Statistical Physics

Dr. A. J. Macfarlane1

Lent 1998

1 A
LT
EXed by Paul Metcalfe – comments to soc-archim-notes@lists.cam.ac.uk.
Revision: 1.10
Date: 1998-06-03 20:13:39+01

The following people have maintained these notes.

– date Paul Metcalfe

Contents

Introduction v

1 Quantum Statistical Mechanics 1

1.1 Introduction to Quantum Statistical Mechanics . . . . . . . . . . . . 1
1.2 Canonical ensemble . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.3 Temperature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.4 Towards thermodynamic variables . . . . . . . . . . . . . . . . . . . 5
1.5 Towards applications . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.5.1 N particle partition function . . . . . . . . . . . . . . . . . . 7
1.5.2 Extensive and intensive variables . . . . . . . . . . . . . . . 8
1.5.3 Density of states . . . . . . . . . . . . . . . . . . . . . . . . 8
1.5.4 Gas of spinless particles . . . . . . . . . . . . . . . . . . . . 9
1.5.5 Entropy and the Gibbs paradox . . . . . . . . . . . . . . . . . 9
1.6 Harmonic oscillator model . . . . . . . . . . . . . . . . . . . . . . . 10

2 Thermodynamics 11
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.2 Applications of dE = T dS − P dV . . . . . . . . . . . . . . . . . . 12
2.2.1 Integrability conditions . . . . . . . . . . . . . . . . . . . . . 12
2.2.2 Specific heats . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.2.3 Adiabatic changes . . . . . . . . . . . . . . . . . . . . . . . 14
2.2.4 Entropy of n moles of ideal gas . . . . . . . . . . . . . . . . 14
2.2.5 van der Waal’s equation . . . . . . . . . . . . . . . . . . . . 15
2.2.6 The Joule effect . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.3 Some thermodynamics . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.3.1 The second law . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.4 Heat flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

3 Grand ensemble methods 19

3.1 The formalism . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.2 Systems of non-interacting identical particles . . . . . . . . . . . . . 21
3.2.1 A little quantum mechanics . . . . . . . . . . . . . . . . . . 21
3.2.2 The partition functions . . . . . . . . . . . . . . . . . . . . . 22
3.2.3 Classical limit . . . . . . . . . . . . . . . . . . . . . . . . . 24
3.3 Black body radiation . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.4 Degenerate Fermi gas . . . . . . . . . . . . . . . . . . . . . . . . . . 26
3.4.1 Heat capacity at low temperature . . . . . . . . . . . . . . . . 27
3.5 Bose-Einstein condensation . . . . . . . . . . . . . . . . . . . . . . . 28

iii
iv CONTENTS

3.6 White dwarf stars . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

4 Classical statistical mechanics 31

4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
4.2 Diatomic gases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
4.3 Paramagnetism . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
4.4 Specific heats . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
4.5 Weak interparticle forces . . . . . . . . . . . . . . . . . . . . . . . . 34
4.6 The Maxwell distribution . . . . . . . . . . . . . . . . . . . . . . . . 35

Approximations 37
Introduction

These notes are based on the course “Statistical Physics” given by Dr. A. J. Macfarlane
in Cambridge in the Lent Term 1998. These typeset notes are totally unconnected
with Dr. Macfarlane. The recommended books for this course are discussed in the
bibliography.
Other sets of notes are available for different courses. At the time of typing these
courses were:
Probability Discrete Mathematics
Analysis Further Analysis
Methods Quantum Mechanics
Fluid Dynamics 1 Quadratic Mathematics
Geometry Dynamics of D.E.’s
Foundations of QM Electrodynamics
Methods of Math. Phys Fluid Dynamics 2
Waves (etc.) Statistical Physics
General Relativity Dynamical Systems

They may be downloaded from

http://home.arachsys.com/˜pdm/ or
http://www.cam.ac.uk/CambUniv/Societies/archim/notes.htm
or you can email soc-archim-notes@lists.cam.ac.uk to get a copy of the
sets you require.

v
 Copyright (c) The Archimedeans, Cambridge University.
All rights reserved.
Redistribution and use of these notes in electronic or printed form, with or without
modification, are permitted provided that the following conditions are met:

1. Redistributions of the electronic files must retain the above copyright notice, this
list of conditions and the following disclaimer.
2. Redistributions in printed form must reproduce the above copyright notice, this
list of conditions and the following disclaimer.
3. All materials derived from these notes must display the following acknowledge-
ment:

This product includes notes developed by The Archimedeans, Cambridge

University and their contributors.

4. Neither the name of The Archimedeans nor the names of their contributors may
be used to endorse or promote products derived from these notes.
5. Neither these notes nor any derived products may be sold on a for-profit basis,
although a fee may be required for the physical act of copying.
6. You must cause any edited versions to carry prominent notices stating that you
edited them and the date of any change.

THESE NOTES ARE PROVIDED BY THE ARCHIMEDEANS AND CONTRIB-

UTORS “AS IS” AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING,
BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABIL-
ITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO
EVENT SHALL THE ARCHIMEDEANS OR CONTRIBUTORS BE LIABLE FOR
ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSE-
QUENTIAL DAMAGES HOWEVER CAUSED AND ON ANY THEORY OF LI-
ABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUD-
ING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
OF THESE NOTES, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAM-
AGE.
Chapter 1

Quantum Statistical Mechanics

1.1 Introduction to Quantum Statistical Mechanics

Statistical mechanics deals with macroscopic systems of many particles. Consider an
isolated system S of gas in a vessel whose walls neither let heat in or out and is subject
to no mechanical action. No matter how it is prepared, it is an experimental fact that
S reaches a steady state (a state of thermodynamic equilibrium) in which a set Σ of
(rather few) thermodynamic variables are constant. Σ includes the pressure P , volume
V , temperature T , total energy E and entropy S.
Any state of S (before or after thermodynamic equilibrium is reached) contains
particles with positions and momenta changing in time. We cannot possibly analyse
this in detail — even if we knew the forces and initial conditions, and could solve the
resulting system there is no hope that we could organise usefully the vast amount of
data.
At best, we aim to treat the possible states of motion of S by some averaging
or statistical procedure that allows us to predict values of the variables in Σ that are
constant in states of thermodynamic equilibrium of S.
We use quantum mechanical ideas to approach the subject. Consider

• microstates of S, the stationary quantum states |ii.

• macrostates, which correspond to states of thermodynamic equilibrium.

The latter are not states in the quantum mechanical sense, but do involve the vast
numbers of microstates.
The ergodic hypothesis (which is provable for some systems) is that S passes in
time through all its possible states compatible with its state of thermodynamic equilib-
rium. This allows us to replace time averages for a single system S with averages at a
fixed time over a suitable ensemble E of systems, each identical to S in its macroscopic
properties. The most probable state of E reveals which microstates of S contribute to its
macroscopic states of thermodynamic equilibrium and yield excellent approximations
to the values of its thermodynamic variables.
Let E, N be the total energy and total number of particles of S respectively. One
meets three types of ensemble.

• microcanonical : each member of E has the same values of E and N .

1
2 CHAPTER 1. QUANTUM STATISTICAL MECHANICS

• canonical : each member of E has N particles but E is not fixed, although the
average total energy is time independent.
• grand : neither E nor N is fixed, although the averages are.

For the latter two, fluctuations about the average values are found to be very small
and all three types of ensemble give the same thermal physics. This course studies
mainly canonical ensembles. We will do systems with identical bosons/fermions via
the grand ensemble for technical simplicity.
The fact that averaging procedures give reliable estimates of values of physical
variables and apparently different procedures yield equivalent results is due to the effect
of the very large number of particles involved. For instance, consider the probability
pm of obtaining N/2 + m heads from N coin tosses.

  r
−N N 2 − 2m2
pm = 2 N ∼ e N ,
2 +m πN

using Stirling’s formula1 for N  m  0. If N = 1023 and m = 1017 then the

11
exponential term e−10 is effectively zero.

1.2 Canonical ensemble

We study a system S of N particles in a volume V with N and V fixed. The microstates
of S are the (complete orthogonal) set |ii with energy eigenvalues Ei . We assume the
spectrum is discrete but with possible degeneracy.
We associate with S a canonical ensemble E. This is a very large number A of
distinguishable replicas of S, S1 , . . . , SA . Suppose that in a possible state of E there
are ai members of E in the microstate |ii. Then

X X
ai = A and ai Ei = AE. (1.1)
i i

The average energy of the members of E is thus the fixed value E.

Given a set {ai } (a configuration of E) there are W (a) = A!ai ! ways of realising
i
it.
Proof. We can distribute a1 systems in the microstate |1i over E in aA1 ways. Then

we can distribute the a2 states in |2i in A−a 1

a2 ways (and so on). Thus

  
A A − a1 A!
W (a) = ··· = Q .
a1 a2 i ai !

We assign equal a priori probability to each way of realising each possible con-
figuration of E and thus there is a probability proportional to W (a) of realising the
configuration {ai }.
1 log n! 1
∼ n log n − n + 2
log 2πn
1.2. CANONICAL ENSEMBLE 3

We will associate the state of thermodynamic equilibrium of S (for the fixed values
N , V , E) with the configuration of E that is most probable in the presence of the
constraints (1.1).
To compute this suppose A  0 such that ai  0 (negligible probability attaches
to configurations where this fails). For large n, Stirling’s formula allows log n! ∼
n (log n − 1) and so

X X
log W ∼ A log A − A − ai (log ai − 1) = A log A − ai log ai .
i i

We seek to maximise this subject to the constraints (1.1). We use two Lagrange
multipliers α and β and solve

!
∂ X X
0= log W − α ai − β ai Ei
∂aj i i

and so log aj + 1 + α + βEj = 0. Thus

aj = e−1−α−βEj . (1.2)

ai = e−1−α Z, defining the canonical partition func-

P
We eliminate α via A = i
tion

X X
Z= e−βEi ≡ Ω(Ej )e−βEj , (1.3)
i Ej

where Ω(Ej ) is the degeneracy of the energy level Ej .

The fraction of members of E in the microstate |ii in the macrostate of thermody-
namic equilibrium is

ai 1
ρi = = e−βEi . (1.4)
A Z
This is the Boltzmann distribution, and may be thought of as the probability of
finding |ii in the state of thermodynamic equilibrium.
We define the average hXi of a physical variable X taking the value Xi in the state
|ii by

X
hXi = ρi X i . (1.5)
i

1 P
For instance hEi = A i ai Ei = E (reassuringly).
Z is very important. It leads directly from quantum mechanical data to calculation
of the thermodynamic variables for S in thermodynamic equilibrium. For instance,

∂ log Z
E=− . (1.6)
∂β
4 CHAPTER 1. QUANTUM STATISTICAL MECHANICS

Here, holding V fixed corresponds to keeping all the Ei fixed.

For A large (as in all cases of interest) the most probable state is overwhelmingly
so. It gives in effect the average over all possible states of the ensemble. The idea
of associating average values with actual physical predictions depends on the possible
variances being negligible.
We can calculate the variance in the energy in a similar way. Note that E = − Z1 ∂Z
∂β
and so

2
1 ∂2Z

∂E 1 ∂Z 2
=− + 2 = −hE 2 i + hEi2 = − (∆E) .
∂β Z ∂β 2 Z ∂β

For typical large systems E ∝ N , and as E depends smoothly on β we expect

∂E
∂β ∝ N as well. Thus
√
|∆E| N 1
∝ = N−2 ,
E N
which is very small for very large N .

1.3 Temperature
Given two systems Sa and Sb with fixed volumes Va , Vb and numbers of particles Na ,
Nb respectively we place them in contact to form a composite system Sab such that
energy can pass from one to the other and a state of thermodynamic equilibrium is
reached. We make a canonical ensemble Eab for Sab by distributing A replicas of Sa
and A replicas of Sb independently across the A members of Eab so that each member
of Eab has a component of type Sab .
Suppose that Sa has microstates |ii of energy Eai , Sb has microstates |σi of energy
Ebσ and Sab has microstates |i, σi of energy Eai + Ebσ .2
Suppose the |ii occurs ai times, |σi occurs bi times and |i, σi occurs ciσ times in
Eab . Then we have

X X
ai = A bσ = A (1.7)
i σ
X X X
ai = ciσ bσ = ciσ ciσ = A (1.8)
σ i iσ

We demand that the average energy or Eab be fixed at

X X
AE = ai Eai + bσ Ebσ (1.9)
i σ
X
AE = ciσ (Eai + Ebσ ) . (1.10)
iσ

Now the configuration {ciσ } arises in

2 If the interaction of particles of S with those of S is negligible, except insofar as necessary to allow a
a b
state of thermal equilibrium to be reached.
1.4. TOWARDS THERMODYNAMIC VARIABLES 5

A!
W (c) ∝ Q (1.11)
iσ ciσ !

ways and the configuration {ai }{bσ } arises in

A! A!
W (a)W (b) ∝ Q Q (1.12)
i a i ! σ bσ !

ways. We can calculate the most probable distribution of {ciσ } by maximising

(1.11) subject to (1.8) and (1.10). We get

ciσ
ciσ = e−1−αab −β(Eai +Ebσ ) ρiσ = .
A
We can also maximise (1.12) subject to (1.7) and (1.9) to get

ai
ai = e−1−αa −βEai ρi =
A
bσ
bσ = e−1−αb −βEbσ ρσ = .
A
Defining

X X X
Zab (β) = e−β(Eai +Ebσ ) Za (β) = e−βEai Zb (β) = e−βEbσ
iσ i σ

we see that Zab (β) = Za (β)Zb (β) and ρiσ = ρi ρσ . A common value for β
characterises the state of thermal equilibrium of Sab and it is natural to assume that β
is some function of temperature T and that the average energy (the energy) of a system
is a function of N , V and T .
1
We define T by β = kT where k is a constant.
As this argument applies to any two systems, k should be a universal constant
(Boltzmann’s constant) the same for all systems when a universal definition of T is
used. We will define T for a standard system (a thermometer) and for other systems by
putting them into a state of thermal equilibrium with our standard system.
Recall the ideal gas law, P V = N kT for a sample of N molecules.3 k is the same
because equal volumes of all gases at fixed temperature and pressure are observed to
have the same number of molecules. T is in degrees Kelvin.

1.4 Towards thermodynamic variables

Recall that in quantum mechanics, spinless particles in a cube of side L have energies
~2 2 3 − 32
En = 2mL 2 |n| , where n ∈ N . Note that E ∝ V and we generalise this to
Ei = Ei (V ). Consider Sgas in thermodynamic equilibrium in a container of volume V
and change V by a small amount by slowly moving one of its walls.
3 This applies to all gases at sufficiently low density.
6 CHAPTER 1. QUANTUM STATISTICAL MECHANICS

The Ei will change according to Ei 7→ Ei + ∂E ∂V δV and changes in which the ρi

i

do not change will be examined first.

The change in energy is supplied by the work done on S by a piston applied to
the right hand wall. For slow motion the system passes through successive positions
of thermal equilibrium and then the applied force just balances the pressure force P A.
Thus Pthe work done to the system is −P Aδl = −P δV . δE = δW = −P δV . Now
E = i ρi Ei and so
X ∂Ei 1 ∂
P =− ρi = log Z. (1.13)
i
∂V β ∂V

P Consider next changes Pin which the ρi are allowed to vary. Then δE = −P δV +
i Ei δρ i , and note that i δρi = 0.
Define the entropy

k
S= log Wmax that is, W at thermal equilibrium
A !
k X
= A log A − ai log ai
A i
! (1.14)
k X X
= A log A − ai log ρi − log A ai
A i i
X
= −k ρi log ρi at thermal equilibrium.
i

We see that

 
X 1
δS = −k δρi log ρi + ρi log ρi
ρi
X
= −k δρi (−βEi − log Z + 1)
1 X
= Ei δρi
T i
P
and thus i Ei δρi = T δS. We have defined S such that

δE = T δS − P δV or in terms of exact differentials,

(1.15)
dE = T dS − P dV.

We might regard this as arising from the comparison of ensembles with infinitesi-
mally different states of thermal equilibrium. (1.15) states the First Law of Thermody-
namics4 , written in the form

dE = δQ + δW, (1.16)

where δQ is the heat supplied to S and δW is the work done on S.

4 You can’t win...
1.5. TOWARDS APPLICATIONS 7

∂E ∂E
(1.15) also shows that T = and P = − ∂V
∂S .
2
For a gas of spinless particles, E = i Ei ρi ∝ V − 3 and so P = 32 VE , giving
P
P V = 32 E. Combined with the result for a perfect gas, E = 23 N kT (see later), we
have Boyle’s law P V = N kT .
We want to be able to calculate S from the partition function. Now, at thermal
equilibrium,

X
S = −k ρi log ρi
i
X
= −k ρi (−βEi − log Z)
i
1
= E + k log Z.
T
Define the free energy F = E − T S, then

F = − β1 log Z. (1.17)
Also,

1 ∂ log Z ∂ ∂F
S=− + k log Z = (kT log Z) = − . (1.18)
T ∂β ∂T ∂T
We have a very tangible definition of S, as proportional to log Wmax , where Wmax
is the number of microstates of S contributing to the state of thermal equilibrium.
Given the partition function (1.3), we can calculate the thermodynamic variables
in a state of thermal equilibrium, P (by (1.13)), E (by (1.6)), F (by (1.17)) and S (by
(1.18)).

1.5 Towards applications

1.5.1 N particle partition function
Consider a composite system with Hamiltonian H = H1 + H2 and H1 and H2 inde-
pendent (they commute). Let H1 |ai = E1a |ai and H2 |αi = E2α |αi. Then as H1 and
H2 commute, H|a, αi = Eaα |a, αi, where Eaα = E1a + E2α . Then the 2 particle
partition function is
X X X
Z= e−βEaα = e−βE1a e−βE2α = Z1 Z2 .
a,α a α

Let Z ≡ ZN describe the sum over states of a gas of N very weakly interacting
particles. If these are supposed independent (distinguishable) then Z = z N , where
z is the “one particle partition function”. This allows easy calculations, but fails for
systems of identical (indistinguishable) bosons or fermions. For the latter it seems best
to use grand ensemble methods.
Later we will consider a gas of diatomic molecules, H = HT + Hrot + Hvib , where
HT describes the centre of mass motion, Hrot describes rotation about the centre of
mass and Hvib describes vibration along the axis. We see that the one particle partition
function is z = zT zrot zvib .
8 CHAPTER 1. QUANTUM STATISTICAL MECHANICS

1.5.2 Extensive and intensive variables

Extensive variables are proportional to the amount of matter in a system at fixed tem-
perature, whereas intensive variables are independent of the amount of matter.
In general, N , V , S and E are extensive and T and P are intensive.
Think initially of a system made up of two independent subsystems. Then V =
V1 +V2 . In a state of thermal equilibrium the systems have the same β (and hence same
T ), and so T is intensive. In a state of equilibrium there is mechanical equilibrium and
so the systems have the same pressure, and so P is intensive.
As Z = Z1 Z2 we see that E = E1 + E2 , S = S1 + S2 and so on. A logical ex-
tension of this argument gives that V , E and S are proportional to N and so extensive.
This can fail if volume energies do not swamp surface energies or if intermolecular
forces are neither weak nor short range.

1.5.3 Density of states

Consider a spinless particle in a box of side L. Instead of using solutions ψ of the
Schrödinger equation such that ψ = 0 on the walls we use periodic boundary condi-
2 2
k
tions, ψ(x + Li) = ψ(x) (and so on). Thus we can use ψk ∝ eık·x with  = ~2m .
Lk 3
The periodicity gives that 2π ∈ Z . There is one such state per unit volume in n-
space and in the continuum limit, there are d3 n states with n in the range n → n + dn.
L 3 3
There are thus d3 n states with k in the range k → k + dk and hence 2π


d k states
with k in the range k → k + dk.
We see that

Z  3
X L
→ d3 k (1.19)
i
2π
V
Z
= 4πk 2 dk
(2π)3
Z  1
V 2m 2 2md
= 4π
(2π)3 ~2 2~2
  32 Z
2m 1
= 2πV  2 d (1.20)
h
Z
= dg().

We insert a factor gS = (2S + 1) to cope with massive particles of spin S, and

  23
2m 1
g() = gS 2πV 2
h2

gives the density of states. We use (1.20) in isotropic contexts, but (1.19) is needed
in kinetic theory.
In two dimensions we consider a square of side L and A = L2 enters in the rôle of
2m


V . We find that g() = h2 gS 2πA is independent of .
1.5. TOWARDS APPLICATIONS 9

21
For relativistic particles of rest mass m we have  = m2 c4 + ~k 2 c2 and use
(1.19) to get
Z ∞
4π
Z Z
3 2

1
d k = 4πk dk = d 2 − m2 c4 2 . (1.21)
(~c)3 mc2
We also supply a factor gS = 2S + 1 for a particle of spin S with nonzero rest
mass. For photons, m = 0, gS = 2 and so

 3 Z ∞ ∞
V 4π 8πV
X Z
→2 2 d = ν 2 dν. (1.22)
i
(2π)3 ~c 0 c3 0

1.5.4 Gas of spinless particles

1
Consider N spinless particles in a cube of side length L = V 3 . Then the single particle
partition function
X Z ∞
z= e−βi → d g()e−β ,
i 0

1 2m

32
where g() = DV  2 and D = 4π h2 . Setting y = β we get

∞   32
2mπkT
Z
3 1
−y
z = DV (kT ) 2
dy y e2 =V . (1.23)
0 h2

Now z = z N and so log Z = N log z. Thus

∂
− 23 log β = 23 N kT.


E = −N
∂β

This is the classical result: there is an energy 12 kT associated with each degree of
freedom of each particle. Combining with P V = 23 E we get P V = N kT .

1.5.5 Entropy and the Gibbs paradox

Using Z = z N , (1.23) and (1.18) we have
 
2mπkT
S = N k log V + 23 N k log + 32 N k,
h2
which is not extensive, as V ∝ N . This is Gibbs’ paradox.
N
If instead we use Z = zN ! , (1.23), (1.18) and Stirling’s formula we have
 
2mπkT
S = N k log N V
+ 32 N k log + 32 N k
h2
and we see that S is extensive.
Often an additive constant in S does not matter and (1.23) gives all thermodynamics
otherwise correctly. It reflects a normal classical view of how to treat indistinguishable
N
particles. Taking Z = zN ! cures this but has no foundation in classical statistical
physics.
10 CHAPTER 1. QUANTUM STATISTICAL MECHANICS

1.6 Harmonic oscillator model

Model a crystal by placing one atom at each point of some regular lattice with N sites.
In 1D, simplify by taking a simple harmonic oscillator of frequency ωr instead of the
rth atom.
Now ZN = r z(ωr ), and using En = ~ω(n + 21 ) for the simple harmonic oscil-
Q
lator with frequency ω we get
β~ω
e− 2
z(ω) = .
1 − e−β~ω
Even if ωr = ω for all r, the “atoms” are distinguishable as there exists one at each
lattice site. This is a one dimensional version of the Einstein solid. We find that
E ∂ log z ~ω
=− = 12 ~ω + β~ω .
N ∂β e −1

If ~ω  kT this gives the classical result NE

− 12 ~ω = kT , with energy 21 kT per
degree of freedom per “atom”.
If ~ω  kT then β~ω  1 and N E
− 12 ~ω ≈ ~ωe−β~ω , which tends to zero as
T → 0.
Quantum statistical mechanics “knows” when to count the full classical 21 kT value
per degree of freedom: do it only when T is large on some scale set by the problem. In
this case the critical temperature Tc = ~ω
k .
Chapter 2

Thermodynamics

2.1 Introduction
Consider a volume V of gas with a fixed number N of particles. The state of thermal
equilibrium of this gas is characterised by T and V .
Our work with quantum statistical mechanics has produced some concepts natu-
rally:

• E: the total energy of the system, which arose as the average energy and has
negligible fluctuations.
• S: the entropy. This has a clear significance as the number of microstates con-
tributing to the macrostate of thermal equilibrium.
• equations of state: for instance P V = N kT .
• dE = T dS − P dV , which is true for any thermodynamic change between two
infinitesimally close states of thermodynamic equilibrium.

All thermodynamic variables for the sample of gas we are talking about can be
regarded as a function of two suitably chosen independent variables. In the case of the
energy, E arises as E(S, V ).
We also found that dE = δQ + δW , where δQ is the heat supplied to the system
and δW is the work done on the system.
For finite changes from the initial to the final state conservation of energy gives
∆E = ∆Q + ∆W . This is the first law of thermodynamics, equivalent to the conser-
vation of energy.
Suppose we change from a state with E1 (S1 , V2 ) to E2 (S2 , V2 ). In general, we
expect
Z V2
∆W ≥ − P dV
V1

because of wasted effort (against friction or in producing convection or turbulence).

This gives us
Z S2
∆Q ≤ T dS.
S1

11
12 CHAPTER 2. THERMODYNAMICS

These inequalities hold as equalities for reversible changes, that is changes which
are quasi-static and non-dissipative.
Quasi-static changes are changes which are done so slowly as to pass through states
arbitrarily close to a state of thermodynamic equilibrium, so that all thermodynamic
variables are well-defined throughout the change.
Some (obvious) irreversible processes are fast piston movement, free expansion of
a gas into a vacuum and the mixing of samples of different gases.
In induced/allowed changes in real life with δQ = 0, δS > 0 and so the entropy
increases.

2.2 Applications of dE = T dS − P dV
2.2.1 Integrability conditions
The internal energy E is seen as E(S, V ) within
   
∂E ∂E
dE = T dS − P dV = dS +
∂S V ∂V S
∂2 E
Thus T = ∂E ∂E



∂S V and P = − ∂V S . dE is an exact differential, so ∂S∂V =
∂2E
∂V ∂S and
   
∂P ∂T
− =
∂S V ∂V S

This is a Maxwell relation.

The free energy F = E − T S is a natural function of T and V , as
dF = dE − T dS − SdT = −SdT − P dV.
Thus S = − ∂F ∂F



∂T V and P = − ∂V T . This gives the Maxwell relation
   
∂S ∂P
= .
∂V T ∂T V
H = E + P V defines the enthalpy H(S, P ) and G = E − T S + P V defines the
Gibbs function G(T, P ).
These give four Maxwell relations, but they are interdependent.
E, F , H and G are all extensive. If (exceptionally) G = 0 (true for a gas of
photons) then G = N µ and so the chemical potential µ can vanish for arbitrary finite
N.
We need some rules for shunting partial derivatives around. Consider z = z(x, y).
Then

   
∂z ∂z
δz = δx + δy and hence
∂x y ∂y x
   
∂x ∂z
1= and
∂z y ∂x y
     
∂x ∂z ∂z
0= + .
∂y z ∂x y ∂y x
2.2. APPLICATIONS OF DE = T DS − P DV 13

This can be rewritten in a slightly different form:

   
∂x ∂z
=1
∂z y ∂x y
      (2.1)
∂x ∂y ∂x
= −1.
∂y z ∂x z ∂z y

When we wrote down E, F , H and G we wrote them in terms of the mathematically

natural independent variables in each case. In practice, we would like E as a function
of T and V . Recall T dS = dE + P dV , so that

   
∂E ∂E
T dS = dT + dV + P dV and so
∂T V ∂V T
   
∂S ∂E
T = and
∂T V ∂T V
   
∂S ∂E
T = + P.
∂V T ∂V T

∂2 S ∂2S
Also, we know that ∂T ∂V = ∂V ∂T and so
   
∂E ∂P
=T − P.
∂V T ∂T V

∂E


For a perfect gas, P V = N kT and so ∂V T
= 0. Thus E is a function of T only.
Given the laws of thermodynamics we need two inputs from experiment to specify
all thermodynamic variables completely. For instance, for a perfect gas we need the
equation of state and E = CT .

2.2.2 Specific heats

An amount of gas with NA (Avogadro’s number) of molecules is called a mole. The
ideal gas law tells us that at fixed temperature and pressure, a mole of any ideal gas
occupies the same volume. Written in terms of moles, the ideal gas law for n moles of
gas is

P V = nNA kT = nRT. (2.2)

R is called the gas constant and is the same for all ideal gases. In fact, all gases are
essentially ideal at sufficiently low density.
For a sample of 1 mole of ideal gas we define the specific heat (or heat capacity per
mole) using δQ = T dS under suitable conditions.
∂S ∂S



Define CV = T ∂T V
, the specific heat at constant volume and CP = T ∂T P
,
the specific heat at constant pressure. Now

T dS = dE + P dV = dH − V dP

and so
14 CHAPTER 2. THERMODYNAMICS

   
∂S ∂E
CV = T =
∂T V ∂T V
  (2.3)
∂H
CP = .
∂T P
For one mole of ideal gas we know that E = CT and so C = CV . To find CP we
need to use P and T as the independent variables.
Now
   
∂S ∂S
dS = dT + dV
∂T V ∂V T
        
∂S ∂S ∂V ∂V
= dT + dT + dP .
∂T V ∂V T ∂T P ∂P T
∂S ∂S ∂S ∂V





Thus ∂T P
= ∂T V
+ ∂V T ∂T P
and so
       
∂S ∂V ∂V ∂P
CP = CV + T = CV + T . (2.4)
∂V T ∂T P ∂T P ∂T V
This allows the calculation of CP − CV from the equation of state: for an ideal gas
CP − CV = R. Define the ratio of specific heats γ = C CP
V
. Note that γ = CVC+R
V
>1
for an ideal gas.
Statistical mechanics gives CV = 23 R, 52 R, . . . (for monatomic, diatomic (etc)
gases). Thus γ = 53 , 75 , . . . .

2.2.3 Adiabatic changes

These are (defined as) changes which are reversible and satisfy δQ = 0. We refer to n
moles of ideal gas using E = nCV T and P V = nRT . Now
0 = RdE + RP dV
= RnCV dT + RP dV
= CV (P dV + V dP ) + RP dV
= CP P dV + CV V dP.
Thus 0 = γ dV dP γ
V + P and so P V is constant on adiabatics.
γ
Note that adiabatics P V constant are steeper than isothermals P V constant.

2.2.4 Entropy of n moles of ideal gas

We start (as usual) from
T dS = dE + P dV
dV
= nCV dT + nRT .
V
dT dV
Thus dS = nCV + nR ,
T V
and so S = nCV log T + nR log V + c1 . Thermodynamics cannot determine the
constant c1 , and does not care that S is not explicitly extensive. S = nCV log P V γ +c0
and so (as expected), S is constant on adiabatics (isentropics).
2.2. APPLICATIONS OF DE = T DS − P DV 15

2.2.5 van der Waal’s equation

We get a better agreement with experiment by replacing the perfect gas law with

n2 A
 
P + 2 (V − nB) = nRT, (2.5)
V
where A, B and R are strictly positive constants.
• Molecules are not treated as point particles, and as P → ∞ at constant T , V →
nB, which is the residual volume of all the molecules.
2 2
• P + nV 2A reduces the real gas pressure by an amount nV 2A , due to intermolecular
attractive forces. If these are short range then the smaller V is the more important
these become.

2.2.6 The Joule effect

Consider the apparatus shown, with adiathermal walls and containing n moles of gas at
volume V1 , pressure P1 and temperature T1 . Pull back the partition and allow the gas
to expand (irreversibly) into the total volume V2 , and then to reach a state of thermal
equilibrium specified by P2 , V2 and T2 .
As δQ = 0 (adiathermal) and δW = 0 (no work is done), dE = 0 and so for a
perfect gas
T2 = T1 and S2 − S1 = nR log VV21 > 0 for V2 > V1 .
For a van der Waal’s gas we still have dE = 0 (which is true in general) and so E
stays constant. Now

∂E
 

∂T ∂V T
=− ∂E


∂V E ∂T V
− T ∂P


P ∂T V
=
nCV
2
n A 1
=− 2 for van der Waal’s gas
V nCV
< 0.
We can show that CV is independent of V for a van der Waal’s gas and we suppose
that CV is also approximately independent of T . We can now integrate this equation to
get
 
nA 1 1
T2 − T1 = − < 0.
CV V2 V1
T decreases because some of the molecular kinetic energy is lost in the expansion
against attractive intermolecular forces.
16 CHAPTER 2. THERMODYNAMICS

2.3 Some thermodynamics

Suppose a sample S of n moles of perfect gas is put in thermal contact with a heat bath
B and compressed reversibly from volume V1 to volume V2 . If the gas is perfect then
E = E(T ) = nCV T and so E stays the same. Now

Z V2
∆Q + ∆W = 0 ⇒ ∆Q = −∆W = P dV = nRT log VV12 .
V1

Thus ∆W > 0 for compression (V2 < V1 ) and ∆Q < 0. Heat is given out to B
by S. Now ∆S = nR log VV21 and so the entropy of S decreases. However, because the
isolated universe of S and B has ∆S = 0 the entropy of B must increase.

2.3.1 The second law

(Kelvin) No process can continuously (by going round a cycle) extract heat from a heat
bath and perform an equal amount of work.
(Clausius) No process can continuously transfer heat from a colder to a hotter body.1

Take a sample S of perfect gas around a closed Carnot cycle by means of adiabatics
and isothermals (T2 > T1 ). On AB the heat bath at T2 supplies heat ∆Q2 > 0 to S at
temperature T2 . On CD the heat bath at T1 supplies heat ∆Q1 < 0 to S at temperature
T1 .
E = E(T ) is unchanged over one complete cycle so that the work done in a cycle
by S is −W = ∆Q1 + ∆Q2 by the first law (as δQ = 0 on adiabatics).
This agrees with the Kelvin statement of the second law, (−∆Q1 ) > 0 of heat is
wasted at the low temperature heat bath. The efficiency η is defined

−W ∆Q1
η= =1+ < 1.
∆Q2 ∆Q2

On adiabatics T V γ−1 is constant and so

T2 VAγ−1 = T1 VDγ−1 and T2 VBγ−1 = T1 VCγ−1 .

VA VD
Thus VB = VC . We know that

∆Q2 = nRT2 log VVB

A
> 0 and ∆Q1 = nRT1 log VVD
C
= −nRT1 log VVB
A
.

T2 −T1
Thus η = T2 , and clearly 0 < η < 1. This can be generalised to other cycles.
1 You can’t break even either... (see footnote on page 6)
2.4. HEAT FLOW 17

2.4 Heat flow

Let S1 be a sample of n1 moles of ideal gas at temperature T1 in a fixed volume V1 .
Then E1 (T1 ) = n1 CV1 T1 = x1 T1 . Similarly for S2 . Put S1 and S2 in thermal contact
allowing no change in V1 + V2 . Suppose the state of thermal equilibrium is reached at
a temperature T . Then (by conservation of energy),

x1 (T1 − T ) + x2 (T2 − T ) = 0

and we can solve for T . As for the entropy, ∆S1,2 = x1,2 log TT1,2 and ∆S1 +∆S2 ≥ 0.
18 CHAPTER 2. THERMODYNAMICS
Chapter 3

Grand ensemble methods

3.1 The formalism

Our approach is a direct extension of the approach we adopted earlier for the canonical
ensemble. Given a system S of fixed volume we construct a grand (canonical) ensem-
ble G with a large number A of distinguishable replicas of S in microstates |ii of S
with Ni particles and energies Ei . Suppose that ai members of G are in the microstate
|ii so that
X X X
ai = A ai Ei = AE ai Ni = AN.
i i i

We have thus fixed the average energy and number of particles of members of the
ensemble. Each configuration {ai } of G defines a state of the ensemble and we assign to
each configuration equal a priori probability. We associate the state of thermodynamic
equilibrium of S with the most probable configuration of G in the presence of our
constraints. We find this by maximising the same log W as before to get
!
∂ X X X
0= log W − α ai − β ai Ei − γ ai N i .
∂ai i i i

This gives us that

ai = e−(1+α) e−β(Ei −µNi ) ,

where we have defined the chemical potential µ by βµ = −γ. We now define the
grand partition function

X
Z= e−β(Ei −µNi ) . (3.1)
i

The fraction of members of G in the microstate |ii is

ai e−β(Ei −µNi )
ρi = = .
A Z

19
20 CHAPTER 3. GRAND ENSEMBLE METHODS

The grand ensemble average is

X
Ō = ρi Oi
i

and Ē = E and N̄ = N (which is, on the whole, a good thing). As before we can
use Z to calculate thermodynamic variables:

 
1∂ log Z
N= , (3.2)
β ∂µ β,V
 
∂ log Z
E − µN = − . (3.3)
∂β µ,V

As before, from consideration of changes at constant ρi we find that δE = −P δV

and

 
X ∂Ei 1 ∂ log Z
P =− ρi = . (3.4)
i
∂V β ∂V β,µ

More general changes in which δρi 6= 0 obey

X X X
δρi = 0 Ni δρi = δN δE = −P δV + Ei δρi .
i i i

Defining the entropy

k X
S=− log Wmax = −k ρi log ρi
A i

gives that

X X 1
δS = −k δρi log ρi − ρi δρi
i i
ρi
X
= −k δρi (−β(Ei − µNi ) − log Z) .
i

This gives the fundamental thermodynamic relation

T δS = δE + P δV − µδN, (3.5)

yielding the first law dE = δQ + δWmech + δWchem , where δWchem = µδN is the
work done in adding δN particles to the system S. The chemical potential is therefore
given by

 
∂E
µ= . (3.6)
∂N S,V
3.2. SYSTEMS OF NON-INTERACTING IDENTICAL PARTICLES 21

Returning to the entropy S we see that

S = kβ(E − µN ) + k log Z (3.7)
which can be put in the same form as (1.18), namely

 
∂kT log Z
S= . (3.8)
∂T µ,V

A similar argument to that used to define temperature in 1.3 gives that two systems
placed in thermal and “diffusive” contact will reach a thermodynamic equilibrium char-
1
acterised by common values of β = kT and µ.
We keep the idea that at fixed NA , NB energy flows from the system with the
higher temperature to that with the lower temperature until thermodynamic equilibrium
is reached. We add that if the contact allows the diffusion of particles then in the state
of thermodynamic equilibrium with constant µ there is (on average) no diffusion.
We return to the first law, dE = T dS − P dV + µdN . We can view E as a function
of S, V and N , and when (as for many large systems), E, S and V are extensive we
get
E(λS, λV, λN ) = λE(S, V, N )
and so
     
∂E ∂E ∂E
E=S +V +N
∂S V,N ∂V S,N ∂N S,V
= T S − P V + µN.
We define the grand potential Ω for a state of thermodynamic equilibrium by
Ω = E − T S − µN
= E − µN − (E − µN + kT log Z)
= −kT log Z.

Thus Z = e−βΩ . Now Ω also equals −P V and so

P V = kT log Z (3.9)
allows the calculation of the equation of state from the grand partition function.
The state of thermodynamic equilibrium corresponds to the most probable state
in G. Some thermodynamic variables arise from averages over microstates weighted
by the probability of finding them in the state of thermodynamic equilibrium. The
averages are effectively averages over all possible states of the ensemble because the
macrostate of thermodynamic equilibrium dominates overwhelmingly. They are also
very sharp (for the same reason).

3.2 Systems of non-interacting identical particles

3.2.1 A little quantum mechanics
We will treat such systems with only one type of particle. When interactions are ne-
glected then the wavefunction of each stationary state of the system Ψ(x1 , x2 , . . . ) is
22 CHAPTER 3. GRAND ENSEMBLE METHODS

obtained by either symmetrization (for bosons) or antisymmetrization (fermions) of the

product φ1 (x1 )φ2 (x2 ) . . . of one particle wavefunctions.
Consider spin 0 bosons in a cube of side L. Then the one particle wavefunctions
are
  32
2
ψn (x) = sin n1Lπx sin n2Lπx sin n3Lπz
L

with corresponding energies

~2 2
En = |n| .
2mL2
For identical bosons we cannot use wavefunctions φ1 (x1 )φ1 (x2 )φ2 (x3 ) which cor-
respond to particles 1 and 2 in state 1 ≡ n1 and particle 3 in state 2 ≡ n2 , but we must
use instead the symmetrized wavefunction
r
1
Ψ= (φ1 (x1 )φ1 (x2 )φ2 (x3 ) + φ1 (x1 )φ2 (x2 )φ1 (x3 ) + φ2 (x1 )φ1 (x2 )φ1 (x3 )) .
3!
All we can say about this is that there are two particles in state |1i and one particle
in state |2i. Ψ is fully determined by the fact that φ1 is used twice and φ2 used once.
For spin 21 fermions we have

φ(x) → ψn (x)α or ψn (x)β,

where α is “spin up” ↑ and β is spin down ↓ with H and En independent of α and
β. We can use determinants to build the antisymmetric wavefunctions.
For instance, given ψ1 (x, µ) and ψ2 (x, µ) (where 1 ≡ n1 , 2 ≡ n2 and µ = α or
β) we get the antisymmetrized version
r
1 ψ1 (x1 , µ1 ) ψ1 (x2 , µ2 )
Ψ(x1 , µ2 , x2 , µ2 ) = .
2! ψ2 (x1 , µ1 ) ψ2 (x2 , µ2 )

Similarly for three particles,

r ψ (x , µ ) ψ1 (x2 , µ2 ) ψ1 (x3 , µ3 )
1 1 1 1
Ψ= ψ2 (x1 , µ1 ) ψ2 (x2 , µ2 ) ψ2 (x3 , µ3 ) .
3!
ψ3 (x1 , µ1 ) ψ3 (x2 , µ2 ) ψ3 (x3 , µ3 )

Ψ reflects the Pauli exclusion principle which forbids any two of the ψi or any two
sets (x, µ) from being the same.
Each Ψ is determined fully by the number of times each one particle wavefunction
is used in the product term that we (anti)symmetrize.

3.2.2 The partition functions

Let the one particle wavefunctions of the particles of S be ψr (x) with energy r . Sup-
pose that in the microstate |ii, nr of the particles have wavefunction φr (x).
Then
X X
Ni = nr and Ei =  r nr .
r r
3.2. SYSTEMS OF NON-INTERACTING IDENTICAL PARTICLES 23

We see that the microstate |ii of S is fully determined by i ≡ {nr }. We obtain the
full set of microstates of S by letting the nr vary without restriction over their allowed
range of values. (Thus Ni and Ei cannot themselves be restricted — this is why we
use the grand ensemble method.)
We can now write down the grand partition function

X
Z= e−β(Ei −µNi )
i
X
= e−β(n1 1 +n2 2 +···−µn1 −µn2 −... )
n1 ,n2 ,...
YX
= e−βnr (r −µ) .
r nr

For fermions we have that nr can only be zero or one, so that

Y 
Z= 1 + e−β(r −µ) . (3.10)
r

We use this to find

 
1 ∂ log Z X 1
N= = .
β ∂µ β,V r
eβ(r −µ) + 1

Now

X nr e−β(r −µ)
n¯r =
n1 ,n2 ,...
Z
 
1 ∂ log Z
=− s 6= r
β ∂r β,µ,s
1
= β( −µ) .
e r +1
th
P
We see that N = r n̄r . n̄r is the average occupation number of the r one
particle state at thermal equilibrium. This is the Fermi-Dirac distribution.
For bosons, 0 ≤ nr < ∞ and
Y 1
Z= −β(
.
1−e r −µ)
r

This gives that

1
n̄r = . (3.11)
eβ(r −µ) −1
This is the Bose-Einstein distribution.
If there are gr one particle states of energy r we can write the average number of
particles with energy r as
gr
n̄(r ) = .
eβ(r −µ) ∓ 1
24 CHAPTER 3. GRAND ENSEMBLE METHODS

for N large we pass to the continuum limit,

X Z
→ dg(),
r

1
where g() is the density of states factor, DV  2 as in section 1.5.3, where D =

32
2π 2m
h2 gS and gS = 2S + 1 is the spin degeneracy.
The average number of particles with energy in the range  →  + d is

g()d
n()d = .
eβ(−µ) ∓ 1
Thus in the continuum limit, the grand partition function Z is given by
Z ∞
log Z = ∓ g() log(1 ∓ e−β(−µ) ) d.
0

We can now use the results of section 3.1 to find things like N and E,

∞ ∞
g() d g() d
Z Z
N= β(−µ)
E= .
0 e ∓1 0 eβ(−µ)∓1
1
In three dimensions, g() = DV  2 and so (integrating by parts) we see that

∞ 3
2 2
Z
log Z = DV β(−µ)
0 3e ∓1
2
= β.
3
1
Combining this with (3.9) we find that P V = 32 E. The 2
3 comes from the  2 factor
in g().

3.2.3 Classical limit

1
For a volume V of N particles with energy E, we use g() = DV  2 and so
∞ 1 ∞ 3
 2 d  2 d
Z Z
N = DV β(−µ)
and E = DV β(−µ)
.
0 e ∓1 0 e ∓1
3 5
Putting z = β we find N = DV (kT ) 2 I 12 (−βµ) and E = DV (kT ) 2 I 32 (−βµ),
with
Z ∞
z n dz
In (y) = .
0 ez+y ∓ 1

If ey  1 we can neglect the 1 in the denominator and find that n() ∝ e−β g().
We can approximate the integrals (expand the integrand) to get

eβµ
 
E ∼ 23 N kT 1 ∓ √ + . . . .
4 2
3.3. BLACK BODY RADIATION 25

3
√
π
In the lowest approximation E = 23 N kT and N = DV (kT ) 2 2 .
The condition e−βµ  1 is thus
  32
2mπkT V
1 if gS = 1.
h2 N

This is a classical limit — it holds when h is very small on a scale defined by m

and the mean energy per particle 32 kT . This condition is satisfied at low density N V
and
at high temperature.
It is true for all real gases except helium at very low temperature and very high
density. Most real gases liquefy before quantum mechanical effects set in.
It fails to holds for electrons in solids due to the fact that their effective mass inside
the solid is much less than the free electron mass.
We can use our formulae for E and N to get the lowest order quantum correction
to the equation of state — recall that P V = 32 E, so

NP 1
P V = N kT ∓ 3 √ .
(kT ) D 8π
2

3.3 Black body radiation

1
Consider a cubic cavity of side L = V 3 inside a perfectly black body, the walls of
which are maintained at a fixed temperature T . The atoms of the perfectly black body
absorb all photons incident on them and independently emit photons such that even
in thermodynamic equilibrium the number of photons varies significantly. Thus the
total number of photons is not conserved and so no constraint can be applied to them.
Therefore µ = 0.
The number of photons depends on the temperature. We must consider a gas of
photons in a volume V at a temperature T . These quanta of the electromagnetic field
are relativistic bosons of rest mass 0.
Our previous work applies, setting µ = 0. Thus

1 X r
n̄r = and E = .
eβr − 1 r
eβr − 1

We pass to the continuum limit using (1.22) to get

∞ ∞
8πV ν 2 dν 8πV hν 3 dν
Z Z
N= 3 and E = 3 ,
c 0 eβhν − 1 c 0 eβhν − 1

where  = hν. This is Planck’s law. We expect classical behaviour at high T or

hν hν
low ν, where e kT − 1 ≈ kT and

8πV 2
dE = ν dνkT = kT g(ν) dV,
c3
which agrees with the classical result of energy kT per normal mode of radiation.
This is the classical equipartition of energy. Using this result, dE 2
dν ∝ ν gives a diver-
gent energy, which is called the ultraviolet catastrophe.
Using the full formula for the energy we find that
26 CHAPTER 3. GRAND ENSEMBLE METHODS

∞
E 8π z 3 dz
Z
e= = (kT )4
V (hc)3 0 ez − 1
= σT 4 .
5 4
8π k
This is Stefan’s law, and σ = 15h 3 c3 is Stefan’s constant. We see that the energy

density is a function of T only.

Using (3.7) with µ = 0 we see that

E
S = k log Z +
T
4E S 4 3
and we can evaluate k log Z = 3T . The entropy density V = 3 σT → 0 as
T → 0. The energy
  43
4 3S 1 1
E = V σT = 1 ∝ V − 3 at constant entropy.
4 (V σ) 3

∂E 1E 1 4


The pressure P can now be calculated from P = − ∂V S
= 3V = 3 σT .
Finally the density of photons, n = N
V ∝ T 3
(see example sheet 3).

3.4 Degenerate Fermi gas

This is the extreme quantum limit. It occurs for

• electrons in solids,

• electrons in white dwarf stars,

• neutrons in neutron stars and

• nucleons in nuclear matter.

For N electrons in a volume V at a temperature T we have

Z ∞ Z ∞
N= g()F () d and E = g()F () d,
0 0

√ 2m

32
where g() = DV , D = 4π h3 and

1
F () = .
eβ(−µ) +1

The range 0 ≤ e−βµ < ∞ is allowed without blowup of F . The region e−βµ  1
is the classical realm and we expect that quantum mechanical effects will be most
pronounced at low temperature, eβµ  1.
We assume that µ is finite and positive and µ = EF = kTF is constant at T = 0.
We further assume that µ = EF + O(kT ) for small T . It can be shown that µ =
EF + O(kT )2 .
3.4. DEGENERATE FERMI GAS 27

Now
(
β(−µ) 0  < EF
lim e =
T →0 ∞  > EF ,

so that

(
1  < EF
lim F () =
T →0 0  > EF .

Thus at T = 0 F () = θ( − EF ). At T = 0 it is energetically most favourable for

electrons to fill up the one particle energy eigenstates (two electrons at a time, one spin
up and one spin down) with increasing energy according to the Pauli principle, until
the Fermi energy is reached. EF is the highest energy occupied at T = 0.
We can now perform the integrals for N and E to get

3
N = DV 32 EF2 (3.12)
5
2 2 3
E = DV 3 EF = 5 N EF . (3.13)

We can solve (3.12) for EF to get

  23
3N 2
EF = ∝ V −3 .
2DV
5
The equation of state P V = 23 E becomes P V 3 = const. Classical kinetic theory
gives P → 0 as T → 0. The Pauli principle requires particles to have non-zero
momentum and so there is pressure even at T = 0.
In fact
 5
2 2 N 3 h2
P = 33 ∝ m−1
3 V 2m

and so at T = 0 the pressure is much bigger for lighter particles.

3.4.1 Heat capacity at low temperature

We have

Z ∞
N= g()F () d and
Z0 ∞
E= g()F () d.
0

We want to find C = ∂E ∂


∂T V
(which is proportional to the usual CV ). We take ∂T
of both of the above to find
∞
∂F
Z
C= g() ( − EF ) d.
0 ∂T
28 CHAPTER 3. GRAND ENSEMBLE METHODS

At low temperature ∂F∂T is very like a delta function and so the region  ≈ EF
dominates the integral.
We approximate g() ≈ g(EF ) and µ ≈ EF . This gives us

∞
g(EF ) eβ(−EF )
Z
2
C∼ d( − EF )
2 ( − EF )
kT 2 0 eβ(−EF ) +1
∞ 2 z
z e dz
Z
= g(EF )k 2 T 2.
−βEF (ez + 1)

We know that EF is finite and positive, so that −βEF → −∞ as T → 0. We therefore

approximate the lower limit of the integral by −∞.
∞
z 2 ez dz
Z
C ∼ g(EF )k 2 T 2.
−∞ (ez + 1)
π2
This leaves a convergent integral whose value is 3 , so that

π2 T
C∼ Nk ,
2 TF
where TF (the Fermi temperature) is defined by EF = kTF . This approximation
is expected to be good for T  TF .

3.5 Bose-Einstein condensation

Recall the result
Z ∞ 1 Z ∞ 1 √
 2 d 3 z 2 dz 3 π
N = DV β(−µ) − 1
= DV (kT ) 2
z−βµ − 1
= DV (kT ) 2 f (−βµ),
0 e 0 e 2

where
∞
X enβµ
f (−βµ) = 3 .
n=1 n2

f (−βµ) is convergent iff βµ ≤ 0. Assuming this, the Bose-Einstein denominator

does not vanish for any point in the range of integration.
It is easy to see that f has a maximum value (of 2.612) at βµ = 0 and decreases
monotonically as βµ decreases through negative values from βµ = 0.
Now suppose that N V is fixed. Our equation for N is OK as T decreases since
f (−βµ) can increase by βµ getting less negative. When βµ reaches 0 problems occur
as f cannot increase any more. This will be at T = TB , given by
√
N 3 D πf (0)
= D̂(kTB ) 2 D̂ = .
V 2
Our equation for N appears to fail beyond this. Why?
1
More care is called for in the passage to the continuum limit. The  2 factor in g()
assigns 0 weight to the particles in the  = 0 state. This would be no problem for
3.6. WHITE DWARF STARS 29

fermions (which have the Pauli principle), but no law stops bosons from condensing
into the  = 0 state if need be. We write N = N0 + NC , where N0 is the number of
particles in the  = 0 state and NC is the number of particles in  > 0 states. Then for
T > TB
N √
= D 2π f (−βµ).
V
N 3
For T = TB , V = D̂(kTB ) 2 , defining TB . For T < TB then
  32
NC 3 NC T
= D̂(kT ) 2 , giving = .
V N TB

3.6 White dwarf stars

White dwarfs are abnormally faint: they are stars in which the hydrogen supply has run
out so that they are composed mainly of helium.
Typically,

T ≈ Tsun = 107 K
M ≈ Msun , and
ρ ≈ 107 ρsun .

We regard them as being a mass of helium at T ≈ 107 K and under extreme com-
pression. For T ≈ 107 K, kT ≈ 103 eV is much greater than the binding energy of
electrons to helium, so that thermal collisions completely ionize all the atoms, produc-
ing an electron gas.
To a good approximation this may be viewed as a Fermi gas in the degenerate
(negligible T ) limit. This may be seen by computing TF , the natural temperature scale
of the problem (the Fermi temperature). Now
 23
h2

3N
kTF = EF = ≈ 2 × 106 eV.
2m 8πV
11
Since N V is so large TF ≈ 10 K  T .
The helium nuclei neutralise the charge of the star and produce the gravitational
attraction which counteracts the extreme zero-point pressure of the electron gas (which
dominates the zero-point pressure from the nuclei).
We can calculate the total energy E = Eelec (R) + Egrav (R) and find the radius R0
of the star by minimising this.
Treating the electrons relativistically we have
EF
N 8π
Z
1
= 3 3  d(2 − m2 c4 ) 2 .
V ~ c mc2
√ N
Let 2 − m2 c4 = mc2 x, so that the integral for V can be done to get

N 8π x3
= 3 3 (mc2 )3 F .
V ~ c 3
30 CHAPTER 3. GRAND ENSEMBLE METHODS

The integral for the energy density is

Z xF
E 8π p
= 3 3 (mc2 )4 x2 dx 1 + x2
V ~ c 0
 
and so E
N = 34 mc2 xF + x1F + O(x−2
F ) . We know that V ∝ R3 and so xF ∝
R−1 . Hence
a γ
E= + bR − ,
R R
where a, b and γ are positive constants. This only has a minimum if γ ≥ a, which
leads to the Chandrasekhar upper limit on the mass of a white dwarf.
Chapter 4

Classical statistical mechanics

4.1 Introduction
Recall the result Z = z N for a system of distinguishable non-interacting particles. It
can be shown that in the classical limit
X X
z= e−βr = hr|e−β Ĥ |ri
r r

gives
Z
z = h−3 d3 p d3 q e−βH(p,q) , (4.1)

where Ĥ(p̂, q̂) is the quantum mechanical Hamiltonian and H(p, q) is the classical
Hamiltonian as a function of classical variables p and q.
The average for the system of a physical variable f (p, q) is
R 3 3
d p d q f (p, q)e−βH(p,q)
hf (p, q)i = R .
d3 p d3 q e−βH(p,q)
 
log z
We can see that hH(p, q)i = − ∂ ∂β agrees with the result
V
   
∂ log Z ∂ log z
E=− = −N .
∂β V ∂β V

Now

e−β
i=1 H(pi ,qi ) ,

N  3
d p d3 q
Z Y 
N
Z=
i=1
h3
PN
where i=1 H(pi , qi ) is the Hamiltonian of the N -particle system.
p2
For a monatomic gas with H = 2m we have
Z 3   23
V − βp
2
V 2πm
z= 3 dp e 2m = 3 .
h h β
This result shows that E = 32 N kT , which is the common classical result of energy
1
2 kTper degree of freedom per particle.

31
32 CHAPTER 4. CLASSICAL STATISTICAL MECHANICS

4.2 Diatomic gases

We will study the rigid dumbbell model as shown.

We want to write down the partition function Z = z N . To apply (4.1) we need the
Hamiltonian H. To find the Hamiltonian we first write down the Lagrangian
 
L = 21 mẋ2 + 12 I θ̇2 + sin2 θφ̇2 .

(V ≡ 0 in this rigid case.) Note that the Lagrangian splits neatly into translational
motion of G and rotational motion about the centre of mass.
We define the generalised momenta pα = ∂∂L q̇α :

pi = mẋi , pθ = I θ̇ and pφ = I sin2 θφ̇.

Now
X p2 p2 p2φ
H= q̇α pα − L = + θ + ,
α
2m 2I 2I sin2 θ

and
1
Z
z= d2 p d3 q dpφ dφ dpθ dθ e−βH = zt zr ,
h5

where
  32
1 2mπ
Z
3 3 − 12 mβp2
zt = 3 d pd qe =V
h βh2

and
 p2
1 β
p2θ + sinφ  = 8π2 IkT .
Z
− 2I 2θ
zr = 2 dpφ dφ dpθ dθ e
h h2

We evaluate the above integral by first doing the pθ , pφ and φ integrals and then
doing the θ integral.

23 8π2 IkT
Thus z = zt zr = V 2mπkT
h2 h2 . This gives E = 25 N kT .1

4.3 Paramagnetism
Each molecule of an N molecule solid acts as a little magnet fixed at its own lattice
site and free only to rotate about it. Each molecule has a dipole moment m and gives
1 Which ought to be expected!
4.4. SPECIFIC HEATS 33

a contribution −m · B to the energy when in an applied magnetic field B = (0, 0, B).

Now
p2 p2 p2φ
H= + θ + − mB cos θ
2m 2I 2I sin2 θ
and so
2πI 2 sinh y
z= 2π
h2 β y
where y = mβB. We can calculate the magnetisation of the solid:
M = (0, 0, M ) = N hM i.
Only the third component hm cos θi is non-zero:
 
−1 ∂ log z
= m coth y − y −1 .


hm cos θi = β
∂B β

N m2 B
For small y (high T ) we thus have Curie’s law, M = 31 N my = 3kT .

4.4 Specific heats

The result E = 52 N kT for a diatomic gas is found to be accurate at sufficiently high
temperatures. E is found to be 32 N kT at lower temperatures. It is as if the rotational
degrees of freedom are frozen out. This is explained by quantum mechanics.
1 2
We have Z = z N and z = zr zt . We will look at zr , using Hr = 2I L (the
quantum mechanical angular momentum operator). Now
l(l + 1)~2
Hr |l mi = |l mi
2I
for m = −l, −l + 1, . . . , l − 1, l for each of l = 0, 1, 2, . . . . Thus
∞
X l(l+1)Tr
zr = (2l + 1)e− T (4.2)
l=0
2
~
with Tr = 2Ik . Tr is typically about 50K and is experimentally accessible.
For Tr  T (most gases at normal temperatures) we can turn this sum into an
integral to get
 2
T 8π
zr = = kT
Tr h2
as before. For Tr  T all of the terms with l 6= 0 in (4.2) are exponentially small
and we take only the l = 0 term to get zr ≈ 1 — there is no rotational contribution to
the energy (or heat capacity). The contribution to the energy from rotational motion is
therefore N kT if T  Tr or 0 if T  Tr .
For high temperatures, E rises to 72 N kT due to vibrational motion along the axis
of the dumbbell. There is an extra term in the Hamiltonian
1 2 mω 2 Q2
Hv = P +
m 4
and an extra factor zv in the one particle partition function.
34 CHAPTER 4. CLASSICAL STATISTICAL MECHANICS

4.5 Weak interparticle forces

Consider a classical gas of N molecules with Hamiltonian
N
X p2r X
H= + φ(Rrs ),
r=1
2m r<s

where Rrs = |qr − qs |. We use

N
d3 pr d3 qr −βH
Z Y
Z= e
r=1
h3
  3N
2πmkT 2 N
= V K
h2
= Z0 K,

where Z0 = z0N and z0 is the partition function for one molecule of a spinless gas
of N non-interacting molecules and

d3 qr e−β
r<s φ(Rrs ) .

Z Y
K = V −N
r

We suppose that λrs = e−βφ(Rrs ) − 1 is small (weak interaction) and we treat our
results to lowest non-trivial order.
Then
Z Y !
X
K = V −N d3 qr 1 + φ(Rrs )
r r<s
N (N − 1)
Z
=1+ d3 qa d3 qb λab (no summation).
2V 2

This integral can be evaluated using the change of variables x = qa +q

2
b
and y =
qa − qb , so that
Z Z  
d3 qa d3 qb λab = V d3 y e−βφ(y) − 1 ≡ V f (T ),

N 2 f (T )
defining f (T ). Then K = 1 + 2V (as N − 1 ∼ N ) and

N 2 f (T )
log Z = N log V + + stuff which does not depend on V .
V
 
∂ log Z
Using βP = ∂V we get
β
  ∞
N f (T )
Z  
P V = N kT 1− with f (T ) = 4πy 2 dy e−βφ(y) − 1 . (4.3)
2V 0

If we take a specific φ we can recover the van der Waals’ equation of state (see page
15).
4.6. THE MAXWELL DISTRIBUTION 35

Consider the φ shown.

Then the contribution to f (T ) from region 1 where φ is infinite is

Z d
4πy 2 dy = −2b.
0
For large T and weak attractive forces the contribution from region 2 is
Z ∞
4πy 2 dy (−βφ(y)) = 2βa.
d
a and b are positive constants. Now f (T ) = −2(b − βa) and (4.3) becomes
N 2a
   
Nb
P + 2 V = N kT 1 + .
V V
For NV b small we can (approximately) take 1 + NVb to the left hand side of the
equation to get
N 2a
 
P + 2 (V − N b) = N kT
V
and the correspondence with (2.5) is completed by setting A = N 2 a and B = N b.
Note that B is the volume of all the molecules.

4.6 The Maxwell distribution

2
p
Consider a gas with (one particle) Hamiltonian 2m .
The number of molecules in the region p → p + dp is
N e−βH βp2
n(p) d3 p = R −βH 3
= Ce− 2m .
e d p
Now
Z Z Z
1 2
N= n(p) d3 p = m3 C 4πv 2 dve− 2 βmv = N f (v) dv,

which defines the Maxwell distribution of speeds,

1 2
f (v) = const × v 2 e− 2 βmv .
f (v) is the probability of finding
R ∞ a particle with speed in v → v + dv. The constant
is (of course) chosen to make 0 f (v) dv = 1. We can define averages in the obvious
way, that is
Z ∞
hg(v)i = f (v)g(v) dv.
0
36 CHAPTER 4. CLASSICAL STATISTICAL MECHANICS
Approximations

Stirling’s formula
We derive the approximation for Stirling’s formula from the Γ function using Laplace’s
method.
Recall that

Z ∞
Γ(z) = tz−1 e−t dt
0

and Γ(n + 1) = n!. Then

Z ∞ Z ∞
n! = tn e−t dt = en log t−t dt.
0 0

t
Let v = n, so that
Z ∞
n! = nn+1 en(log v−v) dv.
0

, so
2
n −1− (v−1)
2
n(log v−v)
Now for large n, e ∼e
Z ∞
nu2 √
n! ∼ nn+1 e−n e− 2 du = 2πne−n nn .
−∞

This is Stirling’s formula.

37