TABLE I: Road Map of Ensembles

ensemble macrostate probability distribution thermodynamics

microcanonical E, V, N pi = 1/Ω S(E, V, N ) = kln(Ω)

canonical T, V, N pi = exp(−βEi )/Z F (T, V, N ) = −kT ln(Z)

grandcanonical T, V, µ pi = exp(−β(Ei − γNi )/Z Φ(T, V, µ) = −kln(Z)

The Grand Canonical Ensemble

POCKET BOOK REFERENCE: GRAND CANONICAL ENSEMBLE

Canonical ensemble: System and a (very large) heat bath/reservoir in equilibrium. Fixed
T, V, N.

The probability of finding the system in a microstate i with energy Ei :

Ei −µNj
e− kT
pij = (Gibbs distribution).
Z

Canonical partition function:

X Ei −µNj
Z(T, V, µ) = e− kT .
ij

Grand potential of the canonical ensemble:

Φ(T, V, µ) = U − T S − µN = −kT lnZ(T, V, µ),

GRAND CANONICAL ENSEMBLE

Consider a system which is coupled to a heat and particle reservoir, that is it can exchange
energy and particles with its environment. Suppose that that the energy and particle number
of the system is much smaller than that of the reservoir, NS ≪ NR and ES ≪ ER . Consider
an ensemble of N such systems and let nij is the number of systems that have energy Ei and
Nj particles. Using the Lagrange multiplier method, we can determine the most probable
distribution set, n∗ij :

1
 TABLE II: Ensemble distributions with a given particle number and energy.
Number of ensembles Energy Number of particles

n11 E1 N1
n12 E1 N2
n21 E2 N1
. . .

Each distribution (see the Table) must satisfy the following constrains:
X X X
nij = N Ei nij = N hEi Nj nij = N hNi
ij ij ij

There are two indeces since neither the energy not the number of particles are the same
in all systems. The probability of the distribution {nij } is proportional to the number of
ways, W {nij }, the given distribution can be formed. The total number of ways W {nij } to
generate a certain distribution {nij } is
N!
W {nij } = Q . (1)
ij nij !

The most probable distribution is the one which maximizes the weight factor W {nij }.
To find the maxima, again, it is simpler to work with ln(W ):
X
ln(W ) = ln(N !) − ln(nij !) (2)
ij

We are in the N → ∞ and ni → ∞ regime, so we can use the Stirling formula.

X
ln(W ) = N ln(N ) − nij ln(nij ) (3)
ij

To determine the maxima of ln(W ) subject to the constrains we use the Lagrange multiplier
technique. First we form a functional by adding the constrains multiplied by Lagrange
multipliers to ln(W )
X X X
L[nij , α, β, γ] = ln(W ) + α(N − nij ) + β(N hEi − nij Ei ) − γ(N hNi − nij Nj )
ij ij ij

To find the minima subject to the constrains one has to require that
∂L ∂L ∂L ∂L
= 0, = 0, = 0, = 0.
∂ni ∂α ∂β ∂γ

2
simultaneously fulfilled. The last three conditions are nothing but the constrains, while
using the first condition we obtain

−ln(nij ) − 1 − α − βEi − γNj = 0,

that is
n∗ij = Ce−βEi −γNj .

From the first constraint we obtain

1 −βEi −γNj
pij = e ,
Z
where
X
Z= e−βEi −γNj ,
ij

is the grand canonical partition function. Physical quantities:

1 X ∂ln(Z)
hEi = Ei e−βEi −γNi = − ,
Z ij ∂β

1 X ∂ln(Z)
hNi = Nj e−βEi −γNi = − ,
Z ij ∂γ

1 X ∂Ei −βEi −γNi 1 ∂ln(Z)

hP i = e =
Z ij ∂V β ∂V
The differential of the grand partition function is then
∂ln(Z) ∂ln(Z) ∂ln(Z)
dln(Z) = dβ + dγ + dV = −hEidβ − hNidγ + βhP idV.
∂β ∂γ ∂V
By adding d(βhEi + γhNi) to both sides we have

d (ln(Z) + βhEi + γhNi) = βdhEi + γdhNi + βhP idV,

or by reordering
1 γ
dhEi = d (ln(Z) + βhEi − γhNi) + dhNi − hP idV,
β β
Comparing that to the thermodynamic equation

dE = T dS − P dV + µdN

we identify
1 µ hEi µhNi
β= γ=− S = kln(Z) + − .
kT kT T T

3
Using the last equation we can also express the grand potential as

Φ = E − T S − µN = −kT ln(Z).

We can also write Z in a different form using

X Ei −µNj X µNj X Ei X µ
Nj
Z(T, V, µ) = e− kT = e kT e− kT = e kT Z(T, V, Nj ),
ij j i j

or by introducing the fugacity

µ
z = e kT

we have
X
Z(T, V, µ) = z Nj Z(T, V, Nj ).
j

In case of noninteracting indistinguishable particles, using

1
Z(T, V, N) = Z(T, V, 1)N ,
N!

we can write the grand partition function as

∞ ∞
X X 1 N
Z(T, V, µ) = z N Z(T, V, N) = z Z(T, V, 1)N = exp{zZ(T, V, 1)}.
N =0 N =0
N!

Example
Describe the ideal gas in grandcanonical ensemble!

Solution
For an ideal gas, the single particle canonical partition function is
r
V h2
Z(T, V, 1) = 3 λ= .
λ 2πmkT

Therefore
V
Z(T, V, µ) = exp{zZ(T, V, 1)} = exp{z }
λ3
The grand potential is
V
Φ = −kT ln(Z) = −kT z
λ3
The equations of state:
 
∂Φ V 5 µ
S(T, V, µ) = =z k( − ),
∂T V,µ λ3 2 kT

4
 state # N ǫn

1 no electron 0 −∆

1 electron up 1 0

1 electron down 1 0

 
∂Φ 1
p(T, V, µ) = = kT z 3 ,
∂V T,µ λ
 
∂Φ V
N(T, V, µ) = =z 3
∂µ T,V λ
Example
Many impurity atoms in a semiconductor exchange energy and electrons with the electrons in
the conduction band. Consider the impurity atoms to be in thermal and chemical equilibrium
with the conduction band, which can be considered to be an energy and particle reservoir.
Assume that ∆ is the ionization energy of the impurity atom. Find the probability that an
impurity atom is ionized.

Solution
Suppose that one and only one electron can be bound to an impurity atom. Because an
electron has a spin, both spin orientations up and down are possible. An impurity atom has
three allowed states: state 1 without an electron (atom ionized), state 2 with an electron
attached with spin up , and state 3 with an electron attached with spin down. We take the
zero of energy to correspond to the two bound states. The microstates of the system are
summarized in the Table:
The grand partition function of the impurity atom is

Z = eβ∆ + 2eβµ (4)

The probability that the atom is in state #1 (ionized)

eβ∆
P = (5)
eβ∆ + 2eβµ

Example
Consider an ideal gas of volume V , with N particles initially. The gas is exposed to a
clean surface, with no particles bound initially. Both the surface and the gas are free to

5
exchange energy with a reservoir of temperature T . The surface has N ≪ M distinguishable
non-interacting sites which can bind a gas molecule by a van der Waals interaction. Model
the gas when bound as a harmonic oscillator, and derive an expression for the fraction of
particles remaining at equilibrium in the gas at temperature T . Discuss high and low
temperature limits.

Solution
The number of gas atoms is Ng and the number of surface atoms is Ns , the total number of
particles is constant, N = Ns + Ng . The chemical potential of the ideal gas is
Ng 3
µg = kT ln( λ)
V
Now we turn to the writing down the grand canonical partition function for the gas molecules
on the surface. We assume that each surface site holds either zero or one particle. For the
N = 0 particle case, the energy is zero, ǫ = 0. For the N = 1 case, there are energies ǫj
available to the harmonic oscillator energy levels. The grand canonical partition function
for one site is
X ǫj
Z(T, V, µ) = 1 + e− kT z = 1 + Zz,
j

where Z is the canonical partition function of the single particle harmonic oscillator
e−β~ω/2
Z=
1 − e−β~ω
Since we are assuming the sites are distinguishable, the M-site partition function is

Z(T, V, µ) = (1 + zZ)M ,

and the grand potential is then

Φ(T, V, µ) = −kT Mln(1 + zZ).

We can now calculate how the number of particles at the surface depends on the surface
chemical potential:  
∂Φ zZ
Ns = N(T, V, µ) = =M
∂µ T,V 1 + zZ
Inverting this, we can infer the chemical potential for the particles at the surface as a function
of the temperature and volume:
 
1 Ns
µs = kT ln
Z M − Ns

6
We have two equations:
the particles are conserved
Ns + Ng = N,

at equilibrium
1 Ns Ng 3
µs = µg → = λ,
Z M − Ns V
that is
V N − Ng
= Ng .
Zλ3 M − N + Ng
Because N ≪ M we have N − Ng ≪ M and with this
V N − Ng
= Ng
Zλ3 M
The number of gas particles at equilibrium:
VN
M Zλ3
Ng =
1 + M VZλ3

OCCUPATION NUMBERS AND QUANTUM STATISTICS

We now develop the formalism for calculating the thermodynamic properties of ideal
quantum systems. The absence of interactions between the particles of an ideal gas enables us
to reduce the problem of determining the energy levels Es of the gas as a whole to determining
ǫk , the energy levels of a single particle. Because the particles are indistinguishable, we
cannot specify the microstate of each particle. Instead a microstate of an ideal gas is specified
by the occupation numbers nk , the number of particles in each of the single particle energies
ǫk . If we know the value of the occupation number for each state, we can write the total
energy of the system as
X
Es = nk ǫk
k
The set of nk completely specifies a microstate of the system. The partition function for an
ideal gas can be expressed in terms of the occupation numbers as
X P
Z(V, T, N) = e−β k nk ǫ k
,
{nk }

where the occupation numbers satisfy the condition

X
N= nk
k

7
The grand partition function is
∞
X
Z(V, T, µ) = z N Z(V, T, N)
N =0

Now let’s define the grand partition function Zk for (all) the particles in the kth single
particle state:
X
Zk (V, T, µ) = e−βnk (ǫk −µ) ,
nk

The grand partition function, for non interacting particles is

Y
Z(V, T, µ) = Zk
k

where the product is taken over the possible energy levels. Simple calculation shows that
the above two definitions of the partition function are identical.
A fundamental result of relativistic quantum mechanics is that all particles can be clas-
sified into two groups. Particles with zero or integral spin such as 4 He are bosons and have
wave functions that are symmetric under the exchange of any pair of particles. Particles
with half-integral spin such as electrons, protons, and neutrons are fermions and have wave
functions that are antisymmetric under particle exchange. The Bose or Fermi character of
composite objects can be found by noting that composite objects that have an even num-
ber of fermions are bosons and those containing an odd number of fermions are themselves
fermions. For example, an atom of 3 He is composed of an odd number of particles: two
electrons, two protons,and one neutron each of spin 1/2. Hence, 3 He has half-integral spin
making it a fermion. An atom of 4 He has one more neutron so there are an even number of
fermions and 4 He is a boson.
The difference between fermions and bosons is specified by the possible values of nk . For
fermions we have
nk = 0 or 1. (fermions).
The restriction states the Pauli exclusion principle for noninteracting particles two identi-
cal fermions cannot be in the same single particle state. In contrast, the occupation numbers
nk for identical bosons can take any positive integer value:
nk = 0, 1, 2, (bosons).
We will see in the following sections that the fermion or boson nature of a many particle
system can have a profound effect on its properties.

8
Example: Calculate the partition function for an ideal gas of N = 3 identical fermions in
equilibrium with a heat bath at temperature T . Assume that each particle can be in one of
the four possible state with energies ǫ1 , ǫ2 , ǫ3 and ǫ4 .
The possible states are
(1) n1 = n2 = n3 = 1 and n4 =0
(2) n1 = n2 = n4 = 1 and n3 =0
(3) n1 = n3 = n4 = 1 and n2 =0
(4) n2 = n3 = n4 = 1 and n1 =0
The partition function is given by

Z(N = 3) = e−β(ǫ1 +ǫ2 +ǫ3 ) + eβ(ǫ1 +ǫ2 +ǫ4 ) + e−β(ǫ1 +ǫ3 +ǫ4 ) + e−β(ǫ2 +ǫ3 +ǫ4 )

For fermions, the grand canonical partition function for the kth energy level can be easily
calculated as nk either 0 or 1.

1
X
Zk (V, T, µ) = e−βnk (ǫk −µ) = 1 + e−β(ǫk −µ)
nk =0

For bosons nk is not restricted, can be any integer:

∞
X
−βnk (ǫk −µ) −β(ǫk −µ) −2β(ǫk −µ)
X
nk
Zk (V, T, µ) = e =1+e +e + ... = e−β(ǫk −µ)
nk =0 nk =0

This geometrical series is convergent for e−β(ǫk −µ) < 1. This has to be true for all ǫk , so we
have to require that ǫk − µ > 0, i.e. ǫk > µ, or by setting the lowest energy to ǫ0 = 0:

µ < 0(bosons)

(In contrast, the chemical potential can be either positive or negative for fermions.) The
summation of the geometrical series now gives
1
Zk (V, T, µ) =
1− e−β(ǫk −µ)
The grand potential for fermions

Φk (V, T, µ) = −kT lnZk = −kT ln(1 + e−β(ǫk −µ) )

The grand potential for bosons

Φk (V, T, µ) = −kT lnZk = kT ln(1 − e−β(ǫk −µ) )

9
 2

1.5

<n>
1

0.5

0
-3 -2 -1 0 1 2 3
(Ε−µ)/κΤ

FIG. 1: Fermi-Dirac, Bose-Einstein and Maxwell-Boltzmann distributions

The average occupation numbers for Fermi-Dirac

∂Φk 1
hnk i = = β(ǫ −µ)
∂µ e k +1

The average occupation numbers for Bose-Einstein

∂Φk 1
hnk i = = β(ǫ −µ)
∂µ e k −1

Note: If 1 ≪ eβ(ǫk −µ) then

hnk i = e−β(ǫk −µ)

(Maxwell-Boltzmann).
Figure 1. shows the FD, BE and MB distributions. The FD distribution converges to 1
for low energies as expected. The BE distribution converges to “infinity” when the energy
approaches the chemical potential. Both the FD and the BE distribution converges to the
classical MB distribution for high energies (or high temperatures). In the MD distribution
the occupation number is small (fraction on one).

10