APAS 5110. Internal Processes in Gases. Fall 1999.

CLASSICAL STATISTICAL MECHANICS OF IDEAL GASES

Andrew J. S. Hamilton, JILA, University of Colorado, Boulder

1. Statistical Mechanics versus Thermodynamics

Statistical Mechanics is the modern microscopic theoretical foundation for the classical macroscopic
phenomenology of Thermodynamics. I take the point of view that classical Thermodynamics, notably
its axiomatic foundation, the venerable three Laws, is mostly of historical interest, so don’t expect to
read too much about classical Thermodynamics here.

2. Definition of Ideal Gas

An ideal gas is a gas in which the particles interact so weakly that their eigenstates are effectively
those of isolated particles. Still, the particles interact to the extent that they are able to exchange energy
and other conserved properties. The definition of ideal admits the possibility that the particles obey an
exclusion principle.
The basic simplifying feature of ideal gases is that the single-particle states of the system can be
treated as independent subsystems. Probabilities are multiplicative over independent subsystems (that’s
what independent means).
The subvolumes of an ideal gas may also be treated as independent systems, since the states of
particles in one subvolume do not affect the states of particles in any other subvolume.

3. Many-Particle States versus Single-Particle States

In quantum mechanics, the state of a system is specified by its wavefunction. For example, to
the extent that the room you are sitting in constitutes a closed system (which is obviously a lousy
approximation), you could regard the room and everything in it as being in a certain state, having
a certain wavefunction. Most likely, the wavefunction is not a definite eigenstate, but rather some
superposition of eigenstates. In a closed system, it is always possible to find a complete orthonormal set
of eigenstates of the system. The ‘number of states’ of the system is the number of distinct eigenstates, or
if you like the dimensionality of the Hilbert space. This number of states is independent of how you choose
the orthonormal basis of eigenstates, just as the number of dimensions of a vector space is independent
of how you choose the basis.
 Any many-particle wave function can be expressed as a suitably (anti)symmetrized sum of products of
single-particle wave functions. Thus it is possible to reduce the problem of the counting of states for the
entire system to the problem of counting the states of single particles. This is most particularly useful
in ideal gases, where the single-particle wavefunctions are by definition the same as those of isolated
particles.
When I refer to a state of a system, I shall (or should, at least) hereafter always mean the many-
particle state of the entire system under consideration. When I want to refer to the states of single
particles I shall say so explicitly.

4. Thermodynamic Variables and Conserved Quantities

The statistical properties of a gas in thermodynamic equilibrium (see §6 below) are specified com-
pletely by the values of additive conserved quantities, such as energy, and particle number. To each
conserved quantity there corresponds a thermodynamic variable. The jargon is to call the conserved
quantities extensive variables, and the associated thermodynamic variables intensive variables. In an
ideal gas, extensive variables are proportional to volume, while intensive variables are independent of
volume (in a nonideal gas there is in general some nonlinear dependence of extensive variables on volume,
for any finite volume). Examples are:

Conserved Quantity Thermodynamic Variable

Extensive Intensive
Energy Temperature
Number Chemical Potential
Volume Pressure
Momentum Bulk Velocity
Angular Momentum Angular Velocity

5. Thermodynamic Variables and External Parameters

In the previous section, §4, volume was construed as a conserved quantity. Actually, volume is not so
much conserved, as fixed by external means, namely a box, or an infinite potential well. More generally,
the eigenstates of a system may be affected by various other externally imposed parameters, for example
by an externally imposed electric, magnetic, or gravitational field. Just as there is a thermodynamic
variable corresponding to each conserved quantity in the system, so also there is a thermodynamic variable
associated with each externally imposed parameter of the system. In some cases the thermodynamic
variable has a name, for example:

External Parameter Thermodynamic Variable

Volume Pressure
Surface Area Surface Tension
Magnetic Field Magnetic Susceptibility

Note that the table above lists surface area as an external parameter. Actually, surface area is
irrelevant for ideal gases, since their single-particle states are by assumption unaffected by the presence

2
of surfaces, from which it follows that ideal gases have zero surface tension.

6. Definition of Thermodynamic Equilibrium

A closed system, with given values of the total energy, number, and other conserved or externally
imposed quantities, is in thermodynamic equilibrium if all the (many-particle) states of the system
consistent with the specified total energy, number, etc., are occupied with equal probability.
A nonclosed system is in thermodynamic equilibrium if it is part of a closed, possibly infinite in the
limit, system in thermodynamic equilibrium.
Question (6.1): This definition of thermodynamic equilibrium is surely daft, since it is absurd to
believe that a real system is going to cycle through all its possible states with equal probability even over
the age of the universe. Discuss.

7. Microscopic Reversibility and Thermodynamic Equilibrium

The tendency of systems to approach thermodynamic equilibrium is a physical consequence of micro-

scopic reversibility, also known as detailed balance, which states that the rate (transition probability
per unit time) for one state to convert into another state equals the inverse rate. Microscopic reversibility
arises from the symmetry with respect to time reversal of the quantum mechanical matrix elements that
describe the transition between incoming and outgoing states.
For example, consider a system with just two possible states, A and B, which are allowed by conser-
vation laws to transform one into the other

A→B . (7.1)

Microscopic reversibility says that the rate A → B equals the inverse rate A ← B. Therefore, if you
wait long enough, the system will spend half its time in state A, and half its time in state B. But equal
probabilities of being in state A or state B is just the condition of thermodynamic equilibrium.
It is ironic that the inexorably irreversible approach of systems toward thermodynamic equilibrium
is ultimately a consequence of time reversibility.
Question (7.1): Read Pathria pages 4 and 32-36, or similar text. What is the relation between
microscopic reversibility and Gibbs’ phase space approach?

8. In Practice, When is Thermodynamic Equilibrium True?

Consider a process
A→B (8.1)
connecting two states A, B of a system. Then A and B will be in relative thermodynamic equilibrium (A
and B are equally probable) if the process and its inverse are faster than any other processes connecting
A or B to other states of the system.
Examples:
1. Elastic or Coulomb collisions between electrons,

e+e→e+e , (8.2)

3
 (note this process is its own inverse) is usually the fastest process that redistributes energy between
electrons. Elastic collisions are for example much faster than inelastic collisions,

e + X → e + X∗ , (8.3)

which involve collisional excitation or deexcitation of atoms X. As a result, nonrelativistic electrons

in astrophysical plasmas almost always have a thermal, otherwise known as Maxwellian, distribution
of energies. Which is a profoundly useful convenience.

Question (8.1): Relativistic electrons generally do not have a thermal distribution of energies. Why
not? Do the electrons in a cathode ray tube have a thermal distribution? Why?

2. At high enough density in a sufficiently ionized gas, the dominant process for populating and
depopulating the levels of molecules/atoms/ions is collisional excitation and deexcitation by (a
Maxwellian distribution of) electrons

e + X → e + X∗ . (8.4)

Collisional (de)excitation is faster for example than any process involving radiation. As a result,
the levels of molecules/atoms/ions at high density are driven into thermodynamic equilibrium with
the colliding thermal electrons. Similarly, the processes of collisional ionization and its inverse,
three-body recombination,
e + X → e + e + X+ , (8.5)
dominate the ionization and recombination of a plasma at high enough density, being faster for
example than photoionization or radiative recombination. This accounts for the fact that the
particles in the atmospheres of stars are usually close to thermal equilibrium even though the
radiation field may be far from equilibrium.

Question (8.2): In a gas at low density, the dominant process for exciting levels of molecules/atoms/ions
is often collisions, whereas the dominant deexcitation process is radiative decay (emission of a photon).
Do you expect the levels then to be in thermal equilibrium?
Question (8.3): In the low density interstellar hydrogen gas (HII region) surrounding a hot, UV-
emitting, OB star, the main processes which cause the hydrogen to ionize and recombine are the mutually
inverse processes of photoionization and radiative recombination

H + γ ↔ H+ + e .

Are H and H+ likely to be in thermodynamic equilibrium? Explain. [Hints: In this problem there are at
least two things to think about: Is the radiation field in thermodynamic equilibrium? What about the
fact that neutral H has many bound levels, and what about transitions between these levels? Note that
it is reasonable to assume that elastic (Coulomb) collisions between H, H+ , and e establish a Maxwellian
distribution of velocities of these particles at some temperature. You may also assume that it is a good
approximation (though this is not guaranteed true) to ignore collisional processes which excite or ionize
H.]

4
9. Temperature

Suppose you have two systems, each of which contains a very large number of particles, and each of
which is in thermodynamic equilibrium. Now put the two systems together, in such a way that energy
can transfer between them, but conserving the total energy E = E1 + E2 . What happens? You are
probably used to the classical notion that what will happen is that energy will be exchanged between the
two systems until their temperatures are equal. Let’s investigate more closely.
For the purposes of the argument that follows it is necessary to assume that the states of system 1
are independent of the states of system 2. This assumption is true by hypothesis for an ideal gas, and it
is true in a nonideal gas if the two systems are each large enough that the interaction between them is
negligible compared to the interaction of particles within each system. This is (one of the reasons) why
I started out by postulating that each system contains a very large number of particles.
At any rate, according to the general argument of microscopic reversibility, §7, the system will aspire
toward a condition where all the states of the combined system are occupied with equal probability.
Because the states of the two systems are being assumed to be independent, the number of states Ω of
the combined system such that system 1 has energy E1 and system 2 has energy E2 , is just the product
of the number of states, Ω1 and Ω2 , of the two systems
Ω(E1 , E2 ) = Ω1 (E1 )Ω2 (E2 ) . (9.1)
The energy E = E1 + E2 of the system may be distributed between systems 1 and 2 in all possible ways,
from one extreme in which all the energy is in system 1, to the other extreme in which all the energy is
in system 2, to anything in between. However, not all divisions of energy are equally probable. The most
probable division of the energy E between the two systems occurs where the number of states Ω(E1 , E2 )
attains a maximum with respect to variations of the energy E1 :
∂ ln Ω(E1 , E − E1 ) ∂ ln Ω1 (E1 ) ∂ ln Ω2 (E − E1 ) ∂ ln Ω1 (E1 ) ∂ ln Ω2 (E2 )
= + = − =0, (9.2)
∂E1 ∂E1 ∂E1 ∂E1 ∂E2
where the third expression of equation (9.2) follows because dE2 = d(E − E1 ) = −dE1 at constant total
energy E. I have written the derivatives in equation (9.2) as partial derivatives because the variation
with respect to energy is being done with other conserved quantities, such as number and volume, being
held fixed. If the two systems are each large, as is being assumed, then the number of states Ω(E1 , E2 ),
subject to energy conservation E1 + E2 = E, is in practice an extremely highly peaked function of E1 , so
that the probability that E1 deviates even slightly from the most probable value is minute (see question
[21.1]). Thus in thermodynamic equilibrium the division of energy between the two large systems is such
that equation (9.2),
∂ ln Ω1 ∂ ln Ω2
= , (9.3)
∂E1 ∂E2
is satisfied with high precision almost all the time. One can say that equation (9.3), the condition for the
most probable state, is asymptotically exact in the limit of two infinitely large systems. This equality
(9.3) of partial derivatives between two large systems in mutual thermodynamic equilibrium suggests
that the partial derivatives might have something to do with the classical notion of temperature T , and
indeed this is so. It turns out that the correct definition of absolute temperature T is
1 ∂ ln Ω(E)
≡β≡ , (9.4)
kT ∂E
with k Boltzmann’s constant (the existence of k is a historical artifact; if temperature is measured in
units of energy, then k = 1). That equation (9.4) is an appropriate definition of temperature will become
clear below, eq. (14.1), when comparison is made to the classical equations of thermodynamics.

5
10. Other Thermodynamic Variables

The previous section dealt with two large systems that were allowed to exchange energy. More
generally, one can consider two large systems that are allowed to exchange any other conserved quantity,
such as number, momentum, etc. All the arguments of the previous section carry through essentially
unchanged, and one arrives at the conclusion that, to each additive conserved quantity C, there exists a
thermodynamic variable γ defined by
∂ ln Ω(C)
γ≡ , (10.1)
∂C
analogously to equation (9.4). The thermodynamic variables γ have the property that they are equal in
large systems that are in mutual thermodynamic equilibrium. The partial derivative with respect to the
conserved quantity C in equation (10.1) is taken with all other conserved quantities being held constant.
If there are externally imposed parameters C of the system, such as a uniform magnetic field, then
again formula (10.1) defines the thermodynamic variable γ corresponding to the parameter C.
The collection of equations (10.1) for each conserved quantity or external parameter C can be written
as a single equation for the total derivative of ln Ω
X
d ln Ω = γdC (10.2)
C

the summation being taken over all the different conserved quantities or external parameters C of interest.
For example, the thermodynamic variable corresponding to conservation of number N is
µ ∂ ln Ω(N )
− ≡α≡ , (10.3)
kT ∂N
where µ is the chemical potential. If there are several different species of particle that are conserved,
for example electrons, muons, protons, helium nuclei, ..., then there is a different chemical potential µX
µX ∂ ln Ω(NX )
− ≡ αX ≡ (10.4)
kT ∂NX
associated with the conservation of the number NX of each species X.
Volume V may considered as an external parameter, but it can also be regarded as a conserved
quantity; for example one can imagine two systems separated by a sliding piston, such that the two
systems can exchange volume, the total volume being fixed. The thermodynamic variable associated
with conservation of volume V is
P ∂ ln Ω(V )
≡ , (10.5)
kT ∂V
where P is the pressure. The partial derivative in equation (10.5) is evaluated with all other conserved
quantities, such as energy and number, being held constant.
That the definitions (10.3) and (10.5) are appropriate definitions of chemical potential µ and pressure
P will become clear below, eq. (14.1), when comparison is made to the classical equations of thermo-
dynamics. For the moment all that one can say is that they are constants that characterize systems in
thermodynamic equilibrium.
Combining equations (9.4), (10.3), and (10.5) yields the total derivative of ln Ω for a system in
thermodynamic equilibrium in which energy E, number N , and volume V are considered to be the
conserved properties of the system:
1 µ P
d ln Ω = dE − dN + dV . (10.6)
kT kT kT

6
 Question (10.1): Take two distinct number-conserving particle types X and Y in a system in thermo-
dynamic equilibrium. Which of the following is true/false/meaningless:

(a) TX = TY ;

(b) µX = µY ;

(c) PX = PY ?

Explain.

11. Linear Momentum, Bulk Velocity, and Internal Energy

If conservation of total momentum P is included, then equation (10.6) for the total derivative of ln Ω
is modified to
1 µ P
d ln Ω = dE + γ · dP − dN + dV , (11.1)
kT kT kT
where γ is the (for the moment unknown) thermodynamic variable associated with the conserved momen-
tum P . However, it is always possible to transform to a frame of reference in which the total momentum
is zero, P = 0. In the zero momentum, or rest, frame, the system is isotropic (it looks the same in all
directions) so the associated thermodynamic variable γ, which is a vector, must also be zero in the rest
frame (which direction could it possibly point in?). Thus, transformed to the rest frame, P = 0, equation
(11.1) goes over to
1 µ P
d ln Ω = dU − dN + dV , (11.2)
kT kT kT
where U is the energy of the system in the rest frame, commonly known as the internal energy of the
system. Back in the moving frame, moving at say velocity v with respect to the rest frame, the total
energy E is the sum of the internal energy U and the bulk kinetic energy EKE ,

E = U + EKE . (11.3)

According to a result well known from mechanics, changes in the bulk kinetic energy and momentum of
a system are related by dEKE = v · dP , which is true quite generally in the relativistic as well as the
nonrelativistic domain. Thus the differential dU of the internal energy can be written

dU = d(E − EKE ) = dE − v · dP . (11.4)

Inserting this result (11.4) into equation (11.2), and comparing with equation (11.1), one concludes that
the thermodynamic variable γ associated with momentum conservation is
v
γ=− . (11.6)
kT
Since γ is a constant characteristic of the entire system, the velocity v must be a constant of the system.
In other words, a system in thermodynamic equilibrium necessarily moves with a uniform bulk velocity.
In statistical mechanics it is convenient to work with internal energy U rather than the total energy E,
since it is thereby possible to dispense with all appearance of momentum. You should be aware however
that the fact that internal energy U can be treated as a conserved quantity arises from the conservation
of both energy and momentum, not just energy.

7
12. Rotation

In the previous section it was shown that a system in thermodynamic equilibrium, in which the parts
of the system can exchange linear momentum, is necessarily in uniform bulk motion. It’s an obvious guess
then that if the parts of the system can also exchange angular momentum, then the system should be in
uniform bulk rotation as well as uniform bulk motion, and indeed this is the case. However, a rotating
system is not quite so simple because, whereas the rest frame of a system in uniform bulk motion is
still inertial, the corotating rest frame of a system in uniform rotation is noninertial, with an apparent
centrifugal force directed away from the rotation axis. The consequence is that the internal energy of a
rotating system is modified by the inclusion of a centrifugal potential energy.
Consider then a system in thermodynamic equilibrium, in which the conserved quantities of the system
are taken to be energy E, momentum P , and angular momentum J . For simplicity, let number N and
volume V be held constant, so that the dN and dV terms in the equation for d ln Ω can be omitted. Then
the total differential of the logarithm of the number of states looks like
1 v ω
d ln Ω = dE − · dP − · dJ ??? (12.1)
kT kT kT
where ω is some constant whose physical significance has yet to be established. Equation (12.1) is written
with question marks after it because it can’t really be right as it stands. The trouble is that the value
of the angular momentum J depends on where the origin of the coordinate system is chosen, whereas
the number of states Ω should obviously be independent of the origin. It’s only the angular momentum
J that has this problem: the energy E and linear momentum P are independent of the origin of the
coordinate system. Now the angular momentum J can always be decomposed into the intrinsic angular
momentum J 0 about the center of mass of the system, plus the bulk angular momentum R × P , if the
center of mass is at position R with respect to the origin:
J = J0 + R × P . (12.2)
Thus to make equation (12.1) independent of the origin, one needs to use the intrinsic angular momentum
J 0 in place of the angular momentum J . So equation (12.1) should correctly be written
1 v ω
d ln Ω = dE − · dP − · dJ 0
kT kT kT
1 v ω
= dE − · dP − · d(J − R × P )
kT kT kT
1 v−ω×R ω
= dE − · dP − · dJ . (12.3)
kT kT kT
Note that a term −(ω/kT ) · P × dR seems to have been lost in going from the second to the third
expression of equation (12.3). That term is however always zero since the change dR in the center of
mass of the system from one moment to the next is always parallel to its momentum P . Now divide the
system into a number (two or more) of large independent subsystems, labeled a, with centers of mass at
positions Ra . Then
X 1 v a − ω a × Ra ωa
X 
d ln Ω = d ln Ωa = dEa − · dP a − · dJ a . (12.4)
a a kTa kTa kTa
Equating (12.3) and (12.4) for arbitrary variations of the additive quantities dEa , dP a , and dJ a , one
concludes that
kTa = kT , (12.5)
v a − ω a × Ra v−ω×R
= , (12.6)
kTa kT
ωa ω
= , (12.7)
kTa kT

8
for all the subsystems a. (If you don’t like the way I just said that, you can subdivide the system into just
two parts and proceed as in §9.) Equation (12.5) says the temperatures Ta are equal for all subsystems,
and equation (12.7) implies that ω a are equal for all subsystems. Equation (12.6) then implies that the
velocity v a of subsystem a is
va = v + ω × ra , (12.8)
where
r a ≡ Ra − R (12.9)
is the center of mass of subsystem a relative to the center of mass R of the entire system. Equation
(12.8) says that the velocity of subsystem a is the sum of a uniform motion at velocity v and a uniform
rotation at angular velocity ω about the center of mass of the system. All of which goes to show that
the thermodynamic variable associated with conservation of angular momentum is
ω
− , (12.10)
kT
where ω is the bulk angular velocity of the system. Since ω/kT is a constant characteristic of the entire
system, it follows that a system in thermodynamic equilibrium is necessarily in uniform angular rotation.
The above argument leaves unanswered the question, what is the internal energy U of a rotating
system? The internal energy is uniquely defined by two requirements: (1) the internal energy of a
subsystem is its energy evaluated in the rest frame of the subsystem; and (2) internal energy is additive
over independent subsystems. It follows that the internal energy must be defined as the sum over particle
kinetic plus potential energies evaluated in frames that are locally at rest. Which is to say, the internal
energy must be the energy evaluated in a system comoving, at velocity v, and corotating, at angular
velocity ω, with the system as a whole. Suppose that the system is comprised of particles of mass m at
positions r relative to the center of mass, having velocities u0 in an inertial frame, and u with respect
to the comoving, corotating rest frame, so that

u0 = u + v + ω × r . (12.11)

In the comoving, corotating rest frame, one has

X X
mu = 0 , mr × u = 0 , (12.12)

and in addition X
mr = 0 , (12.13)
since the particle positions r are defined relative to the center of mass. In the comoving, corotating rest
frame, the energy is the sum of particle kinetic energies UKE plus the centrifugal potential energy Urot
(see e.g. Landau and Lifshitz, Mechanics, §39) plus any additional energy Uother associated with internal
degrees of freedom of particles (atoms) or with external fields. Let me ignore Uother as being immaterial
to the present argument, writing simply
X mu2 X m(ω × r)2
U = UKE + Urot = − . (12.14)
2 2
The centrifugal energy Urot = − m(ω × r)2 /2 can also be written
P

ω · J0
Urot = − (12.15)
2
which follows from the result that the intrinsic angular momentum J 0 is
X X X
J0 = mr × u0 = mr × (u + v + ω × r) = mr × (ω × r) . (12.16)

9
In the inertial frame, the energy E is
X mu2
0
E =
2
X m(u + v + ω × r)2
=
2
X mu2 X mv 2 X m(ω × r)2 X X X
= + + +v· mu + ω · mr × u + v · ω × mr
2 2 2
X mu2 X mv 2 X m(ω × r)2
= + +
2 2 2
v·P
= UKE + − Urot
2
v·P
= U+ + ω · J0 . (12.17)
2
According to equation (12.17), the internal energy U of a moving rotating system is given in general by

v·P
U =E− − ω · J0 . (12.18)
2
Expressing d ln Ω in terms of internal energy U rather than total energy E, one has, from equations (12.3)
and (12.18),
1 J0 µ P
d ln Ω = dU + · dω − dN + dV , (12.19)
kT kT kT kT
where for completeness the terms in dN and dV , omitted for brevity from equation (12.3), have been
reinserted. Equation (12.19) is the general expression for the total derivative of ln Ω in a system in which
angular momentum J as well as energy E, momentum P , number N , and volume V are considered to
be the conserved properties of the system.
In a rotating system, the centrifugal force is balanced by a pressure gradient. This can be seen as
follows. Equations (12.14) and (12.15) for the centrifugal energy Urot can be used to rewrite equation
(12.19) for d ln Ω in the form (as in equation [12.14], the contribution Uother to the internal energy from
internal degrees of freedom of particles [atoms] or from external fields is here omitted for simplicity)

1 1 X ∂Urot µ P
d ln Ω = dUKE − · dr − dN + dV , (12.20)
kT kT ∂r kT kT
where the summation is over all particles in the system. The quantity
∂Urot
− = m(ω × r) × ω (12.21)
∂r
in equation (12.20) is the centrifugal force on the particle at position r. Now suppose that the system
under consideration is actually part of a much larger supersystem, such that the system is small compared
to the scale over which the centrifugal acceleration (ω × r) × ω varies. For such a small system, equation
(12.20) reduces to

1 N m(ω × r) × ω µ P
d ln Ω = dUKE + · dr − dN + dV , (12.22)
kT kT kT kT
where dr is now the change in the center of mass of the small system. The commutativity of partial
derivatives means that the second derivative ∂ 2 ln Ω/∂r∂V can be written two ways,

∂ 2 ln Ω ∂ ∂ ln Ω ∂P/kT 1 ∂P
 
= = = , (12.23)
∂r∂V ∂r ∂V ∂r kT ∂r

10
or
∂ 2 ln Ω ∂ ∂ ln Ω ∂ N m(ω × r) × ω ρ(ω × r) × ω
   
= = = . (12.24)
∂r∂V ∂V ∂r ∂V kT kT
Note that in going to the final expressions of equations (12.23) and (12.24) the thermodynamic variables
kT and ω are not differentiated, since they are independent of both r and V in a system in thermodynamic
equilibrium. Equating (12.23) and (12.24) shows that

∂P
= ρ(ω × r) × ω , (12.25)
∂r
which is the usual equation of hydrostatic equilibrium, with centrifugal force balanced by a pressure
gradient. The situation is entirely analogous to that posed in question (27.1), where the system un-
der consideration is supposed to be subject to a large scale external gravitational force g. Here, the
gravitational force is replaced by a centrifugal force, with identical consequences.
In the more general case where the system is not small compared to the scale over which the centrifugal
force varies, the pressure P in equations (12.19) or (12.20) is to be interpreted as the pressure at the
boundary of the confining volume V , since the volume V is after all only fixed by its boundary conditions.
Because the system is no longer isotropic (there’s a preferred direction in the system), the pressure P
will not in general be the same at all points on the boundary, so that the P dV effect of a change in the
volume will depend on whereabouts on the boundary the volume is changed.
In the remainder of these notes the bulk rotation of systems will not be explicitly considered further,
since almost invariably it is possible to restrict to the case of systems that are small compared to the
scale over which the centrifugal force varies. The thermodynamics of such small systems, expressed in
the comoving, corotating rest frame of the system, is the same as that of stationary, nonrotating systems,
except for the appearance of a large scale centrifugal force which is balanced by a pressure gradient. The
centrifugal force is equivalent in its effects to a gravitational force.

13. Entropy

If you read the previous four sections, you may have noticed that the logarithm of the number of states,
ln Ω, whose partial derivatives with respect to conserved quantities yield thermodynamic variables, is
playing a rather prominent role. Bowing to the inevitable, one defines entropy S as Boltzmann’s constant
k times ln Ω (again, the presence of the factor k is an unnecessary historical artifact: if temperature is
measured in units of energy, then k = 1):
S ≡ k ln Ω . (13.1)

The concept of entropy, which can be considered as a measure of the degree of disorder of a system, is
both peculiar and fundamental to statistical mechanics. The deceptively definite definition (13.1) tends
to obscure the fact that entropy is an intrinsically statistical thing, unlike tangible things like energy and
number. If you look at it too closely, entropy has a tendency, like the Cheshire cat, to vanish away till
all that’s left is the smile.
Question (13.1): Consider a system whose states are labeled by the numbers 0000000000 to 9999999999
inclusive. What does it mean to say that the system is in thermodynamic equilibrium? Is the state
4356932540 more probable than the state 9999999999? Is the number 4356932540 more random than the
number 9999999999? What is the entropy of the system in thermodynamic equilibrium?
Entropy, as defined by equation (13.1), has a number of basic properties:

1. Entropy is additive over independent subsystems;

11
2. If you put two systems together, their entropy always increases;

3. Entropy is a maximum in thermodynamic equilibrium;

4. Entropy is zero if the number of states accessed by the system is one.

Property 1 and Property 4 follow immediately from the definition (13.1) of entropy. Property 2, the
increase of entropy, is an ultimate consequence of a system’s tendency to try and explore all its accessible
states with equal probability, which is a consequence of microscopic reversibility, §7. If you put two
systems together, then the number of states of the combined system in which conserved properties, such
as energy, are allowed to be exchanged, is always greater than the number of states of the system in which
the energies, etc., of each system are constrained to have certain fixed values. Property 3 is essentially a
corollary of property 2. Note that it is possible for a system to have no maximum entropy, that is, there
are an infinite number of states with finite energy, number, etc. In this case the system can never attain
thermodynamic equilibrium: the system is unstable, perpetually evolving in a way that explores more
and more of the states of the system.
Question (13.2): According to the argument of the above paragraph, it should be possible to lower
the entropy of a system in thermodynamic equilibrium inside a box, say, by inserting an impermeable
partition that divides the box into two parts. The insertion of the partition fixes the energy and number
in the two sides, reducing the number of states compared to the case where energy and number may
fluctuate. Is this the makings of a perpetual motion machine? Explain.
Question (13.3): If entropy is a maximum in thermodynamic equilibrium, doesn’t that mean that
dS = kd ln Ω = 0, rendering equation (10.6) meaningless? Explain.

14. The Two Fundamental Equations of Thermodynamics

Substituting the definition (13.1) of entropy into equation (11.2) yields

T dS = dU − µdN + P dV , (14.1)

which is the familiar first fundamental equation of classical thermodynamics.

Now in an ideal gas the subvolumes of the system are by hypothesis independent subsystems, so
that ln Ω, which is additive over independent subsystems, must be proportional to the volume V of the
system. If all the thermodynamic variables, temperature T , chemical potential µ, pressure P , of the
system are considered fixed, then according to equation (10.1) the additive conserved quantities of the
system must also be proportional to the volume of the system. In other words S ≡ k ln Ω, U , and N are
all proportional to V at fixed T , µ, and P . It then follows from equation (14.1) that, in an ideal gas,

T S = U − µN + P V , (14.2)

which is the second fundamental equation of thermodynamics.

Question (14.1): Could I not, by accident or whim, have chosen a different set of conserved quantities
and arrived at an equation different from (14.2)? For example, if I had forgotten that N is conserved, I
might have written down T S = U + P V without the µN term. Or, there might be some new conservation
law arising from some hitherto unsuspected symmetry of the universe that I don’t know about. Maybe
you should allow for the possibility that surface area might be a good conserved quantity. So is equation
(14.2) always right, sometimes right, or what? Explain.

12
15. Thermodynamics in terms of Other Variables

From the two equations (14.1) and (14.2) can be derived a whole battery of tautologous equations.
Many texts devote an inordinate amount of space defining things like work functions, free energies, and
enthalpies, and figuring out their partial derivatives with respect to this and that. Do not be misled.
Such ponderations contain no more information than is already contained in equations (14.1) and (14.2).
You will see that equations (14.1) and (14.2) contain 7 variables, but only 3 are independent, entropy
S being some function (determined by microphysics) of U , N , and V , and the thermodynamic variables
T , µ, and P being determined in terms of partial derivatives of S with respect to U , N , and V . It is
evident that instead of the variables U , N and V , one might prefer to express everything in terms of
some other triplet of variables, such as T , P , and V . In view of all the different possible permutations it
is not surprising that a large number of equations can be generated. It is pointless to try to remember
all the equations. However, you do need to understand how to make the mathematical transformation of
partial derivatives from one set of variables to another. The calculations of the next section, §16, provide
a practical illustration of how to do this.
Question (15.1): Equation (14.1) specifies the partial derivatives of any one of S, U , N , and V with
respect to the other three. Define the free energy F ≡ U − T S, and write down an equation that
specifies the partial derivatives of any one of F , T , N , V with respect to the other three.
Question (15.2): Define the specific heats CV at constant volume, and CP at constant pressure, by

T ∂S T ∂S
CV ≡ , CP ≡ , (15.1)
∂T N,V ∂T N,P

and define the adiabatic index γ as the ratio of specific heats

CP
γ≡ . (15.2)
CV
Show that
∂ ln P ∂ ln P
=γ . (15.3)
∂ ln V S,N ∂ ln V T,N

(Hint: use the method of Jacobians, eqs. (16.8)-(16.11) of the next section). An ideal Boltzmann gas
satisfies P V = N kT (eq. [36.37]). Show that in an adiabatic expansion (i.e. expansion at constant
entropy S and number N ), the pressure P and volume V of an ideal Boltzmann gas vary according to

P V γ = constant . (15.4)

13
16. Stability of Thermodynamic Equilibrium

Thermodynamic equilibrium is characterized by the property that entropy is a maximum. However,

the equilibrium conditions derived in §§9–12, namely that the temperature and other thermodynamic
variables should be the same for large systems in mutual equilibrium, were based only on the requirement
that entropy be an extremum. For the equilibrium to be stable requires also that the entropy be a
maximum.
It should be remarked here that there are in general other constraints on the thermodynamic variables,
such as the positivity of temperature and pressure in most systems, which do not arise directly from the
requirement that entropy be a maximum, but rather are associated with the specific character of the
states of the system. These matters will be addressed in §33, after the formalism of ensembles and
partition functions has been properly developed.
The discussion below deals specifically with the conditions imposed by the requirement that the ther-
modynamic equilibrium be stable with respect to variations in the internal energy U and the volume V ,
the number of particles N being taken as fixed. This is the most interesting case, and it is prototypical.
Amongst other things, the calculation provides a good example of the manipulation of partial deriva-
tives in thermodynamics. The condition for an extremum of entropy to be a maximum with respect to
variations of energy and volume is that
∂2S ∂2S ∂2S
δ2S = (δU )2
+ 2 δU δV + (δV )2 < 0 (16.1)
∂U 2 ∂U ∂V ∂V 2
for arbitrary small perturbations δU and δV of the energy and volume away from thermodynamic equilib-
rium. The stability condition (16.1), which is quadratic in the variables δU and δV , will be true provided
that the two conditions
∂2S
<0 (16.2)
∂U 2 V
!2
∂2S ∂2S ∂2S
− <0 (16.3)
∂U ∂V ∂U 2 ∂V 2
hold.
Question (16.1): In writing down the stability condition (16.1), I should properly have considered the
stability between two systems allowed to exchange energy and volume, or at least the stability between
the parts of a system. Show that equation (16.1) is still the correct condition.
The first stability condition (16.2) requires that

∂2S ∂1/T 1
= =− <0, (16.4)
∂U 2 V
∂U V T 2C V

where CV is the specific heat at constant volume

T ∂S ∂U
CV ≡ = . (16.5)
∂T V ∂T V

Equation (16.4) implies that the first condition for stability of thermodynamic equilibrium is that the
specific heat at constant volume must be positive

CV > 0 . (16.6)

Question (16.2): Can you think of a physical system where the specific heat is negative? What do
you think might happen in that case?

14
 To reduce the second stability condition (16.3), first note that the partial derivative ∂ 2 S/∂U ∂V can
be written two ways:
∂2S ∂(1/T ) ∂(P/T )
= = , (16.7)
∂U ∂V ∂V U ∂U V
which follows directly from the fundamental equation (14.1). Equation (16.7) is a particular example of
one of Maxwell’s relations between partial derivatives of thermodynamic quantities. Besides applying
the Maxwell relation (16.7), it is convenient to transform from U , V variables to 1/T , V variables. The
transformation from one set of variables to another is most easily accomplished using Jacobians. The
Jacobian is defined as the determinant of the matrix of partial derivatives, which for two variables is

∂(u, v) ∂u/∂x ∂v/∂x ∂u ∂v ∂u ∂v

≡ = − . (16.8)
∂(x, y) ∂u/∂y ∂v/∂y ∂x ∂y ∂y ∂x

The Jacobian has the properties

∂(v, u) ∂(u, v)
= − , (16.9)
∂(x, y) ∂(x, y)
∂(u, y) ∂u
= , (16.10)
∂(x, y) ∂x y
∂(u, v) ∂(u, v)/∂(t, s)
= . (16.11)
∂(x, y) ∂(x, y)/∂(t, s)

It is the last property, (16.11), which makes the Jacobian the handy way of transforming to new variables.
The second stability condition (16.3) reduces as follows:
!2
∂2S ∂2S ∂2S ∂(1/T ) ∂(P/T ) ∂(1/T ) ∂(P/T )
− = −
∂U ∂V ∂U 2 ∂V 2 ∂V U ∂U V ∂U V ∂V U
∂(1/T, P/T )
=
∂(V, U )
∂(1/T, P/T )/∂(1/T, V )
=
∂(V, U )/∂(1/T, V )
∂(P/T )
∂V 1/T
= −
∂U
∂(1/T ) V
1 ∂(P/T )
= <0. (16.12)
T 2 CV ∂V T

Since the specific heat CV is positive, according to the first stability criterion (16.6), the second stability
condition, equation (16.12), reduces to
∂P
<0. (16.13)
T ∂V T
With a few bizarre exceptions (33), temperature T is positive, so that the stability condition (16.13) is
the same as ∂P /∂V |T < 0. Condition (16.13) is a constraint on the equation of state [that is, on the
function P (T, N, V )] of the system: it says that pressure must decrease in an isothermal expansion.
Question (16.3): Show that
(∂P/∂T |V )2
CP − CV = − . (16.14)
∂(P/T )/∂V |T
(Hint: my proof involves one application of one of Maxwell’s relations, and one transformation of variables
by the method of Jacobians). Hence (you may quote eq. [16.12]) or otherwise show that the second

15
stability condition (16.3) is equivalent to the requirement that the adiabatic index γ ≡ CP /CV is greater
than unity
γ>1. (16.15)

17. Some Examples Derivable by Scaling Arguments

The physics of any particular thermodynamic situation is contained in the function S(U, N, V ), or
more generally S(C) as a function of all the conserved properties or external parameters C of interest.
The fundamental equations (14.1) and (14.2) do not specify what that information is. In general, the
functional form of S must be determined from a detailed investigation of the microphysics. However,
there is an interesting class of problems where one can draw conclusions about the functional form of S
by simple scaling arguments, without looking too closely at precise details. The arguments below are not
intended to be terribly rigorous; the idea is to give you a taste of what the thermodynamics of realistic
systems looks like. Along the way we meet a hoary old problem known as the N ! problem.
Consider first the thermodynamics of so-called free particles. You may recall from quantum mechanics
that the number of states dΩfree of a free particle in an interval d3xd3p of phase space is

gspin d3xd3p
dΩfree = , (17.1)
h3
where the gspin factor is the number of different spin states of the particle, and h is Planck’s constant. For
nonrelativistic particles, particle momentum p is proportional to the square root of the particle energy
, p ∝ 1/2 , while for relativistic particles, particle momentum is proportional to particle energy, p ∝ .
Considering that the energy available to each particle is  ≈ U/N , one can conclude from equation (17.1),
just by scaling arguments, that the number of states of each free particle must behave like

Ωfree (each particle) ∝ V (U/N )q , (17.2)

where q = 3/2 for nonrelativistic particles, q = 3 for relativistic particles. If the states of single particles
are completely independent of one another (in particular, they satisfy no exclusion principle; read the
paragraph containing eqs. [28.6]-[28.8] for a clarification of this point) then Ωfree is multiplicative over
single particle states, so that

Sfree (U, N, V ) ≡ k ln Ωfree (U, N, V ) = N k {constant + ln[V (U/N )q ]} ??? (17.3)

You will notice that I have put question marks after equation (17.3). This is because equation (17.3)
contains a piece N ln V that seems to violate the expectation that entropy should be extensive. The
difficulty is attributable to the fact that in the counting of states I have effectively treated the particles
of the system as distinguishable objects. If in fact the particles are indistinguishable, then the number
of states must be reduced by some permutation factor. Scaling arguments imply that, if entropy is to be
a truly extensive quantity, then equation (17.3) must be modified to

Sfree (U, N, V ) = N k {constant + ln[(V /N )(U/N )q ]} . (17.4)

You will note that the modification from (17.3) to (17.4) amounts to a reduction in entropy by kN ln N ≈
k ln(N !), which is about what you might expect for a permutation factor. This is the famous N ! problem.
Pathria calls it the Gibbs paradox, and discusses it on pages 24-27. For clarification and rigorous
resolution of the problem, wait till the formalism of partition functions has been developed. See especially
§§28, 31, and 32.

16
 From equation (17.4) it follows that
1 ∂Sfree qN k
= = , (17.5)
T ∂U N,V U

P ∂Sfree Nk
= = . (17.6)
T ∂V U,N V
I have omitted the expression for −µ/T = ∂S/∂N because it involves an unknown constant from equation
(17.4), which my scaling arguments were insufficient to determine. Equations (17.5) and (17.6) yield the
familiar equations of state for free particles

U = qN kT , (17.7)

P V = N kT , (17.8)
with q = 3/2 for nonrelativistic particles, q = 3 for relativistic particles. In an adiabatic experiment on
a system of free particles, where the energy U and volume V are allowed to vary, but entropy Sfree and
number N are fixed, equation (17.4) implies that V U q must remain constant. In view of the relations
(17.7) and (17.8), the constancy of V U q in an adiabatic experiment implies the familiar result

P V γ = constant , (17.9)

where the adiabatic index γ (see question [15.2]), is

q+1
γ= , (17.10)
q
which is equal to 5/3 for nonrelativistic free particles (q = 3/2), and equal to 4/3 for relativistic free
particles (q = 3).
The above scaling arguments can be extended to the case where the particles have internal degrees
of freedom that can be approximated as collections of classical simple harmonic oscillators (things on
springs). In the days before quantum mechanics the only way people knew how to treat the internal
degrees of freedom of atoms was to suppose that they behaved somehow like simple harmonic oscillators.
This philosophy led for example to the notion of the oscillator strength of an atomic transition, which
is essentially ‘how many electronic harmonic oscillators an atomic transition is equivalent to’. Probably
you will recall that in quantum mechanics a simple harmonic oscillator has discrete uniformly spaced
energy levels. In the classical limit, the spacing of energy levels goes to zero, so that the energy levels
form a continuum. Thus the basic feature of a classical simple harmonic oscillator (sho) is that the
number of states dΩsho in an interval of energy d is a constant

dΩsho ∝ d . (17.11)

It follows from equation (17.11), by scaling arguments, that Ωsho ∝ U/N for a single harmonic oscillator,
or more generally, for a particle whose internal degrees of freedom can be modeled as a system of f
independent harmonic oscillators,

Ωint (each particle) ∝ (U/N )f . (17.12)

Note that the internal states of the particle are here independent of volume, which will be true physically
if the volume available to the particle is much larger than the size of the particle. The combination of the
supposedly simple harmonic internal degrees of freedom of the particles, plus their translational degrees
of freedom, leads to an expression for the total entropy of the N -particle system as
n o
S(U, N, V ) = Sfree + Sint = N k constant + ln[(V /N )(U/N )q+f ] . (17.13)

17
All the results (17.5) to (17.10) follow as before, but with q replaced by q + f . In particular,

U = (q + f )N kT , (17.14)

P V = N kT . (17.15)
The adiabatic index γ is now
q+f +1
γ= . (17.16)
q+f

That each harmonic oscillator makes a contribution kT to the energy U , according to equation (17.14)
(N particles, each comprising f oscillators, contributes f N kT to U ), is in accordance with a theorem
known as the equipartition theorem. The equipartition theorem states that each quadratic term in
the Hamiltonian of the system makes a contribution kT /2 to the energy. A harmonic oscillator has two
quadratic terms, one in the momenta and one in the coordinates, yielding kT per oscillator. The number
of quadratic terms in the Hamiltonian is sometimes called the number of degrees of freedom of the
system.
Question (17.1): A rigid rotator has the property that it has a fixed moment of inertia I (which is in
general a tensor in more than one dimension). Classically, a rigid rotator has angular momentum J = Iω
and energy  = Iω 2 /2 where ω is the angular velocity. Quantum mechanically, the angular momentum is
quantized in units of h̄. Show that in the classical limit of large angular momentum the number of states
of a one-dimensional rigid rotator in an interval d of energy is a power law in energy

dΩrot ∝ df . (17.17)

What is the index f of the power law? What is the corresponding result for an n-dimensional rigid
rotator? What is the adiabatic index of a system of free particles each of which is an n-dimensional rigid
rotator (you may quote eq. [17.16] if you wish)?
Question (17.2): The adiabatic index of air at room temperature and pressure is found to be γ ≈ 1.4.
Assume that the internal degrees of freedom of air molecules at room temperature arise from their
rotational degrees of freedom, and assume that the molecules can be modeled as rigid rotators. Using
your answer to question (17.1) above, what can you deduce about the nature of air molecules? What
would you expect the adiabatic index of air to be:

(a) at temperatures <

∼ 10 K, at which neither rotational nor vibrational levels of air molecules are
excited;

(b) at temperatures >

∼ 1000 K, at which both rotational and vibrational levels of air molecules are
excited?

Assume that each vibrational degree of freedom can be approximated as a simple harmonic oscillator.
Diatomic molecules have 2 rotational and 1 vibrational degree of freedom, while molecules containing
N ≥ 3 atoms have 3 rotational and 3N − 6 vibrational degrees of freedom.

18
18. Ensembles

So far the discussion has been confined to closed systems, or to very large systems in thermal contact.
However, such considerations leave unanswered some essential questions of detail. For example, one might
consider a single hydrogen atom inside a system in thermal equilibrium, say in the atmosphere of a star,
and ask what is the probability that the atom is in each of its possible states? It is no longer possible
to say blithely that the atom will be in its most probable state, because there’s a good chance it won’t
be. It is to deal with such questions that one introduces the notions of canonical and grand canonical
ensembles.
An ensemble is a collection states of a system, in which each state is supposed to occur with some
specified (not necessarily equal) probability. An ensemble is a convenient mathematical fiction that
formalizes the physical idea that a system evolves though many states, and that the thing of interest is
not so much the exact state of the system at any instant, but rather the distribution of probabilities of
finding the system in each of its many possible states. The hypothesis of thermodynamic equilibrium
(defined in §6 above) defines certain special ensembles, which are termed canonical.
(a) Microcanonical Ensemble
A microcanonical ensemble is the ensemble of (many-particle) states of a system having a definite
internal energy U , number of particles N , and volume V , each state being considered to occur with the
same probability.
The microcanonical ensemble is characterized by U , N , V .
(b) Canonical Ensemble
A canonical ensemble is the ensemble of (many-particle) states of a system in contact with a vast
thermal reservoir, such that the system has a definite number of particles N and volume V , but the
energy of the system may be exchanged with that of the reservoir; each state of the entire system plus
reservoir is considered to occur with equal probability.
The canonical ensemble is characterized by β ≡ 1/kT , N , V .
(c) Grand Canonical Ensemble
A grand canonical ensemble is the ensemble of (many-particle) states of a system in contact with
a vast thermal reservoir, such that the system has a definite volume V , but both the number of particles
and the energy of the system may be exchanged with that of the reservoir; each state of the entire system
plus reservoir is considered to occur with equal probability.
The grand canonical ensemble is characterized by β ≡ 1/kT , α ≡ −µ/kT , V .
(d) Other Canonical Ensembles
In general, one can consider systems that are able to exchange various other conserved quantities
with some vast thermal reservoir. For example, one can consider a piston, containing a fixed number
of particles, lying inside a vast reservoir, such that the volume of the piston can change, but the total
volume of the system plus reservoir is fixed. The set of states of the system will constitute a canonical
ensemble provided that each state of the entire system plus reservoir occurs with equal probability, which
is the same as saying that the entire system plus reservoir is in thermal equilibrium.
A Slightly Subtle Point Concerning Nonideal Gases
At this point it’s appropriate to mention a slightly subtle point about the definition of canonical
ensembles, which makes no difference in ideal gases, but it does affect nonideal gases. Although the
system is supposed to be in contact with a vast reservoir, the wavefunctions of the system are considered
to be evaluated subject to the boundary conditions imposed by the system volume V . If you like, the

19
system is connected to the reservoir only by a thin pipe. Thus (grand) canonical ensembles of nonideal
gases can experience, for example, surface effects arising from the finiteness of the volume V .

19. State Occupation Probabilities

The point of introducing the ensembles just defined was to allow us to determine the probability that
a system (e.g. a hydrogen atom in a stellar atmosphere) is in any particular state. So let’s do that.
(a) Microcanonical Ensemble
In a microcanonical ensemble, the probability Pj of the system being in any particular state j is by
definition the same for all states, so is just the reciprocal of the number of states Ω(U, N, V ) of the system:
1
Pj = . (19.1)
Ω(U, N, V )

Kind of trivial, huh? Also kind of uninformative.

(b) Canonical Ensemble
Matters liven up in the canonical ensemble. The question is: what is the probability Pj that the
system in a canonical ensemble is in a particular (many-particle) state j, with energy Uj ? The hypothesis
that all states of the combined system plus reservoir are equally likely implies that the probability Pj
is proportional to the number of states of the system plus reservoir such that the system is in state j.
Now the number of states of the system such the system is in state j is obviously just one. The only
constraint on the reservoir is that its energy should be U − Uj , where U is the total conserved energy of
the entire system plus reservoir. Thus the probability Pj of the system being in state j is proportional
to the number of states Ωres (U − Uj ) of the reservoir having energy U − Uj :

Pj ∝ Ωres (U − Uj ) . (19.2)

Since the reservoir is by hypothesis vast compared to the system, it is to be expected that the system
energy Uj is very small compared to the total energy U of the system plus reservoir. This suggests
that the obvious thing to do with equation (19.2) is to try to expand it as a power series in the ‘small’
quantity Uj . However, this must be done with due care and attention. The thing to note is that the
number of states Ωres is multiplicative over vast, hence independent, subsystems of the vast reservoir.
Hence ln Ωres increases linearly with the huge volume Vres of the reservoir, ln Ωres ∝ Vres . For a reservoir
with a given temperature, equation (9.4), energy also increases linearly with the volume of the reservoir,
Ures ≈ U ∝ Vres . Thus, while the first derivative of ln Ωres with respect to energy is a constant independent
of the reservoir volume, ∂ ln Ωres /∂U = 1/kT , the second derivative decreases inversely with reservoir
volume, ∂ 2 ln Ωres /∂U 2 ∝ 1/Vres , the third derivative decreases as the inverse square of reservoir volume,
∂ 3 ln Ωres /∂U 3 ∝ 1/Vres
2 , and so on. In the limit of infinite reservoir volume, V
res → ∞, the second and
higher derivatives of ln Ωres with respect to energy are all zero, ∂ n ln Ωres /∂U n → 0 as Vres → ∞ for
n ≥ 2. So, the Taylor expansion in Uj of the logarithm of equation (19.2) yields, in the limit of infinite
reservoir volume,
∂ ln Ωres (U ) Uj
ln Pj = ln P0 − Uj = ln P0 − , (19.3)
∂U kT
where the state 0 is the ground state of the system, with zero energy, U0 = 0. According to equation
(19.3), the probability Pj that a system in a canonical ensemble is in state j with energy Uj is proportional
to
Pj ∝ e−Uj /kT , (19.4)
the famous Boltzmann factor.

20
 Question (19.1): Read Pathria’s comments at the top of page 53 on expanding Pj as a Taylor series
in Uj . Is he right, or is he pulling wool? Try expanding Pj as a Taylor series without first taking the
logarithm, and see what happens.
To effect the normalization of the probabilities Pj from equation (19.4), note that the sum of proba-
P
bilities over all states j of the system must be unity, Pj = 1. Hence

e−Uj /kT
Pj = , (19.5)
Z(N )
where Z(N ) is the so-called partition function for N particles, defined by

e−Uj /kT ,
X
Z(T, N, V ) ≡ (19.6)
states j

the summation being over all states j of the system in the canonical ensemble.
(c) Grand Canonical Ensemble
The probability Pj that the system in a grand canonical ensemble is in a state j with energy Uj and
number Nj is derived by essentially the same argument as for the canonical ensemble. The probability
Pj is now proportional to the number of states of the reservoir having energy U − Uj and number N − Nj

Pj ∝ Ωres (U − Uj , N − Nj ) (19.7)

which is the grand canonical analog of canonical equation (19.2). Arguments identical to those following
equation (19.2) in the canonical case lead to the grand canonical analog of equation (19.3):
∂ ln Ωres (U, N ) ∂ ln Ωres (U, N ) Uj µNj
ln Pj = ln P0 − Uj − Nj = ln P0 − + , (19.8)
∂U ∂N kT kT
where the state 0 is the ground state of the system, with zero particles and zero energy, Uj = Nj = 0.
According to equation (19.8), the probability Pj that a system in a grand canonical ensemble is in state
j with energy Uj and number Nj is proportional to

Pj ∝ e(−Uj +µNj )/kT . (19.9)

The normalization of the probabilities Pj in equation (19.9) is again effected by the requirement that the
P
sum of the probabilities Pj over all states of the system must be unity, Pj = 1. Hence the probabilities
Pj are
e(−Uj +µNj )/kT
Pj = , (19.10)
ZG
where ZG is the grand partition function, defined by
X
ZG (T, µ, V ) ≡ e(−Uj +µNj )/kT , (19.11)
states j

the summation being over all states j of the system in the grand canonical ensemble. The relation between
the grand partition function ZG , eq. (19.11), and the N -particle partition functions Z(N ), eq. (19.6), is
easily seen to be
∞
X
ZG (T, µ, V ) = Z(T, N, V )eµN/kT . (19.12)
N =0

(d) Other Canonical Ensembles

The arguments concerning the probabilities in the case of the canonical ensemble generalize without
ado to any other kind of canonical ensemble you might care to contemplate. In general, the probability

21
Pj of finding a system in state j in an ensemble that is able to exchange various conserved quantities C
with a vast reservoir is
e− Σ γCj
Pj = P − Σ γCj
, (19.13)
states j e
P
where the sums γCj are taken over the conserved quantities C exchanged with the reservoir, and γ,
equation (10.1), are the thermodynamic variables corresponding to the respective conserved quantities.
Question (19.2): Someone throws a normal 6-sided die many times. After a large number of throws,
you are told that the mean of the numbers thrown so far is N̄ , which is a number between 1 and 6, not
necessarily 3.5. What is the probability distribution of the numbers 1 to 6 thrown? (The probabilities are
no longer the a priori probabilities 1/6, because you have been told some information about the system.
You should find that the system is characterized by a quantity that has the character of a ‘temperature’,
and you should find an implicit equation for the temperature. You need not solve this implicit equation
explicitly, but you should show that the solution is unique. Draw a graph, with labeled axes, of the
‘temperature’ versus N̄ .)

20. Partition Functions

The partition functions of the (grand) canonical ensembles were introduced in the previous section,
equations (19.6) and (19.11). Partition functions are so important that I can’t let their introduction go
by without giving them a section all of their own, so here it is. All I want to say is, partition functions
play the starring role in going from microphysics to macroscopic thermodynamics.
Question (20.1): Show that the (grand) partition function defined by equation (19.6) (eq. [19.11]) is
multiplicative over independent systems with the same temperature (and chemical potential). Suppose
the states of molecules can be can be decomposed as a combination of free translational motion, rotational,
vibrational, and electronic states, each of which modes is independent of the other (i.e. the wavefunction
is a simple product of free, rotational, vibrational, and electronic states). What is the relation between
the total single particle partition function Z(1) and the partition functions Zfree (1), Zrot (1), Zvib (1), and
Zelec (1) of the individual modes?
Question (20.2): The hydrogen atom has bound state energy levels En = −1/(2n2 ) with n = 1, 2, ...
in atomic units e = h̄ = me , the nth level being 2n2 -fold degenerate. Show that the one-particle partition
function of the bound states of the H-atom is formally divergent. Explain the physical origin of this
divergence, and what might be done to overcome it. Why is it often sufficient, in problems not requiring
high accuracy, to include in the H-atom partition function only the contribution of the ground state?
Question (20.3): Discuss the population of the n = 2 level of hydrogen. At approximately what
temperature does this population reach a maximum? Why are Balmer lines strongest in A stars? When
I say strong, do I mean strong in emission or in absorption?

21. Energy and its Fluctuations

Equations (19.5) and (19.10) specify the entire distribution of probabilities of states of different
energies in a (grand) canonical ensemble. From this information it is possible to deduce the moments
of the energy distribution, in particular the mean energy Ū of the ensemble, and the variance h∆U 2 i of
energy fluctuations.
The mean energy Ū of the (grand) canonical ensemble is found by differentiating the (grand) partition

22
function with respect to −1/kT . For the canonical ensemble,

∂Z(N )
Uj e−Uj /kT =
X X
= Uj Pj Z(N ) = Ū Z(N ) , (21.1)
∂(−1/kT ) V states j with Nj =N states j with Nj =N

while for the grand canonical ensemble, with µ/kT and V held constant,
∂ZG X X
= Uj e(−Uj +µNj )/kT = Uj Pj ZG = Ū ZG . (21.2)
∂(−1/kT ) µ/kT,V states j states j

Hence one obtains an equation for the mean energy Ū of the (grand) canonical ensemble as a derivative
of the (grand) partition function

∂ ln Z(N ) ∂ ln ZG
Ū = = . (21.3)
∂(−1/kT ) V ∂(−1/kT ) µ/kT,V

Equation (21.3) is the first fundamental link from the microphysics of the partition function to macroscopic
thermodynamics.
Higher derivatives of the partition function with respect to −1/kT yield higher moments of the energy.
For example, the second derivative of the logarithm of the partition function with respect to −1/kT yields
the second central moment of energy. For the canonical ensemble,
" #2
∂ 2 ln Z(N ) ∂ 2 Z(N ) ∂Z(N )
= − = U 2 − Ū 2 = h∆U 2 i . (21.4)
∂(−1/kT )2 V
Z(N )∂(−1/kT )2 V
Z(N )∂(−1/kT ) V

The same derivation works for the grand canonical ensemble, if the derivatives are carried out with µ/kT
and V held fixed. Thus the variance of energy is

∂ 2 ln Z(N ) ∂ 2 ln ZG
h∆U 2 i = = . (21.5)
∂(−1/kT )2 V
∂(−1/kT )2 µ/kT,V

The third and fourth derivatives of ln Z with respect to −1/kT give

∂ 3 ln Z(N ) ∂ 3 ln ZG
h∆U 3 i = = , (21.6)
∂(−1/kT )3 V
∂(−1/kT )3 µ/kT,V

∂ 4 ln Z(N ) ∂ 4 ln ZG
h∆U 4 i − 3h∆U 2 i2 = = . (21.7)
∂(−1/kT )4 V
∂(−1/kT )4 µ/kT,V

The moments h∆U 2 i, h∆U 3 i, h∆U 4 i − 3h∆U 2 i2 , and so on, which are successive partial derivatives of
ln Z, are called irreducible moments. They have the property of being additive over independent
subsystems, unlike the raw moments hU n i. The additivity is a consequence of the additivity of ln Z over
independent subsystems.
By now you are beginning to get some idea of the power of the partition function.
Question (21.1): A gaussian, or normal, distribution is defined by the property that the third and
higher irreducible moments are zero. Show that the distribution of fluctuations of energy is asymptotically
gaussian for large systems, in the sense that

Ū  h∆U 2 i1/2  h∆U 3 i1/3  (h∆U 4 i − 3h∆U 2 i2 )1/4  ... . (21.8)

How does the width of the gaussian change with volume in an ideal gas? (Hint: use the fact that irreducible
moments are all proportional to volume, which is true as long as the subvolumes are independent of each

23
other, which is true in particular for an ideal gas. You should find that the irreducible moments, far
from going to zero, all diverge. Clarify then what is meant here by ‘asymptotically gaussian’.) Justify
the sentence beginning ‘If the two systems are each large ...’ just before equation (9.3).
By the way, if you just did the above question, then you have just proved some version of the Central
Limit Theorem. Not bad after all, huh?
Question (21.2): Relate the variance of fluctuations h∆U 2 i of energy in the canonical ensemble to the
specific heat CV at constant volume. Interpret physically your result.

22. Number and its Fluctuations

In the canonical ensemble, the number of particles is just N , a known, set quantity. In the grand
canonical ensemble the number N is variable. Its mean and higher moments can be determined by
successive differentiations of ln ZG with respect to µ/kT , in much the same manner as the moments of
energy were derived in the previous section by successive differentiations of ln Z with respect to −1/kT .
An equation for the mean number N̄ of particles in a grand canonical ensemble is gotten by differen-
tiating the grand partition function with respect to µ/kT , with T and V held constant:

∂ZG X X
= Nj e(−Uj +µNj )/kT = Nj Pj ZG = N̄ ZG . (22.1)
∂(µ/kT ) −1/kT,V states j states j

Hence follows an equation for the mean number of particles N̄ of the grand ensemble as a derivative of
ln ZG :
∂ ln ZG
N̄ = . (22.2)
∂(µ/kT ) −1/kT,V
Equation (22.2) is the second fundamental link from the microphysics of the partition function to macro-
scopic thermodynamics.
Higher moments of number are obtained from higher derivatives of ln ZG with respect to µ/kT :

∂ 2 ln ZG
h∆N 2 i = , (22.3)
∂(µ/kT )2 −1/kT,V

∂ 3 ln ZG
h∆N 3 i = , (22.4)
∂(µ/kT )3 −1/kT,V

∂ 4 ln ZG
h∆N 4 i − 3h∆N 2 i2 = , (22.5)
∂(µ/kT )4 −1/kT,V

and so on, exactly paralleling equations (21.5) to (21.7) for the moments of energy. The irreducible
moments h∆N 2 i, h∆N 3 i, and h∆N 4 i − 3h∆N 2 i2 are again additive over independent subsystems.

23. Joint Fluctuations of Energy and Number

The fluctuations of energy and number are in general correlated. For example, the cross-correlation
h∆U ∆N i of energy and number in the grand canonical ensemble is obtained by a technique that by now
is becoming familiar, namely by differentiating ln ZG once with respect to −1/kT , then once with respect

24
to µ/kT :

∂ 2 ln ZG ∂ 2 ZG ∂ZG ∂ZG
= −
∂(−1/kT )∂(µ/kT ) ZG ∂(−1/kT )∂(µ/kT ) ZG ∂(−1/kT ) ZG ∂(µ/kT )
= U N − Ū N̄ = h∆U ∆N i . (23.1)

Now you know, so let’s move on.

24. Volume

The third fundamental link (the first two being eqs. [21.3] and [22.2]) from the microphysics of the
partition function to macroscopic thermodynamics is established by considering the derivative of the
partition function with respect to volume. In the (grand) canonical ensembles the volume V of the
system is a definite quantity, which is not exchanged with the reservoir. However, just as one imagines
thermodynamic operations that change the energy, number, and volume of a microcanonical ensemble,
so also it is possible to imagine thermodynamic operations that change the volume of the system in the
(grand) canonical ensembles.
For the canonical ensemble, it suffices to recognize that changing the volume V , with number of
particles N and temperature T fixed, has an effect on the system, which can be construed as a change
in the energy levels of the states of the ensemble (see §34, eq. [34.9] ff.), so that the derivative of the
partition function with respect to volume is not necessarily zero. While it is customary in textbooks at
this point to make arguments about how all this is related to the notion of pressure and work being done,
I shall simply write down the partial derivative of ln Z(N ) with respect to volume V as
∂ ln Z(N ) P
= , (24.1)
∂V −1/kT kT

which I regard as the definition of pressure P in a canonical ensemble. That the definition is appropriate
becomes clear when comparison is made to the classical equations of thermodynamics.
For the grand canonical ensemble I again write down the partial derivative of ln ZG with respect to
volume V as
∂ ln ZG P
= , (24.2)
∂V −1/kT,µ/kT kT
which I regard as the definition of pressure P in a grand canonical ensemble. A more physical explanation
of what’s going on here appears in §34.
Notwithstanding the apparent similarity of equations (24.1) and (24.2) between the canonical and
grand canonical ensembles, matters are more interesting in the grand case. As you know, partition func-
tions are multiplicative over independent systems in mutual thermodynamic equilibrium. In particular,
the logarithm of the partition function is additive over the subvolumes of an ideal gas in thermody-
namic equilibrium. In other words the logarithm of the grand partition function ln ZG of an ideal gas is
proportional to volume in a system with a given T , µ,
PV
ln ZG = , (24.3)
kT
where the constant of proportionality, P/kT , follows from equation (24.2). The same statement cannot
be made about the partition function Z of the canonical ensemble, for in that case the increase of volume
called for in equation (24.1) is carried out with N rather than µ held fixed, so that the ‘new’ volume will
not in general be in thermodynamic equilibrium with the ‘old’ volume.

25
25. Entropy Revisited

The definition (13.1) of entropy S as Boltzmann’s constant times the logarithm of the number of
states, k ln Ω, is fine for a microcanonical ensemble, but it’s not quite good enough for the canonical
or grand canonical ensembles, where it’s not so obvious how to count different states that occur with
different probability. However, each state of the entire system plus reservoir occurs with equal probability,
so the entropy
Stot = k ln Ωtot (25.1)
of the whole system plus reservoir is well-defined. Now entropy is additive over independent systems,
so it seems a good idea to impose the requirement of additivity on the entropies of the system and the
reservoir. In other words, entropy needs to be defined so that it is the sum of its parts, the entropy S of
the system plus the entropy Sres of the reservoir:

Stot = S + Sres . (25.2)

If the energy (and number) in the reservoir were exactly fixed, life would be very simple: the entropy
of the reservoir would be Sres = k ln Ωres (Ures , Nres ). In reality the energy (and number) in the reservoir
fluctuate slightly, because of the small amount of energy (and number) which the reservoir exchanges
with the system. The obvious thing to do then is to define the entropy of the reservoir as an average over
the fluctuating energy (and number)
X
Sres ≡ k ln Ωres (Ures , Nres ) = k Pj ln[Ωres (U − Uj , N − Nj )] , (25.3)
states j

where the summation is over all states j of the system, and U , N represent the fixed total energy and
particle number of the entire system plus reservoir. The notion of defining entropy by an average becomes
especially plausible when you consider that the thermodynamics of (grand) canonical ensembles involves
the average energy Ū (and average number N̄ ) of the system, in place of the fixed energy U and number N
of a microcanonical ensemble (see eqs. [21.3] and [22.2]). But now according to equation (19.2) or (19.7),
the probability Pj that the system is any one state j is proportional to Ωres (U − Uj , N − Nj ). In fact,
considering that the total number of states of the system plus reservoir is Ωtot = j Ωres (U − Uj , N − Nj )
P

it must be that
Ωres (U − Uj , N − Nj )
Pj = . (25.4)
Ωtot
Substituting equation (25.4) into equation (25.3), and using equations (25.1) and (25.2), yields
X X
S = Stot − Sres = k ln Ωtot − k Pj ln[Pj Ωtot ] = −k Pj ln Pj . (25.5)
states j states j

Equation (25.5) is so important that it is worth writing down again all by itself:
X
S = −k Pj ln Pj . (25.6)
states j

Equation (25.6) is often taken as the definition of entropy in a general ensemble, since it works quite
generally in situations where the definition S = k ln Ω is inadequate. The definition (25.6) of entropy is a
fundament of the mathematical discipline of Information Theory. The demonstration that entropy defined
by equation (25.6) inevitably increases in closed Hamiltonian systems, Boltzmann’s famous H-theorem,
was one of the profound achievements of statistical mechanics in the nineteenth century.
Question (25.1): Show directly that the entropy defined by equation (25.6) is additive over indepen-
dent systems, that it reduces to the definition S = k ln Ω in the case of a microcanonical ensemble, and
that it is zero if the system accesses only a single state j.

26
 In the case of the canonical ensemble, inserting the probabilities (19.5) into the expression (25.6) for
entropy yields " #
S X e−Uj /kT e−Uj /kT Ū
=− ln = ln Z(N ) + . (25.7)
k states j
Z(N ) Z(N ) kT

In the case of the grand canonical ensemble, the probabilities (19.10) inserted into the expression
(25.6) for entropy yield
" #
S X e(−Uj +µNj )/kT e(−Uj +µNj )/kT µN̄ Ū
=− ln = ln ZG − + . (25.8)
k states j
ZG ZG kT kT

26. Free Energy Minimization

Entropy (25.6) strives to attain a maximum in a closed system. However, (grand) canonical ensembles
are not closed systems, being able to exchange energy (and number) with a vast reservoir. Therefore the
entropy of the system is not generally a maximum in thermodynamic equilibrium. Only the total entropy
of the system plus reservoir is maximized in thermodynamic equilibrium.
In a canonical ensemble, the aspiration of the system to maximize its entropy is constrained by the
requirement that its temperature equal that of the vast reservoir, T = Tres . This means that any variation
of the entropy of the system is accompanied by an exchange of energy with the reservoir in such a way
that
∂S δ Ū δ Ū
δS = δ Ū = = (26.1)
∂ Ū T Tres
for small variations about equilibrium. In other words,
!
1 Ū
δS − δ Ū = δ S − =0, (26.2)
T T

which is to say that the entropy of the canonical ensemble is not maximized, but rather the quantity
S̄ − Ū /T is maximized. Equivalently, the free energy F

F ≡ Ū − T S (26.3)

is minimized. It should be remarked that the principle of free energy minimization is more limited than
the principle of entropy maximization, since the former can be applied only to systems that are already
partially equilibrated in the sense of having a definite temperature. By contrast, entropy is defined, by
equation (25.6), for arbitrary ensembles.
The principle of free energy minimization is useful for practical computations of thermodynamic
equilibrium in complicated mixtures of reacting species (see §35, paragraph containing eqs. [35.13]-
[35.15]). That is, one writes down the free energy of a system as a function of temperature T , volume V ,
and the numbers NX of the various particle species X of interest, and then finds numerically the minimum
of the free energy with respect to variations of the numbers NX .
Question (26.1): What quantity is maximized/minimized in a grand canonical ensemble?

27
27. Summary of Thermodynamical Equations for the (Grand) Canonical Ensembles

(a) Canonical Ensemble

The properties of the partition function Z(N ) of the canonical ensemble, as defined by equation (19.6),
can be summarized in two equations (gotten from eqs. [21.3], [24.1], and [25.7]):

−1 P
 
d ln Z(N ) = Ū d + dV , (27.1)
kT kT
S Ū
ln Z(N ) = − , (27.2)
k kT
from which follow all the familiar equations of thermodynamics for systems where N is fixed. In par-
ticular, eliminating ln Z(N ) between equations (27.1) and (27.2) leads to the familiar equation (14.1) of
thermodynamics, except with dN = 0 since N is fixed. Equations (27.1) and (27.2) are true in both ideal
and nonideal gases, since none of the relevant arguments in §§18–25 invoked ideality. Equation (27.2)
can be written
−kT ln Z(N ) = Ū − T S , (27.3)
which is the free energy of the canonical ensemble.
(b) Grand Canonical Ensemble
The properties of the grand partition function ZG of the grand canonical ensemble, as defined by
equation (19.11), can be summarized in three equations (gotten from eqs. [21.3], [22.2], [24.2], [25.8], and
[24.3]):

−1 µ P
   
d ln ZG = Ū d + N̄ d + dV , (27.4)
kT kT kT
S Ū µN̄
ln ZG = − + , (27.5)
k kT kT
PV
ln ZG = , (27.6)
kT
from which flow the two familiar fundamental equations (14.1) and (14.2) of thermodynamics in which
number N is also allowed to vary. Equations (27.4) and (27.5) are true in both ideal and nonideal gases,
but equation (27.6) is only true in ideal gases (see the argument in the paragraph containing eq. [24.3]).
The classical thermodynamic quantity (27.5) corresponding to ln ZG is so important that you’d think
it, or something near it, like kT ln ZG , would have a suitably impressive name, but it doesn’t. Pathria
calls ln ZG the q-potential. A close relative is the so-called Gibbs free energy G ≡ U − T S + P V , which
is kT times the logarithm of the partition function that you would get if you considered an ensemble in
which the number of particles N is fixed, but energy U and volume V are allowed to vary by exchange
with a vast thermal reservoir. In some sense, the q-potential is to the Gibbs free energy as Eulerian
mechanics is to Lagrangian mechanics, as a Hamiltonian is to a Lagrangian, as the Schroedinger or Dirac
equations are to the Feynman path integral formalism.
Question (27.1): Consider a system of particles of mass m immersed in a uniform externally applied
gravitational field g. Is thermodynamic equilibrium still possible? Argue that the presence of the gravi-
tational field makes the energies of the states of a parcel of gas a function of its height r. Show that the
partial derivative of the grand partition function, equation (19.11), of a parcel of gas with respect to its
height r is
∂ ln ZG N̄ mg
= . (27.7)
∂r T,µ,V kT

28
Hence show that an ideal gas in thermodynamic equilibrium must satisfy the usual equation of hydrostatic
equilibrium
∇P = ρg , (27.8)
where ρ ≡ mN̄ /V is the mass density (assume equation [27.6]). Do T or µ vary with height?
Question (27.2): A one-dimensional quantum mechanical simple harmonic oscillator has equally
spaced energy levels n = h̄ω(n + 1/2), with one state per energy level.
(a) For a single particle in the SHO, determine explicit expressions for partition function, the mean
energy, and the specific heat.
(b) The classical limit corresponds to h̄ → 0. Find the classical limits of your results for (a).

28. Partition Function of an Ideal Fermi-Dirac Gas

The key property of a system in statistical mechanics is its partition function. Once you know the
partition function of a system, working out the thermodynamics is just a matter of applying equations
(27.1) and (27.2) or (27.4) to (27.6) of the previous section.
A Fermi-Dirac, or Fermi, gas has the property that the particles of the system are indistinguishable,
and no two particles can occupy exactly the same single-particle state. The single-particle partition
function Z(1) of the system is, as always,

e−i /kT ,
X
Z(1) ≡ (28.1)
states i with Ni =1

where I’ve written the energy as  rather than U to remind you that it’s the energy of a single particle.
If the system is ideal and is described by Fermi statistics, how is the partition function Z(2) for say two
particles related to the single-particle partition function? It’s
1
e−Uj /kT = e−i /kT e−j /kT
X X X
Z(2) ≡
states j with Nj =2
2! states i6=j with Ni =1 states j with Nj =1
 
1 
e−2i /kT  .
X
= Z(1)2 − (27.7)
2! states i with Ni =1

The factor 1/2! in equation (28.2) comes in because the two-particle state ij is indistinguishable from j i.
The last term − e−2i /kT in equation (28.2) arises because the Z(1)2 term includes states where both
P

particles are in the same state i; such states are not allowed by the exclusion principle in Fermi statistics,
so the contribution must be subtracted off. In general, the N -particle partition function Z(N ) in an ideal
Fermi gas looks like
1
e−1 /kT e−2 /kT ...e−N /kT .
X X X
Z(N ) = ...
N ! states i1 6=i2 ,...,iN with Ni1 =1 states i2 6=i3 ,...,iN with Ni2 =1 states iN with NiN =1
(28.3)
In other words, Z(N ) is a mess for a Fermi gas. In the particular case where there are g single-particle
states that all happen to have exactly the same energy  (‘equivalent’ states), then Z(N ) for an ideal
Fermi gas would go over to
g!
Z(N ) = e−N/kT (Case of g 1-particle states with same energy) , (28.4)
N !(g − N )!

29
but equation (28.4) is not true in general, because in general single-particle states have various energies .
You can if you want write the 2-particle partition function Z(2), equation (28.2), as (letting Z(T ) denote
the single-particle partition function Z as a function of T )
1
Z(2) = [Z(T )2 − Z(T /2)] . (28.5)
2!
If you really insist, you can also construct formulae like equation (28.5) for Z(N ), but I’ll tell you right
now there’s a better way. The general conclusion is that Z(N ) is not the nicest thing to work with in a
Fermi gas.
The grand partition function ZG on the other hand makes life a lot easier, because the grand partition
function is multiplicative over single-particle states. This is because in the grand canonical ensemble,
knowing that a certain single-particle state does or does not contain a particle provides no information
about any other single-particle state. By contrast, in the canonical ensemble, where the number N of
particles is fixed, knowing that there is a particle in a particular single-particle state immediately tells
you information about the state of the rest of the system, namely that the rest of the system must contain
N −1 particles. So consider a particular single-particle state i. In Fermi statistics, the single-particle state
can contain either zero or one particle. The partition functions for zero and one particles are, trivially,

Zi (0) = 1 , Zi (1) = e−i /kT . (28.6)

The grand partition function ZG,i for the particular single-particle state i is then
1
X
ZG,i = Zi (N )eµN/kT = 1 + e(−i +µ)/kT . (28.7)
N =0

That the grand partition function is multiplicative over single-particle states is the same thing as say-
ing that the logarithm of the grand partition function is additive over single-particle states. Thus the
logarithm of the grand partition function of an ideal Fermi gas is
X h i
ln ZG = ln 1 + e(−i +µ)/kT , (Fermi-Dirac) (28.8)
1-particle states i

which is a lot simpler than Z(N ). Equation (28.8) for the grand partition function also gives you a
prescription for figuring out what Z(N ) is. If you expand ZG as a power series in eµ/kT , then the
coefficients are, according to equation (19.12), just the N -particle partition functions Z(N ). I invite you
to try it.

29. Partition Function of an Ideal Bose-Einstein Gas

A Bose-Einstein, or Bose, gas has the property that the particles of the system are indistinguishable,
and any number of particles can occupy the same single-particle state. The case of Bose-Einstein goes
much the same as Fermi-Dirac. The single-particle partition function Z is defined in the usual way,
equation (28.1). The 2-particle partition function Z(2) is
 
1
e−Uj /kT e−2i /kT  .
X X
Z(2) ≡ = Z(1)2 + (29.1)
states j with Nj =2
2! states i with Ni =1

The factor 1/2! in equation (29.1) is because the two-particle state ij is the same as j i. The last term
P −2i /kT
e in equation (29.1) has to be added in because the Z(1)2 /2! term includes only half the correct

30
contribution to the two-particle state where both particles are in the same single-particle state i. Once
again you can see that Z(N ) is going to be a mess for a Bose gas. In the particular case where there
are g single-particle states that all happen to have exactly the same energy  (‘equivalent’ states), then
Z(N ) for an ideal Bose gas would go over to
(g + N − 1)! −N/kT
Z(N ) = e (Case of g 1-particle states with same energy) , (29.2)
N !(g − 1)!
but equation (29.2) is not true in general.
As in the Fermi case, dealing with the grand partition function makes life a lot easier, because the
grand partition function is multiplicative over single-particle states. For N particles in the single-particle
state i, the partition function is
Zi (N ) = e−N i /kT , (29.3)
so the grand partition function ZG,i for the single-particle state i is
∞
X 1
ZG,i = Zi (N )eµN/kT = . (29.4)
N =0
1− e(−i +µ)/kT
The logarithm of the grand partition function is additive over single-particle states, so the logarithm of
the grand partition function of an ideal Bose gas is
X h i
ln ZG = − ln 1 − e(−i +µ)/kT (Bose-Einstein) . (29.5)
1-particle states i

Question (29.1): Superfluid 4 He is sometimes described as a Bose gas. How is it that a gas whose
elementary particles are fermions, hence obey an exclusion principle, can show bosonic behavior, in which
a large number of particles can occupy the same single-particle state?

30. Partition Function of an Ideal Planck Gas

A Planck gas is the special case of a Bose gas in which number of particles is not conserved. The
obvious example is the case of a gas of photons. Since the number of particles is variable, it is more
appropriate to use the grand canonical ensemble than the canonical ensemble. However, since number
is not conserved, there is no thermodynamic variable µ corresponding to number conservation. The
partition function is obtained from the Bose case (29.5) simply by setting µ = 0:
h i
− ln 1 − e−i /kT
X
ln ZG = (Planck) . (30.1)
1-particle states i

31. Partition Function of an Ideal Boltzmann Gas

Boltzmann statistics is properly obtained as the nondegenerate limit of Fermi or Bose statistics when
the number of particles per single-particle state is very much less than one. In this case, the N -particle
partition function Z(N ) does take a simple form:
e−Uj /kT
X
Z(N ) ≡
states j with Nj =N
1
e−1 /kT e−2 /kT ... e−N /kT
X X X
= ...
N ! states i1 with Ni1 =1 states i2 with Ni2 =1 states iN with NiN =1
1
= Z(1)N . (31.1)
N!

31
The extra terms that appeared in the Fermi and Bose N -particle partition functions, such as in the 2-
particle partition functions (28.2) or (29.1), can be ignored in the Boltzmann case because the probability
of two particles occupying (or trying to occupy) the same single-particle state is negligibly small.
The grand partition function of a Boltzmann gas is properly derived as the limit of the Fermi or
Bose grand partition functions in the limit of small occupation numbers. Since the probability that a
single-particle state i is occupied by N particles goes like eN (−i +µ)/kT , the condition of small occupation
numbers for arbitrary single-particle states, in particular for the ground single-particle state 0 where
0 = 0, is
eµ/kT  1 . (31.2)
In this limit (31.2) of small occupation numbers, the ideal Fermi and Bose grand partition functions, eqs.
(28.8) and (29.5), go over to the grand partition function of an ideal Boltzmann gas,
X
ln ZG = e(−i +µ)/kT = Z(1)eµ/kT (Boltzmann) . (31.3)
1-particle states i

If the grand partition function ZG from equation (31.3) is expanded as a power series in eµ/kT , then one
recovers a second time the result (31.1) for the N -particle partition functions Z(N ) in an ideal Boltzmann
gas.

32. Partition Function of an Ideal Gas of Distinguishable Particles

It is possible to consider formally a gas of distinguishable particles, even though such a gas would
not normally appear in nature. Since each particle is of a different ‘type’, it is more appropriate to use
the canonical ensemble, in which the particle number is fixed, rather than the grand canonical ensemble,
which involves a reservoir presumably containing one each of an infinite number of particle types. The
grand canonical ensemble can nevertheless be used, as long as it is understood that equation (27.6) is no
longer valid.
For a gas of distinguishable particles the N -particle partition function Z(N ) is

e−Uj /kT
X
Z(N ) ≡
states j with Nj =N

e−1 /kT e−2 /kT ... e−N /kT

X X X
= ...
states i1 with Ni1 =1 states i2 with Ni2 =1 states iN with NiN =1
N
= Z(1) , (32.1)

which differs from the Boltzmann case (31.1) by a factor of N !.

Nevertheless, in the real world there are different species of particle, that are indeed distinguishable.
But distinguishable by whom? E. T. Jaynes (https://bayes.wustl.edu/etj/articles/gibbs.paradox.pdf)
proffers the following scenario. Suppose that there are two isotopes of argon that are so similar that to one
experimenter they appear identical. However, a second experimenter knows the secret difference between
the two isotopes. The second experimenter has crafted two membranes, each of which is impenetrable
by one isotope but transparent to the other. The second experimenter prepares a volume with two
equal compartments, one containing only the first isotope, the second containing only the second. The
two compartments are arranged to contain the same number of particles at the same temperature and
pressure. The only distinction between the two compartments is that they carry different isotopes. The
first experimenter, not knowing that the compartments contain different isotopes, assigns an entropy to
the system that is N ln 2 larger (in energy units k = 1, with N the total combined number of particles)
than the entropy assigned by the second. The first experimenter opens a hole in the barrier between the

32
compartments, allowing the isotopes to mix thoroughly, and then closes the hole. The first experimenter,
believing the isotopes to be identical, claims that the entropy has remained unchanged. The second
experimenter, who knows the isotopes are different, concludes that the entropy has increased by the
mixing entropy N ln 2. At this point the first and second experimenters agree on the entropy of the
mixed system. But while the first experimenter concludes that the entropy has remained the same, the
second experimenter concludes that the entropy has increased.
Redo the experiment with the second experimenter at the helm. Once again, the initial volume
contains two equal compartments that differ only in that they carry different isotopes. Instead of opening
a hole in the barrier between the compartments, the second experimenter replaces the barrier with two
adjacent membranes, one impenetrable to the first isotope, serving to confine the first isotope, the other
impenetrable to the second isotope, serving to confine the second isotope. Each membrane is attached to
a piston. The first membrane, which is transparent to the second isotope, feels only the pressure from the
first isotope on one side. Likewise the second membrane feels only the pressure from the second isotope,
on the opposite side. The second experimenter, holding the pistons, allows the two membranes to move,
slowly, adiabatically, away from each other. As the membranes recede from each other, the volume divides
into three parts: the volume between the two membranes fills with a mixture of both isotopes, while the
volumes at the two sides each contain a pure isotope. Each isotope, bounded by its personal membrane,
fills an increasing volume (part mixed, part pure) that in due course expands to fill the entire volume.
At the end of the experiment, both isotopes fill the entire volume, and are fully mixed. But because
the expansion has been adiabatic, P dV work has been done, and the temperature has decreased thanks
to adiabatic cooling. The second experimenter concludes that, because the experiment was adiabatic,
the entropy has remained constant during the experiment. The first experimenter is mystified by the
experiment. Initially, the system contains two equal compartments with equal pressure on either side.
The system appears to be in thermodynamic equilibrium. But as the first experimenter watches, the two
membranes spontaneously, miraculously, move apart, with a high density region appearing between the
two membranes, and regions of lower density on either side. The temperature of the system decreases,
while the membranes push pistons, providing useful work to the outside. The entropy of the system
decreases, apparently violating the second law of thermodynamics. A perpetual motion machine!? At
the end of the experiment, both experimenters agree on the entropy of the mixed system. But while the
second experimenter concludes that the entropy has remained the same, the first experimenter concludes
that entropy has decreased.
Jaynes’ point is that two different observers can consistently count entropy differently depending on
whether or not they can distinguish particles. As long as the first experimenter has no way to distinguish
particles, they count entropy as that of indistinguishable particles. But when the second experimenter
shows their experiment to the first, and shares the secret of the distinguishable isotopes, the second
experimenter revises their measure of entropy, and is able to extract work from what originally appeared
to be a system in thermodynamic equilibrium at maximum entropy.
Experiment indicates that fundamental particles such as photons are fundamentally indistinguishable.
In that case it is fundamentally impossible to lower the entropy of a closed system that has reached
thermodynamic equilibrium.
Question (32.1): If you watch a 3D movie at the cinema, you will be provided with a pair of 3D glasses
whose two lenses transmit respectively only right-handed and only left-handed circularly polarized light.
The two lenses provide a practical realization of Jaynes’ membranes. Yet photons are fundamentally
indistinguishable. Do the 3D lenses provide a way to extract entropy from a system of photons in
thermodynamic equilibrium? Answer. No. A system of photons in thermodynamic equilibrium has each
spin state, right- and left-handed polarization, equally occupied. It is possible to extract entropy from a
system of photons with unequally occupied spins, but that system is not in thermodynamic equilibrium.

33
33. Constraints on Thermodynamic Variables

Conditions for thermodynamic equilibrium to be stable were derived in §16. However, there exist
other constraints on temperature, chemical potential, and pressure which arise not so much from the
requirement of stability, but rather from certain properties of the states of the system.
(a) Temperature
Probabilities of states j in (grand) canonical ensembles go like Pj ∝ e−Uj /kT . Thus systems at positive
temperature have the property that states of higher energy are less probable than states of lesser energy.
Conversely, systems at negative temperature have the property that states of higher energy are more
probable than states of lesser energy. Therefore, if there is no upper bound to the energy of the possible
states of the system, then temperature must be positive, since a negative temperature would require
an infinite amount of energy. Conversely, if there is an upper bound to the energy of states, then the
temperature may be negative.
In realistic systems, where the free motions of particles permits them an unbounded amount of energy,
temperature must be positive. However, it is possible to imagine, even to construct, systems in which
particles have no free motion, being confined for example to the lattice of a crystal. In such a system
each particle may only have a finite number of single-particle energy states, which might for example
be associated with the interaction of the particle spin with an external magnetic field. Such systems
can have negative temperatures if it can be arranged that the more energetic of the spin single-particle
states are more highly populated than the less energetic. It is to be noted that although the state of the
system at negative temperature is more energetic than at positive temperature, it is impossible to achieve
negative temperatures by heating the system. At best, heating can increase the energy to the point where
all single-particle states are equally occupied regardless of energy, corresponding to infinite temperature.
More devious means must be used to achieve a population inversion and hence a negative temperature.
In the example just mentioned, where the energy levels arise from the interaction of spin with an external
magnetic field, a negative temperature might be attained by reversing the direction of the magnetic field,
so that the spin single-particle states of lesser energy suddenly become the spin single-particle states
of higher energy. Ultimately, systems with negative temperature are unstable in nature, because they
drive inevitably toward thermodynamic equilibrium with the external medium, which, having unbounded
energy states, must have a positive temperature.
(b) Chemical Potential
In an ideal Bose gas, convergence of the single-particle grand partition function, eq. (29.4),
∞
X
ZG,i = eN (−i +µ)/kT (33.1)
N =0

for any single-particle state i, in particular for the ground single-particle state 0 where 0 = 0, requires
that the chemical potential be zero or negative
µ
≤0. (33.2)
kT
The zero of energy in equation (33.2) is defined as the energy of the lowest energy state, the ground state.
In an ideal Fermi gas, the grand partition function converges for any value of µ/kT . Large negative
values of µ/kT correspond to the nondegenerate limit, where the mean number of particles per single-
particle state is much less than one. Conversely, large positive values of µ/kT correspond to the degenerate
limit, where the number of particles per single-particle state approaches unity in single particle states
with energy i less than µ.
(c) Pressure

34
 According to the standard relation ∂S/∂V = P/T , eq. (14.1), a gas with negative P/T can increase
its entropy by reducing its volume. In an ideal gas, therefore, the pressure (more strictly, P/T ) must
always be positive, since under a condition of negative pressure the gas would contract spontaneously.
Examination of the grand partition functions ln ZG = P V /kT of ideal Fermi and Bose gases, eqs. (28.8)
and (29.5), shows that indeed P/T is invariably positive in these cases.
Negative pressures are however possible in nonideal gases, and routine in solids. In a nonideal gas
inside a finite container, the contraction of the gas requires the creation of surfaces that may require more
energy to form than is gained from the contraction of the gas. Thus thermodynamic equilibrium may
occur even at negative pressure. As for a solid, like a wire say, chances are you can put it under quite a
bit of tension (negative pressure) before it busts.

34. Occupancy

The occupation number, or occupancy, of a single-particle state i in a grand canonical ensemble is

by definition the mean number N̄i of particles in that state. Occupation numbers can offer an intuitively
appealing and physically elegant alternative to more commonly used measures such as photon intensity
(Krolik and McKee 1978, ApJ Supp, 37, 459, is a fine example). For the purpose of this section, occupancy
numbers provide an opportunity to review the equations of thermodynamics from a slightly different, less
rigorous, but more intuitive, angle.
In a Fermi gas, a single-particle state i can contain either zero or one particle, so the occupation
number N̄i is, from the probabilities (19.9),
1
X e(−i +µ)/kT 1
N̄i = N Pi (N ) = (− +µ)/kT
= ( −µ)/kT . (34.1)
N =0
1+e i e i +1

The same result is obtained more formally by differentiating the grand partition function ZG,i of the
single-particle Fermi state i, equation (28.7), with respect to µ/kT , according to the fundamental equation
(27.6):
∂ ln ZG,i 1
N̄i = = ( −µ)/kT . (34.2)
∂(µ/kT ) −1/kT,V e i +1
A similar calculation for a Bose gas, where the number of particles in a single-particle state i can vary
from zero to infinity, yields the occupation number N̄i
∞ P∞
N e(−i +µ)N/kT 1
N Pi (N ) = PN∞
=0
X
N̄i = (− +µ)N/kT
= ( −µ)/kT . (34.3)
N =0 N =0 e
i e i −1

Again, the same result is obtained more formally as the partial derivative with respect to µ/kT of the
grand partition function function ZG,i of the single-particle Bose state i, equation (29.4). A Boltzmann
gas corresponds to the limit of small occupation numbers N̄i for all single-particle states i in a Fermi
or Bose gas. The limit of small occupation numbers is attained when eµ/kT  1, or equivalently when
µ/kT → −∞. The occupation number N̄i in a Boltzmann gas is then
N̄i = e(−i +µ)/kT . (34.4)
The occupation numbers N̄i of the Fermi, Bose, and Boltzmann gases, equations (34.1), (34.3), and (34.4),
can be summarized in the single formula
1
N̄i = , (34.5)
 1


e(i −µ)/kT + 0
−1


35
where the numbers 1, 0, and −1 correspond respectively to Fermi, Boltzmann, and Bose gases.
The mean energy Ūi of a single-particle state is, obviously,
X
Ūi = i N Pi (N ) = i N̄ , (34.6)
N

which can also be derived formally by differentiating the grand partition function ZG,i of the single-particle
state i with respect to −1/kT , according to equation (27.6).
The total number N̄ and energy Ū in a grand canonical ensemble can be written as sums over the
single-particle state occupation numbers X
N̄ = N̄i , (34.7)
i
X
Ū = i N̄i , (34.8)
i
which look kind of obvious. You can check that the formal partition function approach gives the same
results.
Pressure is a little bit more subtle. Consider the notion of defining the pressure Pi of a single-particle
state (not to be confused with probability Pi ). According to the fundamental equation (27.6), Pi should
be given by
Pi ∂ ln ZG,i 1 ∂ X
= = e(−i +µ)N/kT
kT ∂V −1/kT,µ/kT Z G,i ∂V −1/kT,µ/kT N

∂i 1 X ∂i N̄i

=− N e(−i +µ)N/kT = − , (34.9)
∂V −1/kT,µ/kT kT ZG,i N ∂V kT
so that the pressure Pi associated with the single-particle state i is
∂i
Pi = − N̄i . (34.10)
∂V
The question which equation (34.10) raises is, what is ∂i /∂V ? The change in the energy level i caused
by a change in the volume V is to be evaluated with T and µ held constant, but in fact the energy i
does not depend on T or µ in any case: the energy level is determined by the boundary conditions of the
problem, which is to say that i is, in the present case, only a function of volume V . As long as the gas
is ideal, only the translational states of particles are affected by the volume V ; the internal energy levels
of particles (atoms) are unchanged. For the translational states, the number of states in an interval of
phase volume is gspin d3xd3p/h3 , so it must be that when the volume V is changed the momentum of the
states changes as
V p3 = constant , (34.11)
at least on average. If the scaling relation (34.11) is taken to be literally true for individual free states,
then from the relativistic formula (free + mc2 )2 − p2 c2 = m2 c4 it follows that

∂free pc2 ∂p v∂p vp

− =− 2
=− = , (34.12)
∂V free + mc ∂V ∂V 3V
where v is the velocity of the particle. Equation (34.12) reduces in the nonrelativistic case to
∂free 2free
− = (free  mc2 ) , (34.13)
∂V 3V
and in the highly relativistic case to
∂free free
− = (free  mc2 ) . (34.14)
∂V 3V

36
Actually, equations (34.12)-(34.14) are not necessarily true for every translational energy level, but they
are true on average. The total pressure P is a sum over the pressures Pi of single-particle states, which
from equations (34.10) and (34.12) is
X 1 X
P = Pi = vi pi N̄i . (34.15)
i
3V i

Equation (34.15) is the general formula for the pressure of an ideal gas of free particles. In the nonrela-
tivistic case the total pressure P , equation (34.15), is

1 X 2Ūfree
P = 2free,i N̄i = , (34.16)
3V i 3V

while in the relativistic case the total pressure P is

1 X Ūfree
P = free,i N̄i = , (34.17)
3V i 3V

which agree with the results (17.14) and (17.15) for an ideal gas obtained previously by other scaling
arguments. The above arguments about the pressure were not entirely rigorous, because of the sloppy
application of the scaling law (34.11). However, the same results (34.16) and (34.17) for the total pressure
P can be derived more formally, in the usual way, by differentiating the grand partition function with
respect to volume, according to the fundamental equation (27.6).
In the following two questions you may use the fact that the number of single-particle states in an
interval of phase space is gspin d3xd3p/h3 .
Question (34.1): Derive the Maxwell-Boltzmann distribution, that is, the distribution of number
densities of particles with velocity v in an ideal Boltzmann gas of nonrelativistic free particles of mass m
and total number density ntot in thermodynamic equilibrium at a temperature T :
3/2
4 m

2 /2kT
dn = ntot e−mv v 2 dv . (34.18)
π 1/2 2kT

Question (34.2): Derive the Planck distribution of intensity Bν (energy per unit time per unit area
per unit frequency per steradian)
2hν 3 dν
Bν dν = 2 hν/kT (34.19)
c (e − 1)
for a gas of photons in thermodynamic equilibrium.

35. Equilibria Between Reacting Species

In a system containing many different particles species X, the first fundamental equation (14.1) of
thermodynamics looks like X
T dS = dU − µX dNX + P dV . (35.1)
X

Typically, there are processes, or reactions, that transform some sets of particles into other sets of
particles. For example, a hydrogen atom may be converted by ionization into a proton and an electron,
and conversely, a proton and an electron may recombine to form a hydrogen atom:

H ↔p+e . (35.2)

37
As a result, the numbers NX of each particle species X are not conserved individually, but rather, only
the numbers NA of fundamental constituents A are conserved. In the example (35.2), the numbers NH of
hydrogen atoms, Np of free protons, and Ne of free electrons are not individually conserved, but the total
number of protons, free and bound, which is Np + NH , and the total number of electrons, free and bound,
which is Ne + NH , are conserved. This implies that the chemical potentials µX in equation (35.1) are not
all independent; rather, only the chemical potentials µA of the fundamental conserved constituents are
independent. If the number of fundamental particles of type A that compose particle X is written nAX ,
then the total number of fundamental particles of type A is
X
NA = nAX NX . (35.3)
X

Thus the fundamental equation (14.1) of thermodynamics can also be written

!
X X
T dS = dU − µA d nAX NX + P dV , (35.4)
A X

which properly reflects the conservation of each of the fundamental particle types A. The equivalence of
equations (35.1) and (35.4) implies that the chemical potential µX of particle X is related to the chemical
potentials of its constituents by X
µX = nAX µA . (35.5)
A

In the example (35.2), the chemical potential of the hydrogen atom H, which is composed of a proton p
and an electron e, is the sum of the chemical potentials of the proton and the electron

µH = µp + µe . (35.6)

The mean number N̄X of reacting species X in thermodynamic equilibrium is determined in the usual
way by differentiating the logarithm of the grand partition function with respect to µX /kT , eq. (22.2),

∂ ln ZG
N̄X = , (35.7)
∂(µX /kT ) −1/kT,µY /kT

where the partial derivative with respect to µX /kT is carried out with the chemical potentials µY of all
other particles Y 6= X formally held fixed. Equation (35.7) expresses the mean number N̄X of particles X
as a function of the chemical potentials µX (and T , V ). If the relation (35.5) for the chemical potential µX
in terms of the chemical potentials µA of its fundamental constituents A is now imposed, then equations
(35.7) for all X comprise a set of implicit relations between the numbers N̄X of the various reacting
particle species. In the case of an ideal Boltzmann gas, considered in the next paragraph, it is possible
to eliminate the chemical potentials so as to obtain explicit, rather than implicit, relations between the
numbers NX , but this is not in general possible.
So consider the case of an ideal Boltzmann gas. In an ideal gas, the grand partition function ZG of
the whole system is multiplicative over the grand partition functions ZG,X of each of the particle species
X X
ln ZG = ln ZG,X . (35.8)
X

Substituting into equation (35.8) the expression (31.3) for the grand partition function of an ideal Boltz-
mann gas yields X
ln ZG = ZX (1)eµX /kT , (35.9)
X

38
where ZX (1) is the single-particle partition function of the species X. The number N̄X of particle species
X follows by differentiating the grand partition function (35.9) with respect to µX /kT
∂ ln ZG
N̄X = = ZX (1)eµX /kT . (35.10)
∂(µX /kT )

Equation (35.10) can be inverted to yield the chemical potential µX in terms of the number NX
!
N̄X
µX /kT = ln . (35.11)
ZX (1)

Imposing the relation (35.5) for the chemical potential µX then implies
!nAX
N̄X Y N̄A
= , (35.12)
ZX (1) A
ZA (1)

which is the Saha equation in its most general form. The Saha equation (35.12) determines the numbers
of the different particle species in an ideal Boltzmann gas in thermodynamic equilibrium.
In a nonideal or non-Boltzmann gas, it is not usually possible to obtain explicit relations between the
numbers NX of particle species in thermodynamic equilibrium. A procedure widely used in the literature
to determine the numbers NX is called free energy minimization (see §26). In this procedure one considers
a canonical ensemble, in which the temperature T , volume V , and numbers of fundamental particles NA
are fixed. The numbers NX of particle species are then varied numerically, at fixed T , V , and NA , until
the free energy Ū − T S is a minimum. It is not difficult to see that the minimization of free energy
is equivalent to the condition (35.5) on the chemical potentials of reacting species in thermodynamic
equilibrium. Reexpressed as an equation for the total derivative of the free energy Ū − T S, the first
fundamental equation (14.1) of thermodynamics looks like
X
d(Ū − T S) = −SdT + µX dNX − P dV . (35.13)
X

In particular, the partial derivative of free energy with respect to number NX is

∂(Ū − T S)
= µX . (35.14)
∂NX T,NY (Y6=X),V

The minimum of the free energy Ū − T S at fixed NA is achieved when the abundances NX are such that

∂(Ū − T S) X ∂(Ū − T S)
− nAX =0 (35.15)
∂NX A
∂NA

is satisfied for all species X. Equation (35.15) is exactly the same condition as the relation (35.5) between
the chemical potentials of reacting species.
Question (35.1): Show that the number densities of an ideal nonrelativistic Boltzmann gas of H-atoms,
protons, and electrons in thermodynamic equilibrium satisfy (notation: n ≡ N̄ /V )
3/2
np ne gp ge 2πme kT

= , (35.16)
nH ZH,int h2

where ZH,int is the internal partition function of a hydrogen atom. Argue that if only the ground state
of hydrogen is considered, and if the zero-point of energy of H is taken equal to that of the just-ionized
atom, then
ZH,int = gH eχH /kT , (35.17)

39
where χH is the ionization potential of hydrogen. What are the numbers ge , gp , and gH of spin states?
Question (35.2): The most tightly bound of all nuclei is 56 Fe. Once a star has synthesized its way
to 56 Fe, no more nuclear energy is left. A star in such a condition gets its energy from gravitational
contraction, which causes the star to heat up. Eventually, the temperature gets high enough to dissociate
56 Fe into α particles (4 He nuclei) and neutrons by

56
Fe → 13α + 4n (35.18)

which costs 124.4129 MeV dissociation energy per 56 Fe. Write down the Saha equation that describes the
equilibrium between 56 Fe, α particles, and neutrons. If the original (low temperature) composition of the
star was entirely 56 Fe, argue that nn /nα = 4/13 (assume that the reaction [35.18] is very much faster
than any weak interaction that converts neutrons to protons or vice versa). Show that the temperature
at which 56 Fe dissociates is about 4 MeV, with a weak dependence on density.
Question (35.3): Derive implicit relations between the equilibrium number densities of an ideal gas
of H-atoms, protons, and electrons when degeneracy of the electrons is becoming important (i.e., treat
the H-atoms and protons as ideal Boltzmann gases, but the electrons as a Fermi gas). Include only the
ground state of the H-atom, and assume that the ground state energy remains unperturbed even at high
electron density. Interpret your results physically. Is the approximation of an unperturbed ground state
of the H-atom likely to be a good one? Does the energy of the continuum relative to the ground state,
the ionization potential, increase or decrease as the electron density increases?

36. Summary of Equations of State of Ideal Gases

For reference, this section gives equations for the number density n ≡ N̄ /V , pressure P , and energy
density Ū /V in ideal Fermi, Bose, and Boltzmann gases of particles, in the nonrelativistic and relativistic
limits. The results are obtained using the fundamental equation (27.6), and the grand partition functions
(28.8), (29.5), and (31.3). The most commonly encountered cases in astrophysics are: the nonrelativistic
Boltzmann gas, equations (36.8)-(36.10), or equations (36.37)-(36.39) if internal degrees of freedom of
particles are also to be included; the nonrelativistic degenerate Fermi gas, equations (36.11)-(36.13); and
the Planck gas, equations (36.26)-(36.28).
The general form of the grand partition function of an ideal gas of free particles is

Z  ln[1 + e(−+µ)/kT ] 
 
X gspin 4πp2 dp
ln ZG = ln ZG,i = V e(−+µ)/kT , (36.1)
(−+µ)/kT  h3
− ln[1 − e

1-particle free states i ]

where the upper, middle, and lower rows in the integrand correspond respectively to the cases of Fermi,
Boltzmann, and Bose gases. The quantity gspin is the number of spin states of the particle. For nonrela-
tivistic free particles of mass m, the momentum p and the energy  are related by

p2
= (  mc2 ) (36.2)
2m
while for highly relativistic particles
 = pc (  mc2 ) . (36.3)

The quantities Fn+ (α) and Fn− (α) used below denote the Fermi-Dirac (+) and Bose-Einstein (−)
integrals
Z ∞ n−1
x dx
Fn± (α) ≡ x+α
. (36.4)
0 e ±1

40
Salient properties of these integrals are summarized in the Appendix to this section.
(a) Nonrelativistic Free Particles
The number density n, pressure P , and energy density Ū /V of an ideal gas of nonrelativistic free
fermions or bosons are
N̄ gspin 2π(2mkT )3/2 ±
n≡ = F3/2 (−µ/kT ) , (36.5)
V h3
±
gspin 4π(2m)3/2 (kT )5/2 ± 2F5/2 (−µ/kT )
P = F5/2 (−µ/kT ) = nkT ± , (36.6)
3h3 3F3/2 (−µ/kT )
3P V
Ū = , (36.7)
2
where the + sign on Fn± is for fermions, the − sign for bosons.
In the nondegenerate, Boltzmann limit, µ/kT  0, the number density n, pressure P , and energy
density Ū /V from equations (36.5)-(36.7) go over to

N̄ gspin (2πmkT )3/2 µ/kT

n≡ = e , (36.8)
V h3
gspin (2πm)3/2 (kT )5/2 µ/kT
P = e = nkT , (36.9)
h3
3P V 3N̄ kT
Ū = = . (36.10)
2 2
In the degenerate limit of a Fermi gas, µ/kT  0, the number density n, pressure P , and energy
density Ū /V from equations (36.5)-(36.7) go to

N̄ gspin 4π(2mµ)3/2
n≡ = , (36.11)
V 3h3
gspin 8π(2m)3/2 µ5/2 2nµ h2 32/3 n5/3
P = = = , (36.12)
15h3 5 5m(gspin 4π)2/3
3P V
Ū = . (36.13)
2
The case where µ/kT = 0, corresponding to the case where the number of particles is not conserved,
is also of special interest. In this case, the number density n, pressure P , and energy density Ū /V from
equations (36.5)-(36.7) are
N̄ gspin 2π(2mkT )3/2 ±
n≡ = F3/2 (0) , (36.14)
V h3
± ±
gspin 4π(2m)3/2 (kT )5/2 ± 2F5/2 (0) h2 n5/3 F5/2 (0)
P = F 5/2 (0) = nkT ± = ± , (36.15)
3h3 3F3/2 (0) 3m(gspin 2π)2/3 [F3/2 (0)]5/3
3P V
Ū = . (36.16)
2
where again the + sign on F ± is for fermions, the − sign for bosons.
(b) Relativistic Free Particles
The number density N̄ /V , pressure P , and energy density Ū /V of an ideal gas of relativistic free
fermions or bosons are
N̄ gspin 4π(kT )3 ±
n≡ = F3 (−µ/kT ) , (36.17)
V c3 h3

41
 gspin 4π(kT )4 ± F4± (−µ/kT )
P = F4 (−µ/kT ) = nkT , (36.18)
3c3 h3 3F3± (−µ/kT )
Ū = 3P V , (36.19)
where the + sign on Fn± is for fermions, the − sign for bosons.
In the nondegenerate, Boltzmann limit, µ/kT → −∞, the number density n, pressure P , and energy
density Ū /V from equations (36.17)-(36.19) go over to

N̄ gspin 8π(kT )3 µ/kT

n≡ = e , (36.20)
V c3 h3
gspin 8π(kT )4 µ/kT
P = e = nkT , (36.21)
c3 h3
Ū = 3P V = 3N̄ kT . (36.22)

In the degenerate limit of a Fermi gas, µ/kT  0, the number density n, pressure P , and energy
density Ū /V from equations (36.17)-(36.19) go to

N̄ gspin 4πµ3
n≡ = , (36.23)
V 3c3 h3
gspin πµ4 nµ 31/3 chn4/3
P = = = , (36.24)
3c3 h3 4 4(gspin 4π)1/3
Ū = 3P V . (36.25)

If the number of particles is not conserved, then µ/kT = 0, in which case the number density n,
pressure P , and energy density Ū /V from equations (36.17)-(36.19) are

N̄ gspin 4π(kT )3 ±
n≡ = F3 (0) , (36.26)
V c3 h3
gspin 4π(kT )4 ± F4± (0) chn4/3 F4± (0)
P = F4 (0) = nkT = , (36.27)
3c3 h3 3F3± (0) 3(gspin 4π)1/3 [F3± (0)]4/3
Ū = 3P V . (36.28)
Equations (36.26)-(36.28) include the case of photons, which are bosons, and have gspin = 2. Photons are
so ubiquitous that one commonly writes radiation density (36.28) as

Ū
= aT 4 , (36.29)
V
where the constant a is, from equation (36.27) with (A36.12),

8πk 4 − 8π 5 k 4
a= F4 (0) = . (36.30)
c3 h3 15c3 h3
Radiation pressure (36.27) is
aT 4
P = . (36.31)
3
It is not hard to figure out that the energy flux of photons (energy per unit area per unit time) passing
in one direction through a plane is cŪ /(4V ). From this follows the Stefan-Boltzmann law for the
luminosity L emitted from a surface area A of photons in thermodynamic equilibrium at temperature T ,

L = AσT 4 , (36.32)

42
where σ is the Stefan-Boltzmann constant
ac 2π 5 k 4
σ= = . (36.33)
4 15c2 h3
(c) Particles with Internal Degrees of Freedom
Atoms and molecules have internal as well as translational degrees of freedom. If the gas is ideal, then
invariably the atoms and molecules are well described as Boltzmann particles. For if the density becomes
so high that the electron gas begins to become degenerate, then you can be sure that the energy levels of
the atoms and molecules are strongly perturbed, and that the gas is no longer ideal. Note from equation
(36.8) that the density n at which nonrelativistic particles of mass m become degenerate is proportional
to m3/2 , so that the lighter electrons become degenerate at lower densities than the more massive nuclei.
The grand partition of an ideal Boltzmann gas of particles with internal degrees of freedom is
ln ZG = Zint (1)Zfree (1)eµ/kT , (36.34)
where Zint (1) is the single-particle partition function of the internal states of the particle,
e−i /kT ,
X
Zint (1) = (36.35)
internal states i

and Zfree (1) is the single-particle partition function of the translational degrees of freedom, assumed here
to be nonrelativistic,
gspin 4πp2 dp V gspin (2πmkT )3/2
Z
Zfree (1) = V e−/kT = . (36.36)
h3 h3
The number density n, pressure P , and energy density Ū /V of the ideal Boltzmann gas of nonrelativistic
particles with internal degrees of freedom are then modified from equations (36.5)-(36.7) to
N̄ Zint (1)Zfree (1)eµ/kT gspin (2πmkT )3/2 µ/kT
n≡ = = Zint e , (36.37)
V V h3
P = nkT , (36.38)
3P V
Ū = + Ūint , (36.39)
2
where Ūint is the mean energy of the internal degrees of freedom of a single particle
i e−i /kT .
X
Ūint = (36.40)
internal states i

(d) More than One Particle Type

If there is more than one type of particle X (which may be Fermi, Bose, Boltzmann or any mix
thereof), then number density n, pressure P , and energy density Ū /V are additive over particle types,
P P P
n = nX , P = PX , and Ū = ŪX , provided that they are expressed in terms of temperature T
and chemical potential µ. If the particle species X react, then the relation (35.5) between the chemical
potentials µX must be imposed.
Question (36.1): Write down the grand partition function of an ideal Fermi gas of relativistic free
particles and antiparticles (what is the relation between the chemical potential µ of the particle and that
µ0 of its antiparticle?). Hence obtain expressions for the numbers N̄ and N̄ 0 of particles and antiparticles,
the internal energy Ū , and the pressure P of the gas in terms of the temperature T , chemical potential
µ, and volume V (you should find expressions involving the Fermi-Dirac integrals Fn+ ). Determine the
forms of the equations in the degenerate limit, and in the case where the number of particles equals the
number of antiparticles. In these last two cases, try to obtain expressions for the pressure P in terms of
the density n ≡ N̄ /V and/or temperature T .

43
Appendix to Section 36: Fermi-Dirac and Bose-Einstein Integrals

This Appendix sets out various properties of the Fermi-Dirac (+) and Bose-Einstein (−) integrals
Z ∞ n−1
x dx
Fn± (α) ≡ . (A36.1)
0 ex+α ± 1

The derivative of the integrals Fn+ and Fn− is

∂Fn± (α) ±
= −(n − 1)Fn−1 (α) (n > 1) , (A36.2)
∂α
which can be proven by an integration by parts.
In the limit α → ∞, corresponding to the Boltzmann, or nondegenerate, limit, the integrals Fn+ and
Fn− go over to
Fn+ (α) = Fn− (α) = e−α Γ(n) (α → ∞) , (A36.3)
where Γ is the gamma function. Series expansions, good for α ≥ 0, are
!
−α e−2α e−3α
Fn+ (α) = Γ(n) e − n + n − ... ,
2 3
!
e−2α e−3α
Fn− (α) = Γ(n) e−α + n + n − ... . (A36.4)
2 3

In the opposite limit α → −∞, corresponding to the degenerate limit, the Fermi-Dirac integral Fn+
goes to
|α|n
Fn+ (α) = (α → −∞) . (A36.5)
n
The Bose-Einstein integral Fn− is of physical relevance only for α ≥ 0. An asymptotic expansion of the
Fermi-Dirac integral Fn+ , good for α < 0, is
 
∞
(−α)n 
n(n − 1)...(n + 1 − 2j)2(1 − 21−2j )ζ(2j)α−2j 
X
Fn+ (α) = 1+
n j=1
 
∞  2j
(−α)n  X n(n − 1)...(n + 1 − 2j) 2j π
= 1+ (2 − 2)|B2j | 
n j=1
(2j)! α
" #
(−α)n π2 7π 4
= 1 + n(n − 1) 2 + n(n − 1)(n − 2)(n − 3) + ... , (A36.6)
n 6α 360α4
where ζ(n) is the Riemann zeta-function
1 1
ζ(n) = 1 + + + ... , (A36.7)
2n 3n
and Bn are Bernoulli numbers. Equation (A36.6) is a particular case of the more general result, known
as Sommerfeld’s lemma, that for a function φ(x) that is sufficiently well-behaved, and regular at x = 0,
∞
d2j φ
Z ∞
(dφ/dx) dx
2(1 − 21−2j )ζ(2j)α−2j
X
= φ(−α) +
0 ex+α ± 1 j=1
dx2j x=−α

44
 d2 φ π2 d4 φ 7π 4
= φ(−α) + + + ... . (A36.8)
dx2 −α
6α2 dx4 −α
360α4

The case α = 0 corresponds to non-number-conserving particles, such as photons, or particle-

antiparticle pairs. A useful and easily proven relation between the Fermi-Dirac and Bose-Einstein integrals
for α = 0 is
Fn+ (0) = (1 − 21−n )Fn− (0) (n > 1) . (A36.9)
The Bose-Einstein integral Fn− (0) is
Fn− (0) = Γ(n)ζ(n) , (A36.10)
where ζ(n) is the Riemann zeta-function (A36.7). For even integral n,

(2π)n
Fn− (0) = |Bn | (n even) , (A36.11)
2n
where Bn are Bernoulli numbers. For example,

π2 π4 8π 6
F2− (0) = , F4− (0) = , F6− (0) = . (A36.12)
6 15 63

45