Physics 971

Statistical Mechanics

Lecture 1
Important experimental fact: we can answer many questions concerning systems in or near
thermodynamic equilibrium without solving the equations of motion of the particles in detail.
Thermodynamic Equilibrium
1. A system in contact with a heat reservoir for a long time.
2. A large isolated system.
Experimental Observation
Pressure of a fixed quantity of gas at constant temperature same as in 1914 and in 2014.
Measurements are done over time intervals covering millions of atomic collisions or vibrations --- i.e., a
time average. What is going on with the system during this time?
Consider Hamiltons equation of motion --- position q and momentum p are conjugate variables.
N atoms 3N coordinates and 3N momenta.
Evolution graphically shown as a single orbit in the 6N dimensional space of ps and qs. Such a space
is known as phase space or space.
[p]
[q]

3N momenta
3N coordinates

Experimental Measurement: time averages over a segment of the orbit in the phase space of the system.
Example of phase space:
Hamilton H = T + V =

Linear Harmonic Oscilllator

2

p
1
+ kq 2
2m 2
1

What is the phase space?

H p
+
=
q =
p m
H

=
kq
p =
q
k
q + q =
0
m
q=
A cos (t + )

k/m
=

p=
m A sin (t + )
2

q p
1
+
=
A m A
Energy of the System
E is a constant of motion as the Hamiltonian does not have explicit time dependence.

H
dH H H
= +
q +
p
t
q
p
dt
H H H H H
= +
+

t
q p p q

Hamilton's Eqn. of motion

H
t
= 0 if there's no explict time-dependence
=

E=

1
m 2 A2 (can be calculated from conservation of energy)
2
2

q p
+
=1
A m A

q2
p2
+
=1
( 2E / m 2 ) 2mE

Phase space is an ellipse.

We need to compute time averages over a segment of the orbit in phase space to find any measured
quantities. But time averages are difficult to perform

x (t ) 6N dimensional vector (3N positions, 3N momenta)

: some observable

1
T

t 0 +T

dt ( x (t ) )

T >> , the relaxation time characteristic time required for fluctuations in the properties of the system
to die out.

Since we cannot solve the detailed dynamics for x (t ) we cannot compute the time average.
2

Gibbs: systems appear to randomize themselves between observations (T >> ) . Instead of computing
a time average, imagine a group of similar systems (satisfying all external requirements) but
suitably randomized, and take averages over this group at one time.
The group of similar systems is called an ensemble of systems and is to be viewed as an intellectual
construction to simulate and represent at one time the properties of the actual system as developed in the
course of time.
The ensemble is randomized suitably in the sense that every configuration of q and p accessible to the
actual system in the course of time is represented in the ensemble at one instant of time.
Problem of Demonstrating time average ensemble average is the subject of Ergodic theory, which we
will not go into in detail.

We need to specify an ensemble through the probability density 0 ( x ) for a selected member of the

ensemble to occupy the phase space point x . Then the ensemble average

= dx 0 ( x ) ( x )
normalized
I will introduce here a simple system of a collection of spins to discuss the concept of ensemble.
Question: As q, p move via Hamiltons Eqn. how does the ( p, q; t ) change?
Liouvilles Theorem:

(q, p; t ) d 3 N q d 3 N p = # of systems in the volume element d 3 N q d 3 N p at any time t around the point
q, p

f (q, p) (q, p; t ) d q d
=
(q, p; t ) d q d p
3N

Physical quantity f (q,p) ;

3N

3N

3N

If the total # of systems in the ensemble does not change with time then must satisfy the continuity
equation.

d = d 3 N q d 3 N p
3

Change

d
t
flow out

( v n ) d

( v ) d

=
+ ( v) 0 ;
v ( q , p )
t
space dependent density

We want the density function to be stationary so that < f > does not change w/time (equilibrium)

=
0
t

Thus

( v) =0
3N

i =1

0
( qi ) + ( p i ) =

pi
qi

3N

q p
3 N

qi +
p i + i + i = 0
pi i 1 qi pi
=i 1 =
qi

qi =

H
H
; p i =
pi
qi

Hamilton's equation of motion

(Note that v = 0 for an incompressible fluid; here it corresponds to the conservation of phase space
volume)
3N



qi +
p i =
0
pi
i =1 qi

v = 0

Equivalently,

H H

pi qi
Liouville's Theorem
i =1 qi pi

[, H ] =
0 Poisson's Bracket
3N

Choices of (q, p )
1. ( q, p ) =constant (Microcanonical Ensemble)
Uniformly distributed over the region of phase spaceuniformly distributed over all possible
microstates.
2. (q, p ) = [ H (q, p )] i.e. any function of the Hamiltonian. Then the Poisson bracket is satisfied
We will see later that a natural choice is

(q, p) e H ( q , p )/ kT (Canonical Ensemble)

The Microcanonical Ensemble

Number of molecules N, volume V, and energy E --- all fixed.
Motion of the N particle system will be confined to a (6N-1) dimensional hyper surface given by
--- we will call this surface ( E ) .
H ( q, p ) = E
Instead of fixing the energy exactly at E we consider it to be

1
1

E E H (q, p ) E + E
2
2

Actually this particular form does not matter. One can equivalently consider

( E E ) H (q, p) E ,

or

E H (q, p ) ( E + E )

With a specified macrostate, a choice still remains for the system to be in any one of a large # of
microstates.
Uncertainty or tolerance E must be small i.e., it should not play any role in the thermodynamic limit

N , V , const . This would require E E. On the other hand, we

V

E
E
need E to allow many microstates. So we should have
E E .
N
N
Probability density function

1
1

=
constant C if E E H (q, p ) E + E
( q, p )
2
2

= 0
otherwise

 Equal a priori probabilities for different accessible regions of equal volume in the phase space.

During any significant time in interval the phase point x (t ) will spend equal time intervals in all
regions of ( E ) --- Ergodic Hypothesis
This hypothesis is sufficient to allow us to replace a time average by an unweighted average over the
surface ( E ) . However this is not necessary. A hypothesis that is weaker than the ergodic hypothesis
but still leads to Statistical Mechanics as we know it is called the mixing hypothesis. Mixing refers to
the notion that an initial compact distribution of points on ( E ) very quickly distorts into a convoluted
object which permeates the entire surface while still occupying the same volume as required by
Liouvilles theorem. (For more on this see the book by Balescu).

Lecture 2
Density of States
Consider N distinguishable particles (such as a perfect crystal or a polymer chain --- particles that can be
identified by their position). Volume of the phase space accessible to the system is the integral

1
( E ) =
C

d 3N q d 3N p

E 1 E H ( q , p ) E + 1 E
2
2

1
Make ( E ) dimensionless; ( E ) =
h3 N

'

d 3 N q d 3 N p , where h is the unit of action and

'

means

total # of states with the energy in the specified range.

Choice of h is to make contact with Bohr-Sommerfield semi classical theory
principle p q h

pdq = nh and uncertainty

If we are dealing with indistinguishable particles, we do not distinguish between states which differ only
by the labeling of the particles occupying the various regions of phase space. In this case our measure of
the number of available states in the ensemble becomes
' 3N
1
( E ) =
d q d 3N p
3N

h N!

Historically Gibbs introduced the N! term to remove a spurious entropy of mixing (Gibbs paradox)

0 ( E ) = Total # of states for H E .

0 ( E )
Density of States:
g (E) =
E
If E is small, ( E ) = g ( E ) E

Entropy in Statistical Mechanics

Thermodynamics tells us that entropy is a state function in the sense that the value of the entropy does not
depend explicitly on the past history of the system but only on the actual state of the system. However, in
thermodynamics we do not have the tools which would enable us to understand physically what entropy
means in terms of the condition of the system.
We define=
S k n (without any additive constant. This is absolute value.)

We need the following properties of entropy to be satisfied:

1.

Q
dS is an exact differential =

T
1

S= S1 + S2 (additive)
3. S 0 for an isolated system (thermally closed)

2.

1. Entropy as defined above has a definite value for an ensemble in statistical equilibrium; thus the
change in entropy is an exact differential. If we define ( E ) as a measure of the imprecision of our
knowledge or as a measure of randomness, then entropy also is a measure of these things.
2. Non-interacting system made of two parts with N1 particles and N 2 particles

= 1 2
Phase space of the combined system is the product spaces of the individual parts

1
1st system 2
3

4
2nd system 5
6

Total system will have a total 9 states (such as 1,4; 2,5; etc.)
Then S= S1 + S2 (Note that N! to be replaced by N1 ! N 2 ! )
3. We need to assume that the equilibrium condition of a system is given by the most probable
condition. Then the volume of phase space with equilibrium properties will be a maximum. And the
entropy of a closed system has its maximum when the system is in the equilibrium condition.

Setting up Thermodynamics
S = S ( E , x, N i )
# of i-th chemical species

external parameters like Volume V, Magnetization M

For simplicity, consider x V

S
S
dS =
dE +
E
V
dS =

dV +

i
N i

dN i

dE P 1
+ i dN i
T T T i

and Chemical Potential

i
T

S
N i

Consider N i fixed and compare with the 2nd Law of Thermodynamics

TdS
= dE + PdV

2nd Law of thermodynamics

1 S
=
T E

Then , Temperature

Pressure

Generalizing further, Chemical Potential

i
T

P S
=
T V

S
N i

Entropy of a perfect gas using the Microcanonical Ensemble

N non-interacting point particles

1 3N 2
pi
2m i =1

H ( q, p )

with constraint

E E H (q, p ) E

'
VN '
VN
1
dq
dq
=
dp
dp
=
dp
dp

dp1 dp3 N
1
3N
1
3N
1
3N
h3 N N !
h3 N N !
h3 N N !
with the constraint 2m( E E ) pi2 2mE

=
( E )

The accessible volume is that of a 3N dimensional hyper shell bounded by two hyperspheres of radii

2m( E E ) and

2mE

Volume of a Hypersphere
Equation of hypersphere

x12 + x22 + + xn2 =

R2

Start with the Gaussian Integral

I =

x12 + x22 ++ xn2

e x=
dx

dx1 dxn

)
(=
n

n /2

But in terms of sphere radius R (polar co-ordinate)

I =

e R

Cn R n 1dR

voln element dV V= Cn Rn1dR

0

Make a change of variable t = R 2

n 1
t
Cn t n 2 2
1 1/2
2
C=
e
t
t
dt
e t dt
n 0
2
2 0
C n
= n Gamma Function
2 2

Cn n
n
1 ! if is an integer
2 2
2

2 n /2
2 n /2
=
Cn

n
n
1 !
2
2
Then
=
V ( R)

n /2

n 1
C=
dR
nR

n /2

=
Rn
Rn
n
n

+ 1
!
2
2

Just consider V ( R ) = CR n
Volume of a shell of thickness s

VS = V ( R) V ( R s)
= CR n C ( R s ) n
s n
= CR 1 1
R
ns / R n
= CR n 1 1

n

n

x
Note that: lim n 1 =
e x
n
n
CR as long as n s R.
CR 1 e nS / R
n

Practically the volume of the whole sphere

n 3=
N ~ 1023
R
=
s

2mE

E )
2m ( E ) 2m ( E=
2mE

E
2mE 1 1

1 E E

m
2 E
E

Then the condition n s R leads to N

E
E
E

N E E E

E
N

We have already discussed this before.

VN
3 N /2
3 N /2
2mE )
(
3N
N ! h 3N

!
2

Then ( E ) =

Lecture 3
n 3=
N ~ 1023
R

2mE

E )
2m ( E ) 2m ( E=

=
s

2mE

2mE 1 1

1 E E

m
2 E
E

Then the condition n s R leads to N

E
E
E

N E E E

E
N

We have already discussed this before.

3 N /2
2mE )
(
3N
N ! h 3N

!
2

Then ( E ) =

S k n
=

3 N /2

S
3/2
V
3N
= N n 3 3/2 ( 2mE ) n( N !) n
!
k
h

2
Stirlings Formula n N ! N nN N

3N 3N 3N
3/2
V
n
+
N n 3 3/2 ( 2mE ) N nN + N
2
2
2
h

3/2

3/2
1 5N
3/2 E
V

= N n 3 ( 2 m )
+
3/2
N h
N 3 2

3/2
3/2

4 m

S Nk ln
=
13 V E + 5 Nk
3
h N N 2

S
k

Now

1 S

3
31

= =
nE Nk

Nk =
T E V , N E
2
2E

and

P S
=
T V

= Nk
V
N ,V

PV
= NkT

=
E

3
NkT
2

(1)

3
NkT , one can write,
2
3/2
2

4 m 3
V 5
S = Nk n
kT
+ Nk
2
2
N 2
3 h

h
where the denominator is the "average thermal momentum" of a molecule
=
1/2
( 2 mkT )

Using E =

(Note: actually the r.m.s. velocity is given by v =

8kT
)
m

V / N 5
=
S Nk n 3 + Nk
2
Ratio of volume per particle/volume associate with de Broglie wavelength
1/3

V
(Thus the classical description is valid for R where R
N

is the mean n-n distance).

Chemical Potential of a Perfect gas:

S
=

T N E ,V
We rewrite Eq. (1) as
5
S= Nk n A N 5/2 + Nk
2

5
S
5
= k n[ AN 5/2 ] k + k = k n[ AN 5/2 ]
N
2
2
N3
= + kT n

V
Note that PV = NkT

P 3
Thus,
=
kT n =

kT nP + f (T ) , a very useful expression.

kT

Other Examples of Non-Interacting Systems

2.

Two level systems.

N magnetic moments --- Classical spins; distinguishable
B applied magnetic field

Each spin can be parallel or antiparallel to the field

E=
=
0
AP
0B

0 B =
0
EP =

Two energy levels.

Fix the energy to some value E

= M 0 Note that
2

 N , N + 2, , N
M=

all spin parallel

all spin antiparallel

=
N + # of spins in + 0 state
Now

N = # of spins in - 0 state

=
E N + 0 N 0
M = N+ N

How many possible ways of choosing N particles out of N so that the energy is still M 0 ? This
number is the # of microscopic states.

=
=

N!
( N )!( N N )!
N!
N ! N+ !

(Interchanging among N or N + gives nothing new)

Now

1
1
N =
( N M ) and N + =+
(N M )
2
2
=
S k n
1
1
1
1

k N nN ( N M )n ( N M ) ( N + M )n ( N + M )
2
2
2
2

1
1

Note: Terms N + ( N M ) + ( N + M ) canel

2
2

1 S
1 S
= =
=

T E N ,V 0 M
1
k 1 1
1
=
n ( N M ) ( N M )
2
0 2 2
1
( N M
2

1
1
1
1
1
1

)
(
)
(
N
+
M

n
N
+
M

1
2
2 2 2
( N + M ) 2
)
2

1 k
N M
n
2 0 N + M

1 1 k
N M
= n
T 2 0 N + M
Note for M > 0 ( E > 0) T < 0 (Negative temperature!)

The system is not normal in the statistical sense, but for M < 0( E < 0) the system is normal

N M
= e 2 0 /kT
N +M

N
N+

Now
=

N
e 0 / kT
= / kT
N e 0 + e 0 / kT

N
e 0 / kT
and + = / kT
N e 0 + e 0 / kT
Then

Probabilities of finding any

one particle in the states

Canonical distribution.

N 0 tanh ( 0 / kT )
E=
( N+ N ) 0 =

Specific Heat

C=

dE
dT

= Nk 0 / cosh 2 0
kT
kT
2

Schottky specific heat

anomaly characteristic of
systems with a gap excitation
energy E 2 0

3.

Generalization of the 2-State System

K-States i

ni particles in each state

N!
N!
=

=
n1 ! n2 ! nK ! ni !

=
S k n

= k N n N N ni n ni + n i

= k N n N ni n ni

Constraints:
=
E

ni i
N
=

=i 1 =i 1

Maximize S =
S ni ni i
S
=0
ni

ni

( , are Lagrange's Multipliers)

k {n ni 1} i =
0
1

ni =
e

e i / k

= C e i / k
But ni = N
=
C

N
N
=
i / k
Q1
e

Note also, n C =1

=k (1 + n C ) =k 1 + n

Q1

Then

=
S k { N n N ni n ni }

N i / k
=
k N n N
e
[n( N / Q1) i / k ]
Q1

N
N
N

n
=
k N n N
e i / k +

Q 1 Q 1 Q 1 k

"Q1

Note that the last term is

ni i as ni =

ie

i / k

N i / k
e
Q1

S = k N n N N n N + N nQ1 + E
k

= Nk n Q1 + E
1 S
Note that = = ; ni 's are the only variables, so Q1 is ind of E
T E
E S =
Nk n Q1
E

S=

Nk

n Q1

F
But we know, E TS =
1
E TS =
NkT n Q1
T
We will learn later, F =
kT ln QN =
kT ln ( Q1N ) =
NkT ln Q1

Lectures 4 and 5 (Lecture 4 was partly HW discussion)

Harmonic Oscillators
Distinguishable
Energy of each one (i-th one)
N

Total energy

i =1

ni + hv
2

ni =
0, 1, 2

ni + hv
2

1
N hv + Mhv
Fix E =
2

So that M = n1 + n2 + + nN

and

M=

E N

hv 2

( E ) =
# of ways distributing M particles in N (labeled boxes) (a box may be empty)

This is same as M while balls + (N-1) Black balls which work as the barriers.
Running index 1,2...M+N-1.

( M + N 1)!
( E ) =
M !( N 1) !

To see that this counting makes sense, consider a simple example of M = 6 and N=3.
The above counting says that there are
What are these states?
0
0
6
1
0
5
2
0
4
3
0
3
1
1
4
2
1
3
2
2
2
Total

( 6 + 3 1)!
6!( 3 1) !

= 28states

(3 states of these types; remember the harmonic oscillators are distinguishable)

(6 states)
(6 states)
(3 states)
(3 states)
(6 states)
(1 state)
28 states

(Q: if the harmonic osc. are instead indistinguishable, how many states are there for general M and N?)
Now, going back to Stat Mech.

=
S k n
N >> 1

M >> 1

S= k {( M + N ) n ( M + N ) M n M N n N }
1 S S M
= =
T E M E
k
M +N
=
n

hv M

Use M
=

E N

hv 2

N
E N

+N
E
hv
+

k
k
2
hv
2
= =
n
n

hv E N hv E N hv
2

hv 2
1
E 1

E + + Nhv
N + 2 + hv
1 k
2 = e hv / kT
=
n

1
T hv E 1 hv
E Nhv
2
N 2

hv
1

=
E N hv + hv / kT

1
e
2

Canonical Ensemble
Difficult to control total energy E; fixed temperature is more practical.
On the technical sidemicrocanonical ensemble is difficult to use in practice because of difficulty in
evaluating the volume of phase space or number of states accessible to the system.

Total system can be expressed by a microcanomical ensemble with total Energy Et

t ( Et ) volume of the phase space accessible to the total system for Energy between Et and Et + Et
The probability that the total system is an element of volume dt of the appropriate phase space is

d t = C d t if the energy between Et and Et + Et

=0 otherwise
What is the probability that the subsystem is in d s without specifying the condition of the reservoir?

d s= C d s r
volume of the phase space of then reservoir
(so that total energy is in Et about Et ).
Our task is to evaluate r --- if we know the subsystem is in d s , how much phase space is accessible to
the reservoir? (I will discuss a spin-system example to elaborate)
Entropy of Reservoir

S=
r k n r

Sr / k
=
r e

Now E=
r Et Es
We may take Es << Et
Expand:

Sr ( Er )= Sr ( Et Es )= Sr ( Et )

Sr ( Et )
Es +
Et

Then

1 Sr ( Et )
S (E )
=
r exp r t exp
Es
Et
k
k

Now Et Er

1 Sr ( Et )
=
Et
T
r= A0 e Es / kT
Reservoir
Then
Es / k BT
d =
s C d s A0 e

= A e Es / kBT d s
may be viewed as a normalization constant.
3

For the subsystem then the probability density

( E s ) = A e Es / k B T

Normalizing

( Es ) =

e Es

e Es

e Es
Q

where Q = e Es for discrete energy states

Thermodynamic Functions

Entropy is fixed by the energy independent of whether the system is isolated or in contact with a heat
bath.

Canonical ensemble entropy for mean energy E = Microcanonical ensemble entropy for energy E

( E ) # of states E , E + E
0 ( E ) # of states H E
g ( E ) density of states =

0
E

If E is small enough g ( E ) E = ( E )

=
S k n ( E )
0
( E ) =
(1)
E E

E
What is E -- the range of probable values for canonical ensemble?
Let p ( E )dE be the canonical ensemble probability distribution that the system will have energy in the
range dE at E . Then

p=
(E) (E) g (E)

probability
density

= (E)

density of states
at E (degeneracy for

discrete system)

0 ( E )
E

p( E )

sharply peaked at E and normalized

p( E ) E = 1

discuss

 ( E )
(E) 0
1
E =
E E

(2)

Then from (1) and (2)

( E )=

1
= A1 e E / k BT
(E)

Mean energy E U

=
S k n ( E )
= k n A+

E
T

F = E TS = kT n A A = e F / kT
Partition Function and Helmholtzs Free-energy

Q = e Ei / kT =
i

1
A

F = kT n Q
dF =
SdT PdV + dN

S=
k ( T n Q )

T N ,V
T
N ,V
F

P=
kT

=
n Q

V N ,T
V
N ,T
F

V ,T

n Q

kT
=
=
N
N
Energy

Write =

1
kT

Es e E

E U =
e E

U =+
F TS =
kT n Q + kT (T n Q ) =
kT 2
( n Q ) =

( n Q )
T

1 Q
=

( n Q )

Alternatively, from the definition of Q , E U =

2-Level Systems
Classical magnetic moments in an external magnetic field H; can only be up or down; energy of 2 states
+ 0 H and 0 H or 0 H i where i = 1 The classical statistics should work at high temperatures.

Hamiltonian H = 0 H i
0 H i

=
e

=
QN

{ i }

{ i }

( e

e H i =
i

Make sure you follow the difference between

whole system and

{ i }

0H

) (

+ e 0 H = e 0 H + e 0 H

QN =
e + e
=
Q1N

(Distinguishable)

Q1

which is the sum over all possible i in the

which is the sum over all possible values of the i-th spin i

Energy + and for each moment

e 0 H i

Lecture 6
2-Level Systems (Continued)
Partition Function
Discrete States

QN (V , T ) = e Es

This sum is over all possible states

{s}

gse

Es

This sum is over distinct energy states with degeneracy g s

Ensemble average of any observable, f can be written as

e Es

f
< f >=

{s}

QN (V , T )

Partition Function of a Two Level System from Counting Degeneracy

QN = g s e Es
s

Es = N1 0 + N 2 ( 0 )
N
= N1 + N 2
gs =

N!
N 1! N2 !
N!
e N1 0 + N 2 0
N 1! N2 !

QN =
s

N!

e N e N e N
=
N !( N N ) !
N1

Binomial Expansion

e N 0

N1

N!
e( 2 0 )
N !( N1 N1 ) !

N!
x n (1 + x) N
n !( N n ) !

QN =
e N 0 1 + e2 0

) { (
N

e 0 1 + e 0
=

)} {
N

e 0 + e 0
=

Continuous States
e H
=
(E)
=
< f >
;
QN (V , T )

f e H ( q, p ) d
QN (V , T )

N1

=
QN (V , T )

N !h

3N

e H ( q, p ) d

Or in terms of Density of States, QN (V , T ) =

e E g ( E ) dE

Perfect Gas
H ( q, p ) =
i

Q=
N (V , T )

pi 2
2m

1
N ! h3 N

Gaussian Integral
I ( )
=
Here =

2m

pi 2

d 3=
qi d 3 p i
i

VN
N ! h3 N

e pi 2 /2 mkT dp
i

x 2

e
dx
=

1
Hence,
2mkT
N

1 V
1 V

=
QN (V , T ) =
( 2 mkT )3/2

3
N!h
N ! 3

Thus for non-interacting systems one can write, QN (V , T ) =

h2
N
kT n QN =
F ( N ,V , T ) =
NkT n

V 2 mkT


h2
N
F

=
=
kT
n

N V ,T
V 2 mkT
NkT
F

=
P=

V
V N ,T

3/2

V 2 mkT 3/2 5
F

=
S=
Nk n

+
N h2 2
T N ,V

U=
( n Q )

3
F + TS ) =
NkT
(comes from U =
2

Partition Function from Density of States

QN (V , T ) =

e E g ( E )dE

3/2

Q1N
N!

3N

0 ( E ) compute as we did before in Microcanonical

V N 3 N /2
3 N /2
0 ( E ) = 3 N
2mE )
(
N ! h 3N !

2
0 ( E ) 1 V
g (E) =
=
N ! h3
E
The integral for QN (V , T ) is of the form

( 2 m )3N /2

3N
1!

3N
1
E 2
=

Ae

3N
1
E 2 dE

e E E dE . Note that the function is defined as

x
( + 1) =
e x dx
0

Make a change of variable, E = x .Then one finds,

( + 1)
!
1
e E E dE = +1 e x x dx = +1 = +1
0

QN (V , T )=

3N
1
E 2
e
E
dE

3N
1 !

= A
3N

1 V 2 m
QN (V , T ) =

N ! h3

3 N /2

Same as obtained by integrating over phase space volume

Lecture 7
Single-Particle Density of States
N

Q
We know that one can write, QN (V , T ) = 1 where Q1 is the single-particle partition function. In
N!

terms of single-particle density of states a ( ) , Q1 can be written as

Q1 =

e a ( ) d

Now, our task will be to evaluate this single particle density of states a ( ) .

0 ( P) : # of microstates available to one particle confined to volume V; its momentum p being P

=
0 ( P)

h3

3
d=
q d3 p

h3 0

4 p 2 dp
=

V 4 3
P
h3 3

pP
Now use energy-momentum relation to find 0 ( ) . For a non-relativistic particle, =

P2
and for an
2m

V 4
(2m )3/2 by
h 3

extreme relativistic particle, = Pc . For a non-relativistic particle, 0 ( ) =

3
using =

=
Q1

d 0 ( ) V
P2
. We then find the single particle density of states
=
a ( ) =
2 (2m)3/2 1/2
3
2m
d
h

3/2

V 2 m
V
= 3
e =
a ( )d
3

Q1N
1 V
=
=
Q
N
N ! N ! 3

Other Examples:
Free orientable classical dipoles (Paramagnetism)
N

H =

i H =
0 H cos i

=i 1 =i 1

QN

=
exp

H
cos

0
i

1
i

{ i}
{ i}

e 0 H cosi

i =1

e0H cosi
=
{ i} i 1 =i 1
N
= Q=
1 where Q1

Q1 ( )

H cosi

independent dipoles

e 0 H cos

4 sinh ( 0 H )

0 H

H cos
sin d d (sum changed into integral)
=
e 0
0

Magnetization = mean magnetic moment along the field direction Mz

Mz =

i =1

0 cos i

angular bracket stands for thermal average

F
F

comapre with P =

H
V

1
N
=
+
n QN =
n Q1
H
H

0 H

N
4 1
1
cosh( 0 H )

2 sinh( 0 H ) + 0
4 sinh( 0 H ) 0 H
H

1
=N
+ 0 coth( 0 H ) =N 0 coth( 0 H )

0 H
H

= N 0 L( 0 H )
Langevin Function

=
L( x ) coth x

High temp T large

x 0
or small field H 0
M z N 0

0 H
3

H
1
where x = 0
x
kT
x x3
L( x) +
3 45

N 0 2
H Susceptibility
=

=
3kT

2
C
M z N 0
lim =
=

3kT
T
H 0 H

( N N 0 for per unit volm calculation where N 0 is the Avogadro's #)

Paramagnetism (Quantum)
Magnetic Moment 0 = g B J where J is the angular momentum
Applied field is in the z direction

z = g B m

magnetic quantum #
m = J , J + 1, J 1, J

Curie's law

Q1 ( )
Q1 ( ) =

=
substitute x g B JH
e g B Hm

m = J
m= J

m = J

= e

emx / J = e x + e x e x / J + e x e 2 xJ + + e x e 2 J x / J

/J
1)
( e(2 J +1) x=

x/ J

1
sinh 1 +
2 J

1
x / sinh x

2J

F N

= n Q1 ( )
Mz =
H H
=
M z N=
x g B JH
z

z = 0 BJ ( x) where BJ ( x) is the Brillouin function given by

1

BJ ( x) =
1 +
2J

coth 1 +

2 J

1
x
coth
x
2J
2J

g 2 B2 J ( J + 1)
x 1
H
z =
3kT
C
Even in the Quantum case, susceptibility = , but the Curie constant is changed.
T
1 1
B J ( x) 1 + x
3 J

Classical Limit : J
BJ ( x) coth( x)

g B 0

g B J 0

1 cosh( x / 2 J )
2 J sinh( x / 2 J )

x
0
J
1 1
1
coth( x) L( x)
Langevin Function
x
2J x
2J
where x g=
=
B J H 0 H / kT . So we recover the Classical Expression.

BJ ( x) coth( x)

Molecule in an Ideal Gas (Maxwells Velocity Distribution)

One moleculeremaining ones constitute the reservoir

1 2 p2
=
E =
mv
2
2m

Probability P ( r , p ) d 3r d 3 p that the molecule has position lying in the range ( r , r + dr ) , momentum

( p, p + d p)

P ( r , p ) d 3r d 3 p d 3r d 3 p

( p /2 m )
e

does
not
depend
on r ; r can be
integrated out

P ( p ) d 3 p Probability that momentum is in the range ( p, p + d p) and that r could be anything

P ( p) d3 p

integral

over r

V e p

d 3r d 3 p e p /2 m

/2 m 3

d p

and V is a constant.

Since p = mv, Velocity Distribution is given by P (v)d 3 v e mv /2 d 3 v or,

P(v)= C e mv

/2

normalization constant
Molecule in an Ideal gas in the presence of gravity

p2
H
=
+ mgz
2m
We want to figure out pressure (z). As before,
2

P(r , p ) d 3r d 3 p d 3r d 3 p e ( p /2 m+ mgz )

P( p) d 3 p

d 3r e ( mgz ) d 3 p e

p 2 /2 m

integrate out

( p 2 /2 m ) 3

P( p ) d 3 p= C e
d p

same as before

Probability of finding a molecule between z , z + dz

x, y

P( z )dz

P ( r , p ) d 3r d 3 p

integral integral

= C e mgz dz
Then number density

n( z ) = n(0) e mgz
As Pressure n if Temp is constant

Pressure( z ) = Pressure(0)e mgz

Isothermal Atmosphere

Grand Canonical Ensemble

Es + Er = Et

S r = k n r

Ns + Nr =
Nt
d s ( N s ) =C d s ( N s ) r ( Nt N s )
= C e Sr ( Et Es , Nt N s )/ k d s

S ( E t , Nt )
S ( E t , Nt )
S r ( E t Es , N t
N s ) Sr ( E t , Nt ) r
Es r
Ns +
=

Et
Nt
Sr ( E t , Nt )

1
T

Es
+ Ns
T T

S
S
,
=

T N E ,V
E N ,V

=
Recall

Then d s A e
=

Es N s
+
kT kT d

= A e( N E )/ kT

e / kT ; Grand Potential
( E , N ) = e( + N E )/ kT

Grand Distribution Function

Grand Partition Function

1
e / kT
A

 / kT
Q e=
=

=
N

N s
/ kT N

(e )

( N E N ,s )/ kT
E

/ kT

e N ,s
s

QN (V ,T )

/ kT
Define
=
z e=

fugacity

Q = z N QN (V , T )
N

Connection with Thermodynamics

Similar to what we did for canonical ensemble

S = k n ( N , E )

= k ( + N E ) / kT

= E TS N
= dE TdS SdT dN Nd
d
Recall

S S ( E ,V , N )
S
S
S
dS =
dE +
dV +
dN
E
V
N
dE P

= + dV dN
T T
T

So, dE =TdS PdV + dN

d = PdV SdT Nd



P=

; S=
;N=

V T ,
T V ,
V ,
Perfect Gas

1 V
but QN
e
QN
=
N ! 3
N
N / kT

/ kT V
/ kT V zV / 3
/
!
exp
N
e
Q e =
=
=

3
3

N
V
V
kT n Q = kT e / kT 3= kT z 3

 / kT V
zV

=
e
But N =
=
3

3

= kT N
N3
and = kT n
V

1
N

P=
kT e / kT 3 =
kT

V
V T ,

PV =
NkT

Ideal Gas in Grand Canonical Ensemble (Continued)

Probability of finding a subsystem with n particles where the average # of particles in that
volume V is N
We do not care what the energy of the subsystem is, so

=
Pn

( E , n) d

energy
integrated out

+ n )/ kT E / kT
+ n )/ kT
e(
e(=
e
d
Qn

+ n ) / kT 1 V
Pn = e(
n ! 3

N e
=

Now

V
=

/ kT V

( N )n en / kT

n
1
N)
(
n!
n
1
Pn = e N ( N ) Poisson Distribution
n!

e / kT
Pn =
Since = kT N ,

Normalized such that

n =0

Pn = 1

Now Compute PN
Note that n PN = 0 !
PN = 1 !!

Show that

n =1

nPn = N

1
LHS =
e N n N n Substitute m =
n 1
nPn =
!
n
n 1=
n 1
=
=e

( m + 1)

( m + 1)! N m+1
m =0

Nm
N
N
= e=
m =0 m !
N

Adsorption of a Perfect Gas onto a Surface

Surface has N 0 adsorption sites (distinguishable) --- n gas molecules adsorbed on it. Molecules do not
move around.
Pressure of the gas is changed; how does the coverage =

0 is the adsorption energy per site

n
change?
N0

Consider first just the adsorbed phase (no interaction between adsorbed molecules) in Grand canonical
Ensemble

Q=

N0

n =0

e n / kT Qn

Qn = e E s
{s}
= g s e E s
s

n molecules are adsorbed Es = n 0

Note that: Only one state. What is the degeneracy of this state?

gs =

N0 !
n !( N 0 n)!

Qn =

N0 !
e n 0
n !( N 0 n)!

Another way of looking at it and computing the Canonical Partition function

( Ei =

0 )

E = Ei
i

Qn =
=
=

1
e Ei
n ! {molecules}

n ! mol 1

molecules are indistinguishable

e Ei

mol 2 mol 3

1
e E1 e E2 e En

n ! mol 1
mol 2
mol n

1
N 0 ( N 0 1) ( N 0 n + 1) e + n 0
n!
N0 !
Same as before
e n 0
=
n !( N 0 n ) !

Then Q =

N0

e n / kT

n =0

N0

n =0

N0 !
e n 0
n !( N 0 n)!

n
N0 !
e ( 0 + )

n !( N 0 n)!

+
= 1 + e ( 0 )

N0

+
= kT n Q = kT N 0n 1 + e ( 0 )

0+ )
N e (
1

+
n=
kT N 0

=
e ( 0 ) =0
+
+

1+ e ( 0 )
1+ e ( 0 )
+
N
N n
n
e ( 0 )
n
1
1
=
0 =+
0
=
( 0 + ) =
n
1
+
+
N0 1 + e ( 0 + )
n
n
N0 n
e ( 0 )
e ( 0 )
n
kT n
=
0
N0 n

=
n

kT
kT
n

+ 0 / kT
/ kT
( N0 n ) e
(1 ) e 0

|
Partition function of one single
adsorbate molecule.
(Problem 4.10 of Pathria)

Equilibrium demands ads = gas

0
(1 )

ads =
kT n

gas =
kT n P + f (T )

Then kT n P=
+ f (T ) kT n
0
(1 )

=
kT nP kT n
T)
+ kT n g (
(1 )
"
f (T ) 0

P=

g (T )

or , =

P
f (T )+ 0 }/ kT
where g (T =
) e {
(Problem 4.11)
P + g (T )

Approach the Problem from Canonical Partition Function

Qn =

N0 !
e n
( N 0 n )!n !

F = kT n Qn

=
kT {n N 0 ! n ( N 0 n ) ! n n !+ n}

=
kT N 0n N 0 N 0 ( N 0 n ) n ( N 0 n ) + N 0 n n n n + n + n
=
kT { N 0n N 0 ( N 0 n ) n ( N 0 n ) n n n + n}
F
=
kT {n n + } + n( N 0 n)

V ,T

ads =

N n
=
kT n 0

n
n
=
same as before
ads kT n
N0 n

Chemical Reaction in the Gas Phase

Mixture of different gases

Gas molecules are non interacting (Perfect gas) but chemically reacting
m different kind of molecules
Chemical symbols of these molecules A1, A2 Am

2 H 2 + O2 2 H 2O
2 H 2 + O2 2 H 2O =
0
m

j =1

a j Aj = 0

(1)

A1 H 2 A2 O2 A3 H 2O
a1 = 2

a2 = 1

a 3 = 2

N j = # of A j molecules in the system

| can change as a result of chemical reaction

Conservation of atoms: dN H 2O : dN H 2 : dNO2 = 2 : 2 : 1

Thus dN j a j for all j

(2)

Consider an isolated system. Equilibrium dictates dS = 0

S S ( E ,V , N1, N 2 ) but E and V do not change

0 j dN j =
0
dS =

(3)

[Note: (a) If instead of an isolated system, we consider T and V to be constant, then dF=0 which
yields
j dN j = 0 .

b) Similarly if we consider T and P to be constant, then dG=0 which again

yields
j dN j = 0 . ]

From(2) and (3)

m

j =1

a j j = 0 General condition for chemical equilibrium

(4)

Need to know j --- can use Canonical partition function to compute F and then j

Q1N
Recall QN (V , T ) =
N!
Since there are many species here, we change the notation a bit and write q = Q1 so that qi is the single
particle partition function of the i-th type.
1

Q1,11 Q1,22 Q1,33

(
,
)
Q
=
V
T
=
N
N1 ! N 2 !

Q1,ii
=
N!
i
i

qi i
Ni !

Ni T ,V , N

i =

F
P=

+ kT
n QN
=
V

F=
kT n QN =
kT

( Ni

n q i n Ni !) =
kT

i =
kT {n qi n Ni }

i = kT n

( Ni

n qi Ni n Ni + Ni )

qi
Ni

Now, we know

ai i = 0

( ai

n qi ai n Ni ) =
0
m

n N i i =
n qi i

a1 a2
N=
1 N2

=i 1 =i 1
a1 a 2
q1=
q2 K N (T , V )

Law of Mass Action.

K N Equilibrium constant of the reaction independent of the numbers of molecules N i present (as it
only depends on single particle partition functions qi s raised to ai s.)
a

Ni i
Equivalent way of writing
=1
i qi
V
=
For a Perfect
Gas qi =
and PV NkT
3

Note that, we are neglecting any binding energies here.

a

a
i
Ni i
Pi3
N
=

i
=
[ci P fi (T )]ai

NkT
qi

N
where ci =i
N

Then {ci P fi (T )} i =
1
a

=
ci i
i

P ai
=
g (T ) P
ai
fi (T )

ai

N
Pi
(Note that Pi V= Ni kT where Pi = Partial Pressure. Thus i= c=
or P=i Pci
i
N
P
and the above expression can be written in terms of Pi rather than in terms of ci )
2

Examples
1. H 2 + Cl2 2 HCl

a1 = 1

a2 = 1

a3 = 2

[ H 2 ]1 [Cl2 ]1 [ HCL ]2

Instead of ci we are now using [ ] = concentration

1+1 2
= P=
g (T ) g (T ) Independent of Pressure

2. H + H H 2

[ H ]2 = g (T )
[H2 ] P
At low pressure the system is largely dissociated as atomic hydrogen, whereas at high pressure the
formation of molecular hydrogen is favored. Binding energy favors molecular hydrogen; entropy favors a
large # of particles.
3. Thermal Ionization of Hydrogen Atoms
T is large [ H 2 ] is negligible

H p + + e

0 =ionization potential (ground state of hydrogen atom has 0 energy)

N p Ne
q p qe
= K=
NH
qH
Classical limit p + , e concentrations are small, T large; Coulomb interaction between separated
protons and electrons neglected.
V
3/2

qe = 2

( 2 mkT )

spin degeneracy

=
q p 2=
( 2 MkT ) 3/2 and qH 4
3

N p Ne
NH

e 0
=
V
3

( 2 MkT ) 3/2 e 0 / kT

Saha Ionization Formula (1921)

where we have used M + m M

You must have noted that we have considered that the H-atom is its ground state. What is the
probability that the H-atom is in its 1st excited state even at a temperature of 104 kT ?

1
4

1st excited state energy 0

N1
=
N0
We know, kTroom

1
0 / kT
e4

=
e 0 / kT
1
eV
40

3
0 / kT
e 4

400
N
1 e
N0

When T = 104

Thus,

Troom
T

N1
= e 12 = 6 106
N0

N1st
small even if T ~ 104 K
N0

Molecules with Internal Motion

1) Intermolecular interactions are negligible

2) Classical gas n 3 << 1

Q1N
N!

QN (V , T ) =

(1)
(1)
Qint
Q1 (V=
, T ) Qtrans
(1)
(1)
(1)
(1)
(1)
= Qelec
Qint
Qnuc
Qvib
Qrot

For example diatomic molecules:

6 degrees of freedom for the 2 nuclei
3 translations for center of mass
1 vibration along the axis
2 rotations of the axis in space ( , )

Diatomic Molecules
kT << Edissociation
104 105 K
T << 104 105 K
Practically no molecule in the excited state as well (exception oxygen at T 2000 6000 K )

Ediss 55, 000 K but Eexcited 11000 K

Vibrational States

103 K
k
Almost full contribution for T > 104
No contribution for T < 102

Temperature is not high enough to excite vibrational states of high energyharmonic oscillation.
Partition Function:

Qvib =

n =0

1
n +
2
e

e v /T

v
=
; v
( Cv )vib Nk =

/
T
2
k
T e v 1
2

T v
T v

( Cv )vib Nk
( Cv )vib 0

Vibrational degrees of freedom are frozen

Very high temperature anharmonic effects

Rotational States
1) Heteronuclear molecules AB
Rotational states & states of the nuclei do not interact
Rotational states of a quantum mechanical linear rigid rotator
2
=
rot ( + 1) / 2 I

2
Mr
=
M
0

equilibrium distance

m1m2
m1 + m2

reduced mass

m.i. about a line perp to the axis and through the cm.

(1)
Q=
rot

=
=
r

=0

2
+

2
1
exp
1
(
) ( )

2 IkT

( 2 + 1) exp ( + 1) Tr

=0
2

10 100 K
2 Ik

(1)

T r Continuum classical approximation (check directly by classical calculation HW)

(1)
Qrot

( +1)
( 2 + 1) e

r /T

Substitute x =
( + 1)

= T / r

( Cv )rot = Nk

Then for T r

Better approximation of the sum to integral: Euler-Maclaurin formula

f (=
n)

n =0

f ( x)dx +

1
1
1
f (0)
f (0) +
f (0) + ...
2
12
720

Here f (=
x) (2 x + 1) e x ( x +1) r /T
2

T 1 1
4 r
+ +
+

+
3 15 T 315 T
3
2

1
16 r
=
Nk 1 + +

+
45 T 945 T

(1)
Qrot
=

( Cv )rot

For T large (but not )

When

( Cv )rot > Nk

and thus shows a maximum

T r Lowest approximation From Eq. (1) gives

(1)
Qrot
=
1 + 3 e 2rT + 5 e 6rT

( Cv )rot 12 Nk e2r /T (goes to zero when T r )

T

Putting all of these together

1)

Very low temperature: rotational and vibrational degrees of freedom are frozen

Cv
2)

r T v
Cv =

3)

3
3
(as U
Nk
NkT for a perfect gas)
=
2
2

5
Nk
2

(extra rotational contribution of Nk )

T v
Cv =

7
Nk
2

vibrational contribution

Log-scale

Quantum Mechanical Considerations

System of many non-interacting identical particles:

Single particle in a potential field V ( r )

Schrdinger equation

2 2

+ V ( r ) ( r ) =E ( r )

2m

Two identical particles

2 2
2 2

+
V
(
r
)

(
r
,
r
)
+
2 + V (r2 ) (r1, r2) =E (r1, r2)

1
1
1 2

2m

2m

N
2 2

H=

+
V
(
r

2m j
N
j )
j =1

To compute the N-particle energy eigenstates, it is only necessary to solve the single particle Schrdinger
equation

2 2

+ V ( r ) un ( r ) =En un ( r )


2m

single-particle eigen state #

Complete set of eigenfunctions for the N particle problem can be taken by arbitrary products of singleparticle states
Example

1) (r1, r2 , r3 ) = u1 (r1 ) u2 (r2 ) u3 (r3 )

eigenstate 1 1st particle

Eigenvalue E = E1 + E2 + E3

H N ( r1, r2, r3 ) = E1u1u2u3 + u1E2u2u3 + u1u2 E3u3 Eu1u2u3

2) ( r1, r2, r3 ) = u1 ( r1 ) u1 ( r2 ) u4 ( r3 )

=
E 2 E1 + E4
In these examples, permutation of particles among the single particle states gives the same energy
eigenvalue.
In example (1) N! permutations.

Symmetry Properties
Quantum Field Theory:
Spin integer particles (Bosons) --- symmetric for 2 particle interchange
Spin half-integer particles (Fermions) --- antisymmetric
Bosons



Symmetry ( r1, r2, r3 ) = ( r2 , r1, r3 ) = ( r1, r3 , r2 ) etc.
Symmetrize Permutations

=
( r1, r2, r3 )

u ( r ) u ( r ) u ( r ) + u ( r ) u ( r ) u ( r ) + u ( r ) u ( r ) u ( r )
1 1 4 2 1 3
4 1 1 2 1 3
1 1 1 2 4 3

distinct permutations

3 of them

1
3

normalization

Eigenvalue E 1 + E1 + E4 = 2 E1 + E4

Same as before

Generally, if we have N1 bosons in state 1, N 2 bosons in state 2, ..

K

Ni = N

K States

r1, r2 , r3 rN =

Ni !
i

N!

u1 (r1 )u1 (r2 )u1 (r3 ) u2 (rN1 +1 )u2 (rN1 + 2 ) u K

distinct
permutations

N1 particles

N 2 particles

This wave function is called permanent.

=
E N1E 1 + N 2 E2 N K E K
Fermions
Antisymmetric

( r2 , r1, r3 ) =
( r1, r3 , r2 ) etc.
( r1, r2 , r3 ) =
Note that u1 ( r1 ) u1 ( r2 ) u4 ( r3 ) not allowed
Interchange particles 1 and 2

u1 ( r1 ) u1 ( r2 ) u4 ( r3 ) = u1 ( r1 ) u1 ( r2 ) u4 ( r3 ) Clearly not allowed

Pauli Exclusion Principle: no two Fermions can have the same set of quantum numbers

2 Particles

=
( r1, r2 ) u1 ( r1 ) u2 ( r2 ) u2 ( r1 ) u1 ( r2 )
Eigenvalues

H=
H1 + H 2
H1ui (r 1) =
Ei ui (r1 ) etc.

( H1 + H 2 )( u1u2 u2u1 ) =

E1u1u2 E2u2u1 + E2u1u2 E1u2u1

= ( E1 + E2 )( u1u2 ) ( E1 + E2 ) u2u1
=
( E1 + E2 )( u1u2 u2u1 )
= E1 + E2
E
N particles
Slater determinants

u1 (r1 )

1 u2 (r1 )

( r1, r2 , r3 ) =
N ! u3 (r1 )

u1 (r2 ) u1 (r3 )

u2 (r2 ) u2 (r3 )
u3 (r2 ) u3 (r3 )

Property of determinants
1) Interchange r1 and r2 meaning interchange two columnspicks up a negative sign.
2) If u1 = u2 then two rows are the same and the determinant vanishes.
(symmetric for bosons antisymmetric for
So of all possible N ! states, Nature picks only 1
1
fermions). This is the origin of the
in the classical case.
N!
If the density is high, then the question of having same quantum # arisesclassical statistics breaks
down.

Complications for Homonuclear Molecules such as H2

T r
Classical partition function (directly) azimuthal angle runs from (0, ) because of exchange symmetry.
Low temperature: Coupling between nuclear spin and rotational states

Total wave function must be symmetric or antisymmetric for an interchange of two identical particle.
Consider H 2

I=

1
for both nuclei
2

i) Total spin 0 (spins are anti-parallelspin function is antisymmetric; spin degeneracy 1)

ii) Total spin 1 (spins are parallelspin function is symmetric; spin degeneracy is 3)
In general if the nuclear spins are sn --- total # of states ( 2 sn + 1)( 2 sn + 1) =
s2
Symmetric

s (s + 1)
s (s - 1)
=
=
sn (2 sn + 1)
( sn + 1)( 2sn + 1) antisymmetric
2
2

Figuring it out:
Symmetric fi (1) f j (2) + fi (2) f j (1)

Spin wave functions part of the total wave function with

i = 1, 2,3,.....s
# of ways doing it

s (s -1)
2

but f1 (1) f1 (2) like states are also allowed

So total number of symmetric states are

s (s -1)
s (s + 1)
+s =
2
2
and total number of antisymmetric states

H2
Symmetric
Antisymmetric

I=

1
2

s (s -1)
2

s = 2 I + 1= 2
3
1

Ortho
Para

D2

s = 2 I + 1= 3

I=1

Symmmetric
Antisymemetric

6
3

Ortho
Para

Ortho larger statistical weight

spin wave function symmetric

Back to Considering H2
Nuclei Fermion
Total wave function must be antisymmetric
Symmetric spin = antisymmetric rotational part Ym

odd
(1)
=
Qrot
odd

+1 /T
(2 + 1)e ( ) r

odd

Antisymmetric spin symmetric rotational part

even
(1)
=
Qrot
even

+1 /T
(2 + 1) e ( ) r

even

(1)
(1)
(1)
=
Qrot
3 Qrot
+ Qrot
odd

even

(1)
3Qrot

Ratio of Ortho to Para

northo
odd
= (1)
npara Q
rot
even

T r
(1)
(1)
Qrot
Qrot

odd

Then

even

1 (1)
Qrot
2 classical

northo
T
3 :1
npara

T r
(1)
Qrot
1 + 5 e6r /T +
even

(1)
Qrot
3 e 2r /T
odd

northo 3 3 e2 r /T

9 e2r /T 0

T
6
/
r
npara 1 + 5 e

when

T 0

All Para

When T r , one needs to compute the full partition function and compute specific heat. Specific heat
computed this way dose not compare well with experiments. But the conversion from Ortho to Para timescale is very large. So the gas behaves like a mixture of two gases of pre-assigned ratio. If that is the case
then Cv
=

3
1
( Cv )ortho + ( Cv )para
4
4

and this compares well with experiments. (Dennison 1924)

-------------------------------------------------------------------------------------------------

I =1

D2
Bosons

So Ortho symmetric spin even rotational

Para antisymmetric spin odd rotational

northo
=
n para

6 Q rot(1)
even
(1)
3 Q rot
odd

T 0 all Ortho

T
2 :1

Ideal Quantum Gas

Review of Classical Partition Function (Distinguishable Particles)
2-State Systems

QN =

{ i }

e H i

e H i

{ i }

= e H i
i
i

=
e H + e H
i

= Q1N
Now consider a different way of computing this in terms of single particle states

H 2 =
1 =
H
n1

n2

=
E n11 + n2 2
N= n1 + n2
QN = e Es
{s}
N ! ( n11 + n2 2 )
= '
e
n ,n n1 ! n2 !
1

n=
2 ( N n1 )
QN =
n1

n1

= y

n1

= Q1N

N!
e 1
n1 !( N n1 )!

) ( e )
n1

N n1

N!
1
n1
x n1 y N =
where x e=
and y e 2
n1 !( N n1 )!
n

x 1
N!
N
= y
n1 !( N n1 )! y

1 +

x


N
= ( x + y) = e 1 + e 2
y

How does one handle indistinguishability in classical Statistical Mechanics?

Well that came as an ad hoc solution by Gibbs:

Q1N
N!
N!
but the
degeneracy factor was still used in the partition function !
ni !
QN =

In general

degeneracy factor

N ! ni i
QN = '
e
non-interacting
{ni } ni !
i

) (

n1 n2
N!
e 1
e 2
!

n1 ,n2 1 2

' n !n

can be shown explicitly

(see, for example, 3-state systems)

= e 1 +e 2 +e 3 +

= e i

= Q1N
3
terms

n1 ,n2 ,n3

n1 ,n2

= zN

N!
x n 1 y n2 z n3
n1 !n2 !n3 !

N!
x n 1 y n2 z N n1 n2
n1 !n2 !{ N (n1 + n2 )!}

n1 ,n2

= zN
n1

N!
x 1 y

n1 !n2 ( N ( n1 + n2 ) )! z z
1 x

n1 ! z

n1

n2

n2

N!

n2 !{ N ( n1 + n2 )}!

n2

=z

n1

n1

1 x

n1 ! z

n1

( N n1 )! y n2
N!
( N n1 )! n2 !( N n1 n2 )! z

1 x

n1 ! z

n1

N ! y
1+
( N n1 )! z

= z 1 +
z

N n1

x z
N!

( N n1 )!n1 ! z y + z

x
=
( z + y ) 1 +

y+z

n1

( x + y + z )N

Quantum Systems (indistinguishable)

Canonical Partition Function

QN =
E

' e n

i i

{ni }

ni i
=
ni

No of particles ni occupying eigenstates i

Indistinguishable: System degeneracy factor is unity. This makes it very hard to compute the Canonical
Partition Function

Grand Partition Function

ni

e N QN
=

Q
=

=
N 0=
N 0

i ni

=
' ze
N =0 {ni } i

QN

where z =
e

Note the double sum with the constraint: the first sum is over ni under a fixed value of N and then over
all possible values of N. When N , this double sum is the same as summing over n i independent of
one another i.e. without the constraint.

To show that in Grand Partition Function

{n1,n2 ,}
Consider only 2 states n1, n2 where n1 + n2 =
N
N =0

'

n1

n2

Suppose N=6
With constraint n1 + n2 =
6 : these points are on the top line
with

N=5 2nd line

N=4 3rd line etc.

But you would have gotten all these (not really!) points by allowing

n1 = 0, 6
independently
n 2 = 0, 6

Actually, this would work only in the line N .

Then,

Q=

n1

n2

ze 1

) ( ze )
n1

n2

=
e ( i )ni =
Qi So the Grand Partition Function Factorizes!
i n
i

Fermi Dirac Statistics

0,1
ni =

QFD =
1+ e ( i )
i

Bose - Einstein statistics

ni = 0,1, 2

1
QBE =
; < i for convergence
i 1 e ( i )

Lectures 15-17
Grand Potential

FD =
kT n QFD =
kT n 1 + e ( i )

Fermi Dirac

(
)
)

=
BE kT n 1 e ( i )

Bose-Einstein
Together, = kT

n 1 e ( i

(upper sign BE)

Also, = E TS N
but G =
F G
F + PV
=
= PV

kT n 1 z e i

=
PV

nQ =

E=
U=

z e i + a
i

e ( i )
N=

=
kT
( i )

i 1 e
1
=

e ( i ) 1
i

N =
i

1
z 1e

+a

a = 1 BE
a = +1 FD

How about ni ?
Since the Grand partition function factorizes and N =

Occupation numbers
=
ni
FD
BE

1
1
=
1 i

z e + a e( i )/ kT + a

ni 1

< i (all i ) otherwise ni < 0

When = 0 , n0 diverges

ni

BE condensation

Classical Statistical mechanics (Maxwell-Boltzmann or MB) is recovered when a=0:

/ kT
n =e ( i )
i

Note than the effect of a is negligible when e( i ) / kT 1

ni 1 (Low density)

Now you can see why Gibbs fix works. If you treat them as distinguishable you get an extra
N!
in the partition function. But this degeneracy factor becomes just N !
degeneracy factor of
ni !
i

as ni ! 1 when ni 1 (recall that 0! = 1 ).

i

Also since e( i ) / kT 1 for all values of i , it follows that

e / kT 1

e / kT 1 z 1

Consistency Check: Ideal Gas

Or,

N3
N3

n
n
=
kT
or,

kT
V
V

N3
1
V

n 3 1

=
n

N
number density
V

Another way of looking at it:

mean interparticle distance ( 3d )

1/3

V

N

n 1/3
n 1 3
n 1

n 3 1
nT 3/2 1
n small
conditions for recovering Classical or Maxwell-Boltzmann Statistics.
T large

Ideal Bose Gas

=
kT

PV

n 1 e ( i )

i
i

PV
+
=
n 1 e ( i ) =
n 1 z i ; z =
e
kT
i
i
If ni is not negative z 1 i.e., < 0 .

1
1
=
1 i
( i )
e
1 z e 1
=
N
=
ni
E ni i
=
ni

Also, z 0

Single Particle Density of States

For large V the single particle states i of a system are almost continuous.

f ( i )
i

a ( ) f ( )d

Consider V = Lx L y Lz and L .
Periodic boundary conditions (to eliminate reflections) ( x + Lx , y , z ) =
( x, y , z ) etc.

ik r
Free=
particle e=
k

k x ( x + Lx ) = k x x + 2 nx k x =

2
nx
Lx

# of states for k x to be in between (k x , k x + k x ) , n x =

# of states k , k + dk

3
Lx L y Lz
k x k y k z
a k d=
k
( 2 )3

( )

Lx
k x
2


=
a k d 3k

( )

V
=
d 3k
3
( 2 )

( 2 )

4 k 2dk

element of volume in k-space

a ( ) d =

( 2 )

2m

2m 1 1/2

d
2

dk

2 2

k
2m

non-relativistic

V (2m )3/2 1/2

a ( ) d =

d
4 23

(Recall, classically
( P)
=

( ) =

V 4
( 2m )3/2
3 3
h

0 ( )
a ( ) =
=

1
V 4 3
3
3
d
q
d
p
P
=

h3 p P
h3 3

3 V 4
( 2m )3/2 1/2
3
2 h 3

3/2
V
3/2 1/2 V (2m)
=
2

2
m

1/2
(
)
3
2 3
h
4

Equation of State
Sum over i can be replaced by integral in the Equation of State

PV
=
a ( ) d n 1 ze
kT

=
N

=
ni

1
z 1e + i 1
1

a ( )d
e

( )

" f BE ( )

E=

a ( )d f BE ( )
BE distribution function

2 V ( 2m )

3/2

N=

h3

1/2 d
)
e (
1

This integral can be evaluated by substituting = x

N =V

2
h3

(kT )3/2 (2m)3/2

x1/2 dx

z 1e x 1

General Form g n ( z ) =

x n1dx
1
(n) z 1e x 1
0

g n ( z ) = g n1 ( z ) (We will need it later)

z
2

3
3 1 1 1
N=
V 3 (kT )3/2 (2m)3/2 g3/2 ( z )
Note : = =

2
2 2 2 2
h

Recursion Relation z

( 2 mkT )1/2
V
V=
g3/2 ( z )

g 3/2 ( z )
h
3

1
N
g3/2 ( z )
=
V 3
Note that z 1 , so we can expand g3/2 ( z )

Multiplying both denominator and numerator by ze x ,

ze x
1
n 1
gn ( z) =
x dx
( n)
1 ze x
0

1
=
x n1dx ze x

( n)
0

) =0 ( ze x )

1
1
x
n 1
=
x
dx
ze

n
n
(
)
(
)
= 1 = 1
0

(Substituting lx = y )

1
y n1e y dy
z
z
=
n1
n
( n) 1 =
1
=
0

gn ( z)

n1 y
since (n) =
y e dy
0

Finally then, g n ( z ) =z +

z2
2n

z3
3n

x n1e x dx

Maximum value of g n ( z ) is when z=1

Converges for n > 1 =

(2)

g n (1) =1 +

;=
(4) 4 / 90;

1
2n

1
3n

+ = (n)

g=
(3 / 2) 2.612
3/2 (1) =

How does this convergence affect the physics?

N
1
3 2.612
V

( 2 mkT )3/2 2.612

h3

experimental
quantity

(When you vary T, changes and so does z. z becomes = 1 and then stays at 1).
So we can choose small enough T such that the inequality is broken!! What is going on?
Bose-Einstein Condensation
Note that

is valid only if no state is macroscopically occupied. So when this inequality is

broken, some state is macroscopically occupied.

Lowest Temperature T = Tc when the equality is still satisfied (with z = 1 ; actually z = 1 for
all T < Tc )

h 3n
Tc
=

2.612

2/3

1
2 mk

h2 n
Tc =
2 mk 2.612

2/3

Which state is macroscopically occupied? Bosons can occupy any state macroscopically. Makes sense
for = 0 state though.

z
1
=
as z 1
z 1 1 z
z e
1
Also note that a ( ) ~ 1/2 which means a ( ) = 0 for = 0 and we have not taken the = 0 state
Check: ni =

When=
i 0, =
N0

1 i

into proper account. So what? What is one state worth in a continuum? Problem here is that this state
is macroscopically occupied.

=
So we better write N

a ( ) d f BE ( ) + N 0

N ex

N N0
1
=
g3/2 ( z )
V
3

When T = Tc , N 0 = 0 , and when T < Tc , N 0 > 0

At Tc ,

N
~ Tc3/2 g3/2 (1)
V

(1)

N N0
~ T 3/2 g3/2 (1)
V

Below Tc ,

(2)

N N0 T
Divide Equation (2) by (1) to get
=

N
Tc
T 3/2
N 0 = 1 N
T Tc
Tc

3/2

T Tc

Revisit Equation of State and Other Thermodynamic Quantities

PV
2
=
3 (2m)3/2V 1/2n 1 ze d n(1 z )
0
kT
h
( =
0 term)

P
2
1
=
3 (2m)3/2 1/2n 1 ze d n(1 z )
0
kT
V
h

Even when N 0 is a large fraction of N

N0
z
=
z
N=

0
N0 + 1
1 z

The last term

N0
N 0 n N 0
1
1
1
1
n 1
n N 0 =
n
=
0 for macroscopic system
=
V N0 + 1
V
N0 + 1 V
V N0
k = 0 Condensation (momentum space) does not contribute to pressure.
Now substitute x =

2 ( 2mkT )
1/2
P
=

x n 1 ze x dx

3
0
kT
h
Integrating by parts,
3/2

ze x 2 3/2
2 3/2

x
x dx
0 1 ze x 3
3
0
The first term vanishes, and the second term can be written as

n 1 ze x

2 x3/2 dx
2
= g5/2 ( z ) (5 / 2)

x
1
3 0 z e 1
3
3 1
3
3 1
2
as (5 / 2)= (3 / 2)= (1/ 2) and (1/ 2)=
= g5/2 ( z )
2 2
2
2 2
3

P
1
=
g5/2 ( z )
kT 3

High temp limit:

g n ( z ) z as z 1; and

N3

V
P
Below the critical =
point,
kT

Classically = kT n

P
1 / kT
e
=
kT 3

P
1
3 N

N
= =
kT 3 V
V

1
1.342
=
g (1)
3 5/2
3

Isotherms (or P-V Characteristics)

Recall the definition of critical temperature:

N=

g3/2 (1)

Instead of setting T = Tc , we keep T fixed and set V = Vc

V
Then N = c g3/2 (1)
3

(1)

When there is no condensation,

P
1
g5/2 ( z );
=
kT 3

P N g5/2 ( z )
=
kT V g3/2 ( z )

V
, Then
N
g ( z)
Pv =
kT 5/2
g3/2 ( z )

Replace

N
1
= 3 g 3/2 ( z )
V
1

by

N
1
V g3/2 ( z )

Write v =

No condensation v>vc

When condensate starts forming v=vc

Pvc = kT

Note that vc is a function of temp.

g5/2 (1)
g3/2 (1)

Substitute T from Eq.(1)

P v5/3
=
c

g5/2 (1)
h2
c0 , some constant
=
2 m { g (1)}5/3
3/2

(2)

N N0
1
= 3 g3/2 (1)
V

P
1
N N 0 g5/2 (1)
and
=
=
g
(1)
5/2
V
kT 3

g 3/2 (1)
1
g (1)
3 3/2
NN0 N
N N0
1

Divide Eq.(3) by=

(1)
=

=
1
VN
V
V
Vc
c
V
g (1)
v<vc ,

When condensate forms

3/2

P
N g5/2 (1)
=
kT Vc g3/2 (1)
kT g5/2 (1)
, independent of v
P=
vc g3/2 (1)

From ( 4 ) and ( 5 ) ,

BE Condensation Revisited in a General Way

For BE condensation, the most important thing to understand is the behavior of the integral

a ( )d

1
1

z e

when z 1

(This integral is actually N .)

If I ( z = 1) diverges, everything is fineno condensation.

I ( z= 1)=

1
a ( )d
e 1

Consider a ( ) ~ m

Now if ~ k n , d ~ k n1dk
d
a (k )dk ~ k d 1dk ~ k d 1 n1 ~ k d n d
k

(3)
(4)

(5)

a ( )d

d n
~ n d i.e.,

m=

d n
n

When is large, e dominates hence there is no divergence.

Consider small and c is some cut off where this approximation cease to work.
c

1
1) C m d
I (z =
c m1d ~ m
0
0
1 + 1 0

I ( z = 1) diverges if m 0 ( m = 0 logarithmic divergence.) No condensation.

For condensation
i.e.,
or

m0
d-n
0
n
d>n

For non-relativistic particles, ~ p 2 ~ k 2 i.e., n = 2 . That means no BE condensation in d=2 but BE

condensation takes place at d=3.
For relativistic particles, = pc ~ k1 i.e. n = 1. That means BE condensation can take place in both d=2
and d=3.

Ideal Fermi Gas

PV
n Q= n 1 + ze e i
kT
ei
1
N
ni
z e / kT and 0 z <
=
=
where
=
1 e i
+1
ei z e
i
N =

f FD (e )

a (e ) de
e )
e (
+1

V (2m)3/2 1/2 V (2m)3/2

=
e
2p e 1/2
2 3
3
h
4p

=
a (e )

V (2m)3/2
a (e ) =
gs
Ce 1/2 ; g s spin degeneracy = 2 s + 1
2p e 1/2 =
3
h
N
V

e 1/2 d e
(2m)3/2 2
g=
substitute x be
s
e b (e m ) + 1
h3
0

= gs

(2m)3/2 2
h3

2 m kT
= gs

h2

-3/2

x1/2 dx

z -1e x + 1

0
3/2

x1/2 dx
z -1e x + 1

x n-1dx
1
1
=
g s 3 f3/2 ( z )
where f n ( z ) =
( n ) z -1e x + 1

0
e
a (e ) d e n (1 + ze ) d e

PV
=
Equation
of state
kT

e
1/2
=
C e d e n 1 + ze d e

2
= C e +3/2n 1 + ze e
3

) 0 C

z ( )
1 + ze

e e

2
e de
x e

= C=
1 e
3
1
z
e
+
0
3/2

2
1
=C 3/2
3

x3/2 dx
1 + z 1e x

g sV

f5/2 ( z )

2 3/2
e
3

Internal Energy U =

nQ

e a (e )

z ,V
gV
P 3
3
3
kT s3 f5/2 (=
z)
kT V =
PV
=
kT 2
2
2

de
1 e

z e

Density low, temperature very high (non - degenerate)

+1

nh 3
g s ( 2 mkT )

3/2

<< 1

Nl3
=
<< 0
Classical
limit m kT n
V

Nl3
z << 1
;
V
Similar to the expansion of g n ( z ) for the Bose gas, here also one can show that

m / kT
=
z e=

f n ( z ) =z

z2
2n

z3
3n

z , for small z

Then

3
V
U = kT g 3 f5/2 ( z )
2

f ( z) 3
3
NkT 5/2
NkT ;
= =
2
f3/2 ( z ) 2

z<<1

2
NkT
PV= U =
3
At very low temperature (degenerate)

T 0
=
n

n 3
gs

1 for e < m0
1

e m )/ kT
e(
+ 1 0 for e > m0

T 0
i) < 0
ii) e > 0

mmm
=
0 where
0 is the Chemical Potential at T 0
0 =
e 0 =

Step Function at T=0

n=
e

/ kT

+1

1
n = / kT
e / kT 0
e
+1

1 as T 0

Definition of Fermi Energy

eF

a (e )d e = N

N=

f FD (e )a (e )d e

1 for e e F and 0 otherwise

a (e )d e

( e m )+1

eF

a (e )d e

2
C e F3/2 = N
3
1
3/2 3
e F=
N
2
C
2/3

2 3
3
6 2n
4

=
e F N=

2 gV (2m)3/2
g
s

2/3

2
2m

Ground state energy (average) (T=0)

E0

eF

e a(e )d e ( since n 1)
=
0

= C

eF

e 3/2 d e

2
3
2 3/2 3
=
C e F5/2 =
N
C eF eF =
5
5
3
5
E0 / N =

3
eF
5

P=

2N
eF 0
5V

Approximate Calculation of the Specific Heat of a Fermi gas

As the temperature is raised form 0 to T each free electron does not gain energy by an amount kT because
most of them are occupying states below Fermi level and by the Pauli principle they cannot be excited to
the states which are already fully occupied. Only a small fraction of electrons with energy close to the
Fermi level that can be excited to empty states lying in the range kT above 0 .
# of excited electrons
N ex (Density of states near the Fermi level) width of energy

a ( F ) kT
gs
=
=

F
1

V (2m)3/2 2
h

1/2
F kT

g s V (2m)3/2
g s V (2m)3/2

2
h

F3/2 kT

2
3 N 4 2 3
kT
h3 2 V g sV (2m)3/2

3 N
3 T
kT
N
=
2 F
2 TF

T 300 K

TF 10,000 K ,

N ex
~ 4 5%
N

T2
3
Then U (T ) ~ Nk
( N ex kT )
TF
2
T
close to the correct result
CV ~ 3Nk
TF
Electron Gas in Metals
3
C=
V T + T

electron
phonon contribution
contribution

CV
= + T 2
T

Magnetic Behavior of an Ideal Fermi Gas

Pauli Paramagnetism;
Alkali metals Pauli 1927
T=0

* magnetic moment

magnetic field

* B
* B

(1)
(2)

=
e KE * B ( Parallel )
e=
KE + * B ( Anti parallel )
|
total energy

N + = # of electrons in group 1; KE= + * B

N = # of electrons in group 2; KE= * B
Spin degeneracy is lifted by magnetic field

=
a (e )d e

2V
=
(2m )3/2 e 1/2 C0e 1/2
h3
|

no g s here
Magnetic moment
=
M * (N+ N )

=
N+

=
N

At T=0

=
N+

a ( + * B ) f FD (
)d
a ( * B ) f FD (
)d

C0 ( + * B )1/2 1 d

2
=
C0 ( F + * B )3/2
3

B is small so * B the lower limit 0 after a change of variable

Similarly

=
N

2
C0 ( F * B )3/2
3

M m *( N + N )
=
3/2
3/2
4mmm
*B
* B
V *(2m) 3/2 3/2
3/2

F 1

F 1 +

3h3
F
F

4m
V *(2m F ) 3/2 3 mm
*B
3 *B
1
1
+

If B is small

2
2

3h3
F
F

M 4 V
lim
m
*2 (2m)3/2 1/2
=
=
F
3
B 0 B
h
= 1 m *2 3 N
F
In contrast to classical Langein paramagnetism where ~
At low temp 0 constant

1
(Curie's law)
T

Pauli Paramagnetism

Physical Picture:
Langevin result: (Probability that an atom parallel to the magnetic field Probability that it is
antiparallel).

e m H e m H

1 + m H 1 + m H
1 + m H + 1 m H
+e
e
= m H (if it is small)
m H

m H

For N atoms this gives a net magnetic moment

N H ~ N 2 H / kT ~ N 2 / kT
But now at a temp T << TF only N ex of the electrons contribute. This means

M ~ N ex H

3 T 2H 3
N

~ N 2 H / kTF
2 TF kT
2

independent of temperature

Fermi Gas : Degenerate Case (Detailed Calculation)

kTF = e F

0 < T << TF

2 V

3/2
0

e 1/2 d e
e m )/ kT
e(
+1

3/2
0

e 3/2 d e
e m )/ kT
e(
+1

g s 3 ( 2m )
h ((
((

mm
(T )

2 V

U ( T ) = g s 3 ( 2m )
h ((
((
C

Integrals of the form I =

in the region

f (
) f FD ( )d where f FD ( ) is a slowly varying function of , except

( ) , define
In order to take advantage of this property of f FD
=
F (
=
( )
)
( )d
and note that F (0) = 0
Then I =

= f

) f FD ( )d
F (

Integrate by parts

( ) F ( )d
) f FD
FD ( ) F (
0
0

bcaus

F (0) = 0
f FD () =0

(
= f FD
) F ( )d
0

dF
d

Expand F ( ) about

1
F (=
) F ( ) + ( ) F ( ) + ( )2 F ( ) +
2

I = I 0 F ( ) + I1F ( ) + I 2 F ( ) +

(
( )d
)d f FD
I 0 = f FD

(
(
)d ( ) f FD
)d
( ) f FD

I1 =

1
I2 =
2

( )

1
2
(
(
)d ( ) f FD
)d
f FD
2

1
Now f FD (e ) =
e ) / kT
e(
+1
Write ( ) / kT =
y

(e ) =
f FD

(e y + 1)

I0 = +

( e y + 1)

e y dy

( e y + 1)

( )e y ; d e = dy

ey

dy

e y + 1, dx =
e y dy
; x=

y = x = 1
dx
=
1

x2
y=
x=

1
y e y dy
dy
I1 =
y
+
=

2
y

e +1
e /2 + e y /2

1 1
y 2e y d y 2
I2 =
+
=
( kT )2

2
2
2
6
y
e + 1
(
(((

" 2 /3

0
odd function
=

Then I =

) f FD ( )d
f (
0

F () +

2
6

( kT )2 F ( )

Where
=
F (
)
( ) d

F =
( )
( )
F =

f (
) f FD ( )d =
)d +
f (

2
6

( kT )2

df
d =

1/2
N= C
f FD ( ) d

f (
)=

0
1/2

Then

1
2
( kT )2 m 1/2
N C e 1/2d e +
=
6
2
0

m
m0

2 (kT )2
1/2
C e d e + e 1/2d e +
12 m01/2

m0
0

small error by

setting 0 here

Note that at T=0

N = C

1/2

2

2 ( kT )
1/2
d +
N = N + C
12 01/2
0
2
1/2
2 ( kT )

d =
0
12 01/2

an valu thor
2
2 ( kT )
01/2 ( 0 ) =

12 01/2

 2 ( kT )
0 =

12 0

(1)

2 kT 2

= 0 1

(degenerate Feri Gas)

12 0

(2)

Internal energy

U = C e 3/2 f FD (e )d e
0
3/2
=
f (e ) e=
f (e )

3 1/2
e
2

2
23
=
U C e 3/2 d e +
kT ) m1/2
(
6
2
0
1 2
mm

2
0
e 3/2 d e + e 3/2 d e + m
kT ) 01/2
(

m0
=C 0
4

small error

2
0)+
U 0 + C m03/ 2 ( mmm
( kT )2 01/2
4

2
2
3/2

2
(kT ) 2
2 1/2
2 1/2
U=
U 0 C mmm

+
=
kT
C
kT
(
)

(
)

0
0
0

12 m0 4
0

2
3
2 3 ( kT )2
2 ( kT )2
2
=N

U U 0 = C 03/2

3
6 2 2 0
4 0

Now N= C

1/2
d = C 03/2

1/2
Then U 0 C=
=
d

0

3
N 0
5

2 ( kT )
3
=
U
N 0 + N
5
4 0

2
5
2
2 kT
( )
kT
3

=
N 0 1 +
C
Nk

V
5
12 0
2 0

Equilibrium between Phases

Two-Phase Co-Existence Curves (single-component)
Sample Phase Diagram

Equilibrium between phases

=
P1 P=
T1 T2
2

1 = 2

For example, all along the condensation curve (i.e., liquid-gas coexistence)

L ( P, T ) = G ( P, T )

1(T , P ) = 2 (T , P )
1(T + T , P +
=
P ) 2 (T + T , P + P )
d 1 =
d 2 Tangential derivative along the curve

For point A
B

Now G (T , P ) = N (T , P )

G =U TS + PV
dG = d U T d S SdT + P d V + VdP =
SdT + VdP
In terms of per particle or molar variables

s dT + v dP =
s g dT + v g dP

v g v dP = s g s dT

dP s
=
dT v

s (at fixed temp) =

L(T )
where L(T) is the Latent Heat
T

dP L(T )
Clausius-Clapeyron Eqn.
=
dT v T

Slope of the coexistence curve:

Solid to liquid transition:
Most case v > 0 but water is an important exception for which v < 0
As the entropy of liquid > entropy of solid, this means the slope is negative for water as shown below:

Liquid Gas Co-Existence Ending at a Critical Point C

For T > TC or P > PC no difference of phases exits, the volume change V between liquid and gas
has approached zero and there exists only one fluid phasethe very dense gas has become
indistinguishable from the liquid.
When there is a critical end point, the concept of different phases itself is arbitraryone can go from
pure liquid to pure gas without any phase transitionsLatent Heat goes to zero as T TC .
No such critical end point exists for solid-liquid transition. Solids and liquids are quantitatively
differentsince they have different internal symmetry. We can only say whether this particular
symmetry exists or notit can appear or disappear only as a whole not gradually.
- Complicated Situation Liquid Crystals

Liquid Gas Transition

Isotherms

For T < TC when the gas is compressed, the gas reaches a specific volume v g where it condenses at a
constant pressure to a liquid specific volume v . For T > TC the gas does not condense, no matter how
much it is compressed.

Order parameter v g v (or density difference g ).

V-T Plane:
Co-existence occupies a whole region of plane and not simply a curve as in the P-T plane. This is
because the volume V, unlike the pressure, is not the same for the two phases.

Total molar volume vo

v o x v + x g v g (due to coexistence)
=

x + xg =
1

x: mole fraction

) (

vo x + xg = x v + xg v g

x v g v o
=
xg vo v

Lever Rule

Van der Waals Eqn. of State for the Liquid Gas Transition
(1873 Ph.D. Thesis)

P = P(v, T )

Equation of State
Ideal gas law failsnon interacting systems do not give phase transition in a classical system (Bose
condensation is the only exception in Quantum Systems).
2

N
NkT
P + a [V Nb ] =
V

or

kT
P + 2 ( v b) =
v

Works away from the critical point, but near critical point fails.
The first term in the equation of state originates from long-range, attractive interactions between gas
molecules (see below) while the second term arises from the excluded volume of the gas molecules as
they are not really point particles as assumed in ideal gas law calculation.

F
P=

Since the extra pressure (compared to the ideal gas case) arising from the 1st term in the
V

equation of state is negative, this would mean the presence of a new attractive free-energy term
(compared to the ideal gas case).

Fextra ~

a N2
a 1
for the extra attractive term N N
V 2
2 V

a
1
. (You need the
term for the
V
V
F
interaction strength to be able to take the thermodynamic limit. Note that
should be finite;
N
F
1 N
which is finite in the thermodynamic limit). Infinite Range Model gives so-called
~ a
N
2 V

Every particle interacts with everyone else with the same strength

mean-field behavior
Rewriting Van der Waals Eqn. of State

kT

v3 b +
P

For T < Tc

2 av ab
0

=
v +
P P

3 solutions (v1, v 2 , v3 ) for fixed P, T

(1)

3 solutions coalesce at v = vc i.e.=

for P P=
C , T TC LHS of Eqn. (1) is ( v vC )

T = Tc

But =
( v - vc ) v3 - 3vc v2 + 3vc2 v - v3c
3

(2)

Comparing (1) and (2)

From (iii) and (ii) dividing

b+

kTc
=
3n c
Pc

a
=
Pc
ab
=
Pc

3vc2
vc3

nc
(i )
=b

3
Then

ab
Pc =
(ii ) =

27b3

and

(iii )
=
Tc

vc =3b
a
27b2
8a
27bk

=
Define reduced variables
=
P P=
/ Pc ,
/ c , T T / Tc
Universal from

1) 1 3 2
P= 8T ( 3

(3)

Law of Corresponding States

Figure Caption: The liquid gas coexistence curves of many fluids can be virtually superimposed when
the temperature and density are scaled by their corresponding critical values Tc and c . Plots of this
kind are manifestations of the law of corresponding states. This plot from Guggenheim (1945) played
an important role in the development of the theory of critical phenomena by showing that the data cannot
be fitted with a quadratic curves van der Waals theory implies, but require the cubic shown here, which
corresponds to the fact that the order-parameter exponent 1 / 3 and not 1/2 as van der Waals theory
implies.
Van der Waals eqn. of state is unphysical

For T < Tc

P
0 is violated
n

stability condition

Maxwells construction
Van der Waals wiggle is replaced by the straight line part such that the area of two regions A and B are
the same.

g
P (vg v ) =
Pdv

vl

Critical Exponents
Start with Eq. (3), get rid of the tildes so that P=
c v=
c T=
c 1 in this set of units.

=
P

8T
3
2
3v 1 v

Lets set v = 1 + v to expand near the critical volume which is 1 in this set of units.

P
=

8T
3

(3 + 3 v 1) (1 + v) 2

8T
3
3
=

= 4T 1 + v
2
3
2
2 1 + v (1 + v )
2

3 (1 + v )

9
27
2 3 2 2 3 4 3
3

= 4T 1 v + v 2 v3 3 1 2 v +
v
v
4
8
2
3 2
2

27

= T 4 6 v + 9 v 2 v3 3 1 6 v + 9 v 2 12 v3
2

27

= ( 4T 3) + (T 1) 6 v + 9 v 2 T 12 v3
2

Now T 1

v 0 ; Lets write T =
1 ( 0)

3
2

Then P (1 + 4 ) + 6 v + 9 v 2 v 3

(7a)

If we set v = 1 v

3
P (1 + 4 ) + +6 v + 9 v 2 + v3
2
Subtracting 0 =
12 v + 3 v3
2
4 (1 T )
4=
v 2 v=

v =
2(1 T )1/2 ( symmetricshape of the co existence curve )
v g ~ 1 + 2(1 T )1/2
v ~ 1 2(1 T )1/2

(7b)

Order parameter

v g v ~ 4(1 T )1/2 ;

T 1

Order parameter behaves like OP t hence the order parameter exponent i.e -exponent for the vdW
gas is

1
(parabolic)
2

v g v 4 (1 T )1/2
1
1
=
g ~ =
~ (1 T )1/2
v v g
v vg
1 4 (1 T )
Experimentally one finds, g ~ T Tc

0.327

so vdW theory does not work near the critical point.

Simple Statistical mechanical Calculation to obtain van der Waals Equation of State
Ornstein (1908)
Interaction between two particles separated by a distance r

=
( r ) q( r )

V
r r0
With q( r ) =
0 r > r0

>0
Hard-core repulsion + infinite-range attractive interaction

Can be solved exactly only in 1-d for particles can be ordered

pi2
+ xi x j
2m 1i< j N

H =

QN ( L, T )

h N

pi2
2 m dp dp
1
N

( xi x j )

dx1 dxN

The N! term does not occur in QN ( L, T ) as particles can be ordered in1d, they are distinguishable.
Then

QN ( L, T ) =
=

N ( N 1)

+ q xi x j
L
e

dx1 dxn

q( x x )
e
dx1 dxn
(((((
(((((((

Qhc

Gas of hard rods of length 0 (Tonks 1936)

Positional marking on the LH corner of each rod. Maximum value of xN is L.

0
__

N___
-th

x1
x2

Qhc

dx1

dx2

dxN e

q xi x j

dxN dxN 1 dx1 e

If they do not overlap qij = 0 and so the integrand is one.

If they do overlap qij = and the integrand is zero.
We will construct conformations such that they do not overlap.
x2 0

Limit on x1

Limit on x2

x3 0

dx1
dx2

particle 1 is already occupying 0 to 0 space.

x4 0

Limit on x3

dx3

..
L

Limit on x N

dx N

( N 1) 0
x 0

2
Qhc =

x3 0

dx1

x4 0

dx2

dx3

( N 1) 0

dxN

Change of variable

u1 = x1
Qhc =

u2 = x2 0

x2 0
0

du1

x3 2 0
0

u3 = x3 2 0 ........
du2

but x2 =
0 u2 ; x3 2=
0 u3..etc.

x4 3 0
0

u N = xN ( N 1) 0

du3

L ( N 1) 0

du N

q xi x j

Qhc =
=

u2
0

u2
0

du1
u3

u3
0

L ( N 1) 0

du2

u4
0

du3

uN

du N 1

L ( N 1) 0

du N

du1 du N

Change the order

L-( N -1) 0

du N

uN

du N -1

u4

du3

u3

du2 2 du1 =
0
0 (

" u2
((((

u32
2
((((((((

"

Then QN ( L, T ) =

N ( N 1)
L

u43
3.2

{L ( N 1) 0 }N
N!

{L - ( N -1) 0 }N
N!

Magnetic Phase Transition

Isotherms

Spontaneous Magnetization M=
M 0 , H 0)
0(M
Order Parameter M 0

Coexistence curve

Magnetic Phase Transition

Curie-Weiss Theory
Mean-Field Theory
Molecular Field Theory

H =
J s is j h s i
<ij >

Ising model

|
unit of energy mB B

If J = 0 spins are independent

Q0 =
=

{s i }

+ h s i

e hs1

e h(s1 +s 2 +s 3 +s N )

1 s 2 =
1
s1 =

e hs 2

1
1
s1 =
s2=
N

hs1
= =
e

s1 =1

[ 2 cosh h]N

kT n Q0 =
NkT n {2 cosh( h)}
F=
1
F
2 Sinh( h) = N tanh( h)

=
+ N kT
M=
h
2 cosh( h)
M
=
m = tanh( h)
N
When J 0 , we will try to write the Hamiltonian in terms of an effective field due to all other spins.
Suppose we can write

H = i hi
i

Then

Q=

e h1s1

e h2s 2

e hN s N

1
1
1
s1 =
s2 =
sN =

= [ 2 cosh( h1 ) ][ 2 cosh( h2 ) ][2 cosh( hN ) ]

If h=
= hN= h then m = tanh( h ) is recovered.
1 h=
2
i hi
Now, H =

hi =h + J j
j

=
h + J
j + J
j
j
j

= applied field + mean-field + fluctuations. Neglect fluctuations and note that m = i

hi h + J 2d m
| co-ordination #

All hi are the same

h + 2dJm
m = tanh

k BT
Set h = 0 and look for spontaneous magnetization

m = tanh(2dJm / k BT )
((
((((

2dJ
2dJ
< 1 , there is only one solution.
Tc =
kT
kB
Now look at the eqn. of state ( h 0) define = Tc / T
h

m tanh
=
+ mt
k BT

For

tanh ( h / k BT ) + tanh ( mt )
1 + tanh ( h / k BT ) tanh ( mt )

tanh

h
m tanh(mt )
=
k BT 1 m tanh(mt )

(Note that is not small!)

x3
Then for small h and m we can expand the tanh-function as tanh x x
.
3
h

k BT

3
m )
(
m m +

m )
(

1 m m
3

3
m(1 ) + m3 1 + m 2
3

3
h
m (1 ) + m3 2 + 3
k BT

Equation of State

}
1)

Set h = 0 to study spontaneous magnetization

t3
0 m (1 ttt
) + m3 2 +
3

T
1t
m2 =

tt
1 (note 0)
t= c
T

t2
tt
1 +
3

Tc T
T T Tc
(1 t )
T
T Tc
m2
3 1 c 3 =
3=
3 c

1
T
Tc
T

T
T
1 1 1 +
3

T T
T
= 1
define t = c
Tc
Tc
1

Then, m 2 3t {1 t} = 3t (1 + t ) 3t + 3t 2 3t
Thus, we see that the order-parameter exponent is =

as t 0

1
--- the same as in a vdW gas!
2

The concept of Universality arises in terms of critical exponents.

Free Energy

h + 2dJm
F = NkT n 2 cosh

k BT

2dJm
F

f =
=
kT n 2 cosh
0)
(h =
N
k BT

Write, x = A(T ) m and look at the expansion of n {2 cosh x}

x 2 x 4

n {2 cosh x}= n e x + e x= n 2 1 +
+
+
!
2 24

x2 x4

= n 2 + n 1 +
+
+

!
2 24

x2

+
n
x
x
careful!
(1
)

x2 x4
x4
n 2 +
+

2((
24
24

n 2 +

1st term
2

2nd term

x
1
x4
2 12

f =
kT n 2 kT

m2 2
kT 4
A (T ) +
A (T )m 4
2
12

=
a0 (t ) + a2 (T )m 2 + a4 (T ) m 4

Landau expansion

(2)

Classification of Phase Transitions: Ehrenfests Scheme

Note that when we derived Clausius-Clapeyron Eq. we moved along the co-existence curve and for each
point we have used 1 = 2
But we have not put any restrictions on

G
V=
;
P T

G
P

or

G
T

G
S =
normal derivatives of G
P P

If these derivates are discontinuous First Order Phase Transition

If these derivates are continuous, but the higher order derivates are discontinuous n-th order transition.
Unfortunately, Nature has not co-operated with Ehrenfests scheme except for the important case of First
Order Phase Transition.
For example, Sp. Heat 2nd derivative
Diverges at the critical point of a magnetic phase transition
Modern Usage 1st order and Continuous Phase Transition
Unambiguous Definition of Phase Transition
In single phase regions of the P-T plane, G (T, P) is an analytic function bounded by co-existence curves
along which the function is not analytic.
Then, Taylor expansion would not converge as one crosses co-existence curves.

Landaus Phenomenological Theory

Near the critical point, free-energy can be expanded in terms of the order parameter m.
Minimize the free-energy in terms of m.

Landau Free-Energy GL ( )
order parameter
Not quite the Gibbs free-energy but will not get into that. In the absence of any external field they behave
quite similarly.

GL ( ) =a0 (T ) + a1(T ) +

a2
a
(T ) 2 + 3 (T ) 3 +
2
3

The expansion is motivated by the expansion of the mean-field equation of state for m small i.e., near Tc
Symmetry: in the absence of applied field GL ( ) is symmetric under

GL ( ) =
a0 (T ) +

a2 ( T ) 2 a4 (T ) 4
+
+
2
4

1. Continuous Transition
a4 is positive
a2 changes sign when T = T0

a2=
(T ) A2 (T T0 )
GL
=0 = s

T < T0 , a 2 is negative

a
A
2=
2 (T T0 )
a2 s + a4 s3 =
0
s = 0 or s2 =
a4
a4

A2
(T0 T )1/2
a4

=
s

is continuous at T = Tc . No Latent Heat

1
2

Tc = T0

= 0 is maximum.
2. First order phase transition
(i) Symmetry survives
(Ferroelectric crystals like Potassium dihydrogen phosphate or KH2PO4 )

a2=
(T ) A2 (T T0 )

a4 is negative.

We need a6 term (positive) in the expansion to produce a first-order phase transition.

GL ( ) =a0 (T ) +

1
1
1
A2 (T T0 ) 2 a4 (T ) 4 + a6 (T ) 6
2
4
6

First Order Phase Transition: Discontinuous jump in the order parameter which is related to
Latent Heat (via the discontinuity in entropy).

Conditions:
1)

GL
=0

minimum

2) At Tc , GL (=
0,=
T T=
s ,=
T Tc )
c ) GL (=

The weak secondary minimum seen above Tc becomes equal strength at Tc and then
becomes deeper i.e. global minimum below Tc .
GL
From
A2 (T T0 ) s a4 s3 + a6 s5 =
0
=0

=
s 0 or A2 (T T0 ) a4 s2 +=
a6 s4 0

(1)

a4 a42 4a6 A2 (T T0 )
s2 =
2a 6
a42 4a6 A2 (T T0 ) for the second minimum to develop.
Let T = T1 when
=
a4 (T1 ) 2 4a6 (T1 ) A2 (T1 T0 ) . Then T < T1 for the double well shape to form.

A2 ( Tc T0 ) a4 (Tc ) s2 + a6 ( Tc ) s4 =
0

Now at Tc From (1)

GL (0, Tc ) = GL ( s , Tc )
1
1
1
a0 (Tc ) = a0 (Tc ) + A2 (Tc T0 ) s2 a4 (Tc ) s4 + a6 (Tc ) s6
2
4
6
1
1
1
A2 (Tc T0 ) a4 s2 + a6 s4 =
0
(3)
2
4
6

Also recall the second condition:

s2 =

From (2) and (3)

3 a4 (Tc )
4 a6

(4)

Substituting s2 in eq.(2)

3
A2 (Tc T0 ) = a4
4

a4
3 2
9 a4
a4 / a6 > 0
a 6 =
16
16 a 2
a6
6

Tc > T0
When both a2 and a4 vanish at the same temp. Tri-critical points.
(ii) Asymmetric case , 3 terms may appear (but term can be eliminated)
Nematic Liquid Crystals

GL ( ) =a0 (T ) +

1
1
1
A2 (T T0 ) 2 + a 3 (T ) 3 + a4 (T ) 4
2
3
4

(2)

Magnetic Phase Transition in the Infinite-range model and its connection to

the Mean-Field Model
J
i j
N 1i< j N

H =

N 1 term necessary for thermodynamic limit to exist

21 2 = (1 + 2 ) 12 22
2

Note: 2 (1 2 + 2 3 + 1 3 ) = (1 + 2 + 3 ) 12 22 32
2

etc.
Hence,
2

J N
J N
H {sss

i2
}=
i +

2N i 1 =
=
2N i 1

but s i2 = 1, so
2

J N
J
H {ss

N
}=
i +
N
2 N i =1
2

QN =

=e

K
2

e H { i }

K
( i )2
2
N
e

define J = K

{ i }

{ i }
a2
e2

x2

+ ax
2

dx
e

1
This is just a Gaussian integral but in this
2
context it is called the Hubbard-Stratanovich Transformation
Use the identity

1/2

K
Choose a =
N

QN

K
2

{ i }

1
2p

si

x 2
K
exp + x
N
2

i dx

Spins are now decoupled and we know that i = 1

=
QN e

Note:

K
2

K
x 2
exp 2 cosh x
dx
N
2

{ i }

1
2p

sss
+ + N )
e ( 1 2

sss
1 =1
2
N

es1 es 2 es 3

sss
1
2
N
N

s i
= =
e

s i

( 2 cosh ) N

New variable m = x( KN ) 1/2

1/2 K

e 2

KN
QN =

Note: b

Km 2

N
n 2cosh( Km )
2

dm

= e c = N n b
c

Saddle Point or Method of Steepest Descent to evaluate the integral:

write =
f ( m) K

m2
n 2 cosh( Km)
2

Minimum value of f(m) contributes the most to the integral when N is very large.
=
f (m) 0,=
m m0
K m0 =

1
2 K sinh( Km0 )
2 cosh( Km0 )

m0 = tanh( Km0 )
K < 1 only one solution
J
K=
= 1 and we find that
c 1
kTc
1/2 K

e 2

KN
QN =

Km 2

N 0 n 2cosh( Km0 )
2

How do we know m0 is the magnetization?

Need to consider the presence of an applied field
1/2 K

e 2

KN
m0 = tanh( Km0 + B ); QN =

2
B=

m 2

N K 0 n 2cosh( Km0 + B )
2

h
kT

F = kT n QN
1 KN K

m02

=
kT n

+
+
Nk
N
n
Km
B
2
cosh(
)

2
2 2 2

F
1
1

=
+ kT N
M=
2sinh( Km0 + B)
h
kT
2 cosh( Km0 + B)
= N tanh( Km0 + B)
M
= tanh( Km 0 + B ) m m0
N