10.

40 Appendix
Connection to Thermodynamics and Derivation of
Boltzmann Distribution
Bernhardt L. Trout

Outline
Cannonical ensemble
Maximum term method
Most probable distribution
Ensembles continued: Canonical, Microcanonical, Grand Canonical, etc.
Connection to thermodynamics
Relation of thermodynamic quantities to Q

0.1

Canonical ensemble

In the Canonical ensemble, each system has constant N, V ,and T.

After equilibration, remove all of the systems from the bath, and put them all
together:
Apply postulate 2 to the ensemble of systems, also called a supersystem.
P
P Let nj = number of systems with energy E j . Also, N = j nj and E tot =
j nj E j .
If we know all E j s, then the state of the entire ensemble would be well-defined.
For example, lets analyze an ensemble with 4 systems, labeled A, B, C, and D,
where
A
E2

B
E3

C
E2

D
E1

Then, E tot =E 1 + 2E 2 +E 3

Also, the distribution of the systems,

n = (n1 , n2, n3 , ...) = (1, 2, 1).
But there are many dierent supersystems consistent with this distribution. In
fact, the number of supersystems consistent with this distribution is
N!
4!

tot (
n) = Q
=
= 12
n
!
1!2!1!
j j

What is the probability of observing a given quantum state, e.g. E j ? In other

words, what is the fraction of systems in the ensemble in the state E j ?
1

10.40: Fall 2003

Appendix

(Ensemble of systems)
N,V,EA

N,V,EB

...

N,V,EC

...

N,V,Ej

(Bath at temperature T)
Figure 1: Canonical ensemble
The answer is

nj
N.

However, it may be the case that many distributions fulfill the conditions of the
ensemble, (N, V , E tot ).
For example, assume that there are two:
n1 = 1, n2 = 2, n3 = 1; tot = 12
and
n1 = 2, n2 = 0, n3 = 2; tot = 6
Then the probability of observing, for example, E 3 is 14 in the first distribution
and 12 in the second.
The probability in the case where both distributions make up the ensemble is:

1 1 12 + 2 6
1
p3 =
=
4
12 + 6
3
In general,
pj =

1
N

tot ( n )nj ( n )
nP

n tot ( n )

where the sum is over all distributions satisfying the conditions of (N, V , E tot ).
Then, for example, we could compute ensemble averages of mechanical quantities:
X
E = hEi =
Pj E j
j

10.40: Fall 2003

Appendix

N,V,EA N,V,EB N,V,EC

...

...

N,V,Ej

Figure 2: Canonical ensemble forming its own bath

and
P = hP i =

pj Pj ,

where p is the pressure.

0.2

Maximum term method

Recall:
pj =
where

1
N

tot ( n )nj ( n )
nP

n tot ( n )

N!

tot (
n) = Q
.
j nj !

As N , nj , for each j.
Thus, the most probable distribution becomes dominant. We can call this dis
tribution,
n .

Let nj = nj in the
n distribution. Then
pj =

0.3

nj
1 tot ( n )nj
=

N tot (
n )
N

Most probable distribution

Which distribution gives the largest tot ?

Solve via method of undetermined multipliers:
Take natural log of tot .

X ! X ! X
N!

ln (tot ( n )) = ln Q
ni ln
ni
ni ln ni ,
=
i ni !
i
i
i

where we have switched the index from j to i and used Stirlings approximation,
which becomes exact as ni :
ln y! y ln y y.

We wish to find the set of nj s, which maximize tot (

n ) and hence ln(tot (
n )) :
#
"
X
X

n ))
ni
ni E i = 0 , j = 1, 2, 3, ...
ln (tot (
nj
i
i
3

10.40: Fall 2003

Appendix

where and are the undetermined multipliers. Carrying out the dierentiation
yields

!
X
ln
ni ln nj E j = 0, j = 1, 2, 3, ...
i

or

nj = N e eE j , j = 1, 2, 3, ...
Recalling that

N =
yields

nj

e eE j = 1

or

e =

eE j .

Also,
hEi =

nj E j
N

and
pj =
where

N e eE j E j
N

E j
Ej
je
= P E
j
je

nj
eEj
= e eEj = P Ej ,
N
je
Q=

eE j

and, as we discussed in the last lecture, is the partition function, the normalization
factor.

0.4

Canonical ensemble continued and connection to thermodynamics

Recall from last time, via the maximum-term method in the canonical ensemble:
P
P
P E
E j
jE
e
Ej
j
j nj E j
j Ne
je
hEi =
=
= P E
j
N
N
e
j
and

pj =
where,

nj
eE j
= e eE j = P E ,
j
N
je
Q=

eE j ,

as we discussed in the last lecture, is the partition function, the normalization factor.
In addition, as we have shown:
E = hEi =

X
j

pj E j

10.40: Fall 2003

Appendix

and
P = hP i =
where P is the pressure.

pj Pj ,

If we dierentiate the equation for hEi,

X
X
dhEi =
E j dpj +
pj dE j
j

X E j
1X
(ln pj + ln Q) dpj +
pj
dV .
j
V N
j

Recall that the pressure,

P =
or

Pj =
This yields for equation 1
dhEi =

(1)

E
V

E j
V

X
1X
1X
ln pj dpj
ln Qdpj +
pj Pj dV .
j
j
j

[Note that

X
d
pj ln pj
j

ln pj dpj +

pj d (ln pj )

ln pj dpj +

Since,

pj

dpj
.
pj

(2)

pj = 1,

dpj = 0.]

Thus, the right term in equation 2 is equal to 0. This yields

1 X
pj ln pj hP idV .
dhEi = d

Recalling from the combined first and second laws (in intensive form, noting that
since N is a constant, intensive and extensive forms are equivalent):
dU = T dS P dV

Since U hEi and p hpi,

1 X
T dS d
pj ln pj .

j
5

10.40: Fall 2003

Appendix

Let
X =

pj ln pj .

Then,
dS =

1
dX.
T

(3)

We know that the left side of the equation is an exact dierential, so the right side
1
must be a function of X. This means that
must be too, and thus, T
dS = (X)dX = df (X).
Integrating,
S = f (X) + const,

(4)

where we can set the arbitrary constant, const, equal to 0 for convenience.
Now we can make use of the additive property of S, and we can divide a system
into two parts, A and B. This yields:
S = S A + S B = f (X A ) + f (X B ).
Note that
X A+B =

(5)

pi,j ln pi,j ,

i,j

where i is the index for the possible states of A and j is the index for the possible
states of B. Then
X
B
A
B
X A+B =
pA
i pj (ln pi + ln pj )
i,j

X
i

A
pA
i ln pi

B
pB
j ln pj

= XA + XB.
Thus, from equation 5,
S = f (X A ) + f (X B ) = f (X A + X B ).
For this to be so,
f (X) = kX,
where k is a constant. Thus,
S = k

pj ln pj .

From equations 3 and 4,

1
= k,
T
and thus,
1
.
kT
We designate k as Boltzmanns constant, a universal constant.
=

(6)

10.40: Fall 2003

0.5

Appendix

Microcanonical, Grand Canonical, and other ensembles

Recalling the formulation for S from equation 6, and noting that in the microcanonical ensemble,
1
pj = ,

where we recall that is the total number of states with the same energy, then
S = k

X
j

pj ln pj = k

X1
1
ln

= k ln (N, V , E).
This is Boltzmanns famous formula for the entropy.
In the Grand Canonical ensemble, the number of particles in each system is
allowed to fluctuate, but is kept constant. This is called the (V , T, ) ensemble.
Also, there are other ensembles, such as (N, P, T ), etc. Note that from an analysis
of fluctuations (Lecture 27), we shall see that in the macroscopic limit of a large
number of systems, all of these ensembles are equivalent.

0.6

Relation of thermodynamic quantities to Q

Recall that
S

= k

pj ln pj

eE j
Q
X
Q =
eE j

pj

Plugging in the formula for pj into that for S yields

S

X eE j

= k

eE j
Q

X eE j E j

ln Q
Q
kT
j

= k
=

ln

hEi
+ k ln Q
T

Recalling our definitions from macroscopic thermodynamics and the fact that U
hEi yields
A= kT ln Q
Similarly,
S
P
U

A
T

= kT

V ,Ni

ln Q
T

V ,Ni

ln Q
V
T,Ni
T,Ni

ln
Q
= A + T S = kT 2
.
T
V ,Ni
=

A
V

= kT

+ k ln Q

10.40: Fall 2003

Appendix

Thus, all thermodynamic properties can be written in terms of the partition function, Q(N, V , T )!
In order to compute Q, all we need are the possible energy levels of the system.
We can obtain these from solving the equations of quantum mechanics.