CHAPTER 1

Theory Of Ensembles

1.1. Fundamental hypothesis

We will consider a general macroscopic system with Hamiltonian

H. Suppose the energy eigenvalue is known to be in (E, E + ∆E) ,
the momenta is known to be in (P , P + ∆P ) and some other quantum
numbers may also be known. We would still have many states (let
us say Ω in number) satisfying these constraints. The first question is,
what is the probability that the system is in one particular state among
the Ω ?
The fundamental hypothesis is that the a priori probability is Ω1 .
This is the only natural hypothesis that can be made at this stage, and
the only one we will need to make.

1.2. Canonical Ensemble

Consider a large system, containing N interacting particles. The

Hamiltonian may be, for example
X p2 X
i
H= + Vij
i
2mi i<j

for a system of nonrelativistic particles interacting via two body forces.

Suppose the system is in thermal equillibrium with a large reservoir.
We consider a large number M of replicas of our system. This is the
ensemble. Individual systems can exchange heat energy only (not par-
ticles).
Total Hamiltonian for the ensemble is

M
X
Hens = Hk + ET h
k=1
where ET h is the thermal interaction energy. We will assume that
ET h << Hk . This is what we mean by a large system. Thus, the total
1
2 1. THEORY OF ENSEMBLES

wavefunction is a product
Y
Ψ= ψk
k

where ψk is the wavefunction of the kth system. Suppose the eigenvalue

E of Hens is known. Since the whole ensemble is an isolated system,
this energy is in fact conserved. If the individual ψk are known, then
the wave function of the entire ensemble Ψ is also known.
Note on the use of Ensembles :
The use of ensembles is purely for technical conve-
nience. Analogous operational techniques are used in
determining (indeed in defining) probability distributions
in general statistical analysis. For example, if one tosses
a coin and tries to learn the probability of getting, say,
a head, then for a finite number of trials, the relative
frequency will fluctuate quite a bit. To get a limiting
relative frequency one has tor repeat the experiment a
large number of times (theoretically an infinite number
of times).One would then define this limiting relative
frequency as “the probability”. In our case the ensem-
ble represents the set of all trials. It’s like using a very
large number of coins and flipping them all together. The
probability we would get in this way would, of course,be
the same as if we were using a single coin and repeating
the observation a large number of times.
All members of a canonical ensemble can exchange
energy with a heat bath. Since it does not really matter
in the least what the bath is made up of, for a particular
member system we will use all the others together as the
heat bath. This in fact is the procedure we would always
adopt.
Suppose, of the entire number M of systems that make up the ensemble,
P
Mk are in the quantum state k. Then we must have k Mk = M and
P
k Mk Ek = E , where Ek is the energy eigenvalue for an individual
system in the state k .
Now we ask the following question - given the set {Mk } , how many
states of the ensemble are possible? Since each system, being itself
 1.2. CANONICAL ENSEMBLE 3

macroscopic, is distinguishable, this number is clearly

M!
Ω({Mk }) = Q
k Mk !

. What, is the probabilty that the systems in the ensemble are dis-
tributed among their various states accoring to the set {Mk } ? By
our fundamental hypothesis, the a-priori probability we are looking
for is ∝ Ω({Mk }) . Which, then, is the most probable distribution ?
This is the distribution for which we have maximum Ω({Mk }) , with
P P
k Mk = M and k Mk Ek = E.
Using lagrange multipliers, the condition we are looking for becomes
∂ ln Ω ∂ X ∂ X
−α Mk − β Mk Ek = 0
∂Mj ∂Mj ∂Mj
where α and β are lagrange multipliers. We will make use of Stirling’s
approximation for the factorial
 N
√
 
N 1
N ! −→N large = 2πN 1 + + ···
e 12N
. In fact for our purposes it will suffice to consider the simpler expres-

N
sion N ! −→N large = Ne since M as well as Mj are large. (In principle
they can be as large as we please, we just have to consider a big enough
ensemble!). Thus we have
X X
ln Ω = M ln M − M − Mk ln Mk + Mk

and so the maximisation condition leads to

− ln M̃j − α − βEj = 0

and thus
M̃j = e−(α+βEj )
for the most probable distribution. Here, α and β are to be obtained
from the constraints. For this distribution the average energy per sys-
tem E can be obtained from
X
E= M̃j Ej = M E

and hence X
E= Pj Ej
j
4 1. THEORY OF ENSEMBLES

where
M̃j 1
Pj = = e−βEj
M Q
is the probabilty that a paricular system is in state j. Here we have
defined the normalisation constant Q as
X
Q= e−βEj
j

. If we fix E , then Pj is independent of M, provided the latter is large.

This is the so called law of large numbers.

Note on Ergodic Hypothesis :

Some times one argues (but as a matter of fact more
or less postulates) that the macroscopic observables, like
pressure, temperature etc. are time averages, rather than
ensemble averages.One then has the burden of proving
the existense of the time average as the observation time
t → ∞ and the equivalence of the two averages. This is
known as the ergodic problem of statistical mechanics.
However, we should realize that as far as we are con-
cerned, the proof of the pudding lies in being able to
derive thermodynamics. So, as long as we have a consis-
tent way of deriving themodynamics, we might look upon
the ergodic problem as a mathematically nice but com-
pletely irrelevant (at least for the purposes of statistical
mechanics) problem.

We will now show that for fixed N and E , the most probable distribu-
tion becomes sharper and sharper as M → ∞ and in the limit becomes
“the distribution”, so to speak. To this end, let us define
X X
f = ln Ω − α Mj − β Mj Ej
∂f
. ∂Mj
= 0 is the extremum condition that we solved to ob-
Mj =M̃j
∂2f
tain the most probable distribution M̃j . The second derivative ∂Mj2
=
− M1j < 0 . So we have really found the maximum.
X X
f˜ = ln Ω̃ − α M̃j − β M̃j Ej
 CONNECTION WITH THERMODYNAMICS 5

is the value of f when Mj = M̃j . Expanding f about Mj = M̃j , we

get
X1 1 1
f = f˜ − (Mj − M̃j )2 + O( 2 )
j
2 Pj M M
and hence
Y − 2P1 M (Mj −M̃j )2
Ω = Ω̃ e j

j
which is the gaussian distribution. From this it is obvious that the
mean value of the number of systems in the state j, Mj equals the
most probable number M̃j . Let us calculate the variance in Mj :
v
u M j − Mj 2
u 2
1
t
2 = √ → 0 as M → ∞
Mj Pj M

The higher moments also vanish as M → ∞ (as a matter of fact even

more rapidly!).
So as M → ∞, the most probable distribution becomes “the distri-
bution”.
1
Pj = e−βEj
Q
is the canonical distribution. Here β is independent of the particular
system considered, so it must depend on the one property of the en-
semble shared by all of it’s members - the temperature. Indeed, one
could even define β as the temperature.

Connection with thermodynamics

1
To make the usual connection we let β = kT , which is really a
definition of T . Here k is the Boltzmann constant. The normalization
constant X
Q= e−βEj
j
is called the partition function and as we shall see, plays a central role
in the theory. A glimpse at it’s utility can be had by considering the
expression for average energy :
X 1 X ∂
E= Pj Ej = Ej e−βEj = − ln Q
j
Q j ∂β
6 1. THEORY OF ENSEMBLES

which we identify with the themodynamic energy E of the system,

E = E.
Let us define the quantity F by

F = −kT ln Q

which we will soon identify with the thermodynamic Helmholtz free en-
∂ F


ergy, FTh . the energy E is seen to be related to F by E = −kT 2 ∂T kT V,N
.
Here the differentiation with respect to T is carried out while keep-
ing the energy eigenvalues Ej constant, hence the restriction to fixed
V and N . The thermodynamic free energy is related to the internal
energy by
FTh = E − T S
and hence
(dFTh )V,N = −SdT
and so  
∂
2 FTh
−kT = FTh + T S = E
∂T kT V,N

and the identification E = E is seen to lead to

 
∂ F − FTh
=0
∂T kT V,N

. Thus the quantity F −F

kT
Th
is independent of the temperature at fixed
V and N . Now as T → 0, the partition function Q tends to ω0 e−βE0
where E0 and ω0 are the energy and degeneracy of the ground state
respectively leading to the limit

F → E0 − kT ln ω0

On the other hand, in the same limit

FTh → E0 − T S0

where S0 is the entropy at zero temperature. Thermodynamics can not

determine S0 - indeed, it can do without it ! So, we take the liberty of
defining :
S0 = k ln ω0
Then, F − FTh → 0 as T → 0 , and hence

(1.2.1) F = FTh
 CONNECTION WITH THERMODYNAMICS 7

at all temperatures and the identification is almost complete. We say

almost, since from the above we can not rule out the possibility that

F = FTh + kT f (V )

where f (V ) is an arbitrary function of V . For a more complete proof

of the identification, (??) we need to demonstrate that the quantity
F −FTh
kT
is also independent of the volume at fixed particle number.
Now, thermodynamics tells us that

dFTh = dE − T dS − SdT = −pdV − SdT

and thus,  
∂FTh
p=−
∂V T
while, on the other hand :
   
∂F kT ∂Z
− =
∂V T Z ∂V T
1 X dEj −βEj
= − e
Z j dV
X dEj
= − Pj
j
dV
 
dEj
= −
dV
When the system is in the j-th energy eigenstate a small quasi-static
change in the volume dV will produce a change in the energy

dEj = −pj dV

where pj is the mechanical pressure corresponding to this state.Thus

 
∂F
− = hpj i
∂V T
However, the ensemble average hpj i is just the thermodynamic pressure
p, that is  
∂F
− =p
∂V T
leading to  
∂ F − FTh
=0
∂V kT T,N
8 1. THEORY OF ENSEMBLES

F −FTh
Thus kT
= C is a constant independent of V and T .

F = FTh + CkT

If we use our F instead of FTh in computing the thermodynamic quan-

tities, we see that only S gets an additional constant contribution −Ck
. Other quantities, like E, P and H = E + P V remain the same.
So, in order to do thermodynamics all we really need is the relation

F = −kT ln Tr e−βH

Thus, as far as thermodynamics is concerned, we do not need all matrix

elements of the Hamiltonian H. This is a tremendous simplification,
particularly for large macroscopic systems where the number of parti-
cles is of the order of 1023 !
P
Remark 1.2.1. It’s only because we want Hi  ETh , that we
want each subsystem to be large. They should be large enough to
ensure the above condition.

Remark 1.2.2. The number of subsystems in the ensemble M has

M
to tend to ∞. This is indeed so that the frequency of the jth state Mj
tends to a finite limit. The fluctuation
Mj2 − hMj i2
→0
hMj i2
as M → ∞ .

Fluctuation in a subsystem (with N finite) at constant

temperature

We define energy fluctuation by

X
(∆E)2 = Pj (Ej − E)2
j
P
where E = Pj Ej is the average energy. We have
j
X X X
(∆E)2 = Pj Ej2 − 2E Pj Ej + E 2 Pj
j j
X
= Pj Ej2 − E 2
j

= E 2 − hEi2
 APPLICATION TO BLACKBODY RADIATION 9

where we have used the definition of the average energy E as well as

P
j Pj = 1.
From

X −βEj
Q = Tr e−βH = e
j
we have
∂ 1 X X
− ln Q = Ej e−βEj = Ej Pj = E
∂β Q j
and
 2 !2
∂ 1 X 2 −βEj 1 X
− ln Q = Ej e − 2 Ej e−βEj
∂β Q j Q j
!2
X X
= Ej2 Pj − Ej Pj
j j
2
= (∆E)

Thus
∂E ∂E
(∆E)2 = − = kT 2
∂β ∂T
This calculation has been done at fixed Ej , hence we have held N and
V fixed. Thus
 
2 2 ∂E
(∆E) = kT = kT 2 CV
∂T N,V
where CV is the heat capacity at constant volume. The relative fluctu-
ation is
∆E 1p 2
= kT CV
E E
Now both E and CV are extensive quantities, E ∝ N and CV ∝ N and
this leads to
∆E 1
∝√
E N
Thus, the larger the system, the smaller the relative fluctuation is going
to be. Using the same technique, we may study fluctuations in other
quantities which depend on the energy.

Application to blackbody radiation

Let us consider a box at temperature T , with linear dimension L,

filled with photons. For this section, we are going to use natural units
c=~=1.
10 1. THEORY OF ENSEMBLES

The momentum eigenvalues of the photons are

~k = 2π~ (m1 , m2 , m3 )
L
with m1 , m2 , m3 integers. Energy eigenvalues are then E = ~k . To
specify the state of the photon completely, we also need to know the
helicity of the photon, which can take the two values of ±1.
We first find out the energy eigenvalues for the whole system. For
this, we neglect photon-photon interactions, i.e. we assume that the
photons move around independently.
X
E= n~k,λ ωk ~
~k,λ

where ωk = c ~k . Here n~k,λ is the number of photons with momentum

~k and helicity λ. The partition function for this system is :
X X h −β P n ω ~ i
Q = e−βEj = e ~ k,λ k
k,λ ~

j {n~k,λ }
X Yh i
= e−βn~k,λ ωk ~
{n~k,λ } ~k,λ
Y X
= e−β~ωk n~k,λ
~k,λ {n~k,λ }
Y 1
=
1 − e−βωk ~
~k,λ

Hence, X
ln 1 − e−βωk ~


ln Q = −
~k,λ

Thus, the average or thermodynamic energy is given by

∂ X ωk e−β~ωk
E=− ln Q =
∂β 1 − e−β~ωk
~k,λ

As the dimensions of the system increase, the discrete momentum

eigenvalues become continuous , and thus the sum on allowed values of
~k can be reduded to an integral
Z
X X 2V
→2 → 3
d3 k
m ,m ,m
(2π~)
~k,λ 1 2 3
 1.3. GRAND CANONICAL ENSEMBLE 11

where we have used the fact that the number of states between mo-
L
mentum component ki and ki + dki is given by ∆mi = 2π~ ∆ki . Thus
the energy is given by
Z
2V ωk ~
E= 3
d3 k β~ω
(2π~) e k −1
- this is the Planck formula. Since probabilities of independent events
are multiplicative, the quantity e−nβωk is proprtional to the probability
of having n photons with energy ω and a given helicity. Thus the mean
number of photons with given momentum and helicity is given by
P∞
D E ne−nβ~ωk
n~k,λ = Pn=0 ∞ −nβ~ωk
n=0 e
"∞ #
∂ X
= − ln e−nβ~ωk
∂(β~ωk ) n=0
1
=
eβ~ωk − 1
Othe thermodynamic functions can also be calculated. For example,
the free energy is
Z
2V
d3 k ln 1 − e−β~ωk


F = −kT ln Q = kT 3
(2π~)
and we can obtain the entropy from S = − ∂F


∂T V
and the pressure from
∂F


p = − ∂V T
.

1.3. Grand Canonical Ensemble

Consider water in equillibrium with it’s vapor. In general, the num-

ber of water molecules will fluctuate. Therefore, we need an ensemble
where we would allow for nonconsevation of particle number.
Suppose the total number of subsystems in the ensemble be M . The
links between the boxes, in this case, represent both energy and particle
exchange (Fig.). Each system can now have an arbitrary number of
particles. Total number of particles in the ensemble as a whole is, of
course, fixed.
Let H(N ) be the Hamiltonian of a system of N particles and volume
V . The total Hamiltonian for the ensemble is
X
HEn = Hsystem + Eexch
12 1. THEORY OF ENSEMBLES

Once again we assume that Eexch  Hsystem so that the wavefunction

for the ensemble factorizes
Y
ΨEn = ψsystem

and the energy of the ensemble is

X
EEn = Esystem

Suppose |j(N )i is the system energy eigenstate corresponding to the

energy eigenvalue Ej(N ) , i.e.

H(N ) |j(N )i = Ej(N ) |j(N )i

here N plays the role of a parameter, describing the number of particles

in a system. Let Mj(N ) be the number of systems in the ensemble in
the state |j(N )i. Clearly we must have
X
Mj(N ) = M
En

and
X
N Mj(N ) = the total number of particles in the ensemble
En
= Mη

which is a definition for η, the average number of particles per system.

Likewise, we may define the average enrgy per system E, as
X
Ej(N ) Mj(N ) = M E
En
P P P
Here the sum over the ensemble En stands for N j(N ) .
{Mj(N ) } denotes a distribution describing the ensemble. Of course,
this does not fix the ensemble wavefunction completely. The number
of ensemble states corresponding to this distribution is
M!
Ω= Q
En Mj(N ) !
Q Q Q
where En = N j(N ) . As in the case of the canonical ensemble, we
now look for the most probable distribution. According to the funda-
mental hypothesis, this is the distribution that maximizes Ω, subect to
the conservation constraints that we have already written down.
 1.3. GRAND CANONICAL ENSEMBLE 13

Taking logarithmic derivatives and imposing the constraints by means

of Lagrange multipliers, we get :
" #
∂ X

ln Ω − Mj(N ) α + βEj(N ) + γN = 0
∂Mj(N ) En

Using Sterling’s approximation we get

X X
ln Ω = M ln M − M − Mj(N ) ln Mj(N ) + Mj(N )
En En

which leads to1

∂
ln Ω = − ln MJ(N )
∂Mj(N )
Thus the extremization of probability gives us


− ln Mj(N ) − α + βEj(N ) + γN = 0

and hence
Mj(N ) = e−α−βEj(N ) −γN
We now define the grand partition function by
XX
QG = e−βEj(N ) −γN
N j(N )

using the first constraint gives

M = e−α QG

and thus the relative frequencies for the most probable distribution are
Mj(N ) 1 −βEj(N ) −γN
= e
M QG
In the limit M → ∞ , the most probable distribution becomes the
distribution. Therefore the probability that a system is in the state

1A better approximation (taking into account further terms in Stirling’s formula)

gives
1 1 X X
ln Ω = ln 2π + ln M + M ln M − M − Mj(N ) ln Mj(N ) + Mj(N )
2 2
En En
1X 1X
− ln 2π − ln Mj(N )
2 2
En En
which leads to
∂ 1
ln Ω = − − ln MJ(N ) ≈ − ln Mj(N )
∂Mj(N ) 2Mj(N )
This is identical to the result obtained using the simpler approximation.
14 1. THEORY OF ENSEMBLES

|j(N )i is
Mj(N ) 1 −βEj(N ) −γN
Pj(N ) = lim = e
M →∞ M QG
the values of β and γ are to be determined from the other two con-
straints.

Interpretation of the Lagrange multipliers

We can now attach one more box to the whole ensemble. The new
entrant can exchange energy only with the members of the initial en-
semble. Thus what we have here is an example of a hybrid ensemble.
We can now deduce that the Lagrange multiplier β, which appears
in the minimization condition via the energy conservation constraint,
must be related to the temperature. Indeed, since the added box par-
ticipates in the energy exchange it will have equal role in determining
β (but not γ, which is fixed by particle number conservation) as the
members of the initial ensemble. So β must be a function of a com-
mon property shared by the new box, hence, of temperature. Thus, we
define the absolute temperature T by
1
β=
kT
The other Lagrange multiplier, γ, is shared by all boxes which can
exchange both energy and particles. It should be a function of the
intensive thermodynamic variables common to these systems, namely
the temperature, T , and the chemical potential, µ

γ = γ(T, µ)

We will have more to say about this in a little while.

Legendre transform of QG

Using the definition of QG we get

(1.3.1)
 
∂ 1 X X
− ln QG = Ej(N ) e−βEj(N ) −γN = Ej(N ) Pj(N ) = E
∂β V,γ QG En En

and
 
∂ 1 X X
(1.3.2) − ln QG = N e−βEj(N ) −γN = N Pj(N ) = η
∂γ V,β QG En En
 ENTROPY 15

Thus at a constant volume, we have

(1.3.3) (d ln QG )V = −E(dβ)V − η(dγ)V

We observe that for the most probable distribution (for large M

and Mj(N ) ) we have
X M
ln Ω = Mj(N ) ln
En
Mj(N )
and thus
X
M −1 ln Ω = Pj(N ) ln Pj(N ) = ln QG + βE + γη
En
P P
where we have used the results En Pj(N ) = 1, En Ej(N ) Pj(N ) = E
P
and En N Pj(N ) = η.
Note that the quantity M −1 ln Ω turns out to be the Legendre trans-
form of ln QG , from a natural function of V ,β and γ to a natural
function of V ,E and η. At constant V , we have

(d[M −1 ln Ω)])V = (d ln QG )V + β(dE)V + E(dβ)V + γ(dη)V + η(dγ)V

= β(dE)V + γ(dη)V

Of course, we can solve the equations (1.3.1,1.3.2) to obtain β and γ as a

function of V ,E and η, and thus express M −1 ln Ω as a function of these
quantities explicitly. The following derivatives will be of importance
later  
∂  −1 
M ln Ω =β
∂E V,η
and  
∂  −1 
M ln Ω =γ
∂η V,E

Entropy

Using the thermodynamic relation dE = T dS − pdV − µdη, where

µ is the chemical potential, we have (dE)V,η = T dS. Hence
   
∂ S 1 ∂  −1 
= =β= M ln Ω
∂E k V,η kT ∂E V,η

and thus we see that

  
∂ S −1
− M ln Ω =0
∂E k V,η
16 1. THEORY OF ENSEMBLES

and so the quantity Sk − M −1 ln Ω = f (V, η) is independent of E. Now

let us take the T → 0 limit. In this limit the ensemble goes over to it’s
ground state, EEn → Eground state . For a nondegenerate ground state,
Ω = 1. We use the third law of thermodynamics, Nernst’s law, that at
absolute zero S → 0 (Remember that this law is really a veto in favour
of non-degenerate ground states!). This leads to f (V, η) = 0 . Even if
we had a degenerate ground state, with Ω = ω0M (this is a definition of
ω0 ), we could have defined the zero temperature entropy as S0 = k ln ω0
and this would have lead to the same conclusion. Thus
S
= M −1 ln Ω
k
 
Again, from thermodynamics, we have T ∂S ∂η
= −µ and thus we
V,E
have the identification
   
∂ S ∂  −1  µ
= M ln Ω =γ=−
∂η k V,E ∂η V,E kT
and thus
µ
γ(T, µ) = − = −βµ
kT
So we have determined the Lagrange multipliers β and γ in terms of
the intensive thermodynamic variables.
To see the thermodynamic significance of the grand canonical par-
tition function, consider
S E ηµ
ln QG = − +
k kT kT
Now, since pV = T S − E + ηµ, we have
pV
ln QG =
kT
Thus knowledge of the grand partition function leads directly to the
equation of state (remeber that the canonical partition function directly
yielded the free energy).
Finally, we have
XX
QG = e−β(Ej(η) −µη)
η j(η)

The summand e−β(Ej(η) −µη) is the relative probability that the system
−βEj(η)
P
has η particles and energy Ej(η) . Since Q(η) = j(η) e is the
 EXAMPLE : A SYSTEM OF FREE PARTICLES 17

canonical partition function for a system of η particles he grand canon-

ical partition function can also be written as
X
QG = e−µβη Q(η)
η

Example : A system of free particles

The total energy of a system of free particles is

X
Esys = n α εα
α

where nα is the number of particles in the state |αi, which has the
energy εα . For bosons, nα can take all nonnegative integer values,
0, 1, 2, 3, . . . , while for fermions it can only take the values 0 and 1.
The label α may be collection of quantum numbers, e.g. momentum,
helicity etc. The energy is given by the relativistic formula
p
εα = p~α2 c2 + m2 c4 − mc2

we won’t need this explicit energy momentum dispersion relation at

P
this point. Of course, the total number of particles is N = α nα .
The grand partition function is
X P
QG = e−β α (nα εα −µnα )
{nα }
XY
= eβ(µ−εα )nα
{nα } α

interchanging the order of the sum and the product leads to

( Q  
β(µ−εα )
YX
β(µ−εα )nα α 1+e −→ Fermions
QG = e = Q  β(µ−εα ) −1

α {n }
α α 1−e −→ Bosons

The equation of state that follows is

pV
= ln QG
kT
X 
ln 1 ± eβ(µ−εα )

= ±
α

where the upper sign stands for fermions and the lower sign stands for
bosons.
18 1. THEORY OF ENSEMBLES

Note that we may write

( Q
(1 + xα ) −→ Fermions
QG = Qα P∞ r
α r=0 xα −→ Bosons

where xα = eβ(µ−εα ) . So xrα is the relative probability that there would

be r particles in the single particle state |αi. For fermions, the average
number of particles in the state |αi is given by
P1
rxrα xα
hnα i = Pr=01 =
r
r=0 xα
1 + xα
while for bosons
xα dxd α ( ∞ r
P
r=0 xα )
P∞
rxrα
hnα i = Pr=0
∞ r = ∞
r=0 xα
P r
r=0 xα
 
P∞ r d 1
= xα dxd α ln ( r=0 xα ) = xα ln
dxα 1 − xα
xα
=
1 − xα
So we get the distribution functions for free particles
1
hnα i =
e(εα −µ)/kT ±1
where the upper sign (+) stands for fermions, while the lower sign (-)
is for bosons.

The Thermodynamic Connection

The only approximation involved in the derivation of the Grand

canonical ensemble was that the particle and energy leakage be small.
This, in itself, does not mean that we will have to confine ourselves to
large volume systems. Even for a finite volume system, the linkage nay
be made small, by reducing the area of the leak, for example.
However, to derive thermodyanamics, we will have to take the in-
finite volume limit. Thermodynamics is valid only for large systems.
This point follows from the next theorem.
Theorem. If the thermodynamic relations

dE = T dS − pdV + µdη

and
G = µη = E − T S + pV
 THE THERMODYNAMIC CONNECTION 19

are both valid, then µ, E/η, S/η are all functions of T and p only (i.e.
they are independent of η)

Proof. We have

dG = dE − T dS − SdT + pdV + V dp
= −SdT + V dp + µdη

But we also have

dG = d(µη) = µdη + ηdµ

Hence,

ηdµ = −SdT + V dp
and thus
S V
dµ = − dT + dp
η η
Now, since dµ is a perfect differential,
  µ must be a function of T and p
∂µ ∂µ
only. Also, Sη = − ∂T p and Vη = ∂p


must be functions of T and p
T
only. Again,
E S V
=µ+T −p
η η η
is also a function of T and p only. 

Since E and S are proportional to η at constant T and p, we are

in effect neglecting surface contributions (For energy, this is the total
surface energy and not just the leakage energy we have neglected in
constructing the grand canonical distribution.). So, therodynamics is
really valid in the limit η → ∞ , i.e. V → ∞.
The next question that concerns us is - under what conditions does
the V → ∞ limit of statistical mechanics reproduce thermodynamics?
Note that we can define an average mechnical pressure pmech by the
realtion  
∂Ej(N )
pmech = −
∂V
This, really, is the pressure one measures experimentally. On the other
hand, while deriving the equation of state
pV
= ln QG
kT
20 1. THEORY OF ENSEMBLES

we have already used the thermodynamic relation for pressure

pV = T S − E + ηµ

So, we could use the relation

pTh V
= ln QG
kT
to define a thermodynamic pressure pTh . In both of these definitions
the V → ∞ limit is understood. If pTh = pmech , the system obeys
thermodynamics, otherwise it does not. So, the question we have been
asking now, can be rephrased as - what conditions are necessary to
ensure that pTh = pmech ? The next theorem answers this very question.
Theorem. pTh = pmech if and only if pTh is a function of T and µ
only.

Proof. We will only prove the sufficiency. Since pTh is a function

of T and µ only, µ, in turn, is a function of pTh and T only.
 
∂Ej(N )
pmech = −
∂V
1 X ∂Ej(N ) −(Ej(N ) −µN )/kT
= − e
QG ∂V
N,j(N )
   
∂ ∂
= kT ln QG = kT ln QG
∂V T,µ ∂V T,pTh
 
∂
= pTh V = pTh
∂V T,pTh

Since ln QG = pThkT
V
the system obeys the laws of thermodynamics
if and only if ln QG = V f (T, µ) as V → ∞, i.e. if and only if the limit
limV →∞ V1 ln QG exists. The large V behaviour of ln QG depends upon
the nature of the law of force among the particles. We would later
obtain rigorous constraints on the force law, for the thermodynamic
limit to exist.
The thermodynamic limit exists for a system of particles interacting
via short range forces. However, if long range forces (e.g. Coulomb or
gravitational) are present, this limit, in general, does not exist. For
example, in a confined plasma, we can not use thermodynamics, but
statistical mechanics still applies. In other words, in a confined plasma
 A SIMPLE EXAMPLE : FREE PARTICLES IN A PERIODIC BOX 21

the mechanical pressure (the pressure that we can measure) is not an

intensive variable - it depends on the volume, too.

A simple example : Free particles in a periodic box

We have already seen that for a system of free particles

X 
ln 1 ± eβ(µ−εα )

ln QG = ±
α

where the sign + (-) stands for fermions (bosons) respectively. If we

assume that the particles are in a periodic cubical box with volume
V , we find that the momentum eigenstates are labelled by ki = 2π l
L i
1/3
where the index i takes on the values x, y, z and L = V is the linear
dimension of the box, and {li } is a set of three integers. To label the
state |αi completely, we will have to specify the helicity also. In the
limit V → ∞, the sum over states become
Z
X V
−→ (2j + 1) 3
d3 k
α
(2π)
where the factor of 2j + 1 is due to the 2j + 1 helicity states of a finite
rest mass particle of spin j. For a cubical box, this formula is easily
derived by directly computing the density of states. However, the final
formula is shape independent2.
We now define the fugacity

z = eµ/kT

2There are two questions involved here :

(1) Shape of the boundary, i.e. whether it is a cube or a sphere, etc.
(2) Nature of the boundary conditions imposed upon the single particle wave
functions , e.g. whether it is constrained to vanish or be periodic at the
boundary.
The single particle wave functions involved would, of course, depend on both the
shape of the boundary and on the boundary conditions (for example, one may get
plane travelling waves, standing waves, sperical Bessel funcions etc. depending
on these factors). However, as V → ∞, the sum to integral conversion formula
is independent of such considerations (To show this for the eigenfunctions of the
Laplace operator, was the famous Poincaré problem. For the sake of convenience,
we will always use a cube and impose periodic boundary conditions. This is what
we mean by a “periodic box”.
22 1. THEORY OF ENSEMBLES

e−Ej(N ) /kT z N we get

P P
Using QG = N j(N )
 
∂ 1 XX
ln QG = N z N e−Ej(N ) /kT
∂ ln z T,V Q G
N j(N )
XX
= Pj(N ) N = η
N j(N )

± ln (1 ± xe−εα/kT )
P
So for our free particle system, for which ln QG = α
the average number of particles is
X (±)2 ze−εα /kT
η =
α
1 ± ze−εα /kT
X 1
= (ε −µ)/kT
α
e α ±1

But η = hnα i where hnα i is the average number of particles in the state
|αi. So, we again find
1
hnα i =
e(εα −µ)/kT ±1
 CHAPTER 2

System of Free Fermions

Let us now consider a system of free fermions with the energy mo-
mentum dispersion relation
p
εα = p~α2 c2 + m2 c4 − mc2

where p~α = ~~kα . The Grand partition function is given by

X
ln 1 + e(µ−εα )/kT


ln QG =
α

Note that 1 and e(µ−εα )/kT represent the relative probabilities that the
single particle state |αi is unoccupied and occupied, respectively. Using
pV P
kT
= ln QG and the sum to integral conversion formula α −→ (2j +
V
R 3
1) (2π)3 d k in the thermodynamic limit, we get
Z
p 2j + 1 3 (µ−ε)/kT


= d k ln 1 + e
kT 8π 3
Since the average number of particles for a energy ε and given helicity,
nε equals the average number of particles with definite momentum ~k
and helicity λ, we have
D E 1
nε = n~k,λ = (ε−µ)/kT
e +1
The total number of particles in the system is
Z
2j + 1
N= V d3 knε
8π 3
and the total energy is
Z
2j + 1
E= V d3 kεnε
8π 3

For nonrelativistic particles, ~k  m the energy momentum rela-

~2 k 2
tion is ε = 2m
where k = ~k . Thus
3/2
√

3 2 2m
d k = 4πk dk = 2π εdε
~2
23
24 2. SYSTEM OF FREE FERMIONS

In the ultrarelativistic limit, on the other hand, we have ~k  m,

leading to ε = ~ck and
4π 2
d3 k = ε dε
(~c)3
2.1. Relation between pressure and energy density

In the non-relativistic case

3/2 Z ∞
√

p 2j + 1 2m
dε ε ln 1 + e(µ−ε)/kT


= 2 2
kT 4π ~ 0
"
∞ Z ∞ 1

µ−ε #
2 3/2  µ−ε
 2 − e kT
= C ε ln 1 + e kT − dεε3/2 kT µ−ε
3 0 3 0 1 + e kT
Z ∞
2 ε3/2
= C× dε ε−µ
3kT 0 e kT + 1
2m 3/2
where C = 2j+1


4π 2 ~2
. On the other hand, we have
Z ∞
E ε3/2
=C dε ε−µ
V 0 e kT + 1
Thus for a gas of non-relativistic fermions we have
p 2 E
=
kT 3kT V
or p = 23 u where u is the energy density.
In the ultra-relativistic case, we have
Z ∞
p
dεε2 ln 1 + e(µ−ε)/kT


= D
kT 0
Z ∞
1 ε3
= D× dε ε−µ
3kT 0 e kT + 1
2j+1
where the constant of proportionality here is D = 2π 2 (~c)3
. Comparing
this with Z ∞
E ε3
=D dε ε−µ
V 0 e kT + 1
1
we see that in this case p = 3 u.

2.2. System of Free Fermions at T = 0

For a fermionic system at T = 0, the occupation probability is

(
0 for ε > µ
nε =
1 for ε < µ
 2.2. SYSTEM OF FREE FERMIONS AT T = 0 25

This is the completely degenerate Fermi-Dirac distribution. The value

of µ for this case (T = 0) is called the Fermi energy εF . Also, we define
the Fermi momentum through
q
εF = ~2 c2 kF2 + m2 c4 − mc2

i.e. εF is the kinetic energy corresponding to the Fermi momentum.

So
Z
2j + 1
E = V d3 kε
8π 3 ε≤ε
Z F
2j + 1
N = 3
V d3 k
8π ε≤εF
2j+1
Thus N = 6π 2
V kF3 and thus
 13   13
6π 2

N
kF =
2j + 1 V
Since the average momentum is of the order of ~kF , the De-Broglie
wavelength is of the order of the inter particle seperation
  13
1 V
λ∼ ∼
kF N
The Fermi energy is given by

2 2
 2
 ~ kF = 6π2 3 ~2 ρ 23 nonrelativistic case

2m 2j+1 2m
εF =  2  13 1
 ~ckF = 6π

~cρ 3 ultrarelativistic case
2j+1

where ρ = NV
is the particle number density.
For the nonrelativistic case one can easily calculate the energy of
the system to be
3 ~2 kF2 3
E= N = N εF
5 2m 5
and using the pressure - energy density relation derived earlier, we get
2E 2
p= = ρεF
3V 5
For the ultrarelativistic case, the corresponding results are
3
E = N εF
4
and
26 2. SYSTEM OF FREE FERMIONS

1
p = ρεF
4
Thus the free fermion gas exerts a nonzero pressure, even at absolute
zero! This purely quatum mechanical feature is called the electron
degeneracy pressure.

2.3. Application to white dwarfs or neutron stars

Astronomical observations reveal that a typical white dwarf star

has a mass of the order of our sun, M ∼ M . From the color, we
know that the temperature is T ∼ 106 − 107 K. Observed luminosity
values lead to a typical value of R ∼ 10−2 R for the white dwarf
radius, which implies that the typical interparticle seperation is of the
order of 10−12 m. Since the interprticle seperation is much smaller
than typical atomic radii, the uncertainity principle says that electron
kinetic energies in a white dwarf is much higher than atomic binding
energies. Hence, the electrons are nearly free! In addition, despite the
high temperature in the white dwarf interior, the high density leads
to a much larger value for the Fermi energy, εF  kT (Typically,
εF ∼ 20 MeV, while kT ∼ 100 eV). Thus, the interior of the white
dwarf can be described as a highly degenerate free electron gas!
In this section we will use order of magnitude estimates to quali-
tatively discuss an important property of the white dwarf stars - the
reltionship between mass and radius. For very low mass, the density is
small. The system behaves like a crystalline solid with constant den-
1
sity. Hence for very small masses, we have R ∼ M 3 . Our interest lies
in the higher mass white dwarfs.
The stars are held in equilibrium by the opposing forces of grav-
itational collapse and internal pressure. For normal stars, the inter-
nal pressure is primarily due to the stellar radiation. White dwarfs,
however, are stars near the end of their evolutionary cycle. Having
exhausted their nuclear fuel, these stars radiate only very weakly (the
necessary energy coming from the release of gravitational energy dur-
ing collapse). The internal pressure in this case comes almost entirely
from the electron degeneracy pressure.
 2.3. APPLICATION TO WHITE DWARFS OR NEUTRON STARS 27

GM 2
In equlibrium, the gravitational force ∼ R2
must balance the hy-
drostatic force ∼ pR2 . Thus,
GM 2
∼ pR2
R2
which leads to
GM 2
∼ pV ∼ E ∼ N εF
R
We could have reached this conclusion via the equipartition theorem,
too.
2
For a nonrelativistic electron gas, εF ∼ ρ 3 . Since N ∼ M we get
5
GM 2 M3
∼ 2
R R
which leads to the mass radius relation
  13
1
R∼
M
Thus, in this regime, the radius decreases with increasing mass. This
continues till the elctrons are squeezed to such a high density that
relativistic effects become important.
For very large mass, we can treat the electron gas as ultrarelativis-
1
tic. In this regime εF ∼ ρ 3 . This leads to
4
GM 2 M3
∼
R R
A more careful analysis shows that this actually gives a critical mass
Mc , at which the radius of the white dwarf star vanishes. Beyond this
mass, electron degeneracy pressure can no longer sustain the star and
it would collapse. This is the Chandrashekhar limit.
Reinstating ~ and c, we get
GM 2
≈ N kF ~c
R
1 M
where kF ≈ ρ 3 and N ≈ 2m n
where mn is the nucleon mass. Here,
we have assumed an average nucleus with 2Z ∼ A. For a neutron
star, N ≈ mMn . This leads to an estimate of the critical mass, ignoring
dimensionless factors of the order of unity
4
GMc2

~c Mc 3
≈
R R mn
28 2. SYSTEM OF FREE FERMIONS

So, the critical mass is given by

  32
1 ~c
Mc ≈ 2
mn G
This gives Mc ≈ 1.7M . Careful calculations lead to the value Mc =
1.4M . Thus our estimate, in spite of it’s crude nature, is actually
quite close to the accurate value!

2.4. Electron gas in a metal

In a metal, the interparticle seperation is of the order of 10−8 cm.

This leads to a Fermi energy εF ≈ 1eV ≈ µ while at room temperature,
1
T ∼ 300K we have kT ≈ 40 eV. Thus the system is degenerate, but not
µ
completely degenerate. The fugacity z = e kT ∼ e100  1. Note that
here the interparticle seperation is of the order of atomic sizes and the
electron energy is of the order of atomic binding energies. To treat such
an electron gas as a free one may seem to be a gross approximation,
but the free electron gas model works remarkably well. The answer
to this puzzle lies in quantum mechanics of an electron in a periodic
potential.
Before we embark on the physics of the electron gas in a metal, let
us take a look at the mathematical machinary required.We would need
the values of integrals of the type
Z ∞
Il = dεnε εl
0

where nε is the Fermi-Dirac distribution function

h ε−µ i−1
nε = e kT +1

Integration by parts leads to

Z ∞ ∞ ∞
εl+1 εl+1 dnε
Z
dεnε εl = nε − dε
0 l+1 0 0 l + 1 dε
and for l ≥ −1 we have
Z ∞ ∞
εl+1 dnε
Z
l
Il = dεnε ε = − dε
0 0 l + 1 dε
 2.4. ELECTRON GAS IN A METAL 29

Now, at T = 0 the Fermi-Dirac distribution function is a step func-

tion (
0 for ε > µ
nε =
1 for ε < µ
and thus the derivative is a delta function
dnε
− = δ(ε − µ)
dε
µ
At finite but low temperatures for which kT  1 the distribution re-
mains nearly a step function, and it’s derivative is almost a delta func-
tion. Indeed, Z ∞
dnε
− dε = nε=0 ≈ 1
0 dε
µ
since e− kT  1. Thus
 
dnε kT
− = δ(ε − µ) + O
dε µ
So
µl+1
 
kT
Il = +O
l+1 µ
In the current problem, however, we do notwish to neglect kT
µ
, which
−2
is of the order of 10 .We therefore start with the exact expression
dnε 1 1
− =
dε kT (e + 1) (e−x + 1)
x

ε−µ
where x = kT
. A change of variables leads to

(µ + kT x)l+1
Z ∞
1
Il = dx
−µ/kT l+1 (ex + 1) (e−x + 1)
(µ + kT x)l+1
Z ∞
1
≈ dx
−∞ l+1 (e + 1) (e−x + 1)
x

µ
where the last step is valid since the integrand is nearly zero for x ≤ kT .
We use the binomial theorem
1
(µ + kT x)l+1 = µl+1 + (l + 1)kT µl x + (l + 1)l(kT )2 µl−1 x2 + . . .
2!
kT
where we have to retain the quadratic term in the small parameter µ
since the linear turm vanishes on integration. This gives
Z ∞
µl+1 1 2 l−1 x2
Il = + l(kT ) µ dx x
l+1 2 −∞ (e + 1) (e−x + 1)
30 2. SYSTEM OF FREE FERMIONS

Now the integral

∞ ∞
x2 x2
Z Z
dx x = 2 dx
−∞ (e + 1) (e−x + 1) 0 (ex + 1) (e−x + 1)
Z ∞

−2
= 2 dxx2 e−x 1 + e−x
0
Z ∞ ∞
X
2 −x
= 2 dxx e (−1)n (n + 1)e−nx
0 n=0
∞
X Z ∞
=2 n
(−1) (n + 1) dxx2 e−(n+1)x
n=0 0
R∞
Thus we need to evaluate integrals of the form 0 x2 e−mx dx . Using
the method of differentiation under the integral sign, we have
Z ∞ Z ∞
∂2 ∂2
 
2 −mx −mx 1 2
xe dx = e dx = =
0 ∂m2 0 ∂m2 m m3
and thus
∞ ∞
x2 π2
Z
n−1 1
X
dx = 4 (−1) =
−∞ (ex + 1) (e−x + 1) n=1
n2 3
So we get the lowest order correction to our integral as
µl+1 1 2 l−1 π
2
Il = + l(kT ) µ + ...
l+1 2 3
Armed with this, we now tackle the question of the value of µ as a
function of temperature in a free electron Fermi gas. Using
Z
2j + 1
N = V d3 knε
8π 3
Z ∞
2j + 1 3 √
= 2
V (2m) 2 dε εnε
4π 0
"  2 2 #
2j + 1 4 3 3 kT π
= V (2mµ) 2 1 + + ...
4π 2 3 8 µ 3
"  2 2 #
2j + 1 3 kT π
= V (2mµ) 2 1 + + ...
6π 2 µ 8
Note that the Fermi energy εF is defined to be the value of µ at T = 0.
Hence, the definition of εF is
2j + 1 3
N≡ 2
V (2mεF ) 2
6π
 2.5. PAULI PARAMAGNETISM 31

and we can relate µ at a finite temperature to εF by

"  2 2 # 23 "  2 2 #
kT π kT π
εF = µ 1 + + ... = µ 1 + + ...
µ 8 µ 12
which can be solved to yield
" 2 #
π2

kT
µ = εF 1 − + ...
εF 12
Since εF is by definition a function of density, we have found µ as a
function of temperature and frequency. The above expression is valid
2
for kT  εF . Since εF ∼ ρ 3 , this expansion is valid when the
quantity T2 is small, i.e. this is a low temperature and/or high density
ρ3
expansion.
We will take a deeper look at the free electron Fermi gas, in a little
while.

2.5. Pauli Paramagnetism

Consider a system of free nonrelativistic electrons in a magnetic field

H. Let us assume that the orbital motion of the electrons is unaffected
by the magnetic field (This, of course, is a gross simplification.Change
in the orbital motion in the presence of a magnetic field is precisely the
cause of such important phenomena as Landau diamagnetism and de
Haas-Van Alphen oscillations).The single elctron energies in the pres-
ence of the magnetic field is

E± = ε ∓ µB H

where the upper (lower) sign refers to electrons with their magnetic
p2
moment parallel (antiparallel) to the magnetic field. Here ε = 2m is
e~
the kinetic energy and µB = 2mc is the Bohr magneton.
The number of electrons with their spin parallel (antiparallel) to
the magnetic field is given by
i−1 Z ∞
X h E± −µ ρ(ε)
n± = e kT +1 → dε E± −µ
ε 0 e kT + 1
where  32
√ 3√

V 2m 2πV
ρ(ε) = 2 ε= (2m) 2 ε
4π ~ h3
32 2. SYSTEM OF FREE FERMIONS

is the density of state for electrons of each kind. To fix µ, note that
the total number of electrons is
Z ∞ " #
1 1
N = n+ + n− = dερ(ε) E+ −µ + E− −µ
0 e kT + 1 e kT + 1
Z ∞  
1 1
= dερ(ε) ε−µB H−µ + ε+µB H−µ
0 e kT + 1 e kT +1
This equation has to be solved for µ(H). Changing the sign of H in this
expression, H → −H shows that µ(H) and µ(−H) satisfy the same
equation. Thus µ is an even function of the magnetic field

µ(H) = µ0 + O H 2

µ0 is the chemical potential in the absence of any magnetic field. At

low temperatures, kT  εF we have seen that it has the expansion
"  2 2 #
kT π
µ0 = ε F 1 − + ...
εF 12
The magnetic dipole moment is

M = µB (n+ − n− )
Z ∞  
1 1
= µB dερ(ε) ε−µB H−µ − ε+µB H−µ
0 e kT +1 e kT +1
Our aim is to compute the magnetization to O(H). Now
Z ∞
ρ(ε)
n+ = dε ε−µB H−µ
0 e kT +1
Z µB H Z ∞
ρ(ε) ρ(ε)
= dε ε−µB H−µ + dε ε−µB H−µ
0 e kT +1 µB H e kT +1
Consider the first integral. It obviously goes to zero as H → 0. To
obtain the leading term, we can replace H and ε in the denominator
of the integrand by zero, which leads to
Z µB H Z µB H
ρ(ε) 1
dε ε−µB H−µ −→ − µ dερ(ε)
0 e kT +1 e kT + 1 0
Z µB H
√ 3
∝ dε ε ∝ (µB H) 2
0
 2.5. PAULI PARAMAGNETISM 33

Thus if we are interested in computing M to first order in H, this

integral may be neglected. Then
Z ∞ Z ∞
ρ(ε) ρ(ε + µB H)
n+ ≈ dε ε−µB H−µ = dε ε−µ
µB H e kT +1 0 e kT + 1
Similarly Z ∞
ρ(ε − µB H)
n− ≈ dε ε−µ
0 e kT + 1
Then
Z ∞
dε
M ≈ µB ε−µ [ρ(ε + µB H) − ρ(ε − µB H)]
0 e kT + 1
Z ∞
2 ρ0 (ε)
≈ 2µB H dε ε−µ0
0 e kT + 1
which is correct to the first order in H. The zero field susceptibility is
2µ2B ∞ ρ0 (ε)
Z
χ = dε ε−µ0
V 0 e kT + 1
3 Z 1
µ2B 2m 2 ∞ ε− 2
 
= dε ε−µ0
4π 2 ~2 0 e kT + 1
 2  23 2
3 3
2 −2
Using εF = 3πV N ~
2m
, we get 2m 2
~ 2 = 3π ε F ρ where ρ = N
V
is the
number density. Thus
1
3 2 − 23 ∞ ε− 2
Z
χ = ρµB εF dε ε−µ0
4 0 e kT + 1
For µ0  kT , we can use the expansion for Il to get
"  2 2 #
3 − 3 √ kT π
χ = ρµ2B εF 2 2 µ0 1 − + ...
4 µ0 24
Using the expansion for µ0 we have
"  2 2 #
3 2 − 32 kT π
χ = ρµ ε 1− + ...
2 B F µ0 24
"  2 2 # 12
√ kT π
× εF 1 − + ...
µ0 12
"  2 2 #
3 ρµ2B kT π
= 1− + ...
2 εF µ0 12
This is known as the Pauli susceptibility.
34 2. SYSTEM OF FREE FERMIONS

2.6. Evaluation of Fermi-Dirac integrals

Consider the integral

Z ∞
f (ε)
I= dε ε−µ
0 e kT +1
for kT  µ, where f (ε) is analytic at ε = µ. First, we make a change
of variables x = ε−µ
kT
to get
Z ∞
f (µ + kT x)
I = kT dx
−µ/kT ex + 1
Z 0 Z ∞ 
f (µ + kT x) f (µ + kT x)
= kT dx + dx
−µ/kT ex + 1 0 ex + 1
"Z Z ∞ #
µ/kT
f (µ − kT x) f (µ + kT x)
= kT dx −x + dx
0 e +1 0 ex + 1

Using e−x1+1 = 1 − ex1+1 we get

Z µ/kT Z µ/kT Z µ/kT
f (µ − kT x) f (µ − kT x)
dx −x = dxf (µ − kT x) − dx
0 e +1 0 0 ex + 1
In the second integral we can replace the upper limit by ∞, the error
µ
introduced thereby being of the order of e− kT .
Z µ/kT Z µ Z ∞
f (µ − kT x) 1 f (µ − kT x)
dx −x ≈ dεf (ε) − dx
0 e +1 kT 0 0 ex + 1
and hence
µ ∞
f (µ + kT x) − f (µ − kT x)
Z Z
I= dεf (ε) + kT dx
0 0 ex + 1
Since f (ε) is analytic at ε = µ , we can expand in Taylor series to get
∞
X (xkT )2n−1 d2n−1
f (µ + kT x) − f (µ − kT x) = 2 2n−1
f (µ)
n=1
(2n − 1)! dµ
Thus
∞ X (kT )2n ∞
f (µ + kT x) − f (µ − kT x)
Z
kT dx = 2
0 ex + 1 n=1
(2n − 1)!
Z ∞
d2n−1 x2n−1
× 2n−1 f (µ) dx x
dµ 0 e +1
 2.6. EVALUATION OF FERMI-DIRAC INTEGRALS 35
R∞ 2n−1
Now 0 dx xex +1 = (2n − 1)! 1 − 22n−1
1


ζ(2n) where ζ(m) is the Rie-
1
mann zeta function . Hence
Z ∞ ∞  
f (µ + kT x) − f (µ − kT x) X 1
kT dx = 2 1 − 2n−1 ζ(2n)
0 ex + 1 n=1
2
 2n−1 
d
× f (µ) (kT )2n
dµ2n−1
Finally, we arrive at
Z ∞ Z µ ∞  
f (ε) X 1
dε ε−µ = dεf (ε) + 2 1 − 2n−1 ζ(2n)f (2n−1) (µ)(kT )2n
0 e kT +1 0 n=1
2
Z µ
π2 0 2 7π 4 000
= dεf (ε) + f (µ)(kT ) + f (µ)(kT )4 + . . .
0 6 360
µ l+1
In particular, for f (ε) = εl we have 0 f (ε)dε = µl+1 , f 0 (µ) = lµl−1 , f 000 (µ) =
R

l(l − 1)(l − 2)µl−3 , . . . and hence

Z ∞
εl µl+1 π2
dε ε−µ = + lµl−1 (kT )2
0 e kT + 1 l+1 6
7π 4
+ l(l − 1)(l − 2)µl−3 (kT )4 + . . .
360 "
l+1
 2
µ π 2 kT
= 1 + (l + 1)l
l+1 6 µ
 4 #
4
7π kT
+(l + 1)l(l − 1)(l − 2) + ...
360 µ

1The
R∞ 2n−1 R∞ −1 P∞
required integral, 0 dx xex +1 = 0 dxx2n−1 e−x (1 + e−x ) = p=1 (−)p+1
R∞ R∞ P∞ p+1 P∞ p+1
× 0 dxx2n−1 e−px = 0 dxx2n−1 e−x p=1 (−1) p2n = (2n − 1)! p=1 (−1)p2n
P∞ p+1 P∞ 1
Now, 1 (−1) 1 1
P

p2n = 1 p2n − 2 p=even p2n = 1 − 22n−1 ζ(2n) where the Rie-
P∞
mann zeta function ζ(m) = p=1 p1m . The zeta function for even integer argument
is related to the Bernouli numbers Bn through
(2π)2s
ζ(2s) = (−1)s+1 B2s
2(2s)!
The Bernouli numbers are defined via their generating function
∞
x X Bn n
= x
ex − 1 n=0 n!
and have the values B1 = − 21 , B2n+1 = 0 for all n = 1, 2, 3, . . . , B2 = 1
6 , B4 =
1 1 1 5 631
− 30 , B6 = 42 , B8 = − 30 , B10 = 66 , B12 = − 2730 etc.
36 2. SYSTEM OF FREE FERMIONS

Problem 2.6.1. Show that the thermodynamic functions for a non-

relativistic degenerate electron gas are given by
"  2  4 #
2 4
π kT π kT
G = N µ = N εF 1 − − + ...
12 εF 80 εF
"  2  4 #
3 5π 2 kT π 4 kT
E = N εF 1 + − + ...
5 12 εF 16 εF
 "  2 #
π2 kT 3π 2 kT
CV = Nk 1− + ...
2 εF 10 εF
 "  2 #
π2 kT π 2 kT
S = Nk 1− + ...
2 εF 10 εF
3
2E V
and p = 3V
where N = 3π 2
(2mεF ) 2 .

Solution :
We start from Z
2j + 1
N= V d3 knε
8π 3
which for a nonrelativistic system becomes
Z ∞
2j + 1 3 √
N= 2
V (2m) 2 dε εnε
4π 0

(we have already done this part of the problem - but this time we are
going to carry out the calculation to the fourth order in the parameter
kT
µ
). Using the general result for Fermi integrals in the last section, we
find
Z ∞ "  2  4 #
√ 2 3 3 π 2 kT 9 7π 4 kT
dε ε = µ2 1 + + + ...
0 3 4 6 µ 16 360 µ
"  2  4 #
2 3 π 2 kT 7π 4 kT
= µ2 1 + + + ...
3 8 µ 640 µ
We get
"  2  4 #
2j + 1 4π 3 3 π 2 kT 9 7π 4 kT
N= V (2mµ) 2 1 + + + ...
8π 3 3 4 6 µ 16 360 µ
Using the definition of the fermi energy
2j + 1 4π 3
N= 3
V (2mεF ) 2
8π 3
 2.6. EVALUATION OF FERMI-DIRAC INTEGRALS 37

we get " #
2 4
π2 7π 4
 
3 3 kT kT
εF = µ
2 2 1+ + + ...
8 µ 640 µ
kT
So, to the fourth order in the parameter µ
, we have
" 2 4 # 23
2 4
 
kT
π 7π kT
εF = µ 1 + + + ...
8
µ 640 µ
"  2    4 #
2 π 2 kT 2 7π 4 1 π 4 kT
= µ 1+ + − + ...
3 8 µ 3 640 9 64 µ
"  2  4 #
π 2 kT π 4 kT
= µ 1+ + + ...
12 µ 180 µ
and we invert this result to get
"  2  4 #−1
2 4
π kT π kT
µ = εF 1 + + + ...
12 µ 180 µ
"  2  4   4 #
π 2 kT π π4 kT
= εF 1 − + − + ...
12 µ 144 180 µ
"  2  4 #
π 2 kT π 4 kT
= εF 1 − + + ...
12 µ 720 µ
 
2
 2 " 2
 2 #−2 4
 4
π kT π kT π kT
= εF 1 − 1− + + . . .
12 εF 12 εF 720 εF
"  2  4   4 #
π 2 kT π π4 kT
= εF 1 − + − + ...
12 εF 720 72 εF
"  2  4 #
π 2 kT π 4 kT
= εF 1 − − + ...
12 εF 80 εF
and thus
" 2 4 #
π2 π4
 
kT kT
G = N µ = N εF 1− − + ...
12 εF 80 εF
The total energy is
Z Z ∞
2j + 1 3 2j + 1 3 3
E= V d kεnε = V (2m) 2 dεε 2 nε
8π 3 4π 2
0
38 2. SYSTEM OF FREE FERMIONS

Now,
Z ∞ "  2  4 #
3 2 5 15 π 2 kT 15 7π 4 kT
dεε 2 = µ2 1 + − + ...
0 5 4 6 µ 16 360 µ
"  2  4 #
2 5 5π 2 kT 7π 4 kT
= µ2 1 + − + ...
5 8 µ 328 µ
Hence
R∞ 3
"  2  4 #
dεε 2 4
E 2 3 5π kT 7π kT
= R 0∞ √ = µ 1+ − + ...
N 0
dε ε 5 8 µ 328 µ
"  2  4 #−1
2 4
π kT 7π kT
× 1+ + + ...
8 µ 640 µ
"  2  4 #
3 π 2 kT π 4 kT
= εF 1 − + + ...
5 12 µ 720 µ
"  2  4 #
5π 2 kT 7π 4 kT
× 1+ − + ...
8 µ 328 µ
"  2  4   4 #
π 2 kT 7π π4 kT
× 1− − − + ...
8 µ 640 64 µ
"  2  4 #
3 5π 2 kT 19π 4 kT
= εF 1 + − + ...
5 12 µ 144 µ
 
2
 2 " 2
 2 #−2 4
 4
3  5π kT π kT 19π kT
= εF 1 + 1− − + . . .
5 12 εF 12 εF 144 εF
"  2  4 #
3 5π 2 kT π 4 kT
= εF 1 + − + ...
5 12 εF 16 εF
and thus
"  2  4 #
3 5π 2 kT π 4 kT
E = N εF 1 + − + ...
5 12 εF 16 εF
 2.6. EVALUATION OF FERMI-DIRAC INTEGRALS 39

The heat capacity is

"  3 #
k 5π 2 π4
 
dE 3 kT kT
CV = = N εF 2 − 4 + ...
dT 5 εF 12 εF 16 εF
 "  3 #
3 5π 2 kT 3π 2 kT
= Nk 1− + ...
5 6 εF 10 εF
 "  3 #
π2 kT 3π 2 kT
= Nk 1− + ...
2 εF 10 εF
To calculate the entropy, note that
2 5
G = N µ = E + pV − T S = E + E − T S = E − T S
3 3
where we have used the result p = 32 VE , valid for a nonrelativistic free
fermion gas
 2 "  3 #
5 π2 kT π 2 kT
T S = E − G = N εF 1− + ...
3 2 εF 10 εF
and hence
" 3 #
π2 π2
 
kT kT
S = Nk 1− + ...
2 εF 10 εF
We could have also calculated the entropy usingthe thermodynamic
relation T dS = dE + pdV which leads to
   
∂S ∂E
T = = CV
∂T V ∂T V
Hence
T
CV (T 0 )
Z
S = dT 0
0 T0
Z T 2
 " 2
 0 3 #
π k 3π kT
= dT 0 N k 1− + ...
0 2 εF 10 εF
 "  3 #
π2 kT π 2 kT
= Nk 1− + ...
2 εF 10 εF
and finally, the pressure is
"  2  4 #
2E 2N 5π 2 kT π 4 kT
(2.6.1) p= = εF 1 + − + ...
3V 5V 12 εF 16 εF
40 2. SYSTEM OF FREE FERMIONS

2.7. The high temperature expansion

So far, we have been concentrating on the degenerate Fermi gas,

that is at temperatures for which kT µ
< 1 (The expansions we have
obtained for various thermodynamic quantities, though valid in this
range, are of course usful only when kT µ
 1). In this section, we
consider an expansion which works well in the other extreme, that is,
µ
the regime kTµ
 1. In this regime, the fugacity z = e kT  1. Since
ε
ε > 0, the quantity ze− kT is less than 1 for all T . Thus we can expand
nε as
ε ∞
ze− kT X lε
nε = ε
− kT
= (−1)l+1 z l e− kT
1 + ze l=1
q2
Assuming a system of nonrelativistic particles obeying ε = 2m
, we get
Z ∞ ∞
2j + 1 2
X lq 2
l+1 l − 2mkT
N= V 4π q dq (−1) z e
8π 3 0 l=1

Using the result2

∞
Z r
2 −λq 2 1 π
dqq e =
0 4 λ3
we get
∞
2j + 1 1 √ 3
X
l+1 z
l
N= V π(2mkT ) 2 (−1) 3
2π 2 4 l=1 l2
which we write as
∞
l
N 2j + 1 X l+1 z
(2.7.1) ρ= = (−1) 3
V a3 l=1 l2
where r
2π
a=
mkT
is a parameter with the dimensions of length. Energy of a particle with
q ∼ a1 is ∼ 2ma
1 kT
2 = 4π . However, the average thermal energy is ∼ kT

and thus a ∼ the thermal de-Broglie wavelength.

2We
R∞ 2 R ∞ dz −z Γ( 1 )
start from 0 dqe−λq = 2√1 λ 0 √ e = 2√2λ = 12 πλ and differentiation
p
z
under the integral sign gives
Z ∞ Z ∞ r
2 ∂ 2 1 π
dqq 2 e−λq = − dqe−λq =
0 ∂λ 0 4 λ3
 2.7. THE HIGH TEMPERATURE EXPANSION 41

Similarly, the total energy is

Z ∞ ∞
2j + 1 2 q2 X lq 2
l+1 l − 2mkT
E= V 4π q dq (−1) z e
8π 3 0 2m l=1
R∞ 2 R∞ 2 −λq 2
Since 0 dqq 4 e−λq = − ∂λ∂ 31
pπ
0
dqq e = 24 λ5
we have
∞ l
E 3 2j + 1 X
l+1 z
= kT (−1) 5
V 2 a3 l=1 l2
Since the system is nonrelativistic, we have
∞ l
2E 2j + 1 X
l+1 z
p= = kT (−1) 5
3V a3 l=1 l2
and we see that the pressure and density are related by
∂  p 
ρ=
∂(ln z) kT
Although derived for the special case of the free fermion gas at high
temperatures, this is really a universal result valid for all systems. To
see this, we note that
 
∂  p  ∂  p  ∂ 1
=z =z ln QG
∂(ln z) kT ∂z kT ∂z V
Ej(N )
since QG = N,j(N ) e− kT z N we have
P

∂  p  1 1 X − Ej(N )
= e kT N z N
∂(ln z) kT V QG
N,j(N )

1
= hN i = ρ
V
So far, we have found the thermodynamic quantities as power series
expansions in the fugacity. What we would like to do is to eliminate
z and write down the expression for, say, pressure as a function of the
density. Since
ρa3 z2
= z − 3 + ...
2j + 1 22
we can solve for z to get
ρa3 ρa3
 
1
z= + + ...
2j + 1 2 32 2j + 1
42 2. SYSTEM OF FREE FERMIONS

Thus
z2
 
2j + 1
p = kT z − 5 + . . .
a3 22
3
   
ρa 1 1
= ρkT 1 + − 5 + ...
2j + 1 2 32 22
3
 
1 ρa
(2.7.2) = ρkT 1 + 5 + ...
2 2 2j + 1
This expansion is valid for
ρa3 < 1
thus for the region where the interparticle seperation > a1 ∼ thermal de-
Broglie wavelength. Thus at a fixed density we can look upon the above
equation as a high temperature expansion, or, at a fixed temperature
this can be regarded as a low density expansion.
The first term in the expansion for pressure, eq (2.7.2) is ρkT -
which is just the pressure of a classical ideal gas. As can be seen, the
first correction is positive. This is because the pressure in an ideal fermi
gas is always more than that in a classical ideal gas.
The expansion for density, eq (2.7.1) was derived under the condi-
tion z ≤ 1. Let us consider the limiting case z = 1. Then, we have
∞
ρa3 X (−1)l
= 3 ≈ 0.765
2j + 1 l=1 l 2

Finally, let us examine the conditions under which the fermi gas
shows it’s two extremes of behaviour. We have seen that the gas is non-
degenerate at high temperature or low density, while it is degenerate
at low temperature or high density. Thus the controlling factor is the
dimensionless quantity ρa3 . If ρa3  1 the system is nondegenerate,
while for ρa3  1 the system is highly degenerate.
We can understand this condition by a simple physical argument.
When ρa3  1 the wave functions of the particles do not overlap much.
So the symmetry or antisymmetry of wave functions have a very small
role to play in the calculation of the energy. So, the system becomes
almost classical.
 2.7. THE HIGH TEMPERATURE EXPANSION 43

We can rewrite the above condition in terms of the Fermi energy

εF and the absolute temperature by considering the ratio
kT 2mkT
=
εF qF
2j+1 4π 3
Since ρ ∼ 8π 3
× q
3 F
we have
kT 2mkT 1
∼ 2 ∼ 2
εF ρ3 (ρa3 ) 3
So, the condition for nondegenerate gas is kT  εF while that for the
degenerate gas is kT  εF
 CHAPTER 3

System of free bosons

For a free boson system, we have

X
N= nα
α

but now nα can take the values 0, 1, 2, . . . , N . The grand partition

function is given by
Y
Y 1
QG = 1 + e−εα /KT z + e−2εα /kT z 2 + . . . =
α α
1 − ze−εα /kT

As in the last chapter, z is the fugacity eµ/KT .

Thus for a system of free bosons we have
p 1 1 X
ln 1 − ze−εα /kT


= ln QG = −
kT V V α
and
∂  p  1 X ze−εα /kT 1 X
ρ= = = n̄α
∂ ln z kT V α 1 − ze−εα /kT V α
Hence, the average number of particles in the momentum helicity state
α is :
1
n̄α = εα −µ
e kT − 1
Also, the total enenrgy is given by
X
E= n̄α εα
α

Calculations similar to those in the last chapter yield the high tem-
perature expansion for a nonrelativistic gas
∞
p 2j + 1 X z l
=
kT a3 l=1 l 52
and ∞
2j + 1 X z l
ρ=
a3 l=1 l 32

45
46 3. SYSTEM OF FREE BOSONS

Note that the difference in the above expressions from that of the
fermion case lies in the alternating signs in the latter. Eliminating
z from these equations leads to
ρa3 1
 
p = ρkT 1 − + ...
2j + 1 2 52
In contrast to the fermion case, we find that in a boson gas the first
correction in the pressure over that of the classical ideal gas is negative.
This is because in the absence of degeneracy pressure the bose gas is
easier to compress than the ideal gas is.
For a system of photons the total number of photons is not con-
served1. Thus, in our derivations we do not need the Lagrange mul-
tiplier that maintains the number conservation constraint. Effectively
this means that we can take µ to be zero. We thus get the same answer
as in the canonical ensemble
1
n̄α = εα
e kT − 1
On the other hand, we could derive the same resulta that we have so far
derived, for systems where the total number of particles are conserved,
by using the canonical ensemble instead. This is because the basic
ststistical hypothesis is the same. Using the grand canonical ensemble
just makes life easier!

3.1. Bose Condensation

Earlier, we had used the replacement

Z
X V
−→ (2j + 1) 3 d3 k
α
8π
to calculate the expressions
N 1 X
ρ= = n̄α
V V α
and
p 2E 1 X
= = n̄α
kT 3kT V V α
for a system of nonrelativistic free bosons, at high temperatures or low
densities.
1This is because their basic interaction is emission and absorption mediated by
jµ Aµ
 3.1. BOSE CONDENSATION 47

When is this replacement accurate? Clearly the low lying states,

particularly the ~k = 0 state, are not counted properly in the integral,
because the volume element d3 k = 4πk 2 dk is small near ~k = 0. If
the occupation number n̄α of the low lying states is O(1), then this
inaccuracy does not make much of a difference as V → ∞. However,
this replacement is going to be a very bad approximation indeed if the
low lying states are macroscopically populated, i.e. n̄α ∼ O(N ) when
N is the total number of particles.
When can the low lying states be macroscopically populated? Con-
sider, in particular, the occupation number in the ~k = 0 state with
εα = 0,
1 z
n̄0 = −µ/kT =
e −1 1−z
where z = e µ/kT
is the fugacity. This means z = n̄0n̄+1
0
. Since 0 ≤
N
n̄0 ≤ N , we have 0 ≤ z ≤ 1+N . For large N this last inequality can
be rewritten in the form 0 ≤ z ≤ 1 − N1 . So the maximum value of
fugacity for a bose gas is
zmax = 1
which can be attained only in the limit N → ∞.
Since, for a bose gas, z = eµ/kT ≤ 1 the chemical potential satisfies
µ ≤ 0. This is in contrast with a system of fermions, for which µ can be
arbitrarily large and positive (at sufficiently low temperature or high
density).
At T = 0 in a bose gas only the ground state would be occupied,
i.e. n̄0 = N . As T increases, the ground state population would be
depleted. For T ≥ 0, we parametrise the ground state occupation
number by the fraction x = n̄N0 . Of course, at T = 0 we have x = 1.
Let us now consider two regimes :
z ∼ 1+O1 1 ∼ 1 − O N1 for N  1


(i) x ∼ 1, n̄0 ∼ O(N ),
(N )
(ii) x ∼ 0, n̄0 ∼ O(1), 0≤z<1
In the 2nd regime we can use the expansions
∞
p 1 X zl
= 3
kT a l=1 l 52
∞
1 X zl
ρ = 3
a l=1 l 32
48 3. SYSTEM OF FREE BOSONS
q
2π
where a = mkT
is the thermal de Broglie wavelength ( we are as-
suming a system of nonrelativistic spinless particles) to calculate the
equation of state. the expansion certainly converges for z < 1. In the
regime (i) , however, z ∼ 1 and we expect a singular behaviour2. To
p
analyze the nature of the singularity, we need expressions for ρ and kT
which are accurate near the singularity (z ∼ 1, x ∼ 1). Let us asume,
for simplicity, that only the ground state is macroscopically populated3,
and seperate out it’s contribution :
Z
n̄0 1
ρ= + 3 d3 k n̄~k
V 8π
Since we are approaching the singularity at z = 1 from below, we can
still use the low density expansion for the second term. thus
∞
1 X zl
ρ = xρ + 3
a l=1 l 32
Since ε0 = 0, even nif it is macroscopically populated, the ground state
does not contribute to the energy density and hence to the pressure.
So ∞
p 1 X zl
= 3
kT a l=1 l 52
To consider the consequence of taking the macroscopic population
of the ground state, let us consider a free boson gas in which the temper-
ature is fixed and we change the density. At sufficiently small density
we have
 
3 1 3
z = ρa 1 − 3 ρa + . . .
22
 
p 1 3
= ρ 1 − 5 ρa + . . .
kT 22
2n̄ has a singularity at z = 1, which is however reached only when N → ∞
0
3To see that this is not a bad assumption, let us consider a system enclosed in
a periodic box of linear dimension L. Some of the low lying states, other than
the ground state will also be macroscopically populated. Let us consider the first
excited state with energy ε ∼ L−2 . In the regime (i) z ∼ 1 means
 
1 1  
n̄ε ∼ ε/kT ∼O ∼ O(L2 ) ∼ O N 2/3
e −1 ε
But V ∼ N - thus the contribution to the density of these states is ρε ∼ O N −1/3 .

Thus for N → ∞, the correction is indeed small.Even if we include the effect of all
the low lying states that are
macroscopically populated, the net correction due to
these terms is still O N −1/3 .
 3.1. BOSE CONDENSATION 49

z is small and we are still far away from the singularity. The system
behaves like an ideal gas. As ρ increases z → 1 and we are approaching
the singularity. At sufficiently high density, we get on the top of the
singularity.
The onset of the singularity can be understood from an estimate
of the critical density. this critical density ρc is the maximum possible
density for which x is still ∼ 0 and z ∼ 1 − δ with δ  0. Thus
∞
3
X 1
ρc a = 3 ≈ 2.612
l=1 l2
and the corresponding critical pressure is given by
∞
p c a3 X 1
= 5 ≈ 1.341
kT l=1 l 2

Both the critical density and the critical pressure are of course functions
of the temperature
3 5
ρc ∝ T 2 p c ∝ T 2
If we increase ρ beyond ρc , p re- Figure
mains constant at pc . This is because 3.1.1. Isotherm
z has already reached it’s maximum for a free bose gas
value of 1 and cannot increase any
further. So the isotherm has the form as shown in the figure (3.1.1).
As ρ increases beyond ρc , more and more particles condense into the
zero momentum ground state. This is known as Bose-Einstein conden-
sation. Since the condensed partickle do not contribute to the energy
density, the pressure remains constant.This is precisely a phase transi-
tion. Since for the phase transition to start, we must have z = 1 which
is possible only when N → ∞4 this occurs only in the thermodynamic
limit. This is a general feature of phase transitions.
The nature of this transition is quite different from the ordinary
gas-liquid phase transition. In this case the isotherm does not have a
branch corresponding to the liquid state. Only analogues of the gas
phase and the two phase region exist.

4For a finite system z is always bounded above by N

1+N
50 3. SYSTEM OF FREE BOSONS

For ρ < ρc (T ) the condensate fraction x = 0. For ρ ≥ ρc (T ) this

fraction can be calculated from
∞
1 X 1
ρ = xρ + 3 = xρ + ρc
a l=1 l 32
to get
ρc
1−x=
ρ
mk 3/2
Since ρc = 2.612


a3
= 2.612 2π
T 3/2 we define a critical temperature
for a given density by
 3/2
mk
ρ = 2.612 Tc3/2
2π
which leads to 3 
T 2
x=1−
Tc
If we keep ρ fixed and change T , condensate fraction is maximum at
T = 0 and vanishes at T = Tc .

3.1.1. Discontinuity of thermodynamic functions in BEC.

Pressure p is continuous across the critical point. Let us check the
∂p
where v = ρ−1 is the specific volume. In the two


continuity of ∂v T
∂p


phase region ∂v T+
is zero. Let us now calculate the value of this
∂p


derivative ∂v T − in the “gas” phase, near the critical point.
Since here we have ρ < ρc , z < 1 and x = 0 we have
∞ ∞
p 1 X zl 1 X zl
= 3 ρ= 3
kT a l=1 l 52 a l=1 l 32
∞ ∞
1 X zl 1 X zl
   
1 ∂p ∂ρ
= 3 = 3
kT ∂ ln z T a l=1 l 32 ∂ ln z T a l=1 l 12
  ∞   ∞
1 ∂p 1 X 1 ∂ρ 1 X 1
= =
kT ∂ ln z T − a3 l=1 l 32 ∂ ln z T − a3 l=1 l 12

Thus ∂ ∂p ∂ρ



ln z T −
is finite but ∂ ln z T −
diverges. Hence
∂p
 

∂p 2 ∂ ln z T −
= −ρ ∂ρ

=0
∂v T− ∂ ln z T −
∂p


Thus, there is no discontinuity in ∂v T
at the bose einstein condensa-
tion point.
 3.1. BOSE CONDENSATION 51
 2 
∂ p
Let us now examine the second derivative ∂v 2 . This is of course
T
again zero in the two phase region. Since, as we have already seen,
∂p


∂ ln z T
= kT ρ we get
 2  " #
∂ p ∂ kT ρ3
= − ∂ρ

∂v 2 T ∂v ∂ ln z T T
" #
ρ2 ∂ kT ρ3
= − ∂ρ
− ∂ρ

∂ ln z T
∂ ln z ∂ ln z T T
  2  
3 ∂ ρ
∂ρ kT ρ
2


ρ2  3kT ρ ∂ ln z T ∂ ln z 2
T
= −
∂ρ ∂ρ ∂ρ 2




∂ ln z T ∂ ln z T ∂ ln z T
 
∂2ρ
3kT ρ4 ∂ ln z 2
=
− kT ρ5 T
∂ρ ∂ρ 3


∂ ln z T ∂ ln z T

Since ∂ ∂ρ


ln z T −
diverges, the first term in the above expression vanishes
as
 we approach Tc . However, the second term
  in2 this
 expression has
∂2ρ ∂ ρ 1
P∞ 1 l
∂ ln z 2
in the numerator and this is given by ∂ ln z2 = a3 l=1 l 2 z
T T
1
and as we approach Tc this tends to a−3 ∞
P
l=1 l which is clearly diver-
2

gent. Thus in  2 
 2  ∂ ρ
∂ p ∂ ln z 2
T−
= −kT ρ5
∂v 2 T − ∂ρ 3


∂ ln z T −
we have the ratio of two divergent quantities. In order to ascertain the
behavioru of the second derivative
 near Tc we must carefully examine
∂2ρ ∂ρ


the nature of the divergences in ∂ ln z2 and ∂ ln z T − , respectively.
T−
Thus the problem reduces to finding out the behaviour of the sums
P∞ 1 l P∞ − 1 l
l=1 l z and l=1 l
2 2 z near z = 1.

To do this we start from the integral representation of the sum5

which does converge :
∞ Z ∞ 2
X zl 4z x dx
3 = √
l=1 l
2 π 0 ex2 − z

5This is connected with the integral representation of the Riemann zeta function
∞ Z ∞ 2
X 1 4 x dx
ζ(s) = = √
l s π 0 e 2 −1
x
l=1
52 3. SYSTEM OF FREE BOSONS

We differentiate this with respect to ln z to arrive at

∞ Z ∞ " #
X zl 4z 1 z
1 = √ dx x2 x2 + 2
l=1 l 2 π 0 e − z (ex2 − z)
As z → 1 the first term in the integral remains finite, while the second
term diverges. Writing z = 1 − ε and retaining the diverging term only,
we get
∞ Z ∞
X zl 4 x2
1 −→z→1 √ dx 2 2
l=1 l
2 π 0 (ex − 1 + ε)
The dominant behaviour in this integral comes from the fact that as
ε → 0 the integrand diverges near x = 0. So we can replace the
integrand by it’s expansion near x = 0 :
∞ Z ∞
X zl 4 x2
−→ z→1 √ dx
(x2 + ε)2
1
l=1 l
2 π 0
Z ∞
4 y2
= √ dy
πε 0 (y 2 + 1)2
Z π
4 2
= √ dθ sin2 θ
πε 0
r r
4 π π π
= √ = =
πε 4 ε 1−z
Taking derivative with respect to ln z again gives
∞
1 π
r
1
X
l
l 2 z −→z→1
l=1
2 (1 − z)3

Now,
∞ ∞
x2 dx −1
Z Z 
2 2
= dx x2 e−x 1 − ze−x
0 e 2 −z
x
0
∞ Z ∞
X 2
= z l−1
dx x2 e−lx
l=1 0
∞ l−1
4 X z
= √ 3
π l2
l=1
Thus we have
∞ ∞
zl x2 dx
Z
X 4z
3 =√
l 2 π 0 ex2 − z
l=1
 3.2. ISOCHORES 53

Thus, the divergent derivatives that we need are given by

  r
∂ρ 1 π
=
∂ ln z T − a3 1 − z
 2 
∂ ρ 1 π
r
=
∂ ln z 2 T − 2a3 (1 − z)3
and hence
1 √
∂2p kT ρ5 a6
 
5 2a3 π
= −kT ρ 1 √ =−
∂v 2 T− a9
π3 2π
 2 
∂ p
So the second derivative ∂v 2 is discntinuous at the critiacal point,
T
the amount of discontinuity being
 2   2   2 
∂ p ∂ p ∂ p kT ρ5c a6
∆ = − =
∂v 2 T ∂v 2 T + ∂v 2 T − 2π

3.2. Isochores

The isochores of a free boson gas are most easily obtained by study-
ing the isothermsfor various values of temperature T , at a fixed specific
volume v (i.e at a fixed density).
3
Recall that the critical density obeys ρc ∝ T 2 . This means that
3
the critical specific volume vc = ρ−1 c ∝ T − 2 . Hence as one goes to
higher temperatures the critical point on the (p-v) diagraqm moves
to the left. In the figure (*), the points α,β and γ are the points of
intersection of a line at constant specific volume with three different
isotherms. As can be seen, α is in the gas phase, β is just on the
critical point, whereas γ is in the two phase region. We obtain the
isochore for that particular density py plotting the pressures at the
points of interaction against corresponding temperature labels of the
isotherms. We also show corresponding points α,β and γ. The point
β corresponds to the critical point. In the region below β, we are in
5
the two phase region where p ∝ T 2 independent of the density. Above
β, we are in the gas phase and here the pressure is a function of both
the temperature and the density. For very high temperatures we have
the classical ideal gas result p = ρkT . If we now consider the isochores
for various values of density, we find that they have identical behaviour
5
in the region below the corresponding critical point ( p ∝ T 2 - with
the constant of proportionality independent of the density). As we
increase the density, of course the critical point moves towards higher
54 3. SYSTEM OF FREE BOSONS
5
temperatures. However, the region lying above the p ∝ T 2 curve is
never reached - it is completely unphysical.6Above the critical point,
5
the isochores deviate from the p ∝ T 2 behaviour.

3.2.1. Discontinuities in the isochores at the BEC tran-

sition point. Since we are considering isochores here, we will now
consider the temperature derivatives of pressure. Let us first consider
∂p

5
∂T v
. In the two phase region, p ∝ T 2 and thus
 
∂p 5 pc
=
∂T v+ 2 Tc
∂p


at the critical point. Let us now consider ∂T v−
, the value of the
derivative as we approach the critical point from the gas phase. To
this end, we note that
       
∂p ∂p ∂p ∂ ln z
(3.2.1) = +
∂T v ∂T z ∂ ln z T ∂T v
and recall that
∞
p 1 X zl
= 3
kT a l=1 l 52
∞
1 X zl
ρ = 3
a l=1 l 32
5
Thus p ∝ T 2 at fixed z
 
∂p 5 pc
=
∂T z− 2 Tc
at the critical point as z → 1−. Again,
 
∂p
= ρkT
∂ ln z T
and ∞
3ρ 1 X zl
dρ = dT + 3 d (ln z)
2T a l=1 l 12

6Contrast this, once again, with the liquid-gas phase transition. There the region
above the two phase coexistence line corresponds to the liquid state. The unphysical
region that vwe speak of here drives home the point that there is no analogue to
the liquid state in BEC.
 3.2. ISOCHORES 55

Thus,     3 ρ
∂ ln z ∂ ln z
= = − 1 P2 ∞
T
zl
∂T v ∂T ρ a3 l=1 12
l
which vanishes as z → 1− (since the denominator diverges). Thus
 
∂p 5 pc 5 pc
= + ρkTc × 0 =
∂T v− 2 Tc 2 Tc
∂p ∂p ∂p




and hence ∂T v−
= ∂T v+
and so ∂T v
is continuous across the
critical point.  2 
∂ p
We next consider the discontinuity in ∂T 2 :
v
 2   2   2 
∂ p ∂ p ∂ p
∆ 2
= 2
−
∂T v ∂T v+ ∂T 2 v−
From (3.2.1) we get
 2          
∂ p ∂ ∂p ∂ ∂p ∂ ln z
= +
∂T 2 v ∂T ∂T z v ∂T ∂ ln z T ∂T v v
Now,     
∂ ∂p 5 p 5 ∂p
=− 2 +
∂T ∂T z v 2T 2T ∂T v
∂p


Since both p and ∂T v
are continuous across the critical point, we have
  
∂ ∂p
∆ =0
∂T ∂T z v
and are thus left with
 2       
∂ p ∂ ∂p ∂ ln z
∆ = ∆
∂T 2 v ∂T ∂ ln z T ∂T v v
     
∂ ∂p ∂ ln z
= −
∂T ∂ ln z T ∂T v v−
But,
3
ρa3 ρ 2 a3
   
∂p ∂ ln z 2 3
= ρkT × − P = − k P∞ zl
∂ ln z T ∂T v T ∞l=1
zl
1
2 l=1 1
l2 l2
56 3. SYSTEM OF FREE BOSONS

and therefore
ρ2
     
∂ ∂p ∂ ln z 3
= − k P∞ zl
∂T ∂ ln z T ∂T v v 2 l=1 12
l
P∞ l 1  
3 2 3 l=1 z l ∂ ln z
2
+ kρ a hP i2
2 ∞ zl ∂T v
l=1 1
l2

As z → 1−,
∞
zl
r
X π
1 →
l=1 l 2 1−z
∞ r
X
l 1 1 π
zl 2 →
l=1
2(1 − z) 1 − z
r
3 ρa3 1 − z
 
∂ ln z
→ −
∂T v 2 T π
So, in this limit
      r
∂ ∂p ∂ ln z 3 2 3 1 π
→ kρ a ×
∂T ∂ ln z T ∂T v v 2 2(1 − z) 1 − z
  3r
1−z 3 ρa 1−z
× × −
π 2 T π
9 ρ 3 a6
k= −
8π T
And thus the discontinuity in the second derivative is
 2 
∂ p 9 ρ3c a6c
∆ = k
∂T 2 v 8π T
3.2.2. Behaviour of the heat capacity at the BEC phase
transition point. The above results derived for the derivatives of
pressure at the BEC transition can be used to deduce the behaviour of
the heat capacity using the fact that for any non-relativistic system of
free particles E = 23 pV . Thus
   
∂E 3 ∂p
CV = = V
∂T v 2 ∂T v
and so  
3 ∂p
∆CV = V ∆ =0
2 ∂T v
 3.3. LIQUID He4 57

across the phase transition point. But

 2 
ρ3c a6c
 
∂CV 3 ∂ p 27
∆ = V∆ = V k
∂T v 2 ∂T 2 v 16π Tc
But ρc a3c = 2.612 and thus
 
∂CV 27 ρc 27 Nk
∆ = kV × (2.612)2 = (2.612)2
∂T v 16π Tc 16π Tc
Thus the CV versus T curve is continuous at T = Tc , but it has a kink
at this point.
5
Below the transition temperature Tc , p ∝ T 2 . Therefore, in this
3
region CV ∝ T 2 independent of the density. On the other hand, at very
high temperatures, p = ρkT and thus CV approaches it’s asymptotic
value CV = 32 N k the classical value.

3.3. Liquid He4

So far we have been discussing the behaviour of an ideal Bose fluid,

where we have a transition to the condensate state at a sufficiently
low temperature. The real life fluid, liquid He4 also shows a phase
transition - to the superfluid phase. The experimental specific heat
versus temperature curve does exhibit a discontinuity in the slope at the
transition point. But this is a logarithmic singularity - a characteristic
of interacting systems. This singularity is also present in other systems
exhibiting phase transitions (for example in the 2-dimensional Ising
model).
Interaction essentially modifies the spectrum of low lying states of
the system. In an interacting Bose system, the elementary excitations
for small momentum k are always phonons, i.e. ε ∝ k. This feature
is essential for superfluidity. Any external probe going through the
medium, would Cherenkov radiate phonons, and thereby lose energy,
only if it’s velocity exceeds the speed of sound in the medium. Oyther-
wise it will move unimpeded and the medium acts like a superfluid.
A noninteracting ideal bose gas does not exhibit superfluidity. Ex-
istence of a macroscopically occupied condensate is not sufficient for
this phenomenon. One must have interactions to modify the low lying
elementary excitations.
The p − T phase diagram of real helium does not bear any resem-
blance to that of the ideal bvose gas that we have considered. However
58 3. SYSTEM OF FREE BOSONS

at p = 1 atmosphere, the calculated Tc for an ideal bose gas of “He-

lium” atoms is ≈ 3.4◦ K which is close to the normal fluid - superfluid
transition temperature ∼ 2.2◦ K. This shows that the condensate may
have something to do about the superfluid transition of He4 .
In a fermi gas, there is no condensation and superfluidity in the
absence of interactions. In the presence of interactions, however, bound
pair excitations are formed. These bound pairs behave like bosons, and
we may ahain have superfluidity. This superfluidity manifests itself as
superconductivity. This seems to be a very general trend in interacting
fermi systems.
 CHAPTER 4

Classical statistics

Consider a system of N interacting identical atoms or molecules.

The Hamiltonian of the system is
X  p2 
i
H= + ui + U (~r1 , . . . , ~r2 )
i
2m
where ui is the internal energy of the i-th atom or molecule. This is
the sum associated with internal energies of rotation, vibration and
electronic excitation. U (~r1 , . . . , ~r2 ) is the potential energy of the sys-
tem (which may include contributions from many body forces). The
coupling of U (~r1 , . . . , ~r2 ) with the rotational and vibrational degrees of
freedom will always be rather small - the coupling to electronic degrees
of freedom may not be. If we confine ourselves to low lying excitations,
however, we may neglect this coupling.
In general, exact diaonalization of H is impossible. So we try to
constriuct approximate eigenstates. In this problem, there ar4e essen-
tially three different length scales
q
• Thermal wavelength, a = mk2πB T ~
1
• Interparticle seperation, d = ρ− 3
• Scale associated with potential U (~r1 , . . . , ~r2 ), λ i.e. it’s ef-
fective range or scattering length (these determine the length
over which the potential changes significantly).

At sufficiently high temperatures and low densities, we always have

a  d and a  λ. this is essentially the classical limit ~ → 0.
Since a  d, we can always choose a length l such that a  l  d.
Now divide the whole volume in smaller volumes of the size ∼ l3 . We
can now construct plane wave solutions confined in each small vol-
ume. Let ψα~k (~r) be the normalized plane wave of momentum ~k in
 
the αth volume. Then ψα~k (~r) is proportional to sin ~k · ~r − R
~ α or
 
cos ~k · ~r − R
~ α inside the αth volume and 0 outside it, where R ~ α is

59
60 4. CLASSICAL STATISTICS

the coordinate associated with the center of the αth volume. Prop-
erly symmetrized (antisymmetrized) products of these wavefunctions
ψα~k (~r) form a complete set of states for a system of bosons (fermions).
The system wavefunction Ψ (~r1 , . . . , ~rN ) can be expanded in terms of
this complete set.
Upto this point, everything we have done is exact. Now recall that
l  a, so that a particle can be localized within a particular small
volume. Again, since d  l the chances of two or more particles being
localized within the same volume are extremely remote. So, of all the
configurations, the ones in which each particle occupies a seperate small
volume will be the dominant one. The corresponding wavefunction
(after suitable syymetrization or antisymmetrization) takes the form
1 X Y
Ψ (~r1 , . . . , ~rN ) = √ (±)P ψαi~ki (P ~ri )
N! P i

where the + amd - signs are for bosons and fermions, respectively.
The summation is over all possible perutations of particle coordinates.
Since there is no overlap between particles belonging to different small
volumes, and ψαi~ki (~r) are normalized, the wavefunction Ψ will be nor-
malized. We now calculate hΨ| H |Ψi and replace the partition function
of the system Tr e−βH by a sum over terms of the form e−βhΨ|H|Ψi . This
is our essential approximation1.
To begin with, let us assume that the internal degrees of freedom
are absent, i.e.
X p2
i
H= + U (~r1 , . . . , ~r2 )
i
2m
Since λ  a, we can always choose our l so that l  λ. In that case
   
~ ~ ~ ~
hΨ| U (~r1 , . . . , ~rN ) |Ψi ≈ U R1 , . . . , RN hΨ| Ψi = U R1 , . . . , RN

and thus
X k2  
hΨ| H |Ψi = i ~ ~
+ U R1 , . . . , RN
i
2m
To calculate the partition function, we will have to sime over all possible
momentum assignments and positions of the small volumes “occupied”
1This does not mean that we are claiming that Ψ is an approximate energy eigen-
state. To calculate Tr e−βH , all we need is a complete set of states which need not
be eigenstates of H. The products of the plane waves localized in small volumes is
a complete set. All we are assuming is that the sum in the trace is dominated by
contribution from one particular state Ψ of this complete set.
 4. CLASSICAL STATISTICS 61

by ~
 the atoms or molecules. total number
 of states
 for Ri in the range
~ i, R
R ~ i and ~ki in the range ~ki , ~ki + d~ki is
~ i + dR
QN QN
i=1 d3 Ri i=1 d3 ki
N
N ! (2π~)
Here, the factor of N ! is essential to avoid double counting while sum-
ming over all possible small volumes. Clearly, some of the changes in
position (assuming we have already summed over all possible momen-
tum labels), would result in mere interchange of the small volumes.,
and therefore not produce a new state (recall that the particles are
identical). There are in fact N ! such permutations, hence the factor of
N ! in the denominator.
So the contribution to the grand partition function from all N par-
ticle configurations (apart from the factor of eβµ ) is :
Z Y 3 " #!
1 d Ri
Z Y X k2  
QN = d3 ki exp −β i
+U R ~ 1, . . . , R
~N
N! (2π~)N 2m
i i i

If we now take the internal energies ui into account, we would get

one extra factor X
q= e−βui
for each atom or molecule. This factor will be independent of i, since
all the atoms or molecules are identical. The summation here is over
all internal degrees of freedom for a given constituent.
Taking the factors q into account, and carrying out the trivial
Gaussian integrals over the momenta, we have
Z
1  q N Y 3
QN = d Ri e−βU (~r1 ,...,~rN )
N ! a3 i

so that the grand partition function is

∞
X
Q= z N QN
N =0
z
We introduce the parameter y = a3
q , which is also called the fugacity.
In terms of y, we have
∞
X yN
Q= QV (N )
N =0
N !
62 4. CLASSICAL STATISTICS

where the configuration integral QV (N ), often called the spatial or

volume partition function, is defined as
Z Y N
QV (N ) = d3 Ri e−βU (~r1 ,...,~rN )
i=1

So to calculate Q we need calculate only the configuration integrals

QV (N ). Later, we will develop a systematic method for calculating
these integrals.

Example : Classical ideal gas

In this case U = 0 since there is no interaction. The configuration

integral can be evaluated trivially to be QV (N ) = V N where V is the
volume of the system. Thus
∞
X yN N
Q= V = eV y
N =0
N !
Lets see what this tells us about the thermodynamics of this sys-
tem. Since, Q = eβpV , we have kBpT = y. Again, the density2 is
 
∂ p
ρ = ∂(ln y) kB T = y and so the equation of state is

p = ρkB T

The Gibbs free energy is

a3 y a3
   
G ≡ N µ = N kB T ln z = N kB T ln = N kB T ln p+N kB T ln
q kB T q
And thus, the entropy is
   
∂G G 5 ∂
S=− = − + N kB + N kB T ln q
∂T P T 2 ∂T p

a3 1
(recall that kB T q
∝ 5 ). Thu8s the total energy is
qT 2
 
3 ∂
E = G + T S − pV = N kB T + N kB T 2 ln q
2 ∂T p

 
2We know that ρ = ∂ p
but y = q
implies ln y = ln z + ln aq3 and hence
∂(ln z) kB T a3 z
 
∂ p
ρ = ∂(ln y) kB T .
 EXAMPLE : CLASSICAL IDEAL GAS 63

From the equipartition theorem, the first term is just the average kinetic
energy. As far as the second term is concerned, we see that
 
∂ 1 X
N kB T 2
ln q = N kB T 2 × 2
ui e−βui
∂T p qkB T int. d.o.f.
P −βui
ui e
= N P −βui = N hui
e
where hui is the average internal energy. So, for a classical ideal gas,
we have
3
E = N kB T + N hui
2
Gibb’s Paradox. Using the equation of state pV = N kB T for the
ideal gas, we get
 
G 5 ∂
S = − + N kB + N kB T ln q
T 2 ∂T p
 3   
a 5 ∂
= −N kB ln p − N kB ln + N kB + N kB T ln q
kB T q 2 ∂T p
   3   
N kB T a 5 ∂
= −N kB ln − N kB ln + N kB + N kB T ln q
V kB T q 2 ∂T p
V
N k ln + N f (T )
N
where
a3
   
5 ∂
f (T ) = kB + kB T ln q − kB ln kB T − kB ln
2 ∂T p kB T q
is a function of the temperature only.
Now consider two different perfect gases A and B. The inter-
moleculer forces are all equal, UAA = UBB = UAB (In the case of ideal
gases all are identically zero). Their atomic (or molecular) masses are
also the same. Despite this, mthe two gases are differnt because they
are observed to be different - i.e. there is some probe to which they
couple differently. The particles that make up the two gases are thus
distinguishable. Truly identical particles cannot be distinguished and
the total wavefunction must in their case be totally symmetric or anti-
symmetric (no matter what the forces are).
Enclose the gases A and B in boxes of volumes VA and VB , in
equilibrium at a temperature T . Let the number of gas molecules NA
and NB be such that NVAA = NVBB , i.e. the densities in the two boxes are
64 4. CLASSICAL STATISTICS

the same. So the total number of particles N = NA + NB and total

volume V = VA + VB .
Assuming that there is no intermolecular interaction, the initial
total entropy is
VA VB
Si = NA kB ln + NB kB ln + (NA + NB ) f (T )
NA NB
- the function f (T ) is the same for the two gases because the internal
structure and masses are the same.
If we now remove the partition between the two boxes, the gases
will mix. Let us calculate the entropy of the final mixture. Since there
is no interaction, the total Hamiltonian is

H = HA + HB

and therefore

QN = Tr e−βHA Tr e−βHB
= QA (NA , V, T )QB (NB , V, T )

Hence the total free energy

F = FA (NA , V, T ) + FB (NB , V, T )

which means that

S = SA (NA , V, T ) + SB (NB , V, T )

and so, the final entropy is

V V
Sf = NA kB ln + NB kB ln + (NA + NB ) f (T )
NA NB
Since V is larger than both VA and VB we find that Sf > Si as we
would have expected from elementary thermodynamics. The so called
“entropy of mixing” is given by
 NA +NB 
V
Sf − Si = kB ln
VANA VBNB
If we had started with VA = VB and NA = NB the the entropy of
mixing is easily seen to be

Sf − Si = N kB ln 2
 EXAMPLE : CLASSICAL IDEAL GAS 65

For truly identical (indistinguishable) gas molecules,

VA VB
Si = NA kB ln + NB kB ln + (NA + NB ) f (T )
NA NB
and
V
Sf = N kB ln + N f (T )
N
NA NB N
But VA
= VB
= V
and hence

Si = Sf

Classical statistical mechanics did not take into account the indis-
tinguishability of molecules - and hence the factor N ! was missing from
the denominator of the expression for QN . This in turn, means that
the expression for the entropy did not have the −N ln N term. Thus,
when Gibb’s originally calculated the entropy of mixing for two gases,
he had

Si = NA kB ln VA + NB kB ln VB + (NA + NB ) f (T )
Sf = N kB ln V + N f (T )

and hence Sf > Si and the entropy of mixing is given by

 NA +NB 
V
Sf − Si = kB ln
VANA VBNB
which is the same expression as before. Trouble is, this is true eeven
when the two gases are identical, which files in the face of the extensive
nature of entropy! This was the celebrated Gibb’s paradox. Gibb’s
had realized that one way of avoiding this problem could be introducing
an ad hoc factor of N ! in the denominator of the classical expression
for the partition function - it was only quantum statistical mechanics
with it’s concept of indistinguishibility that could explain the presence
of this factor.

Calculation of internal partition function.

Monoatomic gases. In monoatomic gases, the only internal degrees
of freedom we have are the electronic degrees of freedom, ui = uele .
To see the effect of these electronic degrees of freedom on the internal
partition function, q, let us first denote the degeneracies of the elec-
tronic ground and first excited states by ω0 and ω1 , respectively. Let
the excitation energy of this excited state (above the ground state) be
66 4. CLASSICAL STATISTICS

u1 . Then
q = ω0 + ω1 e−βu1 + . . .
1
Now, typically u1 ∼ 1 ev and at room temperature kB T ∼ 40
ev. So,
for the first and higher excited states

e−βui  1 for i ≥ 1

Naively, then, one would expect the contributions of the excited states
to the internal partition function to be negligible. As it turns out, this
expectation fails at very low densities!
To see why, let us consider an assemly of Hydrogen atoms. The
spectrum of Hydrogen is well known to be
R
En = −
n2
where R is the Rydberg constant. So the excitation energy to the n-th
level is given by
R
un = R − 2
n
Since the degeneracy of the n-th state is n2 the internal partition func-
tion for a hydrogen atom is given by
∞
2
X
−βR
q=e n2 e−βR/n
n=1

This sum is divergent!!! As can be seen, for large n, the Boltzmann

weight stays almost fixed while the degeneracy keeps on growing with-
out bound - so the summand goes as n2 for large n. This seems to
say that a single hydrogen atom should always auto-ionize! In other
words, at room temperature, the probability that hydrogen will go into
one of it’s highly excited states is very small, but there are so many
of them that the probability of it going into any one of these states is
immense! This trend should be typical of all gases, whereas we know
from experience that the gases auto-ionize very rarely under ordinary
conditions.
This anomaly can be resolved by noting that the divergence exists
because of the overwhelmingly large phase space associated with large
orbits. In an assembly of atoms, however, the largest permissible orbit
will have a radius r which is of the order of the interparticle separation
ρ−1/3 . Since r = n2 a0 where a0 is the Bohr radius, an estimate of the
 EXAMPLE : CLASSICAL IDEAL GAS 67

largest permissible value of n, n0 is given by

n20 a0 = ρ−1/3

Thus, the electronic partition function is then

n0
X βR
qele = 1 + n2 e n2 e−βR
n=2

and the total contribution of the excited states to the electronic parti-
tion function is negligible only when
n0
X βR
n2 e n2 e−βR  1
n=2

For this to be valid, n0 must be small. A good criterion for this could
be
n20 e−βR  1
which means
e−βR
1  1
a0 ρ 3
Hence we can ignore excited states only at sufficiently large densities or
low temperatures. At room temperature, this is a valid approximation
for all gases at normal densities. On the other hand, in almost all
stellar configurations the density is so small and the gas so hot that
the excited state contribution is much larger than that of the ground
state. As a result stellar gases are almost always completely ionized.
Diatomic gases. In this case we have to consider rotational and
vibrational motion as well as the electronic degrees of freedom. The
total internal energy, then is

ui = urot + uvib + uele

And the internal partition function is

q = qrot qvib qele

Now, uvib = n~ω with n = 0, 1, 2, . . . and where ω is the vibrational

frequency3. In this we have assumed that the temperature is sufficiently
low that the vibrations be small, and hence harmonic. Since for the
diatomic molecule the vibrational mode is one dimensional, there is no
3Since we are only interested in excitation energies, we have neglected the zero
point energy of 12 ~ω
68 4. CLASSICAL STATISTICS

degeneracy and so
∞
X 1
qvib = e−βn~ω =
n=0
1 − e−β~ω
Let us now consider the case of a few typical gases. For H2 the
value of ~ω
kB
is 5798 K while that for HCl is 4031 K. This shows that at
room temperature
• Anharmonicity is negligible and
• Vibrational states are hardly excited!
On the other hand, rotational levels are given by
j(j + 1)~2
urot = , j = 0, 1, 2, . . .
2I
where I is the moment of inertia of the molecule. Corresponding eigen-
functions are the rotational spherical harmonics. Degree of degeneracy
of each rotational state is 2j + 1 and each has a parity of (−1)j . Again,
~2
typical values of 2Ik B
are 85 K for H2 and 15 K for HCl. Thus, unlike
the case with vibrations, rotational levels are easily excited at room
temperature.
Homonuclear molecules. To investigate the effect of the rotational
levels, we first consider molecules with identical atoms (e.g. H2 ). In the
ground state the electron spins are antiparallel4. We assume sufficiently
high densities and low temperatures that the excited electronic states
can be neglected. The electronic ground state is nondegenerate and
antisymmetric. As, we have seen, we can also ignore the excited vibra-
tional levels at room temperature. We now consider the nuclear spin
states. Let’s assume that the nuclear spins are 21 (which is certainly
the case for H2 ). Combined spin states can be
• SN = 1 - triplet (degeneracy = 3) . The spin wave function
is symmetric and hence the space wave function has to be
antisymmetric, i.e. j = 1, 3, 5, . . ..

4This is because in order to bind and hence reduce the electrostatic energy, the
electron wavefunctions must have large amplitudes and overlap at the center of
the molecule. Because of the Pauli exclusion principle, the overlap at the center
is large only when the spins are antiparallel. This argument, admittedly, is rather
heuristic - but is works nonetheless. A more precise argument would involve the
sign of the exchange integral. For diatomic molecules the sign is such that the
antiferromagnetic order is favored.
 EXAMPLE : CLASSICAL IDEAL GAS 69

• SN = 0 - singlet (degeneracy = 1) . The spin wave function

is antisymmetric and hence the space wave function has to be
symmetric, i.e. j = 0, 2, 4, . . ..
The SN = 0 and 1 states are called para hydrogen and ortho hydrogen,
respectively. The ortho para transition requires the flipping of nuclear
spins, which is hard to do. So, the transition rate is quite small and
wwe can have pure para-hydrogen, pure ortho-hydrogen or sometimes
a statistical mixture of the two.
For ortho-hydrogen,
 
X j(j + 1)
qortho = 3 (2j + 1) exp −
j=odd
2IkB T

where the factor of 3 comes from the degenercay of the nuclear spin
states. On the other hand, for para-hydrogen
 
X j(j + 1)
qpara = (2j + 1) exp −
j=even
2IkB T

At low temperatures, we need to take just a few terms in each sum

qortho = 3 3e−1/IkB T + 7e−6/IkB T + . . .

qpara = 1 + 5e−3/IkB T + . . .

To find the high temperature expansion, we define x = j(j+1)

2IkB T
. At
high temperatures, x varies almost continuously from term to term in
the sum. The change in x between two successive terms, differing by
∆j, is given by
(2j + 1)∆j
∆x =
2IkB T
Since ∆j = 2 between successive terms of each sum, we have
1 IkB T
1 = ∆j = ∆x
2 2j + 1
we thus have the replacement
Z
X dx
→ IkB T
2j + 1
j=even/odd

Thus, for para hydrogen at high temperatures

Z ∞
qrot ≈ IkB T dxe−x = IkB T
0
70 4. CLASSICAL STATISTICS

and similarly, for ortho

qrot = 3IkB T
Finally, for a statistical mixtre of ortho and para hydrogen,

qrot = 4IkB T

and the average rotational energy is

∂
hurot i = kB T 2
ln qrot = kB T
∂T
This is entirely to be expected from the fact that there are two rata-
tional degrees of freedom, since the equipartition theorem is always
valid for high temperatures.
Heteronuclear molecule. In the case of molecules like HCl, there are
no symmetry requirements and so
∞  
X j(j + 1)
qrot = (2sA + 1) (2sB + 1) exp −
j=0
2IkB T

where sA and sB are the two nuclear spins. At high temperatures this
tends to
qrot = 2 (2sA + 1) (2sB + 1) IkB T
Corrections to these high temperature results can be calculated by using the Euler Mclurin
formula :
n Z n
X 1
f (j) = f (j)dj − [f (n) − f (m)] −
m m 2
X (−1)l Bl h (2l−1) i
f (n) − f (2l−1) (m)
(2l)!
l≥1

where B1 = 61 , B2 = 1
30
,...are the Bernoulli numbers.
X X
qrot = (2j + 1)e−urot /kB T
j
nuclear
spin
X X
= (2j + 1)e−εj(j+1)
j
nuclear
spin

~2
where ε = 2IkB T
. Consider first the heteronuclear case A 6= B. Total number of nuclear spin
states is ω = (2sA + 1) (2sB + 1). In this case there are no exchange symmetry requirements
on the wavefunction, and so all j values are allowed. Anticipating the homonuclear case,
however, we split the sum over j into one over even j and another over odd j.
X ∞
X ∞
X
(2j + 1) e−εj(j+1) = (4m + 1) e−2εm(2m+1) = φ(m)
j=even m=0 m=0
 EXAMPLE : CLASSICAL IDEAL GAS 71

To calculate this sum using the Euler Mclurin formula, we need :

φ(m) = (4m + 1) e−ε2m(2m+1) φ(0) = 1 φ(∞) = 0

φ(1) (m) = 4 − 2ε(4m + 1)2 e−ε2m(2m+1) (1)
φ(1) (∞) = 0
 
φ (0) = 4 − 2ε
φ(2) (m) = [−24ε(4m + 1)] e−ε2m(2m+1) + O(ε2 ) φ(2) (0) = −24ε φ(2) (∞) = 0
φ(3) (m) = −96εe−ε2m(2m+1) + O(ε2 ) φ(3) (0) = −96ε φ(3) (∞) = 0

It’s easy to see that the higher order derivatves are at least of the order of ε2 . Hence
Z ∞
X 1
(2j + 1) e−εj(j+1) = dmφ(m) − [φ(∞) − φ(0)]
j=even 0 2
B1 h (1) i B h
2
i
+ φ (∞) − φ(1) (0) − φ(3) (∞) − φ(3) (0) + . . .
2! 4!
Z ∞
1 d −2εm(2m+1) 1 1
= − dm e + − (4 − 2ε)
2ε 0 dm 2 12
1
− 96ε + O(ε2 )
24 × 30
1 1 1
= + + ε + O(ε2 )
2ε 6 30
Similarly, it can be shown that
X 1 1 1
(2j + 1) e−εj(j+1) = + + ε + O(ε2 )
2ε 6 30
j=odd

i.e. both the sums agree upto linear order in ε. Hence

∞
X 1 1 1
(2j + 1) e−εj(j+1) = + + ε + O(ε2 )
j=0
ε 3 15

and so
∞  
X 1 1 1
qrot = (2sA + 1) (2sB + 1) (2j + 1) e−εj(j+1) = ω + + ε + O(ε2 )
j=0
ε 3 15

where ω = (2sA + 1) (2sB + 1).

For homonuclear molecules A 6= B, the total number of symmetric spin states is 2s + 1 +
1
(2s + 1)2 − (2s + 1) = (2s + 1) (s + 1). These are associated with the j = even (odd)
 
2
states if s is an integer (half odd integer). On the other hand, the number of antisymmetric
states is s(2s + 1) which are associated with the j = odd (even) states if s is an integer (half
odd integer). Since the sums for odd and even j match upto order ε, in either case we have
∞  
X 1 1 1
qrot = [(2s + 1)(s + 1) + s(2s + 1)] (2j + 1) e−εj(j+1) = ω + + ε + O(ε2 )
j=even
ε 3 15

where ω = 21 (2s + 1)2 .

The average rotational energy per molecule is
∂
hEi = kB T 2 ln qrot
∂T
72 4. CLASSICAL STATISTICS

and since
 
1 1 1 2
qrot = ω + + ε + O(ε )
ε 3 15
 
ω 1 1 2
= 1+ ε+ ε + O(ε3 )
ε 3 15
we have
 
1 1 2
ln qrot = ln ω − ln ε + ln 1 + ε + ε + ...
3 15
1 1 2 1 2
= ln ω − ln ε + ε+ ε − ε + O(ε3 )
3 15 18
1 1 2
= ln ω − ln ε + ε + ε + O(ε3 )
3 90
Hence  
∂ 1 1 1 dε
ln qrot = − + + ε + O(ε2 )
∂T ε 3 45 dT
2
dε
and since dT
= − 2Ik~ 2 = − Tε we get
BT
 
1 1 2 3
hEi = kB T 1 − ε − ε + O(ε )
3 45
 CHAPTER 5

Imperfect gases

5.1. Cluster Expansions

Although the method we would develop here can in principle be

applied to a more general case, let us restrict ourselves to the situation
where the molecules of a system interact via a two body potential.
Thus the potential energy f interaction between the molecules is given
by X
UN (~r1 , . . . , ~rN ) = uij (rij )
i>j
where uij is this two body potential. The net potential will be
N
X
U (~r1 , . . . , ~rN ) = UN + ui (~ri )
i=1

where ui is the internal potential energy of the i th molecule.

In what follows we will assume that the intermolecular potential
u → 0 as r → ∞ faster than r−3 1. Usually the long range part of
the intercation will be the Van der Waals force for which u ∼ r−6 . We
would consider sufficiently high temperatures and low densities that we

1This is not a serious restriction. In most of the systems we will be interested in,
i.e. collections of atoms and molecules, the basic interaction is always the Couilomb
potential for which u ∼ r−1 . So we maight naively expect that our current analysis
may not be applicable to this case. However, charged particles always tends to form
neutral groups and outsuide these groups the potential would always be screened
and fall off much faster than r−3 (usually the potential is due to mutual polarization
of neutral groups. These induced multipole potentials tend to fall off at least as fast
as r−6 ). So the current analysis is indeed applicable to these systems. However, in a
system interacting gravitationally there is no such shielding and this analysis cannot
be applied to these systems. In fact for such a system the thermoidynamic limit
does not exist (i.e. kBpT = V1 ln Q is not independent of V in the limit V → ∞).
73
74 5. IMPERFECT GASES

may use the classical grand partition function :

∞ Z N 
eβµN N Y d3 pi d3 ri −βUN −β PNi=1 p2i
X 
Q = q 3
e e 2m

N =0
N ! i=1
h
∞ 3N
X eβµN 2πm  2

= 2
q N QV (N )
N =0
N ! βh
∞
X eβµN q N
= QV (N )
N =0
N !a3N
∞
X yN
= QV (N )
N =0
N!
βµ
where y = e a3 q . Here q is the internal partition function and a =
√ h is the thermal wavelength and
2πmkB T

N
Z Y
QV (N ) = d3 ri e−βUN
i=1

In the curreent section we discuss Mayer’s cluster expansion method

for calculating QV (N ).
Since the potential dies down at large distances, itbis convenient to
introduce
fij = e−βuij − 1
Then fij → 0 as r → ∞ and
Z N
!
Y Y
QV (N ) = d3 ri (1 + fij )
i=1 (i,j)
Q Q
here (i,j) ≡ N ≥i>j≥1 . Since the pair potential uij (~ri , ~rj ) is a function
of rij = |~ri − ~rj | only, fij is clearly symmetric. Of course, since the
terms involved have i > j, only one of fij and fji appear in the integral
Q
above. Let us examine the product (i,j) (1 + fij ) . For example, if
N = 4, then
Y
(1 + fij ) = 1 + (f12 + f13 + f14 + f23 + f24 + f34 )
(i,j)

+ (f12 f13 + f13 f14 + . . .) + . . . + f12 f13 . . . f34

We will use a graphical representation for each term in the expansion.

 5.1. CLUSTER EXPANSIONS 75

7654
0123 7654
0123 7654
0123 7654
0123 7654
0123 7654
0123
1 2 1 2 1 2
7654
0123 7654
0123 7654
0123 7654
0123 7654
0123 7654
0123
3 4 3 4 3 4

=1 = f12 = f13

7654
0123 7654
0123 7654
0123 7654
0123 7654
0123 7654
0123
1 2 1 2 1 ?? 2
??
7654
0123 7654
0123 7654
0123 7654
0123 7654
0123 7654
0123
?
3 4 3 4 3 4

= f34 = f12 f13 = f13 f14

7654
0123 7654
0123 7654
0123 7654
0123
1 2 1 ??  2
 ??
7654 7654 7654 7654

0123 0123 0123 0123

  ?
3 4 3 4

= f12 f13 f24 f23 = f12 f13 f14 f24 f23 f34

In general we define a diagram as the collection of N numbered

points (the vertices), where some pairs of points may be connected by
a line. For each of the line joining the point i with point j we associate
a factor fij . It is obvious that the diagrammatic association with terms
in the integrand is unique.
Next we introduce the concept of a cluster. This is defined to be
a diagram in which every point is connected by at least one path to
every other point. For example

7654
0123 7654
0123 7654
0123 7654
0123
1 2 1 2

7654 7654 7654 7654

0123 0123 0123 0123
is not a cluster but  this one is

3 4 3 4
Since the two body potentials fall off at infinite seperation, a cluster
contributes only when all the points are close together and then this
integral ∝ V where V is the volume of the system. Similarly integral
76 5. IMPERFECT GASES

7654
0123 7654
0123
 
1 2
7654
0123 7654
0123
of a product of two clusters e.g.  ∝ V 2 , the integral of

7654
0123 7654
0123
3 4
 
1 2
7654
0123 7654
0123
the product of four clusters e.g.  ∝ V 4 , and so on.
3 4
7654 7654
0123 0123
A few examples may make this point clear. Let us consider the
1 2
7654
0123 7654
0123
integral of cluster = f12 f13 f34 .
3 4
Z Z
3 3 3 3
d r1 d r2 d r3 d r4 f12 f13 f34 = d3 r1 d3 r12 d3 r13 d3 r34 f12 (r12 ) f13 (r13 ) f34 (r34 )
Z
∝ d3 r1 = V

where in the first step we have substituted the relative vectors ~r12 =
~r2 −~r1 , ~r13 = ~r3 −~r1 and ~r34 = ~r4 −~r3 in place of the original vectors ~r2 ,~r3
and ~r4 (it is easy to verify theat the Jacobian of this transformation
is 1). Due to the short range nature of the functions fij , integrals like
R 3
d r12 f12 (r12 ) are not proprtional to the volume. In particular they

7654
0123 7654
0123
stay finite in the thermodynamic limit V → ∞2.
1 2

7654 7654

0123 0123
As another example consider the cluster  = f12 f13 f23 f24 .

3 4
It’s integral is
Z Z
3 3 3 3
d r1 d r2 d r3 d r4 f12 f13 f23 f24 = d3 r1 d3 r12 d3 r13 d3 r24

×f12 (r12 ) f13 (r13 ) f23 (r23 ) f24 (r24 )

Z
= d3 r1 d3 r12 d3 r13 d3 r24 f12 (r12 ) f13 (r13 )

×f23 (|~r13 − ~r12 |) f24 (r24 )

Z
∝ d3 r1 = V

2It is to achieve precisely this that the function uij has to fall off faster than r−3 .
 5.1. CLUSTER EXPANSIONS 77

7654
0123 7654
0123
1 2
7654
0123 7654
0123
Again consider a product of two clusters, = f12 f34 .
3 4
Z Z
3 3 3 3
d r1 d r2 d r3 d r4 f12 f34 = d3 r1 d3 r12 d3 r3 d3 r34 f12 (r12 ) f34 (r34 )
Z Z
∝ d r1 d3 r3 = V 2
3

Similarly, the integral over a diagram with l clusters is proportional

to V l3. So a cluster decomposition of each term clearly shows us the
power of V this integral is proportional to.
Let us now define the cluster integrals :

b1 = 1
2
Z Y
1 1
b2 = d3 ri f12
2! V i=1
Z Y3
1 1
b3 = d3 ri [f12 f13 + f12 f23 + f13 f23 + f12 f13 f23 ]
3! V i=1
..
.  
Z Yl
11 X Y
bl = d3 ri  fij 
l! V i=1 All l clusters (i,j)

Note for example that the integrand in b3 is the sum of all possible
three clusters
7654
0123 7654
0123 + 7654
0123 7654
0123 + 7654
0123 7654
0123 + 7654
0123 7654
0123
1 2 1 2 1 2 1 2
  
7654 7654 7654 7654
  
0123 0123 0123 0123
  
  
3 3 3 3
The cluster integrals, bl are functions of V and T in general. How-
ever, provided the two body potential falls of faster than r−3 at large
separations, the limit limV →∞ bl (V, T ) exists.

5.1.1. Virial expansion and cluster integrals. In terms of the

cluster integrals, we can write down the equation of state of the imper-
fect gas. This is the basic content of Mayer’s first theorem :
Theorem. Mayer’s theorem # 1
3In field theory, the analogue of the V factor associated with each cluster is the δ
function associated with each connected part of a Feynman diagram.
78 5. IMPERFECT GASES

∞
p X
= bl y l
kB T l=1
∞
X
and ρ = lbl y l
l=1
4

Proof : Consider the decomposition of N point ssuch that there

are nl groups of l each so that N = ∞
P
l=1 nl l with the proviso nl ≡ 0
for l > N . The number of such different decompositions is5
N!
Q nl
l [(l!) nl !]

We now identify each grouping with a sum of diagrams in which one

sums over all possible clusters in each group. For example the grouping
{1, 2, 3} {4} corresponds to

7654
0123 7654
0123 + 7654
0123 7654
0123 + 7654
0123 7654
0123 + 7654
0123 7654
0123
1 2 1 2 1 2 1 2
  
7654 7654 7654
0123 7654 7654
0123 7654 7654
0123 7654
  
0123 0123 0123 0123 0123
  
  
3 4 3 4 3 4 3 4
Now, integrations over different groups are independent, since they
correspond to different clusters. For example, the grouping {1, 2, 3} {4}
in our last example integrates to 3!V b3 × 1!V b1 = 6V 2 b3 . It is easy to
see that each decomposition contributes a factor of
Y
(l!bl V )nl to QV (N )
l

and thus
X N! Y
QV (N ) = Q nl n !]
(l!bl V )nl
l [(l!) l
{nl } l

4In general, different terms in the expansion of QN and hence Q will contain differ-
ent powers of V , depending on the number of associated clusters. We will show that
the series is going to exponentiate in a remarkable way, and in the limit V → ∞ the
exponent is proportional to V (provided uij falls off faster than r−3 at large sepa-
rations). Hence kBpT will be independent of V as V → ∞, and the thermodynamic
limit will exist.
5There are in general N ! ways of permuting the N points. Out of these, the l!
ways of permuting the elements within each of the nl groups do not lead to a new
grouping. Again, the permutations of the nl groups among themselves, in nl ! ways,
do not count as a new decomposition.
 5.1. CLUSTER EXPANSIONS 79
P
with the sum running over all {nl } satisfying l nl l = N . When we
calculate the grand partition function Q, the sum runs over all values
of N from 0 to ∞. Hence, we can relax this restriction over the possible
values of nl and sum over all nl from 0 to ∞.
∞
X yN
Q = QN (V )
N =0
N!
∞ Y ∞
n
X V bl y l l
=
nl =0 l=1
nl !
∞ X ∞
n
Y V bl y l l
=
l=1 n =0
nl !
l

∞ ∞
!
V bl y l
Y X
l
= e = exp V bl y
l=1 l=1

and hence ∞
p 1 X
= ln Q = bl y l
kB T V l=1
and   ∞
∂ p X
ρ= = lbl y l
∂ ln y kB T V,T l=1
Q.E.D.
p
For very large volumes V → ∞ , both and ρ become volume
kB T
independent intensive quantities, since limV →∞ bl (V, T ) exists.

Example 5.1.1. Consider the perfect gas for which uij = 0 and
hence fij = 0. We have b1 = 1 and bl = 0 for all l > 1. Hence
p
=y=ρ
kB T
which is the equation of state of the ideal gas.

Example 5.1.2. Imperfect gas at low density (we will keep up to

O(ρ2 ) terms). Now
ρ = y + 2b2 y 2 + . . .
implies that
y = ρ − 2b2 ρ2 + . . .
80 5. IMPERFECT GASES

and hence
p
= y + b2 y 2 + . . .
kB T
= ρ − 2b2 ρ2 + b2 ρ2 + . . .

= ρ − b 2 ρ2 + . . .

which is the virial expansion upto the second virial coefficient. Hence
we see that the second virial coefficient is
Z
1
b2 = d3 r1 d3 r2 f12 (r12 )
2V
Z
1
d3 r1 d3 r2 e−βU (|~r1 −~r2 | − 1


=
2V
Z
1
d3 r e−βU (r) − 1


=
2

In the previous example we saw that the second virial coefficient

of an imperfect gas involves only the two body cluster integral. Simi-
larly the lth virial coefficient would involve upto only l particle cluster
integrals. So even though the system contains an infinite number of
particles, we need the dynamics of only upto l particle systems to cal-
culate the lth virial coefficient.

5.1.2. The hard sphere gas. Let us now consider a particular

two body potential
(
∞ for rij < d
uij (rij ) =
finite but small otherwise
So, we are considering a gas of hard spheres of diameter d. Then
(
−1 for rij < d
fij (rij ) = u
∼ − kBijT for rij > d
So the second virial coefficient for a gas of molecules interacting via
this potential is
Z Z
1 3 2π 3 1
b2 = d r f (r) = − d − d3 r u(r)
2 3 2kB T r>d
which means that upto O(ρ2 ) we have
 Z 
p 2π 3 1
=ρ+ d + d r u(r) ρ2
3
kB T 3 2kB T r>d
 5.1. CLUSTER EXPANSIONS 81

Let us compare this with the well known Van der Waals equation
of state for one mole of gas
 a
p + 2 (v − b) = RT
v
NA
which can be rewritten in terms of the particle density ρ = v
(where
NA is the Avogadro number) in the form

p + a0 ρ2 (1 − b0 ρ) = ρkB T

where a0 = a
NA2 and b0 = b
NA
. Thus

p ρ a0 2
= − ρ + ...
kB T 1 − b0 ρ kB T
a0
 
0
= ρ+ b − ρ2 + O(ρ2 )
kB T
Comparing this with the virial expansion we have just obtained we get
Z
0 1
a = − d3 r u(r)
2 r>d
1 4π 3
b0 = d
2 3
Or, in terms of the original Van der Waals coefficients
Z
1 2
a = − NA d3 r u(r)
2 r>d
1 4π
b = NA d3
2 3
For attractive forces, u(r) < 0 and hence a > 0, while for repulsive
forces u(r) > 0 and a < 0. The positive constant b can be seen to be
related to the volume occupied by the hard spheres, thus confiming our
intuitive notions about it’s origin.
In the above we have calculated upto the 2nd virial coefficient. If
we retain upto the 3rd virial coefficient, we will see deviations from the
Van der Waals equation. So the Van der Waals equation is correct only
at low density and weal long-range intermolecular potentials.

5.1.3. 3rd virial coefficient. We have

ρ = y + 2b2 y 2 + 3b3 y 3 + . . .
82 5. IMPERFECT GASES

which leads to

y = ρ − 2b2 ρ2 + 8b22 − 3b3 ρ3 + . . .

correct to O(ρ3 ). Thus we have

p
= y + b2 y 2 + b3 y 3 + . . .
kB T
= ρ − b2 ρ2 + 4b22 − 2b3 ρ3 + . . .

And so the third virial coefficient is 4b22 − 2b3 . Let us observe what this
means in terms of the cluster integrals :

d3 r1 d3 r2 7654
0123 7654
0123
Z
1 h i
b2 = 1 2
7654 7654
2!V
0123 0123
 
2 2
 7654 7654
0123 0123
Z 
 

1 
b3 = d3 r1 d3 r2 d3 r3 
3 1 ??? + 1
7654 7654
??? 
0123 0123

3!V
3 3
Now,

7654
0123
 
2
 7654
0123
Z 
 
Z
3 3 3
d r1 d r 2 d r3 
 1 ???
 = d3 r1 d3 r2 d3 r3 f12 (r12 ) f13 (r13 )
7654
0123

3
Z
= d3 r1 d3 r12 d3 r13 f12 (r12 ) f13 (r13 )
Z 2
3
= V d r12 f12 (r12 ) = V (2b2 )2

So,

7654
0123
 
2
 7654
0123
Z 

2 1 
b3 = 2(b2 ) + d3 r1 d3 r2 d3 r3 
 1 ???
7654
0123

3!V 
3

7654
and hence we find that the third virial coefficient is
0123
 
2
7654
0123
Z 

2  
4(b2 )2 − 2b3 = − d3 r1 d3 r2 d3 r3 
 1 ???
7654
0123

3!V 
3
 5.1. CLUSTER EXPANSIONS 83

and using the notation for irreducible cluster integrals (a concept we

will introduce in a moment) this can be rewritten as − 23 β2
So the 3rd virial coefficient is given by 3 particle irreducible cluster.
This, as we shall see, is a general property of virial coefficients.
Definition. A reducible cluster is a cluster that can be sepa-
rated into two disconnected pieces by removing any one of it’s vertices.

7654
0123
A cluster is called irreducible if it is not reducible6.
2
For example, the cluster 7654
0123

1? is reducible since it becomes discon-
7654
0123
??

7654
0123
3
2
nected if we remove the vertex labelled 1. On the other hand, 7654
0123


1 ??
7654
0123
?
3
is irreducible.
We will see that the virial coefficients involve only irreducible clus-
ters. This result is Mayer’s second theorem. We define
Z k+1 !
1 Y X Y
βk = d3 ri fij
k!V i=1 (i,j)
irreducible
cluster of
k + 1 points
Note that the sum is over irreducible clusters of k + 1 particles. Then
we have
Theorem. Mayer’s second theorem
" ∞
#
p X k
=ρ 1− βk ρk
kB T k=1
k + 1
Mayer’s second theorem gives a direct virial expansion in terms of
the βk ’s. The original proof is due to Mayer (Mayer & Mayer ). There
is an elegant derivation due to Ford and Uhlenbeck (Studies in Stat.
Mech 1, 123 (1962)). We will present a proof due to R. Friedberg (J.
Math. Phys. 16 20 (1975)) in the appendix. This proof has a general
framework that can be used in field theory as well.
6These definitions have analogues in field theory. There, of course reducibility
or irreducibility is defined in terms of cutting through lines rather than through
vertices.
84 5. IMPERFECT GASES

Not only the virial equation of state, but all thermodyanamic prop-
erties are available in terms of the irreducible cluster integrals βk . Con-
sider a monoatomic gas for which the electronic excitations can be
ignored. Then the internal partition function q = ω where ω is the
degeneracy of the ground state. Then
 
∂
E = − ln Q
∂β V,βµ
  
2 ∂ p
= kB T V
∂T kB T z
From Mayer’s second theorem
" ∞
#
p X k k
=ρ 1− βk ρ
kB T k=1
k+1
we get
     " ∞
# ∞  
∂ p ∂ρ X
k
X k ∂βk
= 1− kβk ρ − ρk+1
∂T kB T z ∂T z k=1 k=1
k + 1 ∂T z
Again
    " ∞
#
∂ p ∂ρ X
ρ= = 1− kβk ρk
∂ ln y kB T V,T ∂ ln y V,T k=1

which integrates to
∞
X
ln y = ln ρ − βk ρk + f (V, T )
k=1

For a perfect gas, βk = 0 and y = ρ . So f (V, T ) = 0 and hence

ln z = ln y + ln a3 − ln q
∞
3 X
= − ln T + ln ρ − βk ρk + constant
2 k=1

which can be differentiated with respect to T at fixed z to yield

" ∞
#  ∞  
3 1 X
k ∂ρ X ∂βk
0=− + 1− kβk ρ − ρk
2T ρ k=1
∂T z k=1
∂T z

and so
  " ∞
# " ∞   #
∂ρ X 3 X ∂βk
1− kβk ρk = ρ + ρk
∂T z k=1
2T k=1
∂T z
 5.2. A SOLUBLE MODEL : 1 DIMENSIONAL GAS OF HARD RODS 85

and thus
   " ∞   # X ∞  
∂ p 3 X ∂βk k k ∂βk
z = ρ + ρ − ρk+1
∂T kB T 2T k=1
∂T z k=1
k + 1 ∂T z
" ∞   #
3 X 1 ∂βk
= ρ + ρk
2T k=1
k + 1 ∂T z

and so
" ∞   #
3 X 1 ∂β k
E = kB T 2 V ρ + ρk
2T k=1
k + 1 ∂T z
" ∞   #
3 X 1 ∂βk
= N kB T +T ρk
2 k=1
k + 1 ∂T z

Again
∞
!
a3 a3
  X
k
G = N µ = N kB T ln z = N kB T ln y + ln = N kB T ln ρ − βk ρ + ln
ω k=1
ω
Since G = E + pV − T S we have
" ∞
#
ω X
k E + pV
S = N kB ln 3 + βk ρ +
a ρ k=1 T
" ∞
# " ∞   #
ω X 3 X 1 ∂β k
= N kB ln 3 + βk ρk + N kB +T ρk
a ρ k=1 2 k=1
k + 1 ∂T z
" ∞
#
X k
+N kB 1 − βk ρk
k=1
k + 1
" ∞    #
ω 5 X 1 ∂βk
= N kB ln 3 + + βk + T ρk
a ρ 2 k=1 k + 1 ∂T z
"  3 ∞   #
mkB T 2 ω 5 X 1 ∂
= N kB ln + + (T βk ) ρk
2π~2 ρ 2 k=1 k + 1 ∂T z

5.2. A soluble model : 1 dimensional gas of hard rods

Consider a set of N hard rods of length d enclosed in a one dimen-

sional box of length L. the two body potential is
(
∞ for |xi − xj | < d
uij =
0 for |xi − xj | > d
86 5. IMPERFECT GASES
q
Let us assume that the thermal wavelength a = mk2πB T is sufficiently
small so that the classical approximation is O.K (we are using ~ = 1
here).
The canonical prtition function of this system is
N N
" N #!
dxi ∞ Y X k2
Z Y Z
1 i
QN = dki exp −β +U
N ! i=1 2π −∞ i=1 i=1
2m
Z ∞ N Z Y N
1 dk −βk2 /2m
= e dxi e−βU
N! −∞ 2π i=1
Z Y N
1
= N
dxi e−βU
N !a i=1
P
where U = (i,j) uij . The configuration integral looks quite com-
plicated. However, note that every configuration of our system has a
one to one correspondence with each configuration of a system of N
non interacting point particles enclosed in a length L − N d. So, the
configuration integral can be done trivially
Z Y N
dxi e−βU = (L − N d)N
i=1

Thus
1
QN = (L − N d)N
aN N !
and so the free energy is given by

F = −kB T ln QN = −kB T [ln(L − N d) − N ln a − ln N !]

and the pressure is

∂F N N/L ρ
p=− = kB T = kB T = kB T
∂L L − Nd 1 − N d/L 1 − ρd
where ρ = NL is the density of the rods7. This is an exact result. It can
be written as a virial expansion
p
= ρ 1 + ρd + ρ2 d2 + . . .
 
kB T

7Of course this can also be obtained from the grand canonical partition function.
 5.2. A SOLUBLE MODEL : 1 DIMENSIONAL GAS OF HARD RODS 87

We will now show how one can obtain the same result using Mayer’s
theorem. First note that in this case
(
−1 for |xi − xj | < d
fij = e−βuij − 1 =
0 for |xi − xj | > d
So

dx1 dx2 7654

0123 7654
0123
Z Z Z d
1 1
β1 = 1 2 = dx1 dx2 f12 = − dx = −2d
L L −d

Again,

7654
0123
2
7654
0123
Z 

1
β2 = dx1 dx2 dx3 1
7654
???
0123
2!L
3
Z
1
= dx12 dx13 f (x12 ) f (x13 ) f (x23 )
2
where x12 = x2 − x1 , x13 = x3 − x1 and x23 = x3 − x2 = x13 − x12 . Here
the integrand is nonvanishing and equal to -1 for

−d < x12 < d, −d < x13 < d, and − d < x13 − x12 < d

Thus for d > x12 ≥ 0, the integrand is -1 for −d + x12 < x13 < d and
for −d < x12 ≤ 0 for −d < x13 < d + x12 . So, we have
Z d Z d Z 0 Z d+x12 
1
β2 = − dx12 dx13 + dx12 dx13
2 0 −d+x12 −d −d
Z d Z 0 
1
= − dx12 (2d − x12 ) + dx12 (2d + x12 )
2 0 −d
3
= − d2
2
So upto O(ρ3 ) we have from Mayer’s theorem
 
p 1 2 2
= ρ 1 − β1 ρ − β2 ρ − . . .
kB T 2 3
= ρ 1 + dρ + d2 ρ2 + . . .
 

which is the same as our exact solution upto O(ρ3 ).

88 5. IMPERFECT GASES

5.3. Hard sphere gas in three dimensions

For the hard sphere “ideal” gas in three dimensions, we have

(
−1 for |~ri − ~rj | < d
fij =
0 for |~ri − ~rj | > d
Then Z Z Z
1
β1 = d r1 d r2 f12 = d3 r12 f12
3 3
V
In order to get a nonzero contribution we must have |~r12 | < d, i.e. ~r2
must be within a sphere of radius d centered around ~r1 . Thus
4π 3
β1 = − d
3
The next irreducible cluster integral is
Z Z Z
1
β2 = d r1 d r2 d3 r3 f12 f13 f23
3 3
V
To find this integral let us figure out the region where the integrand
gives a nonzero contribution. Let us first fix ~r1 . Then ~r2 must lie
within a sphere of radius d centered around ~r1 . Suppose we have now
fixed ~r2 and |~r2 − ~r1 | = r ≤ d. Then to get a nonzero contribution to
our integral, ~r3 must be within a distance d from both ~r1 and ~r2 , i.e.
it must lie inside the volume V (r), where the two spheres of radius d
centered around ~r1 and ~r2 overlap.
The volume V 2(r) is that of a spherical cap of radius d at a height of
r
2
from the center. Elementary calculus gives us
Z d
V (r)
πdz d2 − z 2


=
2 r/2

r3
   
2 r 1 3
= π d d− − d −
2 3 8
 
2 1 1
= π d3 − rd2 + r3
3 2 24
and hence  
4 1
V (r) = π d3 − rd2 + r3
3 12
Hence
d
5π 2 6
Z
1
β2 = − 4πr2 V (r) dr = − d
2 0 12
 5.3. HARD SPHERE GAS IN THREE DIMENSIONS 89

So, the virial expansion for the 3D hard sphere model is

 
p 1 2 2
= ρ 1 − β1 ρ − β2 ρ − . . .
kB T 2 3
 
2 3

5 2 3 2


= ρ 1 + π ρd + π ρd + . . .
3 18
Let us contrast this with the corresponding Van der Waals equation of
state
p ρ
=
kB T 1 − ρb
2 3
with b = 3 πd which gives
 
p 2 3

4 2 3 2


= ρ 1 + π ρd + π ρd + . . .
kB T 3 9
So we see that the correct third virial coefficient differs from that of
the Van der Waals equation by a factor of 58 !
 APPENDIX A

Friedberg’s proof of Mayer’s second theorem

A critical step inR. Friedberg’s proof of Mayer’s second theorem

requires the association of a dual tree with each cluster.
A dual tree is a connected graph consisting of ns squares and nc
circles joined together by lines that must satisfy

(1) Each line joins a circle to a square (never a square to another

square or a circle to another circle)
(2) the tree must be separated into two disjoint parts if any one
of the lines are cut.

also, all the circles are numbered 1, 2, . . . , nc while the squares are num-
bered 1, 2, . . . , ns . For example

3
7654
0123 7654
0123 7654
0123 7654
0123
1 1 1 2 2 1 1 2 4
are all trees while
3?
 ??
7654 7654
 ??
0123 0123

1 1 2 2 4
is not (it is a loop which can be changed into a tree by cutting a line).
Two trees will be considered identical if they are topologically identical
including the numbers associated with the circles and the squares.
We can represent a cluster by a dual tree. For example

7654
0123
7654
0123 1
2
7654
0123


1 −→ 1?
7654
???
0123
 ??
7654 7654
 ??
0123 0123
3 
2 3

7654
0123 7654
0123 −→ 7654
0123 7654
0123
1 2 1 1 2
91
92 A. FRIEDBERG’S PROOF OF MAYER’S SECOND THEOREM

All irreducible l clusters are represented by the same dual tree,

namely a single square joined to l circles.

7654
0123 7654
0123
1 2
7654
0123 7654
0123
3 4

−→

7654
0123 7654
0123
1 ?? 2
??
7654 7654
0123 0123
?
3 4