Mass Transport Phenomena in Ceramics
-
Upload
others
-
View
0
-
Download
0
Embed Size (px)
Citation preview
MATERIALS SCIENCE RESEARCH
Volume 1: Proceedings of the 1962 Research Conference on Structure
and Properties of Engineering Materials
- edited by H. H. Stadelmaier and W. W. Austin
Volume 2: Proceedings of the 1964 Southern Metals/Materials
Conference - edited by H. M. Otte and S. R. Locke
Volume 3: Proceedings of the 1964 Conference on the Role of Grain
Bound aries and Surfaces in Ceramics
- edited by W. Wurth Kriegel and Hayne Palmour III
Volume 4: Proceedings of the 1967 International Symposium on
Kinetics and Reactions in Ionic Systems
- edited by T. J. Gray and V. D. Frechette
Volume 5: Proceedings of the 1970 Conference on Ceramics in Severe
Environments
- edited by W. Wurth Kriegel and Hayne Palmour III
Volume 6: Proceedings of the 1972 International Symposium on
Sintering and Related Phenomena
- edited by G. C. Kuczynski
Volume 7: Proceedings of the 1973 International Symposium on
Special Topics in Ceramics - edited by V. D. Frechette, W.C.
LaCourse, and V. L. Burdick
Volume 8: Proceedings of the 1974 Conference on Emerging Priorities
in Ceramic Engineering and Science
- edited by V. D. Frechette, L. D. Pye, and J. S. Reed
Volume 9: Proceedings of the Eleventh University Conference on
Ceramic Science devoted to Mass Transport Phenomena in
Ceramics
-edited by A. R. Coope'/" and A. H. Heuer
A Continuation Order Plan is available for this series. A
continuation order will bring delivery of each new volume
immediately upon publication. Volumes are billed only upon actual
shipment. For further information please contact the
publisher.
MATERIALS SCIENCE RESEARCH • Volume 9
MASS TRANSPORT PHENOMENA IN CERAMICS
Edited by
A. R. Cooper and A. H. Heuer Department of Metallurgy and Materials
Science
Case Western Reserve University Cleveland, Ohio
PLENUM PRESS • NEW YORK AND LONDON
Library of Congress Cataloging in Publication Data
University Conference on Ceramic Science, Case Western Reserve Uni
versity, 11th, 1974. Mass transport phenomena in ceramics.
(Materials science research; v. 9) "Proceedings of the eleventh
University Conference on Ceramic Science,
held at Case Western Reserve University from June 3-5, 1974."
Includes bibliographical references and index. 1. Mass
transfer-Congresses. 2. Ceramics-Congresses. I. Cooper,
Alfred R., 1924- II. Heuer, Arthur Harold, 1936- III. Title. IV.
Series. TP156.M3U53 1974 666 75-20154 ISBN-13: 978-1-4684-3152-0
e-ISBN-13: 978-1-4684-3150-6 DOl: 10.1007/978-1-4684-3150-6
Proceedings of the Eleventh University Conference on Ceramic
Science, held at Case Western Reserve University from June 3-5,
1974
©1975 Plenum Press, New York A Division of Plenum Publishing
Corporation 227 West 17th Street, New York, N.Y. 10011 Softcover
reprint of the hardcover 18t edition 1975
United Kingdom edition published by Plenum Press, London A Division
of Plenum Publishing Company, Ltd.
Davis House (4th Floor), 8 Scrubs Lane, Harlesden, London, NW10
6SE, England
All rights reserved
No part of this book may be reproduced stored in a retrieval
system, or transmitted, in any form or by any means, electronic,
mechanical, photocopying, microfilming,
recording, or otherwise, without written permission from the
Publisher
PREFACE
The Eleventh University Conference on Ceramic Science held at Case
Western Reserve University fran June 3 - 5, 1974 was devoted to the
subject of M:l.ss Transport Phenanena in Ceramics. '!his book
follows closely the fonn of the oonference. While the active
participation at the meeting was not reoorded, it is clear that
many of the contributors have benefited fran the ranarks,
suggestions, and criticisms of the participants. Fur thennore, the
session chainnen -- Delbert Day (Univ. of Missouri), WU. IaCourse
{AlfrErl Univ.) , W. Richard ott (Rutgers Univ.) , A.L. FriErlberg
(Univ. of Illinois), v. Stubican (Penn. State Univ.), and R.
Loehman (Univ. of Florida) -- successfully kept the meeting to a
reasonable schedule, but also stimulated the lively
discussion.
The book divides naturally into four sections, focusing on
correlation and ooup1ing effects in diffusion in ionic materials,
understanding of fast ion transport, diffusion and electrical con
ductivity in crystalline and glassy oxides and applications of
diffusion to oxidation and other processes of current
interest.
The editors have benefited fran the cheerful help and assis tance
of many people. !-1rs. Karyn P1etka typed the entire manu script
with unusual accuracy and tolerance. Mr. MakmJd E1Lei1, Ajit Sane,
Leslie M:l.jor and Ms. Jenny Sang provided the subject index. The
authors have been cooperative and understanding and we
ack.now1Erlge our enjoyment in working with them.
The conference receivErl financial assistance fran U.S. Army
Research Office, Durham, N.C., Air Force Office of Scientific
Research, Arlington, Va., and Aerospace Research laboratory, Wright
Patterson Air Force Base, Dayton, Ohio. This support made it
possible to bring together the distinguished group of contri
but.ors to this volt:roe.
Cleveland, Ohio March, 1975
CONI'ENTS
Non-Random Diffusion in Ionic Crystals • . . . . JOM R.
Manning
Correlation Effects in Ionic Transport Processes • . A. D.
LeClaire
Correlation and Isotope Effects for Cation Diffusion in Sirrple
OXides . • • . • • • • • • • • • •
N. L. Peterson and W. K. Chen
Coupling, Cross Tenus, Correlation and Conduction A. R.
Cooper
Irreversible Thermodynamics in Materials Problems F. A. Nichols, G.
P. Marino and H. Ocken
OXygen Redistribution in U02 Due to a Temperature Gradient . . . .
• . • . . . • . . • . • .
D. D. Marchant and H. K. Bcwen
The Thermodynamics of Diffusion Controlled Metamorphic Processes
.•.•..
George W. Fisher
FAST ION TRANSPORI'
Willem Van Gaol
Hiroshi Sato and Ryoichi Kikuchi
Alkali Ion Transport in Materials of the Beta Alumina Family • • •
• • • • • • • • •
Robert A. Huggins
Ionic Conductivity of Doped Cerium Dioxide H. L. Tuller and A. S.
NCMick
lJ::M Temperature Oxygen Transport in
CONTENTS
149
155
177
Nonstoichiametric ce02 • • • • • • • • • •• 187 S. P. Ray and A. S.
Nowick
Self Diffusion Constant Measurement by Continuous Wave Nuclear
Magnetic Resonance • • • •
J. Stepisnik, J. Bjorkstam and C. H. Wei
DIFFUSION AND ELECTRICAL mNDUCTIVITY IN CRYSTALLINE AND GlASSY
OXIDES
On the Interpretation of Lattice Diffusion in Magnesium Oxide • • •
• • • • • • • • •
Bernhardt J. Wuensch
H. P. R. Frederikse and W. R. Hosler
Concentration Dependent Diffusion of W in Ti02: Analysis of
Electronic Effects in Ion Diffusion
o. W. Johnson, J. W. DeFord, and S. -H. Paek
Measurement of Chemical Diffusion Coefficients in
Non-Stoichiometric Oxides using Solid State Electrochemical
Techniques • • . • • • • •
B. C. H. Steele
G. H. Frischat
Oxygen Diffusion in Liquid Silicates and Relation to Their
Viscosity •
Y. Oishi, R. Terai, and H. Ueda
201
211
233
253
269
285
297
CONl'ENTS
Silicon and Oxygen Diffusion in Oxide Glasses Helmut A.
Schaeffer
Alkali Ion Conductivity in Fused Silica Edward M. Clausen
Cation Migration in Electrcx:1e Glasses F. G. K. Baucke
A Structural Model of Ionic and Electronic Relaxation in Glass • •
• • • • • •
J. Aitken and R. K. MacCrone
Molecular Diffusion in Glasses and Oxides J. E. Shelby
APPLICATIONS OF DIFFUSION TO OXIDATION AND Ol'HER PROCESSES OF
CURRENT INTEREST
Mass Transport Phenomena in Oxidation of Metals Per Kofstad
The High-Temperature Oxidation of Hot-Pressed Silicon Carbide • • •
• • • • • • • • •
J. W. Hinze, W. C. Tripp and H. C. Graham
Sintering of Silicon Carbide • • • • SVante Prochazka
Determination of Phase Diagrams Using DIffusion Techniques • • •••
• • • • • • • •
Joseph A. Pask and llhan A. Aksay
Arrbipolar Diffusion and Its Application to Diffusion Creep • • • •
• • • • • •
R. S. Gordon
Crystal Growth and Glass Formation D. R. Uhlmann
The Effect of Diffusion and Shear on Concentration Correlations in
Binary Systems • • • • • •
J. H. Heasley
The Dissolution Kinetics of Lithiated NiO in Aqueous Acid Solutions
• • . . • • •
Chin-Ho Lee, Alan Riga and Ernest Yeager
Subject Index
John R. Manning
ABSTRACT
Non-random diffusion can result both from the presence of atonic
driving forces and from the notion of defects in a crystal.
Defect-related non-random effects appear in two different ways in
the kinetic diffusion equations, as correlation effects and as
defect-wind effects. In the present paper, the origin and magni
tude of these effects during diffusion in a driving force are dis
cussed. Kinetic expressions for the drift velocity <vF> are
derived from expressions for the effective frequencies of
independent atom j unps and are related to the tracer diffusion
coefficient D*. For impurity diffusion in an electric field,
deviations from the Nemst Einstein relation result from
defect-wind effects. Recently developed equations for the
ionic-irnpuri ty drift-rrobili ty when diffusion occurs via
divacancies noving on one of the sub-lattices in the NaCl structure
are summarized. Extensions of the sirople equations derived here to
more complex situations are discussed.
INTRODUCTION
Diffusion in crystals occurs by atoms following more-or-less random
walks. Each atom noves through the crystal by making a series of
elanentary atom jumps from one lattice site to another. An atomic
driving force will provide a bias to the directions of the
individual junps so that they are no longer random in direc tion.
Thus , driving forces, such as electric fields, are one
*Contribution of the National Bureau of Standards, not subj ect to
copyright.
1
2 JOHN R. MANNING
source of non-random effects. If each jUIl"q? has the same bias, as
for self-diffusion in a constant driving force, effects from
driving forces are easily treated by making simple rrodifications
in the random walk equations. Havever, when the biases are not
constant or when other non-random effects occur, such as the
defect-related non-random effects discussed later in this paper,
the analysis is more corrplex.
A category of non-randan effect which can arise even in the absence
of driving forces occurs when a given atom jUIl"q? is influenced by
the directions of previous jurrps taken by atans and defects in the
crystal. 'Ihe resulting "correlation effects" and "defect-wind
effects" can occur either with or without the presence of atomic
driving forces. 'Ihese non-random effects do not provide the same
bias on each jump.
In the present paper, the origin and magnitude of correlation and
defect wind effects will be discussed, particularly for the case
where there is an atonic driving force. Electric field effects will
be emphasized and explicitly discussed. Nevertheless, most of the
equations will apply equally well to effects from other driving
forces and other sources of defect fluxes, such as tem perature
gradients, stress fields, or gradients in concentration or
stoichiorretry.
Electric field effects are empahsized here since they provide a
particularly straight-forward exarrple of a driving force. Also,
electric fields have inportant effects on ions moving in ionic
crystals. In the present discussion, correlation and wind effects
resulting fran motion of individual defects will be emphasized.
Non-stoichiometric crystals with large vacancy concentrations would
require a more canplex treatment since interactions between large
groups of defects then should be considered.
NON-RANOOM DIFEUSION VIA MONOVACANCY MEX.::HANISM
Correlation and defect wind effects depend strongly on the
diffusion mechanism. When diffusion proceeds by the simple inter
stitial mechanism with no atomic driving forces, the individual
interstitial atans can follow truly random walks. 'Ihe kinetic dif
fusion equations then may be obtained from the simple mathematical
theory of randan walks. On the other hand, for most other diffu
sion mechanisms, especially those which require a mobile defect at
a neighboring site, the individual atans do not follow random
walks. 'Ihe direction of an atan jUIl"q? I for exarrple, will
depend on which particular neighboring site is occupied by a defect
at the time of the jump. Consequently, when diffusion of atans
occurs via motion of vacancies, di vacancies, intersti tialcies,
crowdions I or even more-cOITlplex mobile defects, defect-dependent
non-random diffusion
NON-RANDCM DIFFUSION IN IONIC CRYSTALS
will occur [1,2,3].
To illustrate the origin of these defect-dependent non random
effects, let us consider diffusion by the rronovacancy mechanism.
In this mechanism, the elementary atom jump involves the jumping of
an atom into a neighboring vacant lattice site. '!hese vacant
sites, or vacancies, are assu:rred to be present in thenncrlynamic
equilibriun in concentrations which depend on exp (-Ef/kT) where Ef
is the fonnation energy for vacancies, k is Boltzmann's constant
and T is the absolute temperature. In discussing the monovacancy
mechanism, one assu:rres that each vacancy rroves independently in
that it is not bollild to other vacancies and does not interfere
with diffusion processes involv ing oLher vacancies. By contrast,
in the divacancy mechanism, it is assuned that vacancies move as
part of vacancy pairs.
3
With respect to non-random diffusion, the important thing about the
vacancy mechanism, and also the di vacancy mechanism, is that an
atom cannot move lliltil a vacancy arrives at a site neighboring on
it. '!hus, the independent diffusion process is not the single atom
jurrp but instead requires a sequence of jumps. In this sequence,
the vacancy is first created at a vacancy source. '!hen the vacancy
rroves through the crystal by exchanging with the various atans in
the crystal lliltil it arrives at a site neighbor ing on the
diffusing atom i. After arrival, the vacancy can exchange with atan
i one or rrore times. Finally the vacancy will move pennanently
away and be destroyed at a vacancy sink.
'!his sequence of interrelated vacancy jumps may be contrasted to
the situation in the simple interstitial mechanism, where the atcm
jurrps directly fran one interstitial site to another, with th~
direction of each atom junp being independent of the direction of
jump of any other neighboring atcm or defect. en the other hand,
for diffusion via an interstitialcy mechanism, an intersti tialcy
must approach the atom much as described above for diffusion via a
vacancy mechanism.
In principle, one must follav the complete paths of the vacancies
if one wishes to detennine the kinetic diffusion pro cesses
affecting the particular atcm i. In simple situations, how ever,
there are several ways to simplify the calculation.
TO present the defect-dependent non-randan diffusion equa tions in
as simple a fonn as possible while still allowing discus sion of
the influence of driving forces on these equations, atten tion
will be restricted to high-synmetry crystals, such as cubic
crystals, with mirror synmetry across all lattice planes normal to
the diffusion direction. '!he diffusion direction x will be chosen
nonnal to low-index planes separated by a regular inter planar
spacing b, and it will be assumed that the x components of
4 JOHN R. MANNING
the possible atcm jumps are all +b, zero, or -b. Thus, an atom or
vacancy cannot reach a lattice plane that is 2b away without
stopping at a lattice site one interplanar distance <May in that
direction. As examples, <lOO>-diffusion directions in a
variety of cubic struc tures satisfy these rules provided that
only jumps to nearest neigh bor sites are allcwed. Also,
<lOO>-diffusion directions in tetra gonal or orthorhombic
crystals would be sui table. In cubic crystals diffusion is
isotropic, so this arbitrary choice of dif fusion direction does
not make L~e result any less general.
Driving forces can be expected to create a net vacancy flux, which
normally will be distorted in the vicinity of an impurity atom i.
(A distortion is expected unless the diffusion properties of i are
identically the same as those of all other atoms in the crystal. )
Nevertheless, if there is a mirror crystal symnetry plane passing
through i nonnal 'co the flux, the average vacancy concentrations
on this plane will not be altered from those in the absence of a
flux. Thus, sites along this plane will maintain equilibrium
vacancy concentrations and for present purposes can be treated as
effective vacancy sources-and-sinks. Any sequence of vacancy
jUllIJS thus is regarded as terminated when the vacancy reaches
this syrrmetry plane (but not tenninated by exchange with atom i
itself since after such an exchange the atan and vacancy still are
not on the same plane). Further jurrps by the vacancy starting from
the syrrmetry plane will begin a new independent sequence,
uncorrelated to the previous sequence. Since the only way for a
vacancy during a single jurrp sequence to move from one side of
atom i to the other is by exchange with atom i itself, a single
vacancy sequence cannot provide atom i with two consecutive jumps
in the sane sense. Instead jurrps by atom i from a single vacancy
sequence will cause alternate +b and -b displacements and the net
displacement from any sequence will be either ±b or zero.
After one exchange with atom i in a given direction, the vacancy is
in the proper position to cause a reverse jump, moving the atan in
the opposite direction. Such a second jurrp would cancel the effect
of the first jurrp. Similarly, the fourth exchange of the vacancy
with i will cancel the third exchange, the sixth exchange will
cancel the fifth exchange, and so on. After an odd nunber of
exchanges the net displacerrEl1t of i equals b, the interplanar
spacing; and after an even number of exchanges the net displacement
is zero.
After the vacancy has arrived at a site a, neighboring on atom i on
the +x side, it will have a probability P+ of eventually causing a
+b jurrp of atan i. This jurrp may occur on the next jurrp of the
vacancy or the vacancy may first wander away from atom i and only
later return to exchange with it without arrival of the vacancy at
any effective vacancy sink. After a P+ exchange the vacancy will be
on the -x side of atom i in proper position
NON-RANDa-1 DIFFUSION IN IONIC CRYSTALS
to cause a -b jump. The probability of the vacancy causing such a
-b jump (again on any of its subsequent jumps before arrival at an
effective vacancy sink) is defined as P _. If there is a driving
force, P _ will differ from P +; but, in the absence of a driving
force, P_ equals P+ since the only distinction between P+ and P _
is the direction in which the exchange occurs.
The frequency '.1+ with which atom i undergoes a displace:rrent,
always of magnitude b, in the +x direction therefore is:
where vn+ is the frequency of arrival of fresh vacancies at sites a
neighboring on i on the plane on the +x side of i. (These vacancies
then are in the proper position to allow a +b jump of atom i on the
next jump.) The sun involving P+ and P_ tenus in Eqn. (1) gives the
probability of the vacancy causing a net atom displace:rrent +b
after arrival at site a. Thus, '.1+ is the effective frequency of
independent displacements +b.
5
In Eqn. (1), the initial term P+ within the brackets represents the
probability of an initial exchange of an a-site vacancy with atan
i. The second term P~_ represents the probabil ity of a two-junp
sequence where the second junp cancels the effect of the first
jump. Hence, this term appears with a minus sign. The third term
represents the probability of a three-junp sequence, again
providing a net displace:rrent +b of atan i, and so on.
A fresh vacancy is defined as one which in its sequence of jumps
has not previously arrived at a site from which it could cause a +b
or -b j unp of the tracer. As noted previously, any vacancy which
arrives at the mirror symmetry plane passing through atom i normal
to the diffusion direction ends one sequence of jumps and in effect
becomes a new fresh vacancy during its subsequent pa.th. Thus, in
its subsequent path it again can have a first arrival at a site
a.
In Eqn. (1), v + represents a sun over all possible paths from any
vacancy source (or effective fresh vacancy source on the mirror
symmetry plane) to any site a which will allow a +b junp of atom i.
For simplicity, it is assuned in the present equation that all a
sites are equivalent to each other. Then vn+ is a simple scalar
quantity, while still representing a sum over all a sites. This
condition will be satisfied in any crystal which has suf ficient
rotational or mirror symmetry around the diffusion direc tion that
rotations or reflections of the crystal can be applied to bring any
site a into the original position of any other site a and these
rotations or reflections also reproduce in the trans formed
orientation all other lattice sites in the original orien tation.
In such a situation, a vacancy at anyone a site has the
6 JOHN R. MANNING
same effect of diffusion along the x-axis as does a vacancy at any
other a site. This condition will be satisfied by many crystals
(cubic, tetragonal, orthorhanbic) for which the previously
introduced mirror symmetry nor.mal to the diffusion direction
applies.
P+ can be expressed in terms of the possible vacancy jump
frequencies from sites a. In particular, P+ depends on the junp
frequency w2+ for a vacancy on site a to cause a +b jump of atan i
and on the carpeting jump frequencies w+rr for the vacancy to start
a path which leads it away from site a to an effective vacancy sink
without exchangin:j with atom i or returning to any site a. Here
w+rr represents the sum over all jumps (to any sites) which
actually start a vacancy on such a path of non-exchange and
non-return. In w+rr ' one must exclude paths which eventually
return the vacancy to a site a or which involve a direct jump from
one site a to another site a. Detailed calculations sheM that these
a to a transitions do not affect P+.
Physically the reason for omitting a to a transitions may be seen
fran the definition of P +. In the calculation of P +, one must
follow the vacancy until it either exchan:Jes with atan i or
pemanently rroves away by reaching an effective vacancy sink. There
is unit probability that one of these two events will even
tuallyoccur and P+ is the probability of the former, Le., exchange
with atan L The relative probabilities of an eventual w2+ junp or
w+rr junp path are the same regardless of which a site contains the
vacancy. Thus, transitions fran one a site to another do not affect
P+ and these transitions may be anitted fran further
consideration.
The expression for P+ therefore is simply
Also
SummiIB" the series in Eqn. (1) yields
v+ = vrr + P+ (l_P_p+)-1 (l-P_)
(2)
(3)
(4)
This equation may be recast into a form which explicitly displays
the correlation factor and a familiar expression for the jump fre
quency. Upon application of Eqn. (3) relating P+ to I-P+, one
obtains
vrr+ (l-P+) (l-P_)
v+ = w N Z (l-P_P+) w2+ Nva za +rr Va a
(5)
NON-RANDOH DIF.EUSION IN IONIC CRYSTALS 7
'Where NVo: is the equilibrium vacancy concentration at site 0: and
Zo: is the number of equivalent sites 0:. According to Eqn. (5),
the effective jurrp frequency v+ of jurrps +b can be expressed as
the product of four factors,
where
(6)
(7)
(8)
(9)
Here rb+ is the basic jurrp frequency for an atom to jurrp to a
given site in b1.e +x direction. 'Ihis basic frequency is the jurrp
frequency which would apply if the only non-randan effects were
those fran driving forces, i.e., in the absence of defect-related
non-randan effects. These defect-related non-random effects appear
in Eqn. (6) as the oorrelation factor f and the vacancy wind factor
'4.
The basic jurrp frequency rb+ will be affected by driving forces
along the x-axis since these driving forces influence the vacancy
jurrp frequency w2+. By contrast, the correlation factor f is
unaffected by driving forces to first order, since P+ and P_ appear
symmetrically in Eqn. (8). The vacancy wind factor '4 given in Fqn.
(9) is unity if there are no vacancy fluxes. A driving force
nonnally will provide a vacancy flux and make ~ differ fran
unity.
In Fqn. (9), v1l+ is the frequency with which vacancies leave
vacancy sources-and- sinks at sites 11 and travel to sites 0:
neigh boring on the atan, with the 0: sites here being those on
the +x side of the atan. Similarly, w+1I NVo: Zo: is the frequency
with which vacancies follow the reverse paths fran sites 0: to
sites 11.
If there is a net vacancy flux, vacancies will travel more fre
quently in one sense along these paths than the other. Thus, a net
vacancy flux makes '4 differ from unity, whereas in the absence of
such a flux ~ equals unity. This general oonclusion will apply even
when there are interactions between defects.
'lb first order in the driving forces, Eqn. (8) reduces to
1 - P f = 1 + P (10)
where P is the average of P+ and P_. Physically the oorrelation
factor occurs here because a vacancy can exchange with an atom more
than once during a single independent sequence of jurrps.
8 JOHN" R. MANNING
Notian .of the atom itself changes L'le local probability .of a
vacancy being on the +x or -x side .of the atan.
By contrast, the vacancy wind factar arises because .of motian .of
ather atans. When there is a driving farce along the x-axis motion
.of these ather atoms (i. e., ather than the atom i whase jump
frequency v+ is being calculated) will bring vacancies up to sites
ex by paths contributing to v'Tf+ more frequently than motion .of
these ather atoms will carry vacancies in the apposite directian
alang these paths.
Atonic driving farces which bias the directians .of individual atan
jurrps are nat the .only influences which can yield a net vacancy
flux and hence cause G.t ta differ fran unity. For exarrple,
variaus types .of concentration gradients .or stoichianeb:y
gradients can alsa produce net vacancy fluxes. In the present
discussian, however, attentian will be restricted ta vacancy fluxes
resulting from atomic driving farces, and particularly fran
electric fields.
Equatian (6) yields the effective jurrp frequency v+ far an atan
jurrp in the +x directian. A similar equatian is faund far the
effective jurrp frequency v_ in the -x direction,
(11)
where f again is given by Eqn. (8), rb- is .obtained from Eqn. (7)
merely by replacing all subscripts + with subscripts -, where the -
subscript refers ta atOll jumps in the -x directian, and G is
.obtained from Eqn. (9) by replacing subscripts + with subscripts -
where the - subscript refers ta a site in the -x directian fram the
atom.
CALCULATION OF DIFFUSION COEFFICIENT D* AND DRIFT VELCX:ITY
<vF> FRCM EFFEX::TIVE JUMP FREQUENCIES
The basic kinetic diffusian equation far planar diffusian in the
x-direction relates the atom flux J ta the concentratian c and
cancentratian gradient ac/ax .of the diffusing species,
J = -D* (ac/ax) + <vF> c (12)
This equatian contains DNa measureable coefficients, the tracer
diffusian coefficient D* and ti1e drift velacity from atonic
driving farces <vp>' If atonic driving farces are the .only
nan randan effects wh~ch .occur, as far diffusian by a simple
intersti tial mechanism, D* and <vF> are related by the
well-known Nernst Einstein equatian [4],
(13)
NON-RANDCM DIFFUSION IN IONIC CRYSTALS 9
where F is the atanic driving force. For diffusion in an electric
field E
F = q E (14)
where q is the charge of the diffusing ion. The drift rnobili ty 11
is defined as the drift velocity in unit electric field.
Thus,
(15)
and for the specific case of an electric field, the Nernst-Einstein
equation becanes
..L=...3. D* kT
(l6)
~Vhen defect-related non-random effects occur, as for diffu sion
by a vacancy mechanism, it is found that the Nernst-Einstein
equation no longer applies. The deviations which are found can be
described by considering the ways in which the effective junp
frequencies v+ and v_ differ from the basic jump frequencies fb+
and fb-.
In tenns of effective jump frequencies, if v+ and v_ are
independent of location in the crystal, as for diffusion in a
han~eneous crystal with a constant electric field,
D* = ~ b 2 (v+ + v_) (17)
and
<v > = b(v - v ) F +- (18)
These equations are very similar in appearance, the only
differences being the square and half which appear in Eqn. (17) and
the fact that the v's are added in Eqn. (17) but subtracted in Eqn.
(18). Because of this slinilari ty, it is not surprising that D*
and <vF> can be related to one another.
When Eqns. (6) and (11) are inserted into Eqn. (17), one
finds
D* = ~ Zab2f (G+fb+ + GJb-) (19)
'lb first order, G+ and G_ can be written as
G = 1 + ~G (20) ± -
where G is a small quantity calculated from Eqn. (9). Also
fb± = fbo exp(±bF/2kT) (21)
10 JOHN R. MANNING
where fbo is the value of fb+ or fb _ in the absence of atomic
driving force F. Expanding the exponential to first order
yields
fb± = fbo [1 ± (bF/2kT) + • . . . ] (22)
When Eqns. (20) and (22) for G± and fb± are substituted into B:Jn.
(19), one finds
D* = Z b2 f f a bo (23)
When these same equations are substituted into Eqn. (18) for
<vF>' one finds
<VF -> = za b f (G+ fb+ - G_ fb_)
and to first order
(24)
<vF> = za b 2 f fbo [(fb+ - fb_)f~ b-l + (G+ - G_)b-l ]
(25)
(26)
where
(27)
The tenn B, which arises from the vacancy wind factors G+ and G_,
is the tenn which gives rise to deviations from the Nernst-Einstein
relation, Eqn. (l3) or:f.qn. (16). In the special case of
self~iffusion, 1 + B equals f-l. This relationship, especially when
derived by methods other than that given above, has inspired
statements in the literature that <vF> /D* differs fran F/kT
because of the correlation factor. It would appear more appropriate
to say that the deviations are proportional to 1 + B, with B
arising from a defect~ind effect.
'Ihe factor I'.G in B:Jn. (27) depends on the driving forces and
jump frequencies influencing the motion of the vacancy as it
approaches or leaves the vicinity of atom i. In the special case of
self~iffusion, these forces and frequencies are the same as those
for motion of atom i itself. As a result, I'.G for self diffusion
is directly proportional to the force F, and B for self diffusion
is independent of F. Here F is the force which acts specifically on
atom i. For impurity diffusion by a vacancy mechanism, hONever, the
forces contributing to the vacancy flux and to I'.G often differ
from F. Then the expression for B can become rather complex,
depending on the jump frequencies and driving forces for all atoms
and ions in the crystal.
NON-RANDCl1 DIFEUSION IN IONIC CRYSTAlS
Detailed calculations show that the vacancy wind effect for
.impurity diffusion usually differs from that for self-mffusion. In
particular, 1 + B for .impurity diffusion can differ greatly from
f-l and in some cases will even be negative [5].
For diffusion by a monovacancy mechanism, negative values
11
of 1 + B are obtained mainly when the irrpurity and vacancy are
tightly bound together. In an ionic crystal, the force from an
electric field will create a flux of solvent ions on the sub
lattice containing the impurity. '!his ion flux will cause an equal
and opposite flux of vacancies. For vacancies bound to impurity
ions, the vacancy-solvent flux merely moves vacancies fran the
dONn-stream side of irrpurities to the up-stream side. Because of
this vacancy redistribution, the impurities are pro vided with a
greater than randan opportunity of jurt"q?ing up-stream (in a
direction opposite to the solvent ion flux) than of jumping
dONn-stream. '!hen, if the impurity charge is less than twice the
solvent ion charge, the vacancy-irrpurity carplexes may actually
move up-stream, opposite to the normal direction of flow of the
charged solvent and impurity ions.
Even if the drift mobilities of particular irrpuri ty ions do
becane negative, the vacancy wind effect will not lead to a
negative ionic conductivity in the crystal. Physically, this result
can be described in terms of the charge flows carried by the
solvent ions. For example, one can regard the vacancy impurity
canplex as occupying lattice sites that oth.el:wise would be
occupied by two solvent ions. Movanent of this conplex up stream
will be accorrpanied by a countennotion of solvent ions down-stream
with the conplex in effect exchanging places with two solvent ions.
Since the motion of the canplex up-stream occurs only if the
.impurity charge is less than twice the solvent charge, this
interchange of positions actually moves net ionic charge
dONn-stream. A positive value of the ionic conducti vi ty thus is
obtained, as expected.
DEVIATIONS FRCM NERNST-EINSTEIN REIATION FOR DIFFUSION CN FCC
IA'ITICE VIA DIVACANCY MEOfANISM
Recently, results have been obtained [6] for impurity diffu sion
via divacancies on a face-centered cubic lattice, which is the same
lattice as each sub-lattice in the NaCl structure. In this
calculation, it was assumed that the two vacancies in each
divacancy are tightly bound to one another so that the two
vacancies in the divacancy are always nearest neighbors of each
other. '!he di vacancy migrates by means of individual vacancy
jumps which move the divacancy through the close-packed fcc lat
tice without dissociation.
12 JOHN R. MANNING
The simple treatment of Eqn. (1) which was applied to the
monovacancy mechanism must be modified here, since in the ill
vacancy mechanism not all configurations ex which would allow an
atom jump in the +x direction are equivalent. Instead there are two
types of oonfigurations from which at least one of the vacancies in
the divacancy can cause a +x atom jump [7]. In configuration 1,
only one of the vacancies in the divacancy has a different x
coordinate than the diffusing atom whereas in oonfiguration 2 beth
vacancies are on the plane in the +x direction fran the atan. Then
\!+ and \!rr+ in Eqn. (1) becane two-oomponent row vectors. The
carponents of \!+ are the effective jump frequencies for jumps
caused by divacancies in oonfigurations 1 and 2, and the
cc:mponents of \!rr+ are the arrival frequency of fresh divacancies
at configuration of type 1 am 2. The summation involving the P+ and
P_ quantities in Eqn. (1) also must be rrodified, with the
quantities in the summation becc:ming 2 x 2 matrices and including
possible transi tions between configurations 1 and configurations
2.
Wi th this approach, matrix equations similar, but not identi cal,
to Eqns. (6) and (10) can be developed; and it is found that the
diffusion flux J of a dilute impurity diffusing via divacancies in
a oonstant electric field E in an otherwise pure fcc crystal is
given by [6]
cD* q. E J = - -D* ~~ + k~
(3.89wl + 13.26 w3 ) qs
[1 +(7.75Wl + 6.75 w3 ) qi] (28)
where qi is the charge of the impuri ty ion, qs the charge of the
solvent ions on that sub-lattice, WI is the jump frequency for a
jump of a solvent ion where roth before and after the jump both
members of the di vacancy pair neighOOr on the impurity, and w3 is
the jump frequency for a vacancy jump with a solvent ion where
before the jump roth members of the divacancy neighOOr on the
impurity but after the jump only one neighOOrs on the impurity.
'IWo other jump frequencies also are allowed in this model: Wo for
other allowed vacancy-solvent exchanges (with dissociation of the
divacancy not allowed) and w2 for exchange of a vacancy with the
impuri ty . These frequencies w and w do not appear in the expres
sion, equal to 1 + B, inside tRe bra~ets in Eqn. (28), but w2 does
enter into the expression for D*.
Comparison of Eqn. (28) with Eqns. (12), (14), and (15)
yields
y _ qi (3.89 WI + 13.26 w3) qs
D* - kT [1 + (7.75 WI + 6.75 w) qi] (29)
which shows a significant deviation fran the Nernst-Einstein equa
tion.
NON-RANDOM DIFFUSION IN IONIC CRYSTAIS
Fbr self-diffusion, where wI = w3 and qi = qs' this expres sion
reduces to
)1 _ 2.183q D* - kT (30)
13
The techniques used to calculate 6G in the derivation of Egn. (28)
also have been applied to calculate the correlation factor f for
self-diffusion on an fcc lattice by a divacancy mechanism [8],
yielding the result
where
8 WI + 9.15 w3
(31)
(32)
This expression for the correlation factor is similar in form to
that obtained by Mehrer [9], though the numerical coefficients in
this expression for <p are smaller than Mehrer's. Fbr
self-diffusion, where WI = w2 = w3 ' Egn. (31) becomes
-1 f = 0.4581 = (2.183) (33)
Thus, again for self-diffusion,
~ = --.SL D* kTf (34 )
Nevertheless, for impurity diffusion, )1/D* can differ greatly from
q/kTf, as can be seen by canparing Egns. (29) and (31).
An important part of the derivation of both Et]:ns. (28) and (31)
was the establishment of an accurate value for w+7To = w7T+o' the
effective escape or approach frequency of a vacancy from or to the
impurity in the absence of a driving force, equal to the average of
W+7T and W7T • This quantity appears both in the equa tion for P,
which yields f by Egn. (10), and in the equation for 6G, obtained
from Egns. (9) and (20). Thus, although the defect wind effect and
the correlation factor effect are distinct effects, one depending
solely on solvent ion jumps and the other depending to a large
extent on the frequency of impurity ion jumps, the common term
w+7TO enters into both effects.
14 JOHN R. MANNING
MORE GENERAL TRFATMENTS
In the discussion above, several simplifying ass1.lItptions were
made to provide ease of treatment. For exanple, in Eqns. (17) and
(18), it was assumed that v+ and v_ would not vary with position
(diffusion coefficients and driving forces are constant); and
throughout the discussion simple crystal structures and diffusion
directions were assumed. In treating actual problems, these
assumptions are usually not necessary.
It can be shown [10) that diffusion coefficient gradients and
driving force gradients will not affect the kinetic expres sions
for D* and <vF> as measured at a particular plane. Thus ,
even t.hough the average drift of a layer of tracer atans can be
sanewhat affected by such gradients after the tracer atans move
away fran the central plane, the diffusion flux at this central
plane is not affected by these gradients. :fur driving forces
resulting from electric fields, an essentially constant force will
often be present throughout a particular diffusion zone. Then, the
kinetic equation above can be applied directly. Dis cussions of
modifications which are required when there are other driving
forces or effects from diffusion coefficient gradients can be
found, for example, in references 1 and 3.
In more canplex crystals and for diffusion by more canplex defects,
the syrranetry planes used to simplify the above equations often
will not be present. Nevertheless, it is possible in these cases to
use canplete-path equations [11] which when reduced to a matrix
fonn yield equations very similar to Eqn. (6)- (10). These matrix
equations will allow treatment of non-rarn.cm effects by individual
defects in any crystal.
REFERENCES
1. J. R. Manning, Diffusion Kinetics for AtOllS in Crystals (D. Van
tbstrand, Princeton, N. J., 1968).
2 • A. D. IeClaire, in Physical Chenistry, An Advanced Treatise
Vol. X, Solid State, edited by H. Eyring, D. Henderson, and W. Jost
(AcadEmic Press, New York, 1970).
3. J. R. Mmning, in Diffusion (Arrerican Society for Metals, Metals
Park, Ohio, 1973) pp. 1-24.
4. See e.g., A.B. Lidiard in Handbuch der Physik, edited by S.
Flugge (Springer-Verlag, Berlin, 1957), Vol. 20, p. 324.
5. J.R. Manning, Phys. Rev. 139, A2027 (1965); also in Mass
Transport in Oxides, National Bureau of Standards Special
Publication 296, edited by J.B. Wachtman, Jr. and A.D. Franklin
(U.S. (bvernment Printing Office, Washington, D.C., 1968), pp.
53-63.
NON-RANDClJ! DIFFUSION IN IONIC CRYSTALS
6. J. R. Manning, Abstract Bulletin of 'IMS!AIME, Fifth Annual
Sprin;J Meetin;J, Philadelphia (1973) p. 148.
7. R.E. Howard, Phys. Rev. 144, 650 (1966). 8. J.R. Manning, Bull.
Am. Phys. Soc., Ser. II, 18, No.3,
426 (1973). 9. H. M8hrer, J. Phys. F: IVEtal Phys. 2, Lll
(1972).
10. J.R. Hanning, Phys. Rev. 139 A 126 (1965). 11. J.P. Stark and
J.R. Manning, Phys. Rev. ~, 425 (1974).
15
A. D. Le Claire
1. INTRODUcrION
"Correlation Effects" in diffusion and other rrass transport
phenomena were first discussed for solids by Bardeen and Herring
[1] in 1951, although similar phenomena had been recognized as
occurring in neutron diffusion as previously as 1936 [2] and in
gases even earlier [3]. The effects derive from the fact that, in
the nature of the defect processes by which atoms migrate in
solids, the direction of any atom displacement is not at randan but
is influenced in part by the directions in which the previous and
earlier j1.lITpS of that atom occurred. That is to say, successive
jump directions are correlated with one another [4].
The commonest situation is that of vacancy diffusion. SUp pose an
atom has just made one j1.lITp by exchange with a vacancy. Because
the vacancy is adjacent on the original site of the atom, the atom
is rrost likely to make its next j1.lITp back to this site and
least likely to rrake it in the same direction as the first
j1.lITp. Consecutive pairs of jumps with zero net displacement
occur rrore frequently, and those with maximum net displacement
less frequently, than randani the effect is to reduce the diffusion
coefficient, by a factor f, below the value calculated assuming
displacements in any direction occurred with equal probability. f
is called the "correlation factor".
Similar effects occur with other types of defect, as we shall
see.
Correlation effects are closely related to the occurrence of
non-zero cross-terms in the irreversible thermodynamic
for.mulation
17
18 A. D. LE ClAIRE
of nass transport [5,7]: these aspects of the subject are dis
cussed in the following paper by Cooper. In this paper we shall be
concerned with the random walk. fonnulation for this provides the
expressions from which correlation factors can be calculated.
2. RANOOM WALK EQUATIONS [6-8J
It is a straightforward matter to derive a quite general expression
for the net flux of matter J, in a concentration gradient. This
is
2 2 <X> dC <x > dC d <X >
J = c (Xo)-t- - dX 2t" - c (Xo ) dX dC 2t" (1)
+ higher order tenns
c(Xo ) is the concentration of the diffusing species at x = Xo'
where J is evaluated. <X> and <X2> are respectively the
rrean dis plaCEment and mean square displacement of a diffusing
atom after a time t.
Fbr the simple but important cases of tracer diffusion in
effectively chemically homogeneous systEmS, diffusion properties do
not vary with posi tion in the sample and the third tenn in (1) is
zero. Also, <X> and other odd moments are zero. Provided we
can ignore the higher order terms, eqn. (1) reduces to Fick' s law
with
(2)
which is Einstein's equation. The asterisk indicates the restric
tion to tracer type diffusion in homogeneous SYStEmS. There are
similar equations for Dy* and Dz*. In isotropic crystals
-+2 D* = D* = D* = D* = <R > / 6t (3 ) x Y z
~ -+ <R > is the mean square of the total vector displacement
R of an atom in time t.
Equations (2) and (3) apply for any tracer self-diffusion co
efficient in a chEmically homogeneous crystal. In the general case
of diffusion in a chEmical concentration gradient all three tenus
of equation (1) have to be considered in developing expressions for
the intrinsic diffusion coefficients and the chemical interdif
fusion coefficient ~ [7]. These also of course contain correlation
effects, but the correlation factors involved are, to a very good
approximation, numerically the same as those that occur with self
diffusion, so we confine our considerations to this simple
case.
CORRELATION EFFECTS IN IONIC TRANSPORT PROCESSES 19
3. GENERAL EQUATICNS FOR CORREIATIQN FACroRS [7,8J
Each R in eqn. (3) is the vector sum of individual displace ments
h of an atom. Let there be on average N displacem:mts in time t, or
r in unit tline (N = rt). 'Ihen
<R2> = < (l:r. )2> ~
~ ~ ~ ~ ~
If all the ;i have the same magnitude r and if <;i·;i .> (for
a given j) has the same value whichever junp is represented b;?;i,
then all jumps are said to be of the same type. 'Ihis is the case,
for example, for self or impurity diffusion by the vacancy
mechanism in a f .c.c. or b.c.c. crystal or sub-lattice. It is then
convenient to write
+ + 2 <r .• r. . > = r <Cos8 . > (5) ~ ~+J J
where <CoS8j> is the average value of the rosine of the aIBle
between anyone jump am the jth next junp. Equations (3), (4) and
(5) then yield
D* = ! rr2f (6) 6
(7)
f is obviously the "correlation factor", as defined in the
Introduc tion, because if successive displacements take place at
random all the <Cos8y will be zero and f = l.
Generally, for diffusion by a defect mechanism f is not unity. We
have imicated already in the Introduction how correlation effects
cane about with diffusion by vacancies. It will be evident fran
what vas said that <Cos8F will be negative, that <Cos82>
will be positive but snaUer than <CoS8l>' and so on. We
expect the various <CoS8j> to be related and it is easily
swwn that if each junp vector is an axis of at least two-fold
symmetry
<CoS8.> = <CoS8 >j J 1
Equation (7) row becanes
for vacancy diffusion. Clearly fv~l.
More generally, when atans are diffusing by two or more types of
jump it turns out to be more convenient to work with the com
ponents of -;: i along the diffusion direction. If this is the-X=
axis these are xi arrl we put X =2l:Xi in eqn. (2). In place of (4)
we then have <x2> in terms of <Xi> and the <Xi
Xi+j>. Jumps are now of the same type, or equivalent, if for
each of then <xf> arrl <Xi xi+j> have the same values.
Let n be the number of different types and r a the number of jumps
of type a made by an a tom in unit time. We now easily find for D.x
the result
where
D X
f aX
1 n 2 - l: r x f 2 a=l a a aX
= 1 + 2 l: <x X . > Ix2 j a aJ . a
(10)
(11)
Xa is the camnon value of Xi for jumps of type a and Xaj is the x
cc:mponent of the jth jump following an a type jump. fax is the
"partial correlation factor" for diffusion along X associated with
jumps of type a. Generally there will also be a set of fay and f aZ
for diffusion along the y and z direction respectively ..
In terms of an overall correlation factor for diffusion along x,
fx' eqns. (12) may be rewritten
Dx <} l: r a x~) f X (12) a
where
(13)
and cq, = r a/ l: r is the fraction of all jumps that are of type
a. Equation (ll~ ai\:i (13) reduce to eqn. (7) when n = 1.
Obvious applications of these equations are to self diffusion by
vacancies in anisotropic crystals and to solvent diffusion, for
example in a dilute alloy, or doped ionic crystal. In the latter
cases there are different types of jumps because the <Xi
xi+j> will generally depend on the proximity to an impurity atan
ot the par ticular solvent jump represented by xi. c a and n can
generally be readily determined by inspection.
There are similar correlation effects with intersti tialcy dif
fusion, as occurs for example for the cations in A:J halides and
the anions in many fluorite type structures. Figure 1 sho.vs
three
CORRELATION EFFEx:::TS IN IONIC TRAN'SPORT PROCESSES 21
x X X X X X
)~X X
"'- X
(a) (b) ( c )
Figure 1 -- Interstitialcy Diffusion.
consecutive jUIrg?S of a tracer atom. As with vacancy diffusion,
the second junp is more likely than rand.an to be, as shONl1, in
the reverse direction to the first because, in this case, an
intersti tial atom is immaiiately available to induce such a
jUIrg? A jurtp of the tracer in a different direction would first
require inter stitialcy defect migration out of configuration (b),
through the surrounding lattice, so as to bring an atan into one of
the other interstitial sites adjacent to the tracer. lbwever, there
is a difference row fran vacancy diffusion: the third junp of the
tracer out of v.7hichever i-site it occupies after its second junp
(Fig. lc) will occur with equal probability to any of the neighbor
ing and. equivalent lattice (1) sites. Thus it is not correlated
with the secorrl nor, therefore, with the first jump. It follows
that <Cosej> = 0 for all j ~ 2. Also <Cosel> is
non-zero only for correlated consecutive p::tirs, i-+l-+i, of
jurtps. These are half the total number of p::tirs so equation (7)
becanes
fI = 1 + <Cosel >· 1 . ~-+ -+~
(14)
Equation (9) and (14) contain only <Cosel >. In other words a
calculation of f requires consideration of the correlations only
between consecutive junps. The same result is obtained in the
22 A. D. LE CLAIRE
further develop:nent of the more general equations (11) or (13): a
relation analogous to (8) allows the surrmation in eqn. (11) to be
carried out to give a result in terms related only to the <xa
Xal>'
Another carmon rrechanism is that of silrq;lle interstitial dif
fusion, where atans diffuse by junps fran one interstitial site to
another. This occurs in metals for such solutes as H, C, N, 0 etc.
, and is a possible mechanism by which ions in excess of stoichio
metric and accommodated interstitialy, may diffuse. Provided the
concentration of interstitials is low enough all junp directions
are equally probable at every jump so that f = 1. However, at
higher concentrations some jump directions of an atan will be
blocked by adjacent interstitial sites being occupied. SUch
blocking will have the effect of giving <OOS81> a negative
value, because at least the site fran which an atan last junped
must be vacant, for a time anyway. f will therefore decrease below
unity as the concentration increases. In the limit of near full
occupancy of interstitial sites the process becomes the vacancy
mechanism and the correlation factor beaames that for diffusion by
vacancies on the lattice of interstitial sites.
With regard to diffusion by vacancies, as their concentration
increases vacancies additional to the one that effected a jump of
an atom will, with increasiI1<J frEqUency, be found adjacent to
that atan: this can only serve to make more randan the direction of
the atan' s next jump and so increase f y • The lowest value of fv
will occur when the vacancy concentration is low enough for t.lEre
to be negligibly few encounters of a tracer atom with rrore than
one vacancy at a time. This will usually be the case in
stroI1<Jly ionic crystals and in metals, where intrinsic vacancy
concentra tions are :;; 'VIO-3. However, rrany non-stoichianetric
catlpOunds may have vacancy concentrations very much in excess of
intrinsic and these changes in fv may be linportant.
Similar consideration apply to diffusion by interstitialcy jumps.
As the concentration of defects increases we expect suc cessive
jumps of a particular atom to be decorrelated by the prox imity of
other interstitials and fI (eqn. 14) to begin to rise above the
value for isolated defects. As interstitial sites becane
progressively filled, either interstitialcy junps begin to take on
the nature of single vacancy junps that move two atoms at a time,
or the migration process goes over into single atom junps. In
either event the limitiI1<J correlation factor is one for
diffusion by vacancies, and may be greater or less than the
original fi for isolated interstitialcy jumps.
Let us rr:JW consider the correlation factors for the diffusion of
the defects thEmSelves. This is important because in many
CORREIATION EFFECTS IN IONIC TRANSPORT PROCESSES 23
cases ionic conductivity is most sinply thought of in terms of the
migration of the defects rather than the ions.
When defects are present in low enough concentration that they
rarely encounter one another, and provided they are in a pure
lattice or sub-lattice, (i.e. only one species of atom or ion
present), they can obviously jurrp with equal probability in any of
the crystallographically allowed directions and their correla tion
factors will be unity.
If these conditions are not satisfied defect migration will
generally be non-random and the diffusion coefficient for defects
will contain a correlation factor, fda
When there are tw::> or rrore species of atom present a vacancy,
say, will generally exchange with one species of atom or ion more
readily than with the other. This makes the vacancy migration
non-random and can be shown always to give an overall correlation
factor for diffusion of vacancies less than unity. Ibwever, the
partial correlation factors associated with vacancy jumps by -
exchange with the slower moving species may be greater than unity
[9] •
Strong interactions between defects may also serve to induce
additional non-randanness into their migration and further contri
bute to changes in fd fran unity. Such interactions may induce
clustering, or may impose a minimum distance of approach for two
defects, preventing them for example from occupying adjacent sites,
or again, in extrerre cases generate an ordered or near ordered
distribution of defects. There is a good deal of evidence that such
situations often arise with the high defect concentra tions
associated with non stoichiometry in many canpounds [10]. When they
do, defect migration may be very non-random and correla tion
factors accordingly low. With perfect order, to take an extreme
case, any defect jump will put it on a ''wrong'' site and so will
be followed with high probability by a reverse jump. fd will then
be very srrall.
Naturally erough, we expect any non-randanness in the defect
migration to contribute to the non-random migration of the atoms or
ions themselves -- the tv.o must be intimately related. Any effect
that reduces fd from unity will reduce fI or fv and we might expect
the ratio fdlfv I to be roughly constant. There are one or tw::>
cases where this sort of proportionality has been found to hold,
but its general validity is by no means established. (See
later.)
These general remarks have been with regard to the romronly
discussed vacancy, inter$ti tial and intersti tialcy :rrechanisrns
of diffusion. There is evidence that other and rrore carrplex
defects
24 A. D. LE CIAIRE
may need to be invoked in discussing atom migration in many types
of non-stoichianetric carpoW1ds [10], but little study has so far
been made of their modes of migration or of the associated correla
tion effects.
4. THE CAICULATION OF CORRElATION FACI'ORS
Space forbids anything IIDre than a brief :rrention of the three
general methods that have been used to calculate correlation fac
tors. These are (a) random walk methods (b) Monte-Carlo methods and
(c) nett flux methods.
Random Walk Methods [7,8,12-15]
These start from equations like (9), (11) or (14) and calcu late
<Cosel>' <xa xal> etc. directly.
If it is <Cosel> we are dealing with then given an initial
tracer jump from site i to site k
z <Cosel > = E PI Cos (i~k k~l)
1=1 (15)
where z is the coordination nunber and PI the relative probability
that the second junp of the tracer will be from k to 1. Fbr such a
j unp to occur the vacancy, if we assume a vacancy mechanism, will
have had to migrate from i to I, so that to calculate the Pl we
need to know the relative probabilities of the vacancy occurring at
each of the z nearest neighbor sites of the tracer. To find these
we need to consider all possible sequences of junps, or tra j
ectories, the vacancy can pursue through the lattice that start at
i and finish at each of the 1. Systematic means have been devised
for doing this and are described in the references.
Alth:>ugh the random walk has been used in the very large
majority of calculations of f so far rrade, it has two :i.rrportant
limi ta tions .
(i) It must be valid to assume that the defect concentra tions are
sufficiently snall, .::;0.1% say, that all possibilities for
exchanges between a particular atom and a particular defect are
fully exploiterl before another defect approaches that atom.
This limits application of the method in its present fonn to
diffusion in pure substances and stoichiometric catl?Ounds where
only intrinsic defect concentrations prevail. It cannot be
readily
CORRELATION EFFECTS IN IONIC TRANSPORT PROCESSES 25
adapted to calculate f in crystals where defect concentrations are
high.
(ii) It must be possible to specify adequately and simply all
possible vacancy trajectories bebNeen successive junps of a tracer.
This requires that the region of crystal arcund the tracer, through
which the trajectories are bein:] considered, be c::a:tpJsed of
identi cal atoms. If different species of atoms are present the
atomic configuration around the tracer nay be changed by the
passage of a defect. The random walk method, as at presented
fo:r:mulated, has no way of siIrply taking into account such
changes.
The method is therefore restricted also to self diffusion in pure
substances and to diffusion in very dilute solutions. M::>st
estimates of f have been for such cases. It cannot be used for
concentrated alloys or solutions, unless the approximatiIlJ assunp
tions is made, as has been [9], that defects have an average migra
tion rate duriIlJ their trajectories between tracer jurrps.
These inherent limitations of the random walk method are, in
principle, absent from the M:mte-Carlo and the M9an Flux method.
Although these are rot capable of the same precision as the random
walk method, they are becaning important as interest grows in
systems to which the randan walk method is inapplicable.
Monte-Carlo M3thods [11,17,18]
These detenuine f through direct estimates of <R2> or
<x2> and the use of the general definitions of f, for
example
f = <R2> /N<r~> 1
appropriate for isotropic crystals.
(16)
The configuration of an adequately sized crystal, with its defects,
is represented digitally in a canputer array and the actual paths
of a number of atons are simulated, subject to what ever
restriction there nay be on atan jumps arising from the nature of
the defect beiIlJ studied, from interactions between the defects or
between like and unlike atans (ordering effects), and so forth.
with increasing number of individual paths rronitored and/or of
steps within each path, the mean <R2> converges to a limiting
value that is used in egn. (16) to give f.
Nett Flux M3thods [19,20]
These first deter.mine the diffusion coefficient D through a
calculation of the nett flux J in a ooncentration gradient dC/dx
--
26 A. D. LE CIAIRE
D = -J (dC/dX) -- using a rrodel sufficiently detailed to contain
the essential non-randan aspects of atomic migration. The factor f
can usually be identified in the resulting expression.
One starts with sane suitably detailed specifications of the atomic
and defect configurations that occur in the crystal. In a condition
of steady state diffusion under a concentration gradient the
distribution of configurations rEmains in dynamic equilibrium,
constant in time. But there is a net flux across any plane that can
be calculated in tenus of the rates at which atc:ms by jumping move
into and out of the various configurations. Provided the
configurations specified include all those that would be explicitly
or implicitly taken into account in a calculation of, say, the PI
in equation (15), this method yields a D that autanatically con
tains the same factor the randan walk method calculates and iden
tifies as due to correlations between successive atom jumps. The
difference between the two methods is essentially in the averaging
methods they employ, but the way in which configurations can be
specified is particularly convenient for including such effects as
ordering and high defect concentrations.
Diffusion by round vacancy-impurity pairs was first treated this
way [21], the configurations considered being the various
orientations of pairs and their distribution in the steady state.
fure recently Kikuchi and Sato [19,20] have elal::orated the
method, fonnulating it in tenus of the path probability theory of
irre versible phenomena.
5. RESULTS OF CORRELATIO'J FACIDR CAICUlATICNS
Self Diffusion in Pure SUbstances
In very many of these cases there is no more than one defect
frequency involved so that f is just a numerical factor. Sane
values for such cases are shown in Table I.
Results for many more cases of interstitialcy diffusion are
reported in reference 14. In practice, two or more ~s of
interstitialcy jump may be operative, when fI will be a function of
the ratio of their jump frequencies. In the H.C.P. case, fre
quencies for jumps within a basal plane, wA' and between planes,
WE' are assumed equal. For the general case see reference 13.
f is also known for self-diffusion by bound divacancy pairs in
ionic crystals (NaCl and CsCl) as a function of the ratio of
cation- and anion-vacancy jump frequencies [12].
CORRELATION EFFECTS IN IONIC TRANSPORI' PROCESSES 27
TABLE I
Vacancy Diffusion
Dianond Simple Cubic B.C.C. F.C.C. H.C.P.
Hexag. Net (2 Dim) Square Net (2 pim) Trian:]. Net (2 Dim)
Interstitialcy Diffusion
F.C.C. and Cation Diffusion in 'Fluorite' B. C. C. and Cation
Diffusion in 'CsCl'
(110) jumps
= 0.7812 = 0.7815
0.3333 0.4669 0.5601
fI 0.80 1.0 0.666 0.832 1.0 0.912 2/3 32/33 0.9643 0.90
Divacangy Diffusion ~v
F.C.C. [43] 0.458
Exchange, Interstitial Diff.
~
0.40 1.0 0.333 0.622 1.0 1.82 0.333 0.727 1.446 0.6
All these values of ·f are calculated by the randcm walk method and
so are appropriate only for crystals with low defect concen
trations. For crystals with high defect concentrations there have
been very few calculations, but enou:Jh to illustrate many of the
points made in Section 3.
Fig. 2 shows sane results of calculation by de Bruin and Murch
rll], usin:] the M:mte Carlo method, for diffusion by vacancies in
a simple cubic lattice. Curve 1 gives fv for tracer self-diffusion
by non-interacting vacancies and shows the rise fran fva = 0.653
(see Table I) to, ultirrately, unity. Curves (2) and (3) relate to
vacancies between which there are interactions that prohibit any
two vacancies occupying adjacent sites. Curve (3) shows fd for the
vacancy diffusion and Curve (2) shows fv for tracer diffusion. Note
that the ratio fd/fv is rOu:Jhly constant. It is of interest too
that the effects of concentration increase and of interactions do
not become very :important until beyond
28 A. D. LE CLAIRE
atout 5-10%, so the results shown in Table I IrI3.y, in some cases
anyway, be useful beyond the intrinsic range of defect concentra
tions.
de Bruin and :M..lrch have also used the M::>nte-carlo method to
calculate f for diffusion by interstitialcy j'l.lllpS at high
defect concentration, but only for a planar square lattice. f
changes fran 2/3 for isolated interstitialcy j'l.lllps to 0.467 for
vacancy diffusion (Table I) when nearly all interstitial sites are
filled.
These same authors have also calculated f for anion self dif
fusion in U0:2+x crystals by migration of the Willis 2: 2: 2
clustered defect, the most corrplex defect for which correlation
data are available. They find f = 1.587, greater, surprisinJly,
than unity. This is attributed to the unusually canplex se::J:uence
of atom dis placements involved in a Willis cluster junp, but a
convincing description of this sequence has not yet been published.
f was shawn to decrease with increasing density of defects, due to
steric effects assumed in the mcxiel calculation.
The results of both these calculations again show, surprisingly,
that the correlation factors sean to rerrain fairly constant up to
at least a 5-10% concentration of defects.
12
10
~
10 20 30 40 50 60 70 80 90 100
Fract ion of lattice - sites occupied by defects 96
Cbrrelation factors for vacancy diffusion mechanism in a simple
cubic array. [11]
CORRELATION EFFECTS IN IONIC TRANSPORT PROCESSES 29
The effects of interactions that order the atoms and defects are
vividly shown in the results of nett flux calculation by Sato and
Kikuchi [20], using the path probability method. These are reviewed
elsewhere in this volume.
Diffusion of Impurities
By this we mean diffusion of a solute present in law enough
concentrations for each atom to diffuse effectively through pure
solvent. If w2 is the jump frequency for the solute impurity atoms,
the correlation factor f2 for their diffusion turns out, in a large
number of cornrronl y occurring cases, to have the form
(17)
where u contains the frequencies for jumps only of rearest neighbor
and more distant solvent atoms. f2 is small when w2 » u because one
impurity jump is then likely to be followed by second in the
opp0site direction, and <Cos8p -+ -1. When w2 « u, f2 is near
unity -- the many exchanges a vacancy, for example, will make with
solvent atoms between two exchanges with the impurity tend to
randomise the position of the vacancy with respect to the impurity
and make <CoS81> -+ O.
Because of the local influence of the solute atans, solvent jumps
in their neighborhood will differ in frequency fran the value Wo in
pure solvent. Thus a vacancy on a 1st nearest neigh bor (n.n.)
site to a solute moves to another 1st n.n. site with frequency wI
(in f .c.c.) or may dissociate fran the impurity with frequency w3'
w4 denotes the frequency for the reverse association jumps and w5
that for vacancy jumps fran 2nd n.n. to more distant sites. All
more distant jumps are assumed of frequency wo ' as is also w5 in
many cases.
u is known in terms of these frequencies for vacancy and
interstitialcy diffusion in several cammon types of lattice [8,
22,23]. Fbr example, for f.c.c. lattices
(18)
where F is a function of w4/wo' slowly varying fran 2 at w4/wo = 00
to 7 at w4/wo = O. F(w4/wo = 1) = 5.15.
There rray be a binding energy between impurities and vacancies, at
least at 1st n.n. separations. If this is Bl it will be related to
w4 and w3 by
(19)
Solvent Diffusion in Very Dilute Solution
Because of the changed vacancy concentration (due to Bl) and
altered jump fre::ruencies near an impurity, the solvent diffusion
coefficient D (c) in a solution of concentration c will differ fran
its value D (0) in pure sol vent. At low concentrations we
anticipate a relation
D (c) = D (0) (1 + be) (20)
and this is found experimentally. Calculating the correlation
effects entails consideration of a large munber n of jump' types'
corresponding to the many configurations of vacancy, tracer (sol
vent) atan and impurity from and into which tracer jumps can occur.
Assuming the same range of influence of a solute atom as in the
expressions for f2' calculations have been made for f.c.c. [24] (n
= 13) and b.c.c. [25] (n = 15) crystals. Results are expressed in
terms of the 'enhancement factor' b. For f.c.c. crystals, for
example,
(21)
Xl and X2 are mean partial correlation factors associated with WI
and with w3 and w4 jumps respectively and are functions of w3/wl'
w2/wl, and w 4/wo.
Diffusion in Cbncentrated Solutions
The calculation of correlation factors for diffusion in con
centrated solutions is a more difficult problem because of the
large nunber of different atomic configurations and of jump types
that would need to be specified in any accurate trea-t::rrent.
Approx imate values of f have been obtained by assuming an
averag.e jump frequency for a vacancy in its trajectories between
two consecutive exchanges with a tracer [9]. The problem is then
reduced to that of impurity diffusion, with just the additional
feature of corre lation effects in the trajectory motion of the
vacancy, and can be solved with the random walk e::ruations.
The path probability nethod is particularly suited to this more
complex type of problem and more detailed though still approxi
mate calculations, incorporating ordering effects, have been made
using it [19]. There are also sone very recent calculations by the
fbnte-Carlo nethod [42] for random alloys.
6. THE MEASUREMENT AND USE OF CORRELATIQ"J FACTORS [8]
The correlation factor is just a fraction and so is not a very
important contribution to the magnitude of D itself, except
CORRELATION EFFEX::TS IN IONIC TRANSPORl' PROCESSES
in extreme cases. lbwever, it is of course always essential for a
proper analysis of diffusion measurerrents.
31
If f is independent of T, as when only one junp fra;ruency is
involved, it is just an additional factor in A in the Arrhenius
equation D = A exp (-Q/Rl'). However, when more than one fre:JUency
is involved f varies with T and partitions between the A and the Q
in a way that can be significant in theoretical discussion of their
rmgnitudes. Fran the definition of Q as Q = -RdlnD/d(l/T) , there
must always in the equation for Q be a tenn
-C = Riln.f/d (l/T) (22)
'!hen, A must contain the factor (f exp - C/Rl'), a constant if C
is constant.
Theoretical and experimental evidence indicates that C can be an
:important contribution to Q2 for :impurity diffusion, being
carparable with, although less than, the difference between Q2 and
the solvent self-diffusion Q,.
Aside fran such consequences of the variations of f with T, its
rmin importance derives fran its dependence, apart from crystal
structure, on the mechanisn by which atom jurrq;>s occur and,
When there is more than one, on the relative values of all the junp
frequencies concerned. It follows that by experimentally isolating
f we can in principle derive inforrmtion about the junp mechanism
operatin:r in a particular case and, when the llEChanism is known,
inforrmtion on the relative junp frequencies.
'!here are two methods by which a knowledge of correlation factors
can be exploited in this way.
(1) By canparin:J tracer diffusion coefficients with ionic
conductivities. '!his is the rrore pcM'erful and unambiguous
method, but applicable only to good ionic conductors.
(2) By comparin:J the diffusion coefficients of two different
isotopes of the diffusing species -- the Mass Effect method. '!his
is of general application but results are not always canpletely
unambiguous •
Ccroparison of Diffusion and Ionic Conductivity. We can measure for
each sublattice of an ionic crystal the tracer self diffusion
coefficient Dr of the ions and their contribution cr to the total
conductivity. Dr is
1 2 DT = 6" r T r T fT (23)
32 A. D. LE CIAIRE
fT is the number of j unps of a tracer per ooi t time and rT the
dis plaCEment. The conductivity in a pure ionic crystal is due to
the net drift of the lattice defects in the electric field -- each
defect behaves as an effective carrier of charge equal to the ionic
charge q, am
a = nd q ~ (24)
nd is the density of charge carriers and \1 their nobility. We can
also write
1 2 D = -6 f r f (25) q q q q
for the diffusion coefficient Dq of the charge carriers. fq is the
nunber of junps per ooit time of a carrier (defect) and r the
charge displacEment at a jump. fq is the correlation fador for
defect migration.
we now make use of the Nernst-Einstein equation relating Dq and )J
for any type of charge carrier
Dq = kT )J/q
a
r (~)2 r
q
(26)
(27)
The large maJority of practical applications of this equation have
been to D and a in sub-lattices of one atomic species with
intrinsic defect concentrations. In such cases fq = 1 and equa
tion (27) contains only fT' Now consider same special cases.
When the defects are vacancies rT = rq and Nr T = nd r q where N =
number of ions on the sub-lattice per unit volume. Equation (27)
then gives
where
(28)
(29 )
If the defects are intersti tials moving by simple interstitial j
unps, then clearly D'J""Da = 1.
CORRELATION EFFECTS IN IONIC TRANSPORT PROCESSES 33
If the interstitials nove by interstitialcy jumps, two ions move
for each defect junp and NT T - 2 r gl'la • If they are co-linear
junps charge q is displaced rq = 2rT wliile a tracer moves rT.
Then
I D /D = - f (30) T a 2 I
If, hCMeVer, the jumps are non-colinear, rT/rq depends upon their
geanetry. For the possibilities in 1lg Br for example, rTirg = 1378
arrl 13/4. The last column of 'Iable I lists the resultmg values of
Dr/Da (= HR), along with values for other cases of inter sti
tialcy j unps.
The ratio DrVDa is sometimes called 'the correlation factor' in the
present context, but clearly it contains nore than f and this
tenninology is misleading. It derives fran eqn. (28) having been
the first relation to be derived and discussed on this topic, by
Haven [26]. Accordingly the tenn "Haven Ratio", HR, has been
suggested for it.
When there are two or more types of defect present, and mechanisms
for their migration, these contribute additively to Dr, Dq and (J
and the above equations are easily generalized. In all cases, for a
given mechanisn or combination of mechanisms in specified
proportions, there is a unique theoretical value of HR. Thus,
experimental measurements of this ratio are of considerable
diagnostic value in establishing or identifying the operative
mechanism in any particular case. It is to be ranembered that
neutral defects (e.g. anion-cation vacancy pairs) contribute to Dr
but not to a, so HR is larger than in their absence. Similarly, any
electronic conductivity can make HR anomalously small.
fq will generally differ fran unity when there are interactions
between defects or when there is more than one species of ion on
the sub-lattice. In these cases equation (27) is of less general
Vcilue because comparatively little is so far knCMn about fq
•
We have havever quoted a few studies of the way f is changed fran
unity by defect interactions. These of course ~e fT as well, and it
a~s that ~n the ratio fr/fq at least the dras~c effects of defect
lllteract~Ons and order~ng are to some apprec~able extent canceled
out.
Ebr exan:ple, a cancellation is evident in de Bruin and Murch's
calculations (Fig. 2), fran which fylfq '" 0.68. This is close to
the limiting value fv = 0.653 for free "'Vacancies. Similarly, in
Sate arrl Kikuchi's S-Alumina calculations the influence of
ordering on fT arrl on fq is very rruch less evident in their
ratio.
These indications suggest that equation (35) with fg = I may be
still valuable as a help in identifying migration mecfianisms
34 A. D. LE CLAIRE
even when infonnation is lacking on the possible existence and
details of defect interactions, ordering effects etc. in the system
being sttrlied. However, this is a very tentative conclu sion
based on sparse evidence and further study is needed on the extent
of its validity.
Equation (27) is appropriate for pure sub-lattices. When there are
two or more ionic species present, one works with the ratio of Dr
to the mobility \l of each ion. Thus for ions A of charge qA
diffusing by the vacancy mechanism e.g.
(31)
where G contains tracer cnrrelation effects and the effects of non
random defect migration. It is the analogue of frr/f and equals fV
when all ions are identical. Manning [7] has d.ev~lOped expres
S1-ons for G, for the case that A is a dilute i.rrg?urity, in tenus
of the frequencies wO' WI, w3 etc. already defined.
We now mention a few examples of the use of eqn. (27).
The classical illustration of its value is Friauf's analysis of HR
for hJ migration in AgBr [27J. HR ranges from 0.5 at low to 0.65 at
high temperatures. The defects are known to be Frenkel pairs, so a
temperature variation is expected, but HR is incnnsis tent with
any combination of vacancy and interstitial mobility if the
interstitials move only by simple interstitial jurt"ps: HR v..ould
then have to be between O. 78 and 1. 'Ib explain the lower HR one
must conclude there is appreciable migration by intersti tialcy
jumps (see Table I). Detailed analysis showed that HR could be
satisfactorily accounted for only on the assunption of both
colinear and non-cnlinear jumps and the results gave the relative
jump frequencies of these. No other type of experiment could so
readily provide such detailed information on the nature of jump
mechanisms .
Another example of the method is to de:monstra te the suppres sion
of interstitials and augmenting of vacancies in hJBr and Agel when
doped with CdBr2 i this changes HR towards the vacancy value of fv
= 0.78 [28].
In the alkali and Tl-halides, SChottky defects prevail but one
always finds HR > fv' This has been shown to be due to an
appreciable cnntribution to Dr, but not of course to (J, from bound
vacancy pairs [29 ,30J .
An example of current interest is diffusion and conductivity in the
S-Aluninas. For Na and for hJ in single crystal of Na. S
CORRELATION EFFECl'S IN IONIC TRANSPORT PROCESSES 35
Al203 and hJ 13 Al203 respectively, with about 15% excess cation
over stoichianetry, HR has been reported as '" 0.61. This is very
close to the value of 0.6 for interstitialcy diffusion in a two
dimensional trianJUlar lattice and provides evidence for this type
of migration in the high rrobili ty planes of the 13 structure
[31].
In the 13" structure, Na sites lie on a hexagonal lattice with
packing such that interstitialcy diffusion seans unlikely. Fbr
vacancy diffusion HR = 0.33, but this will be a lower limit in
practice because of the high vacancy concentration due to 13" being
always cation deficient.
~asurements of HR have been made for Na migration in poly
crystalline 13" and mixed phase 13 13" sanples [32,33]. At
sufficiently high temperatures (> 'V300°C) all sanples shCM
camon values of HR that are indeed a little above 0.3 and slowly
increasing with tem perature, reaching 0.4 at 'V600°C. Such a
variation is consistent with there being a Na-Na nearest neighbor
interaction energy E of 'V - 0.04 eV, as roughly estimated fran the
Sato and Kikuchi theory. At temperatures below ~ 300°C HR falls
below 0.3 and varies con siderably fran sample to sanple. The
reasons for this are at present not understood.
The Mass Effect or Isotope Effect ~thod. Consider for simpli city
cases of diffusion entailing a unique jump frequency w for the
diffusing species. Fbr two isotopes of this species having dif
ferent masses rna, m13 , junp frequencies will be slightly
different -- wet and wl3 -- and therefore also the correlation
factors fa and f 13 • The ratio of their diffusion coefficients is
then, fran eg.n. (6) for example,
(32)
The f's have the fom of impurity correlation factors, even for
self-diffusion, because the tracer is always an I :impurity I with
jump rate different fran that of the host lattice atans. Thus fran
eg.n. (17)
fa = u/(2wa + u) and fl3 = u/(2w13 + u) (33)
where u is the same in both because it contains only non-tracer
jump frequencies. El:i.mina ting u and one of the f I S between
(32) and (33) gives
(34)
The distinction between fa and f 13 is 'fJ.ON dropped because their
difference is insignificant in this equation. We can 'fJ.ON forget
aJ::x:mt the mass dependence of f and concentrate attention on its
dependence on mechanisms and jump frequency ratios.
36 A. D. LE CIAIRE
Classical statistical mechanical calculations of w lead to the
result [13,35] that
(wa/wS) - 1 = lIK( (mS /ma) 1/2 - 1) (35)
When a migrating atom jurrps, any coupling with the lattice may
induce rrotion also in neighboring lattice atoms, and the lIK fac
tor takes acoount of this. It is the fraction of the total kinetic
energy, associatect with the whole rrotion at the saddle point of a
jump, that actually resides in the migrating atom. Elementary
treatments have often assumed lIK = 1, (i. e. wa /wS = (mS/mct)
1/2) but there is no a priori reason why a jurrp should be such an
effectively one-atom process with no concomitant lattice atom
rrotion.
From (34) and (35) we obtain
(36a)
n - 1/2 mS + (n-l) m
( rna + (n-l) m)
- 1
for processes where n atoms migrate simultaneously, m bein:;r the
average mass of non-tracer atoms. (e.g. n = 2 for interstitialcy
jurrps). E(n) is 'the mass effect for an n atom jurrp process' and
can be calculated for any nonce Da/DS is known.
Equation slinilar to (36a) and (36b) can be derived for diffu sion
where there are several jurrp frequencies or mechanisms and/or
where f is not of the slinple form of eqn. (17) (e.g. Ref.
36).
MeasurEments of DO. /DS are best, and usually, made by diffusing
two radioactive isotopes slinultaneously into the host crystal from
a very thin layer of them deposited on its surface [34]. After
diffusion, the sarrple is sliced and the activities co. and cS of
the t:IM::> isotopes determined in each slice. A plot of ln
ca/cS vs ln co. can then be shown to give a straight line with
slope (00. /DS - J,.). Distance and tline measurements are not
required, as in nonnal diffusion experlinents, and for this reason
E (n) can be determined with sane precision, to within a few
%.
We now oonsider the uses to which the basic equations (36a) and
(36b) can be put.
CORREIATION EFF".EX:TS IN IONIC TRANSPORT PROCESSES 37
Obviously E (n) increases with n, but, since both f and ilK are
fractional, only those E(n) '::1 are pe.nnissible, i.e. there is an
upper limit no on n and this imnediately allows certain mechanisms
to be rejected as incanpatible with the measurements. The values
both of f am of ilK must then lie in the range E (no) to 1 and
refer only to mechanisns with n.:>l1o. Unlike the method of the
last sec tion, the mass effect method cannot give unique values of
f, to aid in unambiguous identification of the mechanism, because
theore tical values of ilK are not yet sufficiently well
established. Nevertheless, it can be used to indicate the
possibility of, or to confi:on, a mechanism in many cases. When
this is well estab lished, so that f is known, it gives a value
for ilK, a quantity oontaining significant infonnation on the
dynamics of jurp pro cesses not easily obtained on other ways. W9
give a few examples.
For simple interstitial diffusion in many systems (e.g. C in a-Fe,
Li in W, He in Si02 etc.), E(n ~ 2) > 1 arrl within experi
mental error E (1) = 1. Thus, f = 1, as expected of interstitial
diffusion am, more interestingly, ilK = 1. For Na in Soda glass
E(2) < 1, indicating the possibility of interstitialcy diffusion
[37] •
For self-diffusion in close packed structures of metals and ionic
substances of the NaCl type, where there is substantial evi dence
for a vacancy mechanism, it is found that no = 1 and E (1) < fv,
consistent with this mechanism [8]. Using the known value of fv,
ilK is found to be around O. 9 in metals, a little less, 'V(). 76,
in the few ionic substances that have been studied, (e.g. cation
migration in CoO, NiO) [38] imicating sane lattice atan partici
pation in the jumps.
In sane cases E varies significantly with terrperature, a valuable
indication that more than one mechanism of diffusion is operative
[36,39].
The other main use of mq.ss effect measurements is to obtain, by
determining f, infonnation on relative jump frequencies when f is a
fl.IDction of these. For this we need to krPw ilK. In dilute
solutions it is currently assuned that ilK, even for solute diffu
sion, is the same as that dete:onined from E for the pure solvent
self-diffusion, where f is known.
Values of f determined in this way are best made use of when
canbined with a knavledge of other quantities that depend on the
sarre jump frequencies. For exanple, in dilute f.c.c. solutions b
(eqn. (21» arrl f2 (eqn. (17) am (18» are functions of w3/Wl,
w2/wl arrl w4/wo. So also is the ratio of the .impurity D2 and the
solvent self-diffusion Do for, using eqn. (19).
38 A. D. LE CLAIRE
D2 w2f 2 exp B~kT
Do = Wo fO
W2 WI w4 = _._._. (37)
Thus if b and D2iDo are known, a measurEment of f2 allows numerical
values of these frequency ratios to be obtained.
This procedure has been applied to a few metallic systems (Zn in Ag
arrl in Cu, Fe in Cu) [40] but so far only to one Wnl.C
systEm (Co in NiO) [41]. When measurEments of f are made as a
function of temperature the quantity C (eqn. 22) can also be
determined [40]. SUch results are of central importance in an
understanding of diffusion processes in dilute solutions.
The same can be done for b.c.c. solutions, althoUJh only two
frequency ratios are involved here (because there are no WI
jl.lll'ps) and only two of the experimental quantities are needed
to determine thEm [25].
For impurity diffusion in ionic crystals a useful quantity to
consider alon] with f2 is the ratio of the impurity nobility 112 to
D2, given by equation (31) [7].
When complete data are not available useful but less quanti tive
conclusions can still be drawn.
When such procedures yield positive values of the frequency ratios
they provide confinnation of the vacancy mechanism on which the
equations being used are based. O::casionally, experimental data
are found to be quite incompatible with the equations, for positive
frequency ratios, so the vacancy mechanism alone in such cases is
not tenable. Such is the situation for diffusion of the noble
metals and of Cd in Pb and the current hypothesis is that
interstitial diffusion is responsible. Similar incompatibilities
are found for Co diffusion in a-Zr, a-Ti and y-U [25]. None have so
far been reported for ionic systems.
REFERENCES
1. J. Bardeen and C. Herrin3", In "Aton ~bvements" p. 87 A.S.M.
Cleveland (1951).
2. E. Amaldi and E. Fermi, Ric. Sci. I 13 (1936). 3. In gas e sit
is known as the "persistence of velocities effect". 4 • Extensive
treatments of correlation effects are to be found
in the followin] texts and reviews, references 5 to 8. 5. R.E.
lbv..ard and A.B. Lidiard, Reports on