Chapter 10

Grand canonical ensemble

10.1 Grand canonical partition function

The grand canonical ensemble is a generalization of the canonical ensemble where the
restriction to a definite number of particles is removed. This is a realistic representation
when then the total number of particles in a macroscopic system cannot be fixed.
Heat and particle reservoir. Consider a sys-
tem A1 in a heat and particle reservoir A2 . The
two systems are in equilibrium with the thermal
equilibrium.
– Thermal equilibrium results form the ex-
change of heat. The two temperature are
then equal: T = T1 = T2
– The equilibrium with respect to particle
exchange leads to identical chemical poten-
tials: µ = µ1 = µ2 .

Energy and particle conservation. We assume that the system A2 is much larger
than the system A1 , i.e., that
E2 � E 1 , N2 � N1 ,
with
N1 + N2 = N = const. E1 + E2 = E = const.
where N and E are the particle number and the energy of the total system A = A1 + A2 .
Hamilton function. The overall Hamilton function is defined as the sum of the Hamilton
functions of A1 and A2 :
H(q, p) = H1 (q(1), p(1), N1 ) + H2 (q(2), p(2), N2 ) .
For the above assumption to be valid, we neglect interactions among particles in A1 and
A2 :
H12 = 0 .

117
118 CHAPTER 10. GRAND CANONICAL ENSEMBLE

This is a valid assuming for most macroscopic systems.

Microcanonical ensemble. Since the total system A is isolated, its distribution function
is given in the microcanonical ensemble as
1
ρ(q, p) = δ(E − H1 − H2 ) ,
Ω(E, N )

as in (9.1), with t �
Ω(E, N ) = d3N q d3N p δ(E − H1 − H2 )

being the density of states.

Sub-macroscopic particle exchange. The total entropy of the combined system is
given by the microcanonical expression
� �
Ω(E, N )Δ
S = kB ln , Γ0 (N ) = h3N N1 !N2 ! , (10.1)
Γ0 (N )

where Δ is the width of the energy shell. Compare (9.4) and Sect. 9.4.
The reason that Γ0 (N ) ∼ N1 !N2 ! in (10.1), and not ∼ N !, stems from the assumption that
there is no particles exchange on the macroscopic level. This is in line with the observation
made in Sect. 9.6 that energy fluctuations,
√ i.e. the exchange of energy between a system
and the heat reservoir, scale like 1/ N relatively to the internal energy. We come back
to this issue in Sect. 10.3.2.
Integrating out the reservoir. We are now interested in the system A1 . As we did for
the canonical ensemble, we integrate the total probability density ρ(q, p) over the phase
space of the reservoir A2 . We obtain
�
ρ1 (q(1), p(1), N1 ) ≡ dq(2) dp(2) ρ(q, p)
�
dq(2) dp(2) δ(E − H1 − H2 , N )
=
Ω(E, N )
Ω2 (E − H1 , N − N1 )
≡ (10.2)
Ω(E, N )

in analogy to (9.2).
Expanding in E1 and N1 . With E1 � E and N1 � N , we can approximate the slowly
varying logarithm of Ω2 (E2 , N2 ) around E2 = E and N2 = N and thus obtain:
� �
Ω2 (E − E1 , N − N1 )Δ
S2 (E − E1 , N − N1 ) = kB ln
Γ0 (N )
� � �� � � ��
∂S2 � ∂S 2 �
= S2 (E, N ) − E1 � − N1 � .
∂E2 N2 � E2 = E ∂N2 E2 � E2 = E
N2 = N N2 = N
10.1. GRAND CANONICAL PARTITION FUNCTION 119

Using (5.14), T dS = dU + P dV − µdN , the derivatives may be substituted by

� � � �
∂ Ω2 (E, N )Δ ∂S2 1
kb ln = = (10.3)
∂E Γ0 (N ) ∂E2 N2 = N T
E2 = E

and � � � �
∂ Ω2 (E, N )Δ ∂S2 µ
kb ln = = − . (10.4)
∂N Γ0 (N ) ∂N2 N2 = N T
E2 = E

Expansion of the probability density. (10.3) and (10.4) allow to expand Ω2 (E −

E1 , N − N1 ) as
� � � �
Ω2 (E − E1 , N − N1 )Δ Ω2 (E, N )Δ E 1 N1 µ
kB ln = kB ln − + ,
Γ0 (N ) Γ0 (N ) T T

which leads via E1 = H1 and (10.2) to a probability distribution of the form

� � � �
ρ1 (q(1), p(1), N1 ) ∼ exp − H1 (q(1), p(1)) − µN1 /(kB T ) .

Grand canonical partition function. The constant of proportionality for the proba-
bility distribution is given by the grand canonical partition function Z = Z(T, V, µ),

�∞ �
d3N q d3N p −β[H(q,p,N )−µN ]
Z(T, V, µ) = e , (10.5)
N =0
h3N N !

where we have dropped the index to the first system substituting ρ, N , q and p for ρ1 ,
N1 , q(1) and p(1). The partition function normalizes the distribution function

1 1
ρ(q, p, N ) = e−β[H(q,p,N )−µN ] (10.6)
h3N N ! Z(T, V, µ)

to 1: ∞ �
�
d3N q d3N p ρ(q, p, N ) = 1
N =0

Note that the normalization factor h3N N ! in (10.5) and (10.6) cancel each other. It
is in any can not possible to justify this factor within classical mechanics, as discussed
previously in Sect. 8.2.
Internal Energy. From the definition (10.5) of the grand canonical potential function
it follows that internal energy U = �H� is given by

∂
U − µN = − ln Z(T, V, µ) . (10.7)
∂β
120 CHAPTER 10. GRAND CANONICAL ENSEMBLE

10.2 Grand canonical potential

In Sect. 5.4 we defined the grand canonical potential∗ as

Ω(T, V, µ) = F (T, V, N ) − µN , Ω = −P V ,

and showed with Eq. (5.21) that

∂ � � ∂
U − µN = βΩ = − ln Z(T, V, µ) , (10.8)
∂β ∂β

where we have used in the last step the representation (10.7) of U − µN = ∂ ln Z/∂β.
Grand canonical potential. From (10.8) we may determine (up-to a constant) the
grand canonical potential with

Ω(T, V, µ) = −kB T ln Z(T, V, µ) , Z = e−βΩ (10.9)

as the logarithm of the grand canonical potential. Note the analog to the relation F =
−kB T ln ZN valid within the canonical ensemble.
Calculating with the grand canonical ensemble. To summarize, in order to obtain
the thermodynamic properties of our system in contact with a heat and particle reservoir,
we do the following:

1 - calculate the grand canonical partition function:

�∞ �
d3N q d3N p −β(H−µN )
Z = 3N N !
e ;
N =0
h

2 - calculate the grand canonical potential:

Ω = −kB T ln Z = −P V, dΩ = −SdT − P dV − N dµ ;

3 - calculate the remaining thermodynamic properties through the equations:

� �
∂Ω
S = −
∂T V,µ
� �
∂Ω Ω
P = − = −
∂V T,µ V
� �
∂Ω
N = −
∂µ T,V

and the remaining thermodynamic potentials through Legendre transformations.

∗
Don’t confuse the grand canonical potential Ω(T, V, µ) with the density of microstates Ω(E)!
10.3. FUGACITY 121

10.3 Fugacity
By using the definition of the partition function in the canonical ensemble,
�
1
ZN = 3N d3N q d3N p e−βH ,
h N!
we can rewrite the partition function in the grand canonical ensemble as
∞
� ∞
�
Z(T, V, µ) = eµN β ZN (T ) = z N ZN (T ) , (10.10)
N =0 N =0

where z = exp(µ/kB T ) is denoted as the fugacity. Equation (10.10) shows that Z(T, µ)
is the discrete Laplace transform of ZN (T ).

10.3.1 Particle distribution function

In the canonical ensemble the particle number N was fixed, whereas it is a variable in the
in the grand canonical ensemble. We define with
ZN (T )
wN = eβµN (10.11)
Z(T, µ)
the probability that the system at temperature T and with chemical potential µ contains
N particles. It is normalized:
∞
� �∞
eβµN ZN (T )
wn = = 1.
N =0 N =0
Z(T, µ)

Average particle number. That mean particle number �N � is given by

�∞ �∞ N
N =0 N z ZN (T, V )
�N � = N wN (T, V ) = � ∞ N
, (10.12)
N =0 N =0 z ZN (T, V )

which can be rewritten as

� �
1 ∂
�N � = ln Z(T, V, µ) , (10.13)
β ∂µ T,V

or, alternatively, as
� �
∂
�N � = z ln Z(T, V, µ) , (10.14)
∂z T,V

when making use of the definition of fugacity z.

Chemical potential. Conversely, the chemical potential for a given average particle
number N = �N �,
µ = µ(T, V, N ) ,
is obtained by by inverting (10.13).
122 CHAPTER 10. GRAND CANONICAL ENSEMBLE

10.3.2 Particle number fluctuations

We start by noting that from
� �
∂Ω
Ω = −P V, N = −
∂µ T,V

it follows that � � � �
∂P −1 ∂Ω N
= = . (10.15)
∂µ T,V V ∂µ T,V V

Particle number fluctuations. From (10.12),

� βµN
N N e ZN (T, V )
�N � = � βµN
,
N e ZN (T, V )

it follows that
� � � �
∂N 2 2
= β �N � − �N � , (10.16)
∂µ T,V

where the second term results from the derivative of the denumerator.
Two systems are in equilibrium with respect to the exchange in particle if their chemical
potentials are identical, µ = µ1 = µ2 . It hence makes sense that the the dependence of
the particle number on the chemical potential is proportional to the size of the particle
number fluctuations �N 2 � − �N �2 , which mediate the equilibrium.
Scaling to the thermodynamic limit. Using (10.15) we can recast (10.16) into
� 2 �
∂ P β� 2 2
�
= �N � − �N � . (10.17)
∂µ2 T,V V

The left-hand side of (10.17) is intensive, viz it scales like V 0 , since both the pressure P
and the chemical potential are intensive. For the right-hand side to scale like V 0 for large
volume V we hence need that
�
� √ �N 2 � − �N �2 1
�N 2 � − �N �2 ∼ V , ∼ √ , (10.18)
N N
where we have used that V ∼ N for fixed densities. The relative particle number fluctu-
ations hence vanish in the thermodynamic limit; a precondition for the grand-canonical
and the canonical ensemble to yield identical results.

10.3.3 Stability conditions

The mechanical stability condition for a system implies that its compressibility κT cannot
take negative values, i.e., � �
1 ∂V
κT = − ≥0.
V ∂P T
10.3. FUGACITY 123

In fact, this can be proven within the grand canonical ensemble in the same way as it
was proven in Sect. 9.6 that the energy fluctuations in the canonical ensemble fulfill the
thermal stability criterion
�� �2 �
CV = kB β 2 H − �H� ≥ 0. (10.19)

Intensive variables. We start be defining the free energy per particle, a = F/N , as

F (T, V, N )
a(T, v) ≡ v = V /N , (10.20)
N

where the v is the volume per particle (the specific volume).

– a = F/N is intensive and a function of the intensive variable T . It can therefore not
be a independently a function of the volume V or the particle number N (which are
extensive), only of V /N (which is intensive).

– The circumstance that a is a function of V /N only allows to transform derivative

with respect to V and to N into each other, the reason of using this representation.

Chemical potential. With (10.20) we find

� �
∂F ∂ N F/N ∂a ∂a ∂(V /N )
µ = = = a+N = a+N ,
∂N ∂N ∂N ∂(V /N ) � ∂N
�� �
−V /N 2

when using (5.12), namely that dF = −SdT − P dV + µdN . This relation yields

∂a ∂µ ∂ 2a
µ = a−v , = −v 2 . (10.21)
∂v ∂v ∂v

Note, that µ is intensive.

Pressure. For the pressure P , an intensive variable, we find likewise that
� �
∂F ∂ N F/N ∂a ∂(V /N )
P = − = − = −N ,
∂V ∂V ∂(V /N ) � ∂V
�� �
1/N

which results in

∂a ∂P ∂ 2a ∂µ ∂P
P = − , = − 2, = v , (10.22)
∂v ∂v ∂v ∂v ∂v

where we have used in the last step the comparison with (10.21).
124 CHAPTER 10. GRAND CANONICAL ENSEMBLE

Particle fluctuations in intensive variables. Using intensive variables for (10.17),

� �
β� 2 2
� ∂ 2 P (T, v) ∂ ∂P ∂v
�N � − �N � = = , (10.23)
V ∂µ2 ∂v ∂µ ∂µ
the last relation of (10.22) and (10.15), namely ∂P/∂µ = 1/v, we obtain
� �� �
∂ 2 P (T, v) ∂ 1 1 ∂v 1 ∂V 1 ∂v
2
= , κT = − = − .
∂µ ∂v v v ∂P V ∂P v ∂P
� �� � � �� �
−1/v 2 −κT

Compressibility. Taking the results together we obtain

βv 2 � 2 � V �N 2 � − �N �2
κT = �N � − �N �2 = β . (10.24)
V N N

for the compressibility κT .

– The compressibility is strictly positive, κT ≥ 0, which implies that the system
constricts when the pressure increases. The basic stability condition.
�
– The compressibility is finite if the size of the fluctuations �N 2 � − �N �2 vanishes
in the thermodynamic limit relative to the number of particles N present in the
system.

Response vs. fluctuations. The specific heat CV and the compressibility κT are mea-
sure the response of the system to a change of variables.
– A temperature gradient ΔT to the heat reservoir leads to a transfer of heat,
ΔQ = CV ΔT,
which is proportional to CV . Compare Sect. 3.2.
– An increase in temperature by ΔP leads to relative decrease of the volume by
ΔV
= κT ΔP,
V
which is proportional to κT .
It is not a coincidence, that both response functions, CV and κT , as given respectively by
(10.19) and (10.24), are proportional to the fluctuations of the involved variables.

A system in which a certain variable A is not fluctuating

(and hence fixed) cannot respond to perturbations trying
to change A. The response always involves the fluctuation
�� �2 �
A − �A� = �A2 � − �A�2 .
10.4. THE IDEAL GAS IN THE GRAND CANONICAL ENSEMBLE 125

10.4 The ideal gas in the grand canonical ensemble

We consider a mono-atomic gas system in a volume V in contact with a heat and particle
reservoir at temperature T and chemical potential µ.
Canonical partition function. One calculates first the grand canonical partition func-
tion, which we found in Chap. 9 to be
� �N
1 V h
ZN (T, V ) = 3
, λT = √ ,
N ! λT 2 π m kB T
where λT is the thermal wavelength.
V
Z1 (T, V, 1) = ,
λ3T
(10.25)

Weighted sum over particle number. The grand canonical partition function is
�∞ �∞ � �N �∞ � �N
N
µN 1 V 1 µ/k T V
Z = z ZN (T ) = e kB T 3
= e B 3 ,
N =0 N =0
N ! λ T N =0
N ! λ T

where z = exp(βµ) is the fugacity. The result is

� �
µ/kB T V
Z(T, V, µ) = exp e . (10.26)
λ3T

Note that the factor N ! comes in handy.

The grand canonical potential. The grand canonical potential is then obtained as

V
Ω = −kB T ln Z, Ω(T, V, µ) = −kB T eµ/kB T . (10.27)
λ3T

Particle number. Comparing the expression

� �
∂Ω V
N = − , N (T, V, µ) = eµ/kB T (10.28)
∂µ T,V λ3T

for the particle number N with (10.27) we find

Ω(T, V, µ) = −kB T N, N = N (T, V, µ) . (10.29)

This relation holds only for the ideal gas.

Equation of state. Comparing now
� �
∂Ω 1
− = P (T, V, µ) = kB T eµ/kB T 3 (10.30)
∂V T,µ λT
126 CHAPTER 10. GRAND CANONICAL ENSEMBLE

with (10.28) we find immediately the equation of state

V P = kB T N

of the ideal gas

Entropy. Evaluating the entropy
� �
∂Ω V h
S = − , Ω(T, V, µ) = −kB T eµ/kB T 3 λT = √ ,
∂T V,µ λT 2 π m kB T

we find � � � �
Ω 3 −µ Ω 5 µ
S = − 1+ −Ω = − − ,
T 2 kB T 2 T 2 kB T
which leads with (10.29), Ω = −kB T N , to
� �
5 µ V
S(T, V, µ) = kB N − , N = eµ/kB T . (10.31)
2 kB T λ3T

Sackur-Tetrode equation. Inverting N for µ/(kB T ) we obtain for the entropy with
� �
V 5
S = kB N ln 3
+
N λT 2

the Sackur-Tetrode equation, which coincides with the both the microcanonical and the
canonical expressions, (9.20) and (8.25). The three ensembles yield identical results in
the thermodynamic limit.
10.4. THE IDEAL GAS IN THE GRAND CANONICAL ENSEMBLE 127

10.4.1 Summary ensembles

Phase space integration Thermodynamic potential
� � �
3N 3N Γ
Microcanonical Γ(E, V, N ) = d qd p S(E, V, N ) = kB ln 3N
E<H<E+Δ h N!
ensemble

∂Γ
Ω(E, V, N ) =
∂E
�
= d3N q d3N p δ(E − H)

�
d3N q d3N p −βH
Canonical ZN (T, V ) = e F (T, V, N ) = −kB T ln ZN (T, V )
h3N N !
ensemble � ∞
1
= 3N dE Ω(E) e−βH ZN = e−βF
h N! 0
�∞
Grand canonical Z(T, µ) = N =0 eµN β ZN (T ) Ω(T, V, µ) = −kB T ln Z(T, V, µ)
ensemble
PV
Z = e−βΩ = e kB T
128 CHAPTER 10. GRAND CANONICAL ENSEMBLE