Part II Statistical Mechanics, Lent 2005

Prof. R.R. Horgan

February 14, 2005

1 Statistical Mechanics
1.1 Introduction
The classical and quantum dynamics of simple systems with few degrees of freedom
is well understood in that we know the equations of motion which govern their time
evolution and, in most cases, we can solve these equations exactly or, in any case, to
the required degree of accuracy. Even systems with many degrees of freedom can be
analyzed, for example, many problems in electromagnetic theory, where the field degrees
of freedom are large, can be solved using Maxwell’s equations. However, it remains that
there are physical systems, which are typified by possessing a large number of degrees
of freedom, for which analysis of this kind in inappropriate or, in practice, impossible,
Whilst these systems are made up of identifiable constituents whose interactions are
well known, different concepts are needed to describe the evolution of the system and
the way it interacts with an observer. In particular, the concepts of heat, temperature,
pressure, entropy, etc. must arise from a proper analysis – they are certainly not
evident in the equations of motion, even though knowledge of the equations of motion
is indispensable. These are ideas which come from a notion of statistical averaging
over the detailed properties of the system. This is because they only make sense when
thought of as properties which apply on the large scales typical of the observer, rather
than the microscopic scales of the individual constituents. Statistical mechanics was
developed to address this conceptual problem. It enables one to answer questions like
“how do you calculate the physical properties of a system in thermal contact with large
reservoir of heat (known as a heat bath) at temperature T ?”

Statistical mechanics deals with macroscopic † systems with very many degrees of
freedom. For example, a gas with a large number of particles or a complex quantum
system with a large number of quantum states. A thermally isolated or closed
system is one whose boundary neither lets heat in or out. Consider such a system,
S, that is not subject to any external forces. No matter how it is prepared, it is an
experimental fact that after sufficient time S reaches a steady state in that it can be
specified by a set, Σ, of time-independent macroscopic or thermodynamic, variables
which usefully describe all the large-scale properties of the system for all practical
purposes. The system is then said to be in a state of thermal equilibrium specified
†
By macroscopic we mean that the system is large on the scale of the typical lengths associated
with the microscopic description, such as the inter-atomic spacing in a solid or mean particle separation
in a gas. That being said, it can be the case that systems with as few as a hundred particles can be
successfully described by statistical mechanics.

1
 1 STATISTICAL MECHANICS 2

by Σ. For example, for a gas isolated in this manner, Σ includes the pressure P, volume
V, temperature T, energy E, and entropy S.
Although the state of thermal equilibrium described by Σ is unique and time-
independent in this sense, the individual degrees of freedom such as particle positions
and momenta are changing with time. It is certainly not practical to know the details of
how each degree of freedom behaves, and it is a basic tenet of statistical mechanics that
it is not necessary to know them in order to fully specify the equilibrium state: they are
not of any practical importance to calculating the accessible experimental properties of
the equilibrium system. The task of statistical mechanics is to use averaging techniques
and statistical methods to predict the variables Σ characterizing the equilibrium state.
We are generally able to describe the dynamics of such systems at the microscopic
level. For instance, we can specify the interaction between neighbouring atoms in a
crystal or between pairs of particles in a gas. This description arises ultimately from
our knowledge of quantum mechanics (and relativity), but it may also be legitimate
and useful to rely on classical mechanics which we view as derivable from quantum
mechanics by the correspondence principle. We shall develop the ideas of statistical
mechanics using quantum mechanics in what follows. We define

(a) The microscopic states of S are the stationary quantum states |ii.
 
This is the Dirac ‘bra’ (hi|) and ‘ket’ (|ii) notation for states. The label i stands implicitly 
for the complete set of quantum numbers which specify the state uniquely. For example, 
 
H|ii = Ei |ii. The matrix element of an operator X between states |ii and |ji is hi|X|ji 
 R 
which, in wave-function notation, is the same as dx φ∗ Xφj . Then hi|X|ii is the expectation
 i 
value, hXi, of X.

(b) The macroscopic states of S are the possible states of thermodynamic equilibrium
and are described by the corresponding set of thermodynamic variables, Σ. These are
not states in the quantum mechanical sense but involve a vast number of microstates.

An important idea is the ergodic hypothesis (which has only been proved for a
few systems† ) which states that:

A system S evolves in time through all accessible microstates

compatible with its state of thermodynamic equilibrium.

This means that time averages for a single system S can be replaced by averages at a
fixed time over a suitable ensemble E of systems, each member identical to S in its
macroscopic properties.
An important feature is that the ensemble, E, can be realized in different ways
whilst giving the same results for the thermodynamic properties of the original system.
The major realizations of E are:

• the microcanonical ensemble: each member of E has the same value of E and N .
This is appropriate to an isolated system for which the energy is fixed.
†
There are many examples in one dimension where the ergodic hypothesis does not hold. This is a
current research area
 1 STATISTICAL MECHANICS 3

• the canonical ensemble: each member of E has N particles but E is not fixed but
the average energy over E is fixed. This is appropriate to a system of a fixed number of
particles but which is in thermal contact with a heat bath with which it can exchange
energy. The heat bath is all the other systems of the ensemble.

• the grand canonical ensemble: neither E nor N is fixed for each member but the
ensemble averaged values are fixed. This describes a system in thermal contact with a
heat bath with which particles can also be exchanged.

Members of the last two ensembles represent the original system in the average. The
reason why they all give the same thermal physics is that the fluctuations of E and N
about their average values are very small because the number of particles involved is
very large. For example, consider the probability p m of obtaining N/2 + m heads from
N coin tosses:   r
−N N 2 −2m2 /N
pm = 2 ∼ e ,
N/2 + m πN
using Stirling’s formula (log n! = n log n − n + 12 log 2πn) for N  m  1. We see
√
that
√ only for m  N is pm appreciable in size. Hence the mean fluctuation, m/N ∼
1/ N , is very small.
The density of a volume of gas V , ρ = N/V , is a macroscopic √ quantity on which
thermodynamic variables depend. If the fluctuations in N are O( N ) then
r
ρ
δρ ∼ ,
V
which vanishes as V −→ ∞. Hence, we expect the results from the different ensembles
√
to be the same in the infinite volume limit but to differ by corrections of O(1/ V ).
The large volume limit is always assumed in what follows.
In this course we mainly study the canonical ensemble, but for fermions and bosons
it will be simpler to use the grand canonical ensemble.

1.2 The 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. We assume that the spectrum is discrete with
possible degeneracy.
We associate with S a canonical ensemble E. This is a very large number A of
distinguishable replicas S1 , S2 , S3 , . . . of S. Suppose we were to measure a complete
set of commuting observables for all members of E simultaneously. Then let a i be the
number of members found in state |ii. Then
X X
ai = A , and ai Ei = AE , (1.2.1)
i i

where E is the fixed, time-independent ensemble average energy of the members of E.

In other words ai /A is the (quantum mechanical) probability for finding the system in
microstate |ii. The set {ai } is called a configuration of E, and the number of ways of
partitioning A into the set {ai } is

A!
W (a) = Q . (1.2.2)
i ai !
1 STATISTICAL MECHANICS 4

The important principle underlying statistical mechanics is that we assign equal a

priori probability to each way of realizing each allowed configuration. This means
that the probability of finding configuration {a i } is proportional to W (a). In the light
of the earlier discussion, the probability distribution for the {a i } will be √
sharply peaked
and the fluctuations about the peak value will be very small: ∼ O(1/ A). Since we
can take A as large as we like, the average of any variable over the ensemble will be
overwhelmingly dominated by the value it takes in the most probable configuration,
{āi }. For fixed N, V and E, we shall associate the state of thermodynamic equilibrium
of S with the most probable configuration of E that is consistent with the constraints
(1.2.1).
We find {āi } by maximizing log W (a) w.r.t. {ai }. Using Stirling’s formula, (log n! =
n log n − n + 12 log 2πn), we have
X X
log W ∼ A log A − A − ai (log ai − 1) = A log A − ai log ai , (1.2.3)
i i
P
since i ai = A. We maximize this expression subject to the constraints (1.2.1) by
using Lagrange multipliers. Then we have
!
∂ X X X
A log A − ai log ai − α ai − β ai Ei = 0 , ∀i . (1.2.4)
∂aj i i i

Thus

log aj + 1 + α + βEj = 0 ,
=⇒
aj = e−1−α−βEj . (1.2.5)

We eliminate α using (1.2.1), and define Z, the canonical partition function:

X
A = ai = e−1−α Z ,
i
with
X X
Z = e−βEi = Ω(E)e−βE , (1.2.6)
i E

where Ω(E) is the degeneracy of levels with energy E, and E runs over all distinct
values in the spectrum of the system.
The fraction of members of E found in microstate |ii in the macrostate of thermo-
dynamic equilibrium is
ai 1 −βEi
ρi = = e . (1.2.7)
A Z
This is the Boltzmann distribution and is thought of as the (quantum mechanical)
probability of finding |ii in the state of thermodynamic equilibrium. The average hXi
of a physical observable is then
X
hXi = hi|X|iiρi . (1.2.8)
i

For example,
X 1 X
hEi = hi|E|iiρi = ai Ei = E , (1.2.9)
i
A i
1 STATISTICAL MECHANICS 5

from (1.2.1), as we expect it should.

Z is very important since we shall see that it allows us to calculate the thermody-
namic and large-scale properties of a system in equilibrium from the quantum mechan-
ical description of the system. An important example is that
 
∂ log Z
hEi = − . (1.2.10)
∂β V

Holding V fixed means holding the Ei fixed since these latter depend on V .
As emphasized earlier for the canonical ensemble we can think of each individual
system as being in contact with a heat bath made up of the rest of the ensemble. What
are the fluctuations in energy of a system? Using above results, we have

1 ∂Z 2
 
∂E 1 ∂2Z
= − +
∂β Z ∂β 2 Z 2 ∂β
= −hE i + hEi2 = − (∆E)2 .
2
(1.2.11)

For large systems typically E ∼ N and, as E depends smoothly on β we expect

√
∂E |∆E| N 1 1
∼N , =⇒ ∼ = N−2 ∝ V −2 . (1.2.12)
∂β E N

We have found that energy plays a vitally central rôle in determining the probability
for finding a system in a given state. In principle, the ρ i can depend on other variables
too. For example, the particle number N i , corresponding to the grand ensemble, or
even angular momenta where relevant. In general, such quantities must be be conserved
and observable on the large scale.

1.3 Temperature
Consider two systems Sa and Sb with volumes Va , Vb and particle numbers Na , Nb , re-
spectively. Form a joint ensemble E ab from the separate ensembles Ea and Eb by allowing
all members of both to exchange energy whilst keeping the overall total energy fixed.
By allowing Sa and Sb to exchange energy we are allowing for interactions between
the particles of the two systems. These will be sufficient to establish equilibrium of S a
with Sb but otherwise be negligible. For instance, they take place across a boundary
common to the systems and so they contribute energy corrections which scale with
the boundary area and so are therefore negligible compared with the energy, which
scales with volume. Now form a composite system S ab by joining Sa and Sb together.
Let Sa have microstates |iia with energy Eia and Sb have microstates |iib with energy
Eib . Because the extra interactions can be neglected in the bulk, then S ab will have
microstates |ij >ab = |i >a |j >b with energy Eij ab = E a + E b .
i j
Now, Ea and Eb are separately in equilibrium characterized by Lagrange multipliers
βa and βb , respectively. Also, Eab is an equilibrium ensemble which is characterized by
βab , and so we must have

ρab a b a a b b
ij (Ei + Ej ) = ρi (Ei )ρj (Ej ) , ∀ Eia and Ejb . (1.3.1)

Then a b a b
e−βab (Ei +Ej ) e−(βa Ei +βb Ej )
= , ∀ Eia and Ejb . (1.3.2)
Zab Za Zb
1 STATISTICAL MECHANICS 6

This can only be satisfied if βab = βa = βb and hence Zab = Za Zb .

It is interesting to note that the result (1.3.1) could be given as the definition of
what it means for Sa to be in thermal equilibrium with Sb . This is because we expect
that in equilibrium these probabilities depend only on the energies of the systems.
Then relation (1.3.1) must hold since energies add and probabilities multiply. The only
solution to (1.3.1) is the Boltzmann distribution (1.2.7).
The relation (1.3.1) can be established directly by finding the most probable dis-
tribution for the ensemble Eab . Suppose |iia occurs ai times, |jib occurs bj times and
|ijiab occurs cij times in Eab . Then we have
X X
ai = A , bi = A (1.3.3)
i i
X X X
ai = cij , bj = cij , cij = A . (1.3.4)
j i ij

We only have one constraint on the energy which is that the total energy of E ab is fixed:
X X
AE = ai Eia + bj Ejb , (1.3.5)
i j
X
AE = cij (Eia + Ejb ) . (1.3.6)
ij

The configuration {ai , bj } arises in

A! A!
W (a)W (b) ∝ Q Q , (1.3.7)
i ai ! j bj !

ways and the configuration {cij } arises in

A!
W (c) ∝ Q , (1.3.8)
ij cij !

ways. The most probable distribution of {a i , bj } is obtained by maximizing W (a)W (b)

subject to (1.3.4) and (1.3.6) to get
a ai
ai = e−1−αa −βEi , ρai = ,
A
b bj
bj = e−1−αb −βEj , ρbj = .
A
Note that since there is only one energy constraint the most probable {a i } and {bj }
depend on the same value of β. The temperature, T , is defined by the statement that
two systems in thermal equilibrium have the same temperature. We therefore conclude
that

two systems in thermal equilibrium are described by Boltzmann

distributions with a common value of β which can therefore be
thought of as defining the temperature

The most probable distribution of {c ij } is obtained by maximizing W (c) subject to

the constraints (1.3.4) and (1.3.6) to get
0 a b cij
cij = e−1−αab −β (Ei +Ej ) , ρab
ij = . (1.3.9)
A
 1 STATISTICAL MECHANICS 7

So far β and β 0 do not necessarily have the same value since they are just Lagrange
multipliers imposing the energy constraint in two separate maximizations. However,
the relations between {cij } and {ai , bj } given in (1.3.4) can be only be satisfied if β 0 = β.
It then follows that (c.f. (1.3.1))

Zab (β) = Za (β)Zb (β) , and ρab a b

ij = ρi ρj . (1.3.10)

The temperature can be defined by

1
β = , (1.3.11)
kT
where k is Boltzmann’s constant and T is in degrees Kelvin. Because the analysis
applies to any pair of systems k is a universal constant: k = 1.38 10 −23 Joules/0 K.
This definition ensures that the average energy increases as T increases. For a dilute
gas with N particles this implies Boyle’s law, P V = N kT , as we shall see.

1.4 Thermodynamic ideas

A system undergoes an adiabatic change if

(i) the system is thermally isolated so that no energy in the form of heat can cross the
system boundary: δQ = 0,

(ii) the change is caused by purely by external forces acting on the system,

(iii) the change takes place arbitrarily slowly.

By heat Q we mean energy that is transfered in a disordered way by random processes.

In contrast, work W is energy transfered through the agency of external forces in an
ordered way.
Work can be done on the system by moving one of the walls (like a piston) with an
external force, which corresponds to changing the volume. If the change is adiabatic,
no energy is permitted to enter or leave the system through the walls. Because the
motion is slow the applied force just balances the pressure force P A (A is the area
of the wall), and the work done on the system is then δW = −P AδL = −P δV . If
the system is in a given microstate with energy E i the change is slow enough that it
remains in this state – there is no induced quantum-mechanical transition to another
level. This is the Ehrenfest principle. The change in energy balances the work done
and so
∂Ei
δWi = δEi = δV . (1.4.1)
∂V
For a system in TE the work done in an adiabatic change is the ensemble average of
this result
X ∂Ei X ∂Ei
−P δV = δW = ρi δV =⇒ P = − ρi . (1.4.2)
i
∂V i
∂V

This is reasonable since if P exists then it is a thermodynamic variable and so should

be expressible as an ensemble average.
These ideas are most easily understood in the context of an ideal gas. Consider the
quantum mechanics of a gas of N non-interacting particles in a cube of side L. The
1 STATISTICAL MECHANICS 8

single particle energy levels are E n = h̄2 n2 /2mL2 with n ∈ Z 3 . Note that for given
n, En ∝ V −2/3 – generally, we expect E = E(V ). Then
∂Ei 2 Ei
= −
∂V 3V
2 X 2
=⇒ P = ρi Ei = 3 , (1.4.3)
3V i

where  = E/V is the energy density. This is the formula relating the pressure to the
energy density for a non-relativistic system of particles. We also deduce the energy
conservation equation for an adiabatic change (δQ = 0):

P δV + δE = 0 . (1.4.4)

For the ideal gas an adiabatic change does not change the occupation probabilities
ρi of the energy levels – just the energies of the levels change. This is the Ehrenfest
principle. For systems with Ei (V ) ∝ V −s the ρi are still Boltzmann equilibrium
probabilities in V + δV but with a different temperature – s = 2/3 for the ideal gas.
If the change is not adiabatic then heat energy can enter or leave the system. Since
P
E = i ρi Ei , in the most general case we have
X X
δE = ρi δEi + Ei δρi
i i
or
X X
δE = −P δV + Ei δρi . (note that δρi = 0.) (1.4.5)
i i

To deal with this eventuality we introduce the concept of entropy. Define the entropy,
S, of a system by
k
S = log W (a) , (1.4.6)
A
where, as before, W (a) is the number of ways of realizing the ensemble configuration
{ai }. Since there are A members in the ensemble this is the entropy per system. Note
that the ensemble is not necessarily an equilibrium one. However, with this definition
we see that for an equilibrium ensemble
k
Seq = log Wmax , (1.4.7)
A
and from (1.4.6) and (1.4.7) we deduce that S is maximized for the an equilibrium
ensemble. Using (1.2.2) we get
!
k X
S = A log A − ai log ai
A i
!
k X X
= A log A − ai log ρi − log A ai ,
A i i
=⇒
P
S = −k i ρi log ρi (1.4.8)

P
where we have used ρi = ai /A , i ai = A. This is an important formula which can
be derived in many different ways from various reasonable requirements. However, for
1 STATISTICAL MECHANICS 9

complex systems it needs some interpretation but for the free gas there is no difficulty.
More on entropy shortly. Let us apply the idea to a general change between nearby
states of TE, e.g., an isothermal change (dT = 0).
Change the volume of the system arbitrarily slowly and allow heat energy to be
absorbed/emitted, e.g., by contact with a heat bath. At every stage the system will
have time to equilibrate and so although the ρ i will be time-dependent they will take
their equilibrium values appropriate to the conditions prevailing (i.e., values of V, T ).
In this case we have (S will stand for the equilibrium value unless otherwise stated)
1
ρi = e−βEi (V ) ,
Z(V, T )
=⇒
X
δS = −k δρi (log ρi + 1) ,
i
X
= −k δρi (−βEi − log Z + 1) ,
i
1X
= Ei δρi
T i
=⇒
X
Ei δρi = T δS . (1.4.9)
i

So, for a slow change, using (1.4.5), we get in general

δE = T δS − P δV ,

or, in exact differentials

dE = T dS − P dV . (1.4.10)
We can think of this as a comparison between two infinitesimally different states of
thermodynamic equilibrium. This equation is an expression of the First Law of
Thermodynamics which is the statement of energy conservation:

The First Law of Thermodynamics states that

dE = δQ + δW . (1.4.11)

Here dE is the differential change in the internal energy of S, δQ is the heat energy
supplied to S, and δW is the work done on S (e.g., δW = −P dV ).
For a slow change where the system is instantaneously always in thermal equilib-
rium the change is said to be reversible. This is because no work done is dissipated
as unrecoverable energy such as through frictional forces or stirring of the system. In
principle, the work done on the system can be recovered by allowing it to do work
on another system, for example, a slowly compressed spring. From above, comparing
(1.4.10) and (1.4.11), for a reversible change we have

reversible change =⇒

δQ = T dS . (1.4.12)
1 STATISTICAL MECHANICS 10

Note that neither δQ or δW are exact derivatives even though dE, dS and dV are. For
an adiabatic change we have dS = δQ/T = 0. So the definition of an adiabatic change
may be taken to be dS = 0.
From (1.4.10) we see that
   
∂E ∂E
T = , P = − . (1.4.13)
∂S V ∂V S

For a system in equilibrium we can derive S ≡ S eq from the partition function. We

have
X
S = −k ρi log ρi
i
X
= −k ρi (−βEi − log Z)
i
1
= E + k log Z . (1.4.14)
T

We define the free energy, F , to be

F = − kT log Z =⇒ F = E − T S . (1.4.15)

In differential form we then have

dF = dE − T dS − SdT ,

or, using the first law in the form (1.4.10) for nearby equilibrium states
   
∂F ∂F
dF = − P dV − SdT , =⇒ P = − , S = − . (1.4.16)
∂V T ∂T V

We shall see that the idea of the free energy, F , is a very important one in thermody-
namics. From (1.4.16) we deduce that F is a function of the independent variables V
and T . Similarly, from (1.4.16),

F = F (V, T ) , =⇒ P = P (V, T ) , S = S(V, T ) . (1.4.17)

Note also that E = E(S, V ) since S, V are the independent variables. Thus the ther-
modynamic state of equilibrium characterized by P, V, T, S can be determined from
the partition function, Z, through the idea of the free energy, F . The equation which
relates P, V and T , in either of the forms shown in (1.4.16) and (1.4.17), is known as
the equation of state. Remember, all these quantities depend implicitly on the fixed
value of N and hence on the particle density n = N/V .

1.5 Extensive and intensive variables

Suppose an equilibrium system is placed next to an exact copy and the partition be-
tween them is withdrawn to make the system twice the size. Variables that double in
this process are called extensive and those whose value is the same as the original are
called intensive. For example, N, V, S, E are extensive and P, T are intensive.
 1 STATISTICAL MECHANICS 11

To elaborate. Consider two equilibrium systems at temperature T , S 1 and S2 ,

which are identical in every way except that they have volumes V 1 and V2 , respectively.
Put them next to each other but do not remove the partition. Then for the whole
system, S12 , we have V12 = V1 + V2 , N12 = N1 + N2 . Since, Z12 = Z1 Z2 , we see that
E = E1 + E2 , S = S1 + S2 etc. If these variables are extensive as claimed then these
equalities will hold after the partition is removed. We expect this to be the case if
S1 , S2 are homogeneous (which means S12 will be, too). The particles in each system
are of the same nature and are not distinguishable and so by removing the partition
we do not mix the systems in a detectable way. Hence, the entropy S 12 should remain
unchanged. Then we expect that V, E, S are proportional to N . Clearly, S 1 and S2
have the same T, P and so T, P are intensive.
This expectation can fail, for example,

(a) if surface energies are not negligible compared with volume energies – the partition
matters too much,

(b) there are long-range forces such as gravity. The systems are no longer homogeneous.
Think of the atmosphere – different parcels of air at different heights may have the
same temperature but the pressure and density are functions of height.

1.6 General Remarks

(i) Let S be the sum of two systems, S = S 1 + S2 , neither of which is necessarily in
equilibrium. Although the partition function is not useful we can still show that the
entropy defined by (1.4.8) is additive. Let the level occupation probabilities for S 1 , S2
be ρi , σi , respectively. Then
X
S = −k ρi σj log ρi σj
ij
X
= −k ρi σj (log ρi + log σj )
ij
X X
= −k ρi log ρi + σj log σj
i j
= S1 + S2 , (1.6.1)
P P
where we have used i ρi = j σj = 1.

(ii) From the definition of entropy (1.4.6) we see that the entropy is a maximum for a
system in equilibrium. Indeed, it was the entropy that we maximized to determine the
equilibrium state. More on this later.

(iii) Of the four variables, P, V, T, S, any two can be treated as independent on which the
others depend. We have naturally found above V, T to be independent, but we can
define G = F + P V which leads to

dG = dF + P dV + V dP , =⇒ dG = V dP − SdT . (1.6.2)

G is called the Gibbs function. Clearly, G = G(P, T ) and now V and S are functions
of P, T . This merely means that the experimenter chooses to control P, T and see how
V, S vary as they are changed. The transformation G = F + P V , which changes the
set of independent variables, is called a Legendre transformation.
1 STATISTICAL MECHANICS 12

1.7 The example of a perfect gas

A perfect gas consists of a system of free particles which therefore do not interact with
each other. The gas is in a cubical box with walls at temperature T , and energy will
be exchanged between the particles and the walls when the particle collide (inelasti-
cally) with the walls. The gas will come into equilibrium with the box and will have
temperature T , also.

1.7.1 the density of states

Let the number of particles be N , the box be of side L with volume V = L 3 , and
the mass of each particle be m. We use periodic boundary conditions to impose the
condition of finite volume so that the single particle wavefunction in the box satisfies

ψ(x + Lei ) = ψ(x) , (1.7.1.1)

where the ei , i = 1, 2, 3 are unit vectors spanning R 3 . For free non-relativistic particles
we then have energy eigenstates
1 2π
ψk (x) = √ eik·x with k = n , n ∈ Z3 , (1.7.1.2)
V L

h̄2 k2
p = h̄k , (p) = . (1.7.1.3)
2m
We are interested in the limit L −→ ∞ and so for fixed momentum, p, we have that
|n| −→ ∞ , |n|/L fixed. Since, |n| is large we can treat n as a continuous variable.
The number of states in n −→ n + dn is d3 n which is therefore the number of states
in k −→ k + dk with dk = (2π/L)dn. Thus,
 3
L
The number of states in k −→ k + dk is d3 k . (1.7.1.4)
2π
Because the state label can be treated as continuous we have
 3 Z
d3 k d3 p
Z Z
X L
−→ d3 k = V = V
i
2π (2π)3 (2πh̄)3
Z Z
V V dp
= 4πp2 dp = p2 d
(2πh̄)3 2π 2 h̄3 d
Z
≡ g()d

V 2 dp
where g() = 2 3 p d . (1.7.1.5)
2π h̄
For particles with spin there is an extra factor of g s = 2s + 1. Then, using (1.7.1.3),
we have
V m
g() = gs 3 (2m)
√ ,
2
2π h̄ 2m
=⇒
 3
V 2m 2 1
g() = gs 2 2 . (1.7.1.6)
4π h̄2
1 STATISTICAL MECHANICS 13

The function g() is the density of states. A general analysis in D-dimensions gives
V SD D−1 dp
g() = gs p , (1.7.1.7)
(2πh̄)D d

where SD is the surface area of the unit sphere in D-dimensions, and V = L D . Again
for non-relativistic particles we can use (1.7.1.3) to get
 1
V 2m 2 1
D=1 g() = gs − 2
2π h̄2
 
V 2m
D=2 g() = gs . (1.7.1.8)
4π h̄2
Note that for D = 2 g() is a constant independent of .
For relativistic particles of rest mass m, the analysis is identical and p
we can use the
general formula (1.7.1.7) with the relativistic dispersion relation (p) = p2 c2 + m2 c4 .
In particular, photons have m = 0 and g s = 2 since there are two distinct polarization
states. In this case, and for D = 3 we get
2
g() = V . (1.7.1.9)
π 2 h̄3 c3
Since  = h̄ω we get the density of states in ω −→ ω + dω, g(ω), to be

d ω2
g(ω) = g() = V 2 3 . (1.7.1.10)
dω π c

1.7.2 the partition function for free spinless particles

Consider N particles in volume V = L3 . Then the single particle partition function is
X Z ∞
z = e−βi −→ d g()e−β . (1.7.2.1)
i 0

From (1.7.1.6), and setting y = β, this gives

 3 Z ∞  3
V 2mkT 2 1 2mπkT 2
z = dy y 2 e−y = V . (1.7.2.2)
4π 2 h̄2 0 h2
In D-dimensions a similar results holds and we find from (1.7.1.7)

z ∝ (kT )D/2 . (1.7.2.3)

Now, from before, we have Z = z N and so log Z = N log z . Then

∂ ∂ D D
E = −N log z = − N (− log β + const.) = N kT . (1.7.2.4)
∂β ∂β 2 2
This is an important result of classical statistical mechanics and is called the equipar-
tition of energy which states that

1
The average energy per particle degree of freedom is 2 kT
1 STATISTICAL MECHANICS 14

Combining this result with P V = 32 E from (1.4.3) we get the equation of state for a
perfect gas.
P V = N kT Boyle’s Law. (1.7.2.5)
Note that this result holds in D-dimensions since the generalization of (1.4.3) is P =
2
D .

1.7.3 entropy and the Gibbs paradox

Using (1.4.15) and (1.4.16), which states that
 
∂F
F = − kT log Z , S = − ,
∂T V

we get
3
S = N k log z + N k
2  
3 2mπkT 3
= N k log V + N k log 2
+ Nk , (1.7.3.1)
2 h 2
which is not extensive since V ∝ N . This is Gibbs paradox. The resolution is to
realize that using Z = z N treats the particles as distinguishable since it corresponds
to summing independently over the states of each particle. In reality the particles
are indistinguishable, and we should not count states as different if they can be
transformed into each other merely by shuffling the particles. The paradox is resolved
by accounting for this fact and taking Z = z N /N !. Using Stirling’s approximation, we
find    
V 3 2mπkT 3
S = N k log + N k log 2
+ Nk , (1.7.3.2)
N 2 h 2
which is extensive.
We should note at this stage that dividing by N ! correctly removes the overcounting
only in the case where the probability that two particles occupy the same level is
negligible. This is the situation where Boltzmann statistics is valid and certainly it
works for T large enough or the number density is low enough. In other cases we must
be more careful and we shall see that the consequences are very important.

1.8 The Increase of Entropy

We saw in section 1.4 from the definition of entropy (1.4.6) and (1.4.7), that S is
maximized for an equilibrium ensemble. This is a general principle which states that

For a system with fixed external conditions the entropy, S, is a

maximum in the state of thermodynamic equilibrium.

By “fixed external conditions” we mean variables like V, E, N for an isolated system or

V, T, N for a system in contact with a heat bath at temperature T .
For example, let S1 be a perfect gas of volume V1 with N1 molecules of species 1,
and S2 be a perfect gas of volume V2 with N2 molecules of species 2. Both systems are
 1 STATISTICAL MECHANICS 15

in equilibrium at temperature T . Place them next to each other with the partition in
place. The total entropy is S1 + S2 with
     
Vi 3 2mi πkT 3
Si = kNi log + kNi log + . (1.8.1)
Ni 2 h2 2
| {z }
Bi (T )

Now remove the partition. Because the species are distinguishable, the new system,
S12 , is not initially in equilibrium but initially has entropy S 1 + S2 . The gases mix
and after equilibrium has been attained the partition function of the new system, of
volume V = V1 + V2 , is
z 1 N1 z 2 N2
Z12 = . (1.8.2)
N1 ! N2 !
Note, not (N1 + N2 )! in the denominator. This gives entropy

S12 =
k [(N1 + N2 ) log(V1 + V2 ) − N1 log N1 − N2 log N2 + N1 B1 (T ) + N2 B2 (T )]
     
V1 + V 2 V1 + V 2
= k N1 log + N2 log + N1 B1 (T ) + N2 B2 (T ) . (1.8.3)
N1 N2
So
    
V1 + V 2 V1 + V 2
S12 − S1 − S2 = k N1 log + N2 log > 0. (1.8.4)
V1 V2
The entropy has increased because the gases have mixed. To separate the two species
requires to system to become more ordered – all the boys to the left and all the girls to
the right. This is unlikely to happen spontaneously. From the definition of the entropy
(1.4.6) the more ordered a system is, the smaller is W (a) and the lower is the
entropy.
Notice that S12 is the sum of the entropies that each gas would have if they each
occupied the volume V1 + V2 separately. This is the law of partial entropies which
holds for ideal gases.

Remarks:

(i) Entropy is a measure of information. When a system is in equilibrium we know the

least about it since all available microstates are equally likely. In contrast, the entropy
is low when we know that the system is in one or very few microstates. Information
theory is based on this observation.
P
(ii) The microscopic definition of entropy S = −k i ρi log ρi , for a complex system is
subtle. In particular, reconciling the microscopic and thermodynamic definitions of
an adiabatic change (δS = 0) is not simple but the perfect gas is an ideal system to
examine these ideas.

(iii) When S1 and S2 separately come into equilibrium we maximize S 1 and S2 separately
each with two constraints (for E1 , N1 and E2 , N2 ), making four in all. When S1 and
S2 are combined there are only two constraints (for E 1 + E2 and N1 + N2 ) which are
a subset of the original four. We expect that the maximum of S 1 + S2 subject to
these two constraints will be greater than the maximum attained when it is subject
 2 THERMODYNAMICS 16

to the four constraints (it may be unchanged c.f. Gibbs paradox). Hence, the entropy
will generally increase when the systems are combined. Mixing just corresponds to
removal of some constraints.

(iv) We have not ever discussed how a system reaches equilibrium. In practice, it takes
a certain time to evolve to equilibrium and some parts of the system may reach equi-
librium faster than others. Indeed, the time constants characterizing the approach to
equilibrium may differ markedly. This can mean that although equilibrium will be
established it might not occur on time scales relevant to observation. For example,
some kinds of naturally occurring sugar molecules are found in one isomeric form or
handedness. In equilibrium, there would be equal of each isomer but this is not relevant
to nature.
The description of the system at equilibrium need not include those forces which are
negligible for most purposes. For example, exactly how a gas molecule interacts with
the vessel walls is a surface effect and does not affect any thermodynamic property
in the infinite volume limit. However, it is generally assumed that it is such small
randomizing interactions which allow energy to be transfered between levels, allowing
the most probable state to be achieved. They are vital to the working of the ergodic
hypothesis. This is by no means certain to hold and in one dimension it is quite hard
to establish that it does.
The point is, that if a closed system is not in equilibrium then, given enough time, the
most probable consequence is an evolution to more probable macrostates and a con-
sequent steady increase in the entropy. Given ergodicity, the probability of transition
to a state of higher entropy is enormously larger than to a state of lower entropy.
Chance fluctuations to a state of lower entropy will never be observed.

The law of increase of entropy states that:

If at some instant the entropy of a closed system does not have its
maximum value, then at subsequent instants the entropy will not
decrease. It will increase or at least remain constant.

2 Thermodynamics
Consider a volume V of gas with fixed number N of particles. From the work on
statistical mechanics we have found a number of concepts which enable us to describe
the thermodynamic properties of systems:

(i) E: the total energy of the system. This arose as the average ensemble energy and
which has negligible fluctuations. It is extensive.

(ii) S: the entropy. This is a new idea which measures the number of microstates con-
tributing to the system macrostate. S is a maximum at equilibrium and is an extensive
property of the system. It has meaning for systems not in equilibrium.

(iii) For nearby equilibrium states dE = T dS − P dV .

 2 THERMODYNAMICS 17

(iv) F : the free energy. It is naturally derived from the partition function and, like E, is
a property of the system. For nearby equilibrium states dF = −SdT − P dV . It is
extensive.

(v) All thermodynamic variables are functions of two suitably chosen independent variables:
F = F (T, V ), E = E(S, T ). Note, if X is extensive we cannot have X = X(P, T ).

(vi) Equations of state express the relationship between thermodynamic variables. For
example, P V = N kT for a perfect gas.

It is because E, S, P, V, T, F etc. are properties of the equilibrium state that we are

able to write infinitesimal changes in their values as exact derivatives. The changes
in their values are independent of the path taken from an initial to final equilibrium
state. This is not true of the heat δQ absorbed or emitted in a change between nearby
equilibrium states – it does depend on the path: δQ is not an exact derivative. However,
dS = δQ/T is an exact derivative, and 1/T is the integrating factor just like in the
theory of first-order o.d.es.
Energy conservation is expressed by the fist law:

δE = δQ + δW for infinitesimal change

∆E = ∆Q + ∆W for finite change (2.1)

where δQ(∆Q) is the heat supplied to the system and δW (∆W ) is the work done on
the system. Suppose the state changes from E(S 1 , V1 ) to E(S2 , V2 ) then in general we
expect Z V2
∆W ≥ − P dV (2.2)
V1

because additional work is done against friction, generating turbulence etc. However,
dE = T dS − P dV which implies
Z S2
∆Q ≤ T dS . (2.3)
S1

2.1 reversible and irreversible changes

These inequalities hold as equalities for reversible changes. This will hold if the
change is non-dissipative and quasistatic. Quasistatic means a change so slow that
the system is always arbitrarily close to a state of equilibrium. The path between initial
and final states can then be plotted in the space of relevant thermodynamic variables.
This is necessary because otherwise thermodynamic variables such as P, T etc. are
not defined on the path and we cannot express δW = −P δV as must be the case for
reversibility. In this case, no energy is dissipated in a way that cannot be recovered by
reversing the process.
For an irreversible change the extra increase in entropy is due to the extra ran-
domizing effects due to friction, turbulence etc. A simple example is rapid piston
movement as well as mixing of distinguishable gases and free expansion of gas into a
vacuum. For an isolated, or closed, system ∆Q = 0 but ∆S ≥ 0, the equality holding
for a reversible change. In real life there is always some dissipation and so even if
δQ = 0 we expect δS > 0.
These ideas are compatible with common sense notions. For reversible transitions
the system can be returned to its original state by manipulating the external conditions.
2 THERMODYNAMICS 18

E.g., a spring that has been quasistatically compressed. Since the change is through a
sequence of equilibrium configurations there is no thermodynamic reason why it could
not be traversed in the opposite direction, thus reversing the change. For an irreversible
change, such as mixing of different gases, no such reverse transition can occur. In the
case of mixing it would require the gases to be separated within a closed system. It
must be emphasized that the entropies of the individual parts of the closed system need
not all increase or remain constant, but the total for the whole system must. Of course,
chemical means can be used to separate mixed gases but only at the cost of increasing
the entropy elsewhere since the closed system must include all parts of the experiment.
If this were not the case then, overall, entropy would have decreased.

2.2 Applications
2.2.1 The Maxwell Relations
The variables we have considered so far, F, G, E, P, V, T, S, are all properties of the
equilibrium state. They each depend on an appropriate pair of variables chosen as
independent and are differentiable. These properties are encoded in differential relations
between the variables. Using dE = T dS − P dV we have

Free energy F = E − TS :
=⇒ dF = −SdT − P dV =⇒ F = F (T, V )

Enthalpy H = E + PV :
(2.2.1.1)
=⇒ dH = T dS + V dP =⇒ H = H(S, P )

Gibbs function G = E − TS + PV :
=⇒ dG = −SdT + V dP =⇒ G = G(T, P )

Because these are all exact differentials we have, for example,

! !
∂2E ∂2E
= . (2.2.1.2)
∂S∂V ∂V ∂S

Since    
∂E ∂E
T = , P = − , (2.2.1.3)
∂S V ∂V S
we have    
∂P ∂T
− = . (2.2.1.4)
∂S V ∂V S
This is a Maxwell relation. Another such Maxwell relation arises from
   
∂F ∂F
S = , P = − , (2.2.1.5)
∂T V ∂V T

we find    
∂S ∂P
= . (2.2.1.6)
∂V T ∂T V
Two more Maxwell relations can be derived from G and H giving four in all. However,
they are not all independent. This can be seen by considering z = z(x, y). Then
   
∂z ∂z
δz = δx + δy =⇒
∂x y ∂y x
2 THERMODYNAMICS 19

   
∂z ∂x
1 = and
∂x y ∂z y
     
∂z ∂z ∂y
0 = + . (2.2.1.7)
∂x y ∂y x ∂x z
Multiply the last equation by (∂x/∂z) y and use the second equation to get
     
∂z ∂y ∂x
= −1 . (2.2.1.8)
∂y x ∂x z ∂z y

Substitute any three of P, V, T, S for x, y, z to relate the terms in Maxwell’s relations.

We note that E, F, H, G are all extensive. Note that G explicitly only depends on
the intensive quantities T, P and so its extensive character is due to N and we can
write G = µ(P, T )N , where µ is called the chemical potential. More on this later.
Consider E = E(V, S) and change variables (V, S) −→ (V, T ). Then we have
       
∂E ∂E ∂E ∂S
= + .
∂V T ∂V S ∂S V ∂V T

Using dE = T dS − P dV this gives

   
∂E ∂S
= −P +T ,
∂V T ∂V T

and using (2.2.1.6) (from differentiability of F ) we have

   
∂E ∂P
= −P +T . (2.2.1.9)
∂V T ∂T V

For a perfect gas the equation of state is P V = N kT which implies

 
∂E
= 0. (2.2.1.10)
∂V T

Since we only need two independent variables we have in this case that E = E(T ) – a
function of T only. This is an example of equipartition where E = 32 N kT .

2.2.2 specific heats

Because δQ is not an exact derivative there is no concept of the amount of heat stored
in a system. Instead δQ is the amount of energy absorbed or emitted by the system
by random processes and excluding work done by mechanical or other means. If we
specify the conditions under which the heat is supplied then we can determine δQ using

δQ = dE + δW , (2.2.2.1)

where dE is an exact derivative. It may be the case that absorption/emission of heat

is not accompanied by a change in T . An example, is the latent heat of evaporation
of water to a gas. However, this is a phase change which we exclude from the current
discussion. We can write δQ = CdT where C is the heat capacity of the system.
Clearly, to give C meaning we need to specify the conditions under which the change
takes place. Important cases are where the change is reversible and at fixed V or fixed
P . Then δQ = T dS and we define
   
∂S ∂S
CV = T , CP = T , (2.2.2.2)
∂T V ∂T P
2 THERMODYNAMICS 20

where CV is the heat capacity at constant volume and C P is heat capacity at

constant pressure. Using
dE = T dS − P dV and dH = T dS + V dP . (2.2.2.3)
we have also    
∂E ∂H
CV = , CP = . (2.2.2.4)
∂T V ∂T P
Changing variables from T, P to T, V we have
       
∂S ∂S ∂S ∂V
= + , (2.2.2.5)
∂T P ∂T V ∂V T ∂T P

and using the Maxwell relation (2.2.1.6) and the definitions of C P and CV we have
   
∂S ∂V
CP = CV + T
∂V T ∂T P
=⇒    
∂V ∂P
CP = CV + T . (2.2.2.6)
∂T P ∂T V

For any physical system we expect

   
∂V ∂P
> 0, > 0, (2.2.2.7)
∂T P ∂T V

and hence that CP > CV . This makes sense because for a given increment dT the
system at fixed P expands and does work P dV on its surroundings whereas the system
at fixed V does no work. The extra energy expended in the former case must be
supplied by δQ which is correspondingly larger than in the latter case.
An important example is that of a perfect gas. An amount of gas with N A (Avo-
gadro’s number) of molecules is called a mole. The ideal gas law tells us that at fixed
T, P a mole of any ideal gas occupies the same volume. The ideal gas law for n moles
is
P V = nNA kT = nRT , (2.2.2.8)
where R is the gas constant, and is the same for all ideal gases. Note that at sufficiently
low density all gases are essentially ideal. The heat capacity per mole, c, is called the
specific heat: c = C/n. Then from (2.2.2.6) we have
cp − c v = R . (2.2.2.9)
We define
cp cv + R
γ = = =⇒ γ > 1 . (2.2.2.10)
cv cv
If the number of degrees of freedom per molecule is denoted N F then equipartition of
energy gives that the energy of 1 mole is
NF
E = RT
2
=⇒    
NF NF NF + 2
cv = R, cp = +1 R , γ = .
2 2 NF
(2.2.2.11)
For a monatomic gas NF = 3, for a diatomic gas NF = 5 (3 position, 2 orientation),
etc, and so γ = 5/3, 7/5, . . ..
2 THERMODYNAMICS 21

2.2.3 adiabatic changes

δQ = 0 for an adiabatic change. For n moles of an ideal gas E = nc v T, P V = nRT
and then
0 = RdE + RP dV (extra R for convenience). (2.2.3.12)
As a function of (V, T ) we have
   
∂E ∂E
dE = dT + dV = ncv dT , (2.2.3.13)
∂T V ∂V T

since (∂E/∂V )T = 0 (eqns. (2.2.1.9),(2.2.1.10)) for a perfect gas. (Note, this will be
true for all systems where E does not depend on V at fixed N .) Then
0 = Rncv dT + RP dV
= cv (P dV + V dP ) + RP dV (using P V = nRT )
= cp P dV + cv V dP . (2.2.3.14)
Thus
cp P dV + cv V dP = 0 . (2.2.3.15)
Thus

P V γ is constant on adiabatics (2.2.3.16)

Note that adiabatics, P V γ constant, are steeper than isothermals, P V constant.

We can use these results to calculate the entropy of n moles of ideal gas. We have
T dS = dE + P dV
dV
= ncv dT + nRT
V
=⇒
dT dV
dS = ncv + nR
T V
=⇒
S = ncv log T + nR log V + S0 .
S0 is the unknown constant of integration. Thermodynamics cannot determine S 0 and
does not care whether S is extensive. This is resolved by the more complete approach of
statistical mechanics given earlier. This result is the same as we found before (1.7.3.2)
if all the constant parts are lumped into S 0 . Using P V = nRT we get
S = ncv log P V γ + S00 , (2.2.3.17)
and so, as expected, S is constant on adiabatics (isentropics).

2.2.4 van der Waals’ equation

In reality even a low density gas is not ideal because there are interactions between
the gas molecules. Ultimately, at temperatures low enough these interactions cause a
phase transition to occur and the gas liquefies. We can account approximately for the
interactions by correcting the ideal gas law to give van der Waals’ equation. The form
of the inter-particle potential is shown in figure 1. There is a repulsive, hard-core, part
at small separations and an attractive part for intermediate separations. The effect of
these two regions is qualitatively modeled as follows:
 2 THERMODYNAMICS 22

Figure 1:

(i) The hard-core is represented as an excluded volume since the volume available is re-
duced. We replace V −→ V − b.

(ii) The pressure measured by an observer, P , is due to the particles on the surface hitting
the walls. Whilst in the bulk the attractive forces roughly cancel out because the
particles are in homogeneous environment, this is not true at the surface since they are
pulled back into the bulk. Consequently, P is less than the bulk pressure, P B . Now, P
is given by P = d(momentum/unit-area)/dt ∝ nmv 2 where n is the number density.
The reduction in K.E. (and hence in v 2 ) due to the unbalanced attractions near the
surface is also proportional to n. Hence the reduction in pressure takes the form
a
PB − P ∝ n2 =⇒ PB − P = , (2.2.4.18)
V2
for fixed N since n = N/V .
The ideal gas equation corrected for the excluded volume applies in the bulk so that

PB (V − b) = N kT =⇒

 
a
P + 2 (V − b) = RT . van der Waals’ equation
V
2
a ∝ N , b ∝ N , R = Nk . (2.2.4.19)

2.3 the Second Law of Thermodynamics

Kelvin-Planck: It is impossible to construct an engine that, operating in a cycle, will
produce no other effect than the extraction of heat from a reservoir and the performance
of an equivalent amount of work.

Clausius: It is impossible to construct a device that, operating in a cycle, will produce

no other effect than the transfer of heat from a cooler to a hotter body.

These two versions can be shown to be equivalent.

By a cycle we mean a series of processes operating on a system for which the initial
and final states are the same. Clearly, the most efficient cycle is a reversible one for
2 THERMODYNAMICS 23

Figure 2:

which all the constituent processes are reversible. Such a cycle can be drawn as a closed
curve in the 2D space of the independent thermodynamic variables (see figure 3).
Consider system shown in figure 2 which, acting in a reversible cycle, extracts heat
Q1 from the hot reservoir at temperature T 1 and delivers heat Q2 to the cold reservoir
at temperature T2 and does work W = Q1 − Q2 . Because the cycle is reversible the
total change in entropy is zero. Thus,
Q2 Q1
∆S = − = 0,
T2 T1
Q1 T2
=⇒ Q2 = . (2.3.1)
T1
The law of increase of entropy tells us that for realistic systems which have frictional
losses etc.
Q1 T2
Q2 ≥ . (2.3.2)
T1
This clearly agrees with the Kelvin-Planck statement of the second law which says that
Q2 > 0. The efficiency of the cycle is defined as
W Q1 − Q 2 T1 − T 2
η = = ≤ . (2.3.3)
Q1 Q1 T1
and we see that the reversible cycle (or engine) has the greatest
R
efficiency.
The work done by the cycle is the areaof the cycle: W = P dV .
This important example of a reversible cycle is the Carnot cycle shown in figure 3
in which a sample S of perfect gas is taken around a closed curve in the P V plane by
a sequence of adiabatic and isothermal reversible processes with T 1 > T2 . On AB the
heat bath supplies heat Q1 to S at temperature T1 . On CD the heat bath extracts
heat Q2 from S at temperature T2 . On the adiabatics, BC and DA, T V γ−1 is constant
so that

T1 VAγ−1 = T2 VDγ−1 , T1 VBγ−1 = T2 VCγ−1 =⇒

VA VD
= . (2.3.4)
VB VC
The direct connection of the second law in the form of Clausius with the law of
increase of entropy can be seen as follows. Suppose there exists an engine I which is
3 THE GRAND CANONICAL ENSEMBLE 24

PV=
RT
1

PV
γ
=K

PV
1

γ
=K
PV=R
T2

2
Figure 3:

Figure 4:

more efficient that a Carnot engine, C, and so violates the law of increase of entropy.
Let I drive C backwards as a Carnot refrigerator in the manner shown in figure 4. Note
that no work is supplied or absorbed by the external surroundings since the output,
W , of I is absorbed by C. Then
W W
ηI > η C =⇒ 0
> =⇒ Q0 < Q . (2.3.5)
Q Q

However, the heat supplied to the hot heat bath is then ∆Q = Q − Q 0 > 0 and this has
been wholly transfered from the cold heat bath without the external input of work.
This violates the Clausius statement of the second law. It also shows that the most
efficient cycle is the Carnot cycle. Remember W and Q are path-dependent even for
reversible paths and hence η also depends on the path used to define the cycle.

3 The Grand Canonical Ensemble

3.1 formalism
We follow an identical procedure to before. We consider a grand (canonical) ensemble,
G of A replicas of the system S in microstates |ii of S with N i particles and energies
3 THE GRAND CANONICAL ENSEMBLE 25

Ei . Suppose ai members of G are in microstate |ii so that

X X X
ai = A , ai Ei = AE , ai Ni = AN . (3.1.1)
i i i

As before we associate thermodynamic equilibrium of S with the most probable config-

uration of G subject to the given constraints. We maximize the same log W as before:
!
δ X X X
log W − α ai − β ai Ei − γ ai Ni = 0. (3.1.2)
δai i i i

This gives us
ai = e−(1+α) e−β(Ei −µNi ) , (3.1.3)
where we have defined the chemical potential µ by βµ = −γ. The grand partition
function is then X
Z = e−β(Ei −µNi ) . (3.1.4)
i

The fraction of members in microstate |ii is

ai e−β(Ei −µNi )
ρi = = . (3.1.5)
A Z
The grand ensemble average is
X
Ô = ρi Oi , (3.1.6)
i

which gives Ê = E, N̂ = N .
What we have done is exchange a fixed value of N for a value that is maintained
in the average by choosing the corresponding value for µ – just as we did for E by
choosing β. This has big advantages since it is often difficult to impose fixed N as a
constraint on a system. It is something that we have not needed to explicitly encode
hitherto. By introducing µ the ability to tune N̂ to the right value using a variable
that is explicitly part of the formalism is a huge advantage, as we shall see.
Then we have
 
1
∂ log Z
N = ,
β∂µ β,V
 
∂ log Z
E − µN = − . (3.1.7)
∂β µ,V

Remember, the V dependence is in Ei ≡ Ei (V ).

As before, we considering adiabatic changes in V (e.g., constant ρ i for the ideal
gas). Then, as before, δE = −P δV and so
 
X δE 1 ∂ log Z
P = − ρi = . (3.1.8)
δV β ∂V β,µ

Following the familiar argument, more general changes δρ i 6= 0, in which heat enters
or leaves the system, obey
X X X
δρi = 0 , δρi Ni = δN , δE = − P δV + δρi Ei . (3.1.9)
i i i
3 THE GRAND CANONICAL ENSEMBLE 26

The entropy is defined in general, as before, to be

k X
S = − log W = − k ρi log ρi . (3.1.10)
A i

For an equilibrium ensemble we then find

X X 1
δS = −k δρi log ρi − ρi δρi
i i
ρi
X
= −k δρi (−β(Ei − µNi ) − log Z + 1) . (3.1.11)
i

This gives the fundamental thermodynamic relation

T dS = dE + P dV − µdN , (3.1.12)

which corresponds to the first law dE = δQ+δW mech +δWchem , where δWchem = µdN
is the work done in adding dN particles to the system. Note, we have conformed to the
convention that an exact derivative is denoted as e.g., dE, whereas an inexact derivative
is denoted as δQ. The chemical potential is then given by
 
∂E
µ = . (3.1.13)
∂N S,V

From the definition the entropy in equilibrium is

S = kβ(E − µN ) + k log Z , (3.1.14)

which can be rewritten as  

∂ kT log Z
S = . (3.1.15)
∂T µ,V
A similar argument to the one given earlier in section 1.3 shows that two systems
placed in thermal and diffusive contact (i.e., they can exchange particles through, say,
a permeable membrane) will reach a thermal equilibrium characterized by a common
value of β, µ. The argument is simply that when the systems are separate there are six
Lagrange multipliers leading to equilibrium characterized by (α 1 , β1 , µ1 ), (α2 , β2 , µ2 ),
but that when they are in contact there are only three Lagrange multipliers and hence
common values for these two sets.
The first law states that dE = T dS − P dV + µdN and so E can be viewed as a
function of (S, V, N ) , and when E, S, V and N are extensive (as is the case for many
large systems) we get

E(λS, λV, λN ) = λE(S, V, N )

     
∂E ∂E ∂E
=⇒ E = S + V + N
∂S V,N ∂V S,N ∂N S,V
=⇒ E = T S − P V + µN . (3.1.16)

We define the grand potential Ω for a state of thermodynamic equilibrium by the

Legendre transformation

Ω = E − T S − µN
= E − µN − (E − µN + kT log Z)
= −kT log Z , =⇒ Z = e−βΩ . (3.1.17)
 3 THE GRAND CANONICAL ENSEMBLE 27

We also have that

dΩ = − SdT − N dµ − P dV , (3.1.18)
which means Ω = Ω(µ, T, V ).
In cases where Ω is extensive we get from (3.1.16) that Ω = − P V . Hence, we
find that in these cases

P V = kT log Z . (3.1.19)

This allows the calculation of the equation of state from the grand partition function.
Note, that as for fluctuations in the energy (1.2.12), we find
 
2 1 ∂N
(∆N ) = ∼ N. (3.1.20)
β ∂µ β

Thus
|∆N | 1
∼ N−2 . (3.1.21)
N
Although, the particle number is no longer fixed to be N , the effect of fluctuations
1
around the most probable ensemble vanish like N − 2 . In all practical circumstances
where N is very large we should therefore expect that the results from the grand canon-
ical and canonical ensembles are indistinguishable. However, we have the advantage
in the grand ensemble that N can be made a function of external variable under the
experimenters control: µ, β. We see also that the equation of state is given explicitly
in terms of the grand partition function.

3.2 Systems of Non-interacting Identical Particles

Suppose the system consists of N identical particles. Then labelling them 1, 2, . . .etc.
wrongly distinguishes them. For example, there is no way of distinguishing

ψ(x1 , α1 , x2 , α2 , t) from ψ(x2 , α2 , x1 , α1 , t) , (3.2.1)

experimentally. They must be the same state with the same energy. They may
differ by a phase since this does not affect probabilities or energies. Somehow we must
construct the states to take into account the indistinguishability of the particles. This
is done in one of the two following ways which have far reaching consequences for most
physical systems:

(i) Fermi-Dirac (FD) statistics states that the wavefunction must be antisymmetric
under interchange of any two particles. In the example above this means 1 ↔ 2 with
a phase factor of −1. Particles which obey Fermi-Dirac statistics are called fermions
and include all spin-1/2 particles and in particular the electron.

(ii) Bose-Einstein (BE) statistics states that the wavefunction for identical particles
is symmetric under interchange of any pair of particles. Particles which obey Bose-
Einstein statistics are called bosons which includes the photon.
3 THE GRAND CANONICAL ENSEMBLE 28

Thus we have:

Fermi-Dirac: ψ(x1 , α1 , x2 , α2 , t) = − ψ(x2 , α2 , x1 , α1 , t) ,

Bose-Einstein: ψ(x1 , α1 , x2 , α2 , t) = + ψ(x2 , α2 , x1 , α1 , t) . (3.2.2)

In the separable case for two particles

h̄2 2 h̄2 2
H = − ∇1 − ∇ + V (x1 ) + V (x2 ) , (3.2.3)
2m 2m 2
and the wavefunction becomes
q
1
Ψ = 2 (ψ1 (x1 , α1 , t)ψ2 (x2 , α2 , t) ∓ ψ1 (x2 , α2 , t)ψ2 (x1 , α1 , t)) , (3.2.4)

with (−) for Fermi-Dirac and (+) for Bose-Einstein.

Consider two electrons and neglect the interaction between them. The Hamiltonian
is then separable. Put one in single particle state ψ 1 (x, s, t) and the other in ψ2 (x, s, t).
The total energy is E = E1 + E2 and the correct wavefunction is
q
1
(ψ1 (x1 , s1 , t)ψ2 (x2 , s2 , t) − ψ1 (x2 , s2 , t)ψ2 (x1 , s1 , t)) .
Ψ(x1 , s1 , x2 , s2 , t) = 2
(3.2.5)
If ψ1 = ψ2 then Ψ = 0. This is the Exclusion Principle which states that
“No two electrons may occupy identical single particle states.”
For a system of N electrons with separable Hamiltonian the energy eigenfunctions take
the form
Ψ(x1 , s1 , . . . , xN , sN , t) = Φ(x1 , s1 , . . . , xN , sN , t) e−iEt/h̄ , (3.2.6)
where Φ can be written most conveniently as

φ1 (x1 , s1 ) φ1 (x2 , s2 ) . . . . . . φ1 (xN , sN )

φ2 (x1 , s1 ) φ2 (x2 , s2 ) . . . . . . φ2 (xN , sN )
r . .
1
Φ = . . , (3.2.7)
N!
. .
. .
φN (x1 , s1 ) φN (x2 , s2 ) . . . . . . φN (xN , sN )
P
N
with E = i=1 Ei . Fermi-Dirac statistics are automatically satisfied.
This is the Slater determinant and eigenfunctions of this kind form the complete
basis set of functions for all atomic physics calculations.
E.g., for N = 3 we have E = E1 + E2 + E3 and
r φ1 (x1 , s1 ) φ1 (x2 , s2 ) φ1 (x3 , s3 )
1
Φ(x1 , s1 , x2 , s2 , x3 , s3 ) = φ2 (x1 , s1 ) φ2 (x2 , s2 ) φ2 (x3 , s3 ) , (3.2.8)
6
φ3 (x1 , s1 ) φ3 (x2 , s2 ) φ3 (x3 , s3 )

In the case that the Hamiltonian does not depend on the spin of the electron the
energy eigenstates which differ just by spin orientation have the same energy and same
spatial wavefunction. In this circumstance we may ignore the presence of spin in many
of our calculations except to remember that any given state may be occupied by at
most two electrons: one with spin “up” (+1/2) and one with spin “down” (−1/2).
3 THE GRAND CANONICAL ENSEMBLE 29

3.2.1 the partition functions

S consists of non-interacting identical particles. Let the one particle states of S be φ r (x)
with energies r . Note that our notation is that r stands for the set of quantum numbers
that uniquely labels the state. Suppose that in microstate |ii, n r of the particles have
the wavefunction φr . Then
X X
Ni = nr , and Ei = nr r . (3.2.1.1)
r r

We see that |ii is completely determined by the set {n r }. We obtain the full set of
microstates by allowing the nr to vary without restriction over their allowed range of
values. Thus we cannot fix Ni or Ei directly – this is why we use the grand ensemble
method. The particular values of N and E for a given system are obtained through
choosing the appropriate values of µ and β. This is a strength of the approach.
We then write the grand partition function as
X
Z = e−β(Ei −µNi )
i
X
= e−β(n1 1 +n2 2 +...−µn1 −µn2 −...)
n1 ,n2 ,...
YX
= e−βnr (r −µ) . (3.2.1.2)
r nr

For fermions we have nr = 0, 1 only. Hence,

Y 
Z = 1 + e−β(r −µ) . (3.2.1.3)
r

Then we find  
1 ∂ log Z X 1
N = = . (3.2.1.4)
β ∂µ β,V r eβ(r −µ) +1
Also,
P P
nr e−β s ns (s −µ)
n1 ,n2 ,...
n̂r ≡ n̄r =
 Z

1 ∂ log Z
= − s 6= r (3.2.1.5)
β ∂r β,µ,s
=⇒
1
n̄r = , Fermi-Dirac distribution. (3.2.1.6)
eβ(r −µ) + 1
P
We see that N = r n̄r where n̄r is the mean occupation number of the r-th single
particle state in thermal equilibrium.
For bosons the sum on the nr is unrestricted: 0 ≤ nr < ∞ and
Y 1
Z = . (3.2.1.7)
r 1 − e−β(r −µ)

The same formula (3.2.1.5) applies to calculate n̄ and we get

1
n̄r = , Bose-Einstein distribution (3.2.1.8)
eβ(r −µ) − 1
3 THE GRAND CANONICAL ENSEMBLE 30

If there are gr states of energy r then the average number of particles with energy  r
is
gr
n̄(r ) = β( −µ) . (3.2.1.9)
e r ±1
For large V we can, as before, pass to the continuum limit for free particles confined
to a box and we have X Z
→ dg() . (3.2.1.10)
r

The general result for g() was given in (1.7.1.7). Here we remember that in D=3
  3 1 1
 V 2m 2 2
 g
 s 4π2 h̄2  ≡ KV  2 , non-relativistic particles
g() =


 2
gs V 2π 2 h̄3 c3
massless particles e.g., photons,
(3.2.1.11)
 3
gs 2m 2
where gs is the spin/polarization degeneracy and K = 4π 2
h̄2
.
The average number of particles in the range  →  + d is

g()d
n()d = . (3.2.1.12)
eβ(−µ) ±1
From the definitions for Z in (3.2.1.3) and (3.2.1.7), we find in the continuum limit
that the grand partition function is given by
Z ∞
log Z = ± d g() log(1 ± e−β(−µ) ) , (3.2.1.13)
0

with + for fermions and − for bosons. We can now use the results of section 3.1 to
derive expressions for N and E. Of course, these expressions can be straightforwardly
derived from the appropriate averages using (3.2.1.12). We find
Z ∞ Z ∞
g()d  g()d
N = β(−µ)
, E = . (3.2.1.14)
0 e ±1 0 eβ(−µ) ±1
The expression for N determines µ in terms of N . Generally this is complicated
but µ must be chosen (as must β) to realize the physical value of N . For bosons (−
sign) we must have µ ≤ 0 otherwise the integrand is singular. For fermions there is no
restriction on µ but we shall see that in many applications µ > 0 is natural.
For non-relativistic particles we get
Z ∞ 1
log Z = ±KV d  2 log(1 ± e−β(−µ) )
0
by parts we find
3/2 d
Z ∞
2
= KV 3 β
0 eβ(r −µ) ± 1
=⇒
2
log Z = 3 βE . (3.2.1.15)

Combining this with the equation of state formula (3.1.19) we get P V = 23 E – re-
member (1.4.3). The 2/3 comes from the  1/2 factor. In D dimensions the exponent is
(D − 2)/2, see (1.7.1.7), and so we find P V = (2/D)E in general.
3 THE GRAND CANONICAL ENSEMBLE 31

3.2.2 the classical limit

When the number density is low and/or the temperature is very high we would not
expect that there would be more than one particle at most in any given quantum state.
In this case the restrictions because of particles being identical should become irrelevant
and we should recover the classical results. Now
Z ∞ Z ∞
g()d  g()d
N = β(−µ)
, E = , (3.2.2.1)
0 e ±1 0 eβ(−µ) ±1
and, putting z = β and y = −βµ, we find N = KV (kT ) 3/2 I1/2 (−βµ) and E =
KV (kT )5/2 I3/2 (−βµ), where

z n dz
Z ∞
In (y) = . (3.2.2.2)
0 ez+y ± 1
When ey  1 we can ignore the 1 in the denominator. We then find that

n() ∼ eβµ e−β g() ∀ , (3.2.2.3)

which is the classical result for allR values of  since µ is just chosen to impose that the
total number of particles is N = n()d. Note that for  such that β( − µ)  1 the
classical approximation will also hold. This is true, for example, for the conduction
electrons in a semi-conductor even though e y is not large.
Under what circumstances is ey large? Suppose this to be the case, then
1  
∼ e−(z+y) 1 ∓ e−(z+y) + e−2(z+y) ∓ . . . . (3.2.2.4)
ez+y ± 1
Substituting in the expression for N above gives
 
Z ∞ Z ∞ 
3/2 βµ
 1/2 −z βµ 1/2 −2z

N = KV (kT ) e 
 dz z e ∓ e dz z e + . . .
 =⇒
| 0 {z } |0 {z } 
√ √ √
π/2 π/4 2
√ βµ
!
π 3 e
N = KV (kT ) 2 eβµ 1 ∓ √ + . . . . (3.2.2.5)
2 2 2
The condition that we require is then
 3
−βµ 2mπkT 2 V
e  1 =⇒  1. (3.2.2.6)
h2 N
This means large T and low density N/V . These conditions hold for most real gases
under normal conditions and most gases liquefy before quantum-mechanical effects
become significant. Of course, in order to liquefy, the gas deviates from ideal behaviour
at low enough temperatures since liquefication is due to attractive intermolecular forces.
The density of states calculation must be modified to account for these forces. An
exception is helium for which these forces are small and so the liquefication temperature
is very low.
By similar means the energy of the gas is then
!
3 eβµ
E ∼ N kT 1 ± √ + ... . (3.2.2.7)
2 4 2
 3 THE GRAND CANONICAL ENSEMBLE 32

We have P V = 32 E and so the lowest order correction to the equation of state is

NP 1
P V = N kT ± 3
√ . (3.2.2.8)
(kT ) K 8π
2

3.3 Black-Body Radiation

Any body with temperature T > 0 emits radiation in the form of photons. For example:
(i) a light bulb filament is heated by the electric current to a high temperature and it
glows;
(ii) iron heated in a forge glows red hot meaning that it emits visible red light. As the iron
gets hotter the colour changes toward the blue end of the spectrum, i.e., there are more
photons of shorter wavelength than before, which means they have higher energy;
(iii) bodies at room temperature typically emit infra-red radiation, e.g., people. This is often
referred to as “heat” but really it is only distinguished by being of longer wavelength
than visible light.
Consider radiation trapped in a box with perfectly reflecting walls. The different energy
states for the photons will each be occupied by different numbers of photons. Suppose,
a body of temperature T is introduced into the box.

This body will emit and absorb photons and so change the occupation numbers
of the levels. The box will thus contain radiation in the form of many photons of
varying frequency which are being continually absorbed and emitted by the body, and
these photons will come into equilibrium with it and so will conform to the Bose-
Einstein distribution. Now, the number of photons is not constrained since they are
continually being emitted from (or absorbed by) the body, and hence there is no need
for the Lagrange multiplier γ in the derivation in section (3.1) above. The effect is to
set µ = 0 for photons.
The body is a black-body if it can absorb and emit radiation of all frequencies
(i.e. energies) across a wide spectral range. If it is not a black-body then photons of
some frequencies cannot be absorbed or emitted by the body and so their number will
be fixed at some non-equilibrium value. The corollary is that if radiation is in contact
with a black-body at temperature T , then photons of all frequencies will conform to
the Bose-Einstein distribution with µ = 0. Thus:
 3 THE GRAND CANONICAL ENSEMBLE 33

Radiation in equilibrium with a black-body at temperature T is characterized only

by the temperature T . Such radiation is called black-body radiation.

3.3.1 The Black-Body Distribution

We use the relativistic formula for the density of states (3.2.1.11) with  = h̄ω. Each
photon has two polarization states, g s = 2, and so the total number of states in (ω, ω +
dω) is:
ω2 V
g(ω)dω = 2 3 dω . (3.3.1.1)
π c
where g(ω) is the density of states.
Thus the number density of photons, N (ω), is given using the Bose-Einstein distri-
bution to be (µ = 0):
1
N (ω) = g(ω) βh̄ω , (3.3.1.2)
e − 1
which, using eqn. (3.3.1.1), becomes

V ω2 1
N (ω) = 2 3 βh̄ω
Black-body radiation distribution. (3.3.1.3)
π c e −1

(i) The energy density, E(ω), is given by

E(ω) = h̄ω N (ω)

V h̄ω 3 1
= 2 3 βh̄ω
(3.3.1.4)
π c e −1
This is the Planck radiation law. For high T or small ω we can approximate

eβh̄ω − 1 ∼ βh̄ω =⇒ E(ω) ∼ g(ω)kT . (3.3.1.5)

This is the classical equipartition

R
of energy – 12 kT per degree of freedom. Since g(ω) ∼
2
ω the total energy, E = E(ω)dω, diverges if we use this result. This is the ultra-
violet catastrophe of the classical approach which treats the photon gas as ideal.

(ii) The total energy is then

Z ∞
E(T ) = E(ω)dω
0
Z ∞
V 1
= 3 3 2 (h̄ω)3 d(h̄ω) . (3.3.1.6)
h̄ c π 0 eβh̄ω −1
Setting x = βh̄ω we get
Z ∞
V 1 1
E(T ) = 4 3 3 2 x3 dx .
β h̄ c π 0 ex −1
Hence, (we look up the integral, it is π 4 /15)
 3
4 π2 k k
E(T ) = CV T , C = . (3.3.1.7)
15 h̄c
 3 THE GRAND CANONICAL ENSEMBLE 34

(iii) The energy flux, E(T ), radiated from a black-body is defined as the energy per second
leaving a small hole of unit area in the wall of the box, assuming no inward flux. The
speed of the photons is c, and the number in interval (ω, ω + dω) crossing unit area per
second is
1 c
df (ω) = N (ω) dω ,
V 4
where the factor 1/4 comes from an angular integration over velocity directions. Thus,

1 E(T )
E(T ) = c,
4 V
or

 3
4 π 2 kc k
E(T ) = σ T , σ = Stefan-Boltzmann law. (3.3.1.8)
60 h̄c

Here σ = 5.67 · 10−8 Js−1 m−2 K −4 is Stefan’s constant.

From (3.1.14) with µ = 0

E
S = k log Z + . (3.3.1.9)
T
From (3.2.1.13) we have
Z ∞  
log Z = − dω g(ω) log 1 − e−βh̄ω . (3.3.1.10)
0

Using (3.3.1.1), g(ω) = BV ω 2 and by parts we find k log Z = E/3T and hence

4E 4V
S = = CT 3 . (3.3.1.11)
3T 3
Then
 4
3S 3 1 1 1
E = ( )3 ∼ V −3 at constant entropy . (3.3.1.12)
4 VC
Note: |k| ∝ L−1 , E = c|k| ⇒ E ∝ V −1/3 .
The free energy is
F = E − T S = − 13 E . (3.3.1.13)
The radiation pressure, P , can be calculated from
 
∂F 1 E 1 4
P = − = 3 = 3 CT . (3.3.1.14)
∂V T V

(Remember F = F (T, V ) and E ∝ V since it is extensive.)

By similar calculation the number density of photons is
N
n = ∝ T3 (see example sheet). (3.3.1.15)
V
3 THE GRAND CANONICAL ENSEMBLE 35

3.4 Debye theory of specific heat

3.4.1 the harmonic oscillator
We begin by discussing a single quantum harmonic oscillator of frequency ω at tem-
perature T . Think of it as being in thermal contact with a heat bath. The partition
function is then
1
X 1
−β(n+ 2 )h̄ω e− 2 βh̄ω 1
z(ω) = e = = . (3.4.1.1)
n 1 − e−βh̄ω 2sinh ( 12 βh̄ω)

Then
∂ log z 1 h̄ω
E = − = 2 h̄ω + . (3.4.1.2)
∂β eβh̄ω −1
The average excitation number n̄ is

E − 21 h̄ω 1
n̄ = = βh̄ω . (3.4.1.3)
h̄ω e −1
This is the Bose-Einstein distribution if we interpret each excitation of h̄ω as putting
another boson into an energy level  = h̄ω. In otherwords, the excitations of the
oscillator act like bosons for all practical purposes. (This is how the quantum field
theory of bose particles works – it starts from the oscillator.)
For h̄ω  kT
E ∼ kT , (3.4.1.4)
p 2
equipartition of energy for two degrees of freedom. [H = 2m + 12 mω 2 x2 has two degrees
of freedom classically – p and x.]
Clearly, for h̄ω  kT we recover classical Boltzmann statistics since the “1” in the
denominator can be ignored.

3.4.2 vibrational specific heat of a solid

We shall consider the contribution to the specific heat of a solid from its elastic or
vibrational modes. Other contributions come from the electrons which, in a metal,
behave like a gas and also from internal atomic degrees of freedom such as rotational
modes.
The long wavelength elastic modes are sound waves of which there are three kinds –
two where the displacement is transverse to the velocity and one where it is longitudinal.
We shall not distinguish them and shall assume that they all have the same velocity c s
which is constant and thus independent of frequency. To a good approximation these
modes are harmonic and each sound wave is viewed as an excitation of the corresponding
quantum oscillator. It is labelled by its frequency and wavevector (ω, k) with ω = c s k
(k = |k|), where the wave solution is identical to that already given in (1.7.1.3):
1 2π
ψk (x) = √ eik·x with k = n , n ∈ Z3 , (3.4.2.1)
V L

p = h̄k , (p) = h̄ω = pcs . (3.4.2.2)

This is the same dispersion relation as for massless particles (e.g., photons) and in the
alternative description of the previous section, each excitation of the oscillator can be
viewed as a boson of energy h̄ω with momentum h̄k. These “vibrational” bosons are
 3 THE GRAND CANONICAL ENSEMBLE 36

called phonons, and the elastic excitations of the solid can be described as a gas of
free phonons.
The density of modes is given by the formula for massless particles with g s = 3
3V ω 2
g(ω) = . (3.4.2.3)
2π 2 c3s
The solid contains N atoms and so there are 3N oscillator degrees of freedom. This
gives a maximum frequency ωmax determined by
!1
6π 2 N
Z ωmax 3
dω g(ω) = 3N =⇒ ωmax = cs . (3.4.2.4)
0 V
Thus ωmax gives an ultraviolet cutoff on the energy: h̄ω max .
Calculations are very similar to those of blackbody radiation. In thermal equilib-
rium at temperature T we have the Planck distribution
Z ωmax
h̄ω g(ω)
E = dω . (3.4.2.5)
0 eβh̄ω − 1
Note: E is relative to the zero–point energy 3N h̄ω/2 which is unobservable and can
be subtracted off.
Define the Debye temperature by kTD = h̄ωmax and set x = h̄ω/kT to get
x3
Z TD /T
3V
E = (kT )4 dx
2π (h̄cs )3
2
0 ex − 1
 
TD
= 3N kT D , (3.4.2.6)
T
where D(z) is the Debye function
x3
Z z
3
D(z) = 3 dx . (3.4.2.7)
z 0 ex − 1
Now
(i) T  TD , =⇒ z  1

3z
D(z) = 1 − + O(z 2 ) . (3.4.2.8)
8
Then
E ∼ 3N kT =⇒ CV ∼ 3N k . (3.4.2.9)
This is the classical regime – note there is no ultraviolet catastrophe since there is an
explicit ultraviolet cutoff ωmax .
(ii) T  TD , =⇒ z  1

π4
D(z) = , z  1. (3.4.2.10)
5z 3
then
 3
3π 4 N kT 4 12π 4 T 2π 2 V k 4 T 3
E ∼ =⇒ CV ∼ Nk = . (3.4.2.11)
5TD3 5 TD 5(h̄cs )3
This is the quantum regime.
As in most cases this theory is very approximate since the vibrational properties of
solids are much more complex in general.
3 THE GRAND CANONICAL ENSEMBLE 37

3.5 Bose-Einstein Condensation

For a free boson gas the chemical potential µ is given implicitly in terms of N and T
via
Z 1
∞ 2
N = KV d
0 eβ(−µ) − 1
Z 1
3 ∞ x2
= KV (kT ) 2 dx
e(x−βµ) − 1 0
√
3 π
= KV (kT ) 2 f (−βµ) , (3.5.1)
2
where ∞ nβµ
X e
f (−βµ) = 3 . (3.5.2)
n=1 n2
[This is derived by writing the integral as
Z ∞ ∞
X
dx x1/2 e−nx enβµ ,
0 1

and using Z √
∞ π
1/2 −nx
dx x e = 3 .
0 2n 2
]
This is a power series expansion in the fugacity κ = e βµ , and it converges for
all βµ < 0 . The maximum allowed value of µ is thus µ = 0, for which case the
denominator in the BE distribution vanishes at  = 0 but the singularity is integrable.
It is easy to see that f has a maximum at βµ = 0 and decreases monotonically as βµ
decreases from 0.
Now suppose the density, N/V , is fixed. As T decreases both sides of the equation
for N , (3.5.1), balance since f can increase by decreasing |βµ| (remember µ < 0).
However, this is accomplished by decreasing |µ| since β is increasing. Thus there must
be some T = TC at which µ = 0 and f has reached its maximum value. A further
decrease in T cannot be accommodated by the equation and we seemingly have a
problem with the approach. From (3.5.1) T C is given by
√
N 3 K π
= K̂(kTC ) ,
2 K̂ = f (0) . (3.5.3)
V 2
Note that f (0) = ζ(3/2) = 2.612.
Why does the equation for N fail for T < T C ? The reason is that we must take
more care in passing to the continuum limit when calculating the density of states. The
1
 2 factor in g() assigns zero weight to the  = 0 state. There is only one state with
 = 0 however large V is, whereas for  > 0 the number of states is proportional to V .
Thus, in the limit V → ∞ this state at  = 0 is neglected in the approximation
X Z
→ d g() . (3.5.4)
r

For fermions this is always good since no more than g s fermions can occupy the  = 0
state. However, for bosons the number that can occupy this state is unrestricted. The
 3 THE GRAND CANONICAL ENSEMBLE 38

resolution of the problem for T < TC is to write N = N0 + NC where there are N0

particles in the  = 0 state and NC in the  > 0 states. The total energy, E, receives
no contribution from the N0 particles with  = 0. For T < TC we have (using (3.5.3))
and µ = 0
NC 3
= K̂(kT ) 2 ,
V
 3 "  3#
NC T 2 N0 T 2
= =⇒ = 1− . (3.5.5)
N TC N TC

This phenomenon is called Bose-Einstein condensation and is a phase transition

between the phase where N0 = 0 and the phase where N0 > 0. Formulae of this
kind are typical a such transitions (c.f. the Curie point separating the ferromagnetic
and paramagnetic phases of iron). BE condensation occurs in nature – for example,
superconductivity and superfluidity are related effects. However, it is only very recently
(search the web for “bose condensation”) that low temperature experiments have been
done with bosonic-type atoms and the effect directly observed.
For T < TC we have
Z 3 Z 3
∞ 2 5 ∞ x2
E = KV d = KV (kT ) 2 dx . (3.5.6)
0 eβ − 1 0 ex − 1
As before, the integral is
√ ∞
3 π X 1
ζ(5/2) using 5 = ζ(5/2) = 1.341 . (3.5.7)
4 n=1 n2

From (3.5.3) we have

2 N
KV = √ , (3.5.8)
πζ(3/2) (kTC ) 32
and so
 3
3ζ(5/2) T 2
E = N kT . (3.5.9)
2ζ(3/2) TC
| {z }
A

Using (3.5.5) we get

E = ANC kT . (3.5.10)
For T < TC :

(i) The heat capacity is  

∂E 5E 3
CV = = ∝ T2 . (3.5.11)
∂T V 2T
The specific heat gets a contribution from the condensed bosons since as T changes N C
changes, and so the energy is affected by particles entering or vacating the  > 0 states.

(ii) The entropy is Z

CV 5E
S = dT = . (3.5.12)
T 3T
(iii) The free energy is
F = E − T S = − 32 E . (3.5.13)
 4 THE DEGENERATE FERMI GAS 39

(iv) The pressure is (remember F = F (T, V ))

 
∂F 2E 2
P = − = =⇒ PV = 3E . (3.5.14)
∂V T 3V

This follows since E ∝ V since it is extensive (see also (3.5.10)).

Finally we note that although CV is continuous at T = TC there is a discontinuity

in its derivative (∂CV /∂T )V . This is again typical of a phase transition. This is because
(∂ 2 µ/∂T 2 ) has a discontinuity T = TC since it turns out that

−µ ∼ (T − TC )2 T = TC + 0 , µ = 0 T < TC . (3.5.15)

We have E = E(V, T, µ) and for given N we have

       
∂E ∂E ∂µ ∂E
CV = = + (3.5.16)
∂T V,N ∂T V,µ ∂T V,N ∂µ V,T

Clearly, (∂CV /∂T )V will contain (∂ 2 µ/∂T 2 ) and so will also be discontinuous at
T = TC .

4 The Degenerate Fermi Gas

For values of T and µ for which the Pauli exclusion principle dominates, the fermi gas
is degenerate. It is in the extreme QM limit and cannot be described by a classical
approach. Examples of systems where this is the case are

Electrons in metals at room temperature;

Electrons in white dwarf stars;
Neutrons in neutron stars;
Nucleons in nuclear matter.

For N electrons in volume V at temperature T :

Z ∞
N = g()F () d ,
Z0 ∞
E =  g()F () d ,
0
1
F () = . (4.1)
eβ(−µ) +1

As discussed e−βµ  1 gives the classical limit (e.g., high T , µ < 0). We expect
degenerate QM behaviour for eβµ  1, i.e., low T , µ > 0. Clearly, low T means T  µ.

4.1 Properties at T = 0
At T = 0 write µ = EF = kTF . Then as T → 0:
(
β(−µ) 0 for  < EF
e −→
∞ for  > EF
(
1 for  < EF
⇒ F () −→ (4.1.1)
0 for  > EF
4 THE DEGENERATE FERMI GAS 40

It can be shown (see later section) that µ = E F + O(T 2 )

Thus at T = 0, F () = θ(EF − ). This is a consequence of the Pauli principle
since at T = 0 the electrons will fill the lowest available one-particle states (in pairs
with spin ↑, ↓) until the Fermi energy E F is reached. EF is the highest energy of an
occupied state at T = 0. Then
Z EF 2 3/2
N = KV 1/2 d = KV EF ,
0 3
 2/3
3N
EF = ,
2KV
Z EF 2 5/2 3
E = KV 3/2 d = KV EF = N EF ∝ V −2/3 . (4.1.2)
0 5 5
The equation of state is then

PV = 2
3E =⇒ P V 5/3 = const . (4.1.3)

Classical theory gives P → 0 as T → 0, but the Pauli principle requires that electrons
have non-zero momenta even at T = 0 and so P is non-zero. With all the factors we
get for the pressure
h̄2  2 2/3 N 5/3
 
P = 3π . (4.1.4)
5m V
So P ∝ 1/m and we see that the pressure at T = 0 is much greater for electrons than
for protons.

4.2 White Dwarf Stars

When the hydrogen fuel of a star has been used up, the gravitational forces are no longer
balanced by the pressure of the plasma and the star collapses. The matter becomes
very dense and cold and the electrons form a degenerate fermi gas. For stars below a
certain mass, known as the Chandrasakar limit, the collapse is halted when the pressure
due to the degenerate electrons balances the gravitational pressure. The nucleons will
also contribute but, as found in the previous section, their effect is negligible compared
with that of the electrons. We can estimate the Chandrasakar limit by minimizing the
total energy of the star with respect to its radius.
Since for relativistic particles P V 4/3 = const. we have P ∝ ρ4/3 . ∗ Using the
equations for hydrodynamic equilibrium we find that the gravitational potential energy
can easily be calculated and is given by

3 GM 2
EG = − , (4.2.1)
2 R
where M = N mp is the mass of the star, N is the number of electrons and protons and
mp is the proton mass.
As the star collapses the value of EF will rise and the temperature T is low enough
that T  TF , justifying the subsequent calculation using the degenerate T = 0 formal-
ism. Then Z EF
N 8π p
= d   2 − m 2 c4 , (4.2.2)
V (2πh̄c)3 mc2
∗
For relativistic particles P V = E/3 and at constant entropy E ∝ V −1/3 . Hence, for adiabatic
changes P V 4/3 = constant. This gives γ = 4/3 .
4 THE DEGENERATE FERMI GAS 41

where the general formula for the density of states (1.7.1.7) has been used and m is the
electron mass. This integral can be done using the substitution
p p
mc2 x =  2 − m 2 c4 =⇒  = mc2 1 + x2 ,

to give
 3
N mc x3F
= 8π . (4.2.3)
V h 3
Thus   1/3
h 9 N 1/3
xF = . (4.2.4)
mc 32π 2 R
Similarly, the energy is
 3 Z xF
Ee mc p
= 8π mc2 dx x2 1 + x2 . (4.2.5)
V h 0

This integral can be

√ done exactly, but since x F  1 we can read off the leading terms
by expanding the 1 + x2 to get
 3 !!
Ee mc2 mc 1 1
= 8π x3F xF + +O . (4.2.6)
V 4 h xF x2F

Thus  
Ee 3 1
= mc2 xF + + ... . (4.2.7)
N 4 xF
Then the total energy per nucleon is
E Ee EG
= +
N N N
a c
= + bR − , (4.2.8)
R R
where  1/3
3hc 9 3
a = N 1/3 , c = GN m2p . (4.2.9)
4 32π 2 2
The a/R term is from the relativistic electrons. The value of b can be read from the
above calculation but we do not need the explicit value in what follows. We note that
E only has a minimum w.r.t R if c ≤ a. In this case the star forms a dense cold body
called a white dwarf. If c > a the collapse cannot be halted and the star becomes so
dense that inverse beta decay occurs, e − + p → νe + n, and the electrons and protons
coalesce into neutrons with the violent release of energy in the form of neutrinos. This
is a supernova. The energy burst causes material to be blown off in a spectacular way
and the star’s brightness increases hugely for some days. What is left is a star which is
a degenerate gas of neutrons — a neutron star. This can be rapidly rotating to form
a pulsar. If the star is too massive the collapse cannot be halted even by the pressure
of the degenerate neutrons and a black hole results.
The limit given above for a white dwarf to be stable is
r  3
2 3 λp
N ≤ , (4.2.10)
π 64π 2 lP
4 THE DEGENERATE FERMI GAS 42

where λp = 1.32 10−15 metres is the Compton wavelength of the proton and l P is the
Planck length defined by
s
Gh̄
lP = = 1.616 10−35 metres . (4.2.11)
c3
Thus
N ≤ 2 1057 = 1.7 Nsun . (4.2.12)
A more general calculation gives ∼ 1.5 N sun .

4.3 Heat Capacity of Fermi Gas at Low T

We want to calculate the energy E for given N as a function of T for low T . The
meaning of “low” will be clear in what follows. We will need to calculate µ(T ) as well.
We use eq. (4.1) and to see how to expand around T = 0 we consider the generic form
for these integrals: Z ∞
f ()d
I(T ) = β(−µ)
. (4.3.1)
0 e +1
Make the transformation β( − µ) = z to get
Z ∞ f (µ + kT z)
I(T ) = kT dz
−βµ ez + 1
Z βµ Z ∞
f (µ − kT z) f (µ + kT z)
= kT dz + kT dz . (4.3.2)
0 e−z + 1 0 ez + 1
Now, in the first integral we can write
1 1
= 1− z
e−z +1 e +1
to give
Z µ Z βµ Z ∞
f (µ − kT z) f (µ + kT z)
I(T ) = f () d − kT dz + kT dz . (4.3.3)
0 0 ez + 1 0 ez + 1
Note that in the first term the variable of integration has been transformed back to
. This term is just the result for T = 0 but with upper limit µ ≡ µ(T ). In the
second term we replace the upper limit by ∞ since we certainly require µ/T  1.
This amounts to neglecting exponentially small terms which behave like exp(−µ/T ).
Note that terms of this kind are not expansible around T = 0. [For interest, we are
deriving a power series expansion in T for I(T ). This series is an asymptotic, not
convergent, series. Given the series, resummation techniques designed to recover the
original function naturally give results ambiguous up to non-expansible functions of this
kind].
We thus find
Z µ Z ∞ f (µ + kT z) − f (µ − kT z)
I(T ) = f () d + kT dz . (4.3.4)
0 0 ez + 1
The numerator of the second integral can now be expanded as a Taylor series in z which
can be integrated term by term. This gives
Z µ Z ∞
2 0 z
I(T ) = f () d + 2(kT ) f (µ) dz
0 0 ez +1
z3
Z ∞
1
+ (kT )4 f 000 (µ) dz + . . . (4.3.5)
3 0 ez + 1
4 THE DEGENERATE FERMI GAS 43

The integrals are given by

 
z x−1
Z ∞ 1
= 1− Γ(x)ζ(x) , x>0. (4.3.6)
0 ez + 1 2x−1
where Γ(x) is the Gamma function and ζ(x) the Riemann zeta function. This is derived
by expanding the denominator in the integrand in (4.3.6) (see section 3.5) and using
standard representations for Γ(x) and ζ(x):
Z ∞ ∞
X 1
Γ(x) = tx−1 e−t dt , ζ(x) = ,
0 n=1
nx
∞  
X (−1)n−1 1
= 1− ζ(x) . (4.3.7)
n=1
nx 2x−1

For integer x > 0 we have Γ(x) = (x − 1)! For the values of x of interest the sums
defining ζ(x) can be done using half-range sine and cosine series. For example,
π2 π4
ζ(2) = , ζ(4) = . (4.3.8)
6 90
(Note, integrals for Bose-Einstein statistics are done in a similar way.)
The outcome is
π2 7π 4
Z µ
I(T ) = f () d + (kT )2 f 0 (µ) + (kT )4 f 000 (µ) + . . . .. (4.3.9)
0 6 360
Then
!
2 3/2 π2
N = KV µ + (kT )2 µ−1/2 + O(T 4 ) , with f (z) = KV z 1/2 ,
3 12
(4.3.10)
!
2 5/2 π2
E(T ) = KV µ + (kT )2 µ1/2 + O(T 4 ) , with f (z) = KV z 3/2 .
5 4
(4.3.11)

We must rewrite (4.3.11) in terms of N using (4.3.10) to express µ as a function of N .

This will give a series in T :

µ(T, N ) = EF + AT 2 + O(T 4 ) . (4.3.12)

The coefficient A can be determined by substitution into (4.3.10). However, since we

want the heat capacity C, we can write
    !
∂E 3/2 ∂µ π 2 2 1/2
C = = KV µ + k Tµ + O(T 2 ) . (4.3.13)
∂T N,V ∂T N,V 2

From (4.3.10) we find

   
∂µ π2 k2 T ∂N
= − + O(T 2 ) , since =0,
∂T N,V 6 µ ∂T N,V

and so the sum of the first two terms gives

π2
C = k2 T KV µ1/2 + O(T 2 ) . (4.3.14)
3
5 REMINDER OF CLASSICAL MECHANICS 44

To O(T ) we can use the zeroth-order approximation µ = E F from (4.3.10). To O(T )

we get
π 2 kT π2 T
C = Nk = Nk , (4.3.15)
2 EF 2 TF
where kTF = EF . The expansion is thus good for T  TF .

5 Reminder of Classical Mechanics

Let a system be described by M generalized coordinates q α , with generalized velocities
q̇α and momenta pα , 1 ≤ α ≤ M . For example, a system of N point particles in D
dimensions where the qα are the particle positions and M = N D. The forces acting on
all parts of the system are derived from a potential which is a function of the q α . The
Lagrangian is L(q, q̇) = T − V where T is the kinetic energy term, T = 12 q̇α Tαβ q̇β
(summation convention applies), and V = V (q α ) is the potential energy. The Tαβ may
depend on the qα but not on the q̇α . It is a non-singular matrix.
Lagrange’s equations are
d ∂L ∂L
− = 0, ∀α . (5.1)
dt ∂ q̇α ∂qα
They imply and are implied by Newton’s equations. E.g.,

if T = 21 mẋi ẋi , V = V (xi ) then

d ∂V
(mẋi ) = − ⇒ mẍ = F = − ∇V .
dt ∂xi
Lagranges’s method can be applied to any choice of coordinates, for example, an-
gular coordinates. In plane polar coordinates x = rr̂, ẋ = ṙr̂ + r θ̇θ̂. Then for the
central force V (x) = −µ/r we have

1 2 1 2 2 µ
L = T −V = 2 mṙ + 2 mr θ̇ + ,
  r
d ∂L ∂L µ
− = mr̈ − mr θ̇ 2 + = 0,
dt ∂ ṙ ∂r r2
 
d ∂L ∂L d  2 
− = mr θ̇ = 0 .
dt ∂ θ̇ ∂θ dt

These are well-known results with conserved angular momentum h = mr 2 θ̇.

Define the generalized momenta by
∂L
pα = = Tαβ q̇β ⇒ q̇α = (T −1 )αβ pβ .
∂ q̇α
Note that if L does not depend explicitly on q γ , for some γ, then Lagrange’s equations
(5.1) imply the equation of motion ṗ γ = 0. This means that pγ is a conserved
momentum. In the case of the central potential above p θ is conserved (we denoted it h
there).
The Hamiltonian H is defined by

H = q̇α pα − L , (5.2)
6 CLASSICAL STATISTICAL PHYSICS 45

and is viewed as a function H(p, q) of the coordinates and momenta by elimination of

the q̇α in favour of the pα . In fact, this is just a Legendre transformation of L to change
the independent variables from (qα , q̇a ) to (pα , qα ). Then
1 −1
H(p, q) = 2 pa (T )αβ pβ + V . (5.3)

Remember Tαβ can depend on the qα as does V . H gives the total energy of the system
since we can write (5.3) as

H = q̇α Tαβ q̇β − L = T + V . (5.4)

Now consider the response of H(p, q) to a small change of the independent variables:

∂H ∂H ∂L ∂L
δH = δqα + δpα = δ q̇α pα + q̇α δpα − δqα − δ q̇α . (5.5)
∂qα ∂pα ∂qα ∂ q̇α
Using the definition of pα the δ q̇α terms cancel and we recover Hamilton’s equations:
∂H ∂L ∂H
= − = − pα , = q̇α . (5.6)
∂qα ∂qα ∂pα
There are 2M such first-order equations which are equivalent to the M second-order
Lagrange’s equations above.
As an example consider a light rod of length 2a with equal masses m fixed at each
of its ends, and which rotates about its fixed centre-of-mass G. Each of the particles
has kinetic energy 21 mẋ2 . Write x = a(sin θ cos φ, sin θ sin φ, cos θ) in spherical polars
since each particle moves on the surface of a sphere of radius a. Then

ẋ2 = a2 (θ̇ 2 + sin 2 θ φ̇2 ) , ⇒ T = 1

2 I(θ̇
2
+ sin 2 θ φ̇2 ) , (5.7)

where I = 2ma2 is the moment of inertia of the system. Then

∂L ∂L
pθ = = I θ̇ , pφ = = Isin 2 θ φ̇ . (5.8)
∂ θ̇ ∂ φ̇
If V = 0 then L = T and so
!
1 p2φ
H = p2θ + . (5.9)
2I sin 2 θ

Also have
1 2
H = L ,
2I
where L = x ∧ p = mx ∧ ẋ is the angular momentum vector. Note that since the
Lagrangian does not depend explicitly on φ the corresponding equation of motion is
ṗφ = 0.

6 Classical Statistical Physics

The classical expression for the partition function of a system of N interacting particles
in D dimensions in TE at temperature T is
Z
Z = e−βH(p,q) dpdq , (6.1)
6 CLASSICAL STATISTICAL PHYSICS 46

where p stands for the set of particle momenta p 1 , p2 , . . . , pN and likewise q stands for
the set of particle positions q 1 , q 2 , . . . , q N . The Hamiltonian for the whole system in-
cluding inter-particle interactions and external potential fields is H(p, q). The measure
is defined to be
N
Y
dpdq = dD pi dD qi . (6.2)
i=1

We recognize the Boltzmann factor with the energy evaluated by the classical Hamil-
tonian. This form can be argued from the quantum formalism as follows.

6.1 The Classical limit

We consider the example of one particle in D = 1. The result is easily generalized. The
partition function is X X
Z = e−βr = hr|e−β Ĥ |ri , (6.1.1)
r r

where Ĥ|ri = r |ri. Clearly, |ri is an energy eigenstate. We can use resolution of the
identity to express Z in the form
X Z  Z 
Z = hr| dq |qihq| e−β Ĥ dq 0 |q 0 ihq 0 | |ri ,
r
Z X
= dqdq 0 hq|e−β Ĥ |q 0 i hq 0 |rihr|qi . (6.1.2)
r

But X
hq 0 |rihr|qi = hq|q 0 i = δ(q − q 0 ) , (6.1.3)
r

and so Z
Z = dq hq|e−β Ĥ |qi . (6.1.4)

[ Comment: both equations (6.1.1) and (6.1.4) are explicit realizations of the
expression  
Z = trace e−β Ĥ . (6.1.5)
which treats the operator as the generalization of a finite-dimensional matrix. E.g., the
operator Ô has matrix elements hr|Ô|r 0 i or hq|Ô|q 0 i . It is well known that the trace
is invariant under a unitary change of basis such as we have here, namely, the change
from the complete set |ri to the complete set |qi . It is the sum over diagonal elements
in all bases. ]

Then use Ĥ(p̂, q̂) = K(p̂) + V (q̂) and write

e−β Ĥ = e−βK(p̂) e−βV (q̂) + O(h̄) . (6.1.6)

Ignore the extra, h̄-dependent, terms to get the classical limit. These arise because p̂
and q̂ do not commute.
Then Z
Z = dq e−βV (q) hq|e−βK(p̂) |qi . (6.1.7)
6 CLASSICAL STATISTICAL PHYSICS 47

But
Z
hq|e−βK(p̂) |qi = hq| dpdp0 |pi hp|e−βK(p̂) |p0 ihp0 |qi
| {z }
e−βK(p) δ(p−p0 )
Z
= dp |hq|pi|2 e−βK(p)

eipq/h̄
Z
1
= dp e−βK(p) , since hq|pi = √ . (6.1.8)
h 2πh̄
Thus Z
1
Z = dpdq e−βH(p,q) , (6.1.9)
h
where H(p, q) = K(p) + V (q) is the classical Hamitonian.
From now on we shall omit the factor of 1/h although it is the correct normalization
for Z which comes from the quantum theory and which a classical derivation can never
give. In particular, it correctly sets the origin of the entropy.
The average of a physical variable then takes the obvious form:
R
dpdq f (p, q) e−βH(p,q)
hf (p, q)i = R , (6.1.10)
dpdq eβH(p,q )

so that as before, for example,

 
∂ log Z
E = hH(p, q)i = − .
∂β
Note, that in general the momentum integrals in the partition function are always
Gaussian integrals which can be done in the standard way.

6.2 The Monatomic Gas

Consider a gas of N non-interacting particles, each of mass m, in volume V in D
dimensions. The Hamiltonian H is separable and is given by
N
X p2i
H = , (6.2.1)
i=1
2m

and Z = z N where z is the single-particle partition function

Z  D/2
D D p2 /2m = V 2πm
z = d pd x e −β
= V (2πmkT )D/2 . (6.2.2)
β
Then
D
E =
N kT , (6.2.3)
2
which demonstrates equipartition of energy.
Let f (v)dv be the number of particles in (v, v + dv). Using symmetry and p = mv
we have the measure dD pdD x = CV v D−1 dv and hence from above
2 /2kT
f (v) = CV v D−1 e−mv . (6.2.4)

This is the Maxwell distribution for velocities.

6 CLASSICAL STATISTICAL PHYSICS 48

P m P
a
G Θ

a
Q m y

Figure 5:

6.3 The Diatomic Molecule

We consider the rigid “dumbbell” model shown in figure 5, where the atoms, each of
mass m, are a fixed distance 2a apart. The internal coordinate degrees of freedom are
then (θ, φ), and the centre-of-mass G has coordinate x. There is no potential energy
and so the Lagrangian is given by
 
2
L = T = 1
2 M ẋ + 1
2I θ̇ 2 + sin 2 θ φ̇2 , (6.3.1)

where M = 2m, and the kinetic energy due to the internal motion was derived in section
5 with the moment of inertia I = 2ma2 . Then get the momenta from pα = ∂L/∂ q̇α
and find
∂L ∂L ∂L
pi = = M ẋi , pθ = = I θ̇ , pφ = = Isin 2 θ φ̇ . (6.3.2)
∂ ẋi ∂ θ̇ ∂ φ̇
Then

H = ẋi pi + θ̇pθ + φ̇pφ − T ,

p2 p2θ p2φ
= + + . (6.3.3)
2M 2I 2Isin 2 θ
The partition function for one molecule is then (omitting 1/h 5 factor)
Z
z = d3 pd3 x dpθ dθ dpφ dφ e−βH . (6.3.4)

We can write z = zt zr where

Z  3/2
3 3 −β p2 /2M 2πM
zt = d pd x e = V ,
β
like the monatomic gas, and
Z

p2θ /2I + p2φ /2I sin 2 θ
zr = dpθ dθ dpφ dφ e−β .

Then have s s
Z π Z 2π
2πI 2πIsin 2 θ
zr = dθ dφ
β 0 β 0
6 CLASSICAL STATISTICAL PHYSICS 49

8π 2 I
Z π Z 2π
2πI
= dθ sin θ dφ = . (6.3.5)
β 0 0 β
Thus
z ∝ (kT )5/2 , Z = zN . (6.3.6)
We then find E = 25 N kT which is equipartition of energy with 5 degrees of freedom
per particle – 3 translational and 2 rotational.

6.4 Equipartition in General

Consider a system described by a set of momenta and coordinates p α , qα 1 ≤ α ≤ r,
with Hamiltonian H(p, q). Take H to be of the form

H(p, q) = pα Aαβ pβ + qα Bαβ qβ , 1 ≤ α, β ≤ r , (6.4.1)

| {z }
V (q )

where A and B may depend on the qα . Let H be a homogeneous function of degree 2 in

the pα and a subset of the qα . Denote these variables by {li } = (p1 , p2 , . . . , pr , q1 , q2 , . . . , qs )
where s ≤ r, 1 ≤ i ≤ r + s. Then by Euler’s theorem
r+s
X ∂H
2H = li . (6.4.2)
i=1
∂li
√
Thus, under the change of variables t i = β li we have
1 1
H(l, qs+1 . . . qr ) = H( √ t, qs+1 . . . qr ) = H(t, qs+1 . . . qr ) . (6.4.3)
β β

The partition function is

Z !
Y
Z = = dli dqs+1 . . . dqr e−βH(l,qs+1 ...qr ) , (6.4.4)
i

and under the change of variables we find

 (r+s)/2 Z !
1 Y
Z = dti dqs+1 . . . dqr e−H(t,qs+1 ...qr ) . (6.4.5)
β i

The integral no longer has any dependence on β and so can be done to give an overall
constant. Then we find
Z ∝ β −(r+s)/2 , (6.4.6)
and so  
∂ log Z 1
E = − = (r + s)kT . (6.4.7)
∂β 2
This is general equipartition of energy for (r + s) “degrees of freedom”. For example,
the harmonic oscillator in D = 3 has
1 2 1
H = p + mω 2 q 2 =⇒ E = 3N kT , (6.4.8)
2m 2
which has 6 degrees of freedom.
6 CLASSICAL STATISTICAL PHYSICS 50

6.5 Heat Capacities of Gases

Monatomic Molecules. The classical treatment of H = p 2 /2m gives E = 23 N kT ⇒
CV = 23 N k, which agrees with experiment.

Diatomic Molecules. We found E = 25 N kT ⇒ CV = 25 N k. However, this only

applies at high enough temperature. Find E = 23 N kT when T is low: the rotational
degrees of freedom are ‘cut out’. This is explained by the full QM analysis. We have
Z = z N , z = zt zr . Look at zr using
!
1 p2φ L2
Hr = p2θ + = , (6.5.1)
2I sin 2 θ 2I

where L = x ∧ p is the angular momentum which we must now treat as quantized. Use
eigenstates |l, mi:

h̄2
Hr |l, mi = l(l + 1)|l, mi , l = 0, 1, 2, . . . −l ≤m≤l . (6.5.2)
2I
Then
∞
X X h̄2
zr = e−βEr = (2l + 1) e−l(l+1)Tr /T , Tr = , (6.5.3)
r l=0
2Ik

where (2l + 1) is the level degeneracy and T r , typically ∼ 50o K, gives the critical
temperature. For T  Tr have
X Z ∞
−→ dl .
l 0

Then let x = l(l + 1) ⇒ dx = (2l + 1)dl and get

8π 2 I
Z ∞ T 2I
zr = dx e−xTr /T = = 2 kT = kT , (6.5.4)
0 Tr h̄ h2
which agrees with the earlier classical result.
For T  Tr only the l = 0 term contributes since the other exponents are very large
and negative. Then zr = 1 and there is no contribution to the specific heat.
Note that for T very large the vibrational modes along the axis of the dumbbell will
be excited. These are described accurately by the one-dimensional harmonic oscillator
which has two degrees of freedom and so contributes N kT to give E = 27 N kT .

6.6 Paramagnetism
Each molecule of N molecules in a solid is a little magnet fixed at the lattice site and
constrained only to rotate about it. The magnetic dipole has energy −m · B in an
external magnetic field B = (0, 0, B) as shown in figure 6. Then
!
1 p2φ
H = p2θ + − mB cos θ , (6.6.1)
2I sin 2 θ

and so   Z π
2πI βmB cos θ
z = 2π | {z θ} dθ e
sin . (6.6.2)
β 0
from pφ
6 CLASSICAL STATISTICAL PHYSICS 51

B
P
Θ
magnetic moment m

E = -mB cos Θ

Figure 6:

!  
8π 2 I sinh y
= , y = mβB . (6.6.3)
β y
| {z }
zr

The magnetic moment of the solid, M , is defined by (0, 0, M ) = N hmi. Only the
z-component survives and we find
1 ∂ log Z
M = N hm cos θi = N
β ∂B
1 ∂
= N [log sinh (mβB) − log mβB]
β ∂B
 
1 1
= N mβ coth(mβB) −
β B
 
1
= N m coth y − . (6.6.4)
y

For small y (e.g., high T ) we find Curie’s law

y N m2 B
M = Nm = . (6.6.5)
3 3kT

6.7 Weak Interparticle Forces

A classical gas of N monatomic molecules has
X p2 X
r
H = + φ(urs ) , urs = |xr − xs | . (6.7.1)
r 2m r<s

Then
Z = (2πmkT )3N/2 V N W ≡ Z0 W , (6.7.2)
where Z0 = z0N is the partition function of a perfect gas of N spinless particles, and
Z
1 P
φ(urs )
W = dx e−β r<s . (6.7.3)
VN
Suppose the interaction φ is weak with a possible hard core. Then W can be calculated
in perturbation theory. Let

λrs = e−βφ(urs ) − 1 , (6.7.4)

 6 CLASSICAL STATISTICAL PHYSICS 52

and so
Z Z
1 Y 1 X
W = dx (1 + λrs ) = dx (1 + λrs + . . .) . (6.7.5)
VN r<s V N
r<s

The reason for expanding in λrs rather than in βφ(urs ) is that φ might be very large for
small ranges of urs which would render such an expansion invalid. However, keeping
the exponential form has no difficulty in accommodating such behaviour (such as a
hard core). Then
Z
1 N (N − 1)
W = 1 + 2 d3 xa d3 xb λab . (no sum on a, b) (6.7.6)
V 2
By a change of variables u = (xa − xb ), w = (xa + xb )/2 we have
Z Z
3 3
d xa d xb λab = V f (T ) , f (T ) = d3 u [e−βφ(u) − 1] .
=⇒
N2
W = 1 + f (T ) . (6.7.7)
2V
Thus
log Z = . . . + N log V + log W , (6.7.8)
| {z }
N2
≈ 2V
f (T )

and then have equation of state

 
P ∂ log Z N N2
= = − f (T ) ,
kT ∂V T V 2V 2
=⇒  
N
PV = N kT 1 − f (T ) . (6.7.9)
2V
This is the beginning of the virial expansion which is an expansion in the density:
 2 !
N N N
P = kT 1 + B(T ) + C(T ) + . . . . (6.7.10)
V V V

An example is the van der Waals’ equation which arises from the potential reproduced
in figure 7.

(I) φ = ∞, e−βφ = 0 , for u < d. The contribution to f (T ) is

4πd3
· (−1) ≡ − 2b . (6.7.11)
3

(II) For high T and φ attractive and weak for u > d we get the rest of f (T ) to be
Z ∞
≈ 4πu2 du [ −βφ(u) ] ≡ 2aβ . (6.7.12)
d
6 CLASSICAL STATISTICAL PHYSICS 53

φ
I II

u
d

Figure 7:

Here a and b are positive constants. Then f (T ) = −2(b − a/kT ). Putting this all
together we get
    
N N a N N a
PV = N kT 1 + b − + . . . ≈ N kT 1 + b 1 −
V V kT V V kT
   
N a N
=⇒ P 1 + V 1 − b ≈ N kT
V kT V
!
N 2a
=⇒ P + ( V − N b ) ≈ N kT , (6.7.13)
V2

where, in the last manipulation, we have used P V ≈ N kT + O(1/V ) to eliminate T in

favour of P in the first bracket. This is correct to the order in 1/V to which we are
working. This is the van der Waals’ equation of state argued earlier.