The Effects of Elastic Stress on the Kinetics

of Ostwald Ripening: The Two-Particle Problem

W I L L I A M C. J O H N S O N , P.W. VOORHEES, and D.E. ZUPON

We show that the kinetics of the classical two-particle Ostwald ripening problem can be altered
radically by elastic stress induced by misfit (coherency) strains. The growth rates of the second-
phase particles in the presence of elastic stress are shown to differ not only in magnitude but
also in sign from the classical zero-stress treatment. For example, the larger particle does not
always grow at the expense of the smaller particle, but there are large ranges of the thermo-
physical parameters in which inverse ripening occurs; i.e., the smaller particle grows at the
expense of the larger particle. Elastic stresses also may induce significant morphological changes
in the particles, as well as far stronger spatial correlations between the coarsening particles than
in the classical zero-stress limit. The analysis is cast in terms of an initial value problem, al-
lowing the growth rates of the particles to be determined as a function of readily obtainable
materials parameters. We find that the influence of misfit strains on particle growth rates and
regimes of inverse coarsening is significantly greater than that predicted from elastic energy
analyses. We expect many of the qualitative features of the ripening behavior discussed herein
to be observed in the multibody problem found in nature.

I. INTRODUCTION
T H E kinetics of Ostwald ripening processes often are
described by relationships between an average length
scale of the mixture and a temporal power law with a
positive exponent. These scaling laws can be derived from
an assumption of self-similarity of the microstructure with
time l~l or from a kinetic equation that describes the growth
rate of a second-phase particle with respect to another, t21
For example, Lifshitz and Slyozov (LS) 12'31use a kinetic
equation appropriate for an infinitely dilute array of
spherical particles in a stress-free matrix to predict that
the average particle radius should increase as t ~/3 where
t is time. The LS theory assumes that the mechanism
responsible for the transformation process is the diffu-
sion of mass from regions of high interfacial curvature
to regions of low interfacial curvature. Such a morpho-
logical evolution process is consistent with a dimunition
of the total interfacial area (and, hence, total interfacial
energy with time) and is called Ostwald ripening.
In many systems, however, the ripening process may
not be driven entirely by the interfacial curvature as in
the LS case; and, as a consequence, both the temporal
evolution of the system and the magnitude of the ex-
ponent of the classical power law, which describes the
time dependence of the average particle radius, can be
different. An example is provided by the hydrodynamic
motion accompanying spinodal decomposition in fluids.
In this case, during the late stages of spinodal decom-
position, the curvature of the interfaces between the two
phases induces fluid flow through the dependence of the
pressure at the interface on the curvature of the interface.
WILLIAM C. JOHNSON, Associate Professor, is with the De-
partment of Metallurgical Engineering and Materials Science, Carnegie
Mellon University, Pittsburgh, PA 15213-3890. P.W. VOORHEES,
Associate Professor, is with the Department of Materials Science and
Engineering, Northwestern University, Evanston, IL 60208. D.E.
ZUPON, Metallurgist, is with the Boeing Military Aircraft Company,
Wichita, KS 67277-7730.
Manuscript submitted July 26, 1988.
This flow results in the size scale of the mixture increas-
ing as t instead of the usual t 1/3 law. 14,51 As the mecha-
nism responsible for this flow is generic to the nature of
a fluid, one would not expect to observe the classical
temporal power laws for late-stage processes in these two-
phase fluids.
In this work, we examine the influence of another
phenomenon which is generic to a large class of two-
phase systems, namely, elastic stress in two-phase
solids, on particle growth rates during ripening. As
did LS, we assume that the dynamical nature of the
system is governed by isothermal mass flow through the
matrix phase. However, unlike LS, we relax the as-
sumption of a vanishingly small volume fraction of the
second phase and allow for the presence of stress gen-
erated by misfit strains. To identify the potential mag-
nitude of the elastic effects, we consider the problem of
determining the growth rates of two elastically misfitting
and diffusionally interacting spherical particles with dif-
ferent radii. This simplification allows us to investigate
the fundamental effects of an elastic field on a late-stage
transformation process without the complications of the
multibody problem found in nature. We feel that the rip-
ening characteristics discussed herein for two particles
also must be present in systems containing large ensem-
bles of particles.
II. FORMULATION
A. Physical Assumptions
We consider a two-phase binary alloy system con-
sisting of two coherent spherical particles (/3 phase) in
an infinite matrix (a phase), as depicted in Figure 1.
Both chemical components are assumed to be substitu-
tional. The system is taken to be isotropic in all respects.
The particles are of radii r_ and r+, respectively, and
have an intercenter distance of d directed along the X3
axis. The particle elastic constants, c~k~, may be different
from those of the matrix phase, c~kt, although later we

METALLURGICAL TRANSACTIONS A VOLUME 20A, JULY 1989-- 1175


R+
H--H.+
X3
D
---~Xl
B ..
Fig. 1 - Schematic diagram of the two-particle system being consid-
ered. The particles lie along the X3 direction, and the origin is that
of the bispherical coordinate system.
assume that the Poisson's ratios of each phase are equal.
Far from the particles, the solute concentration in the
matrix is taken to be a constant, and the displacement
field is assumed to approach zero.
Elastic stresses are engendered only by misfit strains,
owing to the difference in molar volumes between the
particle and matrix phases. The partial molar volumes of
the solute and solvent are assumed to be equal so that
no stresses are generated as a result of concentration gra-
dients in the system, lnterfacial stresses, arising from the
excess forces associated with a coherent interface, are
assumed to be negligible. The particle morphology and
system isotropy imply that the system is axisymmetric
about the X3 axis.
If the diffusional or elastic interaction between parti-
1176--VOLUME 20A, JULY 1989
cles is sufficiently large, the morphology of the particles
may begin to depart from sphericity. This means that the
temporal evolution of two strongly interacting particles
is a two-body free-boundary problem, wherein both the
diffusion and elastic fields around two arbitrarily shaped
particles must be determined self-consistently. As this
problem would involve intensive numerical calculations,
we phrase the problem in terms of an initial value prob-
lem; we are interested in the growth rates of the particles
at t = 0 when the particles are spherical.
B. F i e l d Equations a n d Boundary Conditions
1. Elasticity
In the following, we assume that elastic equilibrium
always is maintained. The unstressed matrix phase is used
as the reference state for the measurement of deforma-
tion. The strains, e o, are linearly related to the stresses,
t,~, through H o o k e ' s law:
t~ tz ct
t i/ = cuktekl [ 1]
t~ = Ct~ 13
ijkl(ekt T
ekl) [21
_ _
where e ur is the stress-free transformation strain and re-
peated indices are summed from 1 to 3. For the isotropic
system being considered,
eT
0 = e6 u [3]
where 6 u is the Kronecker delta, and the particle misfit,
e, is a measure of the difference in molar volumes of
the phases, e = ( V ~ - V~,)/3V~,i.
Elastic equilibrium in each phase requires that
tuj = 0 [41
where the c o m m a denotes differentiation with respect to
xj. In addition, the coherency condition requires the con-
tinuity of both displacement and traction across the
particle-matrix interface:
t~n; + t~n~ = 0 [5]
u~ = u~ [61
where fi is the unit normal pointing away from the des-
ignated phase and u is the displacement. Finally, the dis-
placement far from the particles must approach zero.
2. Diffusion
The local flux in the system, j, is expressed with re-
spect to a coordinate system in the frame of the reference
state (unstressed matrix), trj When the solute and solvent
possess identical partial molar volumes, the diffusion field
in the matrix is given by a solution to
Oc
-- = Dc V2c [7]
Ot
where Dc is the diffusion coefficient assumed indepen-
dent of composition, and c is the solute composition
measured in mole fraction.
The boundary conditions for the diffusion problem are
predicated upon the assumption of local equilibrium at
a spherical and coherent particle-matrix interfaceY ,8] The
interfacial concentrations, c ~ and c ~, can be expressed
in terms of the unstressed equilibrium compositions,
METALLURGICAL TRANSACTIONS A
c~ and c~, i.e., those compositions that would be ob- All variables with the dimensions of length are scaled
tained from the (unstressed) phase diagram, a s [91 by a characteristic length, q, which is related to the total
volume of the particles by
[g(x) + 2cr/r] . "n 1 / 3
c~(x) = c7 + [81
x~ - c7)
q = I_ 8~ [171
[g(x) + 2cr/rl
c~ = c~ + [9] where v_ and v+ are the volumes of the respective par-
x~ ~, - c7) ticles. The relative size of the particles is expressed in
g(x) is a function of the elastic field at a point x on the terms of a parameter, R, as:
particle-matrix interface, R = (R+ - R _ ) / ( R + + R _ ) [181
0 ~ 1 (e~ r ~ 1 When both particles are of the same size, R_ = R+ = 1
g(x) = ( e i j - e ~ 1 7 6 + 2 - e~j)t o - 2 tijeij
~ ~ [10]
and R = 0. The choice of q as the relevant length scale
gives rise to a relative capillary length, Lc:
X is related to the second derivative of the free energy
with respect to composition in the absence of stress, 2~r
Lc - [19l
p0k0
[_ci) (01n7t ] x(<~ - c i~) q
X - ci(1 1 + \O--~nc/c, j [1 11 Finally, a dimensionless time, r, is defined as
P0 is the density of lattice sites in the reference state, r = tDc/q 2 [201
k is the Boltzmann's constant, and y is the activity coef-
In terms of these length scales and dimensionless vari-
ficient of the solute in solution. The final boundary con-
ables, Eqs. [1] through [6] which describe the elastic
dition is that far from the particles, the composition
field in the system become:
approaches a constant, c=.
a ~- a ot
3. G r o w t h rates Tij CijklEkl [21]
The local normal interfacial velocity in the reference T~ = C~u(E~, - 6u) [22]
state, v,, is obtained from a flux conservation condition.
Assuming that the diffusion coefficient in the particle is rij,j = 0 [231
much smaller than that in the matrix, the flux conser- T~n7 + T ~ n ~ = 0 [24]
vation equation becomes
U7 = U~ [251
(c t3 - c~)v~ = - D c V c " ft ~ [12]
where Cokt are dimensionless elastic constants of the form
The isotropic growth rate of a particle, d r / d t , which is
related to the rate of change of the volume of a particle,
•2Cijkt
is obtained from the surface average of v,. If we further Cijkl X(C~i -- C7)2 [26]
assume that X ~ = X ~ then subtracting Eq. [8] from Eq.
[9] gives c o - c" = c~ - c7. Thus, the isotropic growth The nondimensional displacement, U, also is required to
rate is approach zero far from the two particles.
The diffusion equation becomes
dr -___D2 ( ( V c . fi~ d s [131 OC
dt 4"a'r2(c~i - c7) JJs --=
Or
V2C [27]

C. N o n d i m e n s i o n a l i z a t i o n subject to the following boundary condition at each point

The characteristic length and time scales allow for the along the particle-matrix interface,
following nondimensionalization. (Nondimensional and Ca(X) = G(X) + L c / R - Ps [281
dimensional quantities are denoted by upper and lower
where
case letters, respectively.) First, a scaled concentration,
C, is defined as 1 1
G(X) = ( & , _ Eo)r,j
0 ~ + 2 r~j(Eo
0 t3_60 )_ ~ r~E~ [29]
C = (c - c ~ ) / ( c ~ - c~) [14]
where c~ is the equilibrium matrix concentration of and
a system consisting of an isolated coherent particle in Ps = (c~ - c 7 ) / ( c ~ - c7) [301
an infinite matrix; c,? sometimes is referred to as the
coherent solvus as contrasted with the unstressed P, is the nondimensional shift in the coherent solvus from
solvus, c~. the incoherent solvus. When R I D < < 1 for both parti-
We define a normalized strain, E 0, and a nondimen- cles, G(X) ~ Ps and the nondimensional interfacial
sional stress, T 0, as composition is determined by the capillary term L J R . If
there is no misfit strain, G(X) = P~ = 0; the particles
Eij = eij/E [15] interact only diffusionally, and the interfacial equilib-
rium concentrations again are established by capillarity.
etij
We refer to this no-misfit-strain limit as the limit of
T~ X(c~ - cT) 2 [16]
capillarity-induced diffusion.

METALLURGICAL TRANSACTIONS A VOLUME 20A, JULY 1989-- 1177

 The dimensionless local and isotropic particle growth D = [R 2_ + A2] I/2 + [R2+ + A2] 1/2 [36]
rates are
The interfaces of the particles of radius, R+ and R_, are
V,, = - V C . h ~ [31] located at/z = /zo and/Z = /z0, respectively (Figure 1).
The ~7coordinate locates points on the surface of spheres

2---1 f vC Od s of constant /Z and can assume the values 0 -< 7/ --< 7r.
dR-R-d~- 47rR 132] The bispherical coordinate system and its relationship
to the Cartesian coordinate system is discussed, for
example, in Reference 10.
Using the orthogonality condition for Legendre poly-
III. SOLUTION
nomials and the boundary condition, Eq. [28[, the coef-
A. Elastic and Composition Fields ficients An and B. are
Since we assume that a gradient in the composition {Q+ sinh [(n + l/2)/zo] - Q; sinh [(n + 1/2)/zo 1}
field does not engender stress, the elastic fields can be A n =
sink [(n + 1/2) (/Zo -/.to )1
determined from'Eqs. [21 ] through [26] once the relative
particle sizes and intercenter distance are established. [37]
However, as the particle and matrix possess different
elastic constants, a straightforward superposition of the {Q; cosh [(n + 1/2)/zo] - Q.+ cosh [(n + 1/2)/zo]}
B n ----
elastic fields of the individual particles is not possible. sink [(n + 1/2) (/zo -/Z~)]
Here, we employ an approximation that is valid only for
small differences in the elastic shear moduli, c44, be- [38]
tween particle and matrix, ](c~ - c~4)/c~41 < < 1 and in where
the limit for which the interparticle separation is not
too small (see Appendix). Although the approximation Q+ = [(2n + 1)Tr]1/2
has a limited range of validity, it provides analytic 7r ot ~
expressions for the strain fields and reproduces the f [C (/-to ,T/) - C~]Pn(cos r/) sm 7/dr/
general trends for the elastic interaction energy between x J0 [c--osh/z-~ --- co--s ~/]---1/5 [ 39 ]
the particles. In the limit ]6 - 11 < < 1, the coherent
solvus is Q ; = [(2n + 1)77"]1/2

6e 2 C~4t~(1 + t') (= [C~(/Zo 07) - CoolP.(cos 7) sin n dn

c~ = c7 + [331 X [401
X(C~ - c'~) [2(1 - 2u) + 6(1 + ~,)] J0 G N/zo - c 0-2
where ~, is Poisson's ratio, assumed to be equal in each B. Growth Rates
of the phases, and 6 is the ratio of the shear moduli of
the particle to that of the matrix: To calculate the local normal growth rates, we note
that
6 = c44/c44 [34]
V + = VC.t/~'= V C . d . (/z=/zo) [41]
We assume that the matrix concentration field can be
described using the quasistationary approximation. V; = V C - t / ~ = - V C . G (/z=/zo) [42]
Therefore, the time derivative of the concentration is set where
equal to zero in Eq. [27]. This approximation is quite
reasonable when the supersaturation is small, as is the [ c o s h / Z - cos T~]3/2 ~
case during coarsening. Laplace's equation for the two- VC- d~ = (47r),/-~-A E (2n + 1)I/2pn(COS7)
n=0
particle system can be solved in series form using the
bispherical coordinate system. The concentration field
for a system axisymmetric about the X3 axis is, in terms X l a n [ s i n k [(n -~- 1/2)/-L]
of the bispherical coordinates, /z and ~7,
+ sinh/z cosh[(n + 1/2)/Z]]

C = C~ + (cosh/Z - cos ~7)t/2 2(cosh/Z - cos 7) d

n--0
+ Bn[cosh [(n + l/2)/z]
{A. cosh [(n + 1/2)/Z] + B. sink [(n + 1/2)/Z]}
(2n + 1 ~ '/2 + s i n k . sink [(n_ + 1 / 2 ) " ] ] / [43]
x \ 47r / P.(cos 77) [35] 2(cosh/z cos r/) JJ

A. and B. are coefficients to be determined, and P. Substituting Eq. [43] into Eqs. [41] and [42], using the
is the Legendre polynomial. In our case, the coordinate result in Eq. [32] and integrating gives
/z locates spheres of radius A csch I/z] with centers at the -A
points ( 0 , 0 , A coth /Z) where A is the dimensionless R+ - v~~"/~-R
Tr~" E (2n + 1) 1/2 (A. + B.) [44]
interfocal distance given by n=O

1178-- VOLUME 20A, JULY 1989 METALLURGICAL TRANSACTIONS A


_ A 1)1/2
2 (2n +
t~=0
In the absence of misfit strains, the interracial con-
centrations, C~ and C"_, are constants and are given by
the classical Gibbs-Thomson equation. The correspond-
ing coefficients A~ = A~ and B~ = B~ in a form required
for Eqs. [441 and [45] become tll'~2j
[ 87r
A~+B~= (2n+ 1)J
A~-B,~= (2s
C. Self-Consistent M e a n - F i e l d Composition
An arbitrary choice of the mean-field composition
usually results in a net flow of mass into or out of the
two-particle cluster. However, in certain situations, it is
appropriate to consider a system in which all of the mass
leaving one particle deposits on the other particle. This
is accomplished by choosing a mean-field concentration
such that the growth of one particle is offset by the dis-
solution of the other, i.e., choosing the mean-field con-
2 "
centration which guarantees that R2_R_ + R+R+ = O.
Using Eqs. [44] and [45], the value of the mean-field
concentration, C ~ , which results in no change in the total
volume of the two-particle cluster, is
S1
C~ - - -
where
SI = L (2n + 1)'/2{I + sinh [(n + 1/2)/Xo] - I,, sinh [(n + l/2)/x~]}
n-0
,2 {1 n=0
+
cosh [(n + 1/2) (/*o - /Xo)] - cosh [(._n_+ ! / 2 ) (/*o + /~)]'~
IV. RESULTS
A . lsotropic G r o w t h R a t e s
Due to the complex dependence of the interfacial con-
centrations on r/, the coefficients of the eigenfunction
expansion for C must be determined numerically. This
requires choosing explicit values for the dimensionless
particle and matrix elastic constants. As an example, we
consider a prototype M-base binary alloy, the elastic
constants of which are nearly isotropic. For polycrys-
talline A1,1~3Jc~4 = 2.65 x 10 ~ e r g s / c m 3, v = 0.347, and
190 = 6 x 1022/cm 3. Poisson's ratio for the particle and
METALLURGICAL TRANSACTIONS A
matrix are assumed to be equal. To estimate the mag-
nitude of the nondimensional elastic constants, we as-
(B,, - A,) [45]
sume that the alloy behaves as an ideal solution in the
absence of stress and that the particle misfit is small,
e = 0.08 pct. (The misfit is representative of that of co-
herent 6' particles in the A1-Li system.) The nondimen-
sional elastic constants employed are C~4 = 7.7 x 10 -4
and v = 0.347 at a temperature of 500 K. Clearly, in-
] ~/2 e (n+l/2)txd [ ( C ~_ _ C.)e(n+l/2)~*6 _ ( C + - C~)e -(n~|/2)~]
sinh [(n + 1/2) (/,~ - /*~)] [46]
e ~"+1/2)~" [(C+ - C ~ ) e -(n+l/2)gg - ( C ~- - C ~ ) e (n+l/2)m+]
1)1 sinh [(n + 1/2) (/,ff - /*0-)1 [47]
creasing the magnitude of the misfit, changing the tem-
perature, or assuming different solid solution behavior
for the alloy may change significantly the magnitude of
the nondimensional elastic constants and the contribution
of the elastic field to the interfacial concentrations. Such
changes in the materials parameters alter the magnitude
of the growth rates and the particle sizes at which the
elastic effects become important but not the qualitative
predictions of the model.
In the limit as D --+ ~, LS showed that the particles
would grow or shrink via the exchange o f mass with a
mean-field concentration at infinity. In certain cases,
it is instructive to compare the growth behavior of the
two-particle cluster with respect to such a mean-field
concentration. For ease of comparison with ripening
[48] theories in unstressed systems, we express the mean-
field concentration in terms of a mean-field particle
radius R=. Assuming an isolated particle of radius R=,
[491
sinh [(n + 1/2) (/*o - / * ~ ) ]
+ ?/S) (--L-o - / [50]
f0~' C~(/Xo,r/)P,( cos 7) (2n + 1) '/2 sin a~ d'o
[51]
[cosh tz0 - cos r/]
N/~ ('~ C~(/,o,r/)P,,(cos r/) (2n + 1) */z sin T/dT/
[52]
J0 [cosh/*~ - cos 7/] 1/2
the relationship between the mean-field radius and mean-
field composition is
C~ = L , / R ~ [53]
This equation follows from Eq. [28] by setting C ~ = C~
and realizing that for an isolated particle, G(X) = P,.
The above definition allows the coarsening behavior of
the two particles to be expressed in terms of their radii
and the mean-field radius. In the classical LS limit, i . e . ,
no diffusional or elastic interactions, all particles smaller
than the mean-field radius shrink, and all particles larger
than the mean-field radius grow. As will be seen, both
diffusional and elastic interactions can alter this behavior.
V O L U M E 2 0 A , J U L Y 1 9 8 9 - - 1179
 1. The effects of interparticle separation 4/--
1.5 --
on particle growth rates
To understand the influence of elastic interactions on
particle growth rates, we first examine the classical no-
misfit-strain limit of capillarity-controlled ripening for 1.0
the case in which the particles are of the same size
(R+ = R = 1 and R = 0). The growth rates for this case
are depicted in Figure 2 as a function of the intercenter o.5
distance, D, for five different mean-field compositions.
If R~ = 1, the mean-field concentration and interfacial
concentrations are equal, and there is no exchange of
mass; the growth rates at all distances of separation are
,re"
zero and C ~ = Lc (R= = 1) is the self-consistent mean-
field composition. If R= < 1, the particles in the cluster
grow with a rate that falls off by about a factor of two -0.5
as the interparticle separation is decreased from infinity.
The decrease in growth rate is attributable to the shield-
ing of the diffusion field of one particle by the other -I.0 / Lc= 0
particle. The opposite situation is present when the mean- 8= I,I R =0
field radius is greater than one. Cao= 0
In the limit Lc = 0 and C~ = 0, the particle growth -1.5 -

rates are controlled entirely by the elastic interaction as , ,ll , I I Ill I I I 1414"~.5'QO
2.0 2.5 5.0 3.5 4.0
shown in Figure 3. In this case, the dimensional value
of the mean-field composition is equal to that of the co- Intercenter Distance, D
herent solvus, i.e., the matrix composition when it is in Fig. 3 - - N o n d i m e n s i o n a l particle growth rates are plotted as a func-
equilibrium with an isolated large coherent particle. tion of interparticle separation in the limit of very large particles where
Consequently, the particle growth rates are zero at large there is no contribution of the capillary terms. Thus, particles grow
interparticle distances. For smaller interparticle dis- or shrink as a result of the interparticle elastic interaction. Growth
rates are negative for elastically hard particles (6 > l) and positive
tances, the manner in which the misfit strains influence
for elastically soft particles (6 < 1).
the growth rates depends upon the degree of elastic in-
homogeneity. If the system is elastically homogeneous
(6 = 1), the growth rate of each particle is zero, inde-
pendent of the interparticle separation. In this limit, the concentration of the other, and there is no elastically in-
elastic field of one particle does not affect the interfacial duced mass flow. When the particles are softer than the
matrix (6 < 1), the particles grow, the magnitude of the
growth rate increasing as the interparticle separation de-
creases or as the degree of elastic inhomogeneity in-
creases. The particles dissolve when they are elastically
1.5
harder than the matrix (6 > 1), the magnitude of the
growth rate again increasing as the intercenter distance
RoD=0.9 decreases or the degree of inhomogeneity increases.
1.0
Clearly, C~ = 0 does not correspond to the self-consistent
;,..- mean-field composition when misfit strains are present
RoD=0.95 and 6 ~ 1.
0.5
The growth rates are symmetric in the degree of elas-
7o_ tic inhomogeneity about 6 = 1, as the elastic field is
x
r
RoD=I.O calculated only to first order in 6. Furthermore, the growth
.d o
rates depicted in Figure 3 and subsequent figures are de-
4-1 termined only to an intercenter distance of about D =
2.2, as the expressions for the elastic fields are not ac-
-0.5 L RoD=1.05
curate when the intercenter distance becomes very small.
However, it should be noted that the general trends ex-
ROD=I.I hibited by the growth rates as the intercenter distance
-I.O becomes very small are in qualitative agreement with those
R=0 of tetragonally misfitting particles in which the elastic
fields can be determined exactly. [141 (The equivalent
-I.5 I III I I I Ill I I I II 4,-- transformation strain of the particles assumes a position-
2.0 2.5 5.0 5.5 4.0 4.5 03 dependent tetragonal symmetry in the approximation used
Intercenter Distance, D here to calculate the strain fields.)
Figures 4 and 5 depict the effects of changing the mean-
Fig. 2 - - N o n d i m e n s i o n a l particle growth rates for the no-misfit-strain
limit are plotted as a function of the interparticle separation. The par-
field composition on the nondimensional particle growth
ticles are of the same size with radii of 1. The mean-field concentra- rates for two equal-sized (R = R+ -- 1) soft (6 = 0.9)
tion is determined by a mean-field radius, R,, according to Eq. [48]. and hard (6 = 1.1) particles, respectively. When D ---->0%

1180-- VOLUME 20A, JULY 1989 METALLURGICAL TRANSACTIONS A

 the growth rate of an isolated particle in a slightly super-
saturated matrix is recovered for all values of the mean-
L C =10-5 field radius and 8. For elastically soft particles that are
2.0
B= 0.9
larger than the mean-field particle (R~ < 1), the positive
growth rates initially decrease as the interparticle sepa-
R=O
ration is decreased owing to diffusional shielding of the
1.5 particles. At smaller intercenter distances, the elastic fields
begin to dominate, and the growth rate begins to in-
RoO = 0 . 9 crease. This gives a nonmonotonic dependence of the
1.0 growth rates on intercenter distance and a minimum in
0
the growth rate usually in the range 4 < D < 7. When
x
"14 RQO=0.95 the particle radii are smaller than the mean-field radius
,IX: 0.5 (R~ > 1), the particles dissolve when the intercenter dis-
tance is large. However, these growth rates change sign
Rao=l.O at smaller intercenter distances. The misfit strains do not
only influence the magnitude of the particle growth rates
but also their sign.
I f the particles are harder elastically than the matrix,
-o.5 I - ~ Roo=1.o5 m the opposite effects to those discussed for soft particles
are observed, as shown in Figure 5. Particles that are
- RoO=I.I larger than the mean-field radius (corresponding to
-I.0 R= < 1) grow at large intercenter distances. As the in-
2.0 2.5 3.0 5.5 4.0 4.5 oo tercenter distance decreases, however, the growth rates
Intercenter Distance, D become negative, and the particles begin to dissolve. If
the interacting particles are smaller initially than the mean-
Fig. 4 - - N o n d i m e n s i o n a l growth rates of equal-sized particles plotted
as a function of interparticle separation for various values of the mean- field radius (corresponding to R~ > 1), a minimum in
field concentration. The relative capillary length is Lc = 10 s, so both the magnitude of the dissolution rate for a certain inter-
the elasticity and capillarity contribute to establishing the interracial particle separation is observed.
concentrations. The elastic interactions dominate at smaller D to sta- The nonmonotonic dependence of the growth rates on
bilize the two-particle cluster, even when the mean-field radius is greater
than those of the two particles and, thus, would have negative coars-
D is a result of the diffusion field decaying as 1/r, where
ening rates in the capillarity only limits. r is the distance from the center of a particle, while the
elastic stress field of a particle decays as 1/r 3. Thus, as
the distance of separation is decreased, the effects of dif-
fusional interactions are felt before those resulting from
1.0 m

elastic interaction. At smaller interparticle separations,

Roo = 0 . 9 elasticity clearly dominates the growth process. For ex-
B
ample, when D = 2.5, 6 = 0.9, and R~ = 0.95, the
0.5
growth rates are almost three times that expected from
m
Rco= 0.95 capillarity-induced diffusion alone.

0 2. The effects of particle radius on particle

R~ = 1.0
~D
growth rates
0 The dependence of the nondimensional isotropic growth
-0.5 RQO=I.05 rates on the relative size of the particles, R, is illustrated
X
+1 in Figures 6 through 9. As all results are symmetric about
~
Rco=l.I R = 0, we refer to the system only in terms of the pos-
-1.0 itive values of R (corresponding to R+ > R_). In all cases,
the total volume of both particles at t = 0 remains fixed
as R is changed. The isotropic growth rates for elasti-
-1.5 cally interacting particles are given by the solid lines,
Lc=IO -5
while those for capillarity-controlled growth are given
by the broken lines.
8 =1.1
-2.0 Figure 6 depicts the isotropic growth rates when the
R=O
mean-field composition is chosen self-consistently. The
growth rates vanish when both particles are of the same
2.0 2.5 5.0 3.5 4.0 4.5 size (R = 0). In the absence of misfit strains and if the
particles are of different size, the larger particle (radius
Intercenter Distance, D
R+) always grows and the smaller particle (radius R )
Fig. 5 - - N o n d i m e n s i o n a l growth rates plotted as a function of inter- dissolves. The magnitude of the growth rates increases
particle separation for the same system depicted in Fig. 4, except the
as the difference in radii between the particles increases.
particles are elastically harder than the matrix. In this case, the
two-particle cluster dissolves at small D, even when their radii are This behavior is qualitatively the same for all capillary
10 pct larger than the average particle radius used to establish the lengths and interparticle separations and is consistent with
mean-field concentration. an interfacial curvature-driven ripening process.

METALLURGICAL TRANSACTIONS A VOLUME 20A, JULY 1989 I 181

 6 ~- LC = 10-6
D =2.5
- B =0,9

4
R_
2
q.D
x
o 2
+1 R+ x

R 0
\
-2 \ \
\
R_\\
0 0.2 0.4 0.6
R i
Fig. 6 - - I s o t r o p i c growth rates plotted as a ftmction of relative par-
ticle size for a self-consistent mean-field composition. The solid and
broken lines correspond to the cases of Lc = 0 and the no-misfit-strain
limit, respectively. Inverse coarsening occurs for 0 < R < 0.695.
When misfit strains are present and the particles are
elastically softer than the matrix, the smaller particle is
found to grow at the expense of the larger particle for
0 < R < 0.695. As will be discussed further, this is a
result of the elastic interaction between particles and is
a process called inverse ripening, i~s,~6} The growth rate
of the smaller particle increases rapidly reaching a max-
imum at about R = 0.45. The dissolution rate of the
large particle increases with increasing R and reaches a
maximum at about R = 0.2. When R = 0.695, the growth
rates of both particles are zero. Thus, there are two dif-
ferent self-consistent mean-field compositions for this
choice of interparticle separation and thermophysical
physical parameters. When R > 0.695, capillary effects
dominate and the classical behavior is recovered with the
large particle growing at the expense of the smaller par-
ticle, although at rates much different from the no-misfit-
strain limit.
Changes in the elastic inhomogeneity of the system,
relative capillary length, mean-field composition, inter-
particle separation, or other thermophysical parameters
alter the dynamic behavior of the system. Figure 7 dis-
plays the effect of choosing a mean-field composition
different from that of the self-consistent mean-field com-
position for a system that is in all other regards identical
to that of Figure 6. With no misfit strains, the behavior
in the two systems is very similar owing to the choice
.... R+
,% +
'
_ %
i
-2 \
L c = 10-6 \
o = z.5 ~,R_
: 0.9 \
-4
- R ~ = 1.0
0.8
I l l\
0 0.2 0.4 0.6 0.8
R
Fig. 7 - - I s o t r o p i c growth rates are plotted as a function of relative
particle size, R. The thermophysical parameters employed are iden-
tical to those of Fig. 6, except the mean-field composition has here
been set by R~. = 1.
of the mean-field composition in Figure 7. Many of the
qualitative features of the dynamics are also similar for
both cases when misfit strains are included, with the small
particle growing and the large particle dissolving over a
large range of R. However, both particles are growing
for R < 0.1, as R= = 1 is not the self-consistent mean-
field composition when R = 0. (The particle growth rates
when R = 0 are also shown in Figure 4 at D = 2.5.)
Both particles are also growing when 0.6 < R < 0.7.
Thus, for a system in which R+ = 4R_ (R = 0.6), the
nondimensional growth rates for R_ and R+ are 4.9 x
10 -6 and - 2 . 7 x 10 -8, respectively. This is in contrast
with the nondimensional growth rates for the limit of
capillarity-induced diffusion of - 6 . 9 • 10 -6 and 0.45 x
10 -6"
The influence of the elastic interaction on the i.sotropic
growth rates becomes less pronounced for fixed Lc as the
interparticle separation is increased. An illustration is given
in Figure 8 for a system possessing the same nondimen-
sional parameters as employed in Figure 7, except that
an intercenter separation of D = 3 is used instead of D =
2.5. The region for which the small particle is growing
is reduced, extending only to about R = 0.45, and the
magnitude of the growth rates is about a factor of six
less. For 0.15 < R < 0.3, the smaller particle is growing

1182--- VOLUME 20A, JULY 1989 METALLURGICAL TRANSACTIONS A

 / L c = 10-6
-- 1
D = 2.5
elasticity o n l y ~ 8=1.1
2 RoD= 1.0

to
_o o _ ____ ,,-- to
x o \
""" "'~ ""~ ~ "~ R+ -- N
+1
,n." x -2 N
+i \
,n.- \
-2
\
\
L c = 10-6 \\NN\\\\ ~
-4 \
D = &O ~-R_-~Y*-R_
-4
8 = 0.9 ~\ \ R_ \R_
Roo= 1.0

-6 t I t I L ~
0 0.2 0.4 0.6
R -8
Fig. 8--1sotropic growth rates plotted as a function of the relative
particle size. The thermophysical parameters are identical to those of
Fig. 7, except that the interparticle separation has been changed to
D = 3. The line of unequal segments is the limit of elasticity-induced
diffusion only, L, = 0.
while the larger particle is dissolving. In this case, in-
verse coarsening is present even though the radius of the
larger particle (R+ = 1.15 at R = 0.3) is greater than
the mean-field radius (R~ = 1); and, thus, in the classical
case, the particle would be growing. Both particles are
growing in the ranges 0 < R < 0.15 and 0.3 < R <
0.45. The classical ripening behavior is recovered for
larger values of R, although the rates are again very dif-
ferent from those of the classical case. Thus, capillarity-
related effects tend to dominate at large R and elastic
effects tend to dominate at small R.
The isotropic growth rates in the limit L,. = 0 are shown
in Figure 8 by the lines consisting of segments of un-
equal length and should be compared to the isotropic
growth rates in the opposite limit of capillarity-controlled
ripening illustrated with the dashed lines. In the limit
Lc = 0, particle growth rates are completely controlled
by the interparticle elastic interaction. As there is no
interfacial energy in this limit, the smaller particle grows
for all values of R with the magnitude of the nondimen-
sional growth rate increasing as the difference in size
between the particles increases. The growth rate of the
larger particle quickly becomes negative, reaching a
minimum around R = 0.3, and remains negative as R
approaches one. In this elasticity-dominated limit cor-
responding to large q, inverse coarsening is observed for
all R.
The dynamics of the system change entirely when the
I ~ t
0.8
- I I I ~ I
0 0.2 0.4 0.6 0.8
R
Fig. 9--Isotroplc growth rates are plotted as a function of the relative
particle size. The system's thermophysical parameters are identical to
those of Fig. 7 except that the particles are elastically harder than the
matrix. The magnitudes of the growth rates are significantly greater
than those in the limit of capillarity-induced diffusion, and no region
of inverse coarsening exists.
particles are elastically harder than the matrix. These ef-
fects are illustrated in Figure 9 for a system that is iden-
tical in all respects to that of Figure 7, except 6 = 1.1
rather than 6 = 0.9. When R = 0, both particles are
dissolving, consistent with the behavior shown in Figure 5
for D = 2.5. When the particles are of different size,
the smaller particle always dissolves and does so at a
significantly faster rate than in the limit of capillarity-
induced diffusion. When R < 0.05, the larger particle
also dissolves, even though it is larger than the mean-
field radius, owing to the elastic interaction. For larger
values of R, the interracial energy effects again become
important and a positive growth rate for the larger par-
ticle is recovered. However, the elastic interaction is still
significant with the growth rate of the large particle sub-
stantially higher than in the zero-misfit limit.
B. Local Normal Growth Rates
The elastic and diffusional interaction between parti-
cles also strongly affects the local normal growth rates,
V,. In the absence of misfit strains, the diffusional
interaction gives nondimensional local normal growth
rates, as calculated from Eqs. [41] through [43], that vary
from about 40 to 120 pct of the average growth rate.
The position dependence of the local normal growth rate

METALLURGICALTRANSACTIONS A VOLUME 20A, JULY 1989--1183

for the particle with radius R+ is depicted in Figure 10 1.0
for the zero-misfit limit, and a mean-field composition
chosen such that both particles are growing. The local D=3
normal growth rate divided by the relative capillary length 8 = 0.9
is plotted as a function of the angle 0 between the par- R=O
ticle normal and the X3 axis. When 0 = 0, the normal 0.8 Roo= 0.95
points directly away from the second particle, while the
normal points directly toward the second particle when
0 = ~-. The isotropic growth rate is given by the broken
line. The particle is growing faster than the average on
0.6
the side away from the other particle and slower than the
average in the direction toward the second particle. These
nonuniform growth rates m a y lead to both shape changes
and motion of the center of the mass of particles. In this
case, the particles clearly move apart, in agreement with 0 0.4
earlier predictions. [12] O f course, at large intercenter dis- X
tances, the local growth rate is uniform over the surface r

and equal to the average value. >

Figures 11 and 12 show the effects of elastic stress on
the local normal interfacial velocities for the same phys- 0.2
ical parameters used in Figure 10. The local growth rates
deviate significantly in sign, magnitude, and angular de-
pendence from the no-misfit-strain limit of Figure 10.
For Lc = 0, the local growth rates are positive and up
to two orders of magnitude greater than the average growth 0
rate for 0 < 0 < ~r/3 and 27r/3 < 0 < 7r, while the
growth rates are negative in the other directions. As the
relative capillary length is increased, significant anisot-
ropies in the local growth rates remain. Even when L(, = -0.2
10 3, in which case capillarity largely controls the mag-
nitude of the average growth rate, the local growth rate
varies by more than a factor of two from the average, 0 7r/2 77
and the functional dependence of V, on 0 is entirely dif-
ferent from that depicted in Figure 10. At large inter- Angle, 8
particle separations, the elastic interaction vanishes and Fig. 1 1 - - T h e orientation dependence of the local normal growth rate
the local normal growth rates approach the isotropic is shown for elastically soft particles. The elastic interaction gives rise
growth rate. to significant shape changes and considerably alters the symmetry of
the local normal growth rates from the no-misfit limit shown in Fig. 10.

Similar anisotropies in the local normal growth rates

,Local Growth Rate are also observed in elastically hard particles, as shown
in Figure 12. In the limit of Lc = 0, the average growth
rate is negative. Locally, the particle dissolves faster than
the average (by up to a factor of about 100) for 0 < 0 <
7r/3 and 27r/3 < 0 < 7r and grows in the other directions.
OJ "-~Av2ge Growth Rate
0
x 3 V. DISCUSSION
._1 The purpose of this work is to assess in a rigorous
manner the potential influence of elastic particle inter-
>=z actions on a particle's coarsening rate. Thus far we have
R=O
shown that the elastic interaction significantly alters the
Rco=0.95 kinetics of two-particle coarsening and allows for the
D=3 possibility of inverse coarsening. As we shall discuss
Capillarity Only further, inverse coarsening can be qualitatively under-
stood in terms of the elastic interaction energy; t~5] in some
I J I I I i I I I systems, two particles of the same size have a lower elastic
0 rr/2 7r energy than an isolated particle. The elastic interaction
Angle, O between particles may also give rise to other phenomena
that can influence microstructural evolution in two-phase
Fig. l O - - T h e orientation dependence of the local normal growth rate
systems. In Section A of the following, we examine how
is shown for the limiting case of capillarity-induced diffusion. De-
viation from the isotropic growth rate is a result of diffusional inter- the elastic and diffusional interaction between particles
actions and leads to particle shape changes and migrations. can lead to particle shape changes and migration. In

1184-- VOLUME 20A, JULY 1989 METALLURGICAL TRANSACTIONS A


0.6
Lc=IO -3
0.4
Q2
o of the particles are in the opposite direction with the par-
X ticles elongating in directions perpendicular to the axes
>= 0 of revolution. These interaction effects may either pro-
-0.2
D=:5
-0.4
B= I.I
R--O
R~o = 0 . 9 5
-0.6 ' ; ' I I I ~ I
' '
0 "0"/2
Angle, 8 translations. Similar motion of second-phase particles has
Fig. 1 2 - - T h e orientation dependence of the local normal growth rate
is shown for an elastically hard particle. All other system parameters
are identical to those used in Figs. 10 and 11.
Part B, we compare the results of the kinetic analysis
with elastic energy calculations of interacting spherical
particles and show that the energy calculations are not
always a good indicator of how the system might evolve.
Finally, in Section C, we discuss briefly how these ef-
fects might influence microstructural development in
coherent solids.
A. Particle Shape Changes and Translations
Figures 10 through 12 show that diffusional and elas-
tic interactions between the two particles have a large
effect on the orientation dependence of the local normal
growth rate. The strong dependence of V, on 0 indicates
that particle interactions may induce significant shape
changes. In the absence of stress, it is likely that the
particles do not change their shapes dramatically, as the
variations in the local normal growth rates with position
along the interface shown in Figure 10 are not as large
as those shown in Figures 1 1 and 12. These results are
in qualitative agreement with two-dimensional numerical
calculations of ripening t|7] in which the full free-boundary
problem was solved in the limit of no elasticity. The nu-
merical calculations show that even for very small par-
ticle separations, the distortions in the shape of two
initially circular particles are mild.
METALLURGICAL TRANSACTIONS A
When misfit strains are present, these shape changes
are much more pronounced than they are, for the same
interparticle separation, in the no-misfit-strain limit. For
example, when the particle is elastically softer than the
matrix, both spherical particles tend to elongate along
the axes of revolution. In fact, the orientation depen-
dence of the local normal growth rates in Figure 1 1 is
consistent with the possibility of each particle splitting
into two particles, assuming that the interfacial velocity
near 0 = rr/2 remains negative as the particle changes
its shape to an ellipsoid. This shape change is in keeping
with the experimental observations of Miyazaki and co-
workers. []81 If the particles are elastically harder than the
matrix (Figure 12), the distortions in the shape of both
mote or hinder those elastically induced shape transitions
(bifurcations) observed for isolated precipitates. 1~91
The orientation dependence of the local normal growth
rates induced by the diffusional and elastic interaction
between particles can also result in the movement of their
centers of mass. For elastically soft particles in the limit
Lc = 0, the normal growth rates are larger for the higher
values of 0, and the particles, in this case, move toward
one another. This is in contrast to the no-misfit-strain
limit wherein the interparticle separation increases for
the same imposed set of conditions. For elastically hard
particles, the normal growth rates are larger in the
low-0 directions, and the particles move apart at a higher
rate than in the absence of stress. Changing the value of
"rr
Lc strongly influences the sign and magnitude of the
been noted by Geguzin and Krivoglaz [2~ for the case of
two elastically interacting voids in a solid. An accurate
assessment of stress-induced particle migrations on ma-
terial microstructure requires a more complete analysis
and is forthcoming. 12|'221
The large position dependence of the local normal ve-
locities observed here raises serious questions as to the
validity of using point (monopole) sources or sinks of
matter located at the centers of the particles to describe
the diffusion fields surrounding two elastically interact-
ing particles. This point source approximation has been
widely employed in coarsening theories to describe the
multibody diffusional interactions which occur at non-
zero volume fractions in systems with no stress, p~3~] This
concern is substantiated by comparing isotropic growth
rates calculated from the full kinetic analysis with the
point (monopole) source approximation for a system with
6 = 0.9. In general, when capillarity largely determines
the interfacial concentrations, the point source approxi-
mation is reasonable even for relatively small intercenter
distances. For example, when D = 3, Lc = 10 4, R =
0.1, and R~ = 1, the isotropic growth rates are - 8 . 1 x
l 0 6 and 7.0 • l 0 -6 for the small and large particles,
respectively. The corresponding values for the point source
approximation are - 7 . 4 x 10 - 6 and 7.3 x 10 -6, or within
about 8 pet of the actual value. (In the absence of stress,
the isotropic and point source growth rates agree to within
2.3 pet.) Increasing the importance of the elastic effects
by decreasing the relative capillary length to Lc -- 10 -5
for the same set of thermophysical parameters yields
VOLUME 20A, JULY 1989-- 1185
growth rates of - 3 . 4 x 1 0 - 7 and 7.8 x 1 0 - 7 for the
small and large particles, respectively. The correspond-
ing growth rates for the point source approximation are
1.9 x 1 0 - 7 and 1.3 x 1 0 - 6 . Thus, not only do the mag-
nitudes of the growth rates differ by almost a factor of
two between the approaches, they differ in sign. In the
limit of Lc = 0, the point source approximation over-
estimates the growth rate of the small particle by 195 pct
and that of the large particle by 660 pct. Similar dis-
agreement can be found at larger intercenter distances,
although the relative capillarity length must be smaller.
Even for D = 10, the point source approximation over-
estimates the isotropic growth rate by 286 pct when L~ =
R = 0 and R~ = 1. Thus, irrespective of the intercenter
distance, once elastic effects become important, the point
(monopole) source approximation yields a poor descrip-
tion of the isotropic growth rates of the particles. The
failure of the point source approximation in modeling the
growth rates is likely a result of the far stronger position
dependence of the interfacial concentrations and the con-
centration gradients at the interface in the presence of
stress than in the classical interfacial energy driven
coarsening limit. Consequently, determining the particle
growth or dissolution rates on the basis of a point (mono-
pole) source approximation, when elastic interactions are
important, should be approached with extreme caution.
B. Comparison to Elastic Energy Calculations
Some of the results obtained from the kinetic analysis
can be understood qualitatively in terms of the elastic in-
teraction energy. For example, when Lc = 0 and Ca = 0,
there is a direct comparison between the results shown
in Figure 3 and Eshelby's elastic interaction energy cal-
culations of two misfitting equal-sized particles. [32] For
an elastically homogeneous and isotropic system, the in-
teraction energy vanishes. This result is reflected in the
particle growth rates which display a behavior that is iso-
morphic with the growth rates obtained in the no-misfit-
strain limit. For isotropic systems with small differences
in shear moduli between the phases, the interaction en-
ergy is proportional to (6 - 1)D -6 at large interparticle
separations. Likewise, the isotropic growth rates are also
found to vary as (3 - I ) D - 6 when Lc = 0. For elastically
soft particles, the interaction energy is negative and the
overall elastic energy of the system is reduced as the
particles begin to interact. The free energy of the two-
particle cluster is lower than that of two isolated parti-
cles, and mass flows toward the two-particle cluster.
The opposite is true for elastically hard particles. The
elastic interaction for these systems is stronger at small
interparticle separations and for larger elastic inhomo-
geneity, causing the magnitudes of the particle growth
rates to change accordingly. It is important to note that
only in this particular case, where Lc = 0 and C~ = 0,
is there such a close correspondence between the elastic
energy calculations and the calculated growth rates. If
either Lc ~ 0 or C~ # 0, the l I D 6 dependence for the
growth rates is usually not observed for all interparticle
separations.
When the mean-field potential is chosen self-
consistently (as in Figure 6), the dependence of the iso-
tropic growth rates on the relative size of the particles
1186--VOLUME 20A, JULY 1989
can also be qualitatively understood in terms of the total
system energy. However, as the relative capillary length
is no longer zero, the system energy is related to the sum
of the interfacial and elastic energies.
The competing effects of the elastic and interfacial
energies are illustrated in Figure 13. It is a bifurcation
diagram (after Reference 15) in which those values of R
that extremize the system energy at fixed interparticle
separation, total particle volume (corresponding to the
self-consistent mean-field concentration), and spherical
particle morphology are plotted as a function of a non-
dimensional particle size, A, for a system in which the
particle is elastically softer than the matrix. A can be
expressed in terms of the nondimensional parameters used
here as [151
4C~(6 - 1) (1 + v) 2
A = [54]
9(1 - v)ZLc
The solid and broken lines depict the energy minima and
maxima, respectively, while the arrows indicate the di-
rection of decreasing energy.
For A < Ac, the interfacial energy dominates the elas-
tic interaction energy and the extremizing solution R = 0
gives a m a x i m u m in the energy. A system initially pos-
sessing a nonzero value of R evolves toward a one-
particle system (R = 1 or R = - 1 ) for the imposed
conditions. Although the elastic energy increases as the
large particle grows at the expense of the small particle,
it is more than offset by the corresponding decrease in
the interfacial energy. As the volume of the two particles
increases (A increases), a new extremizing solution ap-
pears at A = Ac = (4D 2 - 1)5/[12(12D 4 + D 2 - 3)]. [15]
There is a change in the stability of the extremizing so-
lution, R = 0, so that for A > Ac, both R - - 0 and
R -- -+ 1 are energy minima. Thus, if a two-particle sys-
tem possesses an initial value of R sufficiently close to
t
l /
f
f
" 1
I
I1:0 I
A
\
" T
-I
Fig. 1 3 - Bifurcation diagram depicting the relative particle sizes, R,
that extremize the sum of the elastic interaction and interracial ener-
gies, as a function of a nondimensional particle size, A, where A is
proportional to the inverse of Lc. Solid and broken lines refer to lin-
early stable and unstable solutions, respectively, for a system at con-
stant particle volume.
METALLURGICAL TRANSACTIONS A
R = 0, the energy analysis indicates that inverse ripen-
ing occurs; the smaller particle grows at the expense of
the larger particle. If the initial value of R lies suffi-
ciently close to R = 1 or R = - 1, then the large particle
coarsens at the expense of the small particle.
From both the kinetic and energy analyses, the relative
capillary length, Lc, is seen to provide a measure of the
particle size at which the elastic interaction between par-
ticles becomes important. When Lc is large (particles are
small for a given set of thermophysical parameters), the
capillary contribution to the interfacial concentrations
dominates (Eq. [28]), and the usual Gibbs-Thomson
equation is recovered. The driving force for diffusion is
the reduction of the system's interfacial energy, and the
flow of mass in the system reflects this as a larger par-
ticle tends to grow at the expense of a smaller particle
(corresponding to small A). In the opposite limit, Lc = 0,
the capillary contribution to the interfacial concentra-
tions is nonexistent; and, if diffusion occurs, it is a result
of the elastic interaction between particles. One expects
the elastic effects to become more important as the radii
of the particles increase.
Although qualitatively similar trends are observed for
both the energy and kinetic analyses under certain con-
ditions, the quantitative behavior may be much different.
This is easily seen by computing the value of Lc f r o m
the bifurcation point of the energy analysis, and then
comparing it to the value of Lc for which inverse coars-
ening is first observed in the kinetic analysis. Using the
nondimensional matrix elastic constants given earlier with
= 0.9 and D = 5, the bifurcation point corresponds
to a relative capillary length of Lc = 1.4 • 10 -9. The
corresponding point from the kinetic analysis with the
mean-field concentration chosen self-consistently cor-
responds to Lc = 9.4 • 10 -8. Thus, the kinetic analysis
predicts inverse coarsening at particle sizes about 70 times
smaller than the energy analysis. As the intercenter dis-
tance is decreased (volume fraction increases), the elas-
tic effects become important at substantially smaller
particle sizes, with the disagreement between the kinetic
and energetic analyses remaining roughly constant.
One reason for the apparent disagreement between the
energy and kinetic analyses is due to the differences in
the constraints employed in the two treatments. In the
energetic analysis, the intercenter distance is fixed and
the particles are constrained to be spheres, while both of
these constraints are relaxed in the kinetic analysis for
t > 0. Thus, despite the fact that the other conditions
for the energetic and kinetic analyses are identical, the
free-boundary nature of the kinetic problem allows the
system to sample a much larger parameter space in which
the particles can change shape and translate, as well as
exchange mass between one another. Since the kinetic
analysis is valid only for the initial stages of evolution
of the two-particle system, it is conceivable that once
the particles change shape, their isotropic growth rates
may change in such a manner that the range of inverse
coarsening is reduced.
C. Microstructural Development
When the mean-field concentration is not chosen self-
consistently, there is a slow net growth or dissolution of
METALLURGICAL TRANSACTIONS A
the two-particle cluster. While such a net change in vol-
ume of the two particles should not be observed in a two-
particle ripening problem, it is representative of the
coarsening behavior of two particles in a mean field set
by the multiparticle interaction of many particles during
coarsening. For example, a slow net increase in the vol-
ume of two particles is reasonable if the two particles
have average interfacial concentrations which are smaller
than the mean-field concentration. Thus, for C~ ~ C ~
the analysis corresponds to the ripening behavior of two-
particle clusters located in a mean field set by a larger
scale multiparticle diffusional interaction.
The strong elastic interaction between particles when
< 1 is seen to stabilize a particle doublet in comparison
to the no-misfit-strain limit. If the stabilization in the
kinetic analysis is viewed in terms of the coarsening be-
havior of two particles in a mean field set by multipar-
ticle diffusional interactions during coarsening, it is clear
that elastic stresses tend to promote the formation of par-
ticle clusters and, thus, a highly inhomogeneous spatial
distribution of particles. Such clustering would result in
much stronger spatial correlations during ripening than
has been found for the classical no-misfit-strain
c a s e 125'28'33'341 and has been observed experimentally
in m a n y two-phase coherent alloys. [32,35'361 As noted,
this stabilization is in agreement with Eshelby's energy
analysis t321 and in qualitative agreement with those of
Khatchaturyan and coworkers [37,381 for two-dimensional
tetragonally misfitting particles in the absence of inter-
facial energy.
Inverse coarsening in isotropic systems in which 6 < 1
may be expected to influence the microstructural devel-
opment of two-phase alloys through the effects of clus-
tering. Whether these effects are observed experimentally
depends upon the particle size at which the elastic effects
become important, the volume fraction, and the as-
sumption that the interface does not lose coherency. As
the elastic effects become more important with the in-
creasing size of the particles, it is likely that strong spa-
tial correlations will develop in the later stages of the
ripening process. Our preliminary work on an aniso-
tropic Ni-based alloy system indicates that the elastic ef-
fects can become important at smaller particle sizes (larger
values of Lc) than predicted for the isotropic system, even
when the particle and matrix have the same elastic
constants. 140]
Recently, Kawasaki and Enomoto f391 presented a so-
phisticated statistical mechanical treatment of coarsening
in elastically stressed solids with small particle volume
fractions. Their treatment is based on an analysis of
elastic interaction and interfacial energies in isotropic
media 1~,22j and employs a point (monopole) source
approximation for the diffusion problem. Their govern-
ing kinetic equation is determined from a free energy
dissipation argument that couples the elastic interaction
energy between two particles to the growth or dissolu-
tion rate of a particle. They show that elastic stresses
can lead to a stabilization of particles, with respect to
coarsening, when the shear modulus of the particle is
less than that of the matrix. In addition, they report a
new power law for coarsening of the form t ~/2 in the
elasticity-dominated regime (Lc = 0). It remains for
future work to reconcile the kinetic equation used by
VOLUME 20A, JULY 1989 1187