Classical Nucleation Theory

J.H. Perepezko

A. Kinetic Model
1. Clusters of particles originate when two atoms come together.
2. These clusters are in equilibrium with the surroundings and can grow or shrink by
the addition or removal of single particles.

3. Most clusters will break up into single particles, but there wil be some (small)
probability that a cluster will grow large enough (i.e. to the critical size) that the
number of particles that leave the cluster will just equal those that come to it.

4. If the cluster grows larger that this critical size, it will not shrink, but will grow to
a macroscopic size particle.

Each of these propositions can be given a reasonable justification.

1. We expect that simultaneous collisions of more than 3 separate particles is very

unlikely, and that the probability that 3 separate particles collide is much smaller
than that for two particle collisions.

2. As a result of statement 1, it follows that the only way a cluster can grow is by
single particle additions. Since we know that there is a distribution of energy of
the particles in the cluster, we would expect that there is some probability that
some of the particles will have a sufficient amount of energy to break free again.

3. We would expect the probability that a particle can leave the cluster to decrease as
the cluster grows bigger. Qualitatively this can be seen from considerations of the
surface curvature. As the cluster grows, the curvature of the surface will decrease
resulting in an increase in the average number of nearest neighbors for each
surface area. The greater the number of nearest neighbor bonds that must be
broken to free the particle, the less likely it is to break away. This is the origin of
the Gibbs-Thompson effect.

4. At some point in the growth, the probability that a particle will hit the cluster will
just equal the probability that one will break away. Once a cluster has grown
larger that this size, there will be only a small probability that it will decrease in
size since fewer particles will leave the cluster than will hit it.

These arguments can be expressed in a more quantitative manner. Nucleation theory

uses cluster size as a configurational coordinate. A flux in this configurational space
is the net number of clusters/cm3-sec. Going from size n to (n+1) [i.e. nucleation
rate]. Let C(n,t) represent the number of clusters of size n that exist in the system at
time t. Following from the assumptions of the kinetic model we can express the net
rate of change in the concentration of clusters going from size n to size (n+1) as:
 2

(1)
where  is the impingement frequency of a monomer (for nucleation in vapors

)  is the evaporation frequency and A’(n) is the area of space around

the cluster which will hold an impingement monomer. This procedure is called
Detailed Balancing. By definition, at equilibrium c/ dt=0 and C=Co. Thus, at
equilibrium:

If there is no steady state growth of clusters through the size classes (i.e. for steady
state conditions), the preceeding equation can be made more restrictive,

(2)
It should be noted that by using an equilibrium condition,  has been determined by
invoking the principle of time reversibility. This principle requires that at equilibrium
every microscopic process takes place at the same rate as its reverse. While  has
been found for an equilibrium condition, it should have the same value under other
conditions. In fact, the assumption that  is the same in a non-equilibrium situation
as at equilibrium is vital and absolutely necessary for the development of a
quantitative theory of nucleation. The survival of nucleation theory rests on its ability
to withstand repeated onslaughts attacking this assumption.

Substituting for  in the rate equation (1)

 3

The preceeding is the difference-differential equation for nucleation. This equation

can be greatly simplified if we assume that C/Co and Dco vary slowly with n and if
we disregard terms of higher than second order. To proceed, we define

where A(n) is the surface area for size n, A’(n) is the area around a cluster of size n
that can accept a monomer, and D id the diffusion coefficient of cluster in cluster size
space. Now, we expand in a Taylor series about n the following functions (note: the
t’s are implied)

Substituting these into the difference differential equation yields:

(3)

The above is the isothermal nucleation equation and has the form of Fick’s 2nd law of
diffusion in a force field. This observation is consistent with the view that nucleation
is a random-walk process through cluster size space.

We can rewrite the nucleation equation as an equation of continuity as follows:

(4)

where J(n) is the net number of clusters growing from size n to (n+1). Equation (4)
can also be written as:

(5)

random walk term drift velocity, U, term

 4

B. The Steady State

We now consider the possibility that a steady state current of clusters exists without
any change in C (i.e. . In order to produce such a situation, we must
continuously supply monomers to keep C(1) constant and remove all the clusters equal to
or larger than a certain size, n, so that size will not go to infinity. This scheme was first
proposed by Becker and Döring, Ann. Physik, 24, 719 (1935). Using the equation for
the nucleation rate we have.

where the subscript s indicates steady state. Also, from the steady state model we can
write

Integrating the equation for Js yields

In general

Where C(1) is the number of monomers and G(n) is the free energy to form a cluster of
size n. As with the case of vacancies, Co has a sharp minimum when G is maximized,
or when n=nc. Thus, we can set D=Dc and approximate Co by expanding G(n) about nc
to the second order to get

(6)

where
 5

(7)

Z is generally called the Zeldovich Factor.

C. Time Lag

We have not yet made any statements about the speed with which the process of
nucleation takes place. In order to make such estimates we must examine the
variation of C, and Co with the cluster size and introduce a concept, a critical region
of size . The following two graphs show Co, Cs, and Cs/Co schematically as a
function of the cluster size for a supersaturation of about 5.

We see that for a region around nc that Co/n will be very small and that J will be
dependent mostly upon the random walk term. The first problem is to determine a
reasonable width, , for this region i.e., how far can we go from nc in cluster size
before the increasing drift term begins to have a dominant effect on the direction of
cluster growth.

A reasonable estimate for  can be made by re-examining the nucleation equation

as follows:
 6

In the vicinity of nC

now

Therefore

and at n=nc

So that

If we set D=Dc the nucleation equation becomes

where

The above form of the nucleation equation can now be used to see how the
random walk and drift terms interact. Consider what happens to a group of clusters
 7

placed at size nC +/2. From diffusion theory the random walk term will cause a spread
of clusters in the form of a normal curve i.e.

The drift term will have the effect of moving the whole distribution to higher values of x.
We can see these movements quantitatively by solving the nucleation rate equation for C.

Thin film solution for diffusion

A measure of the amount of spread of the clusters is given by the root mean square
deviation of x, i.e.

where

From statistical theory we know that at any time, t, 32% of the clusters are farther from
the center of the distribution than d, 16% in each direction. We want to determine the
location of nC +/2 in such a way so that most of these clusters grow i.e. that the drift
term dominates over the random walk term. For the moment, let us say that if 16% or
less of the clusters become less than nC and thus run the risk of being pulled down in size,
by the drift term on the other side of nC, we have chosen nC +/2 properly. To express
this quantitatively, let
 8

Then if we have the correct /2

Since will have a maximum at t = /2

/2 is the time at which the maximum number of clusters have crossed nc. The inequality
then becomes

A reasonable definition for  is

The same procedure could be used to find nc =’/2, but since the situation is symmetrical,

. Thus, we may define the critical region as being of width  and centered on nc.

Those clusters which are smaller than nc - /2 will be affected mostly by the drift
term and dissolve. Likewise, those clusters larger than nc + /2 will tend to be governed
by random walk.
For completeness we should mention the usual approach to defining . Consider
G(n) as given in the equation for Co(n), i.e.

The dependence of G(n) on n is as follows:

 9

If a cluster reaches a size which has an energy closer than kT to G(nc), then there is a
good probability that is will move over the free energy “hill”. Thus we define  by the
following equation.

Expanding about nc, we get

or

This value of  is very close to the one we found before. The advantage of the previous
treatment was that we obtained a value for the time, /2, which we cannot get from the
last procedure.
We can now postulate that Cs/Co, as shown before, may be approximated by

Then using the steady state equation we get

 10

If we then accept as a reasonable value of , =1/Z, we get the same result for Js as we
had for the Zeldovich treatment.
We have not directly examined the length of time that would exist between a
rapid change in supersaturation and the appearance of visible clusters, but we have all the
information that we need to calculate this delay time.
It can be divided into three parts:
a) The time < to reach a size nc-/2
b) The time D to grow from (nc-/2) to (nc+/2)
c) The time > to grow from (nc+/2) to visible size
Times < and > will be related to the drift velocity which we have already derived, and
the time  will be related to /2.
Examining  first, we remember that /2 was the time for 16% of the clusters to
grow from (nc+/2) to less than nc. No more than this number would ever by smaller
than nc because of the effect of the drift velocity to larger sizes. By symmetry we said
that the same arguments could be applied to clusters growing from (nc-/2) to nc.
Combining these arguments, we see that if a particular cluster grows form (nc-/2) to nc
and then on to (nc+/2), the total time for this growth will be 2/2. Note, we have used
here the principle of time reversibility, i.e. if a cluster shrinks from (nc+/2) to nc in time
/2, the reverse process will take the same amount of time. The time considerations say
nothing about the probability that a particular cluster grows, but only the time the growth
would take if it does occur. The probability of growth is included in the nucleation
equation. In summary we have

(9)

We can proceed to find < from the drift velocity, u. By definition,

So (10)
 11

Here we use the growth velocity –u or alternatively reverse the time axis and consider the
time to dissolve a cluster of size (nc-D/2). In the same manner, we can calculate

(11)

From these times we can also estimate the time S to reach a steady state distribution of
clusters (of sizes up to nc+/2)
(12)
For vapors S ~ 10-6 to 10-7 sec. For solids S is substantially longer.