J. Phys. Chem. Solids Pergamon Press 1961. Vol. 19, Nos. l/2, pp. 35-50. Printed in Great Britain.

THE KINETICS OF PRECIPITATION FROM

SUPERSATURATED SOLID SOLUTIONS*
I. M. LIFSHITZ and V. V. SLYOZOV
Physico-Technical Institute, Kharkov, U.S.S.R.

(Received 10 October 1959)

Abstract-An analysis is made of the process whereby diffusion effects can cause the precipitation
of grains of a second phase in a supersaturated solid solution. The kinetics of this type of grain
growth are examined in detail. Some grains grow, only to be later dissolved; others increase in size
and incorporate further grains that they encounter in so doing. This latter phenomenon of coalescence
is discussed in a new “kinetic” approximation. Formulae are given for the asymptotic grain size
distribution, for the number of grains per unit volume and for the supersaturation as a function of
time. The effects of anisotropy, strain, crystalline order and the finite size of the specimen are
allowed for. It is pointed out that for a material that can be said to be “supersaturated with vac-
ancies”, the discussion can be applied to the vacancies as solute “atoms” which cluster together to
form internal cavities. The practical case of a real, finite crystal is here important, because the
vacancies can in general also escape to the surface. A special analysis is made of this example, and the
results are applied to the theory of sintering.

1. FORMULATION OF THE PROBLEM and the results contain meaningless logarithmic

IN A supersaturated solid solution, diffusion effects
may bring about the formation of grains of a new
phase. Two stages of this process may be distin-
guished. In the first, concentration fluctuations
produce nuclei of the new phase, which grow
directly from the supersaturated medium. The
second stage may be considered to begin when the
grains thus formed have reached an appreciable size
and the degree of supersaturation of the matrix has
become very slight. In such circumstances the
determining process is that of coalescence, i.e. the
growth of large grains of the new phase by the
incorporation of small ones. Fluctuation effects
play a negligible part in the second stage, as new
nuclei, to be of any importance, would need to be
of macroscopic proportions.
The kinetics of this coalescence process are con-
sidered in what follows. Some years ago, TODES(U)
discussed a similar phenomenon and correctly
arrived at various qualitative conc1usions.t Un-
fortunately his quantitative argument is in error
* Translated by R. D. Lown~.
t A similar problem has also been considered
GREENWOOD(~) but not to the same extent.
divergencies. A satisfactory solution was first given
in our paper of 1958, (4)in which the mode of arrival
of the system at the asymptotic distribution of
grain sizes was analysed in detail. The results of
this earlier study are rederived herein by a rather
different method, using the more accurate “kinetic
approximation” which allows the stability of the
asymptotic solutions to be examined. We also con-
sider the effect on coalescence of various factors
such as anisotropy and internal strain.
In the first instance we shall simplify matters by
ignoring anisotropy and considering the grains to
be spherical. The actual shape of real grains may
be taken into account by adjusting certain numer-
ical constants in the relevant formulae.
Fundamental equations
If c, is the concentration of the saturated
solution, and LY= (2a/K”)wC, is a parameter
containing the inter-phase surface tension Q and
the atomic volume wof the solute, the equilibrium
concentration CR at the boundary of a grain is
related to the grain radius by the usual formula:
by cm-,+
;.
CR = (1)
35
36 I. M. LIFSHITZ and V. V. SLYOZOV

The degree of supersaturation is small, so that space. The unknown functions f(p3, t)and x(t) may
C-C,=A<l, and for the present we may be determined as follows : from the equation of con-
ignore the interaction between grains since their tinuity in dimension space,
dimensions are small compared with” the mean
distance between them. Then the diffusion current
of solute across the grain boundary is given, per
unit area, by*
while from the conservation of matter
Qo = Ao+qo = A+q.
Qo is the total initial supe~aturation, making
The radius R of the grain therefore varies with allowance through a term qofor the volume of
time as? material initially in the grains. We have
dR 23
A-%.
dt=z
4 = +rrRto ;fp3dp3.
( 1
Thus for every value A of the supersaturation there 0
exists a critical radius R, = u/A with which a
grain is in equilibrium with the solution. If Bearing in mind that x = Ae/A, it follows that
R > Rc the grain grows, if R < Rc it dissolves.
This obvious mechanism also explains why large
grains can “devour” small ones by incorporation.
Both A and &themselves vary with time.
Writing ha for the initial supersaturation,
K = @rR;oQ,?
R,o = a/A0 for the initial critical radius, and
T = Rio/&, we now go over to dimensionless re- A normalization to unit volume has here been per-
duced magnitudes formed, so that
R
t’ = .L
p=R co T’ 12 -- *fdp3
s
0
Dropping the prime on t, we obtain
is the number of grains per unit volume. It follows
that the distribution function F(p, t) over the
(3) dimension pis related to f(p3,t)by
in which x(t) is a dimensionless critical radius such F(P, 4 4~ = ftpft)dp3,
that x(0) = 1. We now introduce a function i.e.
f(pa, t) which with in a factor +r is the volume JTP, 9 = 3p2f(p3, 6.
distribution function of the grains, and the rate
VR = dp3/dt of growth of the grains in dimension 2. TIME VARIATION OF THE CRITICAL DIMEN-
SIONS AND THE ASYMPTOTIC FORM OF THE
DISTIUBUTION IWNCTION IN HYDRODYNAMIC
* The equation contains the stationary value aC/ar of APPROXIMATION
the concentration gradient at the grain boundary. It is
easily shown that this procedure is valid if the initial
Along an axis in dimension space representing
supersaturation is small (Ao < 1). grain volume, the point pa (which apart from a
t If the grains are aspherical, but in the later stages of factor +r is the volume of some given grain) moves
the process grow in such a way as to preserve their according to the following prescription. Points to
shape due to an anisotropy in c(, then equation (2) and all the left of x3(t) accelerate to the left, until on
quantitative deductions from it remain correct so long as
reaching the origin they fall out of consideration,
R is interpreted as (3 VJ~T~)~!~,with V the grain volume,
and a, CLand .$@ are modified to the extent of numerical the grain having completely dissolved. Points
factors depending on the shape in question. initially to the right of x3(t) accelerate at first to the
 THE KINETICS OF P~CIPITATION FROM S~ERSATU~TED SOLID SOLUTIONS 37

right; but as the degree of supersaturation falls, expressed in terms of the initial distribution func-
X3(t) increases and successively overtakes points tion f$z) thus :
that in the first instance were to the right of it. co
After being overtaken, such points begin to move Ao
l- -e-r/s Ed:KeT
in the opposite direction and finally also disappear Qo J fo(r)z(r, T) dr. (12)
Yom
into the origin. The motion is throughout regular
in the sense that the initial order of a set of points is Here yo(~) is the solution of xLyo(~), ~1 = 0, i.e.
preserved. the lower limit of the range of starting volumes
appropriate to grains that are still undissolved at
time 7.
Both the form of equation (3) and the physical
meaning of x(t) suggest that it would be more The u~~~~o~~G soh&nz
natural to express equations (3)-(5) in terms of a As it will be shown that C&Z,T) is asymptotically
reduced volume independent of fob(y), the unknown functions are
now r&r, T) and y(7).
Z=L, x(0) = 1
The supersaturation A(t) decreases monotonic-
ally with time, and therefore l/x(t) does the same.
XV)
There are therefore only three possibilities for the
that compares p3 with the critical volume d. asymptotic behaviour of y(7) as T -+ co, viz.
Moreover, as t + co, A -+ 0, so that x(t) -+- co; y(T) + CO, 0 or constant. We shall show that the
thus x(t) may be used as a measure of time. It now first two do not apply.
appears that (3), (4) and (5) take on canonical form (i) T(T) -+ COas T --j. co. In this case, x3(t) varies
if time is introduced through the variable more slowly than t. The ultimate number of grains
7 = ln.S(t) per unit volume is associated with a rate &/dT > 0,
(7) which implies
and a new volume distribution function 46 is em- a,
ployed such that
4(x, T)X dz > const. (13)
s
d(z, T) dz = f(p3, t) dp3. (8) 0

Using (6), (7) and (8), equations (3), (4) and (5) The amount of matter in grain form at T -+ co will
become respectively therefore be
q > const.eT-----t co,
(9 7-fm
so that (11) is not satisfied.
(ii) y(7) --f 0 as 7 -+ ot3, i.e. g(t) varies more
dz
- = ??(z,y) = (x1/3-- l)y(T)--x, quickly than t. At sufficiently large times the
dr solution of (10) in this case is

v(T)=3$; ; = (N+-- l)y-2 S --y(T)--x,

7
00 z = ye-7 - 8-7 e”y(T’) d+.
Ao s
1 = Z e-*/a+ Kero &s, T)Xdx, 0
s
\ (11)
From the condition ztyO(~), T] = 0,
#fw-0 =.I$@), +-0 = p3.
ff the solution of (10) under the initial condition Yo(T) = j e+‘y(r’) dT’ = 347);

~17-0 = y be written z(y, T), then (11) may be 0

38 I. N. LIFSHITZ
the amount of matter in grain form at T 4 co is
9 = eT i ~~Y)~(Y, 4 dY G fmfdyfY dY 2o-)oo- 0.
Y&T) 3t 7-)co
Thus equation (12) cannot be satisfied [nor indeed
equation (ll)].
It follows that Y(T) must tend ~~pt~tic~ly to
some constant value ~0, and this we now proceed
to dete~ne~ together w&h the asymptotic form of
#(AX,7). We first remark that at such times as y(t)
may be put constant in equation (9), the general
solution of (9) is
and x an arbitrary function. Substituting this
result into (11) it will be seen that if the first part
is to be asymFtotic~ly time-inde~ndent, x must
have the form
Thus +(x, T) must tend asymptotically to
#@, 4 = f?-7dt)(x,r).
So far we have examined only the asymptotic
dependence of # on 7. We now turn to the deter-
mination of y, for which we write out the asymp-
totic forms of equations (9~(11):
W shows that the stable solution corresponds to a
d2
-= v(2, 5-1= (G/3- l)y -z, with y = const;
dr
WI
Examination of equation (15) shows that to satisfy
(171, @(9(a,r) can be different from zero only in the
and V. V. SLYOZOV
ultimate phase of change of a; for otherwise
co
@z dz
s
0
would be logarithmically divergent, irrespective of
the detailed nature of (0. For such a solution to
exist and to be continuous and finite over the
whole range of changing z, it would appear sufhci-
ent that @ and its first derivative go to zero at some
point x = .ZOand beyond this point join onto a
constant zero. Solutions of this type can formally
be found, and correspond to values of y > ~0.
RG. 1.
However, in Section 3 it will be shown that these
solutions are unstable when the encounters be-
tween different grains are taken into account. It
turns out that the only stable solution is that far
which #@, y) and all its derivatives go to zero at
xs, so that to the right of za the solution is identic-
ally zero. Further inspection of equation (15) now
value of y = ye such that the rate of change
a(g, ~0) has a second order zero at z = ~0, Using,
therefore, the conditions
0,
 THE KINETICS OF PRECIPITATION FROM S~~~ATU~TE~
The present results may be interpreted in the
light of Fig. 1, which shows w(z, r) as a function of
2 for various values of y. If y < ys, the representa-
tive points of all grains move to the left and dis-
appear into the origin. The grains in existence at
any given time comprise asymptotically a volume
distribution in the form of an infinitely long
“tail” to the initial distribution function, falling
off with 1 faster than a-2; therefore, the total
amount of material in grain form tends to zero:
instead of to a constant value. If on the other hand
y > ‘yo, all grains to the right of a point a?l move
towards the point 22, with the result* that
- const. e7 --+ co,
7’u.J
this again contradicting the conditions of the
problem, Thus only the case y = ys remains valid,
@(.a, ~0) has the form
I motion (10).
A s-&=t Y*)-, 2,<zs=q
&* */o)
@(z, ‘yo)=
0, z&X0=+$
(1% dominated by its infinitely long tail and its total
in which
d% YO)= -f&“/o) 2 0; (20) every initial distribution function is transformed
* Even if ME, yo) = 0 for 2 2 21 fit the outset,tbe sink for points with .V> as and a source for points
emountersbetweengrainstreatedin Section3 praiuce
grainsin the regiona > a, to whichthis con&&on then with I < 20. (The latter set of points have their
applies.
SOLID SOLUTIONS 39
$ln(zl/s+3)+9 Ix@--d/3)+
-I- (1 - !&Z1/s)-1
- ln(332-5/se); (21)
A = K-l 3Qo
+ 0.22;.
h = ~~o(l.ll~ co
r zdz
e-e
3
0
& YO)
From a general point of view, we have found
$(s, 4 and r(r) only in the zero approximation,
this being the most interesting in practice. More
accurate expressions for these quantities are de-
rived in the appendices. Here we remark only that
in the accurate equation y = ~0, all points lying to
the right of the point of contact zs = v and mov-
ing to the left are found to be unable to pass across
2s; they are, so to speak, %aught” there with the
result that the quantity of matter
q-e7-co
T-)W
and equation (11) cannot be satisfied. The con-
clusion is that Y(T) tends to ys from below:
It follows that the “leakage” across zs of points
approaching ,zafrom the right takes place more and
more slowly with the passage of time, the exact
rate depending on 8, which like r(r) itself must be
determined from equation (11) and the equation of
In the next approximation, 4(2,~) is not zero to
the right of the “blocking” point 20, but is some
function governed by the initial distribution func-
tion fO(2). However, as T + co, it is increasingly
contribution in equation (11) tends to zero. We
now see that in the region of the blocking point zs,
into a unique universal function, 20 becoming a
sink at the origin.)
40 I. M. LIFSHITZ and V. V. SLYOZOV

The res&s ~rn~r~ed is the probabili~ that a grain shall have a reduced
The basic results of this section may now be volume between z and a+-&. Fig. 2 shows
written out explicitly. The complete volume dis- P*(s1’3~ YO) = p(zz,~0) . (~~/~~1’3), the probability
tribution function has the form that a grain has mean dimension between ~1’3 =
(R/X) and (R/X)+ @lx), as a function of 23’3.
e-e a Further, from the fact that
Ae-7 - = n(T)& YO), .z G zo
Y&5 YO)= g(a, YO) ax
-& = a--0(x1/3-1)
i 0, s 2 a0
we find
(22) 2.
where x1/3_ 1
.&I-l -_ e-V+-ddz = [y;re--3’x]gO= 0,
s s(s* W)
~(7) = [$(z, 7) da = Ae-7 0
(23) (25)
0 from which it follows that
-
is the number of grains per unit volume, and 13= #3X = x(t). (26)

e-*

l
- -_ 332-5/3e. (21/3+3)-7/3($-x1/3)-11/8 exp -(1-$,$/3)-l,
3 < 20
p@, yo) = d%Yd
0; 2 3 .iro

u
m

0
p(z, “to)AZ= f$J(K
0
YO)& = 1) (24)

To obtain the distribution function relating to the

absolute dimension p of the grains, we need only
replace x in (22) by [p3/X3(t)]and divide by

x3(t) = $t. (27)

Inserting the original parameters, and remember-
ing that according to (26), z = R3/R3,we obtain

P’ f(Rf t) = ~(t)~(R~~)~-3,
I.0
n(t) = flQ&-3 = @t-l, with
(28)

If measurements are to be made on a section of the

material it is convenient to transform the volume
0 ti 0.5 distribution function @(a, T) into a distribution
z ‘/3 function over the apparent dimension r of the
FIG. 2. grains as revealed on the surface of the section,
 THE KINETICS OF PRECIPITATION FROM SUPERSATURATED SOLID SOLUTIONS 41

Writing I = (Y/S),this latter function is defined by if at the outset the mean grain dimensions are of
1 r the order of the critical size (Rs N lied)), then this
F(Z,T) dE= @(r, T) dr; @(r, T) = -F -, T latter dimension must appear in the condition.
x ix 1 However, if 7s $ Rd, then the first stage of
and is easily shown to be development consists in a growth of grains directly
from the solution, which goes on until the degree
of supersaturation has so fallen that the critical
F(Z,T) = 2 radius has caught up with the mean grain size
(X N &I) ; thereafter, strictly speaking, coal-
escence supervenes. In this case, Rcl plays the part
(see Fig. 3). of an initial dimension. The critical size is governed
not only by the initial supersaturation, but also by
LO0
-

0.75 -

cr 0,50-
=:r
LLti
i

*
ops --
0
t
FIG. 4.

the number of grain nuclei (where this can be con-

t sidered a fixed quantity). Thus if there are to
0 0.5 IlO I.5

Fxc. 3.
begin with no nuclei, and -0 B Rco = a/Ao,
growth from the solution takes place until
The degree of supersaturation at time t is given 4_ Ao
-rrR13x -.
by 3 710
In this first stage,
dR2
- = Z-B(Ao- +rrzOR3)= 2SAo 1 I
dt t -(~!1

Range ofvahi-iity of the fmmdae and this development occupies a time

The above formulae completely determine the tl N R;s/BAs.
asymptotic distribution as a function of grain
dimension and time. As shown in the appendix, The initial processes of the second stage (coal-
they are valid when escence) have a characteristic time

1 <5-Z = (InX3)2 = 9[ln(R/R~)]a, R&R&. tr, - @/Bz IV tl(ir;/Reo)

In this inequality, Rco = u/Ao is the initial when, as in the situation discussed, ??I @ Rco, i.e.
critical radius for coalescence, corresponding to the i. $- tl. Fig. 4 shows schematically the mean
starting sunersaturation Aa. It will be realised that dimensions as a function of time for this case.
42 I. M. LIFSHITZ
Infueme of other @et&
We have so far ignored a number of second-
order effects such as that of grain shape, crystalline
order, and the elastic strain that results from a
difference between the specific volumes of the
grains and the matrix. These matters are examined
in detail in a companion paper in the Journal of
Solid State Physics,@) in which we show that the
qualitative conclusions of the present study are
unaffected.
For instance, if the grains initially have a
variety of shapes, we find that at sufficiently great
times the only shape to survive is the most favour-
able one, which gives the greatest growth rate. All
the conclusions of this Section still apply in the
asymptotic situation.
Again, if the growth of a grain produces an
elastic deformation of its surroundings, there will
be an effect on the diffusion rate. We find@) that
this may be taken into account merely by replacing
C,, 9 and a by certain effective values:
CL = Cc&r; a’ = ueY; 9 = L@?
respectively, where
cm0
P asymptotic volume distribution was derived by an
= 3[(1+0)+2n(l-u)]
Here 81 is the elastic energy per atom in the
grain, brr the same quantity in the surroundings;
E is Young’s modulus, and n = E~EII; c is
Poisson’s ratio; V is the volume per atom in the
grain, and AV the difference between the volumes
per atom of the grain and the surrounding medium.
The possibility of order in the atomic arrange-
ment we allow for in the case where the energy
required to jump an atom from one sublattice to
the other is rather large, so that there can be said to
exist two essentially unrelated difhrsion fluxes. The
flux of particles to the grain is then treated exactly
and V. V. SLYOZOV
as above, except that 9 is further replaced by
the suffixes 1 and 2 referring to the individual sub-
lattices separately.
These second-order effects, therefore, call for
no more than a re-expression of certain parameters
in the theory. They do not alter the asymptotic
grain size distribution function or influence its
stability; our present solution, therefore, is
universal.
3. ENcouNmRS BETWEENGRAINs,INKINETIc
APPROXIMATION
Equations (1) and (3) are characteristic results
of applying the “hydrodynamic” approximation to
our problem, while equation (10) expresses the
rate of volume change (&/dT) = -g(z, T) at any
time as a function of the mean supersaturation
A(T). These equations, however, make no allow-
ance for the concentration gradient near a grain
that is in a state of changing volume. In fact, due to
local variations of concentration the rates of
volume change are not exactly equal to the mean
value -g.
The complication thus introduced can be over-
come, using the idea of “encounters” between
grains. An encounter is a situation in which two or
more grains fmd themselves separatedby a distance
that is less than their mean linear dimension, and
either unite directly or at least undergo a diffusion
interaction. The discussion may be said to in-
troduce a “kinetic approximation”.
In the treatment of Section 2, the stable
argument which revolved round the leaking of
representative points B across a blocking point
za = 9. Now encounters have the effect of
carrying representative points across this barrier,
and so the whole problem of their influence on the
asymptotic distribution must be examined closely.
This is true even if the over-all supersaturation is
small and encounters are a rare phenomenon.
Treatment of encounters
In the ordinary physical space of the material
we take the origin of coordinates in the middle of
some chosen grain and use a reduced position
 THE KINETICS QF PRRCIP~TATION FROM S~ERSA~~D SOLID SOLUTIONS 43

vectorg = r/&(t). Although the centres of the equality in (18) must contain a term corresponding
other grains are fixed in the lattice, the fact that to the total “encounter integral”
I?&) -c- cx, means that in reduced coordinates all
grains draw closer together and move radially
towards the selected one. Of course, the majority
of the grains disappear before reaching the origin
and do not have encounters.
We also employ a reduced distribution function
f and a reduced grain density ..&”which refer to a
unit of volume such that 55 = 1. They are related
to the earlier quantities C#and n through:
(The factor 6 is necessary, because without it each
encounter would be reckoned twice.) Taking ad-
vantage of the fact that w(z, 2’) is symmetric with
respect to a transposition of its arguments, it is
Since practically the whole excess of the dissolved easy to show that although the number of grains
material eventually collects into grains such that is reduced by encounters, the volume is conserved:
n N e-7, Jtr approaches a constant value as
E-+ CO.It follows that the probability per unit time
of an encounter in a reduced unit of volume be-
comes time-independent. Writing the number of
encounters, per unit time per reduced volume unit,
between grains of dimension in the range x to
a+ d;r and grains of dimension z’ as

an encounter-probabiii~ distribution function

p(.s, s’) is defined.
p{s, z’} may be expressed in the form p =
w(.s, ,s’)/~~, with 78 a characteristic time for en-
counters, and w(z, a’) a reduced volume centred
around a grain z, over which the interaction with
one of the grains z’ may be held to occur, It is not
difficult to show that to order of magnitude,
ZJ w t-f-~‘~ i.e. is given by the over-all reduced
volume of the m grains. T* must be of the order
of unity, inasmuch as all the parameters are of the
order of unity both in the initial condition
.;rlP = s = zs and in the equation
l&z
-s --y-s
a7
Returning for the moment to equation f17), it
will be observed that f is dependent on the small
parameter l/x. To bring out more clearly its
dependence on the small parameter Qs we intro-
duce the function
Qi = Xf.
Using this variable, a complete set of equations
controlling the system can now be written out with
full allowance for encounters. The equations have
the asymptotic form:

that describes the dissolution of the lesser grain.

When two grains meet, the greater of them 3Qo
absorbs the lesser, and their total vohnne is ap- = - x~;l’” ] ~(a--z’, z’)@(x-z’)@(d) dz’-
proximately conserved. A certain fraction of the 0
material may in fact go back into solution, but it
will be obvious that nothing is essentially changed
if we ignore this detail. For if the effects of en-
counters are to be allowed for, the first approximate
44 I. M. LIFSHITZ and V. V. SLYOZOV

00 so that for (3~s)~s+.ye > y > ye a solution of the

1=
s
0
(92 dx, y = const.
They may be solved by successive approximation,
using the fact that Xas is small.
At this point we may usefully notice that both
from the analysis of Section 2 and directly from
physical arguments one must expect that the new
considerations reduce y below ys. For in the terms
of the earlier discussion, grains that are carried
(32)
type mentioned is possible; it is, however, unstable
in respect of the first term in equation (31), since
in the next higher approbation it is
@(& r) = --&+ j YfU(x’, rw dx’1
and diverges at x = 2s. This also shows that (19) is
the only stable solution of (15).
Taking, therefore, (19) as the zero-order solu-
tion, the encounter integral becomes

3Qo
Y&I)] = $,[@o(x, ro)] = ~T;lXa(~), (33)

* J’ W(Z-Z’, Z’)a>o(x-x’)%(z’) dz’-@s(x) I+, z’)ciDo(z’)dz’, s < 2zo

40(z) = i 0 0

across the barrier-point 2s by encounters have a The method of successive approximation must
non-zero probability of leaking back, and this consist in refining the value of
implies y < 7s. We shall now show that the zero-
order approximation is equivalent to taking the -$I%g1+% = - ~r#IJol,
solution of (31) without the first part and with
00
Y = ys, and that this is equation (19). Indeed, the
solution of the homogeneous equation with y < ys 1 = @1X&.
j
does not satisfy (32), as 0

03 In general this will lead to

is then logarithmically divergent. 03

If y > ys, a solution may be constructed that is 1 = $&zb,
continuous and has a continuous first derivative j
0
over the whole range of z, and which is different
from zero only for s < .sl. g(z, y) = 0 for 2 = zl so that g?8= g@, y% (34)
and zs, zs > ~1, otherwise
00
@l = &O+ ~~&.@0(~, ro>le* W;
@,x dz
j z
0
the convergence of
will diverge at zs. In the neighbourhood of z = ~1, m
@(.a, r) behaves like a,%?C&X
s
(s- ~l)-rc$/~z)-~ 0
._Zi+ll= (s- z&-ttr;‘(r-rJ-++lJ
 THE KINETICS OF PRECIPITATION FROM SUPERSATURATED SOLID SOLUTIONS 45

requires c = 0. A refined value of y is found from The next approx~ation to the distribution
the condition function may be obtained by substituting (36) back
00 into the encounter integral; by iteration, the
~zdz = 1. distribution function can be determined to any
f desired accuracy.
0

The desired solution of (34) is accordingly

szo
e-@g-l cla,[@,(z’, ~O)]C?@
dx’,
@(s, Y) = s
z

(35) may be confirmed without difficulty if one
remembers thatf(z, y) and its first derivative go to
zero at z = %a, so that the first equation in (34) is
satisfied over the whole range of x.
Explicitly, we have shown that in first approxi-
mation
~e~~~e on the in&zl ~persat~ration
It is interesting to see that if in each interval
wxo < x < (n+ 1)~s we retain only the terms in
the lowest power of Qs, the solution may be written
(37)

___
3Qo

!
-r,lleAAe-@g-l, z < zo
47r

CD=
i
A=
I% e@l*Xo(z’) dz’ ;
? (36)
3Qo
-7-;le-!b*g--li e**,lbo(z’) dz’ ; zo < z < 2ze
4T .z

‘0. a > 2x0

/

These formulae take into account the fact that
where Ay G$ 1, S,!J increases sharply in the
neighbourhood of ze and thereafter varies
smoothly, with +*(z, r) = $(z, ~0); in this condi-
tion A = +r(QAyfys)-l/a. Xt follows that where
z > zs, f(s) N Qs. There is, of course, a smooth
transition between the two regions z c za and
9(z) in (37) can be determined within each interval;
for instance, in the range 0 < z Q 2~s we deter-
mined it above. Physically, this type of dependence
on Qe may be traced to the fact that the number of
grams of magnitude z is continually augmented by
the agglomeration of smaller grains having total
VohUUe &?a+ xb = Z.

S > zs. The form of (37) makes it apparent that to an

46 I. M. LIFSHfTZ and V. V. SLYOZOV

order of accuracy Qsl the principal ~ntribu~on in free SurfaCe, two ~~~~ng processes will take
equation (34a) comes from the ~st~bu~on func- place. Far from the surface, cavities will grow and
tion in the region RX=
u 30. Using this fact, we can coalesce according to the scheme developed in the
determine the asymptotic dependence of Ay on Qs earlier se&ions; on the other hand, near the border
as foliows: the cavities will tend to dissolve and the vacancies
to diffuse to the surface. (The surface can be re-
garded as a cavity of infinite radius.) This dis-
lodgment of cavities is also connected with the
phenomenon of sintering.
In the remainder of this section we often refer
Now because y = rs--Ar differs little from ye,
(38) may be rewritten expressly to the particular example of vacancy
difksion, although it is to be understood that the
results are equally applicable to the more general
case of a s~lpersaturat~ solution,

1-2
whence Consider the situation in a half-space. Equation
3~2 (3) for the grain growth and the ~strib~~o~ func-
-hy=l27$ 2ln ---1np
4W z-..---- tion (4) will still apply, and so will their equivalents
3/o 3Qo (in Qd2 (9) and (10). However, the conservation law
Qs =: A+9 that underlies equations (5) and (12)
must be replaced by a diffusion equation. Each
point of coordinate I is now a source or sink of
intensity c?@t, and 4 is now @, $1 because the
cavities are on the whole dissolving. The complete
set of equations is properly

AND THE THl3GRY GF S-G

The su~~at~ated cry&al studied in Sections
1-3 was by implication ~nr~st~cted in size, and
therefore contained no gross diffusion currents on
a macroscopic scale. If we now remove the con-
dition of over-all spatial uniformity, the problem is
altered. The most important practical example of
this more complicated situation is provided by an
interphase boundary between the solution and
some other phase, which might be, for instance, a in which a+(~,t) satisges (10) and itself determines
crystal of pure solute, or more generally perhaps a the form of A@, t).
crystal of the same phase whose precipitation is It will be remembered from Section 2 that the
being considered. At such a boundary, the super- analysis led to a unique law for the decay of
saturation Ajg _ 0 = 0 and ma~rwcopic diffusion supersaturation [equation (29)] :
currents exist.
A = &+a.
Vacancies This means in principle that if the asymptotic fall
An important special case is that of a crystal of the supersaturation were to be in some aray
supersaturated with vacancies, in which our dis- retarded, however slightly, a process of limitless
cussion can be applied to the vacancies as salute grain growth would set in and 9 -+ CO. Alter-
“atoms” which cluster together to form cavities natively, an i~d~nitely small artificial increase in
playing the part of our grains. ff such a crystal has the rate of decay below the r-f/s law must fairly
 THE KINETICS OF P~CIPITATION FROM S~ERSATU~~D SOLID SOLUTIONS 47

quickly result in the total dissolution of all grains (39) gives

@ + 0). The mechanism of sintering is obviously
related to this fact, and at least asymptotically
must admit of the following discussion.

Reduction of the problem

There are, asymptotically, three fairly sharply Before imposing the other two boundary condi-
defined regions in a supersaturated half-space: tions it is helpful to notice that al(t) is the flux of
particles at the surface:
(a) the farthest from the boundary, in a range
is < I < CO.Here the influence of the boundary is
negligible, and as with an infinite medium the
cl@) = 53% .
supersaturation is given by (29);
ai i I-o

Asymptotically, therefore, it must be a decreasing

(b) a region & < I < 12 in which the cavities
function of time, which suggests using the
(grains) have a general tendency to dissolve ; and
asymptotic form al(t) = BJts, (s > 0). Application
(c) the region 0 < I < Zl in which there are no of (42) now yields
cavities (grains) at all, but an over-all diffusion of
vacancies to the side.

The Siting distances II(~) and la(t) are situated

well inside the crystal, and the skin layer that is As will appear shortly, P(t) is o(t), that is to say
free from cavities thickens continually.
Strictly speaking, equations (41) and (42) should
be applied only to the intermediate region
Zr < 1~ Z2.However, it will be shown below that It follows that at large t the principal term in (45)
(&-Zr)/~-@ 1. For a first approximation we shall isthatwithn = 0;
accordingly, consider an average depth c(r) =
#t-22), and by imposing some suitable boundary A =:---.z
B1 al(t) * I.
condition at 5 shall evaluate its magnitude and the ta
vacancy concentration in the surface layer
0 < 1 < 5(t). For this purpose the intermediate Substituting now (41) into (39) we find
region may effectively be replaced by a source of al{ = M-1/3,
strength Qs(&/&) at 5, beyond which all the excess
vacancies may be said to be in the cavities. Then Qo db
al=--,
we may say: 9 at
%A @A and therefore
-@z;
-&= R (Bxt)~‘~
5 = 3~/2(f$W.--. -_ m _, ;
QW Ql/2
0

A = At-W/{ = ~mQ~k&V(9t)-!W;
m I 31/2(#)U3X+2. (47)
(44
Order$ of magnitude
A solution of (38) is now required subject to the It remains to determine the width of the inter-
condition A 1I.. 0 = 0. On seeking it in the form mediate region. By (43),
8s: at Ta(t)
A= 2 a&)& --_-=;
i&--l 5 3t --ii--’
48 1. M, LIFSHITZ and V. V. SLYOZOV

where !!‘a is the time xequired to dissolve the
largest grain to be found on the boundary at time t.
Now &,, = +I-&,and As N t2. A simple calcula-
tion shows that in the equation of motion (lo),
y = 3(dt/d‘+) < -*j < yo. The largest grain has
3s = 7, and we deduce that it dissolves in a time
Ta + $t. Thus
limiting case, when Ts < TX, the vacancies are lost
to the surface before cavities can be formed with
any appreciable probability. In the intermediate
situation where a2 N R3,fa, the exact course of
events will be heavily influenced by the initial
distribution function and is not amenable to a
general discussion.

-84z -.1 5. SUMMARY

5 9 1. An investigation is made of the way in which,
in a supersaturated solid solution, grains of a new
By the same token,
phase grow as the result of diffusion processes.
The final stage of the process is examined in
particular. A critical dimension x is determined
such that large grains having dimension p > x
and grow at the expense of small ones with p < x.
al: 1 2. At large times f the critical dimension x tends
-=-----$1, (50) asymptotically to depend on time as Pa, The
L 3Q;ie degree of supersaturation correspondingly falls
with L the mean distance between grains, Con- as t-113and the number of grains as t-1.
sequently 3. A detailed picture of the law of grain growth
sg 9 L g z?. is provided by the distribution function of grain
dimensions. With the passage of time this distri-
Although, therefore, the intermediate region is bution approaches a certain universal function
much thinner than the surface layer its width is [equations (22-24)] irrespective of the initial
considerably greater than both the mean cavity distribution. This function is remarkable in that the
size and the mean distance between cavities tail of the distribution towards infinity has disap-
(grains). The assumptions we made for equations peared, leaving a maximum grain size one-and-a-
(43) and (44) are therefore justified post facto. half times the critical dimension (pmax = $3).
The characteristic times associated with equa- 4. The effects of anisotropy and of the non-
tion (47) are the same as those for the asymptotic spherical nature of the original grain nuclei, like
equations describing coalescence in an infinite the effect of internal strain set up during the decay
medium. of the supersaturated solution, can all be taken
into account simply by the use of “effective”
A further restriction parameters in place of certain quantities in the
If the discussion for a half-space is now taken expression for the critical size. The form of the
over to a specimen of finite size, as of course in distribution function is unaffected by these con-
practice, it must be realized that there is another siderations.
condition to satisfy. For example, consider a plate- 5, The process herein demonstrated for the
shaped layer of thickness a. Three characteristic coalescence of grains can also be applied to the
times may be defined: that for the escape of vac- study of “crust sintering”. Each microscopic
ancies to the surface, TO N a2/9; that for the cavity in a specimen before sintering may be
growth of cavities from solution, TIN &2[9Ao (as regarded as a crystal composed of vacancies. The
shown in Section 2) ; and the characteristic time for cavities near the surface tend to dissolve as a
coalescence, TC N @/Z@u. For our picture to result of their vacancies moving outwards to the
apply, TO must necessarily be much greater than boundary of the specimen, while in the depth of
T,, so that a2 $- Ri/rx; the inequality is written in the material a process of coalescence goes forward
this form because the mean cavity dimensions must essentially as in an infinite medium. The thickness
be those appropriate to the situation after coal- of the “crust” increases with time as 5 N t113
escence has set in, i.e. R $ Rx. In the opposite [equation (47)-j.
 THE KINETICS OF P~CIPITATION FROM SUPERSATURATED SOLID SOLUTIONS 49

APFENDE 1. AN ACCURATE EXPRESHON FOR and correspondingly

THE ASYMPTOTIC BFZLAVIOUR OF ~(7)
As shown in Section
write it
2, Y(T) < yo, so that we may X3 =#t
E-&u+ ...I].
l+

The expansions may be broken off at any point at which

E(T) may be determined from the equation of motion the succeeding term is su&iently small; in particular,
(10): the basic approximations ~(7) = ~0 and x” = (4/g)t are
justified if
dx
- = (&3--1)y(+-2
dr 72 = (In&)2 = (ln$t)2 $- 1.

The fractional correction to ys decreases swiftly with

time, so that the first appro~mation becomes more and
The Iatter term on the right has been so written because more accurate. It is a rather curious fact that the be-
it is important only near z = 20 = y. Put now for con- haviour of the solution near the blocking point neverthe-
venience a = u3. Near z = so or u = uo = Q the less continues to depend on just this correction.
equation becomes
APPENDIX 2. THE DYSTRIBUMON FUNCTION
du IN THE REGION z > zs, AND IN THE
- = -$(u-#)2-+(T).
dr NETGHBOURHOOD OF zo
(i) Introduce the variable
Changing variable toy = (u-+)/C(T) we may write this:

if= I-$

3 dy
--= and notice that as z + co, g(z) -+ z and z -+ eQ, i.e.
2 Ed7 zxs -+ er+dr. The form of the distribution function to the
right of zs must be related to the initial distribution
For the same reasons as those given in the text, the fo(p3) over p3:
existence of a stable solution requires that in this ap-
proximation there should be a “blocking” point on the z x6 + $1~ fo(P”W) -+fo(=+W3*
axis and that the rate of change of s should have a second
order zero at this point. This yields This being the case, we may determine the distribution
function for z > 80 from

x(7+$) -+fo(e7*+7+*,
3
E2(7)4 - , T-+-cc); or,
-----1/o’
dT 472 #(2x,7) =fO(e7+*)e~+!Q-1(z).

from which Because

W
3 dy
2d -+ r3+-tu”(3)32‘ .h@)z dz
s
0
Evidently dy/dr does have a second order zero at the
exists, fo(z) falls off certainly faster than z-s, and in fact
blocking point y = ys = &/3.
normally is exponential in z. It follows that in the zero-
It is now apparent that at sufficiently great times,
order approximation, the volume probability density to
the right of zo is zero. For
Y(T)= Y0(1- p2>.
The expression for the blocking point has been refined to
p(x, T) = +$. =:fo(eT+~)~(“ne-~gg-l(z)
$)‘3 = g+gy’(3)+) = g+$-l*
By continuing with this procedure, an asymptotically
= f0(p)p2e-~g-1(x) 4 0
accurate expansion for y(r) can be derived:
(p = ey**s co).

Of course, the probability density is non-zero to the left

of .%a.
50 1. M. LIFSHITZ and V. V. SLYOZOV

(ii) To derive a more accurate distribution function in
the vicinityof z = IO = 9.
Write= again for convenience H = u3. Near as the
equation of continuity may be expreased as
from which we at once obtain
f E Ag-f(y)~ e IQ-W) dfi e-7 em f~-%4~Y.
(iii) To show that as 7 + co, the integral contribution
from the neighbourbood of zs becomes neghgibly small.
The quantity of material accounted for in this region is

Changing the variable to

y = $@+-@r,

d(% 7) &J = f(r, lfl T) dY

yields

Now from the equation of conduit,

g(r) = (Y-4)2.
The solution must be of the form

f = f [Ln (T 6 JrwA a)].

because to the right of as the distribution is in the form
We now observe that of an infinitely long drawn out t&I, which may be
ignored. Thus

g-l(u) -+ ~~l(y~2 e fg-*(glf@, Evidently B must be zero, since matter must be con-
served. Consequentiythe amount of material represented
given that 1 > [u-z > @{T). In fact in the vicinity of zs as r + Q) is

Tf jg-‘(U)dU = <l-[~(u”+]-1)
0
REFERENCXS
1. TOD~S 0. M., Zkfk khim. 20,7,630 (1946).
Moreover at the same point I = a”” the original distri- 2. TODB 0. AI. and K~~~usncnnv V. V., Zh. _fk.
bution function must be given in zero order approxima- khiar. 21, 3, 301 (1947).
tion by 3. GRT&XXXJ G. W., Actu. .Sfet. 4,243 (1956).
4. LXFSSTZ I. M. and SLYOZOV T.V., Zh. eksp. tear,
dY fizx. 35, 2(8) (1958).
exp -(T-[@-#)]-l), 5. Lrvarrrrz I. M. and SLYOZOV V. V., Zk. Jfz.
fz+& ~yoy~ykk tel. (in press).