Thermodynamics and Statistical Physics

1
B. Zheng
Zhejiang Institute of Modern Physics, Zhejiang University,
Hangzhou 310027, P.R. China

1
author, e-mail: zheng@zimp.zju.edu.cn; tel: 0086-571-87952753; Fax: 0086-571-
87952659

1
Chapter 1

Thermodynamics

2
Chapter 2

Classical statistical mechanics

3
Chapter 3

Canonical ensemble and

grand canonical ensemble

3.1 Canonical ensemble

3.1.1 Canonical ensemble
We consider an isolated system made up of two subsys-
tems
1: N1 , V1 , H1 ,
2: N2 , V2 , H2
with N = N1 + N2 , V = V1 + V2 ,

H = H1 (p1 , q1 ) + H2 (p2 , q2 )

and assume that interactions between two subsystem

are negligibly small.
As discussed before, “a most probable” decomposi-
tion E = Ē1 + Ē2 of the energy dominates.

A special case

N2 ≫ N1 , Ē2 ≫ Ē1

4
Let E = E1 + E2 to be any decomposition of the en-
ergy, fluctuating around the most probable decomposi-
tion E = Ē1 + Ē2 .
The probability of finding system 1 at (p1 , q1 ) and sys-
tem 2 at (p2 , q2 ) is
∝ dp1 dq1 dp2 dq2
If we look at only system 1, we should sum up all pos-
sible microscopic states of system 2,
∫
∝ dp1 dq1 dp2 dq2 = dp1 dq1 Γ2 (E − E1 )
E2 =E−E1

That is, the probability of finding system 1 in a micro-

scopic state within the volume dp1 dq1 at (p1 , q1 ) is
∝ dp1 dq1 Γ2 (E − E1 )
therefore, the density distribution
ρ(p1 , q1 ) ∝ Γ2 (E − E1 )

Since E1 fluctuates only around Ē1 , and Ē1 ≪ E

∂k log Γ2 (E2 )
k log Γ2 (E − E1 ) = k log Γ2 (E) − E1 + ···
∂E2 E2 =E
∂S2 (E2 )
= S2 (E) − E1
∂E2 E2 =E
E1
= S2 (E) −
T
T is the temperature of the whole system, and therefore
also the temperature of system 1. S2 (E) is irrelevant to
system 1.

5
 Question : Why not expand Γ2 (E − E1 )?

Omitting the subscript 1, it leads to

ρ(p, q) ∝ e−H(p,q)/kT

This is the probability distribution of the so-called canon-

ical ensemble. Similar to that of the microcanonical
ensemble, ρ(p, q) of the canonical ensemble depends on
(p, q) also only through the Hamiltonian H(p, q).
The macroscopic conditions for the canonical ensem-
ble:
• N is fixed.
• In contact with a heat reservoir of temperature T ,
and T is fixed.
We define the partition function as the effective volume
in Γ space, i.e., the effective number of the microscopic
states, ∫
1
QN (V, T ) ≡ d3N p d3N q e−H/kT
N!
then the thermodynamics is to be obtained from

QN (V, T ) ≡ e−A(V,T )/kT

where A(V, T ) is the free energy. To show this,

(a) A is extensive
(b) A is related to U by A = U − T S with
( )
∂A
S≡− .
∂T V

6
 Proof: Differentiating the following identity with re-
spect to β = 1/kT ,
∫
1
d3N p d3N q eβ[A(V,T )−H(p,q)] = 1
N!
we obtain
∫ [ ]
1 ∂A
d3N p d3N q eβ[A−H] A − H + β =0
N! ∂β
Since
∂A ∂A ∂T ∂A
β =β = −T = TS
∂β ∂T ∂β ∂T
Therefore
A = U − TS

Please note that now the temperature is already de-

fined from the beginning. Then, the free energy A(V, T )
is a characteristic function, and one may straightforwardly
derive
( ) ( )
∂A ∂A
P = − S=−
∂V T ∂T V
U = A + TS G = A + PV
The free energy A(V, T ) in the canonical ensemble plays
the role of the entropy S(E, V ) in the microcanonical
ensemble.
In principle, the integration of dp dq can not extend
to the entire Γ space, since from the beginning ρ(p1 , q1 )
requires E1 ≤ E. But we can show in the next subsection
that the contribution from the region H ≥ E is negligi-
bly small. In fact, only one value of the energy dominat-
ingly contributes to the partition function. Therefore,

7
practically the integration may extend over all energy.
This provides much convenience for calculations in the
canonical ensemble!
In addition, the free energy A(V, T ) in the canonical
ensemble can be explicitly computed. In contrast, the
internal energy U (S, V ) in the microcanonical ensemble
could be potentially obtained only from the computa-
tion of S(E, V ), and such a solution is usually much more
difficult.
The definition of the partition function QN (V, T ) in
the canonical ensemble is similar to the total volume
Γ(E) of the microscopic states in the microcanonical en-
semble. In fact

k log QN (V, T ) = −A(V, T )/T = S − U/T

Question: why and how?

Hint: only the microscopic states with the most prob-
able energy Ē, i.e., the average energy Ē − U , contribute
to QN (V, T ). Therefore QN (V, T ) = e−Ē/kT Γ(Ē).

Simple argument for the canonical ensemble

Suppose that a large isolated system consists of many
identical quasi-independent subsystems, and each sub-
system is described by the Hamiltonian H(p, q). Accord-
ing to the result in Sec. 2.7 in the previous chapter,
i.e., the Maxwell-Boltzmann distribution, the probabil-
ity distribution of finding a subsystem at (p, q) is

ρ(p, q) ∝ e−βH(p,q)

8
3.1.2 Energy fluctuation
Questions :
• In what sense are micro-canonical and canonical en-
sembles equivalent?
• Why do we need both ensembles?
Any macroscopic observable can be calculated by
∫
1
< O >= dp dq e−βH O
QN N !
For the energy
∫
1
U =< H >= dp dq e−βH H
QN N !
i.e., ∫
dp dq eβ(A−H) (U − H) = 0
Differentiating both sides with respect to β, we obtain
∫
∂U ∂A
+ dp dq eβ(A−H) (U − H)(A − H − T )=0
∂β ∂T
Further
( )
∂A
∵ S = =−
∂T V
U = A + TS
∂U
∴ + < (U − H)2 >= 0
∂β
i.e.,
∂U ∂U
< H 2 > − < H >2 = − = kT 2
∂β ∂T
2
= kT CV

9
In fact, this is a special case of the well-known fluctuation-
dissipation theorem, which relates the fluctuation of energy
to the response to the external change.
∵ <H> ∼N
CV ∼ N
As N → ∞,
< H 2 > − < H >2
→0
< H >2
Therefore, the microcanonical and canonical ensem-
ble are equivalent, in the sense that the energy only
fluctuates around a fixed value in both cases. In the
canonical ensemble, however, the energy is not given
and needs to be computed. Further, this theorem also
shows that the contribution of the states with a very
large energy is negligible.

3.1.3 Boltzmann statistical theory

N identical quasi-independent particles
∑
H= ϵ(pi , qi )
i

Now i labels particles

1 ∑ −βH
QN (V, T ) = e
N ! p,q
( )N
1 ∑ ∏ −βϵ(pi ,qi ) 1 ∑
= e = e−βϵ(pi ,qi )
N ! p,q i N! pi ,qi
1
= [Z(β, V )]N = e−βA(V,T )
N!
10
 Z(β, V )
A(V, T ) = −N kT log
( ) N
∂A
S = −
∂T V
Z(β, V ) N ∂ log Z(β, V ) ∂β
= N k log +
N β ∂β ∂T
( )
Z(β, V ) ∂ log Z(β, V )
= N k log −β
N ∂β
Comparing Z(β, V ) here with Z(β, y) introduced in
chapter 2, the only difference is that the energy for cal-
culating Z(β, V ) here is not bonded by the total energy.
But this difference is negligibly small.

3.2 Some applications of canonical ensemble

3.2.1 Equipartition theorem

1∑ 2
H= ai pi + U ({qi })
2 i
then
1 1
< ai p2i >= kT
2 2
Proof:
∫ +∞ 1 2 −βH 3N
1 −∞ 2 ai pi e d p d3N q
< ai p2i > = ∫ +∞
2 −βH d3N p d3N q
−∞ e
∫ +∞ 1 2 −β 1 a p2
−∞ 2 ai pi e
2 i i dp
i
= ∫ +∞ −β 1 a p2
−∞ e
2 i i dp
i

11
 ∫
1 −1
de−β 2 ai pi
1 2
numerator = ai · p2i
2 βai pi ∫
−1 −β 12 ai p2i +∞
|−∞ − e−β 2 ai pi dpi )
1 2
= (pi e
2β
1
= kT · denominator
2

1 1
∴ < ai p2i >= kT
2 2
Exercise: If H(ξ → ±∞) → ∞, then
∂H
<ξ >= kT
∂ξ

3.2.2 Ideal gases

1 ∑ 2
H = p
2m i i

∫ ∑ 2
1
QN (V, T ) = d3N p d3N q e−β i pi /2m
N!
(∫ )N
VN 3 −βp2 /2m
= d pe
N!
VN
= [q(β)]N
∫N ! ∞ ∫
dp p2 e−p
2
/2mkT
q(β) = dΩ
0
= (π2mkT )3/2

QN (V, T ) = e−A(V,T )/kT

12
 ∴ A(V, T ) = −kT log QN (V, T )
3
= −N kT (log V /N + log π2mkT + 1)
2
( )
∂A
S = −
∂T V
3 3 1
= +N k(log V /N + log π2mkT + 1) + N kT · ·
2 2 T
3 5
= +N k(log V /N + log π2mkT + )
2 2
( )
∂A
P = −
∂V T
N kT
=
V
This is the equation of state
U = A + TS
3
= N kT
2
then
3 π4mU 5
S = N k(log V /N +log + )
2 3N 2
Everything is the same as that from the microscopic
ensemble.

3.3 Ising model

The Ising model is the simplest model for magnetic ma-
terials, exhibiting a phase transition.

13
 Fig. 3.1 shows a square lattice in two-dimensions, de-
scribing a crystal. On each lattice site, there is a mag-
net, described by Si = ±1
The Hamoltonian
1 ∑ ∑
− H=K Si Sj + h Si , Si = ±1
kT <ij>

< ij > : nearest neighbour

J = −kT K : coupling
h̄ = −kT h : external magnetic field

The canonical ensemble

ρ ∝ e−H/kT

Omitting the Gibbs factor, the partition function is

written as ∑
Z= e−H/kT
{Si }

The average magnetization

1 ∑ 1 ∑ −H/kT
M= Si e
Z N i
{Si }

Solvable cases

d=1 there is no real phase transition

d=2 too difficult

Result: as shown in Fig. 3.2

14
 i j

Figure 3.1:

0 T T
c

Figure 3.2:

15
 h=0

T > TC M =0
T < TC M ̸= 0
T = TC M =0 continuous
∂M /∂T discontinuous
T − TC
T . TC M ∝ (−τ )β τ=
TC
The Mean-field approximation

How to make an approximation

• After the approximation, one can perform the cal-
culation
• There is a clear physical meaning, and at extreme
conditions, the approximation should be correct or
roughly correct

∑ ∑ ∑
Si Sj → Si < S >= d < S > Si
<ij> <ij> i

1 1 ∑
− H = − H(Si )
kT kT i
1
− H(Si ) = Kd < S > Si + hSi
kT
∑
e− kT H(Si )
1
Z = ZiN Zi =
Si =±1

16
 1 ∑
Si e− kT H(Si )
1
M = < Si >=
Zi
Si =±1
= tanh(Kd < S > +h)

The self-consistent condition

M =< S >

For h = 0,
M = tanh(KdM )
Since
1
tanh(x) = x − x3 + · · ·
3
1
∴ M = KdM − (KdM )3 + · · ·
3
Graphical solutions: as shown in Figs. 3.3 and 3.4
For Kd < 1, there is only one solution

M = 0.

For Kd > 1, there are two solutions

M =0

and
M = ±M0 .
In an equilibrium state, the free energy should reach its
minimum, and one can prove that M = ±M0 is a physical
solution.
1
∴ KC d = 1, KC =
d
The phase transition takes place.

17
Figure 3.3:

Figure 3.4:

18
 What are the features of the system around KC ?

K = KC , M =0
K = KC + ∆K ∆K small ⇒ M small

1
∴ M = (KC + ∆K)dM − (KC + ∆K)3 d3 M 3
3
1 3
≈ M + ∆KdM − M − ∆KdM 3
3
√
∆K
∴ M = ±M0 = ± 3
KC
Exercise: Derive ∂M /∂∆K, U and CV at and around
KC discuss their behavior at and around KC .

Under what conditions is the mean-field approxima-

tion almost correct?
• Higher spatial dimensions
• Long-range interactions
Advanced methods in statistical physics:

Approximate methods
• Mean-field methods, and improved mean-field meth-
ods
• Variational methods
• Series expansions, (High temperature and low tem-
perature expansions, cumulant expansions. . . )

19
 • Renormalization group methods (Real space, mo-
mentum space)
Exact methods
• Numerical solutions
• Monte Carlo simulations
• Molecular dynamics

3.4 Grand canonical ensemble

Now we allow the number of particles of the system to
be non-conserved. The idea here is similar to that for
deriving the canonical ensemble.

Γ space: spanned by (p, q, N ), N = 0, 1, 2, · · ·

The grand canonical ensemble:

A collection of systems, identical in composition and
macroscopic conditions, but existing in different num-
bers of particles and different microscopic states, with-
out interactions each other.
A member of the ensemble (i.e., a microscopic state
of the system), is represented by a point in Γ space.
The density distribution of these points is denoted by
ρ(p, q, N ).

To find ρ(p, q, N ), we consider a canonical ensemble of

N particles, volume V , and temperature T .
We divide the system into two subsystems:
1: N1 , V1

20
 2: N2 = N − N1 , V2 = V − V1
and consider the special case

V2 ≫ V1 , N 2 ≫ N1

and designate the coordinates of N1 particles in V1 by

{p1 , q1 }, and those in V2 by {p2 , q2 }

H(p, q, N ) = H(p1 , q1 , N1 ) + H(p2 , q2 , N2 )

Note that H(p1 , q1 , N1 ) and H(p2 , q2 , N2 ) here are the same

functions evaluated at different values of arguments.
This is different from the discussion in canonical en-
semble, since we allow interchange of particles in two
subsystems.
If a particle moves from V2 to V1 we change its coor-
dinates from {p2 , q2 } to {p1 , q1 }.

The partition function of the whole canonical system

∫
dp dq −βH(p,q)
QN (V, T ) = e
V N!
∫ ∑
N ∫ ∫
1 N!
= dp1 dp2 dq1 dq2
N! N1 !N2 ! V1 V2
N1 =0
·exp{−β[H(p1 , q1 , N1 ) + H(p2 , q2 , N2 )]}
∑N ∫
1
= dp1 dq1 e−βH(p1 ,q1 ,N1 )
N 1 ! V1
N1 =0
∫
1
· dp2 dq2 e−βH(p2 ,q2 ,N2 )
N 2 ! V2
∑ QN (V2 , T ) ∫
= 2
dp1 dq1 e−βH(p1 ,q1 ,N1 )
N1 ! V1
N1 =0

21
Now the integration over dq1 is restricted to V1 , dq2 to V2 ,
∑
and we need the sum over N1 , N1 · · · . The argument
is similar to the energy decomposition in deriving the
canonical ensemble. The factorial number N !/(N1 !N2 !) is
the number of possible ways to distribute N1 particles
to V1 and N2 particles to V2 .

For example: N = 2
∫ (∫ ∫ ) (∫ ∫ )
dq1 dq2 = dq1 + dq1 dq2 + dq2
V
(∫V1 ∫V2 ) (∫V1 ∫V2 )
= dq1 + dq2 dq1 + dq2
∫ V1
∫ V2
∫ V1
∫ V2
∫ ∫
= dq2 dq2 + 2 dq2 dq1 + dq1 dq1
V2 V2 V2 V1 V1 V1

The denominator N ! is introduced to solve the Gibbs

paradox. Whenever the number of particles is relevant
in the physical phenomena, such a factor is needed.
Here the Gibbs factor is essential, for all QN , QN1 and QN2 !

Therefore, one may define the probability distribu-

tion of finding N1 particles in V1 to be
QN2 (V2 , T ) e−βH(p1 ,q1 ,N1 )
ρ(p1 , q1 , N1 ) =
QN (V, T ) N1 !

Understanding
Taking into account the Gibbs factor,
e−βH(p1 ,q1 ,N1 ) e−βH(p2 ,q2 ,N2 )
ρ(p1 , q1 , N1 , p2 , q2 , N2 ) ∝
N1 ! N2 !
22
If looking only at system 1, one should integrate over
p2 and q2 ,
e−βH(p1 ,q1 ,N1 )
ρ(p1 , q1 , N1 ) ∝ QN2 (V2 , T )
N1 !
The factor
QN2 (V2 , T )
= exp{−β[A(N − N1 , V − V1 , T ) − A(N, V, T )]}
QN (V, T )
ensures the normalization
∑∫
dp1 dq1 ρ(p1 , q1 , N1 ) = 1
N1 =0 V1

Since N ≫ N1 , V ≫ V1

A(N − N1 , V − V1 , T ) − A(N, V, T ) = −N1 µ + V1 P

where µ and P are respectively the chemical potential
and the pressure of the whole system
( )
∂A(N2 , V, T )
µ = (3.4.1)
∂N2
( )
N2 =N
∂A(N, V2 , T )
P = − (3.4.2)
∂V2 V2 =V

Note that µ is a macroscopic parameter

• it is intensive, µ = µ(N, V, T ) = µ(V /N, T )
• In equilibrium µ = µ(V1 /N1 , T ) = µ(V2 /N2 , T )

Omitting the subscript 1.

e−βP V βµN −βH(p,q,N ) 1
ρ(p, q, N ) = e β=
N! kT

23
In the thermodynamic limit, one allows the range of
N to be 0 ≤ N < ∞. Here eβµN /N ! is an additional
term to the canonical distribution e−βH(p,q,N ) , different
from the microcanonical and canonical ensembles. In
other words, even with a same energy H(p, q, N ), the
probability ρ(p, q, N ) does depend on N . However, it is
very important that independent of positive or negative
µ, the Gibbs factor 1/N ! suppresses the probability of
very large N .
Additional to V and T , there is now another indepen-
dent macroscopic parameter N̄ , the average number of
particles. If µ = 0, N̄ is not an independent macroscopic
parameter, and it is fixed by V and T .

Questions : Does the energy nonconservation in the

canonical ensemble lead to an additional macroscopic
parameter?

To obtain thermodynamics, we keep in mind that

there is now one more microscopic variable N . As usual,
we define the grand partition function
∞
∑
Z(µ, V, T ) ≡ eβµN QN (V, T )
N =0

then
βP V = log Z(µ, V, T ) (3.4.3)
This looks already like the equation of state, but µ is
not convenient for experimental measurements. Thus

24
one computes the average number of particles
∞
∑
1
N̄ = < N >= N eβµN QN (V, T )
Z(µ, V, T )
N =0
∂
= kT log Z(µ, V, T )
∂µ

therefore
µ = µ(N̄ , V, T ) (3.4.4)
In fact, such a relation may be also obtained through
replacing N by N̄ in the definition of µ in Eq. (3.4.1).

Question : Can one prove it?

Hint: the average of a quantity is equal to the most
probable value.

Eliminating µ from Eqs. (3.4.3) and (3.4.4), we ob-

tains the real equation of state
P = P (N̄ , V, T )
Let us keep the definition of the internal energy as
the average value of the Hamiltonian. To calculate the
internal energy, we denote y = eβµ ,
∂
U = − log Z(y, V, T ) , U = U (N̄ , V, T ),
∂β y,V
then ( )
∂U
CV =
∂T V,N̄
∫ T
CV
S = dT
0 T
A = U − TS

25
 Compared with the canonical ensemble, we have one
more macroscopic parameter µ, or N̄ . N̄ is more conve-
nient for the experimental measurement than µ. Once
N̄ is fixed, computations of the thermodynamic quan-
tities are the same as in the canonical ensemble, for
example, one may derive all the common ones from the
free energy A(N̄ , V, T ).
Please note that the dependence of the internal en-
ergy U on N̄ is only through µ. The Hamiltonian itself
does indeed rely on the number of particles, N . But N
is a microscopic variable, and it is summed up already
in the calculation of the average.

It is possible to prove that

< N 2 > − < N >2 1
∝ →0
< N >2 <N >
therefore, the canonical ensemble and grand canonical
ensemble are also equivalent.

Exercise: Calculate the relative fluctuation of the

particle number.

3.5 The chemical potential

3.5.1 Thermodynamics
The chemical potential µ is defined such that the free
energy changes by µdN̄ , when the number of particles
changes by dN̄ , at constant T and V . That is, µ is the
free energy carried by a single particle. Therefore,
dA = −P dV − SdT + µdN̄ .

26
Then, we may deduce a generalized form of the first law
from U = A + T S,
dU = −P dV + T dS + µdN̄
In other words, µ is the energy carried by one particle, if
S and V are kept at constant. The energy conservation
naturally should take into account the energy transfer
through the particle exchange.
When µ is positive, it tends to drive N̄ to smaller
values, to lower the free energy at constant T and V ,
and vice versa (??).

Example 1: an ideal gas

QN = V N (π2mkT )3N/2 /N !

3
A(N, V, T ) = −N kT (log V + log π2mkT − log N + 1)
2
(since N ! = N log N − N )
V 3
= −N kT (log + log π2mkT + 1)
N 2
( )
∂A
µ =
∂N V,T
V 3 1
= −kT (log + log π2mkT + 1) − N kT (− )
N 2 N
V 3
= −kT (log + log π2mkT )
N 2
Here we should not forget the factor 1/N ! in the par-
tition function QN , since it solves the problem of the
Gibbs paradox. Otherwise, for example, it leads to a

27
problematic relation µ ∼ log V rather than µ ∼ log V /N .
Also, we may not derive the grand canonical ensemble.
We may write
N
eβµ = (3.5.5)
V (π2mkT )3/2
Therefore, µ is positive or negative, depending on the
temperature and the density of the gas.
In this simple case, one may directly verify that re-
placing N by N̄ , Eq. (3.5.5) just gives µ = µ(N̄ , V, T ) in
Eq. (3.4.4).

Example 2: N identical quasi-independent particles,

1
QN = Z(β, V )N
N!
where Z(β, V ) is the partition function of a single parti-
cle, and 1/N ! is the Gibbs factor. Therefore

−βA = log QN ≈ −N (log N − 1) + N log Z(β, V )

Thus
∂A
−βµ = −β = log(Z(β, V )/N )
∂N
On the other hand, from the theory of the most proba-
ble distribution
N = e−α Z(β, V )
one identifies
α = −βµ

Simple argument for the grand canonical ensemble

28
 Suppose that a large system consists of many identi-
cal quasi-independent subsystems, and each subsystem
is with N particles and described by the Hamiltonian
H(p, q). According to the Maxwell-Boltzmann distribu-
tion, the probability distribution of finding a subsystem
at (p, q, N ) is
ρ(p, q, N ) ∝ e−α−βH(p,q)
where α = −βµb , and µb is the chemical potential of the
large system. On the other hand, µb is nothing but the
free energy of the subsystem. Since the free energy is
extensive, it can be written as µb = N µ, with µ being
the chemical potential of the subsystem. Taking into
account the Gibbs factor,
1 βµN −βH(p,q,)
ρ(p, q, N ) ∝ e .
N!

Exercise: Derive thermodynamics of an ideal gas, i.e.

the equation of state, and U, CV , S, A
Hint: to derive the entropy of the ideal gas, an iso-
choric transformation is not sufficient, and one needs
also an isothermal transformation.

3.5.2 Chemical equilibrium

Suppose there are two kinds of particles in the grand
canonical ensemble, H = H1 + H2 , and the interactions

29
between two kinds of particles are negligible,
∞ ∑
∑ ∞
Z = QN1 QN2 eβ(µ1 N1 +µ2 N2 )
N1 =0 N2 =0

QNi = e−βAi (Ni ,V,T )

∑
2
A = Ai , Ai = −kT log QNi
i=1
∂Ai
µi =
∂Ni T, V, Nj̸=i

For example, in a mixture of water and ice, a molecule

of water can be converted to that of ice, and vice versa.

Assume that number of total particles is conserved,

i.e., N = N1 + N2 is a constant.
In equilibrium, the free energy reaches its minimum.
∑ ∑ ∂Ai ∑
0 = δA = δAi = δNi = µi δNi
i i
∂N i i

From conservation of particle number

∑
0 = δN = δNi
we obtain a condition for equilibrium
µ1 δN1 + µ2 δN2 = µ1 δN1 − µ2 δN1 = 0
i.e., µ1 = µ2 (V /N, T ) as shown in Fig. 3.5

In general, we may have several kinds of particles,

and a reaction
ν1 X1 + ν2 X2 + · · · ν1′ Y1 + ν2′ Y2 + · · ·

30
 We rewrite Y by X, and ν ′ by ν, then
∑
K
νi Xi = 0
i=1
e.g.,
2H2 + O2 2H2 O
ν1 = 2, ν2 = 1, ν3 = −2
Now, conservation of particle number is
δN1 δN2 δNK
= = ··· = const ≡ δN
ν1 ν2 νK
e.g.,
δNH2 δNO2 δNH2 O
= =
2 1 −2
Note that now there is not only one equation for the
conversation of particle number. Any pair of particles
forms an equation. Then, minimizing the free energy A
∑ ∂Ai ∑
0 = δA = δNi = µi νi δN
i
∂N i i

we obtain
∑
K
µi νi = 0
i=1

Exercise:

• With the canonical ensemble, prove that if H(ξ →

±∞) → ∞, then
∂H
<ξ >= kT
∂ξ

31
• With the mean-field approximation of the Ising model,
derive ∂M /∂∆K, U and CV at and around KC , and
discuss their behavior at and around KC .
• Calculate the relative fluctuation of the particle num-
ber in the grand canonical ensemble.
• With the grand canonical ensemble, derive thermo-
dynamics of an ideal gas, i.e., the equation of state,
and U, CV , S, A.
• Problems 7.1, 7.2 and 7.3 in the textbook.
• Discussion
Gibbs paradox

32
V / N
=
1 2

Figure 3.5:

33