Statistical Mechanics

Lecture notes Baruch Horovitz and class of 2007

Contents
1. Ensemble Theory 2
1a. Thermodynamics (Review) 2
1b. Micro Canonical Ensemble (MCE) 5
1c. Ensemble theory - generalities 12
1d. Canonical Ensemble (CE) 14
1e. Grand Canonical Ensemble (GCE) 19
2. Quantum Statistical Mechanics 23
2a. Ensembles for ideal quantum gases 23
2b. Ideal Bose gas 28
2c. Ideal Fermi gas 34
2d. Non- Ideal gases 36
3. Phase Transitions 39
3a. First order transitions 39
3b. Second order phase transitions Mean Field theory 45
3c. Exact results 46
3d. Landaus Theory for second order phase transitions 49
4. Non- equilibrium 53
4a. Kinetic theory and Boltzmanns equation: 53
4b. Brownian motion 56
4c. Fluctuation Dissipation Theorems (FDT) 60
4d. Onsagers Relations 68
Appendix: Langevins equation from a Hamiltonian 72
1
Reference books:
R. K. Pathria, Statistical Mechanics
Landau & Lifshitz, Statistical Physics
K. Huang, Statistical Mechanics
S.-k Ma, Statistical Mechanics
F. Mohling, Statistical Mechanics Methods and Applications
G. H. Wannier, Statistical Physics
1. ENSEMBLE THEORY
1a. Thermodynamics (Review)
Macroscopic state: Set of measurable coordinates of a system with many () degrees
of freedom, e.g. volume V , number of particles N, energy E.
Equilibrium: A macroscopic state that is uniquely determined by a small number of external
forces, e.g. pressure P, chemical potential , temperature T.
Pairs of force f and coordinate x generate work W = fdx (W is not necessarily an exact
dierential, i.e. equation may not be integrated to yield a state function W).
E.g. W = PdV , W = dN
Microscopic degrees of freedom contain heat energy Q, with S the coordinate, T the force.
Q = TdS.
Thermodynamic limit: N , V with N/V const. Macroscopic state, equilibrium
etc. are dened in this limit. Dene: Extensive variables that increase with N, e.g. V, E, S
and intensive variables that are constants in the thermodynamic limit, e.g. N/V, P, .
First Law: Two ways to exchange energy, work or heat: E = Q+W; heat Q is due
to microscopic degrees of freedom. There exists an adiabatic process for which Q = 0.
Entropy S is a thermodynamic coordinate - proof:
2
Adiabatic process Q = 0. Equilibrium determines a
curve E(x) from dE = fdx. Between curves change is
nonadiabatic, is an integration constant.
E = E(x, ). Curves do not cross since f =
E
x
is unique in equilibrium. Also E(x) is single
valued E(x, ) is monotonic in . For constant x, Q = dE = (
E

)
x
d
dE = fdx + d = (
E

)
x
is not unique - can choose = A() with A() monotonic. Choose which is extensive
S N, and dene temperature T =
E
S
as intensive. Assumption in proof: only one
coordinate.
dE = TdS PdV + dN
E = E(S, V, N) is a (single valued) state function,
_
dE = 0. This is the rst law.
Heat is not single valued -
_
TdS depends on how V, N change along the path (
_
Q ,= 0).
Second Law: In a closed system S(t) increases with time. Exchange dE
1
= dE
2
in two
subsystems:
S = S
1
+ S
2
dS
dt
=
dS
1
dE
1
dE
1
dt
+
dS
2
dE
2
dE
2
dt
=
_
1
T
1

1
T
2
_
dE
1
dt
> 0
energy ows from high to low T.
In equilibrium S is maximal T
1
= T
2
dS =
1
T
dE +
P
T
dV

T
dN
If volumes of subsystems exchange dV
1
= dV
2
with T
1
= T
2
(hence

E
terms vanish),
dS = (
S
1
V
1
)
E,N
dV
1
+ (
S
2
V
2
)
E,N
dV
2
= (
P
1
T

P
2
T
)dV
1
P
1
= P
2
equilibrium.
Particle exchange dN
1
= dN
2
dS = (S
1
/N
1
)
E,V
dN
1
+ (S
2
/N
2
)
E,V
dN
2
= (
1
/T
2
/T)dN
1

1
=
2
chemical equilibrium.
TdS > dE+PdV dN = Q irreversible process, i.e. S increases more than its equilibrium
change.
Adiabatic process:
A process in which the energy is changed only by slow variation of external conditions so
3
that at every instant the system is in equilibrium. Furthermore, the system is thermally
isolated - no energy transfer except the external condition.
Adiabatic process is reversible: expand in d/dt, external condition (e.g. volume)
dS
dt
= a + b
d
dt
+
1
2
c(
d
dt
)
2
a = 0: equilibrium condition
b = 0: since
dS
dt
> 0 cannot depend on sign of
d
dt

dS
d
=
1
2
c
d
dt
+ ..... 0 when
d
dt
0.
dS = 0 in adiabatic process.
Note: A reversible process in a closed system is an adiabatic process and S is constant. A
reversible process in an open system is a process for which TdS = dE + PdV dN, i.e.
heat exchange with a reservoir is allowed and dS ,= 0.
Thermodynamic Functions
F = E TS Helmholtz free energy
dF = SdT PdV + dN F(T, V, N)
If an additional variational parameter is present F is minimized to reach dF = 0.
T, V, N xed: dF = dE TdS 0, for < 0 the process is irreversible.
G = E TS + PV Gibbs free energy
dG = SdT + V dP + dN G(T, P, N)
T, P, N xed: dG = dE TdS + PdV < 0 and G is minimized.

= F N
d

= SdT PdV Nd

(T, V, )
T, V, xed: d

= dE TdS dN < 0 and

is minimized.
Extensiveness: E = E(S, V, N)

[
=1
:
E = TS PV + N
F = PV + N
G = N

= PV (1)
4
1b. Micro Canonical Ensemble (MCE)
Basic assumption:
S = k
B
ln (E)
Boltzmans constant: k
B
= 1.38 10
16
erg/deg.
(E) is the number of microstates for a given N, V, E, and states have equal weight :
(E) =

EHr
r are microscopic degrees of freedom of the N particles.
We shall now prove extensivity and maximum property: Consider two subsystems, neglect
their interaction (justied since the number of particles near the surface is small, or surface
energy bulk energy),
Hr
1
, r
2
= H
1
r
1
+ H
2
r
2

(E) =

r
1
,r
2

EH
1
r
1
H
2
r
2

E
1

r
1

E
1
H
1
r
1

r
2

EE
1
H
2
r
2

E
1
exp
_
S
1
(E
1
) + S
2
(E E
1
)
k
B
_
Since S
i
N
i
are large look for maximum in E
1
(equivalent to steepest descent),

E
1
(S
1
(E
1
) + S
2
(E E
1
)) [
E
1
= 0
S
1
E
1
=
S
2
E
2
T
1
= T
2
Expand to 2nd order
S
1
(E
1
) + S
2
(E
2
) = S
1
(

E
1
) + S
2
(

E
2
)
1
2k
1
(E
1

E
1
)
2

1
2k
2
(E
2

E
2
)
2
1
k
i
=
_

2
S
E
2
i
_
E
i
=

E
i
=
(
1
T
)
E
i
=
1
T
2
C
i
C
i
=
_
E
i
T
_
V,N
e
S(E)/k
B
= e
S
1
(

E
1
)+S
2
(

E
2
)

E
1
exp
_
1
2k
1
(E
1

E
1
)
2

1
2k
2
(E
2

E
2
)
2
_
Probability that E
1
,=

E
1
is Gaussian, (E
1

E
1
)
2
)
1/2

k
1

N
5
Dominant term is E
1
=

E
1
with

E
1
such that T
1
= T
2
.
If is the spacing of E
1
levels (e.g. V
2/3
in ideal gas)

E
1

_
dE
1

exp
_
(
1
2k
1
+
1
2k
2
)(E
1

E
1
)
2
_
=
1

_
2
k
1
k
2
k
1
+ k
2
_
1/2
S(E) = S
1
(

E
1
) + S
2
(

E
2
) + O(ln N)
Extensivity, ie. (E) is dominated by one term: (E) =
1
(

E
1
)
2
(

E
2
)
(Note: The only function f() which is additive when =
1

2
is f ln )
Energy uctuation

k
i
= T

C
V
.
Similarly, separate to subsystems with V = V
1
+ V
2
or N = N
1
+ N
2
to obtain dominant
term at

V
i
(hence P
1
= P
2
) or at

N
i
(hence
1
=
2
).
Ideal Gas (no interactions)
Classical: r (p, q). For one particle energy (p):
cl

_
d
3N
q
_

i
(p
i
)=E
d
3N
p V
N
.
Note missing dimensional prefactor.
P
k
B
T
=
_
ln
V
_
N,E
=
N
V
Quantum: r quantum numbers. Use periodic boundary conditions, e.g. e
ip
x
L/
= 1
p
x
=
h
L
n
x
, n
x
integer < n
x
<
=
h
2
2mL
2
(n
2
x
+ n
2
y
+ n
2
z
)
(E) is the number of solutions to

3N
r=1
n
2
r
=
2m
h
2
EV
2/3
S(N, V, E) = S(N, V
2/3
E)
In an adiabatic process V
2/3
E = const, P =
_
E
V
_
N,S
=
2E
3V
PV
5/3
= const
6
(E) irregular - No. of points on surface of 3N sphere.
Instead (N, V, E) =

E
(N, V, E
t
) has reasonable E limit.
Replace by = number of microstates with E

2
<energy< E +

2
(N, V, E; ) =

E
is the volume of 3N sphere with radius R = (2mV
2/3
E/h
2
)
1/2
Volume of n sphere: V
n
= C
n
R
n
with surface area dV
n
= nC
n
R
n1
dR
_

e
x
2
dx =

. . .
_
e

n
i=1
x
2
i

i
dx
i
=
n/2
In spherical coordinates:
_

0
e
r
2
nC
n
r
n1
dr =
_
n
2
_
!C
n
=
n/2
V
n
=

n/2
R
n
(n/2)!
(N, V, E) =

3N/2
(3N/2)!
(2mV
2/3
E/h
2
)
3N/2
=
3N
2E
N : ln N! = N ln N N
ln = N ln
_
V
h
3
_
4mE
3N
_
3/2
_
+
3
2
N + ln
3N
2
+ ln

E
level spacing: n
r
n
r
+ 1 for some r yields E E +
2n
r
+ 1 =
2m
h
2
V
2/3
0 n
r
<
_
2m
h
2
EV
2/3
_
1/2
N
5/6
N
2/3
< < N
1/6
ln /E = O(ln N)
7
ln = ln + O(ln N)
Volume near surface dominates total volume for N 1. Value of is not important.
S(N, V, E) = k
B
ln = Nk
B
ln
_
V
h
3
_
4mE
3N
_
3/2
_
+
3
2
Nk
B
T =
_
E
S
_
N,V
=
2E
3Nk
B
E =
3
2
Nk
B
T
1
2
mv
2
) =
3
2
k
B
T
C
v
=
_
E
T
_
N,V
=
3
2
Nk
B
P =
_
E
V
_
N,S
=
2E
3V
PV = Nk
B
T
C
p

_
(E + PV )
T
_
N,P
=
5
2
Nk
B
P
_
V
T
_
N,P
is excess work at constant P.

C
p
C
v
=
5
3
Gibbs paradox:
S is not extensive. Even mixing gases 1,2 with equal T, n leads to S
total
,= S
1
+ S
2
The mixing entropy is positive:
Nk
B
ln V N
1
k
B
ln V
1
N
2
k
B
ln V
2
= k
B
_
N
1
ln
V
V
1
+ N
2
ln
V
V
2
_
> 0
But this must be reversible !
Quantum mechanics - indistinguishability. Classical limit (h 0) should give
1
N!

S(N, V, E) = Nk
B
ln
V
N
+
3
2
Nk
B
_
5
3
+ ln
_
2mk
B
T
h
2
__
Sackur Tetrude eq. : S is extensive.
Classical derivation ...

cl
=
1
w
0
_
d
3N
q
_

i
p
2
i
<2mE
d
3N
p =
1
w
0
V
N
C
3N
(2mE)
3N/2
8
w
0
is chosen with

h 0

cl

=
1
N!
C
3N
(
2mE
h
2
)
3N/2
w
0
= N!h
3N
Where
1
N!
is the Gibbs correction and h is the volume of one state in the p,q space.
Ex: evaluate w
0
for harmonic oscillator.
Equipartition
x
i
= q
i
or p
i
x
i
H
x
j
)
_
. . .
_
E
1
2
<H<E+
1
2

x
i
H
x
j
d
_
. . .
_
E
1
2
<H<E+
1
2

d
=

E
_
. . .
_
0<H<E
x
i
H
x
j
d

E
_
. . .
_
0<H<E
d
where d =
d
3N
qd
3N
p
N!h
3N
. Noting that
E
x
j
= 0 and integrating by parts
numerator =
_
. . .
_
0<H<E
x
i

x
j
(HE)d =
_
. . .
_
0<H<E
dx
k,=j
[x
i
(HE)]
x
(2)
j
x
(1)
j

ij
_
. . .
_
0<H<E
(HE)d
Assume H(x
j
) and monotonic, e.g.
p
2
2m
or at walls of container.
At boundaries, x
(1,2)
j
, all energy is in x
j
H(x
(1)
j
) = H(x
(2)
j
) = E
therefore, the rst expression on the right side vanish. Returning to x
i
H
x
j
) we have:
x
i
H
x
j
) =
ij

E
_
. . .
_
0<H<E
(E H)dw

E
_
. . .
_
0<H<E
dw
=
ij
_
. . .
_
0<H<E
dw

E
_
. . .
_
0<H<E
dw
=

ij

E
ln
_
. . .
_
0<H<E
dw
=
k
B

ij
(
S
E
)
N,V
=
i,j
k
B
T
concluding:
x
i
H
x
j
) =
i,j
k
B
T x
j
with H(x
j
) monotonic unbounded.
E.g. (Specic examples):
x
i
= x
j
= p
i
p
i
H
p
i
) = p
i
q
i
) = k
B
T
x
i
= x
j
= q
i
q
i
H
q
i
) = q
i
p
i
) = k
B
T
for quadratic Hamiltonians (usually at high T)
H =

j
A
j
p
2
j
+ B
j
q
2
j
9
for such systems we clearly have:

j
(p
j
H
p
j
+ q
j
H
q
j
) = 2H H) =
1
2
fk
B
T
f=No. of degree of freedom number of quadratic terms in H.
Each harmonic term in the quadratic Hemiltonian makes a contribution of
1
2
kT towards the
internal energy of the system and hence a contribution of
1
2
k towards the specic heat C
v
.
Example: molecule with m atoms
C.M: 3 translations + 3 rotations non co-linear molecule
3 translations + 2 rotations co-linear
(To understand the signicance of colinearity, note that quantum levels

2
(+1)
2I
form a
classical continuum if

2
I
k
B
T. However, if I 0 as in a colinear case, at k
B
T

2
I
only
the single ground state is relevant)
3m coordinates no. of vibrations = 3m-6 non-collinear
or = 3m-5 collinear
Translation (
p
2
x
2m
), rotation (
L
2
x
2I
) - 1 quadratic term
Vibration (
p
2
x
2m
+
m
2
x
2
2
) - 2 quadratic terms.
H) =
1
2
[6 + 2(3m6)]Nk
B
T = (3m3)Nk
B
T non collinear
H) =
1
2
[5 + 2(3m5)]Nk
B
T = (3m5/2)Nk
B
T collinear
Virial Theorem (Clausius 1870) ...
The Virial of a system is by denition, the sum of the products of the coordinates of the
various particles and the representative forces acting on them:
1
3N

i=1
q
i
p
i
) = 3Nk
B
T
E.g. ideal gas in box:
10
p ,= 0 only from walls at q
i
= r

p = P

ds
where: P is pressure, ds is surface element and p is
momentum hitting ds (of all particles)
1 =

r
r

p = P
_
s
r

ds = P
_
(

r)dv = 3PV PV = Nk
B
T
Consider now classical particles i, j with 2-body interaction
u(r
i,j
) : r
ij
= [r
i
r
j
[.

p
i
=

r
i

j
u([r
i
r
j
[)

i
r
i

p
i
=

i,j
r
i

r
i
u([r
i
r
j
[),
u(r)
r
=
r
2
r
u
r
2
Sum over pairs:

i
r
i

p
i
=

i<j

u
r
2
ij
[ r
i

(r
i
r
j
)
2
r
i
+ r
j
(r
j
r
i
)
2
r
j
] + wall pressure =

i<j
u
r
2
ij
[
2r
2
ij
..
2r
i
(r
i
r
j
) + 2 r
j
( r
j
r
i
)] =

i<j
r
ij
u
r
ij
=
The net contribution, arising from all the
N(N1)
2
pairs of particles, (N 1):
1
2
N(N 1)r
u
r
) =
1
2
N
2
_ _
r
12
u
r
12
g(r
1
r
2
)d
3
r
1
d
3
r
2
/V
2
=
N
2
2V
_
r
u
r
g(r)4r
2
dr
g(r) pair distribution function
1 =
N
2
2V
_

0
r
u
r
g(r)4r
2
dr 3PV = 3Nk
B
T
PV = NK
B
T[1
2n
3k
B
T
_

0
u(r)
r
g(r)r
3
dr]
Also:
E =
3
2
NK
B
T[1 +
4n
3k
B
T
_
u(r)g(r)r
2
dr]
where the 1st term in the square brackets stands for the Kinetic energy.
11
1c. Ensemble theory - generalities
(p, q, t)d
3N
qd
3N
p is the probability of microstate(p,q), with normalization
_
E=const
(p, q, t)d
3N
qd
3N
p = 1
Average: f) =
_
f(p, q)(p, q, t)d
3N
qd
3N
p
Equilibrium: all observables
<f>
t
= 0

t
= 0. In mechanics we start with a point
p(0),q(0) in the 6N dimensional phase space, follow trajectory and average on time.
Ergodic theorem: Long time average = ensemble average, i.e. probability concept (p, q, t)
is equivalent to mechanics.
Liouvills theorem
p(t), q(t) satisfy Hamiltons equation and dene a velocity in phase space
v = ( q, p)
Net ow of states from volume , with surface , is =

t
_
d

(v n)d =
_

div(v)d =

t
_

d
true for any . Continuity:

t
+ div(v) = 0.
Consider (q
i
, p
i
, t) for a collection of states q
i
(t), p
i
(t)

t
+

i
_

q
i
q
i
+

p
i
p
i
_
+

i
_
q
i
q
i
+
p
i
p
i
_
= 0
q
i
q
i
=
H
q
i
p
i
=
p
i
p
i
and therefore
_
q
i
q
i
+
p
i
p
i
_
= 0 leads to
d
dt
=

t
+ [, H]
poisson
= 0
Local density as viewed on moving points, is constant, hence
d
dt
= 0.
No. of states is conserved incompressible uid.
Equilibrium:

t
= 0

i
_

q
i
q
i
+

p
i
p
i
_
= 0
12
solutions:
(p, q) =
_
_
_
const E
1
2
< H(p, q) < E +
1
2
microcanonical
0 otherwise
In general can depend on constants of motion, e.g. H(p, q).
[H(q, p)] canonical ensemble
In microcanonic (p, q, t) =
1

S = k
B
ln = k
B
_
(p, q, t) ln (p, q, t)d
3N
pd
3N
q
denes entropy also in other ensembles, in equilibrium. Is this form of S valid at non-
equilibrium? but then

dS
dt
= 0
which violates the second law.
Arrow of time
Consider volume expansion III. For N = 10
20
, volume increases by factor 2.
The increase in number of states is 2
10
20
!! all states of I evolve into II, but only fraction
1
2
10
20
of states in II evolve into I. Probability suggests that S(t) increases.
Just probability is not sucient:
Consider IIIIII, increasing volumes. States in II most probably go to III, but where do
states in II come from - most probably also from III !? not I?
Time reversal invariance: a state x
1
in I evolves into x
2
in II. Now reverse all momenta in
x
2
Rx
2
.
Rx
2
is a state in II. Its time evolution yields x
1
in I, i.e. it is possible to nd a state that
lowers entropy.
The diculty with Rx
2
is that it must be prepared accurately; a minute perturbation causes
instability (chaos, the buttery eect). Nature does not allow perfect aiming with accuracy
2
10
20
.
13
need probability + stability, i.e. information on nearby trajectories, coarse grained,
then entropy can increase with time.
Hypothesis: Universe started with low entropy, low entropy radiation from the sun produces
low entropy food etc...
1d. Canonical Ensemble (CE)
Consider a system that is embedded in a larger heat
bath. Energy exchange is allowed, i.e. E
r
is not xed.
r is point in phase space of the system.
Reservoir energy E
R
, E
r
<< E
0
E
R
+ E
r
= E
0
= const
probability of state r P
r

R
(E
R
), i.e. the number of states of the reservoir.
[note: No. of all states with energy E
r
=
r
(E
r
)
R
(E
R
)]
P
r

R
(E
0
E
r
)
ln
R
(E
0
E
r
) = ln
R
(E
0
)
ln
R
E
[
E
0
E
r
+ = ln
R
(E
0
) E
r
P
r
=
e
E
r

r
e
E
r
=
1
k
B
T
is determined by the reservoir. The partition function is dened by
Z
N
(V, T) =

r
e
E
r
Dene F BY Z
N
= e
F(T,V,N)
. Identify F:
0 =

r
e
E
r
+F
=

r
e
E
r
+F
_
E
r
+ F +
_
F

_
V
_
F(T, V ) E(T, V ) T
_
F
T
_
V
= 0
solution of this dierential equation is the free energy F = E TS.
Alternative proof: Assume dominant energy in sum

r
:
Z
N
(V, T) =

r
e
E
r
=

E
(E)e
E
=

E
e
S(E)/k
B
E
14
dominant term at E

E
(S(E)/k
B
E) = 0
S(E)
E
[
E
=
1
T
E

is such that the MCE at E

has the temperature T.

S(E

)/k
B
E

= F
MCE
where F
MCE
is the free energy of the MCE dened with variable E

, V, N.
Note: S(E

(T, V, N), V, N) denes F(T, V, N).

Fluctuation near E

(expanding S around E

):
S(E)
k
B
E =
S(E

)
k
B
E

+
1
2k
B

2
S
E
2

(E E

)
2
+ ... ,
where,

2
S
E
2
=
(1/T)
E
=
1
T
2
1
E/T
=
1
T
2
C
V
,
and T, C
V
are for MCE.
Z
N
(V, T) = e
F
MCE

E
e
(EE

)
2
/2k
B
T
2
C
V
Gaussian.
The weight of E ,= E

decreases rapidly with width,

_
(E E

)
2
)
E

k
B
T
2
C
V
E

N
0.
To nd the partition function, we replace the Sum with Integral,

_
dE,
where is the energy level spacing (as above N
a
ln ln N), and
Z
N
(V, T) =
e
F
MCE

_
e
E
2
/2k
B
T
2
C
V
dE =
e
F

2k
B
T
2
C
V

= e
F
CE
,
F
CE
= F
MCE
+O(ln N),
where F
CE
, F
MCE
O(N).
Insensitivity of thermodynamic results to type of ensemble due to:
1. (E) e
(...)N
e
(...)E
: exponential increase.
15
2. Thermodynamic limit: E, N .
E.g. for ideal gas, (E)e
E
e
(3/2)N ln EE
with maximum at E

=
3
2
Nk
B
T. For E ,= E

,
(E)e
E
practically vanishes.
Note: The energy uctuations can also be identied by evaluating the specic heat:
C
V
=

T
E) =
1
k
B
T
2

r
E
r
e
E
r

r
e
E
r
=
1
k
B
T
2
_

E
2
_
+E)
2
_
,
where

(E E))
2
_
=

E
2
E)
2
_
= k
B
T
2
C
V
N.
Examples:
Classical systems: the general expression for the partition function,
Z
N
(V, T) =
1
N!h
3N
_
e
1(p,q)
d
3N
qd
3N
p,
where, N! is Gibbs normalization, and h corresponds to a volume of one state in the
q, p phase space.
Ideal gas: H =

i
p
2
i
/2m,
Z
N
(V, T) =
V
N
N!h
3N
__

0
e
p
2
/2m
4p
2
dp
_
N
=
1
N!
_
V

3
_
N
,
where,

h

2mk
B
T
, the thermal wavelength.
This corresponds to a deBroiglie wavelength for momentum

mk
B
T i.e.
h/
_
p
2
) = h/

mk
B
T.
F = k
B
T ln Z
N
= Nk
B
T
_
ln N
3
/V 1
_
,
P =
_
F
V
_
N,T
=
Nk
B
T
V
,
S =
_
F
T
_
N,V
= Nk
B
_
ln V/N
3
+ 5/2
_
, as in MCE,
=
_
F
N
_
T,V
= k
B
T ln N
3
/V.
16
Consider N noninteracting molecules with internal energies
i
, n
i
molecules at level
i
are indistinguishable.
Z
N
(V, T) =

n
i

1
n
0
!n
1
!...
e

i

i
n
i
_
_
_
_
_
e
p
2
/2m
d
3
pd
3
q
. .
center of mass
_
_
_
_
N
=
V
N
N!
3N
_

i
e

i
_
N
, using multinomial expansion
=
1
N!
(Z
1
(V, T))
N
, Z
1

V

i
e

i
partition of one molecule.
Note: The multinomial expansion is:
_

i
a
i
_
N
= N!

n
i

1
n
1
!n
2
!...
a
n
1
1
a
n
2
2
... ,
where the sum on distributions n
i
is restricted by

i
n
i
= N.
Diatomic gas:
H =
0
..
electron
+ ( + 1/2)
. .
vibration
+
2
k(k + 1)/2I
. .
rotation
,
Z = e

=0
e
(+1/2)

k=0
(2k + 1)e

2
k(k+1)/2I

s
(2s + 1)
. .
nonidentical atoms
,
where s is the spin of the two atoms.
Z
rot
=
T

_

0
2ke

2
k
2
/2I
dk = 2I/
2
,
Z
rotclass
=
1
h
2
_
e
(M
2
x
+M
2
y
)/2I
dM
x
dM
y
d
x
d
y
= 2I/
2
,
The two atom axis is z and d
x
d
y
= d 4 is the solid angle.
If the two atoms are identical, we need QM:
There is a constraint: s+k needs to be even. [Orbital exchange has ()
k
, spin exchange
has ()
ss
1
s
2
, hence both Fermion and Boson symmetries are obeyed for s +k even].
17
Spin k degeneracy = 2s + 1
H
2
s = 1 odd 3 orthohydrogen
s = 0 even 1 parahydrogen
D
2
s = 2 even 5
s = 1 odd 3
s = 0 even 1
Z
H
2
rot
= 3

k odd
(2k + 1)e

2
k(k+1)/2I
+

k even
(2k + 1)e

2
k(k+1)/2I
,
Z
D
2
rot
= 3

k odd
... + 6

k even
...
T

1
2
2I

s
(2s + 1).
In the last equation, the
1
2
is the classical reduction of angular integration by factor of
2, for identical atoms.
Equipartition:
_
x
i
H
x
j
_
=
_
x
i
H
x
j
e
H
d
_
e
H
d
,
numerator =
_
x
i
H
x
j
e
H
d =
1

_
dx
i,=j
x
i
e
H

x
(2)
j
x
(1)
j
+
1

_
e
H
d
ij
.
In general, the energy H(x
(1,2)
j
) at the boundaries of x
j

_
x
i
H
x
j
_
=
ij
k
B
T.
E.g. ideal gas:
_
p
ix

p
ix
p
i
2
2m
_
=
1
m

p
2
ix
_
= k
B
T,
E =

i
_
p
i
2
2m
_
=
3
2
Nk
B
T.
Extreme relativistic: H =

i
c [ p
i
[,
q
ix
=
H
p
ix
=

p
ix
c
_
p
2
ix
+ p
2
iy
+ p
2
iz
=
cp
ix
[ p
i
[
,
_

i
p
i

q
i
_
=
_

i
c [ p
i
[
_
= H) = 3Nk
B
T (>
3
2
Nk
B
T),
as the power of p smaller, the energy is higher for given temperature.
18
1e. Grand Canonical Ensemble (GCE)
System in heat bath and particle reservoir.
E
r
E
R
, N
r
N
R
E
r
+ E
R
= E
0
= Const.
N
r
+ N
R
= N
0
= Const.
The combined system is microcanonic.
r is a point in phase space of the system, which now includes all N
r
(N
0
). The probability
of a point r with E
r
, N
r
is P
r
P
r

R
(N
0
N
r
, E
0
E
r
)
ln P
r
ln
R
(N
0
, E
0
)
ln
R
N
[
N
0
N
r

ln
R
E
[
E
0
E
r
= ln
R
(N
0
, E
0
) + N
r
E
r
= 1/k
B
T, is imposed by the resevoir
P
r
= e
N
r
E
r
/L
The grand partition function is
L(, T, V ) =

r
e
N
r
E
r
e

(,T,V )
Identify

:
0 =

r
e
N
r
E
r
+

r
_
N
r
E
r
+

_
P
r
= N E +

T
The energy E(, T, V ) shows that the solution to this dierential equation is the theromo-
dynamic potential, since

= F N = E N TS = E N + T

T
Alternatively, if

N
is dominated by one term
L e
N

r,Nfixed
e
E
r
= e
NF

= F N = PV
19
Dene fugacity = e

L(, T, V ) =

N=0

N
Z
N
(V, T)
So that the weight of each N =
N
Z
N
/L.
N =

ln L(, T, V )[
T,V
E =

ln L(, T, V )[
,V
etc.
Ideal gas: with internal energies
i
for each particle,
Z
1
=
V

i
e

i
=
V

3
a(T)
Z
N
=
1
N!
Z
N
1
L =

N
1
N!
Z
N
1
= e
Z
1
PV
k
B
T
= Z
1
= e

3
a(T) = k
B
T ln
P
3
k
B
Ta(T)
N =

ln L(, V, T) = Z
1
=
PV
k
B
T
= k
B
T ln(
n
3
a(T)
)
n
3
_1

Particle (N) uctuations:
Weight W(N) = e
NF(N,V,T)
/L
Maximum at N such that =
F
N
[
N
, i.e.
can
=
F
N
equals if N is chosen at this maximum,
N = N.
For maximum need

2
F
N
2
[
N
> 0:
F(N, V, T) = Nf(v) v =
V
N
For xed V ,

N
=
v
N

v
, hence
F
N
= f(v) v
f
v


2
F
N
2
=
v
2
N

2
f
v
2
Since P(v) =
f
v


2
F
N
2
[
N
=
v
2
N
P
v
(or =

N
) we need (
P
v
)[
T
< 0 for W(N) being
a maximum ( this is Van Hoves theorem, Huang rst edition p.321 - general proof; Huang
second edition p.206 - for the special case of hardcore interaction).
Physically obvious: if P > P
ext
the net force increases V ; now if
P
V
> 0 P increases even
more and equilibrium (P = P
ext
) is further away.
20
CE and GCE are equivalent by this thermodynamic stability criterion (
P
v
)[
T
< 0.
Note however, that at a 1st order phase transition, e.g. the gas-liquid transition, P/v = 0
and GCE is not equivalent to CE, hence large N uctuations are expected.
W(N) W(

N)e

1
2
(NN)
2

N
2
_1
2
=

k
B
TN
v
2
(P/v)

N
2
_1
2
/N 1/

N 0
Energy uctuations:
E =

r
E
r
P
r
=
_
ln L

_
,V
E
2
=
1
L
_

2
L

2
_
,V
E
2
= E
2
E
2
=
_

2

2
ln L
_
,V
=
_
E

_
,V
= k
B
T
2
E
T
[
,V
E (N (, T, V ) , T, V )
T
=
_
E
T
_
N,V
+
_
E
N
_
T,V
_
N
T
_
,V
= C
V
+
_
E
N
_
T,V
_
N
T
_
,V
dE = TdS PdV + dN
_
E (S (N, T, V ) , V, N)
N
_
T,V
= + T
_
S
N
_
T,V
= T
_

N
_
F
T
_
N,V
_
T,V
= T
_

T
_
N,V
(*)
Chain rule (
x
y
)
z
(
y
z
)
x
(
z
x
)
y
= 1
prove by dx = (
x
y
)
z
dy + (
x
z
)
y
dz = 0 which yields (
y
z
)
x
.
_
N((T, ), T, V )
T
_
,V
=
_
N
T
_
,V
+
_
N

_
T,V
_

T
_

=
_
N

_
T,V
_

T
_
N,V
+
_
N

_
T,V

T
=
1
T
_
N

_
T,V
_
T
_

T
_
N,V
_
=
1
T
_
N

_
T,V
_
E
N
_
T,V
from (*)
E
2
= k
B
T
2
C
V
+ k
B
T
_
N

_
T,V
_
_
E
N
_
T,V
_
2
= (E
2
)
can
+ N
2
_
_
E
N
_
T,V
_
2
> (E
2
)
can
21
(E
2
)
1
2
/E 1/

N 0 except at Phase transitions.

Summary: Thermodynamics:
E(S, V, N) dE = TdS PdV + dN
E TS = F(T, V, N) dF = SdT PdV + dN
F N =

(T, V, ) d

= SdT PdV Nd
Statistical Mechanics:
=

r
(E,Nfixed)
1 = e
S(E,V,N)/k
B
MCE
Z =

r
(Nfixed)
e
E
r
= e
F(T,V,N)/k
B
T
CE
L =

r
e
N
r
E
r
= e

(T,V,)/k
B
T
GCE
22
2. QUANTUM STATISTICAL MECHANICS
A system of N particles is described by a wave function (r
1
, r
2
, ...). Since the particles are
indistinguishable, upon exchange of particles r
i
r
j
the wave function acquires a phase e
i
( , r
i
, , r
j
, ) = e
i
( , r
j
, , r
i
, ).
This particle exchange is equivalent to a rotation by in the relative coordinate r
i
r
j
.
Exchanging the particles i, j twice is equivalent to a 2 rotation, which in 3-dimensions is
equivalent to the identity operator [a circle on a sphere can be smoothly deformed into a
point]. Hence e
2i
= 1 and there are two types of particles in nature:
= 0 symmetric for bosons (Bose Einstein statistics (BE))
= antisymmetric for fermions (Fermi Dirac statistics (FD)).
In particular fermions obey paulis exclusion principle [antisymmetric wavefunction at r
i
= r
j
vanishes]. Note also the spin-statistics connection, i.e. integer spin particles are bosons, half
integer particles are fermions. Note also that in 2-dimensions other statistics are allowed;
since the 2 rotation is now a circle with a singular point r
i
= r
j
at its center, the circle
cannot be deformed into an identity operation. In 3-dimension one escapes in the 3rd
dimension avoiding the r
i
= r
j
singularity.
We also dene a Boltzman statistics (MB) where only the particles at the same energy
level are indistinguishable, i.e. for each energy level we add a factor of
1
n

!
.
2a. Ensembles for ideal quantum gases
Micro Canonical Ensemble
We group the dierent states into energy levels

i
with degeneracy of g
i
. Each level is occupied
with n
i
particles. For the statistical treatment
we assume g
i
, n
i
>> 1.
(N, V, E) =

t
n
i

W(n
i
)whereW(n
i
) =

i
W(i) for a given distributionn
i

23
The prime on the sum denotes the following constraints:

i
n
i
= N and

i
n
i

i
= E
For bosons, each energy level can be occupied with n
i
particles within g
i
folders with g
i
1
divisions. Hence we choose n
i
particles from n
i
+ g
i
1 objects, W(i) =
(n
i
+g
i
1)!
n
i
!(g
i
1)!
and for
all energy levels we have
W
B.E.
=

i
(n
i
+ g
i
1)!
n
i
!(g
i
1)!
For fermions we choose n
i
sites from the g
i
available ones, so
W
F.D.
=

i
g
i
!
n
i
!(g
i
n
i
)!
g
i
n
i
>> 1
For the Boltzman statistics any particle can occupy any of the states up to the Gibbs
correction (
1
n
i
!
) for the indistinguishability of the particles in each energy level
W
M.B.
=

i
(g
i
)
n
i
n
i
!
The Gibbs factor correctly accounts for permutations of particles in dieren quantum states,
but also unnecessarily corrects for particles that are in the same quantum state where correc-
tion is not necessary (e.g. one symmetric state only). Therefore n
i
! is an over-correction
and is valid when the density is low n
i
<< g
i
and there is a small chance of all particles
being in the same quantum state. This is also reected in the possible situation g
n
i
i
/n
i
! < 1.
E.g., consider two particles in two states n=g=2:
W
BE
= 3
_
[aa), [bb),
1

2
([ab) +[ba))
_
W
FD
= 1
1

2
([ab) [ba))
W
MB
=
2
2
2!
= 2
1
2
of [aa) or of [bb) is ad hoc.
The entropy of the quantum gas will be
S = k
B
ln
_
_
t

n
i

W(n
i
)
_
_
.
24
We now nd the distribution n

i
that maximizes S. Using the method of lagrange multipliers
we demand

_
ln W(n
i
)

i
n
i

i
n
i

i
_
= 0.
with , to be determined below. Rewrite our 3 cases as
ln (W(n
i
)) =

i
n
i
ln
_
g
i
n
i
a
_

g
i
a
ln
_
1 a
n
i
g
i
_
where a is dened by a = 1 (BE), a = +1 (FD) and a 0 (MB).
Performing the variation yields

i
_
ln
_
g
i
n

i
a
_

i
_
n
i
= 0
so that S is maximized by the distribution
n
i

g
i
=
1
e
+
i
+ a
The value of S at its maximum is therefore
S/k
B
= ln W(n

i
) =

i
n

i
( +
i
) +
g
i
a
ln
_
1 + ae

i
_
The coecients and are determined by the constraints on N and E:

k
B
T
=
1
k
B
_
S
N
_
E,V
= N
_

N
_
E,V
+ + E
_

N
_
E,V
+

g
i
a
ae

i
1 + ae

i
_
_

N
_
E,V
+
i
_

N
_
E,V
_
=
since the last sum is N
_

N
_
E,V
E
_

N
_
E,V
. The coecient is identied by
1
k
B
T
=
1
k
B
_
S
E
_
N,V
= N
_

E
_
N,V
+ E
_

E
_
N,V
+
+

i
g
i
a
ae

i
1 + ae

i
_
_

E
_
N,V
+
i
_

E
_
N,V
_
=
The equation of state is obtained by
1
a

i
g
i
ln(1 + ae

i
) =
S
k
B
+
N E
k
B
T
=
PV
k
B
T
25
so that nally
PV =
_

_
k
B
T

g
i
ln
_
1 e
(
i
)/k
B
T

a = 1
k
B
T

g
i
e

i
= k
B
T

i
= k
B
TN a = 0
_

_
.
Canonical and Grand-Canonical ensembles
In the GCE there are no constraints on N, E. Hence we label by i all quantum numbers
that completely specify a state so that g
i
= 1 and
W
BE
n
i
= 1
W
FD
n
i
= 1 (if all n
i
= 0,1) or 0 otherwise
W
MB
n
i
=

i
1
n
i
!
notice that W
FD
< W
MB
< W
BE
. E.g. with N = 2 particles in two states
BE: [aa), [bb),
1

2
([ab) +[ba)) 3 states
FD:
1

2
([ab) [ba)) 1 state
MB:
1
2
[aa),
1
2
[bb), [ab) 2 states.
Consider briey the CE constrained with

i
n
i
= N,
Z
N
=

t
n
i

W(n
i
)e

i
n
i

i
and for the MB distribution
Z
N
=

t
n
i

i
_
N!
n
i
!
e

i
n
i
_
1
N!
= using multinomial expansion =
_

i
e

i
_
N
1
N!
=
1
N!
(Z
1
(V, T))
N
[Note: with
i
0 we have

n
i
W
MB
n
i
= N
N
/N! for N particles in N states]. For the
BE and FD statistics one has to proceed to the GCE, to avoid the

n
i
= N constraint.
L(, V, T) =

N=0

N
Z
N
(V.T) =

N=0

t
n
i

i
_
e

i
_
n
i
with = e

here the rst sum determines the constraint N on the second sum. However, since anyway
we some on all occupations, we can sum them on each level independently,
L =

n
0
,n
1
,...
_
e

0
_
n
0
_
e

1
_
n
1
=

i
_

n
i
_
e

i
_
n
i
_
26
now using

n
x
n
=
1
1x
we get the result for each of the distributions
L =
_

i
1
1e

i
BE

i
_
1 + e

i
_
FD
ln L = Z
1
=

i
e

i
MB using

x
n
n!
= e
x
_

_
.
The GCE relation is
ln L =
PV
k
B
T
=
1
a

i
ln
_
1 + ae

i
_
and recall a = 1 for BE, a = 1 for FD and a 0 for MB. For N, E we have :
N =
_
(ln L)

_
V,T
=

i
1
1

i
+ a
E =
_
ln L

_
,V
=

i
1

i
+ a
The mean occupation number is given by
n
i
) =
1
L

n
j

n
i

N
e

j
n
j

j
=
1

_
ln L

i
_
,T,j,=i
=
1
1

i
+ a
=
1
e
(
i
)
+ a
The gure illustrates the various cases. For FD, if > and low temperatures the mean
occupation tends to 1. For the BE case, the condition < must be valid for all to
27
maintain n
i
) 0; this leads (see below) to BE condensation. The MB distribution is valid
only when n
i
) << 1 which means

i

k
B
T
>> 1 for all
i
, hence < 0 and [[ k
B
T or 1.
The condition on the density is N

=

i
e

i
=
V

3
which implies high temperature or
low density limit. Note that both BE and FD approach MB at this limit of 1.
2b. Ideal Bose gas
Here
i
= p
2
/2m and the summation index is i = p.
A note on momentum summations: The sum

p
actually stands for a sum on integers
n
x
, n
y
, n
z
that dene p
x
, p
y
, p
z
via periodic boundary conditions. E.g. e
ip
x
L
x
/
= 1 with L
x
the length in the x direction, p
x
=
2
L
x
n
x
, hence for any function f(p)

n
x
,n
y
,n
z
f(p) =
_
L
x
L
y
L
z
h
3
f(p)d
3
p =
V
h
3
_
f(p)d
3
p
In radial coordinates we then have
P
k
B
T
=
4
h
3
_

0
p
2
ln(1 e
p
2
/2m
)dp
1
V
ln(1 )
N
V
=
4
h
3
_

0
p
2
_
1

e
p
2
/2m
1
_
1
dp +
1
V

1
The p = 0 term in the original sum is singled out. It is the number of particles N
0
=

1
in
the single state p = 0. [Equivalently, need e

i
< 1 to allow convergence for L.]
The crucial property of bosons is 1 so as to keep all state occupations n
p
) 0. For
a given density the integral term decreases with T for a xed ; to keep N/V xed must
increase, but it is bound by = 1. The integral is bound at = 1, hence below some T
c
the
integral becomes < N/V and a macroscopic part of N must populate the p = 0 state. This
is Bose Einstein condensation.
The correction in the pressure equation is negligible:
1 =
1
1+ < N
0
>

1
V
ln(1 ) =
1
V
ln(< N
0
> +1)
always
0
The boson thermodynamics are then given by
P
k
B
T
=
1

3
g
5/2
()
28
N
V
=
1

3
g
3/2
() +
< N
0
>
V
g
n
() =
1
(n)
_

0
x
n1
dx
1

e
x
1
=

=1

n
For small we use the expansion to eliminate :

PV
NkT
=

=1
a

(n
3
)
1
(a

= virial coecient)
a
1
= 1, a
2
=
1
4

2
= 0.176, a
3
= 0.0033 .....
For specic heat:
E = (

ln L)
,V
= k
B
T
2
_

T
V g
5/2
()

3
_
,V
=
3
2
k
B
T
PV
k
B
T
=
3
2
PV
since the virial expansion for PV has T(
3
)
1
T
5/23/2
terms.
C
V
Nk
=
1
Nk
_
E
T
_
N,V
=
3
2
_

T
PV
Nk
_
N,V
=
3
2

=1
5 3
2
a

(n
3
)
1
=
3
2
(1 + 0.0884n
3
+ 0.0066(n
3
)
2
+ ...)
C
V
= T
_
S
T
_
N,V
T0
0
hence C
V
looks roughly as:
T
Cv
If n
3
< g
3/2
(1) = (
3
2
) = 2.612
0 < < 1 and
< N
0
>
V
= 0
If n
3
> g
3/2
(1)
<N
0
>
V
= n
g
3
2
(1)

3
> 0
1

2.612
g 3

29
with = 1 and nite density at the single
= 0 level.
Transition at n
3
= g3
2
(1) kT
c
=
2
2
m[g
3/2
(1)]
2
3
n
2/3
When = 1, < N
1
> is negligible since
1
h
2
/V
2/3
< N
1
>
V
=
1
V
1
e

1
1

1
V
V
2/3
h
2
V
0
At T T
c
,
<N
0
>
N
= 1
g
3/2
(1)

3
n
= 1
_
T
T
c
(n)
_
3/2
= 1
n
c
(T)
n
1
TT
c
1
N
0
N
Two uid concept: < N
0
> condensate,
g
3
2
(1)

3
normal component.
P
kT
=
_

_
1

3
g
5/2
() n < n
c
(normal)
1

3
g
5/2
(1) n > n
c
(condensed phase)
P
c
T
5
2
n
5
3
all excess particles beyond n
c
occupy p = 0,
< N
0
>= N N
c
(T), even at n with
no contribution to P. Along AB
P
n
= 0, two
phases coexist with phase A that has n .
1n
A
P
B
P
c
n
5

3
30
This corresponds to a rst order phase tran-
sition, 1/n jumps by (1/n) = 1/n
c
as pres-
sures increases.
T
P
No
realization
Normal
Condensate on line
P
c
T
5

2
Note that there is no realization of P > P
c
(T). The equation of state denes a 2-dimensional
surface in P, T, n; the gure is a projection of this surface on the P, T plane. There are no
points on the surface that project onto the region P > P
c
(T). Consider a xed n and an
initial high temperature where P = nk
B
T and P P
c
(T). Now as T decreases P approaches
P
c
(T) at T
c
, and upon further decrease of T, P stays on the critical curve P(T) = P
c
(T) for
all T < T
c
. This feature is an artifact of the ideal gas; once hard core repulsion is added, at
suciently high density (in a constant T trajectory) the pressure will start to increase.
Clausius Clapeyron:
dP
c
dT
=
S/N
(
1
n
)
S is the jump in entropy. We will evaluate S directly:
S
Nk
B
=
PV
Nk
B
T
+
E
Nk
B
T


kT
=
5
2
PV
Nk
B
T


k
B
T
=
_

_
5
2
g
5/2
()
n
3
ln T > T
c
5
2
g
5/2
(1)
n
3
=
5
2
g
5/2
(1)
g
3/2
(1)
N<N
0
>
N
T T
c
At T < T
c
, S N < N
0
>= N
norm
, no entropy for the condensate. Above the P
c
(T)
line S/N 0, while just below this line
N =
V

3
g
3/2
(1)
S
N
=
5k
B
2
g
5/2
(1)
g
3/2
(1)
, (1/n) = 1/n
c
=
3
/g
3/2
(1)
Consider now the derivative
dP
c
dT
=
5
2
k
g
5/2
(1)

3
obtained from P
c
=
kT

3
g
5/2
(1)
This proves the Clausius Clapeyron relation.
31
Note that S implies a Latent heat TS 1st order transition.
Specic heat (T < T
c
):
C
v
Nk
=
3V
2N
g
5/2
(1)
d
dT
(
T

3
) =
15
4
(
5
2
)
n
3
T
3/2
At T > T
c
,
dC
v
dT

dg
3/2
()
d
=
g
1/2
()

T
c
T
1.5
1.925
C
v
Nk
The shape of C
V
(T) is similar to the transition of
4
He, which looks like the letter .
For
4
He by this theory T
c
= 3.13

K, by experimental results T
c
= 2.19

K.
Note: In a real superuid there is a nite critical velocity v
c
below which there is no dissi-
pation to the ow. This implies that at momentum k and frequency the current states
have excitations = vk (note the transformation to the moving frame e
vtd/dx
) which cannot
decay into the excitations (k) of the system at rest, hence v
c
= min(k)/k. The ideal
gas is therefore not a real superuid since excitation energies are
k
=
2
k
2
/2m, so that
v
c
= 0 at k 0.
With repulsive weak interaction g
one has at k 0
k
= k
_
Ng
V m
_
1/2
,
and critical current is nite either
from this limit or from a roton
minimum (see gure) so that
v
c
=
0
/k
0
.
k
0
k

k
Black body radiation
Photons in thermal equilibrium correspond to bosons with = 0 since there is no conserva-
tion law for photons, hence N minimizes F, i.e. F/N = 0 = .
Photon spectrum is
k
= ck with [k[ = k. Photons have two polarization states, = 1.
Z =

n
k
,
e

n,

k
n
k,
=

k,

n=0
e

k
n
=

k
2
1 e

k
32
ln Z = 2

k
ln(1 e

k
), same as the GCE L with = 0. (Note F =

when = 0).
n
k
) =
1

(
k
)
ln Z =
2
exp (
k
) 1
n() is the number of photons with the frequency .

k
n
k
) =
V
(2)
3
_
4k
2
n
k
)dk =
4V
(2)
3
_
(

c
)
2
d(

c
)n
k
)
_

0
n()d
u() = n() =

3
exp () 1

2
c
3
This is the Plancks formula for energy density.
The total energy
E
V
=
_

0
n()d =

2
k
4
B
15
3
c
3
T
4
c
V
T
3
is unbounded as can con-
tribute.
R is the rate of radiation through a small hole in a cavity. We need to average only velocities
with v
z
> 0
v
z
)
v
z
>0
= c
_
/2
0
cos d(cos )/
_

0
d(cos ) =
c
4
R =
E
V
c
4
=

2
k
4
B
60
3
c
2
T
4
T
4
where is Stefans constant (1879) = 5.670 10
8
Watt/(meter)
2
(Kelvin)
4
.
We use periodic boundary quantization where n has integer entries,

k
= c[

k[ =
c2[n[
V
1/3


k
V
=
1
3

k
V
Therefore
P =
1

V
ln Z =
1
3V

k

k
n
k
) PV =
1
3
E P =
4
3c
T
4
(for nonreletavistic bosons
k
V
2/3

k
V
=
2
3

k
V
PV =
2
3
E).
Note that (
P
V
)
T
= 0 n
2
)
If the photon had a mass m ,then at k
B
T mc
2
Stefans constant would change by a
factor 3/2, inconsistent with experiment photon mass = 0 (i.e. less than the experimental
k
B
T/c
2
.)
Phonons
33
Consider N atoms in a lattice that have 3N vibration modes, labeled by phonons with
wavevector

k and 3 polarizations. In the Debye model: = c[

k[, <
m
3N =
k
3 = 3
V
(2)
3
_
4k
2
dk
_

m
0
f()d, f() =
3
2
2
2
c
3
V
Integrating f() yields the cuto frequency
m
= c(6
2
n)
1/3
; the Debye temprature is T
D
:
k
B
T
D
=
m
.
Z =

n
i
e
(

i
n
i
)
=
3N

i=1
1
1 e

i
where i =

k, polarization.
n
i
) =
1

(
i
)
ln Z =
1
exp (
i
) 1
E =

ln Z =

i
n
i
) =
_

m
0

f()
exp () 1
d =
E =
_

_
3Nk
B
T(1
3
8
T
D
T
+ ...) T T
D
3Nk
B
T

4
5
(
T
T
D
)
3
+ O(exp (
T
D
T
)) T T
D
Experiments on specic heat conrm that noninteracting phonons are normal modes of solids
at low T.
2c. Ideal Fermi gas
PV
kT
= ln L =

p,spin
ln (1 + exp (p
2
/2m))
Where = exp () and
n
p
) =
1
1

e
p
2
/2m
+ 1
so that N =

p,spin
n
p
).
34
P
k
B
T
=
4
h
3
g
_

0
dpp
2
ln (1 + exp (p
2
/2m)) =
g

3
f
5/2
()
where g = 2s + 1 is the number of spin states
n =
4

3
g
_

0
dpp
2
1
1

e
p
2
/2m
+ 1)
=
g

3
f
3/2
()
f
n
() =
1
(n)
_

0
x
n1
1

exp (x) + 1
dx
=

l=1
()
l+1

l
l
n
0 < < (unlike bosons!)
For n
3
1 or 1:
n
3
/g =
2
/2
3/2
+ ..., [n
p
)
1
g
n
3
exp (
p
) is MB form]
P
nk
B
T
=
g
n
3
(
2
/2
5/2
+ ...) = 1 +
1
g2
5/2
n
3
+ ...
This is the virial expansion. Consider next n
3
1
1
g
n
3
=
4
3

[(ln )
3/2
+

2
8
(ln )
1/2
+ ...] + O(1/)
Where e

F
, so that by comparing T 0 terms

F
=

2
2m
(
6
2
n
g
)
2/3
To conrm the T = 0 result n
p
)
1
exp ((
p

F
))1
so that all the states
with
p
<
F
are occupied. Therefore
N = g
V
h
3
_

p
<
F
d
3
p =
gV
(2)
3
4
3
p
3
F

F
= p
2
F
/2m
conrming the result
F
=

2
2m
(
6
2
n
g
)
2/3
.
Expansion at k
B
T
F
:
=
F
[1

2
12
(
k
B
T
E
F
)
2
+ ...]
35
E =
3
5
N
F
[1 +
5
12
(
k
B
T
E
F
)
2
+ ...] =
3
2
PV.
The rst term is g

[p[<p
F
p
2
/2m. The second term corresponds to a fraction k
B
T/
F
of
excited particles with energy k
B
T, hence excess energy is T
2
and c
V
T.
At T=0 the pressure is due to occupied states that have p ,= 0.
2d. Non- Ideal gases
Atoms have on average dipole)
time
= 0 (time average is also equivalent to ensemble average).
However, the dipole moment of one atom induces a dipole on a neighboring atom such that
dipole dipole)
time
,= 0. This energy is
1
r
6
.
Lennard - Jones potential:
U = [(
b
0
r
)
12
2(
b
0
r
)
6
]
b
0
2 5

A,
(1 40) 10
22
Joule
(1 40) 0.1

K
0

b
0 r
U
Virial expansion:
Z =
1
N!h
3N
_
d
3N
pd
3N
re
[

i
p
2
i
/2m+

i<j
U(r
i
r
j
)]
= Z
0
1
V
N
_
d
3N
re

i<j
U(r
i
r
j
)
= Z
0
1
V
N
_
d
3N
r[(e

i<j
U(r
i
r
j
)
1) + 1]
where Z
0
is the non interacting partition function.
36
For low density, each pair contributes independently:
Z = Z
0
[
N(N1)
2
_
(
d
3
r
1
d
3
r
2
V
2
e
U(r
1
r
2
)
1) + 1]
0
1
1e
U(r)
r
F = F
0

1
2
k
B
T
N
2
V
2
_
(e
U(r
1
r
2
)
1)d
3
r
1
d
3
r
2
B(T)
1
2
_
(1 e
U(r)
)d
3
r
F = F
0
+ k
B
T
N
2
V
B(T)
P =
F
V
=
Nk
B
T
V
(1 +
N
V
B(T))
assuming convergence (U ,=
1
r
).
B
>0
..
_
b
0
0
... +
<0
..
_

b
0
... = b a/T
P =
Nk
B
T
V
(1 +
N
V
b) a(
N
V
)
2

Nk
B
T
V bN
a(
N
V
)
2
Van der Waals equation
where V bN is the excluded volume and the second term, a(
N
V
)
2
, is the pressure reduction
from the attractive part of the interaction.
GCE formulation
Low density low fugacity
L =

N
Z
N
(V, T) = 1 + Z
1
+
2
Z
2
+ ...
for both classical and quantum theories dene partition functions for N = 1, 2 particles:
ln L = Z
1
+
2
(Z
2

1
2
Z
2
1
) + ... =
PV
k
B
T
N =

(ln(L)) = Z
1
+ 2
2
(Z
2

1
2
Z
2
1
)
eliminate
PV
k
B
T
= N
_
N
Z
1
_
2
(Z
2

1
2
Z
2
1
) N
_
1 + B
N
V
_
B = V
1
Z
2
1
(Z
2

1
2
Z
2
1
)
37
Evaluate now Z
1
, Z
2
: For Z
1
there are no eects of interactions or of quantum statistics, i.e.
Z
1
=

p
e
p
2
/2m
=
V

3
=
h

2mk
B
T
Consider next Z
2
= Tr e
1
for N = 2, rst the classical problem:
Z
2
=
1
2!
__
d
3
pe
p
2
/2m
_
2

1
h
6
_
d
3
r
1
d
3
r
2
e
U(r
1
r
2
)
=
V
2
6

_
d
3
re
U(r)
From the expression above for B we obtain B(T) =
1
2
_
(e
U(r
1
r
2
)
1)d
3
r
1
d
3
r
2
as in the
previous derivation.
Consider next the quantum problem with the 2-particle Hamiltonian
H =

2
2m
_

2
r
1
+
2
r
2
_
+ U(r
1
r
2
) =

2
4m

2
R

2
m

2
r
+ U(r)
where r = r
1
r
2
, R =
1
2
(r
1
+r
2
). The eigenvalues of H are
P
2
4m
+E
n
where P is the center
of mass momentum, hence
Z
2
=

P
e
P
2
/4m

n
e
E
n
=
2
3/2

3
V

n
e
E
n
Note that E
n
contain information on statistics, i.e. E
n
is restricted to eigenfunctions that
are symmetric in r for bosons, or to antisymmetric ones for fermions.
38
3. PHASE TRANSITIONS
Order Parameter Example T
c
(

K)
Liquid-gas density H
2
O 647
Ferromagnetic magnetization Fe 1044
Antiferromagnetic sublattice magnetization FeF
2
78
Bose condensation superuid amplitude
4
He 2
Superconductivity electron pair amplitude Y Ba
2
Ca
3
O
7
90
Binary uid concentration of one uid CCl
4
C
7
F
14
302
Binary alloy density of one kind on a sublattice Cu Zn 739
Ferroelectric polarization Triglycine sulfate 322
Ferroelastic q = 0 distortion Ni Ti
charge density wave q ,= 0 distortion NbSe
3
59
Metal-Insulator
Percolation fraction of sites in percolating cluster
Denition: n-th order phase transition has a discontinuity in the n-th derivative of a free
energy. Hence a 1st order transition has a jump in entropy (F/T) which is the latent
heat. A 2nd order transition has a jump in C
v
(contains
2
F/T
2
).
3a. First order transitions
T
n
nl
ng
mid point
(assymetry)
T
M
coexisting domains
Tc
Innitesimal change in P (liquid-gas on left gures) or in H (magnetic eld in a ferromagnet
39
on right gures) leads to a jump in density n (on left) or in magnetization (on right). In
P T plane (left) or H T plane (right) rst order transition ends at a critical point T
c
.
T
P

liquid
gas
solid
critical point
triple point
T
H

Consider the liquid-gas phase transition as a prototype of a 1st order transition

Liquid-gas: 1st transition, no symmetry change.
Solid-liquid: 1st order transition, change in translation symmetry.
Fixing V below T
c
leads to liquid/gas coexistence.
Fixing P leads to a jump in V from fully gas phase to fully liquid phase.
40
Clapeyrons equation
g(P, T) =
1
N
G(P, T, N)
where G(P, T, N) is the Gibbs free energy.
g(P, T) = g
2
(P, T) g
1
(P, T)
where g
i
(T, P) is formally continued across T
c
.
Dene s = S/N, v = V/N
s = s
2
s
1
; v = v
2
v
1
_
g
T
_
P
= s;
_
g
P
_
T
= v
41
By the chain rule
_
g
T
_
P
_
T
P
_
g
_
P
g
_
T
= 1
we get
(g/T)
P
(g/P)
T
=
_
P
T
_
g
=
s
v
Along the transition line P(T) we get g = 0 =const.
dP
dT
=
s
v
=
latent heat
Tv
.
Van-der Waals equation
A rough argument:
V
eff
= V b
P = P
kin

a
V
2
where b N is the excluded volume and a N
2
is a measure of attractive forces among
the molecules of the system; a N
2
since there are N
2
/V
2
pairs. Hence
P
kin
V
eff
= (V b)
_
P +
a
V
2
_
= NK
B
T
At T > T
c
we get one solution for
the equation, and for T < T
c
we get
three.
The three roots merge to one
root at inection point P
c
, V
c
, T
c
The parameters a, b are sample specic. To identify them in term of P
c
, V
c
, and T
c
rewrite
the equation in the form
(V V
c
)
3
= 0 =
_
(V b)
_
P
c
+
a
V
2
_
Nk
B
T
c
_
V
2
P
c
By identifying each power of V in this form we write a, b in terms of P
c
, V
c
, T
c
, and with

P = P/P
c
,

T = T/T
c
,

V = V/V
c
we obtain
_

P +
3

V
2
__

V
1
3
_
=
8
3

T
42
which is the law of corresponding states all substances have the same equation of state
when their P, V, T are measured in units of the critical point.
Note the region with
P
V
> 0 is unstable so that

(N)
2
_
0.
Alternative derivation (S.K.Ma p.470)
G
0
(T, P, N) = E TS + PV
where
E =
3
2
NK
B
T a
1
N
N
V
attraction from neighboring N/V atoms, and
S = k
B
ln
_
1
N!
(V b
1
N)
N

3N
_
where b
1
N is the excluded volume and is the thermal wavelength.
Proper G(T, P, N) is obtained at minimum with respect to V which is a redundant variable
for G.
G
0
(T, P, N; V ) =
3
2
NK
B
T a
1
N
2
V
k
B
T
_
N ln
_
V b
1
N
N
3
_
+ N
_
+ PV
If
G
0
V
= 0, we get the Van der Waals equation with a = a
1
N
2
and b = b
1
N .
G
0
shows (see gure) that with 3 solutions 1 is metastable and 1 is unstable (max of G
0
)
For P
1
< P < P
2
gas is stable, liquid is metastable.
For P
2
< P < P
3
liquid is stable, gas is metastable.
43
P
2
is when gas and liquid are degenerate G
0
(V
1
) = G
0
(V
2
).
Maxwells construction
In general G
0
= F(V, T, N) + PV where P(V ) = F/V . At P
2
:
0 = G
0
(V
2
) G
0
(V
1
) =
_
V
2
V
1
[P(V ) + P]dV .
Hence the area between P = P
2
line and the P(V ) curve vanishes this is Maxwells
construction.
An alternative derivation: In terms of F(V ) choose V
1
and V
2
such that parallel tangents
generate one line with
F
V
1
=
F
V
2
.
The line is
F
2
F
1
V
2
V
1
=
F
V
1

P
2
(V
2
V
1
) = (F
2
F
1
) =
_
V
2
V
1
P(V )dV
which is the same Maxwells construction. The line is
realized by coexistence of the two phases with fractions
x and 1 x, respectively (0 < x < 1),
V = xV
1
+ (1 x)V
2
F
line
= xF
1
+(1x)F
2
=
(V V
2
)F
1
+ (V
1
V )F
2
V
1
V
2
< F
Phase separation at P
2
since V
1
< V < V
2
has lower
F
line
then F on the continuous curve.
When V is changed carefully to avoid strong uctua-
tion, metastable phases can persist until they become
unstable; this leads to hysteresis.
44
3b. Second order phase transitions Mean Field theory
Consider ferromagnetism as a prototype second order phase transition. Localized indepen-
dent moments in a magnetic eld B lead to magnetization
M = N
e
B
e
B
e
B
+ e
B
= Ntanh (B) .
Mean eld theory for interactions: neighbors with magnetization M/V induce an addi-
tional eld, B B +
M
V
, interaction strength.
= M = Ntanh
_

_
B +
M
V
__
has M ,= 0 solutions even if B = 0.
If T = T
c
, 1 =
N
V

2
/k
B
T
c
Expand near T
c
:
M =
T
c
T
M
1
3
_

M
V
_
3
N = M (T
c
T)
1/2
.
Susceptibility at T > T
c
: B 0, M 0
M = N
_
B +
M
V
_
=
_
M
B
_
B=0
=
N
2
1 N
2
/V

1
T T
c
.
Microscopic Model
Consider two neighbors of spin 1/2, total spin is S = 0, 1 with energy dierence which denes
the exchange energy J. The requirement for an antisymmetric wavefunction requires a
symmetric orbital for the singlet S = 0 and antisymmetric one for the triplet S = 1, hence
a large energy dierence from the dierence Coulomb interactions. Hence J is much larger
45
then that of interacting magnetic dipoles. Consider
s
i
s
j
=
1
2
_
(s
i
+s
j
)
2
s
2
i
s
2
j
_
=
1
2
S(S + 1)
1
2

3
2
=
_
_
_
3/4 S = 0
1/4 S = 1
.
= Interaction energy = J

t
n.n.
s
i
s
j
+const (n.n. is summation on nearest neighbors).

t
means that each pair is summed once.
If the crystal has an easy axis for the spin, s
i
(s
z
)
i
, hence two models:
H = J
t

i,j)
s
i
s
j
s spin operator: Heisenberg model
H = J
t

i,j)

j

i
= 1: Ising model.
Solution of the Ising model by mean eld:
H
MF
=
1
2
J )

i
no. of nearest neighbors.
1
2
is needed so that

H
MF
_
=
1
2
NJ )
2
,
1
2
N no. of bonds. (e.g. N/2 odd sites each
generates bonds.)
) =
i
) =

i
=

i
e
J)
i
/2

i
=
e
J)
i
/2

j=i

e
J)

j=i

j
/2

j=i

e
J)

j=i

j
/2
= tanh
_
J
1
2
)
_
= kT
c
=
1
2
J.
3c. Exact results
Mapping between systems
1. Lattice gas model to Ising
N
a
atoms occupy sites in lattice with N cells; N
aa
number of nearest neighbors, each with
energy
0
, g(N
a
, N
aa
) is the number of congurations with a given N
a
, N
aa
(a non-trivial
function). Grand partition
L
G
=

N
a

N
a

N
aa
g (N
a
, N
aa
) e

0
N
aa
.
46
Canonical partition of Ising has the same form:
N
+
no. of + spins, N
++
no. of ++ neighbors. Draw one line from all + sites to all their
neighbors:
Number of lines = N
+
= 2N
++
+ N
+
same for N

= 2N

+ N
+
, N
+
+ N

= N
=

i,j)

j
= N
++
+ N

N
+
= 4N
++
2N
+
+
1
2
N

i
= N
+
N

= 2N
+
N
E
Ising
= 4JN
++
+ 2(J B)N
+

_
1
2
J B
_
N
Z = e
N

1
2
JB

N
+
e
2(JB)N
+

N
++
g (N
+
, N
++
) e
4JN
++
Since conguration counting is the same as in lattice gas we get the same function g(N
a
, N
aa
),
hence the two problems are equivalent as in the following table:
Ising Lattice gas
N
+
N
a
4J
0
e
2(JB)

_
1
N
F
Ising
+
1
2
J B
_
P
M = N
+
N

,
N
+
N
=
1
2
_
M
N
+ 1
_
1
v
=
N
a
)
N
order parameter
2. Binary alloy to lattice gas
N
11
, N
22
, N
12
no. of pairs of each type
E
A
=
1
N
11
+
2
N
22
+
12
N
12
As above
N
1
= 2N
11
+ N
12
N
2
= 2N
22
+ N
12
=
N
12
= N
1
2N
11
N
22
=
1
2
N + N
11
N
1
47
N
1
+ N
2
= N
E
A
= (
1
+
2
2
12
) N
11
+
_
(
12

2
) N
1
+
1
2

2
N

, Z (N
1
, T) =

N
11
g (N
11
) e
E
A
Lattice gas Binary alloy
N
a
N
1

0

1
+
2
2
12
F F + (
12

2
) N
1
+
1
2

2
N
Ising Model in 1D: Exact Solution
Z =

1=

2=
...

N=
e
J
N

k=1

k+1
+
1
2
B

k
(
k
+
k+1
)
;
1
=
N+1
Consider the bond (k, k + 1) with the elements
e
J
k

k+1
+
1
2
B(
k
+
k+1
)
This element can have 4 values for
k
= and
k+1
= . Call these elements
P
1,1
, P
1,1
, P
1.1
, P
1,1
and dene a matrix

P =
_
_
P
1,1
P
1,1
P
1,1
P
1,1
_
_
=
_
_
e
(J+B)
e
J
e
J
e
(JB)
_
_
Note that

P is symmetric by the choice of
k
+
k+1
for the B term in the Hamiltonian.

P is dened in a spinor space [), e.g. +[

P[) = P
1,1
.
E.g. for N = 3 the partition Z has 2
3
terms, one of them is
= P
1,1
P
1,1
P
1,1
= +[

P[+)+[

P[)[

P[+)
In general we need
Z =

all [i)=[)
1[

P[2) . . . k[

P[k + 1)k + 1[

P[k + 2) . . . N[

P[N + 1)
=

1
=
N
=
(P
N
)

1
,
N
= Tr(P
N
) =
N
1
+
N
2
48

1
,
2
are the eigenvalues of

P:
2
2e
J
cosh (B) + 2 sinh (2J) = 0
=
1,2
= e
J
cosh (B) [e
2J
+ e
2J
sinh
2
(B)]
1
2

2
<
1
, N ,
1
dominates, hence
1
N
ln Z ln
1
and is an analytic function (except
at T = 0). For the magnetization we have
M =
1

B
ln Z = N
sinh (B)
[e
4J
+sinh
2
(J)]
1
2
= M(B = 0, T > 0) = 0, M(T = 0) = Nsign(B), hence an ordered phase is only at
T = 0. The susceptibilty:
=
M
B
[
B=0
=
N
2
k
B
T
e
2J/k
B
T
[
T0
indicates ordering at T = 0.
No phase tradition in 1-dimension general argument:
Low energy excitations are walls [, energy is 2J (relative to the ground state), while
entropy is S = k
B
ln N since the wall can be positioned at any bond.
= F = E TS = 2J k
B
T ln N < 0 no long range order in 1-dimension at any nite
temperature.
2D Ising model: Onsagers solution (1944)
KT
c
= 2.269J
M (T
c
T)
1/8
near T
c
C
v
ln (1
T
T
c
)
3d. Landaus Theory for second order phase transitions
Partition sum is done rst locally to dene a coarse grained order parameter, slowly
varying. FM(r) is determined by symmetry for small M(r), i.e near a 2nd order phase
transition. Coarse graining involves a nite cluster FM(r) has an analytic expansion.
For ferromagnet, M M, symmetry+analiticity
FM(r) =
1
2
a(T)M
2
(T) +
1
4
bM
4
(T) +
1
2
c(

M)
2
where a(T) = a
t
(T T
c
) so that in Mean Field the transition occur at T
c
.
49
Mean Field: M(r) = M is homogenous, no uctuations.
F
M
= 0 M =
_
a

b
(T
c
T)
F =
_
_
_

1
4
a
2
b
(T
c
T)
2
T < T
c
0 T > T
c
S =
F
T
continuous (S < 0 at T < T
c
, needs correction by uctuations, see below.)
c
V
= T
S
T
has a jump, and is (T
c
T)
0
at T < T
c
.
Consider next B ,= 0 and add M B to F:
T > T
c
, a
t
(T T
c
)M B = 0 =
M
B
[
0

1
TT
c
T = T
c
, bM
3
B = 0 M B
1/3
Consider uctuations at T > T
c
: neglect M
4
, M(r) = V
1/2

k
M
k
e
ikr
_
FM(r)d
3
r =
1
2

k
(a + ck
2
)[M
k
[
2
since
_
e
ikr+ik

r
d
3
r = V
k,k
. Note that M
k
= M

k
for a real M(r).
The weight of k modes is e
(a+ck
2
)[M
k
[
2
. The correlation function is then
< M(r)M(0) >= V
1

k
< M
k
M
k
> e
ikr
= V
1

k
e
ikr

d(ReM
k
)d(ImM
k
)[M
k
[
2
e
(a+ck
2
)|M

k
|
2

d(ReM
k
)d(ImM
k
)e
(a+ck
2
)|M
k

|
2
= V
1

k
e
ikr
_
d([M
k
[
2
)[M
k
[
2
e
(a+ck
2
)[M
k
[
2
_
d([M
k
[
2
)e
(a+ck
2
)[M
k
[
2
= V
1

k
e
ikr

((a + ck
2
))
ln
_

0
dxe
(a+ck
2
)x
< M(r)M(0) >= V
1

k
e
ikr
(a + ck
2
)
=
k
B
T
4rc
e
r/
where =
_
c
a
(T T
c
)

1
2
is the correlation length. Consider next the free energy
Z =

k
_
d[M
k
[
2
e
(a+ck
2
)[M
k
[
2

k
[(a + ck
2
)]
1
50
F = k
B
T[

k
ln (a + ck
2
) + const.]
Therefore
C
V
= T

2
F
T
2

_
1/a
0
d
d
k
(a + ck
2
)
2
+ less singular terms.
a
0
is a lattice constant, d=dimensions. Using k
t
= k
C
V

1
a
2
_
a
0 d
d
k
t
(1 + k
t2
)
2

d

4d
(T T
c
)
d4
2
Validity of mean eld:
M
2
(0)) << M)
2
; evaluate at T > Tc (similar result at T < T
c
) :
M
2
(0))
_
d
d
k
(a+ck
2
)

_
k
d1
dk
(a+ck
2
)
(

a)
d2
[T T
c
[
(d2)/2
Comparing with M)
2
(T
c
T) shows that mean eld is valid at T T
c
if d > 4.
At d < 4 mean eld breaks down at T T
c
Critical exponents:
t =
TT
c
T
c
, T > Tc; for exponents
t

t
t =
T
c
T
T
c
, T < Tc
Exponent Denition Mean Field Fe
,
t
Specic heat c
V
t

2 d/2 0.12
Order parameter t

1/2 0.34
,
t
Susceptibility x t

1 1.33
Order parameters at T
c
B
1/
3 4.2 (Ni)
,
t
correlation length t

1/2
at T
c
M(r)M(0)) 1/r
d2+
0 0.07
Universality: exponents depend only on symmetry of order parameter and dimensionality.
51
Scaling assumption:
is the single length responsible for singular behavior.
obtain relations between exponents.
e.g. M(r)M(0))
1
r
d2+
g(r/) at T
c
need g g(0) ,= 0
In terms of
_
M
, integration on all congurations M(r), and the Hamiltonian H we have for
the susceptibility
=
M)
B
[
B0
=

B

M
M(0)e
(H

M(r)Bd
d
r)

M
e
(H

M(r)Bd
d
r)
[
B=0
=

M(0)M(r)d
d
re
H

M
e
H
=
_
M(0)M(r)) d
d
r
since at T > T
c
the M)
B=0
= 0 term vanishes (from /B of denominator.)

_
r
d+2
g(r/)d
d
r
d+2

d
(t

)
2
= (2 )
Renormalization Group:
Consider the 1-dimensional Ising model Z =

k
=
e
J

k

k

k+1
Sum on all even sites to obtain a new form for Z

2
e
(J
1

2
+ J
2

3
)
= e
(a

+ J

3
)
Identify a
t
, J
t
by two cases:

3
= + case: e
2J
+ e
2J
= e
(a

+J

3
= case: 2 = e
(a

)
Z(k)

odd i
i
=
e
J

odd k

k+2
with the renormalized coupling J
t
given by e
2J

=
2
e
2J
+e
2J
At T
c
H is scale invariant, i.e. J
t
= J T
c
= 0.
52
4. NON- EQUILIBRIUM
4a. Kinetic theory and Boltzmanns equation:
f(r, v, t) is particle density in phase space d
3
rd
3
v. In absence of collisions
r, v r + vt , v +

F
m
t
so that volume A moves to an innitesimally close volume B with f changed only by collisions,
f( r +vt, v +

F
m
t , t + t) f( r, v, t) (
f
t
)
coll
t
Eect of collisions: If one initial particle is in A the result is certainly outside A (up to the
dierential volume of A) this is a loss term; conversely with one nal particle in B,
Rtd
3
rd
3
v = no. of collisions with one initial particle in A

Rtd
3
rd
3
v = no. of collisions with one nal particle in B [= A + O(t)]
(
f
t
)
coll
t = (

R R)t
Assume binary collisions (dilute gas) v
1
, v
2
v
1
t
, v
2
t
v
1
+ v
2
= v
1
t
+ v
2
t
v
1
2
+ v
2
2
= v
1
t 2
+ v
2
t 2
Dene

V =
1
2
( v
1
+ v
2
) , u = v
2
v
1
;

V
t
=
1
2
( v
1
t
+ v
2
t
) , u
t
= v
2
t
v
1
t

V =

V
t
, [u[ = [u
t
[
A collision is dened by an angle in the center of mass, i.e. between u, u
t
.

V , u, are
independent parameters which determine

V
t
, u
t
.
Consider u u + du , as is xed (i.e. d is independent of d
3
u).
For u
t
u
t
+ du
t
the triangle (u, u + du) is similar to (u
t
, u
t
+ du
t
), hence [du
t
[ = [du[
Choosing an orthogonal basis set du
1
, du
2
, du
3
it transforms to a rotated orthogonal basis
with axes equal in magnitude, hence d
3
u = d
3
u
t
Since d
3
V = d
3
V
t
d
3
v
1
d
3
v
2
= d
3
v
t
1
d
3
v
t
2
.
Assumption of molecular chaos: This is the central assumption of the Boltzman equa-
tion.
The number of pairs with velocities ( v
1
, v
2
) about to collide is the product
53
f(r, v
1
, t)d
3
rd
3
v
1
f(r, v
2
, t)d
3
rd
3
v
2
i.e. both particles do not know about each other until they collide, hence no correlation
before the collision.
The scattered ux of v
2
hitting v
1
is
I =
_
d
3
v
2
_
d() [ v
1
v
2
[ f(r, v
2
, t)
where () is the scattering cross section.
f(r, v
1
, t)d
3
v
1
is the number of target particles, therefore Rd
3
v
1
= f(r, v
1
, t)d
3
v
1
I
For

R dene a collision by variables

v
t
1
,

v
t
2
v
1
, v
2
so that v
1
A.
The scattered ux of

v
t
2
on

v
t
1
is
I
t
=
_
d
3
v
t
2
_
d
t
()

v
t
1

v
t
2

f(r,

v
t
2
, t) =
_
d
3
v
t
2
_
d() [ v
1
v
2
[ f(r,

v
t
2
, t)
using microscopic time reversal
t
() = () (spin and internal quantum states are ne-
glected).
f(r,

v
t
1
, t)d
3
v
t
1
is the number of target particles (no integral since ( v
1
,

v
t
2
, ) determine

v
t
1
).
Therefore, Rd
3
v
1
= f(r,

v
t
1
, t)d
3
v
t
1
I
t
.
Since d
3
v
1
d
3
v
2
= d
3
v
t
1
d
3
v
t
2
the probability for the reverse collision process is
R =
_
d
3
v
2
_
d() [ v
1
v
2
[ f(r,

v
t
1
, t)f(r,

v
t
2
, t)
Hence Boltzmans transport equation:
(

t
+ v
1

r
+
1
m

v
1
)f(r, v
1
, t) =
_
()d
_
d
3
v
2
[ v
1
v
2
[ [f(r,

v
t
2
, t)f(r,

v
t
1
, t) f(r, v
2
, t)f(r, v
1
, t)]
Time reversal would imply that f(r, v
1
, t) also solves this equation. However, changing
t t, and all v v in this equation, the left side changes sign (acting on f(r, v
1
, t))
while the right hand side does not change sign (acting on f(r, v, t) with the various v).
Hence time reversal, although valid on the microscopic level, is broken on the macrospcopic
54
level. [If t t and also v
i

v
t
i
the equation is invariant, i. e. reversing the order of
colliding particles; assumption of molecular chaos looses this symmetry.]
Consider the case with no external force F = 0 and no r dependence (e.g. no density waves).
In equilibrium
f
t
= 0, therefore the right side is equal to 0 for all v
1
, which requires
f(v
2
)f(v
1
) = f(v
t
2
)f(v
t
1
)
hence ln f(v) is conserved in all collisions. The conserved quantities are momentum and
energy, therefore
ln f(v) = A +

B v + c (v)
2
= b(v v
0
)
2
+ ln a
f(v) = ae
b(v v
0
)
2
which is the Maxwell distribution (up to a center of mass velocity).
Boltzmans H theorem
As a candidate for entropy, Boltzman dened the function H(t):
H(t)
_
f(v, t) ln[f(v, t)]d
3
v
where f(v, t) is a solution of Boltzmanns equation.
We show now that
dH(t)
dt
0 so that H(t) decreases with time. Dene as a shorthand
f
1
= f( v
1
, t) f
2
= f( v
2
, t)
f
t
1
= f(

v
t
1
, t) f
t
2
= f(

v
t
2
, t)
(2)
Thus we get:
f
1
t
=
_
d
3
v
2
_
d() [ v
1
v
2
[ (f
t
2
f
t
1
f
2
f
1
)
dH
dt
=
_
d
3
v
2
_
d
3
v
1
_
d() [ v
1
v
2
[ (f
t
2
f
t
1
f
2
f
1
)(1 + ln f
1
)
by changing v
1
v
2
, using eq. for
f
2
t
and taking
1
2
of both, we get
dH
dt
=
1
2
_
d
3
v
1
_
d
3
v
2
_
d() [ v
2
v
1
[ (f
t
2
f
t
1
f
2
f
1
)[2 + ln(f
1
f
2
)]
55
Now, using equations for
f

1
t
,
f

2
t
, i.e. f
t
i
f
i
we get
dH
dt
=
1
2
_
d
3
v
1
_
d
3
v
2
_
d
t
()

v
t
2

v
t
1

(f
2
f
1
f
t
2
f
t
1
)[2 + ln(f
t
1
f
t
2
)]
And nally, adding the last two equations
dH
dt
=
1
4
_
d
3
v
1
_
d
3
v
2
_
d() [ v
2
v
1
[ (f
t
2
f
t
1
f
2
f
1
)[ln(f
1
f
2
) ln(f
t
1
f
t
2
)]
Since (x y) ln
x
y
> 0 for all x, y we have
dH
dt
0 (3)
In equilibrium f
t
2
f
t
1
= f
2
f
1
and
dH
dt
= 0.
Therefore, a candidate for entropy is (H) which increases with time and is maximal in
equilibrium. [Note that f(r, v, t) is coarse grained by the assumption of molecular chaos,
therefore initial velocities v
i
and nal velocities v
t
i
are distinguished].
4b. Brownian motion
Brown (1828) - the random motion of pollen particles in a solution.
Einstein (1905) - random walk problem.
Consider one dimension: P
n
(m) is the probability of arriving at coordinate m after n steps,
i.e. one needs
1
2
(n + m) steps to the right and
1
2
(n m) steps to the left,
P
n
(m) =
_
1
2
_
n
n!
[
1
2
(n + m)]![
1
2
(n m)]!
where
1
2
is the equal probability of going right or left. Normalization:

n
m=n
P
n
(m) = 1
Averages: m = 0, m
2
= n. For m n (since
_
m
2
=

n n)) and using Stirlings limit
we get:
P
n
(m)
2

2n
e
m
2
/2n
Dening x = ml (l is a step size in space) and t = n ( is the time for each step)), we get
56
x
2
=
l
2

t = 2Dt
so that D =
l
2
2
is the diusion coecient.
In this continuoum form
P(x) =
1

4Dt
e
x
2
/4Dt
In general, D is dened by the linear response:
j(r, t) = D

n(r, t)
If there is a gradient, then there is a current which responds to reduce this gradient.
The continuity equation is j +

t
n(r, t) = 0. Taking a gradient yields

2
n(r, t)
1
D
n(r, t)
t
= 0
This is the diusion equation. The solution with an initial value n(r, t = 0) = N
3
(r) is
n(r, t) =
N
(4Dt)
3/2
e
r
2
/4Dt
with
_

0
n(r, t)4r
2
dr = N. This solution shows <

r(t) >= 0 and
< (r(t))
2
>=
1
N
4
_

0
n(r, t)r
4
dr = 6Dt
Langevins equation
A particle of mass M is immersed in a medium which produces a random uctuating force
M

A(t).

A(t) has time correlation
_

A(t + )

A(t)
_
which decays fast with .
The correlation time
col
is the time between collisions of medium particles and the particle
of mass M. The medium leads also to irreversibility, i.e. friction.
.
v= v +

A(t)
1/ is the time scale for approaching equilibrium (see below).
We assume
col
<< 1/ so that the average on the distribution of random forces is
_

A(t)
_
=
0, while A
i
(t + )A
j
(t)) =
1
3
C ()
ij
, i.e.

A(t + )

A(t)) = C().
Note that
_

A r
_
= 0, while
_

A v
_
,= 0 (exercise).
Evaluate now r
2
) assuming that the particle is in equilibrium at temperature T,
1
2
Mv
2
) =
3
2
kT:
57
r v =
1
2
d
dt
r
2
r
.
v=
1
2
d
2
dt
2
r
2
v
2
= r (v +

A)
average :
d
2
dt
2

r
2
_
+
d
dt

r
2
_
= 2

v
2
_
= 6kT/M

r
2
_
=
6kT
M
t + c
1
+ c
2
e
t
Initial value r(t = 0) = 0
d
dt
r
2
= 2r v = 0 at t = 0.

r
2
_
=
6kT
M
[t
1

(1 e
t
)]
t << 1/

r
2
_
=

v
2
_
t
2
ballistic
t >> 1/

r
2
_

6kT
M
t diusive
Friction is related to diusion D, D =
kT
M
. is also related to mobility - responce to
constant force , e.g electric eld E for particle with charge e.
.
v= v +

A(t) +
e

E
M
In steady state
.
v= 0 v) = eE/M E
D = kT/e Einsteins relation
So far the random force

A was not explicit. Consider next the velocity uctuations. We
assume now E = 0; if E ,= 0 one needs just to shift v v + e

E/M.
v(t) = v(0)e
t
+ e
t
_
t
0
e
u

A(u)du
v(t)) = v(0)e
t

v
2
(t)
_
= v
2
(0)e
2t
+e
2t
_
t
0
_
t
0
e
(u
1
+u
2
)
A(u
1
)A(u
2
))du
1
du
2
= v
2
(0)e
2t
+
C
2
(1e
2t
)
t should have equilibrium
1
2
Mv
2
) =
3
2
kT C = 6kT/M.
Strength of uctuating force , both uctuations and friction are due to the medium.
58

v
2
(t)
_
= v
2
(0) + (
3k
B
T
M
v
2
(0))(1 e
2t
)
General solution for r
2
) (v
2
(0) ,= 3k
B
T/M)
:
d
dt
2

r
2
_
+
d
dt

r
2
_
= 2

v
2
_

r
2
_
=
1

2
v
2
(0)(1 e
t
)
2

3k
B
T
M
2
(1 e
t
)(3 e
t
) +
6k
B
T
M
t =
=
_
_
_
v
2
(0)t
2
+ O(t
3
) t << 1/
6k
B
T
M
t t >> 1/
Velocity correlations: K
v
() v(t)v(t + ))
= v
2
(0)e
(2t+)
+ e
(2t+)
_
t
0
_
t+
0
e
(u+u

)
C(u u
t
)dudu
t
=
=
_

_
v
2
(0)e
(2t+)
+
C
2
e
(2t+)
(e
2t
1) > 0
v
2
(0)e
(2t+)
+
C
2
e
(2t+)
(e
2(t+)
1) < 0
K
v
() = v
2
(0)e
[[
+ (
3kT
M
v
2
(0))(e
[[
e
(2t+)
)
=
3k
B
T
M
e
[[
at long times t + , t >> 1/ .
Note that replacing v
2
(0) by its equilibrium average yields K
v
() = (3k
B
T/M)e
[[
at all
times t.
Reaching equilibrium at large times K
v
() becomes t independent stationary medium.
Note that for correlations of just one component, e.g. v
x
, we have
K
v
x
() = v
x
(0)v
x
()) =
k
B
T
M
e
[[
.
Below the Fourier transform is needed:

v
x
() =
_

K
v
x
()e
i
d =
2k
B
T
M

2
+
2
59
4c. Fluctuation Dissipation Theorems (FDT)
General properties of averages of stationary medium
Consider a variable x(t) such that K
x
() = x(t) x(t + )) is independent of the time t.
K
x
(0) > 0; [x(t) x(t + )]
2
) = 2[K
x
(0) K
x
()] 0
it follows that [K
x
()[ K
x
(0) is a decreasing function, and usually K
x
()

0. Also
K
x
() = x(t )x(t)) = K
x
(), with shift t t .
Consider now the Fourier transform x() :
x(t) =
_

x()e
it d
2
For a precise derivation one needs x(t) to be nite only in the interval [
T
2
,
T
2
]. The spacing
between distinct values with cos(
T
2
) = 0 or sin(
T
2
) = 0 is =
2
T
. (See Wannier
p. 481). For T consider
K
x
() =
_
d
_
d
t
e
iti

(t+)
x()x(
t
)) /(2)
2
Since this is t independent, perform the integral
_
...
dt
T
K
x
() =
_
d
_
d

2
T
( +

)e
i

x()x(

)
)
(2)
2
=
_
d

[x()[
2
_
e
i
/(2)
2
This is the Wiener-Khinchin theorem:

x
() =
_
x(t)x(t + )) e
i
d =

2

[x()[
2
_

[x()[
2
_
is called the intensity, or the power spectrum of x.
As an example, consider Langevins equation:
(i + )v() = A() [v()[
2
=
[A()[
2
(
2
+
2
)
from the theorem above
v
() =

A
()

2
+
2
which is consistent with previous result:
v
() =
2k
B
T
M

2
+
2
and
A
() =
2k
B
T
M
, the latter corresponds to white noise.
Note also that ix() = v(), hence
[x()[
2
=
[v()[
2

2

x
() =
2k
B
T
M

(
2
+
2
)
2
60
Dissipation and Response functions
An external force adds the Hamiltonian term F(t)x. The Hamiltonian becomes:
H =
p
2
2m
+ V (x; env) F(t)x
where env stands for the coordinates and momenta of the environment. For an explicit
system + environment see the Appendix.
The equations of motion are:
x =
H
p
=
p
m
, p =
H
x
=
V
x
+ F = m x
hence F(t) is indeed an external force.
The rate at which the systems energy changes is
dH
dt
=
=0
..
H
p
p +
H
x
x +... +
H
t
= x
dF
dt
where the sum of the rst two terms on the right hand side vanishes by Hamiltons equations
for x, p; the ... terms are similar derivatives with respect to the coordinates and momenta
of the environment which also vanish due to their corresponding Hamiltons equations.
The dissipation rate is the rate of energy absorption from the external source, averaged
on time. This includes an average on the enviorment x(t)) (e.g. on the random force in
Langevins equation) and on the explicit time dependence in F(t), hence
dE
dt
= x)
dF
dt
The dissipation rate can be expressed in terms of a response function
x
() where the linear
response at small forces F is dened by
x()) =
x
()F()
It is more compact to consider a single frequency F(t) =
1
2
f
0
e
it
+
1
2
f

0
e
it
(although a
Fourier sum can be added)
x(t)) =
1
2

x
()f
0
e
it
+
1
2

x
()f
0
e
it
with

x
() =
x
()
The dissipation is then:
61
dE
dt
=
_
x
F
t
_
=
i
4
[
x
()f
0
e
it
+

x
()f

0
e
it
][f
0
e
it
f

0
e
it
] =
i
4
[f
0
[
2
[
x
()

x
()]

dE
dt
=
1
2
[f
0
[
2
Im
x
()
Consider now the specic case of Langevins equation, where the environment leads to
friction and random force terms. Averaging on the random force,
M(
2
i) x()) = F(), hence the response has the same frequency as the source and
the response function is

x
() =
x())
F()
=
1
M(
2
+ i)
The dissipation rate is therefore proportional to Im
x
() =

M(
2
+
2
)
2
Comparing with
x
() we get the Fluctuation - Dissipation theorem:

x
() =
2k
B
T

Im
x
()
Note: same result holds for
v
(),
v
() with an external source term in the Hamiltonian
F(t)p/M (exercise).
In conclusion,
x
() are uctuations in absence(!) of external force.

x
() linear response and energy dissipation due to external force.
Applications
1. Kapplers experiment (1931) : uctuations in angle of a mirror suspended on a ne wire
with a restoring force

1
2
C
2
_
=
1
2
k
B
T determine k
B
. Note : Variance is independent of
gas density. Equilibrium is achieved after time
col
, 1/.
62
Low density: rare strong deections High density: frequent weak deections
2. A square vane of area 1cm
2
, painted white
on one side and black on the other, is at-
tached to a vertical axis and can rotate freely
about it as shown. Assume that the vane can
support a temperature dierence between its
black and white sides. Suppose the arrange-
ments is placed in He gas at room tempera-
ture and sunlight is allowed to shine on the
vane. Explain qualitatively why : (a) At ex-
tremely small densities the vane rotates. (b)
At some intermediate (very low) density the
vane rotates in a sense opposite to that in (a);
estimate this intermediate density and the cor-
responding pressure. (c) At a higher (but still
low) density the vane stops.
Answers :
(a) Radiation pressure on white > black.
(b) Black heats up and nearby hotter atoms exert higher pressure. Gas is in local equilibrium
with density n 1cm
3
.
(c) Viscosity is eective at
col
<< 1/ and global equilibrium is achieved.
63
3. Electrical circuit:
L
dI
dt
= RI + V (t)
V (t) - voltage uctuations
Langevin analogy:
dv
x
dt
= v
x
+ A
x
(t) => R/L
Correlation of A is determined by
1
2
Mv
2
x
) =
1
2
k
B
T
ML
=
1
2
LI
2
)

v
x
(w)
=
2k
B
T
M

2
+
2

I
() =
2k
B
T
L

R/L

2
+(R/L)
2
<<R/L
=
2k
B
T
R

A
x
()
=
2k
B
T
M

V/L
() =
2k
B
TR
L
2
=>
V
() = 2k
B
TR
K
I
( = 0) = I
2
(t)) =
_

I
()
d
2
Fluctuations measured in interval
1
< [[ <
2
: I
2
)

2
= 2

2
_

I
()
d
2
. The factor
2 accounts for < 0.
I
2
)

2k
B
T
R
(
2

1
) [
1,2
<< R/L]
V
2
)

2
=
2

k
B
TR(
2

1
)
4. Brownian particle :
0
response time of eye, i.e. only frequencies [[ <
2

0
are observed.
v
2
x
)
obs
= [K
v
( = 0)]
obs
= 2
2/
0
_
0

v
()
d
2
=
2k
B
T
M
tan
1
_
2

_
1

10
6
v
2
x
)
obs

4k
B
T
M

1

<<
k
B
T
M
i.e instead of
_
k
B
T/M 10
1
cm/ sec a small fraction of the uctuations is observed
_
v
2
x
)
obs
10
4
cm/ sec.
64
Fluctuation Dissipation Theorem (FDT): quantum version
Fluctuations
Consider a system in equilibrium, no external force. For x (t) (Heisenberg representation)
dene a symmetrized correlation
K () =
1
2
x (t) x (t + ) + x (t + ) x (t)) .
where a subscript x is omitted here, for convenience. The symmetrized correlation involves
a hermitian operator and has an obvious classical limit. note, however, that there are FDTs
for non-symmetrized correlations.
Note that K () = K () and K() is real. The average is dened by
...) =

n
n[ ... [n) e
E
n
/Z.
where Z =

n
e
E
n
is the partition sum.
Introduce a unit operator

m
[m) m[ = 1 and dene (time independent) matrix elements
x
mn
by
n[ x (t) [m) = e
i(E
n
E
m
)t/
x
nm
.
Therefore
K () =
1
2Z

n,m
e
i(E
n
E
m
)t/
x
nm
x
mn
e
i(E
m
E
n
)(t+)/
e
E
n
+ (t t + ) =
=
1
2Z

n,m
e
i
nm

[x
nm
[
2
e
E
n
+ c.c. where
nm
=
E
n
E
m

() =
_
K () e
i
d =

Z

n,m
e
E
n
[x
nm
[
2
[ (
nm
) + ( +
nm
)]
Interchange n m in the rst (
nm
),
() =

Z

n,m
_
e
E
m
+ e
E
n
_
[x
nm
[
2
( +
nm
) =
=

Z

n,m
e
E
n
_
e
(E
m
E
n
)
+ 1
_
[x
nm
[
2
( +
nm
) =
=

Z
_
1 + e

_
n,m
e
E
n
[x
nm
[
2
( +
nm
)
Dissipation
Add coupling to an external force, F(t):

H =

H
0
+

V (t) where

V (t) = F(t) x .
65
We use here Shrdingers representation to avoid the form H
H
(t) ,= H(t). The response
function () gives x(t)) in terms of F(t ) at all previous times, i.e. > 0:
x(t)) =
_

0
()F(t )d.
Assume, without loss of generality,
F(t) =
1
2
f
0
e
it
+
1
2
f

0
e
it
.
Therefore
x(t)) =
1
2
_

0
()
_
f
0
e
i(t)
+ f

0
e
i(t)

d.
The Fourier transform is
() =
_

0
()e
i
d (() = 0 for < 0),
therefore
x(t)) =
1
2
()f
0
e
it
+
1
2
()f

0
e
it
,
since () =

() [() is real]. In terms of the density matrix (t) corresponding to the

full Hamiltonian H, the dissipation rate is
dE
dt
= Tr
d
dt
( H) =
i

Tr[ , H]H + Tr
H
t
(4)
= Tr x
F
t
= x)
F
t
(5)
where the equation of motion of is used, as well as the cyclic property of the trace. The
result is formally the same as in the classical case, except that x) is a quantum expectation
value.
Using the response function and averaging on time,
dE
dt
=
1
2
(()f
0
e
it
+ ()f

0
e
it
)
i
2
(f
0
e
it
f

0
e
it
),

dE
dt
=
i
4
[f
0
[
2
[() +

()] =

2
[f
0
[
2
Im[()].
66
Therefore Im[()] measures the dissipation rate. Since this is positive
Im[()] > 0 for > 0.
Golden Rule for the Dissipation
The rate of transition between states at time T is
W
nm
=
1

_
m[

V [n)dt

2
1
T
=
=
1

1
2
_
dt
_
e
i(E
m
E
n
)t/
x
mn
(f
0
e
it
+ f

0
e
it
)

2
1
T
We note that
1
T

_
e
i(
mn
)t
dt

2
=
1
T
_
T/2
T/2
e
i(
mn
)t
dt
_
T/2
T/2
e
i(
mn
)t

dt
t
=
=
1
T
_
T/2
T/2
e
i(
mn
)t
[2(
mn
)]dt = 2(
mn
).
W
nm
=

2
2
[f
0
[
2
[x
mn
[
2
[(
nm
) + (
nm
+ )].
The energy dissipation rate can now be calculated
Q =
1
Z

n,m
(E
m
E
n
)W
nm
e
E
n
=
=

2
2
[f
0
[
2
1
Z

n,m
e
E
n
[x
nm
[
2
[( +
nm
) (
nm
)] =
=

2
2
[f
0
[
2
1
Z

n,m
(e
E
n
e
E
m
)[x
nm
[
2
( +
nm
).
Noting that
e
E
n
e
E
m
= e
E
n
_
1 e
(E
m
E
n
)
_
= e
E
n
_
1 e

mn
_
,
and using the function with
mn
= , we get
Q =

2
[f
0
[
2
(1 e

)
1
Z

n,m
e
E
n
[x
nm
[
2
(
nm
+ )
67
Im[()] =

(1 e

n,m
1
Z
e
E
n
[x
nm
[
2
(
nm
+ ).
() = coth
_
1
2

_
Im[()]
The coth
_
1
2

_
can be interpreted as the mean uctuation of an oscillator coordinate:

x
2
osc
_
=

2m

(a + a

)
2
_

2
e

1
+ 1 = coth
_
1
2

_
.
Note that in the limit 0 (i.e. for k
B
T >> ) the quantum FDT becomes
()
2k
B
T

Im[()].
which is our previous classical FDT.
4d. Onsagers Relations
Onsagers relations (1931) consider the response to deviations from equilibrium. The re-
sponses are measured by a ow processes x
i
in coordinates x
i
(e.g heat ow, electric current,
mass transfer, etc.), while the deviation from equilibrium corresponds to forces f
i
, e.g
gradients in temperature, potential, pressure, etc. The forces f
i
can be external ones, or
they can form spontaneously as a uctuation.
If the system is close to equilibrium, we assume a linear relation between the forces and the
responding currents:
x
i
=
ij
f
j
,
where summation convention is used (i.e. repeated indices are summed).
ij
are known as
kinetic coecients. [In general, however, the relation is non-local in time and can be written
as x
i
(t)
_
t

ij
(t t
t
)f
j
(t
t
)dt
t
.]
We consider x
i
as coordinates of the entropy S so that the forces are f
i
= S/x
i
. For small
deviations from equilibrium values x
i
we expand near the entropy maximum
S = S( x
i
)
1
2

ij
(x
i
x
i
)(x
j
x
j
)
f
i
=
S
x
i
=
ij
(x
j
x
j
)
68
x
i
=
ij

jk
(x
k
x
k
) .
Consider
x
i
f
j
) = k
B
_

x
i

x
j
e

ij
(x
i
x
i
)(x
j
x
j
)/2k
B
_

ij
(x
i
x
i
)(x
j
x
j
)/2k
B
where the limits are allowed due to fast convergence.
By partial integration x
i
f
j
) = k
B

ij
x
i
x
j
) =
ik
f
k
x
j
) = k
B

ij
.
This is a type of FDT the uctuations x
i
x
j
) are related to the dissipation
ij
.
Consider x
i
()x
j
(0)) with microscopic time reversal (dissipation
ij
and irreversibility
arise after average on the environment, or on the ensemble.)
x
i
()x
j
(0)) = x
i
()x
j
(0)) = x
i
(0)x
j
())
where in the second form both times are shifted by , allowed in equilibrium.

[
0
x
i
(0)x
j
(0)) = x
i
(0) x
j
(0))
ij
=
ji
These are Onsagers relations.
Dene

F =
1
2

ij
f
i
f
j
x
i
=


F
f
i

S =
S
x
i
x
i
= f
i
x
i
= f
i

F
f
i
= 2

F

S > 0, hence

F > 0 for all choices of f
i
, therefore the matrix
ij
is positive denite.
A few more properties of
ij
:
In presence of a magnetic eld B time reversal involves B B, hence
ij
(B) =
ji
(B);
e.g. for the Hall conductance
xy
(B) =
yx
(B).
In presence of angular velocity , similarly,
ij
() =
ji
().
Furthermore, if for some coordinate x
i
x
i
, x
j
+x
j
under time-reversal, then

ij
=
ji
69
Alternative derivation:
Consider f
i
as external forces with a Legendre transformed entropy S

S = S(x
i
x
i
)f
i
.
Since f
i
=
S
x
i
d

S = (x
i
x
i
)df
i
, i.e.

S =

S(f
i
).
Weight of a conguration is now
e

S/k
B
= e
S/k
B
(x
i
x
i
)f
i
/k
B
here S = Sx
i
has all orders of x
i
.
x
i
)
f=0
= 0, i.e. ow is only in response to f
j
,= 0.
x
i
) =
_
x
i
e
S/k
B
(x
j
x
j
)f
j
/k
B
_
e
S/k
B
(x
j
x
j
)f
j
/k
B
=
_
x
i
e
S/k
B
[1 (x
j
x
j
)f
j
/k
B
]
_
e
S/k
B
[1 (x
j
x
j
)f
j
/k
B
]
+ = x
i
x
j
)
0
f
j
/k
B
+O(f
2
)
Here linear response is explicit
ij
= x
i
x
j
)
0
/k
B
as above.
x
i
x
j
)
0
is evaluated with f
i
= 0 and microscopic time reversal can be used as above

ij
=
ji
Application: Bi-metal junctions
Using E
1
+ E
2
=const, and Q
1
+ Q
2
=const we have
dE
12
= dE
1
= dE
2
, dQ
12
= dQ
1
= dQ
2
and since dE = TdS + V dQ we obtain
dS =
dE
1
T
1
+
dE
2
T
2

V
1
T
1
dQ
1

V
2
T
2
dQ
2
= dE
12
. .
x
1
_

1
T
1
+
1
T
2
_
. .
f
1
+dQ
12
. .
x
2
_
V
1
T
1

V
2
T
2
_
. .
f
2

E
12
=
11
_

1
T
1
+
1
T
2
_
+
12
_
V
1
T
1

V
2
T
2
_

Q
12
=
21
_

1
T
1
+
1
T
2
_
+
22
_
V
1
T
1

V
2
T
2
_
70
Seebeck coecient is dened by V = T in an open circuit, i.e.

Q
12
= 0:
V
1
V
2
=

21

22
T
2
T
1
T
=

21

22
T
(to lowest order in gradients)
Peltier coecient , corresponds to T
1
= T
2
and is dened by

E =

Q =

12

22

12
=
21
= T This is Kelvins relation.
71
Appendix: Langevins equation from a Hamiltonian
This appendix considers a microscopic model for an environment that produces the dissipa-
tion and the random forces (t) in Langevins equation
m q + q +
V
q
= (t) .
The derivation is based on Caldeira and Legget, Ann. Phys. 149, 374(1983) and provides a
basis for Brownian motion.
Consider a linear coupling of the particle coordinate q to a set of Harmonic oscillators, with
coordinates Q
i
and momenta P
i
; the linearity is valid when the particles trajectories are
suciently conned.
H =
P
2
2m
+ V (q) +
N

i=1

i
qQ
i
. .
Coupling of oscillators to a heat bath
+

i
_
P
2
i
2M
i
+
1
2
M
i

2
i
Q
2
i
_
. .
Bath

s Energy
+V (q)
Where V (q) is chosen to retain the original minimum of V (q). Minimizing H with respect
to Q
i
gives Q
i
=

i
M
i

2
i
q, hence q dependent potential terms appear

2
i
M
i

i
q
2
(1+
1
2
) which
we wish to cancel by
V (q) =

2
i
2M
i

2
i
q
2
.
The equations of motion for Q
i
are

Q
i
+
2
i
Q =

i
M
i
q(t)e
t
where retarded response is dened by 0
+
, i.e. at t the coupling to the environ-
ment vanishes. The solution is
Q
i
(t) = Q
0
i
(t) +

i
M
i
_
dt
t
q(t
t
)
_
d
2
e
i(tt

)+t

2
i
where Q
0
i
is the solution of

Q
i
+
2
i
Q = 0. Let us shift i and rewrite
Q
i
(t) = Q
0
i
(t) +

i
M
i
_
dt
t
q(t
t
)
_
d
2
e
i(tt

)
(
i
i) ( +
i
i)
e
t
To solve this integral we need to take the upper contour for t > t
t
(see gure, so that e
i(tt

)
vanishes on the upper half circle). The closed contour integration has two poles (see gure)
and therefore
Q
i
(t) = Q
0
i
(t) +

i
M
i
_
t

dt
t
q(t
t
)i
_
e
i
i
(tt

)
2
i

e
i
i
(tt

)
2
i
_
e
t

72
Hence,
Q
i
(t) = Q
0
i
(t)

i
M
i

i
_
t

dt
t
q(t
t
) sin
i
(t t
t
)e
t

= Q
0
i
(t)

i
M
i

2
i
q(t) +

i
M
i

2
i
_
t

dt
t
q(t
t
) cos
i
(t t
t
) (*)
where partial integration is done, and at t
t
we use e
t

0; in the nal form we set

= 0.
Equation of motion for q(t):
m q +
V
q
+
V
q
=

i
Q
i
=

i
Q
0
i
+

2
i
M
i

2
i
q(t)

2
i
M
i

2
i
_
t

dt
t
q(t
t
)cos
i
(t t
t
)
Note the cancelation of
V
q
. Dene the spectral density as
J() =

2

2
i
M
i

i
(
i
)
so that
m q +
V
q
+
2

_

0
d
J()

_
t
dt
t
q(t
t
) cos (t t
t
) =

i
Q
0
i
.
To obtain linear dissipation, as in Langevins equation, we assume an ohmic bath
J() = . Using
_

0
d cos (t t
t
) =
1
2
Re
_

e
i(tt

)
d = (t t
t
)
we get
m q + q +
V
q
=

i
Q
0
i
We identify now the random force (t) =

i
Q
0
i
(t). Free oscillators evolve as
Q
0
i
(t) =
1

Q
0
i
(0) sin
i
t + Q
0
i
(0) cos
i
t
73
With random initial conditions we have Q
0
i
(0)) =

Q
0
i
(0)) = 0 so that Q
0
i
(t)) = 0.
For a thermal distribution (classical oscillators)

Q
0
i
(t)Q
0
i
(0)
_
=

(Q
0
i
(0))
2
_
cos
i
t =
k
B
T
M
i

2
i
cos
i
t
(t)(0)) = k
B
T

2
i
M
i

2
i
cos
i
t =
2

k
B
T
_

0
d
J()

cos t .
With J() = we obtain
(t)(0)) = 2k
B
T(t) ()
i.e. the noise correlation relates to the dissipation .
To check FDT, we need to evaluate the response of (t) to a source external to the oscillators
q
ex
(t); this source may or may not be the particles position q(t). Adding a term q
ex
(t)(t)
to the Hamiltonian, where q
ex
(t) = q
ex
()e
it+t
, the response is then dened by () =
()q
ex
(). Since q
ex
(t) aects the equation of motion for Q
i
(t) in the same way as q(t) in
Eq. (*) above,
(t) =

2
i
M
i

2
i
q
ex
(t)

2
i
M
i

2
i
_
t

dt
t
q
ex
(t
t
)cos
i
(t t
t
)
=
2

_

0
d
t
J(
t
)

t
q
ex
()e
it+t
+
2

_

0
d
t
J(
t
)

t
_
t

dt
t
e
it

+t

cos[
t
(t t
t
)]iq
ex
()
=
2

__

0
d
t
J(
t
)

t


2
_

0
d
t
J(
t
)

t
(
1
+
t
+ i
+
1

t
+ i
)
_
q
ex
()e
it+t
.
Therefore, for a general bath (not necessarily an ohmic one)
Im() = J()
For an Ohmic bath J() = with Eq. (**) we recover the classical FDT. This relates the
uctuations of the bath to the response () of the bath to an external force.
Note that there is a separate FDT relating the uctuations of the particle q(t)q(0)) and its
response
q
() to an external force, i.e. adding to the Hamiltonian a term q(t)
ex
(t). For
V (q) = 0 this was checked in a previous section on Langevins equation. For V (q) ,= 0 it is
nontrivial to nd either the uctuations or the response, yet FDT is guaranteed.
We proceed to check the quantum FDT, allowing for a general J(). Consider the harmonic
oscillators creation a

i
and annihilation a
i
operators,
74

Q
0
i
(t) =
_

2M
i

i
e
i

H
0
t
_
a
i
+ a

i
_
e
i

H
0
t
=
_

2M
i

i
_
e
i
i
t
a
i
+ e
i
i
t
a

i
_
where

H
0
=

i

i
(a

i
a
i
+
1
2
) is the free Hamiltonian for the oscillators. Hence the correlations
become
1
2
_

Q
0
i
(t)

Q
0
i
(0) +

Q
0
i
(0)

Q
0
i
(t)
_
=
=

4M
i

i
__
e
i
i
t
a
i
+ e
i
i
t
a

i
__
a
i
+ a

i
_
+
_
a
i
+ a

i
__
e
i
i
t
a
i
+ e
i
i
t
a

i
__
=
=

2M
i

i
(2n
i
+ 1) cos
i
t
Boson occupation n
i
=
_
a

i
a
i
_
=
1
e

i 1
, hence
K

(t) =
1
2
(t)(0) + (0)(t)) =
=

2
i
2M
i

2
i
coth
_
1
2

i
_
cos
i
t =

_

0
dJ() coth
_
1
2

_
cost
Using J() = J() we identify the Fourier transform

() = J() coth
_
1
2

_
.
Since the response was found as Im() = J(), we obtain the full quantum FDT

() = coth(
1
2
)Im() .
75