Sunteți pe pagina 1din 23

Journal of Nuclear Materials 69 (k 70 (1978) 38-60

0 North-Holland Publishing Company

ATOMIC JUMP PROCESSES IN SELF-DIFFUSION


H. MEHRER
Institut ftir theoretische und angewandte Physik der UniversitiitStuttgart

Max-Planck-Institutftir Metallforschung, Institut ftir Physik, Stuttgart, W. Germany

The atomic jump processes involved in the vacancy mechanisms of self-diffusion in metals are reviewed with particular
attention to disvacancies. The most important measurements which are helpful to separate mono- and divacancy contributions - temperature, mass, and pressure dependence of the diffusion coefficient and correlation effects - are discussed. The
recent experimental progress will be considered also. The extension of direct tracer studies to much lower temperatures has
greatly increased the reliability with which monovacancy properties may be deduced. Amongst the indirect techniques like
nuclear-magnetic-relaxation, Mijssbauer effect and quasi-elastic neutron-scattering, especially nuclear magnetic relaxation
may be considered nowadays as a quantitative tool. In a discussion of individual metals the above mentioned topics will be
illustrated by examples, with emphasis on those metals where a considerable deepening of our understanding of atomic
jump processes has been achieved.

Les processus de saut atomique quimpliquent les mecanismes lacunaires dautodiffusion dans les m&aux sont pass& en
revue en portant une attention particuliere sur les bi-lacunes. Les mesures les plus importantes qui permettent de s&parer
les contributions des monolacunes et desbilacunes, les relations entre coefficient de diffusion et les variables tempkrature,
masse et pression et les effets de correlation sont discuties. Les progris exp&imentaux &cents seront aussi consid&s. Lextension des Etudes directes par traceurs B des tempiratures beaucoup plus basses a augment6 considkrablement la pouibilitd
de d6duire les propri&s des monolacunes. Parmi les techniques indirectes, comme la relaxation mag&tique nucldaire,
leffet Mllssbauer et la diffusion quasi-hlastique des neutrons, sp&ialement la relaxation magn6tique nuclhaire, peuvent
6tre consid&es maintenant comme des outiles quantitatifs. Dans une discussion concernant certains m&aux, les methodes
mention&es cidessus seront ill&r&es par des examples en mettant laccent sur ceux des r&taux pour lesquels an approfondissement considerable de notre comprehension des processus de saut atomique a 6th obtenu.

Die atomaren Sprungprozesse, die beim Einfach- und Doppelleerstellenmechanismus


der Selbstdiffusion in Metallen auftreten, werden unter besonderer Beriicksichtigung des Doppelleerstell~nmechanismus
betrachtet. Die wichtigsten Messungen, die zur Trennung von Einfach- und DoppelleerstellenbeitrLgen hilfreich sind - Temperatur-, Massen- und Druckabhiingigkeit des Diffusionskoeffizienten sowie Korrelationseffekte - werden diskutiert. Der neueste experimentelle Fortschritt wird ebenfalls betrachtet: Die Ausdehnung direkter Tracer-Messungen zu sehr kleinen Diffusionskoeffizienten
hat
die VerlBsslichkeit, mit der Eiienschaften der Einfachleerstelle bestimmt werden kiinnen, stark erhiiht. Unter den indiiekten
Techniken wie Kernspinrelaxation, MGssbauer-Effekt und quasielastische Neutronenstreuung kann insbesondere die Kernspinrelaxation heutzutage als quantitatives Werkzeug angesehen werden. In einer Diskussion einzelner Metalle werden die
oben erwghnten Punkte anhand von Beispielen erkXutert, wobei solche Metalle herausgegriffen werden, bei denen eine Vertiefung des VerstBndnisses atomarer Sprungprozesse erreicht wurde.

fusion (and also diffusion of substitutional impurities)


occurs by a series of exchange jumps of individual
atoms with vacant lattice sites [ 11. For many years selfdiffusion has been interpreted in terms of monovacanties alone. Whereas the monovacancy mechanism

1. Introduction
Self-diffusion in crystals is one of the most important manifestations of point defects in thermal equilibrium. In metals it is generally agreed that self-dif38

39

H. Mehrer/Atomic jump processesin selfdiffusion


indeed dominates over a wide temperature range it
has become clear in recent years that for most metals
a divacancy contribution is observable near the melting
temperature (for reviews see, e.g. [2-5]). Therefore,
when we relate measurabie quantities like the diffusion coefficient or NMR relaxation rates to atomic
properties of the crystal, we learn something about
vacancy-type defects.
In the general discussion of section 2 we consider
the atomic jump processes involved in the mono- and
divacancy mechanism of self-diffusion in metallic
structures with particular emphasis on recent improvements of the theory of divacancy diffusion. The
tracer self-diffusion coefficient due to mono- and
divacancy migration and its dependence on temperature, hydrostatic pressure and isotopic mass will be
discussed in detail for cubic metals. In section 3 some
remarks will be made about the recent progress in
tracer-measurements
of very small diffusion coefficients due to microsectioning techniques and the
resulting deepending of our understanding of diffusion mechanisms. Section 4 contains a critical survey over various indirect techniques for the study of
atomic jump processes in self-diffusion including
nuclear-magnetic-relaxation,
Mossbauer effect and
quasi-elastic neutron scattering. Owing to recent improvements of the theory nuclear magnetic relaxation
especially may nowadays be considered as a reliable
tool for measuring self-diffusion. In section 5 we
turn to a discussion of individual metals with particular emphasis on recent developments and on those
metals for which fairly definite conclusions on defect
properties may be drawn.

2. General discussion of selfdiffusion


2. I _ General remarks and diffusion mee~~isrn
The transport ofmatter which accompanies the
motion of vacant lattice sites can be described by the
so-called macroscopic coefficient of self-diffusion DSD.
It is related to the mean square displacement of the
diffusing atoms and consequently to the jump frequancies and jump distances of the atomic jumps. According to random waik theory (see, e.g. f&7]) we

have

(2.1)
where N is the number of different types of jumps.
r, (a: = 1 ) .... iV) denotes the number of jumps of type
(Ymade by an atom per unit time and Ax, the x-projection of the pertaining jump distance. For a defect
mechanism of diffusion r, is given by
>

r, =ca?,

cm

where c, denotes the atomic concentration of defects


present at thermal equilibrium in a configuration
which permits an o-type jump of a given atom. v, is
the jump frequency involved.
The self-diffusion coefficient obtained from tracer
experiments, DT, is different from DsD. As first
pointed out by Bardeen and Herring IS] a quantitative
measure of this distinction is the co~e~tion factorf.
It accounts for the spatiai correlation between successive jump directions of tracer atoms and leads to a
reduction of the tracer diffusion coefficient with
respect to the mass-transport coefficient. In cubic
crystals we have
DT =fz)=

12.3)
whereas in hexagonal crystals two tensor components
parallel (II)

DT,I = f

11
@D,

and perpendicular

(2.4a)
(1)

DT,~ = f DSW

(2.4b)

to the hexagonal axis must be distinguished. The correlation factor(s) is (are) characteristic for a given diffusion mechanism and may be calculated if the jump
frequencies of the atoms involved are known. The
methods for calculating correlation factors have been
reviewed by LeClaire [9] and Mehrer [lo] and will
not be discussed here. Correlation factors have been
worked out for almost all cases of practical interest
and will be discussed in section 2.3.
2.1 .I. Monovacancy mechanism
In thermal equilibrium the concentration of monovacancies in a monoatomic crystal is given by
CIV =exp(Sj"v/k)exp(-~~~/kT)y

(2.5)

H. Mehrer /Atomic jump processes in self-diffusion

40

with HFv and Srv denoting the enthalpy and entropy


of formation.
In any of the three cubic Bravais lattices the motion of the monovacancy is characterized by the
monovacancy jump frequency to nearest-neighbour
sites in the lattice
(2.6)

where H~v and Sf;2r denote enthalpy and entropy of


motion, and $V the attempt frequency. According
to (2.3) the self-diffusion coefficient of tracers may
be written as
NV

=fiv

CIV hv a2 ,

(2.7)

where a is the cubic lattice constant andfiv the monovacancy correlation factor (fiv = 0.781 in an fee and
fiv = 0.723 in a bee structure [l 11).
In hexagonal close-packed structures two different
jump frequencies must be considered - one (v*v,A)
for jumps within and another (v~~,~) for jumps obliqu
to the basal plane. The components of the tracer selfdiffusion coefficient may be written as
DTG = 3;~ clv vlV,B

c:c

*2v12

=2c%v*v21

>

(2.8)

where v2vfj denotes the jump frequencies which


transform the divacancy from an ith nearest to a jth
nearest neighbour configuration. Using this relation

(2.7a)

C*

and
D;$

parameters contained in it by comparison with experiments.


In the fee structure there are four lattice sites that
are nearest neighbours to both sites of a vacancy pair
on adjacent sites (In-configuration
of the divacancy).
This may be the reason why in the literature it has been
assumed that the divacancy migrates by nearest-neighbour jumps without changing its configuration.
Whereas this simple model of divacancy migration
is indeed most likely, additional possibilities, e.g.
additional bound configurations, may exist. A more
general mechanism which includes bound divacancy
configurations at first- and second-nearest neighbour
sites (with concentrations C:$ and CiF) is illustrated
in the upper part of fig. 1. In thermal equilibrium the
divacancy concentrations are related by

V2Vf2

= @f:v ClV(3%V,A + hv,da

(2.7b)

"2vrr

with a and c denoting the hexagonal lattice constants.


The correlation factor components _f/? and f:v are
functions of the ratio VIV,A/YIV,B and have been calculated by Mullen [ 121 (see also [ 131).
2.1.2. Divacancy mechanism
Diffusion via bound pairs of vacancies is more complex than monovacancy diffusion. In general several
configurations of the pair with different binding
energies may be present in thermal equilibrium and
atomic jumps over more than one type of saddle point
may contribute to its migration even in the case of
cubic lattices. Although various theoretical calculations concerning divacancy configurations and movements have been performed they are based on interatomic potentials that are not sufficiently reliable to
permit a decision as to which one of the various possibilities prevails in a given metal or even in a given
structure. The best approach may thus be to work out
the consequences of a fairly general divacancy mechanism for each structure and to determine the

In

0.31

.2

.I

"2vrr
v2v12

2n

.6

.8
)

.8

.6

.4

.2

-vzytl
Vzvrr

Fig. 1. Divacancy mechanism of self-diffusion inan fee lattice,


correlation factor and maximum isotope effect.

41

H. Mehrer /Atomic jump processes in selfdiffusion

the diffusion coefficient for tracer motion by divacancies may be written as


T

f)sv

-22
- Ja

In

Czv(vav1t

+ v2v12)fzv

(2.9)

The correlation factor.f2v is a function of the ratio


v2~ll/u2~12 shown in the lower part of fig. 1. (For
details of the calculation see [lo] .) In the case of the
simple mechanism (vav 11 jumps prevailing) fav
approaches the value calculated earlier by Howard
[ 141, Bakker [15], and Mehrer [ 161. However, as
soon as the divacancy has an additional migration
mode the tracer motion is less correlated and fav is
temperature dependent instead of being just a number.
In the bee sfruccure there are no lattice sites that
are nearest-neighbours
to both sites of a vacancy pair
on adjacent sites. This means that a In-divacancy cannot even move (by nearest neighbour jumps of the
individual vacancies) unless additional non-nearest
neighbour con~gurations exist. Mehrer El 71 considered three bound divacancy configurations at first-,

ln

2n

second- and fourth-nearest-neighbour


sites (concentrations C&, C$$ and c*,$) and the atomic jump
frequencies shown in the upper part of fig. 2. This
fairly complicated mode of divacancy migration is
not without theoretical support [l&20].
In thermal
equilibrium the detailed balancing relations
3GGv2~12

(2.10a)

=4C%2~21

and
12&2~24

(2SOb)

= GCy2v42

hold and allow the tracer diffusion coefficient to be


expressed in terms of the second-nearest neighbour
configuration according to [ 17 ]

DTv = 2 a2%@2V21

(2.11)

+ v2V24)f2V

The correlation factor f2v is shown in the lower part


of fig. 2 as a function of v2v21~~2v24.

In

0.466

0.3
0

cl.33
.2

.I

.6

-3uL
iV.?L

.6

.B

0.4

0.6

0.8

7.0

0.8

0.6

0.L

0.2

.6 .4 .Z

%w--

Fig. 2. Divacancy rne~~an~rn of

0.2

selfdiffusion

fvzv,*a
-)

VZV,dR
in a bee lattice,

correlation factor and maximum isotope effect.

Fig.

-%,A.4

3. Divacancy mechanism of se~~iffu~on in

vrv,ns

a hcp lattice,

correlation factors parallel and perpendicular to the c-axis

[ 131.

H. Mehrer /Atomic jump processes in self-diffusion

42

The divacancy mechanism in the hcp structure has


been considered by Steiner et al. j13]. As shown in
fig. 3, two bound divacancy can~gurations may be
distinguished: an A-configuration (concentration
C&)
where both vacancies occupy nrlrest-neighbour
sites
in the same basal planes, and a B-configuration (concentration C&V, where the two vacancies are located
in adjacent sites of two nei~bouring
basal planes.
The migration of the divacancy as an entity involves
four atomic jump frequencies shown in fig. 3. The
tensor components of the self-diffusion coefficient of
tracer motion by divacancies may be written as [ 131
D#

(2.12a)

= c2C%v~~,~Bf8v

and

(2.12b)
where the correlation factor components are multivalued functions of the atomic jump frequencies shown
in the lower part of fig. 3.
2.2. Temperature dependence of self-diffusion
Within a limited temperature range self-diffusion
data may often be represented with sufficient accuracy
by an Arrhenius law
(2.13)
where both the pre-exponentia1 factor Dzff and the
activation enthalpy QeT ;e taken as independent of
temperature (k denotes Boltzmanns constant).
By inserting eqs. (2.5) and (2.6) into eq. (2.7), we
obtain for the activation enthalpy of self-diffusion by
monovacancies
Q1v=H,v

M
fHlV

(2.14a)

and for the corresponding

pre-exponential

factor
(2.14b)

of cubic crystals. The interpretation of measured values of Qeff and Dgff in terms of eq. (2.14) has sometimes been called the standard interpretation
of selfdiffusion.
However, deviations from an Arrhenius behaviour
appear to be an almost common feature of self-dif-

fusion in metals. For the so-called anomalous bee


metals like /3-Ti, &Zr and V1 where the deviations are
fairly strong, this has been known for many years (see,
e.g. [l]). Considerable curvatures of the Arrhenius
plot have also been observed for the alkali metals
(see section 5). The smallest curvatures are found in
the fee metals. However, the extension of diffusion
measurements to lower temperatures with the help
of microsectioning techniques (see section 3) and improvements of the experimental accuracy have permitted their observation. A good example is provided by
the self-diffusion data on silver, where four studies
of three independent groups cover almost ten orders
of magnitude in the tracer diffusion coefficient
[21-23,117].
There are several possible causes for a curvature of
the Arrhenius plot of bulk self-diffusion. In cubic
metals the most important ones are * : (i) mono- and
divacancy contributions to self-diffusion, and (ii) temperature dependence of the activation parameters. In
hexagonal metals the components of the tracer selfdiffusion coefficient even for a monovacancy mechanism will in general not obey an Arrhenius law.
When the migration enthalpies for A- and B-jumps are
different, we expect from eq. (2.7) deviations due to
the superposition of two Arrhenius-terms in eq. (2.7b)
and due to the temperature dependence of the correlation factor. In the following subsections we confine
ourselves to cubic crystals and consider each of the
above mentioned reasons for non-Arrhenius behaviour
in some detail. The extension to hexagonal crystals is
easily performed.
2.2. I. Simultaneous action of mono- and divacancies
When both mechanisms operate simultaneously the
tracer diffusion coefficient is given by
DT=D+D&.

(2.15)

Since the monovacancy mechanism has the lower


activation enthalpy it always predominates at lower
temperatures. With increasing temperature D&/DTv
increases. Whereas L>Tv for cubic metals obeys an
Arrhenius Law, D&Z may in general already be a super* A trivial cause for a curvature of the krrhenius plot at low
temperatures is along short circuits like grain boundaries and dislocations. If highly perfect single crystals and/or
the microsectioning techniques discussed in section 3 are
used, the influence

of short circuits may be eliminated.

H. Mehrer /Atomic

jump processes in self-diffusion

position of various Arrhenius terms with slightly different activation enthalpies and may imply a temperature dependent correlation factor [see eqs. (2.9) and
(2.1 l)]. However, since the divacancy contribution is
often only a small correction term in DT it may be
difficult to resolve the details of the divacancy mechanism from an analysis of the temperature dependence. On the other hand correlation and mass effects
discussed in section 2.3 are more sensitive to such
details.
For those fee metals where it is sufficient to consider the simple divacancy mechanism, eq. (2.15)
reduces to a superposition of two Arrhenius terms
DT=Dy

exp(-g)+Di

exp(-z),

(2.15a)

where the abbreviations in eq. (2.14) for the monovacancy parameters and
Q2v = Wyv - H::
D!: = 4f2va2v$

exp

+ H% ,
2sF; + A&v + s %
k

D:v +Qzv---D;v ,
DT

of the activation

DT

enthalpies

H(T) = H(TO) + cwk(T- TO) + Pk(T - TO)2 t ... ,


(2.18)

(2.16)

of the two mechanisms.

2.2.2. Temperature dependence of activation


parameters
In the preceding discussion we have implicitly
assumed that defect enthalpies and entropies are
independent of temperature. However, all equations
remain valid if this assumption is not made. A priori,
there is little reason to exclude the possibility of a
temperature dependence of the defect parameters.
The only thermodynamic
requirement is that the temperature variations of enthalpies and entropies are
related according to

(E),=T(%lp.

Since eq. (2.17) defines a specific heat, a temperature


variation of the defect parameters is equivalent to the
statement that there is an additional specific heat
associated with defect formation and motion. Hence
at temperatures well below the Debye temperature,
where quantum rather than classical statistics must
be used, defect parameters will be temperature
dependent. Above the Debye temperature the defect
parameters are temperature independent as long as
the harmonic approximation can be used. Anharmonicity effects which manifest themselves, e.g. in
thermal expansion, give rise to an increasing relaxation of the defect with increasing temperature. This
means that the defect entropy and because of eq.
(2.17) also the enthalpy may increase with temperature.
Since the expected variations for normal metals
are rather small we may expand the enthalpy in a
Taylor series [3,4] as

(2.1 Sb)

for the divacancy parameters have been used. Hyv


and HFv denote the migration and binding enthalpy
of a nearest-neighbour
divacancy, Syv is the migration and AS,, is the association entropy of the divacancy. & is the pertaining attempt frequency. The
effective activation enthalpy defined by Qeff s - d In
DT/d( l/kZJ is a weighted average
Qeff=Qlv-

43

(2.17)

where T,-,is a reference temperature and o and /3 are coefficients.


De Vries [24] has stressed the possible significance
of the quadratic term in eq. (2.18). However, theoretical estimates by Levinson and Nabarro [25], Girifalco [26] and Flynn [27] indicate that the temperature variation of the formation enthalpies is very
small for close-packed metals (typically of the order
of 0.01 eV between room temperature and melting
point). Moreover, Franklin [28,29], who included
quantum and anharmonicity effects into the statistical mechanical approach to reaction rate theory of diffusion, obtained that the pre-exponential factor 0: of
copper self-diffusion varies by less than 20% over a
range of ten orders of magnitude in DT. The associated
entropy variation according to eq. (2.17) corresponds
to a variation of the activation enthalpy which is less
than 0.02 eV. A procedure how such small variations
can be included into the analysis of diffusion data, if
necessary, has been worked out by Seeger and
Mehrer [2].
Recently Gilder and Lazarus [30] claimed that the
whole curvature in the Arrhenius-plot of self-diffusion is explainable in terms of a single highly relaxed
vacancy-like defect in which the anharmonicity of the

44

H. Mehrer /Atomic jump processes in self-diffusion

means that one has to be careful if one compares


activation enthalpies measured well below the Curie
temperature with measurements in the paramagneti~
region.
Ruth et al. 1321 have proposed an expression in
which the deviation from an Arrhenius law is related
to the ferromagnetic order-parameter R. For monovacancy diffusion, which certainly predominates in
the ferromagnetic region, their result may be written
as

lattice modes gives rise to a large thermal expansion


of the defect. Positive thermal expansion coefficients
of the defect which are as much as 15 times larger
than those of the crystal itself are postulated. However, in the present authors view a decrease of the
defect volume with respect to the atomic volume
should occur rather than an increase when the defect
configuration becomes more relaxed with increasing
temperature. Gilder and Lazarus argue that the large
defect expansion coefficient is supported by the observation that the activation volume of self-diffusion
increases with temperature. However, as outlined in
section 2.3, this effect can be explained in a quite
natural way by the simultaneous contributions of
mono- and divacancies to self-diffusion.
In ferromagnetic metals a temperature variation
of the activation parameters must be expected due
to the infIuence of ferromagnetic ordering. In contrast to the variations discussed above this may be a
big effect. Clear-cut experimental evidence for this
has become available only very recently, since precise
diffusion experiments in the ferromagnetic region
necessitate appropriate microsectioning techniques.
An example is provided by the measurements of
Mehrer et al. [31] on pure iron. The diffusion coefficient in the ferromagnetic region decreases more
rapidly with temperature than an Arrhenius extrapoiation of the paramagnetic data would suggest. This

DT =DTv = 07 exp[-QPva(l

f ~R2)/k7],

(2.19)

where @F denotes the activation enthalpy in the


paramagnetic region and 7 is a dimensionless parameter which may be determined from a comparison
with the experimental data.
2.3. Correlation and the isotope effect
Since the correlation factor is not the same for different diffusion mechanisms its determination may
help to establish the diffusion mechanism (see table 1).
Accurate measurements of DT and DSD according to
eq. (2.3) are in principle capable of giving
f = LPfP

(2.20)

and providing this information. An example for this


are measurements on Li discussed in section 5.

Table 1
Correlation factors and isotope effect for self-diffusion in metallic structures
Face-centered-cubic

Hexagonal-close-packed
Parallel c-axis

Perpendicular c-axis

fiV =0.723 [llf

fiv

fiv see 112,131

EIV =fivAK,v

.Eiv =ftvA&v,B

Eiv see [44,13]

fiv

flv

fh

see fig. 3

E& <g$v

see [I31

-_l. ..._.-

~_.
fiv=0.781

Body-centered-cubic

[ll]

Monovacancy
EIV =fivAK,v

see 1121

simple divacancy
Divacancy
fiv

= 0.468 [ 14-161

Ezv =f2vA&vWW

82v < g2v


&vacancy with partial
dissociation
f2v
see fig. 1
QV < g2v -

see fig. 2
and [17]

&

< f2v

see fii. 3
and {13]

H. Mehrer /Atomic jump processes in self-diffusion

The most important method for obtaining information on fin metals, however, is the measurement of
the isotope effect in diffusion (for a recent review see
[33]). A quantitative measure is the isotope effect
parameter defined by

Dal@-

EE

(mp/m,)i2 - 1

(2.21)

where Da and Dp are the diffusion coefficients of two


isotopes (Yand /I with masses ma and mp. WC consider
in the rest of this subsection the relationship between
the correlation factor and isotope effect parameter
(see also [34]).
For a defect mechanism of diffusion the correlation factor may be written as
Vi) ,

fT =f($,

(2.22)

where VT denote the jump-frequencies


of the tracer
and vi those of the matrix atoms, which are different
because of the different masses of tracer and matrix
atoms * .
In the case of a monovacancy mechanism in the
cubic lattices a unique jump frequency vTv = vT contributes to the tracer motion [see eq. (2.7)] and the
correlation factor has the form [35]
fT =

U(Vj>
VT.

U(Vj) +

(2.23)

As shown, e.g., in [33] eq. (2.21) may then be replaced


by
E,v

=f:v

AK,v ,

is not far from unity, in reasonable agreement with


experimental values. Feit [41], using the dynamical
theory of diffusion, deduced AKiv = 0.5 for sodium.
Recently Bennett [42] obtained from molecular
dynamic calculations utilizing pair-potentials appropriate for solid Ar a value of 0.89 and in preliminary
calculations on sodium [43] about 0.6.
In general several different jump frequencies of the
tracer will be present as for divacancy diffusion or for
monovacancy diffusion in the hexagonal close-packed
lattice. The relationship between the isotope effect
parameter and the correlation factor requires then a
generalization because of two reasons:
(1) Obviously the correlation factor can not be
written as eq. (2.23). As a consequence fT in general
no longer provides an upper limit for E. However,
according to Mehrer et al. [34] a corresponding upper
limit, which we denote by g, may be calculated from
g =f

(2.25)

The quantities g2v for divacancy diffusion in the


cubic lattices are also shown in figs. 1 and 2. In general gzv lies above fiv. Only in the case of a simple
divacancy mechanism in an fee lattice both quantities
coincide, where Bakker [15] and Mehrer [ 161 have
shown that the correlation factor has indeed the form
of eq. (2.23).
A generalization of eq. (2.23) has recently been
discussed in detail in [34]. When the correlation factor has the form
fT=

* In the preceding discussion we had neglected this difference


which is usually small in the case of self-diffusion. fT then
reduces to the so-called geometrical correlation factor fi

*If P(mplm,)1/2
- 1
(mplm,)'i2- 1

(2.24)

where AKrv is the kinetic energy factor introduced


by Mullen [36] and LeClaire [37]. AKiv denotes the
fraction of kinetic energy at the saddle point associated
with tracer motion in the jump direction. By definition it satisfies 0 < AKlv < 1. The deviation of
AKrv from unity is related to the relaxation of the
atoms surrounding the vacancy.
Theoretical estimates of AKrv based on the reaction-rate theory of diffusion by various authors
[38-401 indicate that in close-packed lattices AKrv

45

1 + G(vT, vi)

(2.26)

where G denotes a homogeneous function of degree


one in the tracer jump frequencies, and when quantum
effects in the atomic jump process are negligible, fT
still, provides an upper limit for E. To the authors
knowledge the only example in which eq. (2.26) is
fulfilled is diffusion by divacancies parallel to the
hexagonal c-axis [ 131.
(2) Different AK-factors pertaining to the tracer
jumps over different saddle points must be considered.
An example for this is monovacancy diffusion in the
hcp structure where two AK factors pertaining to
jumps either within or oblique to the basal plane may
be distinguished [44].
For divacancy mechanisms in cubic lattices it may

46

H. Mehrer /Atomic

jump processes in self-diffusion

to a first approximation suffice to use a unique value


AKzv . We then obtain for the isotope effect parameter
E2v

=gzv

AKav

(2.27)

where we expect AK2v < AKtv because of the


stronger relaxation around a divacancy as compared
to a single vacancy in agreement with theoretical calculations of Burton [44a].
When mono- and divacancies contribute simul.
taneously to diffusion the net isotope effect will be a
weighted average

E =fivAKlv

BTV

+g,vAKzv

&

(2.28)

Sincefiv
>gzv (see table 1) and since A&V Q AKrv
we expect E to decrease with temperature even if
AKiv are temperature independent. Due to anharmoniticy effects the AK factors may slightly decrease
with temperature. However, theoretical estimates [39]
suggest that this temperature dependence is small.
2.4. Pressure dependence of self-diffusion
Between the Gibbs free energy G of a solid and
its volume V the following thermodynamic relationship holds :
ac/ap = V,

(2.29)

where p denotes hydrostatic pressure. This relation


enables us to associate an activator delude AV to
each atomic jump processes. For a defect mechanism
of diffusion we have (see, e.g. (11)
AV=VF+p,

(2.30)

where VF denotes the volume of formation of the


pertaining defect ~on~g~ation
and p the migration
volume of the pertaining jump-type. Usually @ is
only a small fraction of the total activation volume.
From diffusion experiments under hydrostatic
pressure an effective activation volume may be determined according to
A Veff = --XT y

t correction

term

(2.31)

from the pressure dependence of DT. The correction


term is related to the pressure dependence of the
attempt frequency and amounts to not more than a

few percent of the first term on the right-hand side


of eq. (2.29) f45]. Experimental values of A Pff may
not identify the diffusion mechanism in a unique way.
They may, however, add evidence in favour of one or
another mechanism,
For a monovacancy mechanism of diffusion in
close-packed cubic metals general arguments [4] as
well as theoretical calculations f46J indicate that
A Vlv is between one-half and three-quarters of an
atomic volume. In the less dense-packed bee metals
the relaxation of the atoms surrounding a vacancy is
stronger and hence the expected values of Al/iv may
be somewhat smaller. For a divacancy mechanism
various activation volumes pertaining to the various
jump-types must be considered in principle. However, theoretical calculations by Schottky et al. [47]
indicate that the activation volume for a nearest-neighbour divacancy in noble metals is approximately
twice the value for the monovacancy. This result suggests that we may characterize to a first approximation the divacancy mechanism by a unique activation
volume A Vzv .
If more than one diffusion mechanism is operating
AVeff will vary with pressure and temperature. Equations for the simultaneous contributions of monoand divacancies have been discussed in detail by
Mehrer and Seeger [45]. The activation volume may
be written as

AVeff = Aviv

DTV

6
+ AVav 3

(2.32)

Since AV2v > AVlv with increasing temperature


eq. (2.32) predicts an increase of AVeff. A fit of
eq. (2.32) to experimental values yields AV,, and
AVzv and may put limits on the mono- and divacancy
cont~butions
to LfT.

3. Remarks on tracer measurements of selfdiffusion


A large amount of the experimental evidence on
atomic jump processes in pure metals is based on
direct measurements of the self-diffusion coefficient.
Direct methods for diffusion measurements in solids
are those in which the diffusion coefficient of tracer
atoms, DT, is obtained by comparing the actual
tracer redistribution resulting from a diffusion anneal

H. Mehrer /Atomic jump processes in self-diffusion

with the diffusion profile predicted from Ficks second


law.
For the boundary condition of a standard tracer
experiment (an infinitely thin tracer layer at
X = t = 0 on a surface of a semi-infinite
specimen)
this solution is

c=&t

exp-4x
t

X2

(3.1)

Here C denotes the tracer concentration, X is the distance from the surface, t is the diffusion time and Co
is a constant. In the most precise diffusion studies the
tracer distribution is measured with the help of an
appropriate serial sectioning technique which permits
material removal in thin layers parallel to the initial
surface and subsequent counting of the radioactivity
in each layer. Such measurements allow one to determine the tracer self-diffusion coefficient to a fairly
high accuracy of a few percent (see, e.g. [ 11) and
provide unambigous information on the diffusion
kinetics.
From the viewpoint of deducing reliable defect
properties from tracer experiments it is important to
base such an attempt on measurements which cover
a temperature range as wide as possible. In the high
temperature region between the melting temperature
T, and about fT,,, mechanical sectioning techniques
can be used. In this region the diffusion coefficient
varies for a typical metal between about 1O- cm2 /s
(near T,,,) and about lo-* cm*/s. In cases where the
half-life of the available tracer isotope limits the time
for the diffusion anneal it may even be difficult to
reach the lower limit.
Fortunately nowadays powerful microsectioning
techniques are available which permit serial sectioning
on a submicron level.
Since the pioneering work of Pawel et al. [48-501
various chemical and electrochemical techniques have
been developed (for a brief survey see, e.g., section 2.2
in [Sl]). These techniques.usually
consist of a twostep polishing process where oxide or halide formation is followed by dissolution of the oxide or halide
film in an appropriate solvent.
In very recent years sputtering methods have been
adapted for serial sectioning in tracer diffusion measurements. Two variants have been developed, one by
Gupta and Tsui [52] and another by Maier and Schtile

41

[53]. Gupta utilizes r.f. Ar-ion backsputtering in a glow


and applied it to lattice and short-circuit
diffusion studies in gold [54,55]. Maier uses ionbeam sputtering by an intense beam of fairly low
energetic Ar-ions (about 500 eV) which hits the specimen at an oblique angle. This method has been
already applied successfully to studies of bulk selfdiffusion in copper [56], nickel [57], iron [3 11,
platinum [ 1161, silicon [58] and silver [ 1171. (For
details of the method and for some results concerning
Fe, Pt, and Si see also the contribution of Mehrer et
al. to this Conference [3 l] .) In the authors opinion
the great advantage of the sputter-sectioning
methods is that they can be used for a great variety of materials including metals, alloys and semiconductors
whereas the chemical and electrochemical
methods
are specific to certain materials. On the other hand
some of the data obtained with electrochemical
sectioning [59,60] even appear to be influenced by experimental difficulties as pointed out in [23,117].
Both groups of microsectioning techniques are in
discharge

10

0,)

0.01
Fig. 4. Penetration plots of nickel self-diffusion obtained by
means of a sputter-sectioning technique [57].

48

H. Mehrer /Atomic

jump processes in self-diffusion

principle capable of measuring very small diffusion


coefficients between lo-l2 cm2/s and about lo-l9
cm2/s within reasonable diffusion times. As an
example fig. 4 shows penetration profiles of 63Ni in
nickel single crystals obtained by Maier et al. [57] with
the ion-beam sputter-sectioning
technique. The
profiles are gaussian over about three orders in activity
drop and they bear a comparison with modern
standard diffusion experiments performed with a
mechanical sectioning technique. The mean diffusion
length 2m
is of the order of 1000 A. From the
slopes of the penetration plots diffusi~ties between
about lo-l8 and lo-l2 cm2/s have been deduced.

4. Indirect techniques for measuring self-diffusion


4.1. Nuclear-magnetic relaxation
Nuclear-magnetic-relaxation
experiments provide
a powerful approach to the study of atomic jump processes in crystals. A necessary requirement is that the
jumping atoms have nuclear spins w > 0. Where applicable these methods may supplement tracer measurements. For materials where suitable radio-isotopes
are not available they may be even unique. Various
NMR tec~iques have been described in the literature
including spin-echoes in a magnetic field gradient 1611,
motional narrowing of the NMR linewidth (which is
essentially a measurement of the spin-spin relaxation
time r2) [62], studies of the laboratory frame spinlattice relaxation time Ir, [61], and measurements of
the spin-lattice
relaxation time in the rotating coordinate system, T1,, 163,641, The relative merits and
weaknesses of these methods, when applied to seff-diffusion studies in metals, have already been discussed
in [2]. We confine the following discussion to measurements of T1 and T,, since these are the most
powerful methods. ?I, measurements, particularly,
permit an extension of the temperature range over
which self-diffusion may be studied to lower temperatures.
In contrast to tracer methods the measurement of
relaxation times is a microscopic technique in the
sense that it provides local information about the
jumping atoms. Both methods are based on the fact
that the nuclear relaxation times are strongly affected
by the jumps of the nuclei if the mean jump-time, F,

is comparable with the reciprocal Larmor frequency


of the spins. This means that in a T, measurement,
where the relaxation occurs in a strong external field
HO, the relationship
7w(J=

(4.la)

is satisfied, where w,-, = -yH, with y denoting the gyromagnetic ratio. The corresponding relation for T,, is
7w1=1.

(4.lb)

Here w1 denotes the Larmor frequency associated


with the rotating field H, . Since under appropriate
experimental conditions the ratio w1 /wa is of the
order of 10-3-104
and since i--decreases exponentially
with temperature, the T,, method is sensitive to
atomic motion in a lower temperature range than the
T1 method. In practice this places restrictions on the
order of magnitude of the self-diffusion coefficient.
In a T, measurement values down to about lo-
cm/s may be obtained whereas in a T1, measurement
values as low as lo-
cm2/s may be deduced in
favourable cases.
Obviously theoretical efforts are necessary to
relate the relaxation times to the mean jump times
and these via definite self-diffusion mechanisms to
the diffusion coefficient. For a brief review of the
problems involved we refer to a paper by Seeger et al.
[65]. The progress achieved in this area in recent
years, especially for isotopic&y pure crystals, will be
briefly reviewed in the rest of this section.
In metals the two main contributions to T1 and T,,
come from the nuclear magnetic dipole-dipole
interaction (subscript dip) and from the coupling of the
nuclear spins to the conduction electrons through the
Fermi interaction (subsc~pt e). The pertaining relaxation rates are additive and hence we may write
1

c=Ta,dip+Tol,ey
where a! = 1, lp. The effect of atomic jump processes
on T, is mainly contained in the dipolar contributions
which under appropriate experimental conditions lead
to a minimum of Tcyas a function of temperature. A
good example is provided by the measurements of
Messer [66] onisotopically
pure 7Li shown in fig. 5
where both the T1 and the TIP minimum have been
observed.
The electronic contributions to the relaxation rates

49

H. ~eh?er /Atomic jump processes in self~iffus~~

and for TIP [73]


1
T l&dip

= ;y+Pz(1+

x (2wo)l *

4-2.0

3.0

1.0

Fig. 5. Nuclear-magnetic-relaxation
according to [66].

5.0

6.0

35 Mffz

7.0 8.0 9.0 fo.0


-dQ+

times T1 and ?I, in 7Li

is well understood.

According to Rorringas relation


[671 l/J-,,, increases nearly proportional to the absolute temperature. In the case of l/TIP,, Korringas
relation is slightly modified and depends on the relaxation field [64,68 ] , as discussed in detail in 1691.
In the theoretical treatment of the ~j~~r relaxation r&es it is expedient to distinguish between relaxations occurring in magnetic field that are much
larger than the local field of the nuclear dipoles
which is typically of the order of 1 Oe (for a precise
definition see [70]) and cases where this is not true.
We refer for the case of low-field relation to a recent
paper by Wolf [71] who extended the Slichter-Alion
theory [64] for low-field rotating frame relaxation to
correlated self-diffusion mechanisms, In the case of
high-field relaxation the relationship between the
dipolar relaxation rates and the self-diffusion coefficient may be deduced in three steps of which step
(i) depends on the quantity to be observed but not
on the diffusion mechanism, whereas for steps (ii) and
(iii) the reverse is true.
(i) The problem of relaxation due to the internal
motion of nuclei was first considered by Bloembergen
et al. [72]. They found that the dipolar relaxation
rates are related to the spectral density functions
J4)(w)(q= 0, 1, 2) of the internal motions (see e.g.
[61]) by treating the magnetic dipole interaction as a
small time-dependent
perturbation acting on the
Zeeman hamiltonian of the spin system. The result
for TI is [72]
1
= #y4A21(1 + l){J()(wo)
Tr,dip

t Jt2)(2wo)}

i4.3a)

l){J@)(2wr)

+ lOJ()(wo)

+ P)
(4.3b)

These relations are applicable for any diffusion mechanism.


(ii) In order to d e d uce the spectral density functions a definite diffusion model must be considered.
Torrey [74] and Eisenstadt and Redfield [75] considered models in which the nuclei perform a random
walk. However, for defect mechanisms of diffusion in
solids successive jumps of nuclei are correlated. In the
case of tracer diffusion the correlation effects have
been discussed in section 2.2. They concern spatial
correlations of successive jump directions. Although
originating from the same physical effect, the influence of correlations between successive jumps is more
complicated in the case of nuclear-magnetic-relaxation.
Since NMR experiments have an inherent time scale
(l/w0 or l/wr)a temporal correlation in addition to
the spatial correlation must be considered. It takes
into account that successive jumps of an atom caused
by the same migrating defect are bunched into groups

1751.
A comprehensive treatment of the influence of
correlated diffusion on nuclear-magnetic-relaxation
has been given recently by Wolf 1761. He obtained
numerical results for diffusion via monovacancies in
bee and fee crystals. The divacancy mechanisms in
cubic crystals and the simultaneous action of monoand divacancies has been considered in [77] and more
rigorously in [77aj. Both Tr,din and Tr,,unshow
minima near temperatures where either wo? = 1 or
wr? % 1. The exact position, shape and width of the
minima as functions of temperature depend on the
diffusion mechanism. From a fit of the theoretical
results to the data the mean jump time ? may be
determined.
(iii) The mean jump time is related to the uncorrelated self-diffusion coefficient by the EinsteinSmoluchowsky relationship which for cubic crystals
may be written as
DSD = b=/fi,

(4.4)
where b is the jump distance. According to eq. (2.22)
the correlated diffusion coefficient is given by

D zfz>S= .

(4.5)

H. Mehrer/Atomic jump processesin sell-diffusion

50

Since eq. (4.5) pertains to the abundant isotope I, a


comparison with tracer experiments usually requires
a correction for the isotope effect. From eq. (2.21)
we obtain
(4.6)
This correction

similar to that of nuclear-magnetic-relaxation


through
diffusional motion before the effects of definite diffusion mechanisms in solids had been properly incorporated into the theory. Utilizing the theoretical
methods developed in this area might help that Mossbauer spectroscopy becomes a quantitative tool for
studies of atomic jump processes.

amounts at best to a few percent.


4.3. Quasi-elasticneutron scattering

4.2. ~~ssbauer effect


The Mijssbauer effect may be considered as the
recoilless emission and absorption of nuclear y-radiation. If the nuclei involved are at rest the width of the
resonance line is determined by the life-time of the
excited nuclear state, TN, through Heisenbergs relation. Atomic jumps lead to a broadening of the line
when the mean jump time 7 becomes shorter than TN,
which for typical Mossbauer isotopes is of the order of
low7 s. This corresponds via eq. (4.4) to diffusion
coefficients of 10m9 cm/s and restricts the application of the Miissbauer effect in diffusion studies to
the ~gh-temperature
region.
Singwi and Sjiilander 1781 considered the effect of
atomic motion on the line-width. Their theory
separates atomic motions into vibrational motion and
into atomic jump processes. Using a random flight
model in the sudden jump approximation they predict that the Mossbatter line is lorentzian with a diffusional broadening, Ae, given by
Ae =(2h/-?)(l

- cr) ,

The investigation of diffusion processes in liquids


and in hydrogen-metal
systems by means of quasielastic scattering of slow neutrons is a well-established
method (see, e.g. [86,87] for reviews). Self-diffusion
in monoatomic solids also gives rise to a broadening
of the quasi-elastic line which should be observable in
a few materials which have fairly high diffusivities
D 2 10e7 cm2/s and large incoherent scattering cross
sections. The great advantage of the method lies in
the fact that it is also able to resolve details of the
atomic jump process like, e.g., the mean jump distance.
According to van Hove [SS] the incoherent quasielastic scattering cross section, Sine, is related to the
Fourier transforms of the correlation functions of
atomic motion. Chudley and Elliot [89] have calculated Sine by separating atomic motions into vibrational and diffusive motion and by treating the latter
within the framework of a random walk model. The
quasi-elastic line is broadened and has a lorentzian
shape

(4.7)

where CY
is a small correction term. A critical comparison between experiments and theory has been
given recently by Janot 1791. According to eq. (4.7)
the activation enthalpy of diffusion may be simply
determined from an Arrhenius plot of Ae versus
reciprocal temperature. However, whenabsolute values of the diffusion coefficients are deduced and compared with tracer measurements, most experiments
indicate that the broadening is smaller than the
value predicted by theory. This is confirmed by a
recent paper by Lindsey [80] who investigated the
self-diffusion of Fe in a-iron stabilized by small
amounts of vanadium. Although various improvements of the Singe-Sj~lander
model have been
proposed (Sl-SS] the theory is still not yet completely satisfactory. It appears that the situation is

Sine

*,-pQ2

=-f(e)

f@12+(Ml2 .

(4.8)

Here hw and &Q denote energy and momentum transfer during the scattering event. The Debye-Waller
factor e -pQ2 determines the intensity of the line. The
energy-width of the line depends on Q like f(Q). For
jump diffusion in a lattice f(Q) is a periodic function
which depends on the lattice geometry and on the
diffusion mechanism. The extrema of f(Q) are directly
related to the atomic jump vectors. For small values
of the momentum transfer (Q + 0) the half-width
approaches 2 h2Q2DSDwhich permits a determination of the macroscopic seif-diffusion coefficient.
Recently Ait-Salem [90] and Goeltz [9 1] have
shown that self-diffusion in sodium indeed affects the
quasi-elastic line. Ait-Salem concludes that self-dif-

H. Mehrer /Atomic jump process in self-diffusion

metal to metal. As a consequence one has to separate


mono- and divacancy contributions when one wants
to deduce reliable defect properties. The reliability of
an analysis of the temperature dependence of the diffusion coefficient in terms of eq. (2.15) is improved
when it is based on measurements over a wide temperature range. Hence, the application of microsectioning and nuclear-magnetic-relaxation
techniques to various metals and the resulting extension
of self-diffusion measurements to low temperatures
represents a considerable progress. The reliability
with which, e.g., the activation enthalpy of monovacancy diffusion may be obtained, has been greatly
increased. The separation of the divacancy contribution may be helped by measurements of the isotope
effect and of the pressure dependence of self-diffussion.

fusion in sodium near the melting point occurs mainly


by nearest-neighbour jumps with a small additional
contribution which may be due to next-nearest-neighbour jumps caused by divacancies whose definite
identification, however, requires further experiments
[9 11. The possibility of using quasi-elastic scattering
for a study of jump processes in self-diffusion has been
demonstrated experimentally. However, the agreement between experiment and theory is not yet perfect. On the other hand it is clear that the theory is
capable of further improvements and that in principle more information could be extracted.

5. Application

to specific metals

Self-diffusion in metals has been reviewed by


Peterson during this Conference and also by various
authors in the past few years [l-5]. Therefore, we
have chosen to limit the present discussion to subjects close to the area of our interest. Special attention is given to metals where a considerable deepening
of our understanding of atomic jump processes and
vacancy properties has been achieved.

5.1.1. Silver
The analysis of self-diffusion data has been
reported in detail elsewhere [23]. It was based on
the most reliable high temperature data [21] and on
two recent measurements of low temperature selfdiffusion 122,231 by means of chemical microsectioning. The Arrhenius plot of self-diffusion is curved. To
reveal the curvature more clearly the deviations of the
data [21-231 from a single Arrhenius term, obtained
by force fitting eq. (2.13) to the data, have been
plotted in fig. 6. The mono- and divacancy contributions deduced from the analysis of the temperature

5.1. Face-centered cubic metals


Self-diffusion in fee metals occurs predominantly
by monovacancies with a divacancy contribution at
higher temperatures whose magnitude varies from

80%-

700

?O 8po

600

51

500

100 -

TKJ

300

o Rothman, Peterson and Robinson


60% -

+
A

Lam, Rothman, Mehrer and Nowicki


Backus, Bakker, Mehrer

+
A

40%4

20%-

-e0

00

-20 % -

+*

09
Y
l Qeff

+
+

0+
~ A
a

= 0.235 cm24
=
1.86 eV

-10% -

I
0.8

I
1

I
1.2

1.1
lo/

1.6
T TK-I
-)

I
1.8

Fig. 6. Deviations of silver self-diffusion data from a single Arrhenius term [23].

H. Mehrer /Atomic jump process in selfdiffusion

SO 800 700
i

600
8

-TTCJ
coo

500
I

300

Ag self-diffusion
D Rorhmon ( Pet~son vnd Robinson
+ tom, Kbthmon, Mehrer und Nowicki
d &c&us, Bokker und Mehnr

600

fly =c1782correlation
factor
for monovacancy

t
14

to7
$
.p

-.

TECl

700

sCaa

800

900

Rothmon
AgtqAgllO

fJv=

0.8

0.468

correlation

factor

0.9

1.0

et al. I19701
isotope

effect

for simple

103/T

divocanc
, mecL*m
1.1
l-K*]
--_)

Fig.

I@@
0.8

1.0

8. Isotope effect in silver. The solid line represents the


fit in terms of mono- and divacancy contributions according to 1231.
1.2

1.1

1.6
lO+T ----c

1.8

Fig. 7. Arrhenius plot of tracer self-diffusion in silver; monoand divacancy con~ibutions according to [ 2 31.

dependence are shown in fig. 7 together with the data.


The monovacancy properties are Qrv = 1.76 eV and
Dt = 0.046 cm2/s. The fairly large divacancy contribution at high temperatures is con~rmed by the observation that the isotope effect decreases with
increasing temperature [21,92]. In fig. 8 the most
accurate isotope effect data are shown [21]. From
the combined analysis of figs. 7 and 8 also a fairly
detailed conclusion with respect to the divacancy
mechanism have been drawn, The divacancy migrates
predominantly by nearest-neighbour jumps which
preserve the nearest-neighbour bond (v2v1 r jumps in
fig. 1) with an additional contribution which is most
likely due to the dissociation to a second nearest
neighbour position (v2vr2 jumps in fig. 1). If the
divacancy migrated by r+vl r jumps only, the pertaining value of the maximum isotope effect would
be 0.468. This low value would require an even
stronger decrease of the isotope effect with increasing
temperature than the observed one. The solid line in
fig. 8 is a fit of eq. (2.26) to the data with gav given

by fig. 1. Using the mono- and divacancy contributions shown in fig. 7. The adjusted parameters are the
kinetic energy factors AKrv = 0.99 and AR2v = 0.95,
which are assumed to be temperature independent.
Reasonable fits, which imply AKzv Q AK, < 1,
have been achieved only when vzvl a jumps have been
taken into account. A ratio ~~v~r/v~vr~ between 0.3
and 0.7 near the melting point is compatible with the
data.
The pressure dependence of self-diffusion has been
studied in [93] at a temperature not far below the
melting temperature. The observed activation volume
A pff = 0.9 atomic volumes is consistent with diffusion be mono- and divacancies. However, a measurement of AVeff over a wide temperature range which
appears to be possible by means of appropriate microsectioning techniques would help to establish values
for AV,v and AVav.
5.12. Copper
Self-diffusion in copper has been measured in the
high temperature region with high accuracy by
Rothman and Peterson [94] and by Bartdorff [95].
Rothman and Peterson also measured the isotope
effect and obtained an average value of E = 0.68 + 0.02

53

H. Mehrer /Atomic jump process in self-diffusion


800

TCCI
100

600
self-diffusion
0

Rothmon

Bartdorff

M&r,

in Cu
and Peterson

1969

1970

and ealssani 1973

Schdc

Lam. Rothman and Nowcki


We&se
and Noack 1973
(NMR)

extrapolation
IO+4

from high tempemtures


according
to
Mehrer

W73

and Seeger

1969
\

5.1.3. Nickel

lo7

0.8

I
1.0

slightly different. The monovacancy contribution is


described by Dl = 0.1 cm2/s and Qrv = 2.04 eV. A
direct determination of Qrv has been performed in
the contribution of Maier and Mehrer [ 1011 to this
Conference. They obtained Qrv = 2.05 eV.
The effective activation volume AVeff = 0.91
atomic volume [ 1021 observed in the high temperature region is consistent with a divacancy contribution.
An extension of the measurements of the pressure
dependence and of the isotope effect would be desirable and might result in unambigous values of Al/,v
and AKIv.

1.2

1.4
10+/z CK-I

1.6

Fig. 9. Arrhenius plot of tracer self-diffusion in copper. Extrapolation from high temperatures according to [96].

which was independent of temperature within the


limited temperature range of the experiments. Mehrer
and Seeger [96] have performed an analysis of the
data of [94]. They came up with a monovacancy contribution which is dominant in the whole temperature range. The small divacancy contribution deduced
in their analysis requires a decrease of the isotope
effect with temperature which is within the error bars
of the data [94]. The extrapolation of the tracer diffusion coefficient to lower temperatures according to
the Mehrer-Seeger analysis is shown in fig. 8.
The extrapolation has been tested in the meantime
by various authors and techniques. Maier [53,97] and
Lam et al. [98] were able to extend tracer measurement of self-diffusion to almost ten powers in DT by
means of either sputter-sectioning
(53,971 or electrochemical sectioning [98]. Weithase and Noack [99]
have found that their nuclear-magnetic-relaxation
measurements are also in close agreement with the deductions of [96].
An analysis similar to that of Seeger and Mehrer
which includes the low temperature data will be published elsewhere [ 1001. The conclusions as to the
mechanisms responsible for diffusion are the same as
those expressed in [96]. The defect parameters are

Self-diffusion in nickel was studied many times in


the past 20 years (for references see, e.g. [57]). In the
Arrhenius plot of fig. 10 the data of Bakker [103],
who performed the most comprehensive study in the
high temperature, regime have been plotted together
with low temperature data obtained by Maier et al.
[57] using sputter sectioning. (See also the pertaining
penetration plots in fig. 4.) The data of Maier et al.
TPCl-lo+

T,,,=US3

1200

1000

800

600

Bakker

Maier,
and

1968
Mehrer.

Schiile

Lenmann

(1976)

Fig. 10. Arrhenius plot of tracer self-diffusion in nickel.

54

H. Mehrer /Atomic jump process in self-diffusion

appear not to be influenced by diffusion along short


circuits in contrast to earlier measurements of low
temperature diffusion in nickel (for details see [57]).
The monovacancy parameters deduced from an
analysis of the data of fig. j are Q,v= 2.88 eV and
0: = 0.92 cm2/s [57]. These pertain to paramagnetic
nickel. In view of the influence of ferromagnetic
ordering on the defect properties observed for iron
[3 11 (see also sect. 2.2) one has to be careful when
these values are compared with vacancy properties
below the Curie temperature (359C).
5.2. Body-centred cubic metals
Body-centred metals are usually divided into normal and anomalous metals according to their diffusion behaviour (see, e.g. [ 1041). The anomalous
metals exhibit strong nonlinearities of the Arrhenius
plot. The normal bee metals, which include the
alkali metals, show no unusual features when compared with fee metals. We restrict ourselves to a discussion of the alkali metals Li and Na for which a
fairly comprehensive set of data exists. These metals
fit into the concept of diffusion through mono- and
divacancies and may well be typical for many bee
metals. Of particular interest is the role of divacancies
which, according to section 2.1 in bee metals, may
be more complex than in fee metals.
5.2.1. Lithium
Lithium provides a particular interesting case for
nuclear-magnetic-relaxation
studies of self-diffusion.
In addition to older NMR studies a number of recent
measurements on pure lithium are available [ 105,661.
We base the following discussion on the data of Messer
[66] which appear to be both the most accurate and
the most comprehensive. The relaxation times for
isotopically pure Li shown in fig. 5 have been converted into macroscopic self-diffusion coefficients by
means of the theory reviewed in section 4.1. The
resulting Arrhenius plot of fig. 11 covers eight powers
of ten in PD.
Direct measurements of the tracer se~f~iffusjon
coefficient DT of 6Li in Li are also available [ 1061.
They had been performed by mass spectroscopy
since radioactive tracers appropriate for radiotracer
studies do not exist. The ratio DT/DSD is shown in
fig. 12 as a function of reciprocal temperature. Ac-

,o?51

2.5

3.0

3.5

;\

1.0

4.5

,a

5.0 5.5
-103K
T

Fig. 11. Arrhenius plot of the macroscopic self-diffusion


coefficient in lithium deduced from fig. 5 [66].

cording to eq. (2.20) this ratio is just the average COTrelation factor of the diffusion process. Below 40C
the data approach the value flv = 0.727 which is
characteristic for a monovacancy mechanism. The
decrease with increasing temperature is fairly direct
evidence for a strong divacancy contribution at higher
temperatures. From eqs. (2.20) and (2.13) we obtain
for the simultaneous action of mono- and divacancies
D%
f=flV~+f2vgg.

(5.1)

Since fiv < flv, f decreases with increasing tempera-

or

-$5

.9-

.7+
.6 .5 .L -

.3 -

:I, , , , ,
2.0

2.2

2.1

(
2.6

2.8

Fig. 12. Correlation factor for se~diffusion

,
3.0

,
3.2
t

3.4
3
j0.K
T

in lithium 166 1.

H. Mehrer /Atomic

ture. Near the melting temperature fis as low as 0.4.


From a comparison of this value with the theoretical
value offa in fig. 2 we may conclude that the most
stable divacancy configuration is the next-nearestneighbour divacancy which migrates predominantly
by the jump sequence ~2~24 + ~2~42.
The mono- and divacancy contributions obtained
from a fit to the data of the macroscopic self-diffusion coefficient are shown in fig. 11. The monovacancy parameters are Q 1v = (0.5 18 + 0.009) eV and
Dy/flv = (0.06 + 0.1) cm2/s. Near the melting point
the divacancy contribution dominates and causes a
considerable curvature in the Arrhenius plot. A curvature due to divacancy contributions is more prominent in the macroscopic self-diffusion coefficient than
in the tracer self-diffusion coefficient because of the
stronger correlation effects for divacancies.
Isotope effects in lithium present an interesting field
of research. Because of the extremely high Debye
temperature with respect to the melting point one
might expect quantum effects in diffusion which are
due to the fact that one has to apply quantum rather
than classical statistics in the whole range of the solid
phase. The mean jump times?i for isotope i have been
determined for 6Li and 7Li by Messer [66] and for
*Li by Ackermann et al. [107] in the temperature
range where diffusion occurs by monovacancies. In
table 2 the results are compared with the classical
equation
-7 i/Tj = (mi/i?lj)2 ,
(5.2)
where mi denotes the mass of isotope i. In contrast
to earlier measurements by means of mass-spectroscopy [ 1061, which indicated an isotope effect of
20-40% between 6Li and Li, the values of table 2
obey approximately classical statistics.

and Smith

Fig. 13. Arrhenius

plot of tracer self-diffusion

100

0.8

75

AVzv i 0.59 5-t

Li

AV,,,=AV,,D:,/D+AV,,D:,/D

of self-

Table 2
Effect of isotopic mass on the mean jump time of lithium
according to (661
6Li

T Fl

50

9
to identify the mechanisms

in sodium.

diffusion measurements of the temperature, pressure, and mass dependence have been performed. The
temperature dependence of the tracer diffusion coefficient was measured by Mundy [108,109]. The Arrhenius plot shown in fig. 5.8 is clearly curved upward.
The pressure dependence was investigated by Nachtrieb et al. [l lo] and Mundy [109] by means of a
radio-tracer technique and by Hultsch and Barnes [ 11 l]
by means of nuclear-magnetic-relaxation.
The effective activation volume deduced from these experiments shown in fig. 14 increases with increasing temperature. The isotope effect data [108,109] plotted
in fig. 15 show a gradual decrease with increasing temperature.
It is tempting to attribute these findings to the
simultaneous contributions of mono- and divacancies
to self-diffusion. A fit of the temperature dependence

LO.6

5.2.2. Sodium
In an attempt

5.5

jump process in self-diffusion

04

8Li

:
0.937

1.088

0.926

1.068

3.0

3.5
103/T

Fig. 14. Effective activation volume


sodium as a function of temperature.

4.0
WI
-

of self-diffusion

4.5

in

H. Mehrer /Atomic

56

0.6

0.5

.
a

Mundy
Mundy,

double jumps of atoms which have been recently


observed by Bennett [42] in molecular dynamics
calculations.
Barr and Smith

E = AK,,I,,
*

jump processes in self-diffusion

0;

/D

AKJv gJv

0:

10

ij&+..F

0.3
0.2
?

0.1

Acknowledgements
I am grateful to Dr. R. Messer for making available
his NMR measurements on lithium prior to publication. I also benefited from discussions with Professor
Seeger and with my colleagues Dr. K. Maier, Dipl.Phys. G. Hettich and Dipl.-Phys. G. Reiu.

References
Fig. 15. Isotope effect of self-diffusion in sodium.

[II N.L. Peterson, in: Solid State Physics, Vol. 22, F. Seitz,
in terms of eq. (2.15) is shown in fig. 13 as a solid line.
The mono- and divacancy contributions shown as
dashed lines are described by 0:" = 3.76 X low3
exp(-0.365
eV/kT) cm2/s and Dzv = [0.292
exp(-0.457
eV/kT) + 8 exp(-0.575 eV/kZJ] f2v
cm2/s, withf2v given in fig. 2. The two activation
enthalpies in the divacancy contribution pertain to
the two saddle points between the various divacancy
configurations. The strong divacancy contribution is
supported by the pressure and isotope effect data.
Fits of eqs. (2.33) and (2.28) to the data are shown
in figs. 14 and 15 as solid lines. In these fits the monoand divacancy contributions deduced above have been
used. The fit to the pressure data yields A V,v =
0.32 52 and A V2v = 0.59 52 (52 = atomic volume). The
numerical values of the activation volumes also support the view that the two mechanisms are indeed
due to mono- and divacancies. The kinetic energy
factors in the fit of the isotope effect data AK,v =
0.68 and AKzv = 0.55 are in reasonable agreement
with theoretical predictions for bee metals mentioned
in section 2.3.
In the immediate vicinity of the melting point the
self-diffusion data reveal a sharp upward curvature of
the Arrhenius plot. A possible interpretation
of this
phenomenon could be second-nearest-neighbour
jumps caused by divacancies as suggested by the
findings of quasi-elastic neutron scattering discussed
in section 4.3. Another explanation could be the

D. Turnbull and H. Ehrenreich, eds. (Academic Press,


New York, 1968).
t21 A. Seeger and H. Mehrer, in: Vacancies and Interstitials
in Metals, A. Seeger, D. Schumacher, W. Schilling and
.I. Diehl, eds. (North-Holland Publ. Company, Amsterdam, 1970) p. 1.
[31 A. Seeger, J. Less-Common Metals 28 (1972) 387.
I41 N.L. Peterson, in: Diffusion (Amer. Sot. Metals, Metals
Park, Ohio) 1973.
[51 N.L. Peterson and W.K. Chen, in: Annual Review of
Material Science, R.A. Huggins, R.H. Bube and
R.W. Roberts, eds. (Annual Reviews Inc. Palo Alto,
1973) p. 75.
[61 Y. Adda and J. Philibert, La Diffusion dans les Solides
(Presses Univ. de France, 1966).
[71 J.R. Manning, Diffusion Kinetics of Atoms in Crystals
(van Nostrand, Princeton, 1968).
VI J. Bardeen and C. Herring, in: Atoms Movements
(A.S.M. Cleveland, 1951) p. 87; and in Imperfections
in Nearly Perfect Crystals, W. Shockley, ed. (J. Wiley,
New York, 1952) p. 261.
[91 A.D. LeClaire, in: Physical Chemistry, vol. 10,
H. Eyring, D. Henderson and W. Jost, eds. (Academic
Press, New York, 1970) ch. 6.
IlO1 H. Mehrer, Habilitation Thesis, University of Stuttgart
(1973).
[Ill K. Compaan and Y. Haven, Trans. Faraday Sot. 52
(1956) 786.
iI21 J.G. Mullen, Phys. Rev. 124 (1961) 1723.
[I31 E. Steiner, H. Mehrer and A. Seeger, Phys. Status Solidi
(b) 75 (1976) 361.
[I41 R.E. Howard, Phys. Rev. 144 (1966) 650.
1151 H. Balcker, Phys. Status Solidi 44 (1971) 369.
[I61 H. Mehrer, J. Phys. F: Metal Phys. 2 (1972) L 11.
[I71 H. Mehrer, J. Phys. F: Metal Phys. 3 (1973) 543.
1181 R.A. Johnson, in: Diffusion in b.c.c. Metals, J.A. Wheeler

H. Mehrer /Atomic jum processes in self-diffusion


and F.R. Winslow, eds. (Amer. Sot. Metals, Ohio, 1966)
p. 357.
[19] J.W. Flocken and J.R. Hardy, Phys. Rev. 177 (1969)
1054.
[20] P.S. Ho, Phil. Mag. 26 (1972) 1429.
[21] S.J. Rothman, N.L. Peterson and T.J. Robinson, Phys.
Status Solidi 39 (1970) 635.
[22] Nghi Q. Lam, S.J. Rothman, H. Mehrer and L.J. Nowicki,
Phys. Status Solidi (b) 57 (1973) 225.
[23] J.G.E.M. Backus, H. Bakker and H. Mehrer, Phys. Status
Solidi
64 (1974) 151.
[24] G. de Vries, Phys. Status Solidi (a) 28 (1975) K 61.
[25] L.M. Levinson and F.R.N. Nabarro, Acta Met. 15 (1967)
785.
[26] L.A. Girifalco, Ser. Met. 1 (1967) 5.
[27] C.P. Flynn, Phys. Rev. 171 (1968) 682.
[28] W.M. Franklin, J. Chem. Phys. 52 (1972) 2659.
[29] W.M. Franklin, in: Diffusion in Solids, AS. Nowick and
J.J. Burton, eds. (Academic Press, New York, 1975) p. 1.
[30] H.M. Gilder and D. Lazarus, Phys. Rev. B 11 (1975)
4916.
[31] H. Mehrer, K. Maier, G. Hettich, H.J. Mayer and G. Rein,
this Conference. See also: Scripta Met. 7 (1977).
[32] L. Ruth, D.R. Sain, H.L. Yeh and L.A. Girifalco, J. Phys.
Chem. Solids 37 (1976) 649.
[ 331 N.L. Peterson, in: Diffusion in Solids, A.S. Nowick and
J.J. Burton, eds. (Academic Press, New York, 1975)
p. 116.
[ 341 H. Mehrer, A. Seeger and E. Steiner, Phys. Status Solidi
(b) 73 (1976) 131.
[35] J.R. Manning, Phys. Rev. 136 (1964) A 1758.
(361 J.G. Mullen, Phys. Rev. 121 (1961) 1649.
[37] A.D. LeClaire, Phil. Mag. 14 (1966). 1271.
(381 H.B. Huntington, M.D. Feit and D. Lortz, Cryst. Lattice
Defects l(l970) 193.
[39] B.N.N. Achar, Phys. Rev. B 2 (1970) 3848.
[40] R.C. Brown, J. Worster, N.H. March, R.C. Perrin and
R. Bullough, Z. Naturforsch. 26a (1971) 77.
[41] M.D. Feit, Phys. Rev. B 5 (1972) 2145.
(421 C.H. Bennett, in: Diffusion in Solids, A.S. Nowick and
J.J. Burton, eds. (Academic Press, New York, 1975)
p. 74.
[43] C.H. Bennett, 19th Colloque de Metallurgic (Saclay/
France, 1976) La Diffusion dans les Milieux Condenses
Vol. II p. 655.
[44) A.P. Batra, Phys. Rev. 159 (1967) 487.
[44] a. J.J. Burton, Phys. Rev. B 9 (1974) 1200.
[45] H. Mehrer and A. Seeger, Cryst. Lattice Defects 3 (1972)l
1461 L. Tewordt, Phys. Rev. 109 (1958) 61.
[47] G. Schottky, A. Seeger and G. Schmid, Phys. Status
Solidi 4 (1964) 419.
[48] R.E. Pawel, Rev. Sci. Instrum. 35 (1964) 1066.
[49] R.E. Pawel amd T.S. Lundy, J. Phys. Chem. Solids 26
(1965) 937.
[SO] R.E. Pawel and T.S. Lundy, J. Electrochem. Sot. 115
(1968) 233.
[51] D. Gupta, Thin Solid Films 25 (1975) 231.

57

[52] D. Gupta and R.T.C. Tsui, Appl. Phys. Lett. 17 (1970)


294.
[53] K. Maier and W. Schiile, Euratom Report EUR 5234 d
(1974).
[54] D. Gupta, Phys. Rev. B 7 (1973) 586.
[55] D. Gupta and K.W. Asai, Thin Solid Films 22 (1974)
121.
[56] K. Maier, C. Bassani and W. Schiile, Phys. Lett. A 44
(1973) 539.
[57] K. Maier, H. Mehrer, E. Lessmann and W. Schtile, Phys.
Status Solidi (6) 78 (1976) 689.
[58] H.J. Mayer, H. Mehrer and K. Maier, in: Radiation
Effects in Semiconductors, N.B. Urli and J.W. Corbett,
eds. Institute of Physics Conf. Series No. 31 (Bristol
and London) 1977, p. 186.
[59] A.G. Vardiman and M.R. Achter, Trans. MS AIME 245
(1969) 178.
[60] C.T. Lai and H.M. Morrison, Can. J. Phys. 48 (1970)
1548.
[al] A. Abragam, The Principles of Nuclear Magnetism (Oxford Clarendon Press, London, 1961).
[62] I. Ebert and G. Seifert, Kernresonanz in Festkorpern
(Akad. Verlagsgesellschaft Leipzig, 1966).
[63] D.C. Ailion and C.P. Slichter, Phys. Rev. Lett. 12 (1964)
168.
[64] C.P. Slichter and DC. Ailion, Phys. Rev. 135 (1964)
A 1099.
(651 A. Seeger, D. Wolfand H. Mehrer, Phys. Status Solidi
(b) 48 (1971) 481.
[66] R. Messer, Dr. rer. nat. thesis, University of Stuttgart
(1976);
R. Messer, Proc. 19th Colloque Ampere, Heidelberg,
1976, H. Brunner, K.H. Hauser and D. Schweitzer, eds.,
p. 533.
[67] J. Korringa, Physica (Utrecht) 16 (1950) 601.
[68] D.C. Ailion and C.P. Slichter, Phys. Rev. 137 (1965)
A 235.
[69] P. Jung, Diploma thesis, University of Stuttgart (1973).
(701 L.C. Hebel, Solid State Phys. 15 (1963) 409.
[71] D. Wolf, Phys. Rev. B 10 (1974) 2724.
[72] N. Bloembergen, E.M. Purcell and R.V. Pound, Phys.
Rev. 73 (1948) 679.
[73] D.C. Look and I.J. Lowe, J. Chem. Phys. 44 (1966)
2995.
[74] H.C. Torrey, Phys. Rev. 92 (1953) 962.
[75] M. Eisenstadt and A.G. Redfield, Phys. Rev. 132 (1963)
635.
[76] D. Wolf, Phys. Rev. B 10 (1974) 2710.
[77] E. Cavelius, Phys. Status Solidi (b) 65 (1974) 181.
[77a] D. Wolf, Phys. Rev. B15 (1977) 37.
[78] K.S. Singwi and A. Sjolander, Phys. Rev. 119 (1960)
863; 120 (1960) 1093.
(791 C. Janot, J. de Phys. 37 (1976) 253.
(80) R. Lindsey, Phys. Status Solidi (b) 75 (1976) 583.
[81] R,C. Knauer, Phys. Rev. B 3 (1971) 567.
(821 K. Sorenson and G. Trumpy, Phys. Rev. B 7 (1973)
1791.

58

H. Mehrer /Atomic

jump processes in self-diffusion

[83] S.V. Karyagin, Phys. Lett. 49A (1974) 183.

[84] M.C. Dibar-Ure and P.A. Flinn, Appl. Phys. Lett. 23


(1973) 587.
[85] M. Ron and F. Hornstei. , J. Phys. Colloq. 35 (1974)
C 6-505.
[86] T. Springer, Quasielastic Neutron Scattering for the
Investigation of Diffusive Motions in Solids and Liquids
(Springer-Verlag, 1972).
[87] W. Gissler, Ber. Bunseu Ges. 76 (1972) 770.
[88] L. van Hove, Phys. Rev. 95 (1954) 249.
[89] C.T. Chudley and R.J. Elliot, Proc. Phys. Sot. 77 (1961)
353.
[PO] M. Ait-Salem, Dr. rer. nat. thesis, Technische Hochschule Aachen (1976).
[Pl] G. Goltz, private communication.
[92] P. Reimers and D. Bartdorff, Phys. Status Solidi 50
(1972) 305.
[93] F.R. Bonanno and C.T. Tomizuka, Phys. Rev. A 137
(1965) 1264.
(941 S.J. Rothman and N.L. Peterson, Phys. Status Solidi 35
(1969) 305.
[95] D. Bartdorff, Dr. rer. nat. thesis, FU Berlin (1971).
[96] H. Mehrer and A. Seeger, Phys. Status Solidi 35 (1969)
313.
[97] K. Maier, Dr. rer. nat. thesis, University of Stuttgart
(1973).
[98] N.Q. Lam, S.J. Rothman and L.J. Nowicki, Phys. Status
Solidi (a) 23 (1974) K 35.
[PP] M. Weithase and F. Noack, Z. Phys. 270 (1974) 319.
[ 1001 K. Maier, Phys. Status Solidi (a), in press.
[lo1 ] K. Maier and H. Mehrer, this Conference.
[102] M. Beyeler and Y. Adda, J. de Phys. 29 (1968) 345.
[103] H. Bakker, Phys. Status Solidi 28 (1968) 569.
[104] Diffusion in BodyCentered Cubic Metals, Conference
in Gatlinbury, Tennessee (ASM, 1965).
[ 105 ] R. Messer and F. Noack, Appl. Phys. 6 (1975) 79.
[106] A. Lodding, J.N. Mundy, and A. Ott, Phys. Status Solidi
38 (1970) 559.
[107] H. Ackermann, D. Dubbers, M. Grupp, P. Heitjans,
R. Messer, and H.J. Stockman, Phys. Status Solidi (b)
71 (1975) K 91.
[lOS] J.N. Mundy, L.W. Barr and F.A. Smith, Phil. Mag. 14
(1966) 785.
[lOP] J.N. Mundy, Phys. Rev. B 3 (1971) 2431.
[llO] N.H. Nachtrieb, J.A. Weil, Catalan0 and A.W. Lawson,
J. Chem. Phys. 20 (1952) 1189.
[ill] R.A. Hultsch and R.G. Barnes, Phys. Rev. 125 (1962)
1832.
[112] For an introductory discussion see H.B. Callen, Thermodynamics (Wiley, New York, 1960).
[ 1131 See, for example, the papers by N.L. Peterson,
R.W. Siege1 and R.W. Balluffi in this volume. For an
introductory discussion of these quantities see
C.P. Flynn, Point Defects and Diffusion (Oxford, 1972).
[114] Y. Fukai, this Conference.
[115] R. Vargas, M.B. Salamon and C.P. Flynn, Phys. Rev.
Lett. 37 (1976) 1550.

[ 1161 G. Reiu, H. Mehrer and K. Maier, Phys. Status Solidi


(a), to be published.
[ 1171 J. Bihr, H. Mehrer and K. Maier, submitted to Phys.
Status Solidi.

Discussion
D. Lazarus: I should like to emphasize that the general
picture of a strong divacancy contribution at high temperatures, while resonable, should still be questioned because:
(1) there are no isotope effect measurements for
D 5 lo-13 cm*/s;
(2) the Hart mechanism may contribute to D at low temperatures, causing an enhancement in the monovacancy contribution; and
(3) the Ailion-Slichter NMR measurements on sodium,
over an enormous range of temperatures, do not indicate any
curvature compared to high-temperature tracer data.
D. Gupta: The contribution of the Hart-type dislocation
enhancement to the lattice diffusion should be kept in view
while examining the data for curvatures in the Arrhenius plot.
It has been our experience that such effects may become significant in the 10-l * -10-l 6 cm* diffusivity range for
annealing times of a few hours. At lower temperatures and
diffusivities, < 10-l 6 cm*/s, the effect is always negligible
even for high dislocation densities of 10 lines/cm* and the
quality of the data improves.
H. Mehrer: (1) It is true that there are no isotope effect
measurements for lower temperatures and diffusivities. The
reason for this is that such measurements are difficult to perform. It appears, however, that nowadays isotope effect
experiments at lower temperatures could be done by means
of the microsectioning techniques discussed in section 3 of
my paper. Such measurements, when performed under conditions where a dislocation enhancement of the lattice diffusivity is negligible, would indeed be highly welcome. They
would provide unambiguous information about the kinetic
energy factor AK1 V of the monovacancy mechanism.
(2) I agree with the comments of Professor Lazarus and
Dr. Gupta that a possible influence of diffusion short circuits
on the observed diffusivity must be kept in mind and, if
necessary, must be eliminated to obtain reliable values of
the bulk diffusivity. This has been done in all the examples
discussed in my paper.
The influence of dislocation short circuits may be eliminated if highly single crystals and/or the microsectioning
techniques discussed in section 3 are used. Whether dislocation pipe diffusion influences the value of the diffusivity
deduced from the gaussian segment of the penetration profile
in a standard tracer experiment depends upon the conditions
under which the experiment is performed. When the mean
lattice diffusion length, a,
is large as compared to the
mean dislocation distance, 7, the observed diffusivity may
be enhanced through d,rocation pipe diffusion (Hart-type

H. Mehrer /Atomic

59

jump processes in self-diffusion

enhancement). When a microsectioning technique is used diffusion may be investigated under conditions where
fi
<< a. The penetration profile then consists of a gaussian segment which is mainly due to lattice diffusion and of
a so-called tail due to dislocation pipe diffusion, thus permitting an unambigous separation of both contributions. In
- a
cases where m
- this separation may indeed be difficult
as pointed out in the comment of Dr. Gupta.
(3) According to my knowledge there are no NMR measurements on sodium by Ailion and Slichter. Presumably the
third comment of Professor Lazarus refers to measurements
on lithium [DC. Ailion and C.P. Slichter, Phys. Rev. 137
(1965) A 235; see ref. (68) of the present paper]. It is true
that these measurements cover almost the same temperature
region as the NMR data of Messer discussed in detail in section 5.2.1 of my paper. The very recent data of Messer are,
however, by far more comprehensive since both Tf, and TI
have been measured for various relaxation fields also. On
the other hand there is no discrepancy between the raw data
of both investigations for the same relaxation field. However,
since the pioneering work of Ailion and Slichter the NMR
theory of relaxation through diffusional motion in solids has
been considerably improved. In particular the effects of specific diffusion mechanisms in metals and of temporal correlation between atomic jumps have been included into the
theory asoutlined in detail in section 4.1 of the present paper.
These improvements have already been utilized in the evaluation of Messers data. The arrhenius plot of the macroscopic
self-diffusion coefficient deduced from Messers data is shown
in fig. 11 of the paper. It clearly reveals considerable curvature.
D.N. Seidman: This comment is in response to Professor
Lazarus comment about the existence of divacancies. I would
like to point out that in the case of quenched platinum
[AS. Berger, D.N. Seidman and R.W. Balluffi, Acta Met. 21
(1973) 1231 we have observed directly by means of field-ion
microscopy, both monovacancies and first nearest-neighbor
divacancies. The direct measurement of the divacancy-tomonovacancy concentration ratio enabled us to determine
the Gibbs free binding energy of a divacancy for a quench
temperature very close to the melting point of platinum
[A.S. Berger, D.N. Seidman and R.W. Balluffi, Acta Met. 21
(1973) 1371.

IV. &hilling: What is the basis for assuming the AK factor


to be temperature independent and assigning all the observed
temperature dependence of the Isotope effect to the correlation factor?
H. Mehrer: A priori, there is indeed little reason to exclude
a temperature dependence of the AK factors. Anharmonicity
effects may lead to a decrease of AK with increasing temperature. This effect has been estimated for various metals by
Achar (see ref. [39) of the present paper). His calculations
suggest that the temperature dependence is very small. This
result is supported by isotope effect experiments on palladium [N.L. Peterson, Phys. Rev. 136 (1964) A568]. The
value of A K1 v = 1.04 f 0.05 observed near the melting tem-

perature leaves little room for an increase with decreasing


temperature. However, if necessary, a temperature dependence
of the AK factors could be taken into account in the analysis
of isotope effect data. In view of the limited accuracy of
isotope effect data this has not been done so far.
H.M. Gilder: How sensitive are your very nice multidefect
model best-fit curves of E and A V data to your obsolete
assumption of no temperature dependence of the properties
of each of the defects supposed to be present?
H. Mehrer: The answer to your first question concerning
the analysis of isotope effect data is the same as the answer
to Professor Schillings question.
In the fit of the effective activation volume of sodium as
a function of temperature shown in fig. 14 of the present
paper in terms of mono- and divacancy contributions a temperature dependence of the individual activation volumes
A V1v and A V~V has indeed been neglected. AV1 v and A V~V
are the only adjusted parameters in fig. 14 since the monoand divacancy contributions to the total diffusivity are those
deduced from the temperature dependence of the selfdiffusion coefficient. I think that it is important to note that
this procedure leads to a consistent interpretation of the temperature, mass, and pressure dependence of self-diffusion in
sodium with a reasonable value of AVzv/AVtv.
Unfortunately the theoretical situation concerning a possible temperature dependence of the activation volumes is
still confusing. Your calculations presented at this Conference
suggest that the vacancy gets bigger with increasing temperature whereas calculations mentioned in Dr. Stotts talk
during this Conference, if I understood him correctly, suggest
the opposite. I feel that, if there is a temperature dependence
of the activation volume, a decrease with respect to the
atomic volume is at least more plausible wince I would
expect that the vacancy due to anharmonicity effects
becomes more relaxed with increasing temperature.

C.P. Flynn: The Gibbs functions ordinarily met in thermodynamic relationships define the enthalpies, volumes and
entropies of bodies according to:
s = -(aG/aT)p:

V = @G/a&:

H = G + ST.

(1)

The specific heat follows as:

cp = (aHfaT), = r(as/ar),

(2)

There are also stability relations [ 1121 that constrain the possible behaviour of G for solid materials to be such that
c, > 0,

c>o,

(3)

where c is any elastic constant. In addition, perfectly ordered


substances conform to the Third Law:
s-*0

asT+0.

(4)

Quantities resembling Gibbs functions are also employed to


define the densities and migration rates of defects in solids
[ 1131. A question of concern at this Conference is whether

60

H. Mehrer /Atomic jump processes in self-diffusion

Fig. 16. Sketch of the surface GM@, T)describing mobility


near an order-disorder transformation.

or not these defect parameters conform to eqs, (l)-(4) and


in what way the equations can thus constrain defect properties.
It should be mentioned first that eqs. (1) are valid for any
arbitrary surface in p, T space provided that H is understood
to be the intercept at T = 0 of the tangent to G for p constant. Eqs. (2) then merelydefine a new quantity Cp in a selfconsistent way. These formulae do not constrain the possible p and T dependence of G in any way, whether or not
C@, 7) happens to be the Gibbs function of some physical
system.
Eqs. (3) pertain to the Gibbs function G(%) of some
specific, stable body, with $ the stress tensor and c the stiffnesses. They apply neither to the migration parameter GM
nor to foxmatian psopesties GF. The GM that describes defect

hopping bears no general relationship to a thermodynamic


Gibbs function, and so eqs, (3) are irrelevant. The third Law,
eq. (4), is equally invalid for the same reason. For formation
properties, GF is the differeence between properties of two
distinct bodies, one imperfect and the other perfect, so
eqs. (3) are once more not applicable. Since @G/a?),-+ 0
for both these bodies, however, eq. (4) wiltgenerally hold
for the case of defect formation.
Thus, in summary of these comments, eqs. f l)-(4) constrain the p, T dependence of GM and GF only through
(X$&7) + 0 as T+ 0. Even this is ordinarily of no practical
concern, being relevant only for T << 8D, with 8D the Debye
temperature.
A related question is raised by Fukai [ 1141. He points
out that GM for hydrogen diffusion in VH;! behaves more
Iike the crystal enthalpy H than the Gibbs function G, These
characteristics become very apparant near the order-disorder
transformation temperature. Note that this observation is in
no way inconsistent with the preceding discussion, which
allows arbitrary p, T dependences of migration properties. In
point of fact, a term proportional to Hdoes appear in the
microscopic theory of GM for interstitial solid solutions, and
the theory has been verified by means of the conductivity
in certain solid electrolytes [115]. Near the order-disorder
transformation GM takes the form sketched in fii. 16. This
system provides a striking example of the freedom with
which migration properties may vary with thermodynamic
state variables. The migration entropy s&f and migration
HM, derived from GM by means of eq. (l), each become
infite at the critical point. We have seen above that GF has
a similar freedom, away from the plane T = 0.

S-ar putea să vă placă și