Chapter 1

Statistical Ensembles
1.1
1.1.1

Principle of statistical physics and ensembles

Partial information to partial result

Statistical systems are complex systems. The systems are so complex that we cannot obtain all the
information to completely characterize the systems. For example, a liter of gas may contain 1022 atoms.
To completely characterize such a system (or more precisely, a state of such a system), we need to known
the three components of the velocity for each atoms and the three components of the position for each
atoms. It is impossible to obtain 6 1022 real numbers to completely characterize the gas.
However, not knowing all the information needed to characterize gas does not prevented us to develop a
theory of gas. This is because we are only interested in some average properties of gas such as the pressure,
volume, temperature. Those properties do not depend on every little details of each atoms. Not knowing
every thing about the atoms does not prevent us from calculating those properties. This is the kind of
problems in statistical physics. In statistical physics we try to understand the properties of a complex
system without know all the information of the systems. This is possible since the properties that we are
interested in do not depend on the all details of the system.

1.1.2

All possible states appear with an equal probability

In statistical physics there is only one principle: All possible states appear with an equal probability. Let
us explain what do we mean by the above statement. Suppose we know certain quantities, such as pressure,
total energy, etc, of a complex systems. But those quantities do not characterize the system completely.
This means that the system has a number of states for which those quantities take the same values. Thus
even after knowing the values of those quantities, we still do not know, among those possible states, which
state the system is actually in. Then according to the principle of statistical physical, we say all the possible
states are equally likely.

1.1.3

Time average and ensemble average

But the system can only be in one state at a given time. What do we mean by all the possible states
are equally likely? There are two points of view. In the rst point of view, we may imagine we have
many copies of the system, all characterized by the same set of quantities, such as total energy, pressure,
etc . But each copy may be in a dierent possible state. Then equally likely means that each possible
1

state appear the same number of times among the copies of the system. The copies of the system is called
ensemble. We have to have an ensemble to even dene the probabilities. Under the rst interpretation,
statistical physics is a science that deal with ensembles, rather than individual systems.
The second point of view only apply to the situation where the environment of the system is independent
of time. In this case we interpret equally likely as that all the possible states appear for the same amount
of time during a long period of time. The second point of view is related to the rst point of view if we
view the system at dierent times as the dierent copies of the system.
The two points of view may not be equivalent. The two points of view are equivalent only when the
system can visit all the possible states, many times, during a long period of time. This is the ergodicity
hypothesis.
Not all systems are ergodic. For a non-ergodic system, statistical physics only apply to its ensemble.
For an ergodic system, statistical physics also apply to the time average of the system.

1.2

Microcanonical ensemble

A microcanonical ensemble is an ensemble formed by isolated systems. All the systems in the ensemble
have the same energy (and possibly some other properties). Here by same energy we really mean that
all systems has an energy which lies within a small window between E and E + E.

1.2.1

Number of states and entropy

Let us rst study a simple example: N spins in magnetic eld. The energy for an up-spin is E = 0 /2
and for a down-spin E = 0 /2.
We like to ask: How many states are there with a total energy E?
Since the total energy is given by E = M 20 (N M ) 20 , where M is the number of up-spin. So
the states with a total energy E are the states with M up-spins. But how many states are there with
N!
M =
M
M = E0 + N2 up-spins? The answer is CN
M !(N M )! . (Here CN is the number of ways to pick M objects
from N objects.) So the number of states with a total energy E is
E

(E) = CN0

+N
2

After obtain the number of states as a function of total energy E, we can dene the entropy of the
system: the entropy is kB time the log of the number of states:
S(E) = kB ln (E),

(1.2.1)

where kB = 1.3807 1016 erg/K = 8.617343(15) 105 eV/K is the Boltzmann constant. (The Boltzmann
constant is conversion factor between energy and temperature: 1K = 1.38071016 erg = 8.617343(15)
105 eV. Or 1eV = 11605K. We will introduce the denition of temperature shortly.)
For a microcanonical ensemble, entropy a function of energy E. So for our N -spin system, the entropy
is

S(E) = kB ln CN0

+N
2

(1.2.2)

M , we can use the Stirlings approximation

To calculate ln CN

ln(n!) = n ln n n +
2

1
ln(2n) + O(1/n)
2

(1.2.3)

S(E)/N

0
0.5

E/E 0 = / 0

0.5

Figure 1.1: The entropy per spin, S(E)/N , as a function of E or the average energy per spin. The
maximum entropy of a spin-1/2 spin is kB ln(2) = 0.69314718056kB .
Thus (see Fig. 1.1)
1
M
kB
S(E) = ln CN

N ln N M ln M (N M ) ln(N M )
M
N M
= M ln( ) (N M ) ln(
)
N
N
=N (f ln f f ln f )

(1.2.4)

0
0
M
where f M
N (or f 1 N ) is the probability for a spin to be up (or down). Since E = M 2 (N M ) 2 ,
E
1
E
1
we have f = 2 + E0 and f = 2 E0 where E0 = N 0 . Thus

[
1
E
1
E
1
E
1
E ]
1
kB
S(E) =N ( +
) ln( +
)(
) ln(
)
2 E0
2 E0
2 E0
2 E0

(1.2.5)

Clearly, from the denition, the physical meaning of the entropy is

number of states with energy E = eS(E)/kB

1.2.2

(1.2.6)

Concept of temperature

To introduce the concept of temperature, let us put two systems of spins together. The system-1 has N1
1,2 be the energies of the two systems at the beginning. The
spins and the system-2 has N2 spins. Let E

total energy is E = E1 + E2 . If we allow the two systems to exchange their energy, then the spins in the
two systems may ip up and down, and sample all the possible states with total energy E. Now we like to
ask what is the probability for the system-1 to have a new energy E1 ? Certainly, the system-2 will have
an energy E2 = E E1 when the system-1 to have a new energy E1 .
The number states with system-1 having an energy E1 and the system-2 having an energy E2 = E E1
is given by tot = 1 (E1 )2 (E2 ), or in terms of entropies:
1

tot (E1 ) = ekB

1
S1 (E1 ) kB
S2 (EE1 )

= ekB

[S1 (E1 )+S2 (EE1 )]

(1.2.7)

Since every possible states are equally possible, the probability for the system-1 to have an energy E1
1

P (E1 ) ekB

[S1 (E1 )+S2 (EE1 )]

(1.2.8)

From Fig. 1.2, we see that when N , P (E1 ) is almost a -function. We can almost say for sure that
1 that it maximizes the total entropy S1 (E1 ) + S2 (E E1 ),
the energy of the system-1 has such a value E
or
1 ) = S (E E
1 )
S1 (E
(1.2.9)
2
3

P(E 1)/P max

0.8

0.6

10
0.4

100

0.2

1000
0
0.5

0.4

0.3

10000

0.2

0.1

E1 / |E |

Figure 1.2: For a system of N1 spins and a system of N2 spins with total energy E, we plot the probability
P (E1 ) for the N1 -spin system to have an energy E1 . Here N2 = 2N1 and N1 = 10, 100, 1000, 10000. The
total energy E is chosen to be N1 0 . P (E1 ) reach its maximum when E1 = E/3.

0.5

/ 0

0.5

Figure 1.3: The relation between temperate T , the inverse temperature with the average energy per spin
.
1 at the beginning is not equal to E
1 , then after we bring the two spin systems together, E1 will shift
If E
1 to E
1 . We see that Eq. (1.2.9) is a condition for equilibrium. It is also the maximum entropy
from E
condition. We have derived the second law of thermodynamics: as an isolated system approach to the
equilibrium state, its entropy always increase (if we dene the entropy as in Eq. (1.2.1)).
If we dene the temperature as
1
S(E)
= kB =
T
E
then the equilibrium condition Eq. (1.2.9) becomes

For our spin system

(1.2.10)

T1 = T2

(1.2.11)

1 ( 1 E0 E )
1 ( f )
1
= kB = kB ln 21
= kB ln
T
0
0
f
2 E0 + E

(1.2.12)

(see Fig. 1.3).

0.008
experiment

(emu/mole Cu)

Curie law

Ca 2Y2 Cu5 O10

0
0

100

200

300

T (K)

Figure 1.4: Curies law for a magnetic material.

1.2.3

Curies law

Eq. 1.2.12 relates the temperature T with the probability distribution of the spins: f and f . We nd
f
= e0 /kB T
f

(1.2.13)

which implies that

1
e0 /kB T
,
f = /k T
,

/k
T
e 0 B +1
e 0 B +1
Note that Eq. 1.2.13 is nothing but the Boltzmann distribution. So from the equal probability principle
and the denition of temperature Eq. 1.2.10, we can derive the Boltzmann distribution.
f =

For a spin-1/2 system in magnetic eld B, 0 = gB B. The total magnetic moment is M = N (f

f ) g2B . We nd
)
(
gB
e0 /kB T
1
M=N

2
e0 /kB T + 1 e0 /kB T + 1
N gB
0
tanh
2
2kB T
gB B
N gB
tanh
=
2
2kB T

For B

kB T
gB ,

we have
M=

We nd magnetic susceptibility =

1.2.4

g 2 2B N
4kB T

g 2 2B N
B
4kB T

1/T . This is the Curies law (see Fig. 1.4).

Properties of entropy

Entropy is an extensive quantity

From
[
1

1
]
1
S(E) =N ( + ) ln( + ) ( ) ln( )
kB
2 0
2 0
2 0
2 0

(1.2.14)

where = E/N is the average energy per spin, we see that entropy is proportional to N , the size of system.
Thus S is an extensive quantity. In contrast, , as the average energy per spin, is an intensive quantity.
The total energy E is an extensive quantity and the temperature T is an intensive quantity.
5

2E
E1

E2

(a)

(b)

E
(c)

Figure 1.5: Total entropy of combined systems.

( E 1) ( E 2)
E1

( E ) ( E )

and the system-1

Figure 1.6: Total numbers of states in the combined system with the total energy 2E
energy E1 .
Entropy and energy window
From the denition of entropy
S(E, E) = kB ln(number of states with energy between E and E + E)

(1.2.15)

we see that entropy also depend on the energy window E. However, in the thermodynamical limit
N , such a dependence can be dropped and we can regard S as a function of E only.
To see this, we consider
S(E, E) =kB ln(number of states with energy between E and E + E)
=kB ln[ (number of states with energy between E and E + E)]
=S(E, E) + kB ln

(1.2.16)

Since S(E, E) N , as long as = O(N n ), kB ln term can be dropped.

Additive property of entropy
Consider two systems both with N spins. The rst system has total energy E1 and the second E2 . The
rst system has 1 = (E1 ) possible states and the second 2 = (E2 ) possible states.
If we put the two systems together, but forbid any exchange of energy between them (see Fig. 1.5a), then
the combined system will has = 1 2 possible states. The entropy of the combined system S = kB ln
is the sum of the sub systems
S = S1 + S2
(1.2.17)
If we allow the two system to exchange energy, the two systems will reach an equilibrium state. The
= (E1 + E2 )/2 in the equilibrium state. The equilibsubsystem will have the same average energy E
The number of possible states become
rium state of the combined system will have a total energy 2E.
6

 = (E1 )(2E
E1 ), it is clear that the
> = (E1 )(E2 ) and the equilibrium state has

E1
a higher entropy (see Fig. 1.6). Thus reaching equilibrium always increase entropy (the second law of
thermodynamics).
After the two systems reach the equilibrium, we now forbid the energy exchange. The total number
= (E)(
E).
We like to show that ln (E)(
E)
= ln
in the thermodynamical
states is then reduced to
limit, ie the system Fig. 1.5b and the system Fig. 1.5c have the same entropy. As the maximum of the
E1 ), we nd
> /2N

(E1 )2 (2E
, where 2N is the number of possible distinct values of E1 . We also

have < . Thus

> ln
> ln(/2N

ln
)
(1.2.18)
or

S > S > S kB ln(2N )

(1.2.19)

Since S and S is of order N . In large N limit, we can regard S = S .

From Fig. 1.6 we also see that as the system goes from Fig. 1.5a to the equilibrium state Fig. 1.5b or
Fig. 1.5c, the entropy of the system is maximized. Or equilibrium state has maximum entropy.
Entropy and probability distribution
The entropy can be related to probability distribution. To see such a relation, let us interpret the denition
of the entropy as follows: if the states at energy E all have the same probability to appear, then the
entropy is S = kB ln . Now the question is that if the states have unequal probabilities to appear, then
what is the entropy of the system?
From the entropy of the system of N spin-1/2 S = kB N (f ln f f ln f ) and the additive property
of the entropy, we may interpret kB (f ln f f ln f ) as the entropy of a single spin-1/2. Since f and
f are the probabilities for spin in the up and the down states, we see how entropy and the probabilities
are related.
In general, if fi is the probability for the ith state to appear, then the entropy is
S = kB

fi ln fi ,

(1.2.20)

i=1

where is the total number of states. If all states have the same probability to appear, we have fi = 1/
and
S = kB

fi ln fi = kB

i=1

1 ln 1 = kB ln .

(1.2.21)

i=1

We recover the old result.

Reversible and irreversible processes
The system in Fig. 1.5b is evolved from the system in Fig. 1.5a. There are only (E1 )(E2 ) possible

initial states. Thus there will be only (E1 )(E2 ) possible nal states. The system in Fig. 1.5b has

states with energy 2E. But among possible state, only (E1 )(E2 ) of them are the nal states evolved
from the system in Fig. 1.5a. But we have no clue about which are the (E1 )(E2 ) possible nal states.
We lost the information. We only know the total energy of the system, and we only know the state can be
states. This is how the entropy get increased.
in one of the
7

4 states

4 states

7 states

4 states

(a)

(b)

Figure 1.7: The lines represent possible states. The thick lines represent states that actually appear in the
ensembles. (a) represents an irreversible process while (b) represents a reversible process.
The evolution from Fig. 1.5a to Fig. 1.5b is also presented in Fig. 1.7a. The Fig. 1.7b represent a
reversible (or adiabatic) evolution, say, caused by a change in 0 . We see that reversible (or adiabatic)
processes do not change the entropy, since the number of possible states is not changed.

1.3

Application to classical ideal gas

How to count states:

Each state is described by a point in phase space (q, p).
A particle has three degrees of freedom and its state is described by (x, px , y, py , z, pz ).
Consider a N -particle system. How many states with total energy below E. The answer is innity.
We need quantum physics to get a sensible result. Each state in a degree freedom occupies a finite area
qp = h. For the N particle system, the phase space is 6N dimensional. Each h3N volume in the 6N
dimensional phase space correspond to one state. Thus the number of states with total energy below E is
given by

1
N< (E) = 3N
d3N qd3N p
2
h
pi /2m<E

N
V
= 3N
d3N p
h
p2i <2mE

V N S3N ( 2mE)3N /3N

=
h3N
R
where Sn is the solid angle in n dimension and 0 Sn rn1 dr = Sn Rn /n is the volume of a n-dimensional
ball of radius R. The number states between E and E + E is

V N S3N ( 2mE)3N 2 m
(E) = N< (E + E) N< (E) =
E
(1.3.1)
h3N

To obtain Sn , we note

dn xex =
2

Sn rn1 drer

1
2
= Sn (r2 )(n2)/2 dr2 er
2
1
= Sn (n/2) = n/2
2
2

We nd that
Sn =

2 n/2
(n/2)

(1.3.2)

(1.3.3)

The entropy can now be calculated as

1
kB
S(E)

3N
3N
ln N + 3N ln((2m)1/2 /h) + 3N ln
ln(3N/2) + 3N/2 + ln E
2
2
v(2m)3/2
3 2 3
=N ln N + N ln
+ N ( ln
+ ) + ln E
(1.3.4)
h3
2
3
2
=N ln N + N ln v +

where v = V /N is the volume per particle and = E/N is the average energy per particle.
A big problem, the entropy is NOT extensive due to the N ln N term. We need to use a concept from
quantum physics - identical particle. For identical particles

1
N< (E) = 3N
d3N qd3N p
(1.3.5)
h N ! p2i /2m<E
Using ln N ! = N ln N N , we nd
1
kB
S(E) =N ln

v(2m)3/2
3 2 5
+ )
+ N ( ln
3
h
2
3
2

(1.3.6)

For identical particles, the entropy is extensive. The entropy per particle, s, is given by
1
1
kB
s =kB
S/N = ln

v(2m)3/2
3 2 5
+ ( ln
+ )
3
h
2
3
2

v(2m)3/2
h3
v
= ln 3

ln

(1.3.7)

Meaning: average energy per particle. (2m)1/2 the corresponding momentum. = h/(2m)1/2 the
corresponding wave length. v/3 number of wave packets that can be tted into the volume per particle.
Classical gas: v/3 1.
Quantum gas: v/3 1.
(Question: is air at room temperature a quantum gas or a classical gas?)
Thermodynamical function E(S, V, N )
From =

h2
e2s/3kB
2mv 2/3

we get
E(S, V, N ) = N

h2 N 2/3 2S/3N kB
e
2mV 2/3
9

(1.3.8)

The equation of state: The temperature

T =

E 
2
h2 N 2/3 2S/3N kB
N
e
 =
S V
3N kB 2mV 2/3

(1.3.9)

E 
h2 N 2/3 2S/3N kB
2
N
e
 =
V S
3V 2mV 2/3

(1.3.10)

The pressure
P =
We obtain the equation of state

P V = N kB T

10

(1.3.11)