MSE307 Unit 2 November 2018 (A.T.

Paxton) Page 1 of 14

MSE307 Unit 2—Outline of Diffusion Theory

1. Fick’s first law

We will begin our discussion of diffusion in metals and alloys with the phenomenological
Fick’s first law,
dci
Ji = −Di (1.1)
dz
This relates directly to experience and states that the flux of moles of component i, Ji ,
in units of mole m−2 s−1 (the number of moles that pass through a unit area normal to
the flow per unit time) is directly proportional to minus the gradient in concentration.
Here, concentration, ci , is the concentration of component i in moles m−3 . In these notes
we will keep things simple and consider just one dimensional diffusion in the z-direction.
In full vector notation Fick’s first law is

Ji = −Di ∇ci

The constant of proportionality is the diffusion coefficient, or diffusivity. It has the

dimensions of m2 s−1 . These units are shared by the coefficients in the transport of heat
and momentum. The thermal diffusivity, κ, is the ratio of the thermal conductivity, k,
from Fourier’s law and the heat capacity per unit volume; the kinematic viscosity, ν,
is the viscosity, η, from Newton’s law divided by the mass density. All these transport
coefficients, D, κ and ν have the dimensions m2 s−1 .

2. Mobility

At a more fundamental level, we think of the speed of a diffusing particle as the product
of the force acting on the particle and its mobility. For example in semiconductor
materials science and electronic engineering we write for the drift velocity of a carrier
(electron or hole)
dφ
vd = µd E = µd (2.1)
dz
where φ is the electric potential, E is the electric field in V m−1 (or newton per coulomb,
N C−1 ) and µd is the mobility in C m2 s−1 J−1 . This formula would apply also for the
migration of an ion in a fuel cell. The mobility is proportional to the carrier’s charge.

You know from your thermodynamics notes (unit 2D) that fundamentally the driving
force for solid state diffusion is the gradient in chemical potential, not concentration.
So to begin with we are tempted to write down

dµi
Ji = −ci Mi
dz

with M having the units of mole m2 s−1 J−1 . The inclusion of the factor ci makes
sense for dimensional reasons: the units of µd for the case of the charged carrier are
MSE307 Unit 2 November 2018 ([Link]) Page 2 of 14

the same as those of M except that the charge unit is replaced by the mole unit—each
representing the amount of quantity that is responding to the force. Of course to arrive
at a more standard form it is sufficient to define a new mobility, M, by

Mi = ci Mi (2.2)

and then we have

dµi
Ji = −Mi (2.3)
dz
But M has the crazy units of mole2 m−2 s−1 J−1 . We would like to make contact between
the phenomenological Fick’s law (1.1) and the more physically motivated (2.3); in other
words what is the relation between diffusivity and mobility? Recall equation (3.1) from
unit 2D and write
µi = µ◦i + RT ln ai
= µ◦i + RT ln xi + RT ln γi
Then  
dµi d ln xi d ln γi
= RT +
dz dz dz
and so (2.3) can be written
 
d ln xi d ln γi
Ji = −RT Mi +
dz dz
 
d ln xi d ln γi d ln xi
= −RT Mi +
dz d ln xi dz
 
1 dxi d ln γi dxi
= −RT Mi +
xi dz d ln xi dz
 
1 d ln γi dci
= −RT Mi 1+ (2.4a)
ci d ln xi dz
 
d ln γi dci
= −RT Mi 1 + (2.4b)
d ln xi dz
If the solid solution is ideal or it follows either Raoult’s or Henry’s laws, then the activity
cofficient is either one or a constant and so its derivative with respect to anything is
zero and (2.4b) reduces simply to
dci
Ji = −RT Mi ←− Roultian or Henrian solid solution
dz
dci
= −Di
dz
This, or the more general equation (2.4), furnishes us with the relation between the
fundamental formula (2.3) and the phenomenological Fick’s law (1.1). Di is called the
intrinsic diffusion coefficient for the component i in a solid solution, and
 
d ln γi
Di = −RT Mi 1 +
d ln xi
= −RT Mi ←− Roultian or Henrian solid solution
 MSE307 Unit 2 November 2018 ([Link]) Page 3 of 14

3. Binary alloy—intrinsic and chemical diffusion coefficients

Further progress can be made in the case of a binary system. For a binary A-B alloy

dµA
JA = −MA
dz
dµB
JB = −MB
dz

To compare this with Fick’s first law, again I want a relation between the derivative of
µ and the derivative of c. I write, using the chain rule of differentiation, and recalling,
as I did in section 2, from the properties of the logarithm that d ln y = (1/y)dy,

dµA dµA dcA dµA 1 dcA

= =
dz dcA dz d ln cA cA dz

Then
MA dµA dcA
JA = −
cA d ln cA dz
and by comparison with (1.1) and (2.2) I find that the diffusion coefficient is

dµA
DA = MA
d ln cA

and similarly for component B. However the Gibbs-Duhem relation for a binary alloy
leads to†
dµA dµB
=
d ln cA d ln cB
† For a binary alloy, c + c is a constant, so we have dc = −dc . By the definition of the logarithm
A B A B

1 1
d ln cA = dcA d ln cB = dcB
cA cB

From your thermodynamics notes, equation (2.11), the Gibbs-Duhem equation for a binary alloy reads

nA dµA = −nB dµB

and if I divide through by the volume of the body under discussion then

cA dµA = −cB dµB

and this can be written using the previous identites for the binary alloy,

dµA dµB
dcA = − dcB
d ln cA d ln cB

and since dcA = −dcB ,

dµA dµB
=
d ln cA d ln cB
MSE307 Unit 2 November 2018 ([Link]) Page 4 of 14

and so

dµA
DA = MA (3.1a)
d ln cA
dµA
DB = MB (3.1b)
d ln cA

These are the intrinsic diffusion coefficients for the binary alloy. This means that intrinsic
diffusivities in a binary alloy differ only due to the differences in mobility.

The total flux of atoms in a binary alloy is J = JA + JB . This is relative to the crystal
lattice. Since there may also be vacancy flow then the flux of A atoms relative to
the centre of mass of the body is JA take away the total J times the atom fraction
xA = 1 − xB ; considering equations (3.1),

JAC.O.M. = JA − xA (JA + JB ) (3.2)

= xB JA − xA JB
dcA
= − (xB DA + xA DB )
dz
dcA
= −Dchem
dz

and
Dchem = xB DA + xA DB (3.3)

is called the chemical diffusion coefficient, that is the sum of the intrinsic diffusivities
each weighted by the other’s concentration. This exposes very nicely that in general
the diffusion coefficient is a function of concentration and should not be treated as a
constant in Fick’s first and second laws, which consequently should be written

J = −∇ (Dc) First law

∂c
= ∇ · (D∇c) Second law
∂t

If all diffusion is by place exchange and not by vacancy exchange then DA = DB =

Dchem . The chemical diffusion coefficient is that which is most usually measured in an
experiment. For example if a piece of pure copper and a piece of pure nickel are placed
in contact with a flat and smooth interface then after placing in a furnace for some
hours the two elements will be found to have inter-diffused. This is diffusion bonding.
By measuring the amount of Cu on the Ni side and the amount of Ni on the Cu side
at a distance z from the original interface as a function of time and temperature the
diffusivity is measured. Since the two intrinsic diffusivities may be different (3.1), there
may be a total flux of vacancies and so markers, such as rods of an inert metal, may be
found to have moved relative to the ends of the specimen. This is called the Kirkendall
effect and was used in the first instance to demonstrate that vacancies are involved in
diffusion.
 MSE307 Unit 2 November 2018 ([Link]) Page 5 of 14

4. Binary alloy—tracer diffusion coefficient

If there is only one component, that is, if the metal is pure then we are interested in the
tracer diffusion coefficient. The way this is usually measured is by a radioactive isotope
tracer experiment—hence the name. If we use the subscript A for the metal atoms and
a subscript A• for the isotope then the fluxes of the two types of atoms are

dµA dcA
JA = MA
d ln cA dz
dµA• dcA•
JA• = MA •
d ln cA• dz

The atom A and its isotope are chemically identical and so their chemical potentials will
be the same except for differences arising from the different entropies of mixing (due to
their not occuring in the same concentration). So if we put everything except for the
entropy of mixing into the standard state µ◦ we are left with, recalling equation (3.4)
from your thermodynamics notes,

cA cA •
µA = µ◦ + RT ln and µA• = µ◦ + RT ln
c c

where c = cA + cA• , which means that

dµA dµA•
= = RT
d ln cA d ln cA•

The diffusion currents must be equal and opposite and on physical grounds we expect
the mobilities of the atom and its isotope to be identical. Therefore if we set MA• = MA
we see from (3.1) that
DA• = RT MA (4.1)

This relation between tracer diffusion coeffient and mobility is very famous indeed as it
was first written down by Albert Einstein in 1905. Finally for a binary alloy we note
that we have found the relation between the intrinsic and the tracer diffusion coefficients
for components A and B. In view of (3.1)

DB DA 1 dµA
= = (4.2)
DB • DA• RT d ln cA

This is known as Darken’s thermodynamic relation.

5. Relation between chemical and tracer diffusion coefficients in a binary alloy

So far we have defined three diffusivities: the intrinsic, the chemical and the tracer
diffusion coefficient. (There is one more—the so called self diffusivity but we’ll come
to that later.) Darken’s relation (4.2) expresses the connection between intrinsic and
tracer diffusivities. We can also find a relation between chemical and tracer diffusion
 MSE307 Unit 2 November 2018 ([Link]) Page 6 of 14

coefficients in a binary alloy. We defined Dchem in equation (3.3). We can also use the
Gibbs-Duhem relation to show that†
d ln γA d ln γB
= (5.1)
dxA dxB

If I now combine (2.4) and (2.2) with (4.1) and (5.1) if follows very easily that

Dchem = xB DA + xA DB
   
d ln γA d ln γB
= xB D A • 1 + + xA DB • 1 +
d ln xA d ln xB
 
d ln γA
= (xB DA• + xA DB • ) 1 +
d ln xA

This is a very nice connection indeed as it relates the diffusivity as measured using a
diffusion couple and which incorporates the Kirkendall effect with the tracer diffusion
coefficients which describe the trajectories of atoms in the pure substance and finally
which includes a term to account for non ideality of the solid solution. If a solid solution
is ideal (or obeys either Raoult’s or Henry’s law) and if there is only diffusion by place
exchange (that is, there is no vacancy flux) then

Dchem = DA = DB = DA• = DB •

and all diffusion coefficients are numerically equal. This would be the case for example
in the diffusion of nitrogen or carbon in ferrite or in reasonably low concentration in
austenite. It should also hold for hydrogen or oxygen diffusivity in titanium alloys and
many other cases of the diffusion of interstitial impurities.

6. Atomistic aspects

At the atomic scale diffusion is thought to take place either by a vacancy exchange, in
the case of substitutional elements, or by the movement of interstitials. In the first case
there is a severe restriction that for a particular atom to be able to jump it must have
at least one neighbouring site vacant, while in the case of interstitials, as long as these
are sufficiently dilute then we can assume that each neighbouring site is vacant. This
means that for interstitial diffusion the rate is principally determined by the probability
that the atom will jump over the activation barrier; while in the substitutional case this
probability must, roughly speaking, be multiplied by the probability that the site in
†
 
d ln γA
xA dµA = RT (dxA + xA d ln γA ) = RT 1+ dxA
d ln xA
which when put into the Gibbs-Duhem equation becomes
   
d ln γA d ln γB
RT 1 + dxA = RT 1 + dxB
d ln xA d ln xB

and since for a binary alloy dxA = −dxB , (5.1) is proved.

MSE307 Unit 2 November 2018 ([Link]) Page 7 of 14

the direction of jumping is vacant. This second probability is actually extremely small
in metals given that even near the melting point the vacancy concentration is usually
smaller than 10−5 . This is why diffusion of substitutional elements, such as Mn, Ni, Mo,
in steel is orders of magnitude smaller than interstititials such as C or N. The diffusion
of hydrogen in ferrite is probably the fastest known in the solid state. However there
are more exotic possible processes for diffusion that have been proposed, illustrated in
figure 1. Apart from those already discussed there are direct exchange and ring diffusion
and diffusion via substitional or host atoms temporarily occupying interstitial positions;
this is usually regarded as rare since the formation enthalpy of an interstitial defect is
very large—much larger than the vacancy formation energy. But an interstitial may
have quite a low formation energy if one atom is placed between two atoms belonging to
a row of atoms in a close packed lattice direction: the row of atoms can all move a little
to accommodate the interstitial. This row of atoms containing one additional atom is
called a “crowdion”. Crowdions in the [111] direction in the bcc lattice can move rather
easily—but only in the particular h111i direction, the crowdion cannot turn corners!

Figure 1: Cartoon illustrating possible diffusion mechanisms at the

atomic scale. (a) Interstitial solid solution, (b) place exchange and
ring diffusion (no vacancies needed), (c) diffusion of substitutional
atom by an interstitialcy mechanism (the lower inclined line shows
a crowdion), (d) vacancy diffusion.
 MSE307 Unit 2 November 2018 ([Link]) Page 8 of 14

6.1 Equilibrium concentration of vacancies

Since the probability that a substitutional site is vacant is central to the diffusivity we’ll
need to calculate it. It is the same as the fraction of sites that is vacant, namely the
equilibrium concentration of vacancies. I can create a vacancy by removing an atom
from the interior of a crystal and placing it on the surface.† Suppose the crystal has NL
lattice sites of which NV are vacancies. Since the vacancies can be arranged in a large
number of ways, whereas there is only one distinguishable way to arrange the atoms in
a perfect lattice, the configurational entropy is increased from zero to Sc . We calculate
this from Boltzmann’s famous formula (it’s engraved on his tombstone),

Sc = k ln W

k is the Boltzmann constant and W is the number of ways of arranging NV vacancies

among NL sites,
NL !
W =
NV ! (NL − NV )!
As usual we use Sterling’s approximation for ln x when x is very large. For us it’s on
the order of the Avogadro number so it’s very large indeed, so this is not really an
approximation:

ln W = NL ln NL − NV ln NV − (NL − NV ) ln (NL − NV )

The mole fraction of vacancies is

NV
x=
NL
so that
NL − NV
1−x=
NL
Using the last three equations and after some algebra we get

ln W = −NL (x ln x + (1 − x) ln (1 − x))

Then the configurational entropy of the crystal containing NV vacancies is

Sc = k ln W = −kNL (x ln x + (1 − x) ln (1 − x))

This is not the total change in entropy. When I remove the atom this changes the
vibrational frequencies of the surrounding atoms and so it changes the total vibrational
entropy, Sν (also called thermal entropy) of the crystal. Assuming that making further
vacancies simply increases Sν additively (that is, that the change in vibrational fre-
quencies is local) then the total increase in entropy due to the creation of NV vacancies
is
∆S = NV ∆Sν + Sc
† This need not affect the surface energy since I can place the atom at a kink along a terrace edge—this
does’t change the surface energy.
MSE307 Unit 2 November 2018 ([Link]) Page 9 of 14

There is also a change in enthalpy at the creation of each vacancy. At the simplest
possible level this is because, say, z bonds are broken when I wrest the atom from the
interior, but some fewer than z (roughly z/2) bonds are remade when I attach it to the
surface. Hence in the zeroth approximation the enthalpy of formation of one vacancy,
∆H is roughly the cohesive energy per atom of the pure solid. The change in free
enthalpy on the introduction of NV vacancies is therefore

∆Gcryst. = NV ∆H − T ∆S
= NV ∆H − T NV ∆Sν − T Sc
= NV ∆G0V + kT NL (x ln x + (1 − x) ln (1 − x))

where ∆G0V = ∆H − T ∆Sν is the total change in free enthalpy per vacancy having
taken away the configurational entropy contribution. I divide through by NL and I get
the change in free enthalpy per lattice site

∆Gcryst.
= ∆GL = x∆G0V + kT (x ln x + (1 − x) ln (1 − x))
NL

I can find the equilibrium concentration of vacancies, xeq , by finding the value of x that
minimises ∆GL . I take the derivative of ∆GL with respect to x and set the result to
zero. After rearranging the result I get

xeq 0

eq
= e−∆GV /kT = e∆Sν /k e−∆H/kT ≈ xeq
1−x

The last approximation follows because you find in any metal below the melting point
that 1 − xeq ≈ 1.

You can use this procedure to find the equilibrium concentration of any point defect. If
you care to ignore the, possibly small, change in vibrational entropy on the creation of
the defect and in addition neglect the ∆(pV ) term in the enthalpy then you have

xeq ≈ e−∆U/kT

and these days you can calculate ∆U using first principles quantum mechanics.

6.2 Random walk

Even if there is no gradient in chemical potential the atoms are still jumping from site
to site in a solid under the influence of thermal excitation and the collective action of
phonon wavepackets. The simplest example is a dilute interstitial impurity all of whose
z neighbouring interstices are empty. It can choose to jump to any one at random and
it will jump on average at a rate of Γ jumps per unit time. In any case, including the
case of vacancy or other mechanisms we expect that we can assert that

Γ = P zν
MSE307 Unit 2 November 2018 ([Link]) Page 10 of 14

where P is the probability that the site in the direction of the jump is vacant, z is
the coordination number or number of neighbours, and ν is a rate coefficient for the
activated jump process: it is the frequency with which an atom makes a unit jump (see
section 7, below).

If an atom starts out at the origin at time t = 0 and it makes n jumps, and if at the j th
jump it moves a vector distance sj then after the n jumps it has arrived at a position
vector from the origin given by
Xn
S= sj
j=1

The square of the distance that it has jumped is S 2 = S · S. This is

n n−j
n−1 X
X X
2
S = sj · sj + 2 si · si+j
j=1 j=1 i=1
n n−j
n−1 X
X X
= s2j +2 si si+j cos θi,i+j
j=1 j=1 i=1

where θi,i+j is the angle between the vectors belonging to the ith and the (i + j)th jump.
This formula makes no approximations concerning the randomness of jumps, lengths or
angles between jumps, or even the dimension (whether this is 3D, 2D or 1D movement).
In the more usual cases, certainly in fcc metals and almost certainly in bcc metals,
and certainly for interstitial jumping in these metals, the jump distances are all equal,
call it the distance a. (You would have to be careful in hcp metals.) In those special
circumstances, and we will assume such cases from now on)

n−j
n−1 X
X
S 2 = na2 + 2a2 cos θi,i+j
j=1 i=1

We now take a statistical average over the trajectories of many atoms (some small
fraction of an Avogadro number in a typical diffusion experiment) and we write,

n−j
n−1 X
X
2 2
S2 = na + 2a cos θi,i+j (6.1.1)
j=1 i=1

where the overline indicates an average, and cos θi,i+j is the average value of the cosine
of the angle between the directions of the ith and j th jumps. This is a very important
term because it embodies all the correlations that are present in the jumping process.
The trajectory of an atom is not usually even approximately a random walk. This
is particularly true for the jumping of substitutional impurities, since these impurities
will have different interaction preferences with vacancies than the host atoms. Also in
general for the case of vacancy diffusion: if an atom exchanges places with a vacancy, at
the next jump by far the most likely probability that it will reverse that jump because
MSE307 Unit 2 November 2018 ([Link]) Page 11 of 14

that is whre the nearest vacancy is. However if the direction of each jump is selected
entirely randomly then cos θj = 0 and

S 2 = na2 (6.1.2)

According to a kinetic argument for a cubic lattice the diffusivity is

1 2
D= S (6.1.3)
6t
(an argument first put forward by Einstein in 1906). After n jumps in a time t the rate
of jumping must have been Γ = n/t, or t = n/Γ and so for the diffusivity we have

1 2 1
D= Γ a = P zνa2 (6.1.4)
6 6
This truly only applies to interstitial dilute impurity diffusion in fcc and bcc metals, for
which we will have P = 1, z = 12 for fcc and z = 4 in bcc metals. ν is unknown and can
only be measured indirectly, but it is possible nowadays to calculate rate coefficients to
good precision using first principles quantum mechanics—even in the case of hydrogen
in steel where quantum mechanical tunelling is involved.

6.2 Non random walk

We should modify the equation (6.1.4) for the diffusivity to account for the correlation
and other effects neglected thus far. A sensible modification is to write,
1
D= f (sa0 )2 νP P 0 jz (6.2.1)
6
Here, a0 is the lattice parameter and sa0 is the distance actually jumped in the direction
of atom flow. P as before is the probability that the site in the direction of travel is
available for a jump, i.e. vacant, and P 0 is the probability that the atom is residing
on its site just prior to the jump. ν is the rate coefficient, or the number of events per
unit time, for the atom and vacancy to exchange places. z is the atom’s coordination
number and j is the fraction of its neighbours which are placed in the intended direction
of travel. Finally, f is the “correlation factor” which account for the non randomness of
the walk. We’ll come back to how to work out f shortly. First let’s look at the example
of vacancy diffusion in a pure fcc metal. Consider diffusion along a [001] direction:
atoms will jump a distance a0 /2 between successive (002) planes so s = 1/2. We assume
all sites are occupied so P 0 = 1 and the probability that a site in the next (200) plane
is vacant is xeq as we discussed in section 6.1. z = 12 in the fcc lattice but only a third
of the neighbours are in the next (002) plane and so j = 1/3. The diffusivity is hence
 2
1 1 12 1
D= f a0 νxeq = f a20 xeq ν
6 2 3 6

A useful general definition of f comes about in this way. Consider the difference between
the trajectories of a vacancy and of an atom that has been harmlessly tagged in some
MSE307 Unit 2 November 2018 ([Link]) Page 12 of 14

way. The ornithologist tags a bird by attaching a ring to its leg and hopes that this
will in no way affect its behaviour. We could tag an atom by replacing its nucleus
with an isotope of the same metal. Its diffusivity is DA• as we saw in section 4. The
trajectory of a vacancy, at least in a pure metal, is truly a random walk—its diffusivity
is called the “self diffusivity” Dself , of the metal. Self diffusivity is what is measured in
gasses and liquids, or in Brownian motion, in which each particle or molecule, before
jumping, has no memory of where it came from and chooses each available direction in
which to jump with equal probability. The same applies to dilute interstitials in metals
as we have seen. Conversely a substitutional atom is very biased in choosing the next
direction in which to jump: the greatest likelihood is that it will jump straight back
again into the site it just vacated; but occasionally that vacant site will become occupied
by another of its neighbours and our atom will have to stay put until another vacancy
drifts into a neighbouring position. So the tracer diffusivity is essentially different to
the self diffusivity in vacancy diffusion and we can define the correlation factor as

DA• S 2 (A• )
f= = lim n
Dself n→∞ S 2 (vac.)
n

the ratio of the mean square displacements of the tracer and vacancy after n jumps.
Since Sn2 (vac.) = na2 , equation (6.1.3), we have
n−j
n−1 X
X
f =1+ cos θi,i+j
j=1 i=1

which follows from (6.1.1). Actually the memory is not that long term: the atom only
remembers its previous jump and so the direction of the next jump is only correlated
with the immediately preceding jump. Put another way the geometry of the problem
concerns tracer-vacancy pairs and these are all identical except for their orientation.
In that case cos θi,i+j is independent of the index i and we define cos θj as the average
value of the cosine of the angle between the directions of the ith and j th jumps and is
independent of i. In fact
j
cos θj = cos θj−1 cos θ1 = (cos θ1 )j = (cos θ1 )

so it’s just the angle of the first jump raised to the power j. This leads to
n−1
!
X j
f = lim 1 + (n − j)(cos θ)
n→∞
j=1

and θ is the angle between successive jumps as there’s nothing special about the first
jump. The limit can be taken and the series summed analytically. The result is

1 + cos θ
f=
1 − cos θ
The calculation of f is dreary and difficult and involves some probability theory. A
number of people have kindly done this for us and found that correlation factors for fcc
 MSE307 Unit 2 November 2018 ([Link]) Page 13 of 14

and bcc pure metals are 0.781 and 0.727 respectively. So they are not very close to one
and cannot really be ignored.

7. Rate theory aspects

I’ll take you through a simplified and not very rigorous “derivation” of the well known
formula for the diffusivity,
D = D0 e−Q/kT
We want to know the frequency, ν, at which an atom will exchange places with a vacancy.
Put another way, if an atom and a vacancy are neighbours, what on average is the time
τ = 1/ν that I have to wait before the atom and the vacancy change places? This
is a standard question in materials science and chemistry. The event is an “activated
process”. Think of the atom as it is dragged from its lattice and taken adiabatically
into the empty neighbouring site. During this procedure the total energy of the crystal
must increase until a maximum is reached, probably when the atom is half way along
its journey. This arrangement of atoms is called an activated complex in chemistry;
the configuration of all the atoms is at a saddle point in the energy landscape. At any
point in time and at any temperature in a crystal of metal that has an equilibrium
concentration, xeq , of vacancies, there will also be an equilibrium concentration of these
activated complexes. By exactly the same argument that we used in section 6.1 we can
take it that the equilibrium concentration of these is
0
e−∆GM /kT
where, by analogy with ∆G0V , ∆G0M is the free enthalpy of formation of one activated
complex take away its configurational entropy contribution. So the frequency of jumping
is this concentration times the frequency, ν0 , at which an atoms attempts the jump.
Approximately speaking we take it that ν0 is the vibrational frequency of the collection
of atoms in that particular mode of vibration which drives the atom in the direction of
the saddle point configuration.† What’s causing this to happen is the system of phonons
whose amplitude depends on temperature. This gives us
0
ν = ν0 e−∆GM /kT
which we can put into our formula (6.2.1)
1 2
D=
f a νP jz (6.2.1)
6
along with the equilibrium vacancy concentration that we obtained in section 6.2,
0
P = xeq = e∆GV /kT
This results in
1 1 0 0
D= f jza2 xeq ν = f jza2 ν0 e−(∆GM +∆GV )/kT
6 6
1 M V M V
= f jza2 ν0 e(∆Sν +∆Sν )/k e−(∆H +∆H )/kT
6
= D0 e−Q/kT
† Statistical mechanics arguments can be used show that ν = kT /h, where h is the Planck constant.
0
MSE307 Unit 2 November 2018 ([Link]) Page 14 of 14

which is the desired result. Here, ∆SνM and ∆SνV are respectively the vibrational en-
tropies of formation of the actvated complex and an isolated vacancy: these reflect
changes in the phonon spectra when these defects are created. Similarly, ∆H M and
∆H V are the respective enthalpies of formation.

1 M V
D0 = f jza20 ν0 e(∆Sν +∆Sν )/k
6
is defined by the above formula; note that it has the expected dimensions of a transport
coefficient: length squared over time, or m2 s−1 .

Q = ∆H M + ∆H V is called the “activation energy” and note that is is an enthalpy,

not a free enthalpy—the configurational entropy is already incuded in ν as we saw in
section 6.1, and the remaining entropy terms are included in the D0 “prefactor”. The
activation energy is relatively easy to calculate using first principles quantum mechanics.
∆H M can be calculated using a computational technique called the “nudged elastic
band” in which we drag the atomic configuration up the saddle point as it were with
an elastic band which confines images of the system to the reaction coordinate. Total
energies are calculated using the density functional theory. If we are happy to neglect
thermal entropy terms then we can find a reasonable estimate of the diffusivity. On the
other hand the whole estimation can be done using a computer method called molecular
dynamics; this will in principle include entropy and finite temperature terms, but it has
other attendant problems, mostly connected with not having fast enough computers and
thereby having to throw away the proper quantum mechanical theory in favour of very
approximate so called “classical potentials.”

Sources
1. J. W. Christian, The theory of transformations in metals and alloys, Third edition
(Pergamon, 2002)