase in the measured diffusivity by as
much as a factor of 5 (Fleischer, 1985) over that obtained using ERD or SANS.
2. Molecular Weight Dependence of Diffusion

While the earliest known measurements of diffusion of flexible linear chains in
melts were performed by Bueche and coworkers (Bueche, 1968; Bueche et al.,
1956) usinga radioisotope labeling technique, the first systematic series of mea-
surements designed to test the predictions of reptation were performed by Jacob
Klein at the Cavendish Laboratories during the late 1970s (Klein, 1978; Klein
and Briscoe, 1979). These measurements were performed on polyethylene (PE)
TRANSLATIONAL D W M I C S IN
MELTS 263
using infrared microdensitometry. Figure 5a showsa double logarithmicplot of

the self-diffusion of polyethylene from different authors. While virtually all the
published data for polyetheylene follow the M-' dependence predicted by rep-
tation (Bartels et al., 1984; Kimmich and Bachus, 1982; Bachus and Kimmich,
1983; Fleisher, 1984, 1985, 1987; McCall et al., 1959; Klein and Briscoe, 1979;
Klein et al., 1983; Peterlin, 1983; Zupancic et al., 1985), the actual magnitudes
di€fer in some cases. The discrepancy is associated with the NMR data. Some
of the NMR data, like those of Pearson et al. (1987), appear to be higher than
the other sets of data determined using SANS (Bartels et al., 1984) and IRM
(Klein, 1978). It is evidently not related to the technique, as the other sets of
NMR data are in agreement with those data obtained using S A N S and IRM
after they were corrected for polydispersity. The data of Pearson et al. were also
corrected, but the discrepancy still remains. The difference is therefore not re-
lated to polydispersity or to the techniques. One strong possibility if that some
of the measurements were done on hydrogenated or deuterated 174-polybutadi-
ene, which have some minor microstructural differences that differentiate them
from PE (inclusion of some 172-polybutadiene).
The different sets of data on polystyrene also exhibit an M-' dependence
(Bueche, 1957; Kumugai et al., 1979; Kimmich and Bachus, 1982; Bachus and
Kimmich, 1983; Green et al., 1985, 1986; Green and Doyle, 1990; Antonietti
et al., 1984, 1987). The molecular weight dependence of the D* values of PS
is shown in Fig. 5b. The agreement between the different sets of data is much
better thanthat for polyethylene. There are additional results for other pure
homopolymer systems, such as poly(methy1 methacrylate) (PMMA) (Green et
al., 1988, 1989), that also show M-' dependencies for other pure homopolymer
systems (Fig. 5c). There should be a word of caution about comparingD* and
D,. The M-' dependence is strictly valid in the case of a tracer chain diffusing
into a host of sufficiently high molecular weight. In the case of self-diffusion
one measuresthe diffusion ofchains of molecularweight M into a hostof chains
of identical molecular weight. There is a regime of M where D, necessarily
becomes larger than the D* of a chain in a matrix of high molecular weight.
This is attributed to the constraint release or tube renewal effects (Graessley,
1982; Klein, 1986), which are discussed in a later section.
We can now comment of the molecular weight dependence of D* in the
molecular weight regime whereM M,. Data on PE exhibit an M-' dependence
(Pearson et al., 1986), as expected from the Rouse model. This result is realized
after free-volume corrections have beenmade. RecentNMR experiments ofself-
diffusion in polydimethylsiloxane (PDMS) show that for M < M,, D* M", -
which is also consistent with the Rouse prediction (Crosgrove et al., 1992).
There remain some noteworthy observations that are disturbing. In PDMS
(Crosgrove et al., 1992), the molecular weight regime where M > M,, it was
-
determined that D M-", where a is appreciably less than 2. Another note-
264 GREEN
I
1
1o - ' O O A\0
0
0
10-l2 ' 1 I I
1 o2 1 o3 1 o4 1 o5 1 o6
(a) M
Figure 5 (a) A plot showing the molecular weight dependence of tracer diffusion and
self-diffusion in polyethylene melts. The circles represent the SANS data (self-diffusion);
the triangles represent the IRM (tracer diffusion) results, and the solid straight line is a
representation of the data of Pearson et al. using NMR (self-diffusion).
worthy discrepancy is seen in polyisoprene (PI), where the apparent exponent

is 3 and not the expected value of 2. These discrepancies remain unresolved
(Yu, private communication).
Predictions of the magnitude of D based on Eq. (17) are in reasonable agree-
ment with the published diffusion data. In the case of PS, at 174"C, the predicted
prefactor is Do = 0.006 whereas the measured prefactor is 0.0075 cm'/s [ G i =
2 X lo6 dyn/cm' (Ferry, 1980); (R') = 1 X cm', p = 0.96 g/cm3, M, =
31,000, and rlO(Mc)= 1200 (Fox and Berry, 1968)l. In the case of polyethylene,
at 176"C, the experimental value of the prefactor ranges from 0.32 cm2/s(SANS)
to 0.82 cm'/s (IRM), while the calculated value is 0.34 cm2/s [ G i = 2 X lo7
dyn/cm2, (R') = 1 X cm2, p = 0.767 g/cm3, M , = 3800, and qo(Mc)= 0.3
(Bartels et al., 1984)l. At 175"C, Pearson et al. found from the NMR experi-
ments that Do = 1.25 cm2/s.
Having discussed tracer diffusion and self-diffusion in pure homopolymer
systems, we can now address tracer diffusion of chains into a host composed of
two dissimilar yet compatible polymers. 'lbo systems have been investigated.
TRANSLATIONAL DYNAMICS IN MELTS 265
1 0-l2
1 0-13
*
p 10-l~
1 0-15
I
1 0-l6
1 o4 I o5 1 o6
(b) M
I 0
D'(RBS)
D*(ERD)
A D'(FRS)
\
X D' (NRA)
1 o4 1 o5 1 o6 1 o7
(c) M
Figure 5 Continued (b) FRES measurement of d-PMMA of molecular weight M
into high molecular weight h-PMMA (data of Green). (c) Dependence of D* on M for
polystyrene using different techniques. [FRS data of Antonietti et al. (1987); RBS data
of Green et al. (1984); ERD data of Mills et al. (1984).]
GREEN
Green et al. studied the diffusion of d-PS into compatible mixtures of PS and
poly(viny1 methyl ether) (PVME) of varying composition and determined that
the D* for the d-PS chains varied as M-' (Fig. 6) (Green, 1991; Green et al.,
1991). However, the actual magnitude of D* varied with composition and ex-
hibitedaminimum at approximately the 50% composition regime. Figure 7
shows the composition dependenceof D* for d-PS diffusing into the PSPVME
system at a constant temperature abovethe glass transition temperature of each
blend (T - TB= The variation is considerable and is not fully understood
at this point. Compost0 et al. (1990) studied the tracer diffusion of d-PS and of
deuterated poly(xyleny1 ether) (d-PXE) chains into mixtures of PS and PXE of
varying composition. Like the PSPVME system, D* of the d-PS and of the d-
PXE chains diffusing into the mixtures varied as M-*. These values exhibit
a strong compositional dependence. What is interesting abuot the composition
dependence of D* in bothsystems is that it is highlynonlinear. It is clear
nevertheless that the relaxation times of the chains in the blend are considerably
slower than in the pure components. Recent NMR measurements of the relax-
ation the components in PSPVME and PSPXE blends show that the seg-
1 0-l2
m
D(40%0,1070C)
lo"8 3
10"~ ~ ~ ~ " " " " " ' " ' ' " '
-0.2 o 0.2 0.4 0.6 0.8 1
cp
Figure 7 Compositiondependence thetracerdiffusioncoefficient, D*, d-PS
chains diffusing into miscible blends if PSandPVME are at temperatures T - T, =
100°C. [Data Green et al.
mental mobilities of the components in the blends are very slow in comparison
to mobilities in the pure components (Jones et al., 1993). In addition, recent
measurements of the relaxation of the components of a PMMAIpolyethylene
oxide (PEO)mixture using an approach that simultaneously determinesthe in-
frared dichroism and the birefringence also support these findings (Zawada et
al., 1992). Dichroism measurements by Monerrie and coworkers also support
these findings (Lefebvre et al., 1984; Faivre et al., 1985). Therefore it is clear
thatina compatible mixture the components relax, or undergo translational
diffusion, at a rate that is considerably slower than in the mixture. Further, the
relaxation varies nonlinearly with the average composition of the blend. There
is yet to be a clear explanation of this finding. It could be due, in part, to the
composition-dependent interactions between the unlike segments in the mixture.
3. TemperatureDependence
It is very clear from Eq. (17) that knowledge of the temperature dependenceof
qoshouldbe sufficient todetermine the temperaturedependence of The
temperature dependenceof DRepcan be determinedin a straightforward manner.
Equation (17) can be rearranged yield
log(DR,/T) = c(M)- 1% l o (19)

268 GREEN
The relation that G$) = pRT/M, was substituted into (17) (Graessley, 1980).
Thetemperaturedependence of qo is well describedby the Vogel-Fulcher
equation
log ll0 = A + B(T - To)“ (20)
where To and B are the Vogel-Fulcher constants, unique to eachpolymer (Berry
and Fox, 1968). It follows that the temperature dependenceof D,,/T should be
given by
log [(=)DIl,(T,S
(%)l = A’ - B(T - To)-’
where A‘ = B(T,, - To)”. In this equation, T,, is the reference temperature at

which DRep(Tnf) is determined.
This result suggests that log T/D* should have the same temperature depen-
dence as log qo.It turns out that the experimental data in polystyrene (Green
and Kramer, 1986a,b) and in polyisoprene (Nemoto et al., 1984) are consistent
with this. Figure 8 shows the temperature dependenceof the diffusion d-PS
into PS. It also exhibits the same temperature dependence asthat the inverse
o”*
0-15
5 6 10
B/(T-T )
Temperature dependence the D* d-PS of M = 110,000 into PS.The
empirical parameters used for the fit are B = 710”C, To = 49”C, T,, = These are
thesameconstantsusedto fit the PS viscosity data. [Data replotted from Green and
Kramer (1986).]
TRANSLATIONAL
DYNAMICS IN MELTS 269
of the viscosity of PS (Green and Kramer, 1986a,b). Another system where a

similar observation is made is in PMMA.
It should be pointed out that the temperature dependence of log (T/D*) is
different from that of log qoin polybutadiene, as shownby Bartels et al. (1984).
In polybutadiene, the temperature dependence of the melt viscosity is
where the activation energy for flow is 30 kJ/mol. While independent of M , as

expected, it is 8 kJ/mol higher than that for diffusion. As argued by Bartels et
al., this can be rationalized in terms of (17) and(12). The temperature
dependence of DIT is controlled primarily by the temperature coefficient of the
friction factor It turns out that in PS, PI, and PMMA, the temperature depen-
dence of 5 varies by orders of magnitude over a temperature range of 100°C.
The other temperature-dependent factors in Eq. (17), (R') and the density, vary
by only a few percent over the same range. In the case of PBD, however, the
temperature dependence of is not very strong; therefore, that of the density
and that of (Rz) become important. These factors account for the difference in
temperaturecoefficient. alternative explanationhasbeensuggested (Mc-
Kenna et al., 1985) for the difference in temperature coefficient. This argument
is based on the fact that the entanglement interactions between diffusion and
viscosity are different. This is related to the observation that the time dependence
of G(t) (Fig. l), the stress relaxation modulus, is not a simple exponential in
time; instead it is described by a stretched exponential [G@) exp-
(Ngai et al., 1988). The value of the coupling constant n for the viscosity is
different from that for diffusion. The data for PS, PI, and PMMA, however,
appear to refute this alternative explanation.
D. Tube Length Fluctuations and Constraint Release

There is considerable experimental evidence in support of the M -'power law
scaling behavior for the diffusion of a long flexible polymer chain in a host
environment composed of sufficiently long chains. In addition, measurements
of the magnitude of D,, and of G$)are in reasonableagreement withpredictions
of reptation. However, there exist a number of inconsistencies between predic-
tions of the pure reptation model and experiments. One discrepancy of great
-
concern is the prediction that qo M' while experiments showthat qoactually
scales as (Ferry,
1980;
Berry
and Fox, 1968).Furthermore, while the
plateau of G(t) [G(t) = G,(O)+(t)] is correctly predicted, the prediction that the
product of the steady-state compliance J.(O) and is
is not correct. Experiments show that the product is between 2 and As we

mentioned earlier, the idea of treating the chain as diffusing into a fixed tube is
270 GREEN
not strictly valid in many situations. One can imagine a situation in which a
long flexible chain diffuses in an environment where the host chains relax at a
faster rate than the probe chain. In that case, Eq. (14) would no longer be valid.
Figure 9 shows a schematic of the topology of the tube being altered on time
scales smaller than the T~ of the N-mer chain. It is clear from this figure that
upon removal of constraints on the N-mer chain, the chain canundergo a lateral
displacement, which is otherwise prohibited in the simple reptation picture. One
might view this, at one level, as the diffusion of the tube, as the primitive path
is altered in the process. Clearly, however, the effect of the mobility of the host
environment is to enhance the mobility of the N-mer chain.
One of the first attempts to rectify the situation was made by Daoud and de-
Gennes (1979). They noted that in the limit where the molecular weight of the
diffusant is very high, the diffusant should diffuse as a coil in an environment
-
whose viscosity should vary as qo P 3 (P > Pc), where P is the degree of
polymerization of the host chains. Consequently, there should be a correction
of DRcp, which varies in a manner that maybe described by the Stokes-Einstein
-
relation. This correction is predicted to vary as DCR N'nP". There have been
a variety of other theories developed understand the effect of the matrix on
diffusion (Klein, 1986; Wantanabe and Tirrell, 1989; Viovy, 1985; Graessley,
1984; des Cloizeaux, 1988a,b, 1990, 1992; Rubenstein et al., 1987; Rubenstein
The configuration of the chain changes as the topology of its environment

changes when a constraint is released.
TRANSLATIONAL D W M I C S IN MELTS 271
and Colby, 1988; Doi et al., 1987). With the exception of Klein’s work, these
theories address the question of viscoelasticity primarily. Most the constraint
-
release theories suggest that the correction to DRepvaries as DCR P-3. Klein,
-
on the contrary, argued that the dependence is somewhat stronger, DCR P-SR.
Klein’s argument is based on the fact that a single chain may provide more than
one constraint on the N-merchain. Graessley’s theory, for example, asdiscussed
later, assumes that each P-mer chain accountsfor one constraint. The effect of
the interdependence the constraints is to enhance the constraint release (tube
-
renewal) contribution; hence D l C R F S R . Hess (1988b) also addressed the ques-
tion of constraint release in melts using a many-body approach and suggested
that D , - P”. It turns out that there are no experimental data on the effect
constraint release of the diffusion of linear chains that support Hess’s finding.
Below we describe Graessley’s contribution.
The basic idea is that each chain in the system is undergoing a reptative
motion with a characteristic reptation time given byTd(lM) for the probe chain
and Td(P) for the matrix chains. Both the reptation and the constraint release
processes are assumedto be independent;therefore, the total diffusion coefficient
of the N-mer chain is now
D* = DRep+ DCR (24)

where DCR, as described above, is the contibution the host environment. It
has been argued (Wantanabe and Tirrell, 1991) that the assumption that the two
processes are independent is valid only if the conformation of the tube and that
the chain trapped in it remain Gaussian during each successivestep.
The constraints in this model are considered in an idealized manner where
the diffusion the N-mer chain occurs on a cubic lattice with effective con-
straints (the removal of an arbitrary constraint will not necessarily contribute to
alteration the primitive path) per step along the primitive path. Constraints
should relax at a rate proportional to Td. The mean waiting time for the release
the first the constraints is defined as
According to Graessley, the contribution of constraint release to the diffusion

coefficient of the chain is
In the case of a single chain diffusing into a single-component monodisperse

host, the mean waiting time takes on a value of T,,,= (m2/12)2?d.
272 GREEN
Note that an approximationis made whereby +(f) is represented by the first,

and by far the most dominant, term in the series. It follows that the complete
diffusion coefficient for a linear flexible chain of molecular weight M diffusing
into a monodisperse host environment of molecular weight MPis given by
D, = Do(”’ + acRM2M/Mi)
where Dowas defined earlier as Do= (4/15)MdM,kBT/{. It is clear from this
result that the correction to the simple reptation prediction becomes less signif-
icant as MP increases. It is interesting to note that in the case of self-diffusion
(M = P), D, has a slightly higher magnitude thatD*, which becomessignificant,
particularly at lower M. At sufficiently high M, D* = D,.
Green and coworkers have shown that Eq. (27) provides a very good de-
scription of the diffusion in d-PS chains into PS hosts of varying molecular
weights. Shown in Fig. 10a are data that have been fit with Eq. (27). Only one
adjustable parameter has been used, acR,which was found to have a constant
value of 11. Forward recoil spectrometry measurements by Green et al. (1984)
on P “ A melts are also well described by Eq. (27) using a value of = 11
(Fig. lob). IRD measurements in polyethylene(Von Seggren, 1991)also indicate
that Eq. (27) provides a good description of the data. It was, however, found
I I
lo7
(a) P
(a) Data showing the effects of constraint release of d-PS in PS at
The lines drawn through the data were computed using Eq. 0) M =
M= M= M= (A) M = (A)M = 1,800,000.
[Data of Green et al. (1984).]
TRANSLATIONALDYNAMICS
MELTS IN 273
thatalthough was the only adjustable parameter, its value varied with the
molecular weight of the d-PS diffusant. Studies of the polypropylene systemdid
not provide strong evidence for the reliability of Eq. (27) (Smith, 1982; Smith
et al., 1984), butother studies supported this prediction (Tead and Kramer,1988;
Antonietti and Sillescu, 1986).
The effects of constraint release on the tracer diffusion of a homopolymer
into miscible blends have been investigated in PS/PVME (Green, 1991) and in
PS/PXE (Compost0 et al., 1992) systems. The results were found to be well
described by a generalized form of the equation
that accounts for a host of components, 1and 2. In this equation, = ~ ( 1 ) /

~ ( 2and
) = [Me(1)/Me(2)lm.In the absence of the second component(i.e,. cp =
0), Eq. (28) reverts to Eq. (27). The data in Fig. 11 represents the diffusion of
d-PS of M = 200,000 (circles) and of M = 520,000 (squares) into a blends of
PS of molecular weight P = 1.8 lo6 with PVME, where the PVME
P
Continued (b) Constraintrelease data of d-PMMA (M = 519,000) into
PMMA of molecular weight P at [Data of Green et al. (1984).]
274 GREEN
10-12
-
10-13 -
10.14
-
10-15
-
10-16
104
I ....
107105
I I ...a l
106
.. I I **ad
P
Figure Data showing the constraint release of d-PS M= (M) M =
520,0001 into miscible blends of PS of molecular weight P with 40% P V M E of fixed
molecular weight P = 145,000. [Data of Green et al.
molecular weight was fixed at M = 145,000. The broken line was calculated
[Eq. (28)], with the constants cp = 0.4 and the molecular weight between entan-
glements of PS taken to be Me(l) = 18,000 and that of PVME, Me(2) = 12,000.
The constant k was taken to be equal to aCR = 11. While the values of k were
found to be consistent in the PSPVME system, they were found by Compost0
et al. (1992) to vary considerably with composition in the PSPXE! system. A
resolution of this situation will await further experiments and theory.
2. TubeLengthFluctuations
In addition to the constraint release process,theN-mer chain is capable of
undergoing other relaxation processes not described by the original reptation
model such as fluctuations in the length, L, of the primitive path. It has been
suggested that these fluctuations in L might account for the discrepancy between
the experimental and predicted power law dependence of qop o i , 1983; Doi
and Edwards, 1986). Doi has shown that the average fluctuations
(M2)/@)
= Z-In (29)
Recall that 2 is the number of steps on the primitive path (2= L/u). Considering
that qois proportional to the longest relaxation time, one might incorporate the
fluctuations in length in calculating the new relaxation time. According to Doi,
7d(F) m 7d(1 - k/Z1,)’
MELTS IN 275
where k is a numerical constant that is close to unity. This result is obtained by

noting that 7 d L*/&, and 7 d ( F ) (L - AL)’/DR,,.Based on the result in Eq.
-
(30), the new qo,which incorporates the chain length fluctuations, is
qom qo[l - k v 4 e ~ ~ i n 1 3 (31)
where k’ is a new constant. It is important to point out that Eq. approaches
the reptation prediction at extremely large values of M . At smaller valuesof M,
these corrections are important. Experimentally, Colby et al. (1987) have shown
- -
that qo N3.4for M/Me as great as 150; beyond M/Me 200, departures from
the N3.4dependence that are consistent with N 3 are observed. While the data of
Colby et al. suggest that in the limit of very large M/Me the viscosity should
approach M 3 , there is a suggestion that the tube length fluctuations may not
fully account for the discrepancy (O’Connor and Ball, 1992). This is due, in
part, to the large error associated with measurement of the viscosity the very
high M polymers.
O’ConnorandBall(1992) later revisitedthisproblem. As mentioned
earlier, the Doi-Edwards model considers the chain diffusing along the tube in
accordance with one-dimensional Rouse behavior. Only a single diffusion co-
efficient, D R O , and only one Rouse mode, the longest relaxation time, are con-
sidered to be relevant. Furthermore, fluctuations in chain length are also ignored.
O’Connor and Ball recognized that any description of the dynamics the chain
should incorporate the full Rouse relaxation behavior, particularly of the chain
ends. This is important because the release of stresses in the tube occurs at the
current tail (defined by the direction of motion of the chain) of thechain whereas
that due to the head of the chain is not important, provided there are no major
fluctuations in contour length. By carefully rescaling the relaxation times and
all the length scales in the problem, they were able to express the positions of
the chain ends in terms independent coordinates. Consequently, the behavior
of the chain ends could be described in terms of independent Rouse modes.
With the use of only two material-dependent parameters, which are easily
measured, G$) and the monomeric friction factor O’Connor and Ball (1992)
showed, through a computer simulation, that the outstanding discrepancies (qo-
M, J:)G$) prdiction, etc.) in dependence could be accounted for. One the
important results of this work is that the chain contourfluctuations described by
DoiandEdwards(1986)couldnot account for the viscosity power law dis-
crepancy. In fact, O’Connor and Ball demonstrated that by combining the con-
straint release effects with their corrections they could account for the depen-
dence of the viscosity on complete magnitude and molecular weight, entangled
and unentangled. The M3.4power law dependence wasaccounted for in a num-
ber polymer systems; at very large M the M 3 power law dependence is re-
covered. Furthermore, the discrepancy in the value of J:)G$) was also resolved.
It is my opinion that this is not the end of the story! There are details of the
276 GREEN
simulation that are unknown because they were not published. The diffusion of
ring and star molecules is discussed below.
BranchedMolecules
Branched molecules, in the presence of fixed objects, are unable to undergo a
strict reptation process. Their motion is facilitated primarily through fluctuations
in contour length. In a host of linear reptating chains, the constraint release
process is expected to play an important role, more important than in the dif-
fusion of linear chains.
The question of the diffusion of star-shapedmolecules wasfirst addressed by
deGennes (1979) and subsequently by a number of other authors (Graessley,
1982; Helfand and Pearson, 1983; Klein, 1986; Pearsonand Helfand, 1984; Doi
and Kunuzu, 1980). For the sake of clarity we begin with the motion of a star
molecule off = 3 the schematic of which is shown in Fig. 12. The star
undergoes translation in the presence of fixed obstacles. It follows that transla-
tional motion of the star can occur only by fluctuations in arm length L,. For
the star in Fig. 12a to move a distance a, a step on the primitive path, arm 1
must retract to the node without crossing any obstacles. We can calculate the
relaxation time, T,, for such a process. Doi argued that if the probability distri-
bution of a chain of N, segments, L, - (Ls), is Gaussian, then the motion L,
can be considered to be Brownian and occurring within a harmonic potential
k,T
= (S) (L, -
If one considers this an activated process, then the disengagement time associ-
ated with the chain end going from a point (L,) to L, = 0, as shown in Fig. 12b,
12 Schematic depicting the mechanism of diffusion of a star molecule.

-
is given by
T~ exp[(3/2)~,(b/a)~] (33)
alternative manner in which one might arrive at the same result, as shown
by deGennes, is to consider the probability of an arm of N, segments retracting
along its contour without enclosingany obstacles. Such a probability is
34) P(NJ exp(-yNs)
where y is a constant that depends on M,. The rate at which the arm retracts,
deGennes argues, should be given by
1hs= P(Ns)k@J (35)
Therefore,
(36) 7 s cc Td(NS) exp(yNs)
There have been a number of predictions for the form of Eq. (36) where
7 s - N: e x p ( - W (37)
The exponentk has beenassigned values of 3 (deGennes, 1979;Doi and
Kunuzu, 1980); 0 (Graessley, 1982); 312 (Pearson and Helfand, 1983); and 1.9
0.1 (Needs and Edwards, 1983).
When the arm completely retracts (Fig. 12c), the center of mass of the chain
diffuses a distance equivalent to the primitive path. During this process, the
chain is forced to drag the other two arms that same distance. Since the diffusion
coefficient is defined as
D a2/7, (38)
(38) predicts that
The foregoing result is for a three-arm star. Doi pointed out that as the star
diffuses, it needs to withdraw - 2 arms to the branching point. The activation
energy for such a process is
Therefore the general diffusion coefficient is
70
278 GREEN
In general, one might write the diffusion coefficient for an farm star as
D,,, = c”Dre, exp[-s(f - 2WeMsI (42)

where c,, c, and c, are constants.
We may now consider the situation in which the star chains diffuse into a
matrix of linear flexible chains of molecular weight MP.The constraint release
process will play a significant role in the translational dynamics of this system.
Under these conditions the total diffusion coefficient is
D* = D,,,, + D m (43)
where DCRis given by Eq. (25).
Few measurements have been done in this area (Kline et al., 1983; Bartels
et al., 1986; Antonietti and Sillescu, 1986; Shull et al., 1988; Crist et al., 1989;
Fleischer, 1985). Experimentally it is well established that star molecules diffuse
much more slowly than linear chains of comparable dimensions and that the
diffusion rate of the stars depends exponentially on the length of the arm. The
actual molecular weight dependence of the preexponential is less certain at this
point. Many of the data appear to be well describedby assuming that the preex-
ponential factor is independent of molecularweight. Shull showedthat D,,,
varies exponentially with the molecular weight per arm for stars of different
lengths diffusing into microgel matrices (Fig. 13). These microgels relax suffi-
1 0-l1
U
10”‘
l0” *
0-l
0 10000 20000 30000 40000 50000
M
a
13 length dependence of the D* of afour-armstarmolecule.Thedata

are exponential as expected. [Data of Shull et al.
TRANSLATIONAL DWAMICS
MELTS IN 279
ciently slowly that they simulate the behavior of linear chains of infinite length.
Consequently the arm retractionmechanism is significant. Bartells et al.and
Klein et al. also demonstrated this exponential dependenceof the diffusivity of
PE star molecules diffusing into high molecular weight matrices.
The constraint release mechanism is also shown to be veryimportantin
facilitating motion in linear matrices of sufficiently short length. As one might
anticipate, this release process plays a considerably more significant role in the
translational diffusivity of star-shaped molecules. Shown in Fig. 14 is the de-
pendence of the diffusivity of star-branched molecules on the molecular weight
of the chains in the matrix. The data are well described by Eq. where the
constraint release contribution is described by (Klein, 1985)
DCR= D ~ M - 1 P - 5 R (44)
This is a stronger dependece onP than the expected F 3 . This discrepancy might
be related to the interdependence of the constraints discussed by Klein earlier.
Molecules
As pointed out by Kline (1985), ring molecules can assume a number of con-
formations as shown in Fig. 15. In Fig. 15a the ring can enclose obstacles, in
which caseit will be trapped and be able to undergo translational motion,unless,
10 I
Y
%c 1 -
0.1
1 0' 1Ob os
14 Datashowingthe effect of constraintrelease thediffusion of a star

molecule of M. = 60,000. D m vanes as P-'.', as [Replotted data Shull et al.
280 GREEN
of course, by some constraint release process. There are, however, untrapped

conformations shown in Figs. 15b (ramified configuration) and 15c (linear con-
figuration). The probability that the ring molecule will not be trapped by the
chains in the matrix depends on the size of the molecule and varies as e-consc "R,
where NR is the number of segments that compose the molecule. It is therefore

expected that a fraction of the chains will be trapped. If the ring is to undergo
translational motion, it has to be of the linear configuration as shown in Fig.
1 5 ~Therefore
. the diffusion coefficient of a ring molecule is
DR(NR) DRep(NR/2) (45)
Rubenstein (1987) made a similar prediction. If the chain diffuses into a host
environment of linear chains of molecular weight M,,, then the total difision
coefficient is
D* = ~ ( N R+ )DcR(M~) (46)
Studies of the d f i s i o n of ring polymers into different host environments
have been performed in PS (Mills et al., 1987; Tead et al,. 1992) and in PDMS
(Crosgrove et al., 1992). The experimental situation is far less certain for rings
than for stars. The NMR measurements (Crosgrove et al., 1992) the self-
diffusion of ring molecules show a molecular weight dependence that is appre-
ciably less than F*. The exponent shows no tendency to approach -2 with
increasing N . This is a surprising result. It is possible that the PDMS system is
-
unusual consideringthat the authors foundDs N-",where a < 1, in the regime
where an exponent of a = 2 is anticipated. It is also noteworthy that rings of N
less than about 20 units diflFuse at a rate that is faster than that of linear chains
of a comparable number of units. At larger values of N , the rates are comparable.
These data are shown in Fig. 16.
The PS measurements by Tead et al. (1992) show clearly that thereis a lower
limit of the ring diffusion coefficient in linear matrices that was not observed
by Mills et al. in anearlier study. They showed that for large enough host chains
D* became independent of host molecular weight, consistent with the predic-
tions of (25) (Fig. 16). This is reasonably strong evidence in support of a
15 Possible configurations of ring molecules in a host of linear chains.

TRANSLATIONALDYNAMICS IN MELTS 281
U
n
10”2 l O Om
0
m.
0
1 0-13 m 0 0
00
0
10”‘ 4
Constraint release of ring molecules diffusing into ring matrices (0)and
into linear matrices N dependence of the self-diffusion of PS chains. [Data of Tead
et al. (1992) replotted.]
reptation type of mechanism. additional point that is worth mentioning is

that the diffusivity of linear chains into linear matrices is identical to that into
rings of the same molecular weight.
The D* of rings into PS microgels of P = 490,000 were found to vary with
M with an exponent much larger in magnitude than -2. The data of Teal et al.
are replotted in Fig. 17 as D*M& versus M to illustrate the large deviation.
This result is somewhatof a mystery.It is clear that the diffusivity of ring
molecules is an area that awaits further exploration given the limited, and to
some degree controversial, data that are currently available.
In the section that follows we addressthe question of interdiffusion in poly-
mers where the diffusivities of the chains become highly dependent of concen-
tration and of the strength of the thermodynamic interactions between them.
W. INTERDIFFUSION
In the foregoing section we showed that the tracer diffusion coefficient of a
single chain into a high molecular weight matrix environment was determined
primarily by two parameters, the longest relaxation time, T,, (recall T~ -and
the molecular weight. It was also noted that when the chains that compose the
matrix are able to relax sufficiently fast, the topology of the tube is altered on
time scales faster than the longest relaxation time of the diffusant. Under these
282 GREEN
1 o4 1 o5 1 os
17 The D* of d-PS rings into PS microgels of P = 490,000. The data of Teal

et al. (1992) are replotted D*Mzgversus M .
circumstances, a correction has to be made to the tracer diffusion coefficient to

account for the effect of the host chains.
In polymer mixtures the interdiffusion coefficient D is of great practical im-
portance. It influences adhesion and bonding between dissimilar polymers, flow
and viscoelastic properties, and phase separation in polymer blends. The inter-
diffusion coefficient not only depends on T~ and N, it is also highly concentra-
tion-dependent. Tracer diffusion and self-diffusion coefficients are entropically
driven, whereas D is influenced by gradients in chemical potential. The effect
of the thermodynamicsof the system on D can, within the Flory-Huggins model
(Flory, be characterized by x.
In a two-component system the excess free energy of mixing is proportional
to x+A+B, where +A is the volume fraction of component A and +B is that of
component B. It was first pointed out by Flory (1952) that the free energy of
mixing per segment can, within a mean field approximation, be described by
TRANSLATIONALDW’ICS
MELTS IN 283
The first two terms on the right represent the combinational entropy of mixing,
and the third represents the enthalpic and noncombinatorial entropy of mixing.
It is clear from this equation that the combinatorial entropyof mixing for poly-
mers is extremely low, varying as UN, where N is the degree of polymerization
of the chain. This is in contrast to small-molecule systems (recall that N for
polymers is typically a few hundred to a few thousand). Therefore itis expected
that the mutual diffusion coefficient will be highly influenced by the value of
x. When < 0, mixing is favored and, as we will see later, the magnitude of
the mutualdiffusioncoefficient(the interdiffusion coefficient) is greatly en-
hanced over the case where x = 0. This behavior can be described as a ther-
x
modynamic “acceleration” of D. When > 0, mixing is influencedby the
combinatorial entropy such that a system canremain in a region of single-phase
x
stability provided that 0 < < S,where
As shown later, D(+) will undergo a “thermodynamic slowing down” under

x
these conditions. Of course, when > xS,mixing is not favored.
Our discussion on interdiffusion will concentrate on two-component mixtures
for simplicity. Wemay begin by considering twopolymer layers, A and B,
separated by an interface. Further, we will imagine that the chains are arranged
on a quasi-lattice, where eachcell occupies a volume When the chains diffuse
across the interface, an important requirement is that the total flux across the
interface must be zero. There is a flux, JA, due to the A chains and another, JB,
due to the B chains. It is clear from the early experiments, beginning with the
Kirkendall (1948) marker experiments in metallic alloys, that in diffusion ex-
periments of this nature there are other fluxes in addition to the diffusive fluxes,
J(species A) and J(species B), present that necessarily compensate in the event
that J(species A) # J(species B). In the case of metallic alloys, it is a vacancy
flow. In these landmark metallic alloy experiments it was demonstrated that the
motion of the marker at the original interface of the dissimilar metals provides
direct evidence of this vacancy flow.
In the case of polymers, one should expect similar behavior. If JA # JBand
there are no other fluxes present, then a pressure gradient proportional to a
gradient in chemical potential will develop at the interface. It is expected that
these gradients must be relaxed in the system. For polymers, however, where a
lattice does not exist in reality, a vacancy mechanism of the sort that exists in
metals would not be appropriate. However, any mechanism that accommodates
a bulk flow to release the buildup in pressure should suffice. In this regard we
can, as Onsager showed earlier, write down the fluxes as linear combinations of
generalized forces (gradients in chemical potential), where the constants of pro-
284 GREEN
portionality are the Onsager (or mobility) coefficients

JA = -MAVpA
J B = -MBVpB
Jv = MAVFA + MBVCLB
Here we formallyrefer to the compensating bulkflow as a "vacancy" flow and
designate the flux as J, Vpi are the gradients in chemical potential. We have
formally introduced the assumption that Vc~y= 0, which guarantees that "va-
cancies" are at equilibrium everywhere. We have also assumedthatthe
diagonal terms M m = MBA= 0. The following expressionfor the interdiffusion
coefficient was derived by Kramer et al. (1984):
where xs is defined by Eq. (48). This equation was derived by first obtaining
expressions for the chemical potential gradients in terms of gradients in con-
centration. That was followed by noting that the total flux segments of
JA(tot), across a fixed interface must be conserved, i.e.,
JA(tot) = "AVPA + +(MAVPA + MBVpB) (51)
and that
Eq. (50) follows. By writing the Onsager coefficients in terms of the Rouse
segmental mobilities, Kramer et al. (1984) showed how M. (50) could be ex-
pressed in terms of the tracer diffusion coefficients of species A and B in the
mixture.
The Onsagercoefficients can be written in terms of the segmental mobilities,
BA and BB, as
MB = BBcB (54b)
where C, = +/aand cB = (1 - +)/aare the concentrations of species A and
B. These relations follow from the fact that the flux is proportional to the dif-
MELTS IN
TRANSLATIONAL
DYNAMICS 285
fusional velocities, where the constant of proportionality is the concentration.

Alternatively,the flux is also proportional to the chemical potential gradient
(generalizedforce), where the Onsagercoefficients are the proportionality
constants.
Equation (50) &n be written in terms of the tracer diffusion coefficients by
noting that the polymer segment mobilities, B, and BB, can be described by the
curvilinear Rouse segment mobilities. Therefore,
D(+) = - +PT(X. - (55)
where
DT = - +)DXNA + +DzNB (56)
This result was derived independently by Sillescu using a more
general approach, where he took advantage of the Hartley-Crank equation. The
form of Eq. (56) indicates, for example, that if DX >> D:, then D(+) is con-
trolled by D:, the faster moving species. Therefore, if a Kirkendall-type marker
experiment wasperformed, one would expect that there would be a net flow of
species A in the direction of the more slowly diffusing species B and the marker
would move in the direction opposite to that which species A (faster) moves.
This result has been identified as the “fast7’-mode theory of diffusion and has
been criticized for reasons discussedbelow.
Long polymer chains are highly entangled and diffuse by reptation, and for
this reason it is not natural to define a lattice, as one does in metallic systems.
Therefore it is difficult to envision a vacancy mechanism that would alleviate
any pressure gradients that might develop. This problem is compounded by the
fact that polymers are highly incompressible. Therefore one might imagine a
case where for the faster A chain to reptate there must be “free” space ahead
of it, and this space can be created only if the slower B chain reptates. Conse-
quently, the difEusive process should be controlled by the more slowly moving
species. This led to anotherproposal, the “slow7’-mode theory (Brochard-Wuart
et al., Brochard et al., Brochard-Wuart and deGennes, Binder,
One can, in fact, arrive at the prediction for the slow-mode theory by invok-
ing the condition that J, = 0. Brochard and coworkers and Binder arrived at the
result that the expression for DT is
This equation shows that when D: >> D:, the process is dominated by D ; , the
more slowly diffusing species. As we show later, marker experiments conducted
in polystyrene systems are strongly in favor of the fast theory (Green et al.,
subsequent proposal by these authors suggested that the problem is
286 GREEN
actually a length scale-dependent one whereby, over a length scale

L = [D? Td(B)]'R
the more slowly moving species is swelled by the faster diffusing species. This
process occurs by the fast-mode process. However, over much longer length
scales, or longer times, the diffusion process commences according tothe slow-
mode theory (Brochard-Wyart and deGennes, We will show that subse-
quent experiments do not support this view.
Further support for the predictions of the fast-mode theory was provided by
two additional studies. One study by Jordan et al. addressed the concern of the
length scale dependence. I R M was used to determine the concentration profile
of polyethylene chains of very different molecular weight, one of 32,000 and
the other of 520,000. They were allowed to interdfise at distances of macro-
scopic lengthscales (on the order of many micrometers). It was determined that
the diffusion process was determined by the faster diffusing species, in support
of the fast-mode predictions. The concentrationprofiles were asymmetric, which
is consistent with the fact that the faster diffusing species penetrate more deeply
into the slower moving species. Concentration profiles were calculated using
both predictions, and the results could be rationalized only in terms of the fast-
mode theory. Composto et al. also addressed this problem but used
a different approach. They studied the molecular weight dependenceof the mu-
tual diffusion coefficient at a given concentration in the PSPXE system and
-
determined that D N&?, a result that can be rationalized only in terms of the
fast-mode theory. There have been two noteworthy studies that support the slow
theory, the work of Garbella and Wendorf in PMWpoly(viny1ydine
fluoride) (PVDF)and that of Murschall et al. in polyphenylmethylsilox-
ane (PPMS)PS. As discussed by Composto et al. the analysis of the
results that enabled them to arrive at their conclusions has been shown to be in
error. Therefore, the overwhelming experimental evidence is in support of the
original conclusions of the marker experiments of Green and coworkers.
The question of thermodynamic slowing down in polymers was first dem-
onstrated by Green and Doyle who used ERD to study interdif-
fusion in d-PSPS mixtures of sufficiently high molecular weight. It turns out
that the replacement of one component in a mixture with its deuterated coun-
terpart can affect the phase equilibrium properties of the mixture. This effect
could become important for polymers of sufficiently high molecular weight at
finite compositions. The origins of this isotope effect are reasonably well un-
derstood (Bates and Wignall, 1986a,b; Bates et al., The isotopic substi-
tution results in slight differences in segment volumeV and atomic polarizability
a between the one polymer and its isotopic counterpart. It has been shown that
knowledge of a and V will enable oneto calculate x. Because of this unfavorable
segmental interaction, the d-PSPS system exhibits an upper critical solution
TRANSLATIONAL DE?VAMICS
MELTS IN 287
temperature (UCST). The stability limit for the system that Green and Doyle
studied was xs(aC) = 2.1 the degree of polymerizationof the PS chains
was 8.7 lo3 and that of d-PS, 9.8 lo3. Green and Doyle showed that the
mutual diffusion coefficient in the d-PSPS system experienced a minimum, or
critical slowing down, in the middle of the concentration regime at the critical
composition. the temperature of the experiment approached the UCST, the
degree of slowing down increased. This is shown in Fig. 18. The lines drawn
through the data were calculated using the fast-mode prediction [cf. Eq. (50)].
The only fitting parameter was x, whose temperature dependence is well de-
scribed by
x = 0.22(+0.01)T1 - 3.2(+0.4) (59)
which is in excellent agreement with independent measurements using
The UCST was determined for this system to be approximately 140°C.
The PSPXE system investigated by Composto and coworkers is character-
izedby x < 0. In addition to determining the compositionaldependence of
interdifhsion, where they showed a thermodynamic “acceleration” of the pro-
cess (Fig. 19), they also examinedthe temperaturedependence of D. They
showed how the value D(+ = 0.55) changed in relation to D* as a function of
<p
18 “Thermodynamic slowing down” D in d-PSPS mixtures.[Data
Green et al. (1987).]
288 GREEN
x
temperature (Fig. 20). For -c 0, D(+) > D*, and as the temperature increased
and the system approached the lower critical solution temperature (LCST), x >
0 and D(+) became smaller than D*.
Thus far the content of this review has been devoted to diffusion of homopoly-
mers of different architectures into homogeneous pure homopolymer or miscible
blend host environments. equally interesting problem is associated with the
diffusion of a homopolymer chain or a chain composed of blocks segments
distinct chemical structure, a block copolymer, into a block copolymer host.
Below we address these issues.
Block copolymers comein a variety ofarchitectures, which include diblocks,
triblocks, and star blocks, shown in Fig. 21. We focus our attention on diblocks,
which are composed of two chemically distinct monomers, A and B. The phase
behavior of diblocks is determined primarily by the degree of polymerization
N, the overall volume fractionofpolymer andthe segmental interaction
x.
parameter At sufficiently high temperatures and chain lengths, the copolymer
16”’
0.0 0.2
VOLUME FRACTION PS
Thermodynamic “acceleration” of D in the PSPXE system is represented

by diamonds. The circles represent the tracer diffusionof d-PS chains in the mixture and
the squares that of d-PXE. [Figure reproduced with permission from Composto et al.
(1988).]
TRANSLATIONALDYNAMICS IN MELTS 289
T (K)
600 560 520
' I
+-
-
"
-
-
-
IO
IOOO/(T-T,), (K-')
20 Temperaturedependence of diffusion. (0)D; D* of d-PS; (0) D*
of d-PXE. [Figure reproduced with permission from Composto et al. (1988).]
21 Thedifferentarchitecturesthatblock copolymers possess (a) diblocks;

@) triblock, (c) blocks.
290 GREEN
is completely disordered; hence the microstructure can be characterized as ho-

mogeneous. In this regime the radius of gyration of the copolymers R, NIn. -
As the temperature is lowered (i.e., x increases) the number of A-B contacts
is reduced. Of course, this is accompanied by a loss of combinatorial and con-
formational entropy. In the limit xN >> 1, well-developed ordered phases are
formed, and the interfacial region between the phases is narrow. This is the
called strong segregation limit. Here the chains are stretched and the interdomain
spacing varies as NZ3.Depending on the value off, the volume fraction of the
A segment, different ordered phases may form. With increasing values off, the
following phases are observed: body-centered cubic arrays of spheres of the
minor component in a host of the major component, an ordered array of cyl-
inders of the minor component, an ordered bicontinuous double diamond lattice
of the minor component, and finally, when f 1/2, a lamellar phase is formed.
Close to the transition where the system goes from disordered to ordered, the
phases are weakly segregated; this is the so-called weak segregation regime.
Here the radius of gyration varies as NIn. At this point the different micro-
structures still exist, but the interfacial region is comparatively broad. A com-
plete thermodynamic theory was developed by Leibler (1980) that addresses the
weak segregation limit of diblock copolymers.
Below we discuss the results of a theory developed by Fredrickson and Mil-
ner (1990) to address the diffusion of an A-B diblock copolymer of degree of
polymerization N, and volume fraction of A phase, f, into a host composed of
an A-B diblock copolymer that forms a lamellar phase. The radius of gyration
of the diffusing copolymer is R, = Ninb16, where b is the statistical segment
length. The A and B blocks are assumed, for simplicity, to have the same seg-
ment length, and each monomer occupies a volume b3. In theory, the lamellae
are oriented perpendicular to the z direction, the direction of mobility of the
tracer. A number of situations are addressed.
1. The lamellar spacing of the host is very large in comparison to that of the
tracer of f # 112. As the tracer difises into the copolymer host, the A
segment located at aposition r experiences a chemical potential p(r) =
kBTx+B(r)while that felt by a B monomer is -p(r). The net force that the
copolymer experiences froma periodic potential is
F = 2uk&&N(l - 2f) sin(&) (60)
where a is the amplitude of the potential (it is assumed to be much less
thanunity),and is awavenumberthat characterizes the scale of the
compositional variation in the melt, which is formally defined as k,, =
,&ln. The parameter (see Fig. 22) is dimensionless and can be varied
by changing the molecular weight of the copolymer host or by the addition
of homopolymer molecules. As expected, the diffusivity is anisotropic, and
TMSUTZONAL DYNAMICS IN MELTS 291
22 Schematic the diffusing copolymer in different lamellae hosts. (a) The

diffusant is small in comparison to the domainsize. (b) The diffusant size is comparable
tothe size the domain structure. (c) The interdomain spacing is small compared to
the dimensions of the diffusant.
the diffusion coefficient is described using a tensor

D = D,ZZ+ D,,,@ - (61)
where is a unit vector in the direction. D,, is the d f i s i o n coefficient
in the direction parallel to the lamellae; henceforth it will be denoted by
Dl. The prediction for D, is
D, = Dl/[lg(4IZ (62)
where Io(a)is a modified Bessel function and a = &@(l - 2f). In the
limit of small (a l),a single d f i s i o n coefficient is predicted,
DID1 = 1 - (2/3)uz(xN)'(l - 2f)' + O(a4) (63)
and when a >> 1 the prediction is
292 GREEN
DID, = (2/3)[1 +T exp(-2la()] +--- (64)

These predictions are meant to apply even in the limit where the tracer is
a homopolymer = 1 or 0).
2. Thedomain spacing isvery large (x,, << l ) , butthe tracer is symmetric
(f = 1/2). Here the segments of the copolymer experience potentials that
are exactly equal and opposite. Therefore the net force on the chain van-
ishes. The calculation proceeds by using an electrical analogy where the
segment is considered to be negatively charged and the B segments posi-
tively charged. The chain is then considered to be in the presence of an
electric field. The effective diffusion coefficient is calculated to be
DID, = 1 - 4.6 + O(P4)
10-3a4(xNc)4x~ (65)
-
where P (R, N&x h)’.
3. The domain spacing is comparable to or smaller than that of the tracer (x,,
2 (Figs. 22b and 22c). In this situation, in particular when > 1, more
than one period of the potentialis experienced on length scales on the order
of the radius of gyration of the tracer as it diffuses. In the calculation the
tracer is assumed to behave as a Rouse chain (unentangled). For the case
where IaxNcl << 1, and for arbitrary values of and f, the diffusion coef-
ficient is predicted to be
D/Dl = 1 - 0.67a2(~Nc)X(x,,, + 0[(a~N~)~] (66)
where K is a dimensionless function of and off. If the tracer is a homo-
polymer = 0 or l),then at small K FJ 1 - xd3, and at large it
crosses over to K FJ 15.8.q-’.
More recently Helfand (1992) considered the case of diffusion in strongly
segregated block copolymer hosts. The study, however, concentrated on calcu-
lating the activation energy that a copolymer chain must experience as it tries
to overcome the force exerted on it as it diffuses through the domains. There
were no predictions of diffusion with which one might compare results. Figure
23 shows the schematic of a diblock copolymer attempting to diffuse across a
copolymer domain whose dimensions are comparable to its own. I t is clear that
the segment of the chain that diffuses into the phase with which it is incom-
patible undergoes considerable stretching during this process. Figure 23a shows
the chain initially in a configuration where its segment is in the phase and
its B segment is in the B phase. The segment then has to difhse across the
B domain to get to the adjacent domain (Figs.23band23c). During this
process a portion of the segment is stretched considerably beyond normal
conditions as the A segment to get completely into the adjacent A phase.
This is shown in Figs. 23d and 23e. Figure 23f shows the final configuration.
TRANSLATIONAL DI?VAMICS IN MELTS 293
23 A schematic of a diblock copolymer chain diffusing across a domain of

unlike chemical structure.
294 GREEN
Few experiments have been performed that address the diffusion of homopo-
lymer chains or block copolymer chains into copolymer hosts. The most recent
is that of Ehlich et al. (1992), where an FRS study of self-diffusion in the PS-PI
copolymer systemwas performed. Their results showedan anisotropy of diffusion.
There were not sufficient results to make a true comparison with theory.
Shul et al. (1991) used FRES to study self-dif€usion in poly(ethy1ene propyl-
ene)/polyethylethylene (PEP-PEE) copolymer in a temperature range above and
below the order-disorder transition They observed no marked change in
the temperature dependence diffusion at theODT. They noted considerable
anisotropy in diffusivity. They showed that diffusion perpendicular to the lamellae
was orders of magnitude lower than diffusion parallel to the lamellae.It was not
possible to make a careful comparison with theory because of a lack complete
understanding of the domain structure of the copolymer host.
Green et al. (1989) studied the diffusion PS and PMMA homopolymer
chains into symmetric PS-PMMA diblock copolymer hosts. They showed that
for chains whose molecular weight M was equal to or less than half of that of
the copolymer host their tracer diffusion coefficient varied as D* = C/M2.C is
at least an order of magnitude lower than that measuerd for PS diffusing into
PS or of PMMA into P " A (Fig. 24). This is due, in part, to the fact that the
domains of the host are randomly aligned (tortuous), as shown in Fig. 25. The
I I I
2 3 4
N
24 Molecular weight dependence of the diffusion of d-PS into PS into
a symmetric copolymer of N = 824(A),and a copolymer of N =
FREE SURFACE
NEAR SURFACE
REGION
BULK
25 Domainstructure of an unannealedcopolymer.
temperature dependenceof D* for PS diffusing into PS was identical to that of

PS diffusing into the copolymer. These data are shown in Fig. 26. The same
observation was made for PMMA. From these observations it was concluded
that diffusion occurred primarily into the phases where the interactions were
favorable. As with the other studies, it was not possible to make quantitative
comparisons with theory because of a lack of understanding of the domain
structure. Green et al. (1989) also studied the case where the copolymers were
annealed for a long period of time to allow the domains to align parallel to the
surface that the chains were forced todiffuse perpendicular to the orientation
the lamellae (Fig. 27). They observedthe M-' dependence butalso observed
that the coefficient C' was appreciably smaller (C' << C) in this situation than
in any of the cases mentioned earlier. This result is not unexpected. Like the
other studies, it is not possible to make comparisons with theory because the
case addressed by Fredrickson and Milner applies only to short-chain tracers
and Green et al. studied the molecular weightdependence of long chains, which
are expected to diffuse by reptation.
It is clear that considerable more experimental work and theory are needed
in this area.
VI. CONCLUDING
While it appears that the first notable attempts to understand translational dy-
namics of melts began with Bueche in the mid-1950~~ the truly significant ad-
vances in the area were made by deGennes withhis introduction of reptation in
1971 and later by Doi and Edwards in the late 1970s. It is important to reem-
phasize that reptation is purely phenomenological. It imagines the existence of
a chain undergoing curvilinear motion in an imaginary "tube" along a fictitious
296 GREEN
-1Q
-11
-
-12
-13
9,
- -14
-1 5
-1 6
-17
26 Temperaturedependence d-PS diffusing PS (opensymbols)and

into a copolymer (A).
path, the primitive path. The tubeis defined to accountfor entanglement effects,
which in turn were introduced to account for the existence of a rubbery plateau
in melts. Despite its phenomenological nature, reptation is of great appeal, pri-
marily because of its simplicity and because of its success at predicting quali-
tative features about the diffusion and viscoelasticity of melts.
Despite the success of reptation, there are some aspects that one might find
disturbing. This is, in part, because of its phenomenological nature and because
of modifications that involve additional assumptions about the nature of the
fictitious tube (constraint release, tube length fluctuations, etc.), which have been
made to account for many experimental observations.
Considerable computer simulations have been performed to understand the
nature of the dynamics of melts and to provide validity for the central assump-
FREE SURFACE
27 Schematic of the near-surface structure of a copolymer that has been an-

nealed for a longer period of time.
tions of reptation(Kolinski et al., 1986, 1987a,b;KremerandBinder,1984;

Kremer and Grest, 1990; Muthukumar and Baumgartner, 1989). Although the
early Monte Carlo studies of Kolinski et al. found evidence for reptation in the
limit where a single chain was allowed to diffuse into amatrix of fixed obstacles,
they raised serious doubt about the validity of reptation as a viable mechanism
for the dynamics of melts. On the other hand, subsequent molecular dynamics
simulations by Kremer and Grest, the first of this type in dense polymer melts,
did provide evidencethat suggested that the reptation picture may be valid. The
work of O’Connor and Ball suggests a similar conclusion.
In addition to reptation, a numberof other theories have been proposed. Many
of these invoke tube-like concepts and give predictions similar to those of rep-
tation. recenttheorybasedon the introduction of ageneralizedLangevin
equation to describe the dynamics was developed by Schweizer (1989). The
concept of the tube is not introduced, and the effect of the host chains in a melt
on the motion of a central chain is accounted for by a generalized dynamic
friction function. This function is evaluated using mode-mode coupling ideas
that allow one to evaluate correlated collision and feedback processes in dense
molecular systems that give rise to confinement-type effects that one might ex-
pect in polymer melts. This theory recovers the essental reptation predictions in
the long-chain limit.
It is clear that reptation continues to bethe subject of debate. What is indis-
putable about it is that it is by far the most successful of the theories that have
been proposed. The reader is referred to a review by Lodge et al. (1990) that
presents a critical evaluation of its consequences.
This review hasprovided an overview of the progress that has been made in
diffusion in melts. Earlier reviews of this topic are those of Pearson (1987),
Tirrell (1984), Kausch and Tirrell (1989), and Klein (1990). Considerable the-
298 GREEN
oretical and experimental work still needs to bedone. There are unresolved
questions regarding why anomalous power law exponents are observed in PI
and PDMS. Critical tests theories for diffusion chains architectures other
than linear chains also need to be pursued. There are also open questions con-
cerning diffusion in miscible blends, particularly the anomalous concentration
dependence tracer diffusion. Our understandingof diffusion in block copoly-
mers is clearly very poor in comparison to that diffusion in pure homopoly-
mers and blends. In the final analysis, our overall progress in this field will
require not only continued activity in the area theory but also the design of
new experimentsthat will give us further insight into the details the dynamical
process.
This work wasperformed at Sandia National Laboratories and supported by the

U.S. Department Energy under contract number DE-AC04-76DP00789.
Antonietti, M., and H. Sillescu (1985). Macromolecules, 18, 1162.

Antonietti, M,. and H. Sillescu (1986). Macromolecules, 19, 798.
Antonietti, M., J. Coutandin, Grutter, and H.Sillescu (1984). Macromolecules, 17,
798.
Antonietti, M., J. Coutandin, and H. Sillescu (1986). Macromolecules, 19, 793.
Antonietti, M,. K. J. Folsch, and H. Sillescu (1987). Makromol. Chem., 188, 2217.
Bachus, R., and R. Kimmich (1983). Polymer, 24, 964.
Bartels, C. R.,B. Crist, and W.W. Graessley (1984). Macromolecules, 17, 2702.
Bartels, C. R.,B. Crist, Jr., L. J. Fetters, and W.W. Graessley (1986). Macromolecules,
19, 785.
Bates, F. S., and G . D. Wignall (1986a). Macromolecules, 19, 934.
Bates, F. S., and G . D.Wignall (1986b). Phys. Rev. Lett., 57, 1425.
Bates, F. S., G. D. Wignall, and W. C. Koehler (1985). Phys. Rev. Lett., 55, 2425.
Berry, G. C., and T. G. Fox (1968). Adv. Polym. Sci., 5, 261.
Binder, K (1983). J. Chem. Phys., 79, 6387.
Binder, K. (1987). J. Colloid Polym. 265, 273.
Brochard, F., J. Joufhoy, and P. Levinston (1984). Macromolecules, 17, 2925.
Brochard-Wyart, F., and P. G. de Gennes (1986). Europhys. Lett., 1, 221.
Brochard-Wyart, F., J. Jouffroy, and P. Levinson (1983). Macromolecules, 16, 1638.
Bueche, F. (1968). J. Chem. Phys., 48, 1410.
Bueche, F.,W. M. Chasin, and P. Debye (1956). J. Chem. Phys., 20
Callaghan, P. T.,and D.N.Pinder (1980). Macromolecules, 13,
Callaghan, P. T., and D.N. Pinder (1981). Macromolecules, 14, 1334.
Callaghan, P. T., and D.N. Pinder (1982). Macromolecules, 17, 431.
Chaturvedi, et al. (1990). Appl. Phys. Lett., 56, 1228.
MELTS IN
TRANSLATIONAL
DYNAMICS 299
Chu, W. K, J. W.Mayer,andM.A.Nicolet(1978). BackscatteringSpectrometry,

Academic, New York.
Colby, R. H., L. J. Fetters, and W.W. Graessley (1987). Macromolecules, 20, 2226.
Composto, R. J., E. J. Kramer, J. W. Mayer, and D. M. White (1986). Phys. Rev. Lett.,
57, 1312.
Composto, R. J., E. J. Kramer, and D. M. White (1987). Nature, 328, 234.
Composto, R.J., E. J. Kramer, and D. M. White (1988). Macromolecules, 21, 2580.
Composto, R. J., E. J. Kramer, J. W. Mayer, and D. M. White (1990). Macromolecules,
31, 2320.
Crank, J. (1975). The Mathematics of Diffusion, 2nd ed., Clarendon Press, Oxford.
Crosgrove, T., P.C. Griffiths, J. Hollingsthurst, R. D. C. Richards, and J. A. Semlyen
(1992). Macromolecules, 25, 6761.
Daoud, M., and P. F. de Gennes (1979). J. Polym. Sci., Polym. Phys. Ed., 17, 1971.
Darken, L. (1948). Trans. Am. Inst. Min. Metall. Eng., 175, 184.
de Gennes, P. G. (1971). J. Chem. Phys., 55, 572.
de Gennes, P. G.(1979). Scaling Concepts inPolymerPhysics, CornellUniv.Press,
Ithaca, NY.
de Gennes, P. G. (1982). J. Chem. Phys., 76, 3316, 3322.
des Cloizeaux, J. (1988a). Europhys. Lett., 5, 437.
des Cloizeaux, J. (1988b). Europhys. Lett., 6, 475.
des Cloizeaux, J. (1990a). Macromolecules, 3992.
des Cloizeaux, J. (1990b). Macromolecules, 25, 835.
Doi, M. (1981). J. Polym. Sci., Polym. Lett., 19, 265.
Doi, M. (1983). J. Polym. Sci., Polym. Phys. Ed., 21, 667.
Doi, M., and S. F. Edwards (1978a). J. Chem. Soc., Faraday Trans. 2, 74, 1789, 1802,
1818.
Doi, M., and S. F. Edwards (1978b). J. Chem. Soc., Faraday Trans. 2, 75, 38.
Doi, M., and S. F. Edwards (1986). The Theory of Polymer Dynamics, Oxford Univ.
Press, Oxford, UK.
Doi, M., and N. Y. Kunuzu (1980). J. Polymer Sci. Polym. Lett. Ed., 18, 775.
Doi,M.,W. W. Graessley, E. Helfand, and D. S. Pearson (1987). Macromolecules, 20,
1900.
Faivre, J. P., B. Jasse, and L. Monerrie (1985). Polymer, 26, 879.
Ferry, J. D. (1980). V o l e a s t i c Properties of Polymers, 3rd ed., Wiley, New York.
Fixman, M. (1988). J. Chem. Phys., 89, 3892,3912.
Fleischer, G. (1983). Polym. Bull., 9, 152.
Fleischer, G. (1984). Polym. Bull., 11, 75.
Fleischer, G. (1985). Makromol. Chem., Rapid Commun., 6, 403.
Fleischer, G. (1987). Colloid Polym Sci., 265, 89.
Fleischer, G. (1992). Macromolecules, 25, 4210.
P. J. (1953). Principles of Polymer Chemistry, Cornell Univ. Press, Ithaca, NY.
Flory, P. J. (1969). Statistical Mechanics of Chain Molecules, Interscience, New York.
Fredrickson, G. H., and S. Milner (1990). Matel: Res. Soc. Symp. 1990, 177, 169.
Garbella, R. W., and J. H.Wendorff (1986). Makromol. Chem, Rapid Commun., 7,591.
Gilmore, P.T.,R. Falabella, and R. L. Laurence (1980). Macromolecules, 13, 880.
GREEN
Graessley, G. G., and S. F. Edwards (1981). Polymer, 22, 1329.

Graessley, W. W. (1974). Adv. Polym. Sci., 16, 1.
Graessley, W. W. (1980). J. Polym. Sci., Polym. Phys. Ed., 18, 27.
Graessley, W. W. (1982). Adv. Polym. Sci., 47, 68.
Graessley, W. W. (1983). Roy Soc. Chem. Faraday Div.; Faraday Symp., 18, 1.
Green P. E (1991). Macromolecules, 24, 3373.
Green, P. F., and B. L. Doyle (1986). Phys. Rev. Lett., 57, 2407.
Green, P. F., and B. L. Doyle (1987). Macromolecules, 20, 2471.
Green, P.F.,andP.F.Doyle (1990). Ion beam analysis of thin polymer films, in New
Characterization Techniquesfor Thin Polymer Films, H. M. Tong and L. T. Nguyen,
Ws., Wiley, New York.
Green, P. F., and E.J. Kramer (1986a). Macromolecules, 19, 1108.
Green, P.F.,andE. J. Kramer (1986b). J. MateE Res., 1, 202.
Green, P. F.,P. J. Mills, C. J. Palmstrom, J. W. Mayer, and E. J. Kramer (1984). Phys.
Rev. Lett., 53, 2145.
Green, P. F., C. J. Palmstrom, J. W. Mayer, and E. J. Kramer (1985). Macromolecules,
18, 501.
Green, P. F.,P. J. Mills, and E. J. Kramer (1986). Polymer, 27, 1063.
Green, P.F., T. P. Russell, M. Granville, and R. Jerome (1988). Macromolecules, 21,
3266.
Green, P.F., T. P. Russell, M. Granville, and R. Jerome (1989). Macromolecules, 22,
908.
Green, P. F., D. B. Adolf, and L. R. Gilliom (1991). Macromolecules, 3377.
Helfand, E. (1992). Macromolecules, 25, 492.
Helfand, E., and D. S. Pearson (1983). J. Chem. Phys., 79, 2054.
Hess, W. (1986). Macromolecules, 19, 1395.
Hess, W. (1987). Macromolecules, 20, 2587.
Hess, W. (1988a). Macromolecules, 21, 2620.
Hess, W. (1988b). Macromolecules, 21, 2587.
High, M. S., P. G. Painter, and M. M.Coleman (1992). Macromolecules, 25, 797.
Jones, E. B., A. Jones, and Ingerfield (1994). J. Polym. Sci.: Polym. Phys. Ed.
Kausch, H. H., and M. Tirrell (1989). Ann. Rev. Mater Sci., 19, 341.
Kimmich, R., and R. Bachus (1982). Colloid Polym. Sci., 260, 911.
Kirkendall, E. D. (1942). Trans. Am. Inst. Min. Metall. Eng., 147, 104.
Klein, J. (1978). Macromolecules, 11, 852.
Klein, J. (1986). Macromolecules, 19, 105.
Klein, J. (1990). Science, 250, 640.
Klein, J., and B. J. Briscoe (1979). Proc. Roy. Soc. Lond. A, 365,53.
Klein, J., D. Fletcher, and L. J. Fetters (1983). Nature, 304, 526.
Kolinski, A., J. Skolnick, and R. Yaris (1986). J. Chem. Phys., 1922.
Kolinski, J. Skolnick, and R. Yaris (1978a). J. Chem. Phys., 86, 7164.
Kolinski, J. Skolnick, and R. Yaris (198%). J. Chem. Phys., 86, 7174.
Kramer, E. J., P. F. Green, and J. W. Mayer (1984). Polymer, 25, 473.
Kremer, K, and K Binder (1984). J. Chem. Phys., 81, 6381.
Kremer, K, and G. S. Crest (1990). J. Chem. Phys., 92, 5057.
MELTS IN 301
Kremer, K., G. S. Grest, and I. Carmesin Phys. Rev. Lett., 61,

Kumagai, Y.,H.Watanabe, K. Miyasaka, and T. Hata J. Chem. Eng. Jpn., 12,
Lefebvre, D., B. Jasse, and L. Monerrie Polymer, 25,

Leibler, L, Macromolecules, 13,
Lin, Y.-H. Macromolecules, 17,
Lm,Y.-H. Macromolecules, 19,
Lodge, T.P., N. A. Rotstein, and S. Prager Adx Chem. Phys., 74.
McCall, D.W., D. C. Douglass, and E. W. Anderson J. Chem. Phys., 30,
McKenna, G. B., K. L. Ngai, and D. J. Plazek Polymer, 26,
Marrucci, G. J. Polym. Sci., Polym. Phys.Ed., 23,
Mills, P. J., P.F. Green, C. J. Palmstrom, J. W.Mayer,andE. J. Kramer Appl.
Phys. Lett., 53,
Mills, P. J., J. E. Mayer, E. J. Kramer, G. Hadziiaounnou, P. Lutz, C. Strazielle, Rempp,
and A. J. Kovacs Macromolecules, 20,
Murschall, U., R.W.Fischer,andHerkt-Maetzky J. Polym. Sci.: Polym. Lett.,
Muthukumar, M.,and A. Baumgartner Macromolecules, 22,

Needs, R. J., and S. F. Edwards Macromolecules, 16,
Nemoto, N., M. Landry, I. N. Icksam, and H. Yu Polym Commun., 25,
Ngai, K. L., A. K Rajagopal, and S. Teitler J. Chem. Phys., 88,
O’Connor, N. T. P., and R. C. Ball Macromolecules, 25,
Pearson, D. S. Rubber Chem. Technol., 60,
Pearson, D. S., and E. Helfand Macromolecules, 17,
Pearson, D. S., G. Ver Strate, E. von Meerwall, and F. C. Schilling Macromol-
ecules, 20,
Rouse, P. E. J. Chem Phys., 21,
Rubinstein, M. Phys. Rev. Lett., 59,
Rubinstein, M., and R. H.Colby J. Chem. Phys., 89,
Rubinstein, M., E. Helfand, and D. S. Pearson Macromolecules, 20,
Schweizer, K S. J. Chem. Phys., 91,
Shull, K R., E. J. Kramer,G.Hadziiaounnou,M.Antonietti,andH.Sillescu
Macromolecules, 21,
Shull, K R., E. J. Kramer, F. S. Bates, and J. H. Rosedale Macromolecules,
Sillescu, H. Macromol. Chem., Rapid Commun., 5,

Sillescu, H. Macromol. Chem, Rapid Commun., 8,
Skolnick, J., A. Kolinski, and R. Yaris Acc. Chem. Res., 20,
Smith, B. A. Macromolecules, 15,
Smith, B. A., E. T. Somulski, L. P. Yu, and M. A. Winnick Phys. Rev. Lett., 52,
Struglinski, M. J., and W. W. Graessley Macromolecules,

Tead, S. F., E. J.Kramer, G. Hadziioannou, M Antonietti,H.Sillescu, P. Lutz,and
Strazielle Macromolecules, 25,
Tirrell, M. Rubber Chem. Technol., 57,
302 GREEN
Van Asten, J. G., and S. R. Lustig (1992). Macromolecules, 25, 5069.

Viovy, J. L. (1985). J. Phys. (Les Ulk), 46, 847.
von Meerwall, E. (1991). J. Non-Crys. Solids, 131, 735.
von Meerwall, E. D., E. J. Amis, and J. D. Ferry (1985). Macromolecules, 18, 260.
vonMeerwall, E.,J Grigsby, D. Tomich,and R. VanAntwerp (1982). J. Polym. Sci,
Polym. Phys. Ed., 20, 1037.
von Seggern, J., S. Klotz, and H.-J. Cantow (1991). Macromolecules, 3300.
Watanabe, H., and M. Tirrell (1989). Macromolecules, 22, 927.
Ye, M., R. J. Composto, and R. S. Stein (1990). Macromolecules, 23, 4830.
Zawada, J. M. Ylitalo, G. G. Fuller, R. H.Colby,andT. E. Long (1992). Macro-
molecules, 25, 2896.
Index
Activation energy, Case II,

Activity, Cellulose, cuprammonium
Adsorption, regenerated,
Affinity, Cellulose regenerated,
After-effect function,
Aging, Chemical potential,
Anti-plasticizer, Chromatography,
Anisotropy, capillary column,
Averaging methods, gas-liquid,
Axial dispersion, gas-solid,
Cluster
connectivity,
Binary mixtures, shape, 85
Block copolymers, size,
Bond angle, 55 percolation,
Bond length, 5 5 , 5 8 Cohen and Turnbull,
Branched polymers, Configuration,
Burnett coefficients, accessible volume, 80
bias, Monte Carlo,
random coil,
Cannonball solid, unoccupied volume,
Capillary model, Conjugate force-flux,
303
304 INDEX
Connected overlapping spheres, [Diffusion]

Conservation of species equation, multicomponent,
non-Fickian,
Constraint forces, pure fluids,
Constraint release, statistical mechanics,
surface,
Correlation function Diffusion coefficient,
end-to-end vector, binary,
length, concentration dependent,
Critical Einstein,
lower critical solution Fickian,
temperature Green-Kubo,
upper critical solution interdiffusion coefficient,
temperature (UCST),
Cross-coupling coefficients, measurement techniques,
Cross-kinetic coefficients, mutual,
Cross-linked polymers, polystyrene/polyvinylmethyl
Crystallinity, ether,
reptation,
Crystallization, polymers, ring-shaped polymers,
self,
Crystallization, solvent induced,
NMR,
polyethylene,
Darken equation, predicted,
Deborah number, star-shaped polymers,
Desorption, tracer diffusion coefficient,
Devolatilization,
Diffraction pattern, prediction, polymethyl methacrylate,
Diffuse reflection,
Diffusion polystyrene,
amorphous polymer, transition state theory,
anomalous, transport,
collective, unentangled chain,
flux, Dihedral angle,
hopping mechanism, Dispersion coefficient,
Knudsen, Dissolution,
low temperature, Dividing surface,
mechanism, Dry molding,
memory-dependent, Dual-mode sorption,
molecular simulation, Ductile failure,
molecular view, Dusty gas model,
INDEX 305
Dynamic loss, tangent, 233 Glassy polymer, 184

Dynamic modulus, 233 configuration, 70
molecular mechanics, 74
transition, 57, 60, 156, 162, 198,
Elastic diffuse scattering, 19 234-236
Elastic recoil detection, 260
Electrostatic field, 196
Elution times, 200 Half-time, 174
Ensemble, microcanonical, 16 Heat of sorption, 98
Equilibrium, local, 4, 9, 14 Heat treatment, 245
Equilibrium, mechanical, 8 Henry’s law, 98, 176
Exponents Hierarchical modeling, 139
critical, 48 Heredity, 187
scaling, 48 Hole-filling, 194
Extended chain, 218, 223
Interaction potential
Fiber, 221, 222 hard core, 18
Fibril, 223, 224 hard sphere, 16
Fick’s law, 174 Lennard-Jones, 16, 28, 50
Finite difference, 18, 181 London-van der waal, 26
Crank-Nicholson, 182 particle-porewall, 18
forward time central space, 182 Interchain
semi-implicit, 182 rotation, 55
Finite element, 182 vibration, 55
Flexibility chain, 57 Interfacial instability, 197
Flory-Huggins, 150, 255, 285 Interpenetrating network, 202
Fluxes, molecular, 6 Interpore connectivity, 37
Fluctuations, temperature, 6 Isotropy, 10, 11
Foams, polymer, 195
Folded chain, 218, 222, 223
Force balances, 189 Jump balances, 183
Frames reference, 6
Frames, flow, 8
Free volume, 57, 155, 216 Kevlar, 223, 225
Friction factor, monomeric, 251, Kinetic theory, 4
254,258, 269 Kinetic temperature, 29
Fringed micelle, 218, 222
Fujita model, 155
Lamella, 197, 218, 223
Langmuir adsorption isotherm, 192,
Gibbs-Duhem, 8, 12 194
306 INDEX
Laplace transform, 189 [Molecular simulation]

Lateral order, 222, 223 scope, 68
Linear response theory, 6, 24 sorption, 67
Local twisting motion, 237 isotherm, 96
Long-time tail, 21, 45, 48, 49 Molecular weight
Loop, 223 critical, 252
Loretz gas, 48, 49 Monte Carlo simulation, 40, 76
concerted rotation, 76
configuration bias, 76
Master equation, 124 Gibbs ensemble, 103
Maxwell-Boltzmann, 20 kinetic Monte Carlo, 125
Melting, 234 reptation, 76
Melt molding, 231 sorption, 102
Membranes, 174, 225 Moving boundary problems, 183
desalination, 196 Multicomponent diffusion, 165
hyperfiltration, 196
reverse osmosis/asymmetric, 196,
197, 225, 226 Nematic, 220
skin, 196 Newton’s laws of motion, 17, 56
ultrafiltration, 196 Nodule, 221
Micro Brownian motion, 233, 235 Noncrystalline, 217, 219, 220, 221,
Microfibril, 223, 224 233, 245
Micro phase separation, 232 Nonequilibrium statistical
Microscopic reversibility, 124 mechanics, 4
Microvoids, 192 Nonequilibrium thermodynamics, 4
Mobility, 12 Nuclear magnetic resonance, 260,
Mode theory 262
fast, 285 Nuclear reaction analysis, 260
slow, 285 Numerical solution, 181
Modulus
plateau, 252, 268, 269
shear stress relaxation, 252 Onsager coefficients, 283
Molecular dynamics simulations, Onsager reciprocal relationship, 4, 5
12, 28, 40, 55, 62, 75 Orientation, 223-224, 244, 245
nonequilibrium, 118 Orientational decorrelation, 88
sorption equilibria, 102 Orthogonal collocation, 183
unconstrained technique, 57 Oscillatory, 178, 189
Molecular model, 72
Molecular simulation
amorphous cell, 72 Packing density, 220
diffusion, 67 Pair distribution function, 77
mechanics, 74 Partition coefficient, 14, 15, 31, 242
INDEX 307
Phenomenological coefficients, Pore density,

Plastic deformation, Pore radius distribution,
Plasticization Porous media,
molecular view, asymmetric membranes,
glass transition,
Percolation, cannonball solid,
Periodic imaging, dusty gas model,
Permeability, interpore connectivity,
Permeation, percolation, 44,
coefficient, Swiss cheese solid,
Poisson process, Positron annihilation spectroscopy,
Polyblends,
Polymers Potential energy,
aging, Power law,
amorphous, Primary particle,
annealing, Primitive path,
branched,
foams,
materials,
molecular engineering, Radio labeling,
semiatomistic, Radius, mean,
semi-crystalline, cutoff,
star-shaped, maximum pore,
molecular models, pore distribution,
validation of, Random media,
Poly-a-olefine, Random overlapping spheres,
Polyethylene, Random nonoverlapping connected
Polyethyleneterephthalate, spheres, 50
Random nonoverlapping spheres,
Polymethyl methacrylate, Random walk,
Poly-paraphenyleneterephthalamidel Relaxation
Kevlar, diffusion time,
Polystyrene, function, diffusional,
Polyvinyl acetate, molecular,
Polyvinyl methyl ether, segmental,
shear stress modulus,
Pore, stretched exponential,
fluid time, longest,
binary, Reptation,
continuum solvent, Monte Carlo,
mixtures, Ring-shaped molecules,
single component, Rouse model,
308 INDEX
Rutherford backscattering Stress-strain constitutive equation,

spectrometry, 260 192
relaxation modulus, 252 Structure
multilayer, 225, 226
regularity, 220
Structure factor
Scaling theory, 21 dynamic, 90
Scattering functionintermediate, 90 fine, 211, 213, 221
Secondary particle, 226 first, 213
Segregation limit fourth, 213, 225
strong, 290 second, 213, 230
weak, 290 third, 214
Self-sharpening, 180 Swelling, 189, 1944
Shear stress, 252
Silica, 50
Simulations, 13, 62
Skin concentration, 190, 191 Takayanagi’s model, 214, 218,
Slip, flow, 26 219
diffusive, 36 Tessellation
viscous, 29 Delaunay, 83
Small angle neutron scattering, 260 Voronoi, 83, 203
Smectic, 220 Thermal diffusion ratio, 9
Solubility, 190, 242 Thermal motion, 233
Widom test particle, 98 Thermodynamic acceleration, 283
Sorption, 174 Thermodynamic slowing down,
differential, 180 283, 287
heat, 98 Tie molecule, 223
dual-mode, 159 Times
Fermi gas, 100 characteristic molecular, 87
integral, correlation function, 6, 21
isotherm, 96 Transition free energy perturbation,
molecular dynamics, 102 130
molecular simulation, 67 Transition state theory of diffusion,
Monte Carlo, 102 121
sigmoidal, 185, 190 macrostate, 124
statistical mechanics, 91 prediction, 133
thermodynamics, 91 transition path, 127
Specular reflection, 19, 24 Transition rate constant for jumps,
Spherulites, 197 130
Stefan-Maxwell, 9, 31 Transport phenomena, 173
Strain, 189 Tube, 254
Stress, 189 length fluctuation, 274
INDEX 309
Ultracentrifuge, 204 Volume

accessible, 80
clusters, 81
van der Waals, 242 free, 111, 112
Vector percolation, 81
end-to-end, 258 redistribution, 87, 111
Velocity autocorrelation function, 7, unoccupied, 80
107
Velocity Verlet, 18
Virus removal filter, 213 Waiting time distribution, 49, 50
Viscosity Wave-vector dependence, 6
zero shear, 252, 253, 256, 259, Wet molding, 230
267, 269 Williams, Landel and Ferry, 148,
Vogel-Fulcher, 268 198,235

Diffusion in Polymers 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

Diffusion in Polymers PDF

Încărcat de

Drepturi de autor:

Formate disponibile

DIFFUSION

Diffusion in polymers/ edited by P. Neogi.

This book is printed on acid-free paper.

Copyright 1996 by Marcel Dekker, Inc. All Rights Reserved.

Marcel Dekker, Inc.

Current printing (last digit):

PRINTED IN THE UNITED STATES OF AMERICA

Another development lies in the area of transport phenomena. It is no longer

The Pennsylvania State University, University Park,

University College Dublin, Dublin, Ireland

Fukuoka Women’s University, Fukuoka, Japan

University of Missouri-Rolla, Rolla, Missouri

M. Air Products & Chemicals, Inc., Allentown,

In this equation is the local fluiddensity xi is the mass fraction of

Postulate I: Microscopically large though macroscopically smallvolume

where Ji and Xi are the conjugate f l u e s and forces.

Onsager's reciprocal relations: with a suitable choice conjugate forces and

For isotropic mediaandunder conditions of mechanical equilibrium, two

where the angular brackets represent averaging overthe unperturbed equilibrium

to the corresponding expressionprovidedby linear response theory (Green,

i.e., a kinetic coefficient determined by the mechanical properties j , in contrast

where qij(t)is the velocity correlation function (VCF)

Now consider the condition of mechanical equilibriumas expressedby the

where P is the local pressure in the system. TreatingV+,, - F, as the dependent

(V+, - Fi) - kTiV In T (i = 1, . . . , v)

where 6, is the Kronecker delta.

medium itself also treated as one ofthe components in the multicomponent

2. Density and temperature variations within a microscopic volumeelement of

where = n,,i.e., the numberdensityof the fluid within the medium(fluid

1. Microscopic properties, which cannot bereadilymeasuredexperimentally

The discussion in Section III is presented in three parts. As an introduction

SIMULATION AND MODELING OF DIFFUSION IN

where Kf is the partition coefficient for the fluid f defined by

ary of the control volume under investigation). Equation (37b) corresponds to

where $(t) is the instantaneous center-of-mass velocity of the pore fluid as a

and (2) the Lennard-Jones (12-6) interaction potential

The principal objective in the (classical) MD method is to solve Newton’s

For the hard-sphereinteractiongivenbyEq. (45) the force, F,,acting on

(vijo r,, < (48a)

ri(t + At) = ri(t) + At v,(t) + -21 At”,@) (50a)

vi(t + At) = v,(t) + -21 At [a,(t) + ai(t + At)]

Avi = - 2 7(vio riw)

of the fundamental cell (below

where M is the number of independent time origins employed in the averaging

velocity components of the center-of-mass of the fluid using the expression

where k = xy or Due to the absence of interparticle collisions, cross-

section shown in Fig. 2 (nf - -

where ri is the radial positionofthefluid particle within the pore, eiWis

and in the limit R, -

5 Model cylindricalpore structures. (a) Particle-pore wall continuum interactions.

by the solid in Eq. (49) to give

In the molecular dynamics simulation of a Lennard-Jones pore fluid subject

(1987)] should also be incorporated in the simulation code to improvethe com-

confirm the existence of viscous slip (nonzero D% in the limit h -

VTFJ - F, = (VTF, - FJ)B (60)

and vice versa. Also noting that in general

with L:;) given by

The Stefan-Maxwell form of the flux equations may also be employed to

The mutualdiffusioncoefficientsmay also be expressedin terms of the

D:: = k,T~(')l/n,Ly~ (65c)

where IL"I = Ly:Lfi - L??.

where Llf' is given by (42) and

Note that in the thermodynamic limit N