Unit 5

Ensembles in quantum statistical

physics.

5.1 Introduction to the quantum statistical physics

In quantum mechanics, it is not possible to specify simultaneously the coordinates
and momenta of a system of particles due to the Heisenberg’s uncertainty principle.
Now, the maximum information we can get from a quantum system is encoded in
its wavefunction |ψi (in the bra-ket notation). They are elements of a Hilbert space
with an inner product between two arbitrary wavefunctions |ψi and |χi which we will
denote as hψ| χi ≡ hχ|ψi∗ . The measurable quantities are represented as Hermitian
linear operators  ≡ † called observables, which act on the space of wavefunctions.
In a measurement of an observable, we can obtain an eigenvalue of its associated
operator, but now the outcome is not deterministic, but probabilistic since the
possible outcomes have a probability to appear which depend on the wavefunction
prior to the measurement (after it, the wavefunction collapses to the projection of
the original wavefunction on the eigenspace associated to the measured eigenvalue).
So, it is important to note that in quantum statistical physics we have two different
sources of the probabilistic description: one from the statistical physics postulates,
and the other one from the quantum description of the system. A quantity which
defines the typical value of an observable  at a given state |ψi can be the quantum
average, defined as hψ|Â|ψi.
An important operator in conservative quantum systems is the Hamiltonian Ĥ
since it determines the time evolution of the wavefunction through the Schrödinger

1
Ensembles in quantum statistical physics. 2

equation:
∂|ψi
= Ĥ|ψii~ (5.1)
∂t
where we are using the Schrödinger picture of the quantum mechanics. For the
quantum statistical physics, another important operator is the projector |ψihψ|.
The time evolution of this operator can be obtained easily from Eq. (5.1) as
∂|ψihψ|
i~ = Ĥ(|ψihψ|) − (|ψihψ|)Ĥ ≡ [Ĥ, |ψihψ|] (5.2)
∂t
where [Â, B̂] = ÂB̂ − B̂  is the commutator of the operators  and B̂.
From a practical point of view, it is convenient to represent wavefunctions and
operators by (infinite-dimensional) vectors and matrices. For this purpose, we use
complete set of orthonormal wavefunctions {|φn i}, which have the following prop-
erties: X
hφn |φm i = δn,m |φn ihφn | = Iˆ (5.3)
n

where δn,m is the Kronecker’s delta symbol, and Iˆ is the identity operator in the
Hilbert space. This set can be obtained as the set of eigenfunctions of observables,
such as the Hamiltonian. So, an arbitrary wavefunction and operator can be ex-
pressed as
X X
|ψi = hφn |ψi|φn i ≡ cn |φn i (5.4)
n n
X X
 = hφn |Â|φm i|φn ihφm | ≡ Anm |φn ihφm | (5.5)
n,m n,m

So, the coordinates {cn } and the matrix elements {Anm } define the wavefunction
|ψi and Â, respectively. Another representation can be obtained by considering the
following complete set of states for a system of N particles:
|rαi ≡ |r1 α1 i ⊗ . . . ⊗ |rN αN i (5.6)
where |ri αi i represents a state for the particle i to be at the position ri and with a
set of quantum numbers αi associated to internal degrees of freedom such as spin.
In this case, the representation of |ψi is
X Z XZ
|ψi = dr1 . . . drN ψα1 ,...,αN (r1 , . . . , rN )|rαi ≡ drψα (r)|rαi (5.7)
α1 ,...,αN α

In order to obtain the statistical physics of quantum systems, we proceed in a

similar way as we did in the classical case. Now, the wavefunction of a N -particles
Ensembles in quantum statistical physics. 3

systems, where N is of order of Avogadro’s number, encodes too much information of

the system, while macroscopically its state can be determined by a few parameters.
So, for each macrostate we will have a set of microstates (an ensemble) compatible
with the parameters which define the macrostate. The first postulate of the (Quan-
tum) Statistical Physics is that the macroscopic value of a physical quantity will be
average on the ensemble of the quantum averages of the associated observable in the
states of each element of the ensemble (which are different realizations of the same
macroscopic system):
Amacro = A = hψ|Â|ψi (5.8)
An important operator in the Quantum Statistical Physics is the density matrix ρ̂,
defined as
ρ̂ ≡ |ψihψ| (5.9)
Now, A can be expressed in terms of  and ρ. By using any representation or Â
and |ψi, Eq. (5.8) can be recast as
X X
A= c∗m Amn cn = Amn cn c∗m (5.10)
n,m n,m

On the other hand, the representation of ρ̂ is

ρnm = cn c∗m (5.11)

So, we can rewrite Eq. (5.10) as

!
X X X X
A= Amn ρnm = Amn ρnm = (Âρ̂)mm ≡ T r(Âρ̂) = T r(ρ̂Â) (5.12)
m,n m n m

where T r stands for the trace of the operator (which isR independent of the repre-
sentation). This expression is the quantum analogue of dqdpAρ which we used in
Classical Statistical Physics. So, the density matrix is the quantum version of the
probability density in Classical Statistical Physics. Indeed, by averaging Eq. (5.2)
on the ensemble, we get that the time evolution of ρ̂ is
∂ ρ̂
i~ = [Ĥ, ρ̂] (5.13)
∂t
which is the von Neumann equation. Note that this is the quantum version of
Liouville’s equation, since [Â, B̂]/(i~) converges to the Poisson’s bracket {A, B} in
the classical limit.
Ensembles in quantum statistical physics. 4

For equilibrium ensembles, ρ̂ is independent of time, so the left-hand side of

Eq. (5.13) vanishes. Thus, ρ̂ and Ĥ commute, and it is possible to find a common
set of eigenfunctions for both operators. In this representation, ρ̂ is diagonal, i.e.
ρnm = δn,m ρnn ≡ pn δn,m , and consequently
X
A= Ann pn (5.14)
n

So, pn can be regarded as the probability of the system to be in the eigenstate n of

both Ĥ and ρ̂. Now let’s consider a closed and isolated system, which corresponds to
the microcanonical ensemble. The second postulate of Quantum Statistical Physics
states that, under these conditions, all the microstates are equally probable (a priori
equal probability postulate). In the representation mentioned above, this means
than
1 X
ρnm = δn,m δR,n (5.15)
Ω R:E<E <E+∆E
R

where ∆E  E, i.e. pn = 1/Ω if E < En < E + ∆E, and zero otherwise. The
normalization condition
T r(ρ̂) = 1 (5.16)
implies that Ω is the number of microstates with energy between E and E+∆E. This
result is general for any representation. However, were we have used a general repre-
sentation, we would have obtain the same result only if we couple the equal a priori
principle with a random a priori phases postulate, i.e., that cn c∗m = ei(θn −θm ) /Ω = 0
if m 6= n. Thermodynamic properties can be obtained by using the expressions ob-
tained above, in a similar way as they are introduced in Classical Statistical Physics:
 
1 ∂ ln Ω
S = kB ln Ω , β ≡ = (5.17)
kB T ∂E X,N
   
1 ∂ ln Ω 1 ∂ ln Ω
Yα = , µi = − (5.18)
β ∂Xα E,Xγ6=α ,N β ∂Ni E,X,Nj6=i

The canonical ensemble is introduced in a similar way as in the classical case.

Consider a closed an isolated system which is composed by two closed subsytems A1
and A2 in thermal contact. The total Hamiltonian of the system can be expressed as
Ĥ = Ĥ1 + Ĥ2 + Ĥ12 . As in the classical case, Ĥ12 can be neglected with respect the
Ĥ1 and Ĥ2 . Under this approximation, the eigenstates of the Hamiltonian, which
satisfy Ĥ|φn i = En |φn i, can be written as

|φn i = |φn1 i ⊗ |φn2 i (5.19)

Ensembles in quantum statistical physics. 5

where Ĥ1 |φn1 i = En1 |φn1 i and Ĥ2 |φn2 i = En2 |φn2 i. On the other hand, the energies
of the total system can be written as

En = En1 + En2 (5.20)

Now, suppose that A2 has much more degrees of freedom than A1 . By summing
over the states of A2 , we find that the probability of finding the system A1 in an
energy state R of energy ER is

Ω2 (E − ER )
p1 (R) = (5.21)
Ω(E)

where Ω2 is the number of microstates of the system A2 . Now we Taylor expand Ω2

around ER = 0, and find the following expression for p1 :

e−βER
p1 (R) = (5.22)
Z
where the canonical partition function is written as
X
Z= e−βER (5.23)
R

This sum is over all the possible states of the system A1 . Alternatively, we may sum
over the possible energy levels, which in general have a degeneracy g(ER ). So,
X
Z= g(ER )e−βER (5.24)
ER

This corresponds to the probability of a state in the canonical ensemble, which

alternatively can be expressed in terms of the density matrix

e−β Ĥ1  
ρ̂ = , Z = T r e−β Ĥ1 (5.25)
Z
The thermodynamic properties of the system can be obtained in the same way as
in the classical case:
 
∂ ln Z

E=− , S = kB ln Z + βE (5.26)
∂β X,N
   
1 ∂ ln Z 1 ∂ ln Z
Yα = , µi = − (5.27)
β ∂Xα T,Xγ6=α ,N β ∂Ni T,X,Nj6=i
Ensembles in quantum statistical physics. 6

Finally, the description of a quantum system in the grand-canonical ensemble can

be obtained in a similar way as in Classical Statistical Physics. Now the probability
of an energy state R with energy ER and number of particles NR when the system
is in contact with a particles and thermal bath characterized by a temperature T
and parameters α is
e−αNR −βER
p(R) = (5.28)
Q
where the grand-canonical partition function Q has the expression
∞
X X ∞
X
−αNR −βER
Q= e e ≡ e−αNR Z(T, X, N ) (5.29)
N =0 R:NR =N N =0

This energy state probability is consistent with a density matrix

e−αN̂ −β Ĥ  
ρ̂ = , Q = T r e−αN̂ −β Ĥ (5.30)
Q
The thermodynamic properties of the system can be obtained in the same way as
in the classical case:
   
∂ ln Q ∂ ln Q
E=− , Ni = − (5.31)
∂β X,α ∂αi T,X,αj6=i
!  
X 1 ∂ ln Q
S = kB ln Q + βE + αi N i , Yγ = (5.32)
i
β ∂Xγ T,Xδ6=γ ,α
P
Finally, ln Q = β γ Y γ Xγ , which for hydrostatic systems, reduces to

βpV = ln Q (5.33)

where p is the pressure.

5.2 Indistinguishability of particles

The indistinguishability of particles in quantum physics is a inherent property of
identical particles. Unlike in Classical Mechanics, where we can track the identity of
the particles through their time evolution, in quantum physics this is not true. The
reason for this fact is that, when two particles collide, by the uncertainty principle
we loose information of their identities in the collision process, so in the outcome
Ensembles in quantum statistical physics. 7

we do not know the particles identities. Thus, it is not important which particle is
in which state, but how many particles there are in each state.
The indistinguishability introduces constraints on the wavefunctions the N −particles
system. If we denote ξi = ri αi , the state of the total system is represented by the
wavefunction ψ(ξ1 , . . . , ξN ). Now, if we permute two particles, e.g. the particles i
and j, the state of the system must not change, in the sense that all the measurable
quantities are the same. This means that the wavefunction must change only by a
phase factor:
ψ(. . . , ξj . . . , ξi , . . .) = eiα ψ(. . . , ξi , . . . , ξj , . . .) (5.34)
If we permute again these particles, we get that

ψ(. . . , ξi . . . , ξj , . . .) = e2iα ψ(. . . , ξi , . . . , ξj , . . .) (5.35)

Thus, e2iα = 1 and eiα = ±1. So, identical particles in quantum physics fall into
two categories:

• Bosons: the wavefunctions are symmetric under a permutation of particles:

ψ(. . . , ξi . . . , ξj , . . .) = +ψ(. . . , ξi , . . . , ξj , . . .) (5.36)

• Fermions: the wavefunctions are antisymmetric under a permutation of par-

ticles:
ψ(. . . , ξi . . . , ξj , . . .) = −ψ(. . . , ξi , . . . , ξj , . . .) (5.37)

All elementary particles in Nature are either bosons or fermions. The spin-statistics
theorem due to Pauli states a connection between the spin of the particle and the
statistics which obeys: particles with half-integer spin are fermions, and particles
with integer spin are bosons. The proof of this theorem is beyond the scope of the
course, and it is based on the analysis of relativistic quantum field theory. There are
non-relativistic “proofs” based on the connection of particles exchange and rotations,
but they are flawed, in the sense that it is possible to obtain symmetric states for
fermions and viceversa. Finally, we must mention that in strong-correlated electrons
phenomena in Condensed Matter Physics as fractional quantum Hall effect involv-
ing quasi-particles associated to elementary excitations which are neither bosons or
fermions, and which are named anyons. This is also beyond the scope of this course.
The antisymmetry of the wavefunctions for fermions have as a consequence the
Pauli’s exclusion principle. If we choose that ξi = ξj , then Eq. (5.37) leads to

ψ(. . . , ξ . . . , ξ, . . .) = −ψ(. . . , ξ, . . . , ξ, . . .) (5.38)

Ensembles in quantum statistical physics. 8

so
ψ(. . . , ξ . . . , ξ, . . .) = 0 (5.39)
i.e. two fermions cannot be at the same point and the same internal state at the
same time. This can be extended for any representation of the wavefunction.
Possible wavefunctions for bosons and fermions are restricted to those which are
either symmetric or antisymmetric with respect to their arguments, respectively.
This has a consequence on the statistical properties of quantum systems of bosons
of fermions. In particular, in general the procedure we used to take into account the
indistinguishability in Classical Statistical Physics will be no longer valid. We will
discuss this issue in more detailed way for ideal systems in the next Sections.

5.3 Ideal systems

Now we are going to consider ideal systems composed by identical particles of only
one species where the interaction between them can be neglected. In this case, the
eigenstates of the total energy and number of particles can be obtained from the
eigenstates of the single-particle Hamiltonian ĥ as

ĥ|φr i = r |φr i (5.40)

If particles were distinguishable, the eigenfunctions of the Hamiltonian Ĥ = N

P
i=1 ĥi
are
|Φr1 ,...,rN i = |φr1 i ⊗ . . . ⊗ |φrN i (5.41)
where |φri i is the single-particle energy eigenfunction with energy ri of the particle
i. The energy associated to |Φr1 ,...,rN i is given by

Er1 ,...,rN = r1 + . . . + rN (5.42)

Now, we take into account the indistinguishability of the particles. This means that
the space of eigenfunctions is reduced with respect to the distinguishable case. For
example, the subspace spanned by the linear independent eigenfunctions in the set
|ΦP(r1 ,...,rN ) i, where P(r1 , . . . , rN ) is a permutation of r1 , . . . , rN (all of them with
the same total energy E) is replaced for bosons by a single eigenfunction, which is
obtained by symmetrization of the set of eigenfunctions for the distinguishable case.
For fermions, on the other hand, they are replaced by a single eigenfunction, which is
obtained by the antisymmetrization of the distinguishable eigenfunctions (the Slater
determinant) if all the values of ri are different. However, these states are forbidden
by the Pauli’s exclusion principle if there are at least two equal values of ri . For
example, for a two-particles system, the possible states are |φr1 i ⊗ |φr2 i and |φr2 i ⊗
Ensembles in quantum statistical physics. 9

|φr1 i if r1 6= r2 , and otherwise |φr1 i⊗|φr1 i. The set of possible normal states spanned √
by the two eigenfunctions for r1 6= r2 reduce to (|φr1 i ⊗ |φr2 i ± |φr2 i ⊗ |φr1 i)/ 2
for bosons (+) and fermions (−). On the other hand, if r1 = r2 , |φr1 i ⊗ |φr1 i is an
admissible state for bosons, but it is forbidden for fermions. To represent the set of
eigenfunctions of the total energy which preserve the symmetry or the antisymmetry
under particles permutation, we use as quantum numbers the occupation numbers
nr of each single-particle state r. So, these states will be denoted by |n1 , n2 , . . .i. The
energy of number of particles associated to these states are given by the expressions
X X
ER ≡ En1 ,n2 ,... = nr r , NR = nr (5.43)
r r

where we used R to identify theQstates of the total system. We note that each of
these states correspond to N !/ r nr ! states of a distinguishable system. Thus, if
we divide this expression by N !, which was the procedure to obtain the number of
independent states of a system of indistinguishable particles from the distinguishable
case in Classical Statistical Physics, we only get the correct expression if all the
values of nr are either 0 or 1. Otherwise, we underestimate the number of states
for bosons, and overestimate them for fermions, since ni can take any non-negative
integer value for bosons, but for fermions nr can only be either 0 or 1. Thus, the
main differences between the Classical and Quantum Statistical Physics will emerge
when states with multiple occupancy in single-particle states are relevant for the
evaluation of the statistical properties.
We start evaluating the canonical partition function:
N
X XX Y
Z= e−βER = . . .0 e−βr nr (5.44)
R {n1 } {n2 } r

where the prime denotes that the sums over the possible occupation numbers of
the single-particle energy states are subject to the constraint n1 + n2 + . . . = N .
This fact complicates the evaluation of the statistical properties in the canonical
ensemble. However, we can evaluate them by taking into account the equivalence
between equilibrium ensembles. So, we can solve the problem in the grand-canonical
ensemble, where the chemical potential is tuned in order to assure that N = N . The
grand-canonical partition function can be obtained as
∞
X X XX Y
Q= e−αN e−βER = ... e(−α−βr )nr (5.45)
N =0 R {n1 } {n2 } r

where now there are no restrictions for the sums. We can exchange the order of
Ensembles in quantum statistical physics. 10

sums and products as  

Y X
Q=  e(−α−βr )nr  (5.46)
r {nr }

Now, we can evaluate the thermodynamic properties of these systems in terms of

the mean occupation number or distribution nr , which can be obtained as
P P
−βER −αNR −β r r nr,R −α r nr,R
P P
R nr,R e n r,R e −1 ∂Q
nr = = R =
Q Q βQ ∂r
P 
−(α+βr )nr
−1 ∂ ln Q −1 ∂ ln {nr } e
= = (5.47)
β ∂r β ∂r
From Eq. (5.43), we get that
X X
E= r nr , N =N = nr (5.48)
r r

For hydrostatic systems, βpV = ln Q. Alternatively, we may use that βp = ∂ ln Q/∂V .

The only terms which depend on V are the energies r . For example, the energies of
a particle confined inside in a rectangular parallelepiped of dimensions Lx , Ly and
Lz are: " 
2  2  2 #
~2 π 2 mx my mz
r ≡ mx ,my ,mz = + + (5.49)
2m Lx Ly Lz
with mx , my , mz = 1, 2, 3, . . .. If the box is a cube, then Lx = Ly = Lz = L and
V = L3 , so
~2 π 2  2 2 2
mx ,my ,mz = (mx ) + (my ) + (m z ) (5.50)
2mV 2/3
So, the pressure is
 
1 ∂ ln Q X 1 ∂ ln Q ∂r X ∂r
p= = = − nr (5.51)
β ∂V r
β ∂r ∂V r
∂V

For the ideal gas in a cubic box of length L,

∂r ~2 π 2  2 2 2 2r
=− (mx ) + (my ) + (mz ) = − (5.52)
∂V 3mV 5/3 3V
So,
X 2r 2
pV = nr = E (5.53)
r
3 3
Ensembles in quantum statistical physics. 11

Note that this expression is valid for the classical ideal gas.
Finally, although it is not easy to obtain the exact expression for the canonical
partition function, we can use the equivalence between equilibrium ensembles to get
its asymptotic behaviour. As we discussed in the classical case,
ln Q ≈ ln Z(N ) − αN → ln Z = ln Q + αN (5.54)
for the value of α corresponding to the condition N = N .
Now, we are going to obtain the mean occupation number for bosons and fermions.
For fermions, the values of nr are 0 or 1, and then
1
X
e−(α−βr )nr = 1 + e−α−βr (5.55)
nr =0

So, X
ln Q = ln(1 + e−α−βr ) (5.56)
r

Substituting this expression in Eq. (5.47) we get that

e−α−βr 1
nr = −α−β
= β(r −µ) (5.57)
1+e r e +1
where we have used that α ≡ −βµ. This is the Fermi-Dirac distribution (see Fig.
5.1). Note that for   µ, nr ≈ 1, while for   µ, nr ≈ 0. The Fermi-Dirac
distribution takes intermediate values between 0 and 1 on a range of values of 
of order of kB T . At  = µ nr = 1/2. In this context, the chemical potential µ is
also called the Fermi level. For T = 0, the Fermi-Dirac distribution becomes a step
function, and in this case µ(T = 0) ≡ µ0 is called the Fermi energy.
Now we turn to the bosonic case. The values of nr are all the non-negative
integers, and then
∞
X 1
e−(α−βr )nr = −α−β
(5.58)
n =0
1−e r
r

for  > µ, but diverges otherwise. This means that the values of µ must be smaller
than 0 , where 0 is the single-particle ground state energy. So,
X
ln Q = − ln(1 − e−α−βr ) (5.59)
r

Substituting this expression in Eq. (5.47) we get that

e−α−βr 1
nr = −α−β
= β(r −µ) (5.60)
1−e r e −1
Ensembles in quantum statistical physics. 12

4.5 Fermi-Dirac distribution

Bose-Einstein distribution
4 Maxwell-Boltzmann distribution

3.5
3
nr

2.5

2
1.5
1

0.5
0
βµ−4 βµ−2 βµ βµ+2 βµ+4
βε

Figure 5.1: Fermi-Dirac, Bose-Einstein and Maxwell-Boltzmann distributions.

where we have used that α ≡ −βµ. This is the Bose-Einstein distribution (see Fig.
5.1). Note that it diverges as  → µ+ .
For hydrostatic systems, ln Q = βpV . Since
X
ln Q = ± ln(1 ± e−α−βr ) (5.61)
r

where the positive sign stands for fermions, and the negative for bosons, we get that
kB T X
p=± ln(1 ± e−α−βr ) (5.62)
V r

5.4 The quantum ideal gas

Let’s suppose an ideal gas of particles with internal degrees of freedom which are
not coupled with the translational Hamiltonian. So, the single-particle states can
be expressed as φ(r)χ(α), where α stands for the internal degrees of freedom as,
for example, spin (in this case, g = 2s + 1, where s is the particle spin). The
single-particle energy levels arise only from the translational contribution, so we
Ensembles in quantum statistical physics. 13

will denote by g their degeneracy from internal degrees of freedom. Now, the single-
particle Schrödinger equation is

~2
 2
∂2 ∂2

∂
− + + φ = φ (5.63)
2m ∂x2 ∂y 2 ∂z 2

which must be solved in a rectangular parallelepiped of sides Lx , Ly and Lz . As the

wavefunction must vanish outside this volume, wavefunction continuity means that
φ must vanish on the boundary. The translational part of a particle eigenfunction
has the form
     
mx πx my πy mz πz
φmx ,my ,mz (x, y, z) = A sin sin sin (5.64)
Lx Ly Lz

where A is a normalization factor and mx , my and mz are positive non-zero integer

numbers. The energy associated to this eigenfunction is
"   2  2 #
2
~2 π 2 mx my mz
r ≡ mx ,my ,mz = + + (5.65)
2m Lx Ly Lz

In the thermodynamic limit, N , Lx , Ly and Lz diverge, but N/V → n finite, where

V = Lx Ly Lz . Thus, the gaps between nearest-neighbour levels decrease, so the
single-particle energy spectrum becomes the continuum set (0, +∞) in the thermo-
dynamic limit. For large V , we can approximate the sums over single-particle energy
states by an integral on energy as
X Z ∞
... ≈ dD() . . . (5.66)
r 0

where D() is the density of states (DOS). First, we note that the translational
states are not eigenfunctions of the linear momentum operator, but eigenfunctions
of the absolute values of its components, i.e. |px |φ = ~mx πφ/Lx and so on. This is
reasonable, since the translational part of the eigenfunctions are stationary waves,
which are obtained by superposition of two travelling waves with opposite momenta
(corresponding to the physical picture of a particle bouncing back and forward
from the walls). So, how many translational states have an absolute value for the
i−component of the linear momentum between |pi | and |pi | + ∆|pi |? This is related
to the values of mi :

D(|px |, |py |, |pz |)∆|px |∆|py |∆|pz | = D(mx , my , mz )∆mx ∆my ∆mz (5.67)
Ensembles in quantum statistical physics. 14

Now, D(mx , my , mz ) = 1, since the values of (mx , my , mz ) are the nodes of a simple
cubic lattice of spacing 1. On the other hand, ∆mi = Li ∆|pi |/~π. So,
Lx Ly Lz 8V
D(|px |, |py |, |pz |) = 3
= 3 (5.68)
(~π) h
where h = 2π~ is the Planck’s constant. Now, the distribution of the translational
states with linear momentum modulus p is obtained as
Z
D(p)∆p = √ d|px |d|py |d|pz |D(|px |, |py |, |pz |)
p≤ p2x +p2y +p2z ≤p+∆p
Z
8V
= √ d|px |d|py |d|pz |
h3 p≤ p2x +p2y +p2z ≤p+∆p

V
≈ 3
4πp2 ∆p (5.69)
h
where the integration is restricted to an octant of an spherical shell or radius p
and width ∆p  p. This expression, although it has been obtained for specific
geometry and boundary conditions, are valid asymptotically (i.e. up to extensive
terms proportional to V ) for any container shape and other boundary conditions.
Finally, in order to get D(), we note that  = p2 /(2m) and that there are g states
for each translational state. So
√
√ 2m3 √
 
 dp  4πV
D() = gD(p = 2m)   = g  (5.70)
d h3
With this expression
Z ∞ Z ∞
D()d D()d
N= , E = (5.71)
0 eβ(−µ) ± 1 0 eβ(−µ) ± 1
Z ∞
2E
D() ln 1 ± e−β(−µ) d =


βpV = ± (5.72)
0 3kB T
where the + (−) sign stand for fermions (bosons), and the last equality is obtained
by integration by parts.

5.5 The classical limit: the Maxwell-Boltzmann

statistics
Now we would like to know in which limit we recover the classical limit. First,
we note that both the Fermi-Dirac and the Bose-Einstein distributions, for large ,
Ensembles in quantum statistical physics. 15

behave as
1
nr = ≈ e−β(r −µ)  1 (5.73)
eβ(r −µ)
±1
If we assume this is valid for any r , we get the Maxwell-Boltzmann statistics,
which indeed converges to both Fermi-Dirac and Maxwell-Boltzmann statistics for
 − µ > 2kB T (see Fig. 5.1).
Now this approximation will be valid for any  > 0 for high temperatures and/or
low densities. For a fixed temperature and number of particles, we have seen for the
ideal gas that, as the volume increases, the number of energy states which are in a
fixed range of energies, e.g. between 1 and 1 + ∆, also increases. Here, ∆  1
but much larger than the gaps between consecutive energy levels. So, if volume is
large enough, the number of states can exceed the number of particles, and as
X
N= nr (5.74)
r

then nr  1 in this range. But as 1 is arbitrary, this is valid for any . The way to
get this is by shifting µ to values much smaller than the single-particle ground state
0 . Alternatively, by fixing N and V , nr increases as β decreases (fixing µ). So, if
temperature is high enough, the mean number of particles with energy between 1
and 1 +∆ can exceed N . The only way to avoid that is shift µ to smaller values. In
particular, this will be valid if µ  0 , so again the Maxwell-Boltzmann distribution
will be valid for all the values of .
Now, we are going to see the values of the thermodynamic properties in the
Maxwell-Boltzmann statistics. The number of particles is given by
X X
N =N = nr ≈ e−α e−βr = zζ (5.75)
r r

where z is the fugacity and ζ the single-particle canonical partition function. Note
that this is the expression we obtained in the classical grand-canonical ensemble.
On the other hand
X X
βpV = ln Q = ± ln(1 ± e−α−βr ) ≈ e−α−βr = zζ = N (5.76)
r r

where we used that exp(−α − βr )  1 and that ln(1 + x) ≈ x for small x. So, we
recover the equation of state of the ideal gas, pV = N kB T , as in the classical case.
Finally, in this case it is possible to obtain exactly the canonical partition function
in the Maxwell-Boltzmann statistics. Since
∞ ∞
zζ
X (zζ)N X
Q=e = = z N Z(N ) (5.77)
N =0
N ! N =0
Ensembles in quantum statistical physics. 16

we get that
ζN
Z= (5.78)
N!
which is the classical canonical partition function for indistinguishable particles.
For a spinless monoatomic ideal gas, we will recover the classical expression.
First, we have to evaluate ζ for large V . In general
∞
! ∞ 2 π2

∞
!
2 2
X − β~ π m2x β~
X − 2 m2y β~2 π 2
X X − m2
ζ= e−βr = e 2mL2x  e 2mLy  e 2mL2z z (5.79)
r mx =1 my =1 mz =1

Let’s focus on one of the factors in the right-hand side. The gap between two
consecutive exponents is ∆, which is
β~2 π 2
∆= (2mx + 1) ≈ 10−20 (2mx + 1) (5.80)
2mL2x
for room temperature, Lx ∼ 0.1 m and m ∼ 10−25 Kg. On the other hand, assuming
that the system obeys Maxwell-Boltzmann statistics, equipartition dictactes that
 ∼ kB T , so mx ∼ 1010 . Thus ∆ ∼ 10−10  1. So, the relative difference of two
terms in the sum with respect to one of them is given by 1 − exp(−∆) ≈ ∆  1.
Under these circumstances,
∞ 2 2 Z ∞ 2 2
r r
X − β~ π2 m2x − β~ π2 m2x Lx 2πm L x 2πm Lx
e 2mLx ≈ dmx e 2mLx = = = (5.81)
m =1 0 2π~ β h β Λ
x

where Λ is the thermal de Broglie length. Repeating this calculation for the other
factors, we get finally that
Lx Ly Lz V
ζ= 3
= 3 (5.82)
Λ Λ
which is the classical expression, where now h can be identified as the Planck’s
constant. This result can also be obtained from the expression
Z ∞
ζ≈ dD()e−β (5.83)
0

We have previously argued that the Maxwell-Boltzmann statistics is valid for

low densities and high temperatures. We are going to provide a more quantitative
criterion for ideal gases. Since e−α−βr  1 for all r , this must be also valid for  = 0.
So, in the Maxwell-Boltzmann limit e−α = z  1. For the Maxwell-Boltzmann
statistics, z = N/ζ, and for the ideal gas, z = N Λ3 /V = nΛ3 . So, Maxwell-
Boltzmann statistics will be valid if y ≡ nΛ3  1. y is called the degeneracy
Ensembles in quantum statistical physics. 17

parameter, and note that for the ideal gas, this parameter is small for low densities
or low values of Λ, which corresponds to high temperatures.
The condition nΛ3 can be physically rationalized in two ways. First, we argued
that Maxwell-Boltzmann statistics is valid if the number of available single-particle
states is much larger than the number of particles, so multiple occupancy is statis-
tically infrequent. We can estimate this in the following way. The phase volume
associated to one particle is V p3 ∼ V (2mkB T )3/2 , where we have estimated that for
typical occupied states,  = p2 /2m ∼ kB T . Thus, the number of quantum states can
be obtained semi-classically by dividing the phase volume of a particle by h3 . So,
Maxwell-Boltzmann statistics is valid if V (2mkB T )3/2 /h3  N , which is equivalent
to nΛ3  1. An alternative explanation is based on particles being represented by
wave packets with size λ, which is related with the linear momentum
√ of the particle
through Heisenberg’s principle λ ∼ h/p. Typically p ∼ mkB T , so λ is approxi-
matelly equal to the thermal de Broglie wavelength Λ. Now, Maxwell-Boltzmann
statistics is valid if the wave packets associated to different particles do not overlap.
As the particle density of the gas is uniform, the typical distance between close par-
ticles is d, which is obtained from the condition N/V ∼ 1/d3 , i.e. d ∼ n−1/3 . Thus,
the non-overlapping condition means that d  Λ, which again reduces to nΛ3  1.
For weakly degenerate systems, i.e. those with small y, we can introduce sys-
tematically corrections with respect to the Maxwell-Boltzmann expressions for the
thermodynamic properties as power expansions of y. For example, the first cor-
rection beyond the Maxwell-Boltzmann limit can be obtained from the following
approximation for the grand-canonical partition function
X e2(−α−βr )
X 
−α−βr −α−βr
ln Q = ± ln(1 ± e )≈ e ∓ (5.84)
r r
2

where we have used that ln(1 + x) ≈ x − x2 /2 for small x. If we denote ζ(T ) as the
single-particle canonical partition function of the system for a temperature T , this
expression can be recast as

e−2α ζ(T /2)

ln Q ≈ e−α ζ(T ) ∓ (5.85)
2
In the thermodynamic limit of an ideal gas, we can evaluate these sums as integrals,
in a similar way we did in the Maxwell-Boltzmann limit. So,

e−2α
 
V −α
ln Q ≈ 3 e ∓ 5/2 (5.86)
Λ 2
where Λ is the thermal de Broglie length for temperature T . The number of particles
Ensembles in quantum statistical physics. 18

satisfies the condition

e−2α
 
∂ ln Q V −α
N =N =− = 3 e ∓ 3/2 (5.87)
∂α Λ 2

This is a quadratic equation for z = e−α , which has as solution

√ √ y2
 q 
−α
e = ∓ 2 −1 + 1 ∓ 2y ≈ y ± 3/2 (5.88)
2

where y = nΛ3 (the other branch leads to large values of z or negative unphysical
values). The mean energy and pressure of the gas have the following expressions

e−2α
 
∂ ln Q 3 V −α
E=− = kB T 3 e ∓ 5/2 (5.89)
∂β 2 Λ 2

and
e−2α
 
V −α
pV = kB T 3 e ∓ 5/2 (5.90)
Λ 2
Substituting Eq. (5.88) in Eqs. (5.89) and (5.90), and keeping terms of order of y 2 ,
we get that
3  y   y 
E = N kB T 1 ± 5/2 , pV = N kB T 1 ± 5/2 (5.91)
2 2 2
where the positive (negative) sign stands for fermions (bosons). Finally, other quan-
tities can be obtained in a similar way
   
∂E 3  y  5 y
CV = = N kB 1 ∓ 7/2 , S = N kB − ln y ± 7/2 (5.92)
∂T N,V 2 2 2 2

5.6 Bibliography
1. J. Javier Brey Abalo, J. de la Rubia Pacheco y J. de la Rubia Sánchez,
Mecánica Estadı́stica (UNED, 2011).

2. F. Reif., Fundamentals of Statistical and Thermal Physics (Waveland Pr. Inc.,

2008).

3. P.K. Pathria and P.D. Beale, Statistical Mechanics, 3rd edition (Elsevier,
2011).