Chapter 1

Ficks Laws and Macroscopic

Diusion
1.1 Introduction
In the rst part of this course we will discuss diusion, which is the fun-
damental process for atomic transport in solid state transformations and
reactions. We start with a mathematical description of diusion due to a
concentration gradient. We see that random atomic jumps act to smooth
out spatially varying concentrations, with an atomic ux which is propor-
tional to the concentration gradient. Later in the course we show that this
behavior is a special case, and that, more generally, the atomic ux will be
proportional to a gradient in chemical potential.
1.2 Ficks Laws
1.2.1 Ficks First Law
We rst examine a simple one-dimensional model of diusion. We consider
diusion of a trace amount of an impurity (or tracer) in a single-phase alloy.
If the planer density of impurities at a position z is given by (z) (measured
in atoms/cm
2
), and if the spacing between adjacent planes is z, then the
volume concentration of impurities c(z) is given by:
c(z) =
(z)
z
1
2 CHAPTER 1. FICKS LAWS AND MACROSCOPIC DIFFUSION
Further, we assume that we have one-dimensional random nearest neighbor
jumps, and the diusing atoms are chemically identical to, but distinguish-
able from the host atoms. This is the case for radioactive isotope tracers, for
example. We dene J
+
to be the ux of atoms to the right from the plane
at z to the one at z + z. This is given by:
J
+
=
1
2
(z)
where is the mean jump frequency, and the factor of 1/2 accounts for the
jumps being able to go in either the plus or minus z direction. We can also
dene J

as the ux of atoms to the left from the plane at z +z to the one

at z, and we nd:
J

=
1
2
(z + z)
If we assume that is not a function of concentration, then the net ux J is
given by:
J = J
+
J

=
1
2
[(z + z) (z)]
=
1
2
z [c(z + z) c(z)]
=
1
2
(z)
2
c(z + z) c(z)
z

1
2
(z)
2
c
z
= D
c
z
(1.1)
where in the last step we have assigned:
D =
1
2
(z)
2
The quantity D is known as the diusivity, and for the three-dimensional
case we nd:
D =
1
6
(z)
2
Equation 1.1 which relates the concentration gradient to the ux is known
as Ficks rst law. As we go into this course, we will nd that Ficks rst
1.2. FICKS LAWS 3
law does not always hold, but it is in fact a special case of the more general
statement that a ux will be driven by a gradient in chemical potential. In
many cases, the gradient in chemical potential is proportional to the gradient
in concentration.
Lets discuss units. In the cgs system, the various quantities have the
units:
J
_
atoms
cm
2
s
_
= D
_
cm
2
s
_
c
z
_
atoms
cm
4
_
Actually, the ux is a vector quantity in that you might want to know
the direction as well as magnitude of the atomic ow. Hence, Ficks rst law
can be written as a vector equation:
J = Dc
where is the gradient operator.
1.2.2 The Conservation Equation
In practice, atomic uxes are often a dicult quantity to measure. In solids,
it is more typical to measure concentration prole changes which result from
thermal treatment. Hence it is helpful to have an expression to describe
how a ux will change a concentration prole. Consider the situation in Fig.
1.1, where we have a bar with concentration prole c(z), giving rise to ux
J(z) = Dc/z. We consider the volume element Az where A is the cross
section area. If there is curvature in c(z) then the uxes at z and z +z are
not equal. We will have a ux into our volume element given by J(z) and a
ux going out of J(z +z). Thus the rate of change in concentration in the
volume element due to uxes is:
c
t
=
J(z) J(z + z)
z
so in the limit of small z we have
c
t
=
J
z
In three-dimensions the vectorial relationship is:
c
t
= J (1.2)
4 CHAPTER 1. FICKS LAWS AND MACROSCOPIC DIFFUSION
where the quantity J is known as the divergence of J, and where now the
ux is represented by a vector J which contains information on the direction
and magnitude of the atomic ux.
Equation 1.2 is known as the conservation equation, and it holds if the
species we are considering is neither being created or destroyed. If the species
is not conserved, then we must add a source term to Eqn. 1.2. For example,
we are sometimes interested in the diusion of vacancies which can be created
or destroyed at a surface or interior defect. If p(r, t) represents the rate of
production of the species of interest as a function of position and time then
the conservation equation becomes:
c
t
= J + p(r, t)
It is important to recognize that the conservation equation will hold even
when Ficks rst law does not.
1.2.3 Ficks Second Law
If we combine Ficks rst law
J = Dc
with the conservation equation
c
t
= J
we nd:
c
t
= Dc (1.3)
Equation 1.3 is known as Ficks second law. We will frequently consider the
situation where the diusivity D is independent of concentration, and hence
position, so that Ficks second law becomes:
c
t
= D
2
c (1.4)
where
2
is known as the Laplacian operator. In a Cartesian coordinate
system the Laplacian is:

2
=

2
x
2
+

2
y
2
+

2
z
2
1.2. FICKS LAWS 5
c
z
J
z
J(z) J(z + z)
z z + z
Figure 1.1: Schematic of concentration and ux as a function of position near
a volume element.
6 CHAPTER 1. FICKS LAWS AND MACROSCOPIC DIFFUSION
The form of the Laplacian and other vector operators for dierent coordinate
systems are summarized in a handout from MSE 205.
Equation 1.4 is a partial dierential equation which is second order in
the spatial coordinate and rst order in time. Hence a unique solution will
be characterized by a statement of the initial state of the system and two
boundary conditions.
1.2.4 Cylindrical Example Using Ficks First Law
For the rst example we consider diusion through a pipe. This is a common
experimental situation for determining diusion rate and its compositional
dependence. We have a pipe of inner radius r
1
and outer radius r
2
(Fig. 1.2).
We somehow conspire to maintain a constant concentration c
1
on the inner
pipe wall at r = r
1
. This can be done, for example, by maintaining a constant
chemical potential in the gas inside of the pipe. We also assume that the
pipe is long enough so that there are no end eects, and that we have waited
a long time, so the concentration in the pipe is not changing, i.e.
c
t
= 0
This later assumption is known as a steady-state condition and with it the
concentration becomes independent of time, i.e. c(r, t) c(r).
It is natural to use cylindrical coordinates for this problem, and due to the
symmetry of the geometry and boundary condition Ficks rst law becomes:
J = D
dc
dr
r
where now we take the total derivative since we are at steady-state, so the
concentration is not a function of time, and the symmetry of the problem (ig-
noring end eects) dictates that it is also not a function of spatial coordinates
other than the radius.
A steady-state ux within the cylinder will give rise to a mass ow outside
the cylinder. In the experimental situation, we measure the rate at which
atoms appear outside the cylinder m(atoms/s). If, inside our pipe, we draw
an imaginary cylinder with radius r
1
< r < r
2
we see that the total number
of atoms passing through this cylinder must be the same for any value of r
within this range. That is, the ux times the area must be m. This yields:
m = (Dc) (2rh)
1.2. FICKS LAWS 7
h
r
r
r
2
1
Figure 1.2: Schematic of pipe showing length h, inner radius r
1
, and outer
radius r
2
.
= 2rhD
c
r
If D = D(c) we can integrate to nd:
_
c
c
1
dc =
_
r
r
1

m
2rhD
dr
c c
1
=
m
2hD
_
r
r
1
dr
r
=
m
2hD
ln
r
r
1
We see that a plot of c(r) versus ln r will give a straight line. By measuring
the mass ow m and the concentration at two dierent r-values, we can
obtain D. For example if we measure c
1
and c
2
, the concentration at the
inner and outer radii of the cylinder, we obtain:
D =
m
2h(c
2
c
1
)
ln
r
2
r
1
In the case that the diusivity depends on concentration (D = D(c)), we
8 CHAPTER 1. FICKS LAWS AND MACROSCOPIC DIFFUSION
cannot perform the integration as above. We have:
m = 2rhD(c)
c
r
which can be solved for D(c) to nd:
D(c) =
m
2rh
c
r
=
m
2h
c
ln r
This gives us a way to determine D(c). We measure the concentration c(r) as
a function of radius r. If D depends on concentration a plot of c versus ln r
will not yield a straight line, but from the slope of the line we can determine
D(r) which can be transformed to D(c). This is shown schematically in
Fig. 1.3.
1.3 The Thin Film Solution
The next example we consider is the solution of Ficks second law with the
initial condition of an innitely thin lm placed between two semi-innite
media. This is often a good approximation to real situations, and it forms the
basis for constructing solutions to more complex situations. The boundary
conditions are that the composition must go to zero at z = . The solution
is:
c(z, t) =
m
B

4Dt
exp
_

z
2
4Dt
_
(1.5)
where m
B
is the total amount per area of the diusing species, D is the
composition-independent diusivity, and z is the distance from the original
interface. It can be shown by direct substitution that Eqn. 1.5 solves Ficks
second law (Eqn. 1.4):
c
t
= D

2
c
z
2
The solution in Eqn. 1.5 also has the correct normalization, that is, the total
amount of diusing material is m
B
:
_

c(z, t) dz = m
B
1.3. THE THIN FILM SOLUTION 9
c
ln r
ln r
1 ln r
2
c
1
c
2
D
ln r
ln r
1 ln r
2
D
1
D
2
Figure 1.3: Schematic showing concentration as a function of ln r result-
ing from a steady state cylindrical diusion experiment, and the position-
dependent diusivity D(r) which can be extracted.
10 CHAPTER 1. FICKS LAWS AND MACROSCOPIC DIFFUSION
Furthermore, we match the boundary conditions, and, at t 0, our solution
has the values:
for |z| > 0, c 0 as t 0
for z = 0, c as t 0
which states that initially we have a nite amount of material in an innitely
narrow band. This is the initial condition for the problem we set out to solve.
The concentration has a Gaussian spatial prole, with width w (distance
between inection points) given by:
w = 2

2Dt
This width is a measure of the diusion distance. The concentration at the
origin (z = 0) is given by:
c(0, t) =
m
B

4Dt
which decreases linearly with 1/

t.
It is instructive to examine the behavior of this solution relative to Ficks
laws. In Fig. 1.4 the composition, ux, and time rate of change of composition
are plotted as a function of distance at some xed time. In region where the
slope of the concentration is positive, the ux, given by:
J = D
c
z
,
is negative, so that material is moving to the left. The opposite is true
where the slope of the concentration is negative. The diusing atoms move
down the concentration gradient. By Ficks second law (Eqn. 1.4), the rate
of accumulation of material is proportional to the second derivative of the
concentration with respect to position. Thus, where the curvature of the
concentration is positive, diusing atoms are accumulating. The opposite is
true where the curvature is negative.
1.4 Solutions Based on the Thin-Film Solu-
tion
There are a class of solutions based on the thin lm solution which are
examples of the general problem solving method known as Greens Functions.
The general recipe is:
1.4. SOLUTIONS BASED ON THE THIN-FILM SOLUTION 11
c
-2 -1 0 1 2
z
J
-2 -1 0 1 2
z
Flux
to Left
Flux
to Right

c
/

t
-2 -1 0 1 2
z
Accumulating
Leaving
Figure 1.4: Composition, ux, and time rate of change of composition for
the thin lm solution.
12 CHAPTER 1. FICKS LAWS AND MACROSCOPIC DIFFUSION
Find the response of the system to a delta function input. In the case
of the diusion equation this amounts to solving Ficks second law for
the innitely thin lm with unit total material (m
B
= 1).
Integrate this response over actual input. In the case of the diusion
equation, this amounts to integrating over the initial composition.
This technique can be applied to cases when the problem can be framed
with an innite system. In a homework problem, you will show that:
c(z, t) =
1

4Dt
_

f(z

) exp
_
(z z

)
2
4Dt
_
dz

solves the diusion equation, where f(z) is the initial distribution for an
innite system, i. e. :
f(z) = c(z, 0)
and the diusivity is not a function of composition (D = D(c)).
As an example of the use of this method, we consider the situation of
a pair of semi-innite solids which initially have dierent composition. We
arbitrarily set the composition in one equal to 0 and the other equal to c
0
,
so that the initial condition is:
c(z, 0) =
_
0 z < 0
c
0
z 0
The solution is then:
c(z, t) =
1

4Dt
_

f(z

) exp
_
(z z

)
2
4Dt
_
dz

=
c
0

4Dt
_

0
exp
_
(z z

)
2
4Dt
_
dz

We can cast this integral into a standard form with the variable substitution:
=
z z

4Dt
so that
d =
dz

4Dt
1.4. SOLUTIONS BASED ON THE THIN-FILM SOLUTION 13
and the solution becomes:
c(z, t) =
c
0

4Dt
_

z/

4Dt
e

2
_

4Dt
_
d
=
c
0

_
_

0
e

2
d +
_
z/

4Dt
0
e

2
d
_
This can be represented as a special function known as the error function
erf(y) dened as:
erf(y)
2

_
y
0
e

2
d
This function has the properties:
erf(0) =
2

_
0
0
e

2
d = 0
erf() =
2

_

0
e

2
d = 1
In terms of these denitions, the solution is:
c(z, t) =
c
0
2
_
1 + erf
_
z

4Dt
__
It is interesting to consider what we have done using this solution method.
We have used a collection of innitely thin slabs to represent a semi-innite
solid, and have let the material in each slab diuse into a Gaussian prole
in a manner predicted by the thin lm solution. The composition prole is
then a sum of these Gaussians. This is illustrated schematically in Fig. 1.5.
It is also interesting to consider the time evolution of this solution. The
value of the composition at the origin for all times is c
0
/2, as is the value at
innite time at all positions. That is:
c(0, t) =
c
0
2
c(z, ) =
c
0
2
The composition prole spreads as a function of time. For a homework
problem you show that the points z
1
and z
2
dened by:
c(z
1
, t) =
c
0
4
c(z
2
, t) =
3c
0
4
14 CHAPTER 1. FICKS LAWS AND MACROSCOPIC DIFFUSION
1.0
0.5
0.0
c
-10 -5 0 5
z
Figure 1.5: Schematic of solution to the step composition prole. Dotted
lines are the solutions for thin slabs, and the sum of these solutions, solid
line is the actual error function solution.
spread apart with the distance between them being proportional to the square
root of time. So again, we nd that the diusion distance is proportional to

Dt.
We can generalize the error function solution for other cases. The func-
tion:
c(z, t) = A + B erf
_
z

4Dt
_
solves the diusion equation. If we can choose constants A and B to meet
the initial condition and boundary conditions, we have a solution.
Before leaving this solution method we make some general comments.
First of all, we have assumed that the diusivity is not a function of com-
position (D = D(c)). This is often a bad assumption if there are large
composition dierences in the problem. However this solution is often used
to estimate the behavior of a system assuming some average diusion con-
stant. Furthermore, we have also assumed that the system is innite. This
assumption can also lead to trouble as explored in one of the exercises. In
general this solution is valuable for the case that the diusion distances are
small relative to the dimensions of the system.
1.5. SEPARATION OF VARIABLES 15
1.5 Separation of Variables
We now turn to a technique which is best used for describing a system at
long times, where it is approaching homogenization. We start by assuming
that the solution can have the form:
c(z, t) = Z(z) T (t) (1.6)
where Z(z) is a function only of position z, and T (t) is a function only of
time t. We note that this is not always possible, and in fact the solutions
we have considered so far are not separable, since we have found solutions of
the form:
c(z, t) = f
_
z/

t
_
Nonetheless we proceed. Inserting Eqn. 1.6 into the diusion equation:
c
t
= D

2
c
z
2
we obtain:
Z
dT
dt
= DT
d
2
Z
dz
2
1
DT
dT
dt
=
1
Z
d
2
Z
dz
2
(1.7)
Each side of Eqn. 1.7 is only a function of one variable, and since both x and
t are independent variables which can vary freely, the only way that Eqn. 1.7
can hold is if both sides are equal to a constant, for which we choose
2
.
Looking at the t dependence we nd:
1
T
dT
dt
=
2
D
Integrating we nd:
_
T
T
0
dT
T
=
2
D
_
t
0
dt
ln
T
T
0
=
2
Dt
T = T
0
e

2
Dt
(1.8)
16 CHAPTER 1. FICKS LAWS AND MACROSCOPIC DIFFUSION
We see from Eqn. 1.8 that if is real ( R), then as long as D > 0 then
composition inhomogeneities will decay with time.
The behavior as a function of position is determined by the equation for
Z(z), which is:
d
2
Z
dz
2
+
2
Z = 0
which has solutions:
Z(z) = A

cos z + B

sin z
So our solution for the composition is:
c(z, t) = Z(z)T (t) = (Acos z + B sin z) e

2
Dt
(1.9)
where we have absorbed the factor T
0
into A and B.
In fact, we are free to choose any number of solutions of the form of
Eqn. 1.9, and we can write:
c(z, t) =

n=
(A
n
cos
n
z + B
n
sin
n
z) e

2
n
Dt
(1.10)
where A
0
is the average composition.
At time t = 0 we see that the initial composition prole is:
c(z, 0) =

n=
(A
n
cos
n
z + B
n
sin
n
z)
which is just the Fourier series expansion of the initial composition prole.
We can represent any periodic function with this series, and we can represent
any function on a limited range, say from L to L, by a periodic extension,
forming a function with period 2L. In either case, we nd that
n
= n/L
so we have:
c(z, 0) =

n=
_
A
n
cos
nz
L
+ B
n
sin
nz
L
_
(1.11)
We can use the standard Fourier analysis tricks to nd A
n
and B
n
, namely,
multiply both sides of Eqn. 1.11 by sin mz/L and integrate from L to L.
Due to the orthagonality of the sin functions we nd:
A
n
=
1
2L
_
L
L
c(z, 0) cos
nz
L
dz
1.5. SEPARATION OF VARIABLES 17
Using similar tricks we can nd:
B
n
=
1
2L
_
L
L
c(z, 0) sin
nz
L
dz.
The coecient A
0
is given by
A
0
=
1
2L
_
L
L
c(z, 0) dz
which is just the average of the initial composition.
Lets apply this formalism to a problem. We consider a periodic structure
as shown in Fig. 1.6, where the initial composition prole is a square wave,
made up of slabs of thickness L with composition alternating between c
0
and
0. The coecient A
0
is just the average:
A
0
=
1
2L
_
L
L
c(z, 0) dz =
c
0
2L
_
L
0
dz =
c
0
2
We integrate to nd the coecients A
n
and B
n
, and for A
n
we nd:
A
n
=
1
2L
_
L
L
c(z, 0) cos
nz
L
dz
=
c
0
2L
_
L
0
cos
nz
L
dz
=
c
0
2n
sin
nz
L

L
0
= 0
and for B
n
:
B
n
=
c
0
2L
_
L
0
sin
nz
L
dz
=
c
0
2n
cos
nz
L

L
0
=
c
0
2n
(1 cos n) =
_
c
0
n
n = odd
0 n = even
So we can nd the Fourier series representation for the initial composition
prole:
c(z, 0) =
c
0
2
+
2c
0

j=0
_
1
2j + 1
_
sin
2j + 1
L
z
18 CHAPTER 1. FICKS LAWS AND MACROSCOPIC DIFFUSION
c

c
L 2L 3L 4L 5L
z
Figure 1.6: Square composition wave.
The sum of the rst several terms is plotted in Fig. 1.7 a. Note that the
terms decrease in amplitude as the frequency increases.
It is a simple manner to include the time dependence by using Eqn. 1.10,
and we nd:
c(z, t) =
c
0
2
+
2c
0

j=0
_
1
2j + 1
_
sin
(2j + 1)z
L
exp
_
_
_

_
(2j + 1)
L
_
2
Dt
_
_
_
This is plotted in Fig. 1.8 for dierent times. Note that the higher order
terms go away fast. Even at a time as small as 0.01L
2
/D, which corresponds
to a diusion distance of L/100, the wiggles associated with high order terms
are gone. The prole quickly begins to look like a sine wave, as only the rst
order term is important for longer times. The physical reason behind this
behavior is clear. Higher order terms correspond to higher spatial frequencies,
which have higher concentration gradients and hence faster diusion.
We can quantify this behavior by examining the ratio of successive terms.
If we dene R
j,k
to be the magnitude of the k
th
term relative to that of the
j
th
term. We nd:
R
j,j+1
=
_
2j + 1
2(j + 1) + 1
_
exp
_
_
_
_

_
[2(j + 1) + 1]
L
_
2
+
_
(2j + 1)
L
_
2
_
_
Dt
_
_
=
_
2j + 1
2j + 3
_
exp
_
8(j + 1)

2
Dt
L
2
_
1.5. SEPARATION OF VARIABLES 19
-0.5
0.0
0.5
F
o
u
r
i
e
r

C
o
m
p
o
n
e
n
t
s
1 0
z/L
1.0
0.5
0.0
c
(
z
)
1 0
z/L
Figure 1.7: Plot of the rst several terms of the Fourier series expansion of
a square wave. a) The individual Fourier components. b) The sum of the
components.
20 CHAPTER 1. FICKS LAWS AND MACROSCOPIC DIFFUSION
1.0
0.5
0.0
c
(
z
)
2 1 0
z/L
t = 0
1.0
0.5
0.0
c
(
z
)
2 1 0
z/L
t = 0.01
1.0
0.5
0.0
c
(
z
)
2 1 0
z/L
t = 0.05
Figure 1.8: Plot of the rst twelve terms of the Fourier series expansion of a
square wave at dierent times. Time is in units of L
2
/D.