Vous êtes sur la page 1sur 80

Contents

1 Prelude 5
1.1 Heat and temperature . . . . . . . . . . . . . . . . . . . 5
1.2 The rst law of thermodynamics . . . . . . . . . . . . . . 5
1.3 The Carnot Engine . . . . . . . . . . . . . . . . . . . . . 5
1.4 The second law of thermodynamics and entropy . . . . . 5
2 The formal statistical mechanics theory 7
2.1 Entropy, temperature and pressure of ideal gas . . . . . . 8
2.2 Energy exchange contact between two systems, and the
relation between entropy and temperature . . . . . . . . 11
2.3 Energy and volume exchange contact between two sys-
tems and the relation between entropy and pressure . . . 13
2.4 Energy and particle exchange contact between two sys-
tems and the relation between entropy and chemical po-
tential . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.5 The rst law of thermodynamics . . . . . . . . . . . . . . 15
2.6 The second law of thermodynamics . . . . . . . . . . . . 15
2.7 The third law of thermodynamics . . . . . . . . . . . . . 16
2.8 The canonical ensemble . . . . . . . . . . . . . . . . . . . 16
2.9 The partition function Z and the Helmholtz free energy F 18
2.9.1 Paramagnet . . . . . . . . . . . . . . . . . . . . . 21
2.9.2 Maxwell distribution . . . . . . . . . . . . . . . . 22
2.9.3 Equal-partition theorem . . . . . . . . . . . . . . 23
2.10 The grand canonical ensemble . . . . . . . . . . . . . . . 25
2.11 The grand partition function and thermodynamic function 26
2.11.1 Molecular adsorption -The hydrogen storage
problem . . . . . . . . . . . . . . . . . . . . . . . 27
1
3 Black body radiation 31
3.1 Statistical mechanics of harmonic oscillators . . . . . . . 32
3.2 Quantum statistical mechanics of harmonic oscillator . . 33
3.3 Classical description of the E-M eld . . . . . . . . . . . 34
3.4 Quantum statistical mechanics of the cavity E-M eld . . 36
3.5 Acoustic phonon contribution to the specic heat . . . . 37
4 Ideal Fermi gas 41
4.1 Particle in a box . . . . . . . . . . . . . . . . . . . . . . 41
4.2 Grand canonical ensemble treatment of Fermi particle in
a box . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
4.3 The particle number constraint and the chemical potential 42
4.4 The total internal energy and the specic heat . . . . . . 47
4.5 The magnetic susceptibility . . . . . . . . . . . . . . . . 48
4.6 The physics of while dwarfs (optional) . . . . . . . . . . 50
5 Ideal Bose gas 55
5.1 The grand canonical ensemble treatment . . . . . . . . . 55
5.2 The particle number constraint . . . . . . . . . . . . . . 56
5.2.1 zero temperature macroscopic occupation of the
k = 0 state . . . . . . . . . . . . . . . . . . . . . 57
5.2.2 Non-zero temperature Bose Einstein condensation 57
5.3 The amazing work of indistinguishability (optional) . . . 60
6 Phase transitions 63
6.1 Example: critical phenomena of an easy-axis ferromagnet 65
6.2 Universality . . . . . . . . . . . . . . . . . . . . . . . . . 66
6.3 Scaling . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
6.4 A simple model for easy-axis ferromagnet: the Ising model 67
6.5 The predictions of mean-eld theory . . . . . . . . . . . 70
6.5.1 The spontaneous magnetization, the exponent . 70
6.5.2 The exponent . . . . . . . . . . . . . . . . . . . 72
6.5.3 The exponent . . . . . . . . . . . . . . . . . . . 73
6.5.4 The exponent . . . . . . . . . . . . . . . . . . . 73
6.5.5 Scaling . . . . . . . . . . . . . . . . . . . . . . . . 75
6.6 The Landau theory of phase transitions . . . . . . . . . . 76
6.6.1 First order phase transition . . . . . . . . . . . . 78
2
Statistical mechanics aims to provide a deductive method which
leads us from the microscopic physical world to the macroscopic world
starting from the atomic or molecular structure of matter and the fun-
damental dynamical principles of the microscopic world and combining
them with the logic of probability theory. The purpose of this course is
to learn how one applies statistical mechanics to understand the macro-
scopic behavior of various physical systems.
3
4
Chapter 1
Prelude
1.1 Heat and temperature
1.2 The rst law of thermodynamics
1.3 The Carnot Engine
1.4 The second law of thermodynamics
and entropy
5
6
Chapter 2
The formal statistical
mechanics theory
Let us start with the ideal gas, but looking at it from a microscopic point
of view. A microstate of the ideal gas is specied by the coordinates
and velocities (or equivalently the momenta) of all gas particles, i.e.
Microstate r
1
, ..., r
N
, p
1
, ..., p
N
. (2.1)
The space spanned by the microstate is called the phase space, it has
6N dimensions and is a continuous space. In this space the number of
states fall within an innitesimal volume element d
3
r
1
...d
3
p
N
around a
particular microstate r
1
, ..., r
N
, p
1
, ..., p
N
is
d =
1
N!h
3N
d
3
r
1
...d
3
p
N
. (2.2)
Here the
1
N!
is to ensure that the particle are identical hence so
long as the values of the 3N coordinates and 3N momenta are the
same resuing the particle labels does not lead to new state. For
example r
1
, r
2
..., r
N
, p
1
, p
2
..., p
N
is regarded as the same state as
r
2
, r
1
..., r
N
, p
2
, p
1
..., p
N
. Note that here we have performed the
1 2 reshuing. The factor
1
h
3N
(h is the Planck constant) accounts
for the Heisenbergs uncertainty principle which says the minimum er-
rors we can have in determining,say, the x-position and x-momentum
of a particle is bounded by p
x
x h Therefore the position and mo-
mentum value of a particle has fundamental fuzziness dictated by the
7
Heisenberg uncertainty principle. Hence per h
3N
in the phase space we
can have one state. (Note that h has the dimension of length times
momentum, hence h
3N
has the right unit to be the volume element in
the phase space.
2.1 Entropy, temperature and pressure of
ideal gas
The Hamiltonian governing the ideal gas is
H(r
1
, ..., r
N
; p
1
, ...p
N
) =
N

i=1
p
2
i
2m
. (2.3)
We note there is no interaction between the particles, and the Hamil-
tonian does not depend on the particle coordinates.
1
Question: How many state are there if the total energy of the whole
gas is between E and E + E? The answer is
W =
_
E

N
i=1
p
2
i
2m
E+E
d (2.4)
Note that H(r
1
, ...p
N
) = E denes a hyper surface in the phase space
on which the energy is a constant. To calculate this quantity we rst
note that the constraint in the integral does not aect the coordinate,
so the 3N coordinate integration can be done easily. It gives V
N
where
V is the volume of the container in which the ideal gas is conned.
What we left to do is the momentum integral
_
E

N
i=1
p
2
i
2m
E+E
d
3
p
1
...d
3
p
N
. (2.5)
To do this integral let us review a more familiar situation
_
Rx
2
+y
2
+z
2
R+R
dxdydz. (2.6)
1
An interaction term will look like
1
2

i =j
V (r
i
r
j
) where r
i
is the coordinate
of the ith particle.
8
This is the volume of the spherical shell between radius R and R+R.
For small R the answer is 4R
2
R. Here 4 is the surface area
a sphere of radius 1. How about in two dimensions? The answer is
2RR. Here 2 is the length of a circle of radius 1. In fact the
following formula applies generally to any dimension
S
D
R
D1
R (2.7)
As we have calculated in class,
S
D
= 2

D/2
(D/2)
(2.8)
is the volume of the spherical shell in D-dimension and S
D
is the area of
the hypersurface with radius 1 in D dimensions. Using this formula we
can compute Eq. (2.5). This is has been done in class and the answer
is
S
3N
E
3N1
2
(2m)
3N
2
(2.9)
Put this together with the factor of V
N
and
1
N!h
3N
we obtain
W =
_
V
N
h
3N
N!
2
3N/2
(3N/2)
(2m)
3N/2
(E)
3N1/2
_
E. (2.10)
In textbooks you will often encounter a quantity called density of
states. Its denition is the total number of states in energy interval
E and E + E is
W = (E)E. (2.11)
Hence according to Eq. (2.10)
(E) =
_
V
N
h
3N
N!
2
3N/2
(3N/2)
(2m)
3N/2
(E)
3N1/2
_
. (2.12)
If we take the logarithm of W we obtain (as N )
ln W = N
_
C
1
+ ln
V
N
+
3
2
ln
E
N
_
. (2.13)
9
(Here we have taken into account that ln E/N > 0 as N .
Here C
1
=
5
2
+
3
2
ln
_
4
3
_
3 ln h +
3
2
ln m. Now suppose we hold volume
and particle number constant and ask what change to ln W would an
energy change E E + dE induce? The answer is
dW =
3N
2E
dE. (2.14)
We recall from the introductory prelude that for ideal gas E =
3
2
Nk
B
T
which means
dW =
1
k
B
T
dE, (2.15)
or equivalently
dE = T(k
B
dW). (2.16)
We also recall that for xed volume according to the rst law of ther-
modynamics
dE = dQ = TdS (2.17)
hence we can identify dS with k
B
dW hence up to a reference constant
we can identify
S = k
B
ln W. (2.18)
This last equation was written in Boltzmanns tomb stone.
From Eq. (2.18) we see that the entropy is a function of E, V and
N. Here are some properties of S for ideal gas
S
E
=
3k
B
N
2E
=
1
T
S
V
=
k
B
N
V
=
k
B
N
Nk
B
T
P
=
P
T
. (2.19)
Therefore
S
E
=
1
T
S
V
=
P
T
. (2.20)
These equation are derived for ideal gas. The question is whether they
hold in general. In order to answer this question we consider system in
thermal and mechanical equilibrium.
10
Figure 2.1: Ludwig Boltzmann
2.2 Energy exchange contact between two
systems, and the relation between en-
tropy and temperature
When two systems I and II brought in thermal contact with each other
(meaning they can exchange energy) the total energy satises
E
I
+ E
II
= E (2.21)
and is xed. In this way, as a function of time, the system roams
through the microstates of I and II under the constraint Eq. (2.21)
According to the equal weight assumption all states of I and II satisfying
Eq. (??) will be visited equally frequently.
Let (E) be the density of states of the combined the system , and

I
(E
I
),
II
(E
II
) be the density of states of system I, II respectively.
The following equality holds
W = (E)E =
_
E<E
I
+E
II
<E+E
dE
I
dE
II

I
(E
I
)
II
(E
II
)
= E
_
dE
I
dE
II
(E
I
+ E
II
E)
I
(E
I
)
II
(E
II
)
= E
_
E
I
<E
dE
I

I
(E
I
)
II
(E E
II
). (2.22)
Therefore the probability that system I will have energy between E
I
11
and E
I
+ dE
I
is given by

I
(E
I
)
II
(E E
I
)dE
I
E
(E)E
. (2.23)
The function
I
(E
I
) increases very rapidly with E
I
, and
II
(E E
I
)
decreases very rapidly with E
I
, so the above probability is very sharply
peaked at certain E

I
The energy partition (E

I
, EE

I
) is to be expected
almost certainly in the equilibrium state of I and II. If two systems I
and II initially have E
0
I
and E
0
II
, after they are brought together the
combine system explore all possible energy partitions consistent with a
xed sum. As time elapses the most probable partition will be realized,
and after that the energy of the two subsystems no longer changes.
Thus we say a equilibrium is established between I and II.
The condition

I
(E
I
)
II
(E E
I
)dE
I
E = max (2.24)
is equivalent to
S
I
(E
I
) + S
II
(E E
I
) = max. (2.25)
Thus the as the systems moves toward equilibrium, the entropy always
increases. Another consequence of Eq. (2.25) is
S
I
E
I
[
E

I
=
S
II
E
II
[
E

II
=EE

I
. (2.26)
Suppose system I is an ideal gas then the left hand side will be equal
to
1
T
I
. The question is what should the right hand side be equal to. To
answer that question we go back to daily experience, when two systems
are in thermal contact, they become equilibrated when the temperature
of the two system become the same. Because of this we conclude
S
II
E
II
[
E

II
=EE

I
=
1
T
II
. (2.27)
Since system II is arbitrary (it does not have to be an ideal gas) we
conclude that
1
T
=
S
E
holds for any system.
12
2.3 Energy and volume exchange contact
between two systems and the relation
between entropy and pressure
If the contact between two system I and II allows an exchange of energy
and volume in such a way that the combined system visit all states with
a xed E
I
+ E
II
, V
I
+ V
II
with equal frequency.
To nd the most probable partition of E and V, we need to maximize
S
I
(E
I
, V
I
) + S
II
(E E
I
, V V
I
) (2.28)
The most probable E
I
and V
I
satisfy

E
I
S
I
(E
I
, V
I
) + S
II
(E E
I
, V V
I
) = 0

V
I
S
I
(E
I
, V
I
) + S
II
(E E
I
, V V
I
) = 0 (2.29)
We already know the rst equation implies T
I
= T
I
I. How about the
second equation it reduces to

V
I
S
I
(E
I
, V
I
) =

V
II
S
II
(E
II
, V
II
). (2.30)
Suppose system I is an ideal gas, then we know the left hand side is
equal to p
I
/T
I
. Because when two systems are in mechanical contact
they become equilibrated when the pressure the is same. Hence we
conclude the left hand side is equal to p
II
/T
II
(recall that T
II
= T
I
).
Thus
S
V
= p/T is also true in general.
2.4 Energy and particle exchange contact
between two systems and the relation
between entropy and chemical poten-
tial
If the contact between two system I and II allows an exchange of parti-
cles as well as energy in such a way that the combined system visit all
13
product states with a xed E
I
+E
II
and N
I
+N
II
with equal frequency,
then the combined system satisfy the dynamics of microcanonical en-
semble.
The principle of equal weight says that each one of the
(E, N)E = E
N

N
I
=0
_
E
I
E
dE
I

I
(E
I
, N
I
)
II
(E E
I
, N N
I
)
(2.31)
is equally likely to be realized. The probability for system I to have
energy between E
I
and E
I
+ dE
I
and particle number N
I
given by

I
(E
I
, N
I
)
II
(E E
I
, N N
II
)dE
I
E
(E)E
(2.32)
The most probable value of E
I
and N
I
maximize the numerator of the
above equation or equivalently maximize
S
I
(E
I
, N
I
) + S
II
(E E
I
, N N
I
). (2.33)
The solution E

I
and N

I
satisfy
T
I
(E

I
, N

I
) = T
II
(E

II
, N

II
)

N
I
S
I
(E

I
, N

I
) =

N
II
S
II
(E

II
, N

II
)
where E

I
+ E

II
= E and N

I
+ N

II
= N. (2.34)
If system I is an ideal gas it turns out the left hand side of the second
equation is
I
/T
I
. (For the ideal gas

T
= k
B
_
3
2
ln
E
N
+ ln
V
N
+ C
1

5
2
_
. (2.35)
.) Because from empirical facts we know when to system can exchange
particle, equilibrium will be reached when the chemical potential be-
come equal. Hence we conclude S/N = /T is generally true.
14
2.5 The rst law of thermodynamics
From previous discussion we have
S
EV,N
=
1
T
S
V E,N
=
P
T
S
NE,V
=

T
. (2.36)
Now we are ready to ask what change will E E + dE, V V + dV
and N N + dN induce upon S. The answer is
dS =
S
EV,N
dE +
S
V E,N
dV +
S
NE,V
dN. (2.37)
Use Eq. (2.36) we can rewrite the above equation as
dE = TdS pdV + dN. (2.38)
This is the rst law of thermodynamics. (This is slightly more general
than you used to see where dN is assumed to be zero. In that case
dE = TdS pdV.)
2.6 The second law of thermodynamics
This law is interpreted by a probabilistic concept in statistical mechan-
ics. When a certain inhibition (for example a dividing wall between two
gas chambers) is removed the entropy of the whole system will almost
certainly increase. This is because the whole system is an isolated sys-
tem and the principle of equal weight apply. Increasing entropy means
increasing the probability of realizing certain macrostate. When the
number of particles is large, the most probable macrostate has a dom-
inant probability to be realized then others.
15
2.7 The third law of thermodynamics
Entropy is given by
S = k
B
ln W (2.39)
where W is the number of microstate consistent with the macroscopic
condition of the system. Because W 1 the entropy is always non-
negative. In thermodynamics when we speak of zero entropy what we
really mean is that entropy is not of order N or V , i.e., not extensive.
2
For a quantum system this requires the ground state degeneracy D be
negligible compared with #
N
or #
V
. Under that condition as T 0
the system goes into its ground states and S 0.
However, in nature they are systems which do show zero-point en-
tropy (at least to the lowest temperature we have studied). A famous
example is ice. Ice is a molecular crystal made of H
2
O molecules. In
an ice crystal the oxygen form a diamond lattice, and the hydrogen
atoms sit between the oxygen atoms on the link of the diamond lat-
tice. However they do not sit at the center of the link. Because of the
stoichiometry, H
2
O, for the four links stemming from each oxygen, two
of them have a hydrogen atom sitting closer to the oxygen in question
and the hydrogen on the other two links sit closer to other oxygens. It
turns out that for such a lattice of N oxygens there are approximately
1.56
N
distinct hydrogen distribution patterns. For each pattern, each
oxygen has two closer hydrogens and two farther hydrogens. This gives
rise to a residual entropy Nk
B
ln(1.56). Interestingly down to the low-
est temperature, this entropy persists indicating that the degeneracy is
not lifted.
2.8 The canonical ensemble
Come back to the situation where two systems are allowed to exchange
energy only. When system II in the previous subsection is much big-
2
At non-zero temperature the entropy is extensive. The zero-point entropy can
be determined by extrapolating the nite temperature result to T = 0. Because
of the necessity of extrapolation, a non-extensive zero point entropy would requires
the determination of nite temperature entropy to accuracy O(1/N), which is of
course almost impossible.
16
A system emersed in a heat bath.
Figure 2.2:
ger than system I we call it an energy reservoir, or in short heat
bath(Fig. (4.1)).
Let l be a microstate of system I and
II
(E) be the density of
states of the heat bath. According to the principle of equal weight the
probability of realizing state l, P
l
, is proportional to

II
(E
t
E
l
)E = e
S
II
(E
t
E
l
)
k
B
e
S
II
(E
t
)
k
B
e

E
l
k
B
S
II
E

E
t
= e
S
II
(E
t
)
k
B
e

E
l
k
B
T
.
(2.40)
In the above E
t
is the total energy of the combined system, T =
(
E
S)[
1
Et
, and S
II
is the entropy of the heat bath. Since the heat bath
is very large compared to the system under consideration E
t
>> E
l
.
As the result we can expand S
II
(E
t
E
l
).
Therefore
P
l
e
E
l
/k
B
T
, or P
l
= constant e
E
l
/k
B
T
. (2.41)
Because the probability must sum to unity when all possibilities are
accounted for, we have

l
constant e
E
l
/k
B
T
= 1, (2.42)
which implies
constant
1
Z =

l
e
E
l
/k
B
T
. (2.43)
17
For the ideal gas
Z =
1
N!h
3N
_
d
3
r
1
...d
3
p
N
e
(
p
2
1
2m
+...+
p
2
N
2m
)
=
V
N
N!h
3N
_

2
m
_
3N
(2.44)
From now on, we abbreviate
1
k
B
T
. (2.45)
The normalization constant Z of the probability distribution is called
the partition function. The ensemble dened by the probability dis-
tribution Eqs (2.41 and 2.43) is called the canonical ensemble.
2.9 The partition function Z and the
Helmholtz free energy F
The canonical ensemble partition function can be rewritten in terms of
the density of states
Z(T, V, N) =
_

0
dE(E, V, N)e
E
. (2.46)
By the denition of statistical entropy we can write the above equation
as
Z(T, V, N) =
_

0
dEe
S(E,N,V )/k
B
e
E
=
_

0
dEe

ETS(E,N,V )
k
B
T
e

TS(E

,N,V )
k
B
T
_

0
dEe

2
E
S(E

,N,V )(EE

)
2
2k
B
= e

TS(E

,N,V )
k
B
T
_

0
dEe
K(EE

)
2
2k
B
= e

TS(E

,N,V )
k
B
T
_
2k
B
/K. (2.47)
Here K =
2
E
S(E

, N, V ). In the above E

is the solution of

E
S(E, N, V )[
E
=
1
T
. (2.48)
18
Thus to leading order in N
k
B
T ln Z(T, V, N) = E

TS(E

, V, N) F(T, V, N). (2.49)


The function F is obtained by performing the following transform on
the entropy function
F(T, V, N) [E TS(E, N, V )]
E the solution of
1
T
=
E
S
(2.50)
In thermodynamics this is precisely how we get the Helmholtz free
energy F from the entropy. Again what we do is to rst solve for
1
T
=
S(E, V, N)
E
. (2.51)
The right hand side is a function of E,V,N. Solve this equation implicitly
for E we will obtain E as a function of T,V,N. Once this function is
determined we replace every E in E-TS(E,V,N) by this function. As
the result we have
F(T, V, N) = E

(T, V, N) TS(E

(T, V, N), V, N). (2.52)


Let us practice doing what describe above for the ideal gas. Here
we know, from Eq. (2.13) that
S(E, V, N) = k
B
N
_
C
1
+ ln
V
N
+
3
2
ln
E
N
_
. (2.53)
On the other hand from Eq. (2.44) we have
k
B
T ln Z = k
B
T ln
_
V
N
N!h
3N
_

2
m
_
3N
_
Nk
B
T[C
2
+ ln
V
N
+
3
2
ln (k
B
T)]. (2.54)
Here C
2
+3/2 = C
1
+
3
2
ln
3
2
(recall thatC
1
=
5
2
+
3
2
ln
_
4
3
_
3 lnh+
3
2
ln m.)
Now let see whether we can reach Eq. (2.54) by doing the transfor-
mation discussed earlier. First we set up the equation
1
T
=
S
E
. Simple
computation show this equation is
1
T
=
3
2
Nk
B
E
. (2.55)
19
Thus E

=
3
2
Nk
B
T. Now we replace E every where in ETS(E, V, N)
by
3N
2
k
B
T the result is
E TS(E, V, N)
3N
2
k
B
T Nk
B
T
_
C
1
+ ln
V
N
+
3
2
ln
3
2
k
B
T
_
=
3N
2
k
B
T Nk
B
T
_
C
1
+
3
2
ln
3
2
+ ln
V
N
+
3
2
ln(k
B
T)
_
=
3N
2
k
B
T Nk
B
T
_
C
2
+
3
2
+ ln
V
N
+
3
2
ln(k
B
T)
_
= Nk
B
T[C
2
+ ln
V
N
+
3
2
ln (k
B
T)] (2.56)
which coincide with the expression for F.
Now an interesting question arise. What change will T t +
dT, V V + dV and N N + dN induce upon F? Let dE

=
E

(T + dT, v + dV, N + dN) E

(T, V, N) we have
dF = F

(T + dT, v + dV, N + dN) F

(T, V, N)
= dE

dTS(E

(T, V, N), V, N) T
S
E

dE

T
S
V
dV T
S
N
dN
(2.57)
The rst and third terms on the right hand side cancel on account of
S
E

= 1/T. Thus we have


dF = SdT pdV + dN. (2.58)
So
S =
F
T V,N
p =
F
V T,N
=
F
NT,V
. (2.59)
However you might ask wouldnt the entropy so obtained be a func-
tion of T, N, V rather than a function of E, N, V ? To get back the
original E dependence of the entropy we note that

ln Z = E (2.60)
20
We can use this equation to solve T as a function of E and substitute
the result into the entropy expression obtained by dierentiating F.
2.9.1 Paramagnet
The paramagnet is a disordered collection of magnetic moments. The
simplest model describing paramagnet is
H = hm
0
n

i=1

i
. (2.61)
Here h is the applied magnet eld, and
i
= 1 describing whether
the moment align or anti-align with the external magnetic eld. The
kind of magnet in which the magnetic moments can only point along
or opposite to a xed direction is called an easy-axis magnet. The
partition function is given by
Z =

i
=1
e
hm
0

i
= [2 cosh(hm
0
)]
N
. (2.62)
The averaged magnetic moment is given by
M =

j

j
)e
hm
0

Z
=

j
=1

j
e
m
0
h
j

j
=1
e
m
0
h
j
= Nm
0
tanh(m
0
h). (2.63)
The magnetization is dened as
m =
M
N
= m
0
tanh(m
0
h). (2.64)
The magnetic susceptibility is given by
(h) =
h
m =
m
2
0
k
B
T
sech(hm
0
/k
B
T). (2.65)
The zero-eld susceptibility is therefore

0
=
m
2
0
k
B
T
. (2.66)
21
2.9.2 Maxwell distribution
Consider a gas of interacting molecules:
H =
N

i=1
p
2
i
2m
+
1
2

i,=j
v(r
i
r
j
). (2.67)
An important property of the classical partition function is that
Z =
1
N!h
3N
_
d
3
p
1
...
_
d
3
p
N
e

i
p
2
i
/2m
_
d
3
r
1
...d
3
r
N
e
(1/2)

i=j
V (r
ij
)
=
1
N!h
3N
Z
p
Z
r
. (2.68)
In the above Z
p
and Z
r
represents the momentum and space integral
respectively. The probability that the molecule has a velocity v
1
, ..., v
N
is given by
1
N!h
3N
e
(mv
2
1
+...+mv
2
N
)/2
Z
r
1
N!h
3N
Z
p
Z
r
=
e
(mv
2
1
+...+mv
2
N
)/2
Z
p
. (2.69)
The probability that the rst molecule will have velocity v is given by
_
d
3
p
2
...d
3
p
N
e
(mv
2
+...+mv
2
N
)/2
Z
p
=
e
mv
2
/2
_
d
3
p
1
e
p
2
1
/2m
=
e
mv
2
/2
(2mk
B
T)
3/2
(2.70)
The same is true for molecule two,three...N. Thus the probability that
any molecule has velocity v is given by
P(v) =
e
mv
2
/2
(2mk
B
T)
3/2
. (2.71)
The probability distribution in Eq. (2.71) satistises
_
d
3
pP(v) = m
3
_
d
3
vP(v) = 1. (2.72)
22
Therefore it does not have the proper normalization for a velocity dis-
tribution function. Instead

P(v) = m
3
P(v) =
m
3
e
mv
2
/2
(2mk
B
T)
3/2
(2.73)
has the right normalization and is the velocity distribution function.
Eq. (2.73) is called the Maxwell velocity distribution. The Maxwell
distribution satises
_
d
3
v

P(v) = 1 (2.74)
The left hand side of the above equation can be expressed in terms of
polar coordinate as
_

0
4v
2

P(v) = 1 the integrand
4v
2
m
3
e
mv
2
/2
(2mk
B
T)
3/2
(2.75)
is called the Maxwell speed distribution.
Note that the fact that molecular distribution obeys the Maxwell
distribution is a consequence of classical statistical mechanics and is
independent of whether the molecules interact or not !
2.9.3 Equal-partition theorem
The equal-partition theorem is the consequence of the following mathe-
matical statement. Suppose a variable x obey the Gaussian probability
distribution, i.e.,
P(x) =
1
_
2
a
e

a
2
x
2
(2.76)
The expectation value of x
2
is
1
_
2
a
_

x
2
e

a
2
x
2
=
2
_
2
a

a
_

a
2
x
2
=
2
_
2
a

2
a
= 2a
1/2

a
a
1/2
=
1
a
(2.77)
23
This is the reason why the average energy of the ideal gas is
2
2
Nk
B
T.
This is because the momentum distribution of any gas particle is the
Maxwell distribution
P(p
x
, p
y
, p
z
) =
e
p
2
x
/(2mk
B
T)

2mk
B
T
e
p
2
y
/(2mk
B
T)

2mk
B
T
e
p
2
z
/(2mk
B
T)

2mk
B
T
. (2.78)
If we calculate
p
2
x
2m
we can use the math formula derived above and
replace a by
1
mk
B
T
. Thus

p
2
x
2m
=
1
2m
mk
B
T = 1/2k
B
T. (2.79)
Therefore

p
2
2m
=
p
2
x
2m
+
p
2
y
2m
+
p
2
z
2m
=
3
2
k
B
T. (2.80)
In fact the equal partition theorem state more generally that ev-
ery quadratic, independent degrees of freedom in the Hamiltonian con-
tribute
1
2
k
B
T (2.81)
to the average energy. The classical partition function read
Z =
1
A
_
dq
1
...dq
M
_
dp
1
...dp
M
e
H(q,p)
, (2.82)
where
1
A
is an overall factor that is independent of temperature. An
independent, quadratic freedom appear in the Hamiltonian as
H =
C
2

2
+ W(rest of the 2M1 variable). (2.83)
The partition function is therefore
Z =
1
A

2k
B
T
C
_

d rest of the variablese
W
. (2.84)
24
The average energy is given by

E =
1
A
_
d
C
2

2
_
d...e
H
Z
=
_
d
C
2

2
e
C
2
/2
_
de
C
2
/2
=
k
B
T
2
. (2.85)
The importance of the result is that it is independent of the spring
constant C, and for that matter any other details of the Hamiltonian!
Here is an application of the equal-partition theorem. When the
temperature is very high so that classical statistical mechanics applies,
the equal partition theorem predicts that, to the extent the atoms in
a solid can be treated as an elastic network, the contribution to the
average energy of a crystal due to the vibration of its constituent atoms
is equal to
3nNk
B
T. (2.86)
(Recall that each atom has 6 degrees of freedom, 3 from the momentum
and 3 from the position.) In Eq. (2.86) N is the number of unit cell,
and n is the number of atoms per unit cell. According to this prediction
the specic heat due to lattice vibration is equal to
3nk
B
. (2.87)
This called the Dulong-Petit law.
2.10 The grand canonical ensemble
Next we consider the situation where energy and particle exchange are
allowed. When the system II is much bigger than system I we say it is a
particle and energy reservoir. Let
II
(E
r
, N
r
) be the density of state of
the reservoir, and E
l
, N
l
are the energy and particle number associated
with a particular microstate of system I. According to the principle of
equal weight the probability of realizing state l, P
l
, is proportional to

II
(E
t
E
l
, N
t
N
l
)E

II
(E
t
E
l
, N
t
N
l
)E

II
(E
t
, N
t
)E
= e
S
II
(E
t
E
l
,N
t
N
l
)S
II
(E
t
,N
t
)
k
B
. (2.88)
25
In the above E
t
, N
t
are the total energy and total particle number of
the combined system and S
II
is the entropy of the reservoir. Since
the reservoir is very large compared to the system under consideration
E
t
>> E
l
and N
t
>> N
l
. As the result we can expand the exponent as
S
II
(E
t
E
l
, N
t
N
l
) S
II
(E
t
, N
t
) E
l

E
S
II
N
l

N
S
II
=
E
l
T
+
N
l
T
. (2.89)
Here we have used (
N
S
II
) = /T. As the result
P
N,l
=
1

e
(E
l
N)
, (2.90)
where

l,N
e
(E
l
N)
. (2.91)
The ensemble dened by the probability distribution Eqs (2.90 and
2.91) is called the grand canonical ensemble.
2.11 The grand partition function and
thermodynamic function
The grand partition function is given by
(T, V, ) =

N
e
N
_
dE(E)e
E
=

N
e
N
_
dEe
(ETS(E,V,N))
e
(E

TS(E

,V,N

)N

)
. (2.92)
Here E

and N

are the solutions of

E
S(E, V, N) =
1
T
and =
N
F(T, V, N). (2.93)
Thus
J(T, V, ) = k
B
T ln (2.94)
26
is the thermodynamic function obtained from the Legendre transform
J(T, V, ) F(T, V, N) N[
replace N by the soution of =
N
F
(2.95)
2.11.1 Molecular adsorption -The hydrogen stor-
age problem
The usage of hydrogen as future fuel has been a subject of much dis-
cussion in recent years. One of the key road block for is the storage
problem. The DOE target hydrogen storage for year 2015 is 9% in
weight. Since hydrogen is the lightest element, in order to achieve this
target it is necessary for the storage media to have the following prop-
erties: (1) high surface area, (2) appropriate adsorbing binding energy
so that for each explored surface the coverage 1.
This problem can be analyzed using what we have already learned.
Consider an adsorbent surface having N sites each of which can adsorb
one gas molecule. In the following we shall model the H
2
gas as an ideal
gas.
3
We treat the H
2
gas as the energy and particle reservoir that set
the chemical potential and the temperature for the adsorbates. By
denition the reservoirs energy and particle number is far greater than
the substrate. The temperature pressure and the chemical potential
of the ideal gas exerts on to the absorbed molecules can be computed
from
1
T
=
S
E
,
p
T
=
S
V
,

T
=
S
N
. (The pressure of the ideal gas is
controlled by a piston as shown in Fig. (2.3)). These dierentiations
gives
1
T
=
3Nk
B
2E
,
p
T
=
Nk
b
V
, and Eq. (2.35), respectively. Assuming that
an adsorbed molecule has energy
0
compared to one in the free state,
the goal is to determine the coverage as a function of P.
According to Eq. (2.35) the chemical potential exerted by an ideal gas
is given by

T
= k
B
_
3
2
ln
E
N
+ ln
V
N
+ C
1

5
2
_
. (2.96)
3
H
2
is not an ideal gas because the molecule has internal rotation and vibration
degrees of freedom. Lets ignore such complications.
27
Figure 2.3:
If we replace E by
3
2
NK
B
T,
V
N
by
k
B
T
P
and insert the value for C
1
and
combine the sum of log into the log of the product we obtain

k
B
T
= ln
_
P
k
B
T
_
h
2
2mk
B
T
_
3/2
_
.
(2.97)
This is the chemical potential the ideal gas exerted onto the molecules
that are absorbed on the substrate.
The statistical mechanics of the adsorbed molecules is described
by the grand canonical ensemble with chemical potential given by
Eq. (2.97). For n adsorbed molecules distributed among N adsorb-
ing sites there are
N!
n!(N n)!
(2.98)
dierent adsorbing patterns. Each of them is a microstate of the
adsorbed system. The grand partition function of the substrate is
=
N

n=0
N!
n!(N n)!
e
n(
0
+)
=
_
1 + e
(
0
+)
_
N
. (2.99)
The probability of n molecules being adsorbed is
e
n(+
0
) N!
n!(Nn)!

. (2.100)
28
Thus the average number of adsorbed molecule is
n =
1

n=0
ne
n(+
0
)
N!
n!(N n)!
=
1

ln =
N
e
(
0
+)
+ 1
.(2.101)
Since is the chemical potential of the particle reservoir which, in this
case, is the ideal gas, we have, according to Eq. (2.97)
e

=
p
k
B
T
_
h
2
2mk
B
T
_
3/2
. (2.102)
Substitute this result into Eq. (2.101) we obtain
=
n
N
=
p
p + p
0
(T)
(2.103)
where
p
0
(T) =
_
2mk
B
T
h
2
_
3/2
e

0
/k
B
T
k
B
T. (2.104)
When p = p
0
(T) in the above equation the coverage is = 1/2. Put
in the number for the H
2
mass, planck constant, Boltzmann constant
..etc we obtain
p
0
(T) = 115091 atm
_
T
T
room
_
e

0
/k
B
T
. (2.105)
The material science question is what material we use for the sub-
strate. In our language this translate into what kind of
0
we need.
There are two constraints. (1) We would like to achieve considerable
coverage at around room temperature and around ambient pressure.
(2) We would like the substrate to release a substantial fraction of the
adsorbate when the temperature is heated from room temperature to
say 100
o
C. Lets say we would like
(T
room
, 2atm) = f
(1.4T
room
, 2atm) =
2
3
f (2.106)
29
0.0 0.2 0.4 0.6 0.8 1.0
Release Fraction 0.0
0.2
0.4
0.6
0.8
1.0
Coverage
0.0 0.2 0.4 0.6 0.8 1.0
Release Fraction 0.0
0.2
0.4
0.6
0.8
1.0

0
eV
Figure 2.4:
0
and f as a function of the release fraction.
These conditions translate into
2
2 + 115091e

0
/26
= f
2
2 + 115091 1.4 e

0
/(1.426)
=
2
3
f. (2.107)
Here I have used the fact that k
B
T
room
26meV and have expressed
0
in unit of meV. We would like to solve the above equations for f and
0
.
It turns out that
0
= 0.4361eV and f = 0.997 solve the above equation.
Thus we need to search for a material which can bind H
2
molecule at
the binding energy of 0.436 eV. The above result was obtained for the
release fraction 1 2/3 = 1/3. For dierent release fractions the result
for
0
and f are slightly dierent. In Fig. (2.4) I plot
0
and f as a
function of the release fraction.
Thus a high surface area material which binds the hydrogen
molecule at binding energy in the range of 0.3-0.5 eV is what we are
looking for.
30
Chapter 3
Black body radiation
Quantum statistical mechanics works exactly the same as its classi-
cal counterpart. The biggest dierence is the specication of the mi-
crostate. In most of the cases we shall encounter, each microstate is
the energy eigenstate of a Hamiltonian. For example the canonical
ensemble partition function of a quantum system is given by
Z =

k
e
E
k
(3.1)
Here k labels the eigenstate and E
k
is the eigen energy of the state.
The grand partition function is given by
=

k
e
(E
k
(N)N)
(3.2)
where E
k
(N) is the eigen energy of state k when the particle number
is N. The relation between the partition function and the free energy
are exactly the same as we talked about before.
If you have not taken introduction to quantum mechanics it is the
right time for you to read the suggested materials in bspace.
31
3.1 Statistical mechanics of harmonic os-
cillators
A harmonic oscillator is governed by the following Hamiltonian
H =
p
2
2m
+
K
2
x
2
. (3.3)
Classically the oscillation frequency is given by
_
K/m. The classical
equation of motion (the Newton equation) is given by

2
x
t
2
+
2
0
x = 0 (3.4)
where
0
=
_
K/m is the oscillation frequency. According to the equal
partition theorem the average kinetic and potential energy each con-
tributes k
B
T/2 to the total energy, hence
H = k
B
T. (3.5)
For a collection of independent harmonic oscillators (each with a dif-
ferent frequency) the hamiltonian is given by
H =
N

i=1
p
2
i
2m
i
+
K
i
2
x
2
i
. (3.6)
The important thing about the equal partition theorem is that regard-
less of the values of m
i
and K
i
each modes contribute k
B
T to the
total internal energy, as the result the total energy of N modes of har-
monic oscillator is Nk
B
T. The specic heat due to these oscillators
is
C =
E
T
= Nk
B
. (3.7)
32
3.2 Quantum statistical mechanics of har-
monic oscillator
According to the quantum theory of harmonic oscillator, each mi-
crostate is labeled by a non-negative integer n.
1
The energy associated
with such microstate is given by
E
n
= (n +
1
2
) h
0
(3.8)
where
0
=
_
K/m is the classical oscillation frequency. Physically n
is the number of oscillation quanta present in the nth excited state.
Each oscillation quantum carries energy h
0
hence n quanta carry en-
ergy n h
0
. The ground state energy h
0
/2 is the energy due to the
zero point oscillation. The partition function of a quantum harmonic
oscillator is given by
Z =

n=0
e
(n+1/2) h
0
= e
h
0
/2

n=0
e
nh
0
= e
h
0
/2
1
1 e
h
0
.(3.9)
In arriving at the nal result we have used the formula for geometric
series

n=0
x
n
=
1
1 x
(3.10)
for [x[ < 1. The average number of oscillation quantum is given by
n =

n
ne
(n+1/2)h
0
)
Z
=

n
ne
nh
0

n
e
nh
0
=

( h
0
)
ln
_
1
1 e
h
0
_
=
e
h
0
1 e
h
0
=
1
e
h
0
1
. (3.11)
The total energy associated with such an oscillator is given by
E = (n + 1/2) h
0
= h
0
/2 +
h
0
e
h
0
1
. (3.12)
1
I will ask the TA to review the quantum mechanics of harmonic oscillator in
the discussion session.
33
For k
B
T >> h the above result is approximate by
E h
0
/2 + k
B
T. (3.13)
Aside from the zero point energy the second term, namely the energy
due to the thermal excitation, is the same as the equal partition theorem
prediction. For a collection of N harmonic oscillators Eq. (3.12) is
modied to
E =

i
_
h
i
/2 +
h
i
e
h
i
1
_
, (3.14)
where
i
is the oscillation frequency of the ith oscillator.
3.3 Classical description of the E-M eld
In the absence of charge and current density, the magnetic eld of the
EM oscillation satises Maxwells wave equation

2
B
t
2
c
2

2
B = 0. (3.15)
In a cavity with linear dimension LLL the magnetic eld satises
the boundary condition B = 0 on the walls. In that case B can be
expanded in terms of normal modes
B =

n
1
,n
2
,n
3
B
(n
1
,n
2
,n
3
)
sin(n
1
x/L) sin(n
2
y/L) sin(n
3
z/L) (3.16)
where n
1,2,3
are positive integers. (The negation of n
1,2,3
reverses the
sign of the sine function hence does not give rise to a new mode).
In the following we shall introduce a vector
k =

L
(n
1
, n
2
, n
3
) (3.17)
to replace the triplet (n
1
, n
2
, n
3
). After so doing Eq. (3.16) becomes
B =

k
B
k
sin(k
x
x) sin(k
y
y) sin(k
z
z) (3.18)
34
where k
x
, k
y
, k
z
denotes the three component of k. In the three dimen-
sional space spanned by k
x
, k
y
, k
z
the allowed k lies in the rst octant.
They form a simple cubic grid with lattice constant /L. Note that
the origin in this k-space is excluded because the corresponding B
k
is zero. Substituting Eq. (3.18) into Eq. (3.15) we obtain

2
B
k
t
2
+ c
2
k
2
B
k
= 0. (3.19)
Eq. (3.19) looks exactly the same as Eq. (3.4), namely, the equation
of motion of a simple harmonic oscillator. Moreover since there is a
normal mode for each k we have a collection of independent simple
harmonic oscillators. An important dierence between Eq. (3.19) and
Eq. (3.4) is that the coordinates of the normal mode in Eq. (3.19)
is a vector not a scalar. However we could write down Eq. (3.19) for
each component of B
k
. In doping so we convert Eq. (3.19) into three
equations, one for each component of B
k
. However do B
k
really have
three independent components? The answer is no. To see that we recall
the magnetic eld obeys B = 0 (there is no magnetic monopole).
Substitute Eq. (3.16) into this equation we obtain
k B
k
= 0. (3.20)
hence for each k the only allowed B
k
components are transverse to
k. Therefore for each k there are two independent components of B
k
.
Another thing to note is the oscillation frequency implied by Eq. (3.19)
is

k
= c[k[. (3.21)
According to the equal partition theorem the total thermal energy of
the cavity EM eld is
E = 2

k
k
B
T (3.22)
here the prefactor 2 is due to the two independent direction of B
k
for
each k. Since there are innitely many allowed k, the above result
suggests the total internal energy of an cavity EM eld is innite!
35
3.4 Quantum statistical mechanics of the
cavity E-M eld
Quantum mechanically the total energy (omitting the zero point oscil-
lation energy) associated with the cavity EM eld is given by
E = 2

k
n
k
h(k) (3.23)
where
n
k
=
1
e
ck
1
. (3.24)
When L is very large, the allowed k points form a very dense grid.
Under the condition that n
k
does not change appreciably if we move k
from a lattice point to its nearest neighbor, we can replace the sum in
Eq. (4.8) by an integral
E = 2
_
d
3
k
(/L)
3
h
k
e
h
k
1
(3.25)
Here we need to remember that the k integral is over the rst octant and
the reason we divide by (/L)
3
is because for every (/L)
3
in k-space
there is one k point. Because the integrand only depends on the modu-
lus of k we can change d
3
k to the radial integral
4
8
k
2
dk (4/8 because
we only integrate over the rst octant). In addition since k = / hc we
can change the radial k-integral into an integral over the frequency.
E =
_

0
dD()
h
e
h
1
. (3.26)
Here the photon density of states can be determined as follows
D()d = 2
4
8
_

c
_
2
d
c
(/L)
3
. (3.27)
Therefore
D() = L
3

2
c
3

2
. (3.28)
36
Substitute this result into Eq. (3.29) we obtain
E = V
_

0
d
_
h
3
c
3

2
_
1
e
h
1
, (3.29)
where V = L
3
is the total volume. The energy density is given by
U =
E
V
=
_

0
d
_
h
3
c
3

2
_
1
e
h
1
. (3.30)
The integrand of Eq. (3.30) is referred to as the blackbody spectral
function in the literature:
(, T) =
_
h
3
c
3

2
_
1
e
h
1
. (3.31)
This universal function describes all emission spectrum of cavity EM
eld. By changing the integration variable to y = h and performing
the following dimensionless integral
_

0
dx
x
3
e
x
1
=

4
15
(3.32)
we obtain the energy density of blackbody radiation
U =
_
8k
4
B

5
15c
2
h
3
_
T
4
. (3.33)
This is called the Stefan law.
3.5 Acoustic phonon contribution to the
specic heat
The quanta of lattice vibration is called phonon. A solid is made up of
atoms arranging in a periodic array. The unit cell is a group of atoms
which, upon periodic repetition, generate the whole solid. The acoustic
vibration is a dynamic displacement in which atoms in the same unit
cell displace by the same amount in the same direction. The optical
vibration is the type of dynamic displacement where atoms in the same
unit cell are displaced by dierent amount and/or dierent direction.
37
Like EM eld, the vibration in a solid is characterized by wavevector
satisfying k =

L
(n
1
, n
2
, n
2
). The associated displacement is given by
u(R, t) =

k
u
k
(t) sin(k
x
R
x
) sin(k
y
R
y
) sin(k
z
R
z
), (3.34)
where R is the position of atoms in the periodic arrangement. The
Fourier coecient u
k
(t) satises the equation of motion of harmonic
oscillators. However unlike EM eld there is no no constraint on the
direction of u
k
(such as k u
k
= 0). As the result there are three
independent directions for u
k
, two perpendicular to k (called transverse
modes) and one parallel to k (call the longitudinal modes).
The dispersion of acoustic phonon is very similar to that of photon.
(k) = c
l
k for longitudinal modes
= c
t
k for transverse modes. (3.35)
Here c
l
and c
t
are the longitudinal and transverse phonon velocities.
Following Eq. (3.28) the number of phonon modes in frequency range
+ d is given by
D()d = V
_
1
c
3
l
+
2
c
3
t
_

2
2
2
d. (3.36)
Here the factor of 2(1) multiplying
1
c
3
t
(
1
c
3
l
) reects the fact that for each
wavevector there are two (one) independent transverse (longitudinal)
modes. There is however a constraint on the total number of normal
modes. For a solid with volume V there are N = V/a
3
unit cells (a is
the lattice constant). In each unit cell the atoms can displace in three
dierent directions, hence there are in total N = 3N modes. Debye
incorporates this constraint by imposing a maximum frequency
D
so
that
D() = V
_
1
c
3
l
+
2
c
3
t
_

2
2
2
for < h
D
= 0 for > h
D
. (3.37)
He chooses
D
so that
_

D
0
dD() = N V
_
1
c
3
l
+
2
c
3
t
_

3
D
6
2
= 3N. (3.38)
38
In terms of
D
Eq. (3.37) read
D() =
9N
2

3
D
for <
D
= 0 for >
D
. (3.39)
Using Eq. (3.39) we compute the average energy to be

E =
_

D
0
9N
2
h
3

3
D
h
e
h
1
d. (3.40)
At temperature where
h
D
k
B
T
>> 1 (3.41)
Eq. (3.40) gives

E
3k
4
B

4
5 h
3

3
D
NT
4
. (3.42)
The heat capacity derived from that is
C =
T
E =
12k
4
B

4
5 h
3

3
D
NT
3
. (3.43)
The T
3
specic heat is often taken as the hallmark of acoustic phonons.
39
40
Chapter 4
Ideal Fermi gas
4.1 Particle in a box
We focus on spin 1/2 Fermi particles conned in a L L L box of
volume V = L
3
. This time, however, we practice a dierent boundary
condition - the periodic boundary condition:
(x + L, y, z) = (x, y, y), (x, y + L, z) = (x, y, y),
(x, y, z + L) = (x, y, y). (4.1)
Under this boundary condition the eigenfunction of a particle in the
box is given by

k
=
1

L
3
e
ikr
(4.2)
where
k =
2
L
(n
1
, n
2
, n
3
), (4.3)
where n
1,2,3
can take all integer values. The associated eigen energy is

k
=
h
2
[k[
2
2m
. (4.4)
Thus in the three dimensional space spanned by the allowed k is a
simple cubic grid with lattice constant 2/L.
41
4.2 Grand canonical ensemble treatment
of Fermi particle in a box
First we focus on the grand canonical ensemble. The quantum states
are characterized by
[n
k
(4.5)
where n
k
is the fermion occupation number of the k, state. Paulis
exclusion principle tells us that each n
k
is either 1 or zero.
The partition function is given by
=

k,
1

n
k
=0
e
((k))n
k
=

k,
_
1 + e
((k))
_
. (4.6)
Here = 1/2 is the z-component of the spin of the fermion. The
average number of particle n
k
is given by
n
k
=
1

n
p}
n
k
e
((p))np
=
1

1
n
k
=0
e
((k))n
k
_
_

1
n
k
=0
e
((k))n
k
_
=
1

_
1 + e
((k))
_
_
1 + e
((k))
_
=
e
((k))
_
1 + e
((k))
_
=
1
e
((k))
+ 1
(4.7)
Eq. (5.6) is called the Fermi Dirac distribution. Note that because the
energy in Eq. (??) does not depend on the spin, n
k
is independent of
.
4.3 The particle number constraint and
the chemical potential
If the total number of particle is xed (which correspond to the usual
situation of a solid) the following constraint must be satised
N = 2

k
1
e
((k))
+ 1
(4.8)
42
Here the k sum is performed the simple cubic lattice (n
1
, n
2
, n
3
)2/L.
Here the overall factor of two comes form the fact that every fermion
has two spin states. The equality in Eq. (4.8) is made possible by
adjusting the chemical potential . In fact Eq. (4.8) determines the
chemical potential as a function of temperature. When L is very large,
the k point is very dense. Under the condition that n
k
does not change
appreciably if we move k from a lattice point to its nearest neighbor,
we can replace the sum in Eq. (4.8) by an integral
N = 2
_
d
3
k
(2/L)
3
1
e
((k))
+ 1
(4.9)
We divide by (2/L)
3
is because for every (2/L)
3
in k-space there is
one k point.
N = 2L
3
_
d
3
k
(2)
3
1
e
((k))
+ 1
=
L
3

2
_

0
k
2
dk
1
e
((k))
+ 1
=
_

0
dD()
1
e
()
+ 1
. (4.10)
Here D() is the density of states, i.e., the number of k state between
energy and + d, of the free Fermi gas:
D()d =
L
3

2
_

2m
h
_
2
_
m
2
d
h
= 2L
3
d
m
3/2

2 h
3

2
. (4.11)
and
N = 2L
3
m
3/2

2 h
3

2
_

0
d

e
()
+ 1
(4.12)
In the zero temperature limit the left hand side of the above equation
becomes
N/L
3
= = 2
m
3/2

2 h
3

2
_

0
d

(4.13)
43
The solution is

F
=
h
2
2m
(9
4

2
)
1/3
. (4.14)
This energy is often referred to as the Fermi energy. At zero tem-
perature, all free fermion states are occupied below this energy, while
empty above this energy. In terms of
F
and the number of particle N
the density of state is
D() =
3N
2
F

F
, (4.15)
or
D() = D(
F
)

F
. (4.16)
For low but non-zero temperature, the chemical potential deviates
from the Fermi energy. In this case the formula that determines the
chemical potential is given by
= 2
m
3/2

2 h
3

2
_

0
d

e
()
+ 1
, (4.17)
or equivalently
2
3

3/2
F
=
_

0
d

e
()
+ 1
. (4.18)
For low temperatures it is desirable to nd analytic expression for
as a function of temperature. In order to do that let us rst consider
how to evaluate the following integral
_

0
df()

(), where f() =


1
e
()
+ 1
, and (0) = 0. (4.19)
In the case of Eq. (??) () =
2
3

3/2
. Using integration by parts
_

0
df()

() =
_

0
()

f(). (4.20)
44
Figure 4.1:
f

versus .
45
The advantages of Eq. (4.20) is that

f() is a sharply peaked function


(about = ) at low temperatures:
f

=

[e
()
+ 1][e
()
+ 1]
. (4.21)
This suggests that in evaluating the right hand side of Eq. (4.20)
we should expand around = :
() = () + ( )

() +
( )
2
2

() + ... (4.22)
Because Eq. (4.21) is an even function of , all odd powers in
Eq. (4.22) drop out after integration. For even powers we have
_

0
d

[e
()
+ 1][e
()
+ 1]
( )
2n
(k
B
T)
2n
_

dx
x
2n
[e
x
+ 1][e
x
+ 1]
= (k
B
T)
2n
2 (2n)!(1 2
2n+1
)(2n).
(4.23)
Here we have used the fact that
_

dx
x
m
(e
x
+ 1)(e
x
+ 1)
= 2m!(1
2
2
m
)(m). (4.24)
For 2n = 0, 2, 4, 6 the above result
is 1, (k
B
T)
2

2
/3, (k
B
T)
4
7
4
/15, (k
B
T)
6
31
6
/21. Use the above
result in Eq. (4.20) we obtain
2
3

3/2
F
=
_

0

f() =
2
3
_

0

3/2

f()
=
2
3
_

3/2
+
3
8

2
3
(k
B
T)
2
+
3
128
5/2
7
4
15
(k
B
T)
4
+ ..
_
.
(4.25)
Solve the above equation for we obtain
=
F


2
12
(k
B
T)
2


4
80
(k
B
T)
4

3
F
+ ... (4.26)
46
4.4 The total internal energy and the spe-
cic heat
The average energy is given by

E =
_

0
dD()

e
()
+ 1
. (4.27)
The specic heat of the free electron gas at constant volume and con-
stant particle number is dened as
C
V,N
=
T
E(T, V, N) =
1
k
B
T
2
_

0
dD()
( + T

)
[e
()
+ 1][e
()
+ 1]
=
1
k
B
T
2
D(
F
)

F
_

0
d

5/2
( + T

)
[e
()
+ 1][e
()
+ 1]
(4.28)
We can change the integration variable to z = ( )
C
V,N
=
1
T
D(
F
)

F
_

dz
(k
B
Tz + )
5/2
(k
B
Tz + T

)
[e
z
+ 1][e
z
+ 1]

1
T
D(
F
)

F
_

dz
(k
B
Tz + )
5/2
(k
B
Tz + T

)
[e
z
+ 1][e
z
+ 1]

1
T
D(
F
)

F
_

dz
[
5/2
+
5
2

3/2
k
B
Tz + ...](k
B
Tz + T

)
[e
z
+ 1][e
z
+ 1]

1
T
D(
F
)

F
_

dz
T
5/2
F

+
5
2

3/2
F
(k
B
T)
2
z
2
[e
z
+ 1][e
z
+ 1]
(4.29)
Recall that

2
6
k
2
B
T

F
and
_

dx
[e
z
+ 1][e
z
+ 1]
= 1,
_

dx
x
2
[e
z
+ 1][e
z
+ 1]
=
2
/3 (4.30)
Eq. (4.29)reduces to
C
V,N
=
2
3
k
2
B

2
D[
F
]T. (4.31)
47
Using the result in Eq. (4.11) we obtain
C
V,N
= Nk
B
k
B
T
2
F
. (4.32)
Eq. (4.31) is an important formula for experimental condensed matter
physics, because by measuring the heat capacity we can deduce the
density of states at the Fermi energy via
C
V,N
T
=
2
3
k
2
B

2
D[
F
]. (4.33)
4.5 The magnetic susceptibility
The magnetic susceptibility of the free Fermi gas is also an impor-
tant quantity to study experimentally. In the presence of an applied
magnetic eld, the spin up and spin down energy undergoes opposite
Zeeman shift, namely,
For spin up (k) (k)
B
h
For spin down (k) (k) +
B
h (4.34)
In the above
B
is the magnetic moment of each electron. In this case,
the average number of spin up and spin down electrons are no longer
identical.
N
+
=
1
2
_

0
dD()
1
e
(
B
h)
+ 1
N

=
1
2
_

0
dD()
1
e
(+
B
h)
+ 1
. (4.35)
The new constraint on the chemical potential is given by
N =
1
2
_

0
dD()
_
1
e
(
B
h)
+ 1
+
1
e
(+
B
h)
+ 1
_
(4.36)
48
When the magnetic eld is very weak, to linear order in h is the same
as Eq. (4.26). The magnetic moment induced by the applied eld is
given by
M =
B
(N
+
N

) =

B
2
_

0
dD()
_
1
e
(
B
h)
+ 1

1
e
(+
B
h)
+ 1
_
.
(4.37)
For weak applied eld, we can expand the dierence in the integrand
in power of the applied eld
M =
B

B
h
_

0
dD()

f()
=
2
B
h
_

0
dD()

f() =
2
B
h
_

0
dD()

[e
()
+ 1][e
()
+ 1]
.
(4.38)
As usual we expand D() around :
D() = D(
F
)

F
= D(
F
)
_

F
+
_

F
2
( )
_

F
8
2
( )
2
+ ..
_
.
(4.39)
Therefore
M =
2
B
h
_

0
dD(
F
)
_

F
8
2
( )
2
_

[e
()
+ 1][e
()
+ 1]
=
2
B
hD(
F
)
_

dx
_

F
8
2
(k
B
T)
2
x
2
_
1
[e
x
+ 1][e
x
+ 1]
=
2
B
hD(
F
)
_

F
8
2
(k
B
T)
2

2
3
_
hD(
F
)
2
B
_
1

2
24
(k
B
T)
2

2
F
_
. (4.40)
The magnetic susceptibility is dened as
=
M
h
[
h=0
= D(
F
)
2
B
_
1

2
24
(k
B
T)
2

2
F
_
. (4.41)
49
Figure 4.2: Subrahmanyan Chandrasekhar
Combining Eq. (4.33) and Eq. (6.6) we see that as T 0 the ratio
C
N,V
/T

=
2
3
k
2
B

2
B
(4.42)
approaches a universal constant.
4.6 The physics of while dwarfs (optional)
The white dwarfs are cold stars stabilized against gravitational collapse
by the quantum pressure of their electrons. Let the number of nucleons
(protons and neutrons) be N,and the number of electron per nucleon
be q (q 1/2). If R is the radius of the star, the total gravitational
energy is given by the following estimate.
The total mass of the star is MN where M is the nucleon mass.
(We have omitted the electron mass.) The mass density of the star is
given by
M
=
MN
4R
3
/3
. The gravitation energy is given by
E
g
= G
_
R
0

M
4r
3
3

M
4r
2
dr
r
=
3GM
2
N
2
5R
. (4.43)
The total kinetic energy of the electron is
E
K
=
_

0
dD()

e
()
+ 1
. (4.44)
50
Let us assume that we can treat the electron as non-relativistic and
regard them as at T = 0 (we will come back to see whether that is a
good approximation). Under that approximation =
F
where

F
=
h
2
2m
_
9
4
_
qN
4R
3
/3
_
2
_
1/3
=
3 h
2
_
3
2
N
2

2
q
2
_
1/3
4mR
2
, (4.45)
and
E
K
=
_

F
0
D() = 2
_

F
0
3qN
4
F

F
=
3Nq
F
5
=
9(
3
2
)
1/3

2/3
h
2
N
5/3
q
5/3
20mR
2
. (4.46)
Therefore the total energy of the star is given by
E
tot
(R) =
3GM
2
N
2
5R
+
9(
3
2
)
1/3

2/3
h
2
N
5/3
q
5/3
20mR
2
. (4.47)
The optimal radius satises

R
E
tot
(R) = 0. (4.48)
The solution is
R =
_
9
4
_
2/3 h
2
q
5/3
GmM
2
N
1/3
(4.49)
Put in the numbers (G = 6.67 10
11
m
3
s
2
kg
1
) we obtain
R =
7.29 10
25
m
N
1/3
(4.50)
The solar mass is 1.99 10
30
kg which amounts to N = 1.19 10
57
.
This means
R =
6881
(M/M

)
1/3
Km. (4.51)
A white dwarf whose mass is one solar mass has electron density
4.4 10
29
/cm
3
The Fermi energy associated with that is

F
2.1 10
5
eV. (4.52)
51
The electron rest mass is mc
2
5.110
5
eV therefore the Fermi energy
is about 2/5 of the electron rest mass, hence the electrons in the white
dwarf are quite relativistic. In general a white dwarf of mass M
s
has
Fermi energy

F

2mc
2
5
_
M
s
M

_
4/3
. (4.53)
The Fermi temperature of the solar-mass white dwarf is T
F

2.4 10
9
K. The temperature of the white dwarf is about 10
5
K hence
compared to its Fermi temperature it is justiably almost at absolute
zero.
For star whose mass is much bigger than the solar mass, the Fermi
energy given by Eq. (4.53) can become much larger than the rest mass
of the electron. In that case, we should use the ultra-relativistic limit
to compute the average kinetic energy as a function of radius. The
relativistic kinetic energy
(k) hck (4.54)
which give the density of states
D() = 2V

2
2c
3
h
3

2
, (4.55)
the same as that for the photon gas. With the density of states given
by Eq. (4.55) we recompute the Fermi energy in the ultra-relativistic
limit and nd it to be
N
e
= V

3
F
3c
3
h
3

2

F
= hc(3
2

e
)
1/3
. (4.56)
The average kinetic energy is given by
E
K
=
_

F
0
dD() =
3N
e
4
hc(3
2

e
)
1/3
. (4.57)
Substitute N
e
= qN and
e
= qN/(4R
3
/3) into the above equation
we obtain
E
K
=
3 hc
4
_
9N
4
q
4
4
_
1/3
R
. (4.58)
52
Combined with the gravitational energy it yields
E
tot
=
3GM
2
N
2
5R
+
3 hc
4
_
9N
4
q
4
4
_
1/3
R
. (4.59)
We note that the two terms in the above equation have the same radius
dependence. Thus if
3GM
2
N
2
5R
>
3 hc
4
_
9N
4
q
4
4
_
1/3
R
, (4.60)
or
N > 2.0 10
57
M
s
> 1.7M

(4.61)
the gravitational attraction will overwhelm the electron quantum pres-
sure, and the star will gravitationally collapse despite the electron quan-
tum pressure.
When the density of the star get so high that inverse beta decay
e

+ P
+
n + (4.62)
occurs, at that stage almost all the protons and electrons are converted
into neutrons and neutrinos. The neutrino will escape and carry energy
away. Eventually the degeneracy pressure of the neutron ght against
the collapse. In that case we can do sa similar calculation as in the
non-relativistic limit of the electron pressure to get
R =
_
9
4
_
2/3 h
2
GM
3
N
1/3
. (4.63)
(note that compared with Eq. (4.49) we see that m M and q 1.)
This yield, for a star twice the solar mass a radius
R 9.7Km. (4.64)
Using Eq. (4.45) (with q 1 and m M) we can compute the Fermi
energy of such star. The answer is

F
141MeV
1
9
Mc
2
. (4.65)
53
54
Chapter 5
Ideal Bose gas
5.1 The grand canonical ensemble treat-
ment
We focus on spin zero Bose particles in a box of volume V = L
3
, again
under periodic boundary condition. Like the Fermi case the eigenfunc-
tion of any single particle is given by

k
=
1

L
3
e
ikr
(5.1)
where
k =
2
L
(n
1
, n
2
, n
3
), (5.2)
where n
1,2,3
take all integer values. The associated eigen energy is

k
=
h
2
[k[
2
2m
. (5.3)
All of the above eigen information are the same as the Fermi case.
Again the quantum states are characterized by
[n
k
(5.4)
where n
k
is the boson occupation number of the k state.
55
Figure 5.1: Satyendra Nath Bose
We focus on the grand canonical ensemble. The partition function
is given by
=

n
k
=0
e
(k)n
k
+n
k
=

k
_
1 e
((k))
_
1
. (5.5)
The average number of particle n
k
is given by
n
k
=
1

n
p}
n
k
e
((p))np
=
1

n
k
=0
e
((k))n
k
_
_

n
k
=0
e
((k))n
k
_
=
1

_
1 + e
((k))
_
1
_
1 + e
((k))
_
1
=
1
e
((k))
1
(5.6)
Eq. (5.6) is called the Bose Einstein distribution. Because the average
particle number is positive denite for all n
k
therefore
0. (5.7)
5.2 The particle number constraint
The particle number constraint requires
N =

k
n
k
=

k
1
e
(
k
)1
. (5.8)
56
5.2.1 zero temperature macroscopic occupation of
the k = 0 state
Let us start with the zero temperature limit. In this case all the bosons
will sit in the lowest energy ,.i.e.,the k = 0 state. In other words
n
k=0
=N. In order for this to be consistent with the Bose distribution
function we must have
lim

1
e
[[
1
= N. (5.9)
Here we have used the fact that 0. Eq. (5.9) implies that e
[[
=
1 +
1
N
or [[ = ln(1 +
1
N
) =
1
N
. In the last step we used the fact that
we are interested in the thermodynamic limit where N . Solving
the above equation we obtain
[[ =
1
N
. (5.10)
Clearly as and N both approach innity 0. Thus at zero tem-
perature = 0 (this result holds even when N is nite.)
5.2.2 Non-zero temperature Bose Einstein con-
densation
Because the chemical potential can be zero the Bose distribution func-
tion
n
k
=
1
e
(
k
)
1
. (5.11)
can be singular at
k
= 0. The fact that the summand of Eq. (5.8)
can be singular for k = 0 call for a more careful look at the previous
procession from

k
to
_
d
3
k
(2/L)
3
. In fact this replacement is good for all
k except k = 0 Thus we write

k
n
k
= n
0
+

k,[k[>0
n
k
= n
0
+
_
[k[>0
d
3
k
(2/L)
3
n
k
. (5.12)
57
For the second term of the above equation we can proceed as usual and
convert
_
[k[>0
d
3
k
(2/L)
3
n
k
to
_

0
+
dD()
1
e
()
1
. (5.13)
As a result Eq. (5.8) read
N = n
0
+
_

0
+
dD()
1
e
()
1
. (5.14)
Here D() is the density of state, identical to that of the free Fermi gas
D() = L
3
m
3/2

2 h
3

2
. (5.15)
The maximum number of particle that can be accommodated by
the second term of Eq. (5.14) is obtained when we set = 0 in that
term
N

(T) =
_

0
+
dD()
1
e

1
= D
0
_

0
+
d

1
= D
0
(k
B
T)
3/2
_

0
+
dx

x
e
x
1
= D
0
(k
B
T)
3/2

2
(3/2). (5.16)
In the above
D
0
= L
3
m
3/2

2 h
3

2
, (5.17)
and
_

0

x
e
x
1
=

2
(3/2) 2.31516. (5.18)
The density corresponds to N

is

(T) =
m
3/2

2 h
3

2
(k
B
T)
3/2

2
(3/2). (5.19)
58
For a system with a xed number of particle, hence an xed density
, there exists a temperature T
c
such

(T
c
) = . (5.20)
For temperature lower T
c
the total number of particles accommodated
by the k ,= 0 states can no longer account for all the particles. When
that happens a xed fraction of the particles have to be accommodated
by the k = 0 state, or n
0
. The phenomenon that at low temperatures
a innite number of particles all sits in the same state is called Bose-
Einstein condensation. Substitute Eq. (5.19) into Eq. (5.20) we obtain
m
3/2

2 h
3

2
(k
B
T
c
)
3/2

2
(3/2) = , (5.21)
which implies
T
c
=
2 h
2

2/3
k
B
m(3/2)
2/3
. (5.22)
The thermal De Broglie wavelength is dened as
(T) =
h

2mk
B
T
. (5.23)
At the critical temperature

3
(T
c
) = (3/2) 2.61248. (5.24)
If we dene interquartile distance d as
d
3
= 1, (5.25)
then Eq. (5.24) implies
(T
c
)
d
= (3/2)
1/3
1.38. (5.26)
59
For T < T
c
the condensate fraction
f =
n
0
N
=

(T)

(T
c
)

(T)

(T
c
)
= 1
T
3/2
T
3/2
C
. (5.27)
As T T

c
the condensate fraction vanishes as
f(T)
3
2
T
c
T
T
c
. (5.28)
For temperature below T
c
the chemical potential is pinned at zero.
The internal energy is given by
E =
_

0
dD()

e

1
= D
0
_

0
d

3/2
e

1
= D
0
(k
B
T)
5/2
_

0
dx
x
3/2
e
x
1
= D
0
(k
B
T)
5/2
3

4
(5/2)
. (5.29)
The specic heat is given by
C
V,N
=
15D
0

(5/2)
8
k
5/2
B
T
3/2
. (5.30)
The value of (5/2) is approximately 1.34.
5.3 The amazing work of indistinguisha-
bility (optional)
In the above discussion we have seen that for temperature below T
c
the
number of bosons in the ground states become of order N. However
despite the minute energy dierence E =
h
2
2m
(
2
L
)
2
, the number of
atom in the rst excited state is proportional to
1
e
E
1

k
B
T
E
=
k
B
T
h
2
2m
(
2
L
)
2
=
k
B
T
h
2
2m
(
2
d
)
2
N
2/3
. (5.31)
60
As the result the ratio between the number of ground state versus the
rst excited state particle is proportional to
N
2/3
N
0 as N . (5.32)
This is pretty amazing given the fact that as L the thermal energy
far exceed E.
To better appreciate the above discussion, let consider a simper
problem with only two energy states, one has energy zero and the other
energy > 0. First, let consider N distinguishable particle case. The
partition function is
Z =
N

n=0
N!
n!(N n)!
e
n
. (5.33)
Here n is the number of particle in the upper energy states. The above
is a binomial series and can be easily summed to give
Z =
_
1 + e

_
N
. (5.34)
Next, we compute the average number of particle in the upper state
n

N
n=0
n
N!
n!(Nn)!
e
n
Z
=
1

ln Z. (5.35)
The result is
n

= N
1
e

+ 1
. (5.36)
As the result the ratio between the number of atoms in the two states
are given by
N

N
0
=
N
1
e

+1
N N
1
e

+1
= e

, (5.37)
as expected from classical statistical mechanics.
Now let us consider the particles to be indistinguishable bosons. In
this case the partition function is given by
Z =
N

n=0
e
n
. (5.38)
61
Note that the factor
N!
n!(Nn)!
in Eq. (5.33) is missing. This is because
the particle are indistinguishable, i.e., we can not count dierent per-
mutation of the particle (with the same total number of particle in the
upper and lower states) as dierent state. Of course Eq. (5.38) can also
be summed easily to give
Z =
1 e
(N+1)
1 e


1
1 e

as N . (5.39)
The total number of particle in the upper state is then
n

N
n=0
ne
n
Z
=
1

ln Z =
1
e

1
. (5.40)
Note that for all nite temperature T > 0 the number of particles in
the excited state is of order 1 not N. Hence
N

N
0
=
1
e

1
N
1
e

1
0 as N . (5.41)
This striking contrast with the distinguishable particle result is the
act of indistinguishability. More precisely, in the distinguishable par-
ticle case the Boltzmann weight unfavor the occupation of the excited
state. However the
N!
n!(Nn)!
favors it. As the balance between the en-
ergy and entropy factors, the number of particles in the excited state
remains of order N. For bosons, the entropy factor is entirely removed!
62
Chapter 6
Phase transitions
This chapter is about phase transitions, in particular, continuous phase
transitions. The behaviors of various physical measurable quantities
near the critical point of a phase transition are called critical phenom-
ena. The task we are facing is to understand critical phenomena within
the framework of statistical mechanics.
The reason for being interested in phase transitions is because the
concepts we develop in understanding them are widely applicable in
numerous other areas of physics. Among them the concept of renor-
malization group is particularly important.
Renormalization group is not a mathematical group, rather it is a
strategy. In specic, it is a strategy that is useful when we face a system
with innite (or a huge number) degrees of freedom. Huge number of
degrees of freedom does not necessarily makes a problem dicult. The
diculty only arises when these degrees of freedom interact with one
another. For concreteness, let us imagine each degree of freedom is a
bit, or rather, a spin that can point in either the up or down directions.
The renormalization group strategy can be simply stated as follows.
We rst solve the problem involves a few spins. We realize that due
to the spin-spin interaction, the low energy states often involve the
spins in question act collectively together. To be more precise, we seek
for a description of the low energy states of these few spins in terms
of the behavior of a single eective degree of freedom. We then use
the original microscopic Hamiltonian to gure out how do the eective
spins interact. Next, we solve a few eective spin problem and describes
63
the low energy states in terms of a new eective spins. Clearly this
procedure can be continued forever. Some times, if we are lucky, the
interaction between the eective spins are the same as that between the
original spins, except the strength of the interaction is modied. Under
such condition, the iterated spin eective spin procedure described
above traces out a trajectory in the space panned by the parameters
describing various types of interaction in the Hamiltonian. Tremendous
insight can be gained by just analyzing the trajectory as a function of
iterations. For instance by knowing the limiting behavior after a large
number of iterations, we often can make statement about the state of
the system. Furthermore by nding xed points at which the trajectory
bifurcate, and analyze the ow near them we can extract information
about phase transitions and critical phenomena.
There are numerous examples of phase transitions. Some hap-
pen at nite temperature (e.g., ferromagnetic and antiferromagnetic
transition, superconducting phase transition, liquid-gas critical point,
order-disorder transition of adsorbed atoms, etc.) and some happen
at zero temperature (e.g. the metal-insulator transition, superconduc-
tor or superuid-insulator transition, quantum magnetic phase transi-
tion, etc). The latter class of transitions are called quantum phase
transitions. In this chapter we shall concentrate on classical phase
transitions where the transition happens at non-zero temperature.
Let me now give an example of continuous phase transition and
discuss its associated critical phenomena. Consider the easy-axis ferro-
magnet Y FeO
3
. This material has two magnetic phases: the param-
agnetic phase and the ferromagnetic phase. In the paramagnetic phase
the local magnetic moments arrange in a random fashion such that if we
average over them there is no net magnetization. In the ferromagnetic
phase, on the other hand, the local moments align with one another so
that a net magnetization results.
When we study a magnetic system experimentally, temperature T is
an important control parameter. Experimentally it is found that there
exists a critical temperature T
c
, above which the system is paramagnetic
and below which it is ferromagnetic. The behavior of the system in the
neighborhood of T
c
is called the critical phenomena, which we shall
discuss in the following.
64
6.1 Example: critical phenomena of an
easy-axis ferromagnet
The specic heat C
h
diverges as T T
c
. The precise fashion by which
such divergence occurs is
C
h
A
+
(
T T
c
T
c
)

; T > T
c
A

(
T
c
T
T
c
)

; T < T
c
. (6.1)
Experimentally 0.11. The number is called a critical exponent.
In the following we shall dene
t
T T
c
T
c
. (6.2)
The magnetization m is dened as the average magnetic moment
per unit cell. Above T
c
, m = 0. Below T
c
the system develops non-zero
magnetization. However, the direction of such spontaneous magneti-
zation is undetermined until an innitesimal external magnetic eld is
applied. If we dene the spontaneous magnetization m
0
as
m
0
(T) = lim
h0
m(T, h), (6.3)
we nd
m
0
(T) sign(h)B(t)

; T < T
c
m
0
(T) = 0; T > T
c
. (6.4)
For easy-axis ferromagnet 0.32.
The magnetic susceptibility records the response of the system to
an externally applied magnetic eld:
(T)
m(T, h)
h
. (6.5)
Experimentally it is found that as T T
c
the zero-eld susceptibility
diverges according to
(T) C
+
t

; T > T
c
C

(t)

; T < T
c
. (6.6)
65
The value of is 1.24.
From Eq. (6.6) we know that the derivative of m(T
c
, h) is innite at
h = 0. In fact experimentally it is found
m(T
c
, h) D
+
h
1/
: h > 0
D

(h)
1/
: h < 0. (6.7)
Since m/h diverges as h 0, we conclude > 1. The experimental
value for is 4.8.
Experimentally using neutral diraction we can determine the equal
time correlation function between two local moments
G(r; T) m(0)m(r). (6.8)
It is found that
G(r; T) =
E
+
r
d2+
e
r/(T)
; T > T
c
= m
0
(T)
2
+
E

r
d2+
e
r/(T)
: T < T
c
. (6.9)
This dene the critical exponent . The value of is 0.05.
Finally the correlation length dened in Eq. (6.9) also diverges at
T
c
. The singular behavior follows
(T) F
+
t

; T > T
c
F
+
(t)

; T < T
c
. (6.10)
The value of is 0.65.
So we have

0.11 0.32 1.24 4.80 0.65 0.05
6.2 Universality
The reason that we are interested in the above critical exponents is
because they are universal. In other words the numerical value for
66
... for any easy-axis ferromagnet that exhibit a Curie transition is
the same. What is even more surprising is that the same set of numbers
describe the critical behavior near the liquid-gas critical point of, say,
CO
2
, as long as the following translation is made:
liquid gas critical point : P T
c
ferromagnet paramagnet critical point : h T mm
0
.
(6.11)
Apparently the critical exponents for the easy-plane and isotropic
ferromagnets dier from those of the easy-axis ones. Therefore the no-
tion of universality only exists in the weak sense. The purpose of our
subsequent lectures is to discuss the origin of such universality and ex-
plore dierent universality classes.
6.3 Scaling
Experimentally it is found that
C
h
(T, h) = [t[

S
c,
(h[t[
y
)
m(T, h) = [t[

S
m,
(h[t[
y
)
(T, h) = [t[

S
,
(h[t[
y
)
G(r, T, h) =
1
r
d2+
S
G,
(r[t[

; h[t[
y
), (6.12)
where y = , and S
C,
are scaling functions whose form depend on
whether t > 0 or t < 0.
6.4 A simple model for easy-axis ferro-
magnet: the Ising model
Consider the following Hamiltonian
1 = J

ij)
S
i
S
j
, (6.13)
67
where S
i
= 1,J > 0, i, j run through the sites of a simple cubic lattice,
and ij denotes the nearest neighboring lattice sites. In subsequent
discussions we shall be interested in understanding the predictions of
the statistical mechanics of the above model (the Ising model).
In statistical mechanics we study the expectation values of various
physical quantities. To be more precise, a physical quantity O is a
function of the S
i
s, i.e. O[S
i
]. In the above and for the rest of the
course we shall us the symbol O[S
i
] to denote that O depends on all
the S
i
s.
The expectation value of O is dened as
O

[S
i
]
O[S
i
]e
7
:
, (6.14)
where = 1/k
B
T, and
:

[S
i
]
e
7
. (6.15)
For example when we want to compute the average magnetization we
have
O =
1
N

i
S
i
, (6.16)
where N is the total number of lattice sites (of course we shall be
interested in the thermodynamic limit where N ). As another
example when we want to compute the average spin-spin correlation
function we have
O(r) =
1
N

i
S
i
S
i+r
. (6.17)
In the presence of external magnetic eld Eq. (6.13) becomes
1 J

ij)
S
i
S
j
h

i
S
i
. (6.18)
Thus
m =
1
N

i
S
i
=
1
N
Tr[S
i
e
7
]
:(T, h)
=
1
N
:
h
1
:
=
1
N
ln :
h
=
f
h
. (6.19)
68
In the above Tr is short for

[S
i
]
, and
f(t, h)
k
B
T
N
ln : (6.20)
is the Helmholtz free energy density. Applying one more h-derivative
to Eq. (6.19) we obtain
=
m
h
=

2
f
h
2
=
1
N

2
:/h
2
:
(
:/h
:
)
2

=

N

i,j

Tr[S
i
S
j
e
7
]
:(T, h)

Tr[S
i
e
7
]
:(T, h)
Tr[S
j
e
7
]
:(T, h)

=

N

i,j
[S
i
S
j
S
i
S
j
]. (6.21)
Eq. (6.21) is the famous uctuation-dissipation theorem.
Now let us get some feeling about the the Ising model. Clearly as
T 0 the spins S
i
will align to minimize the energy. The ground
state has two-fold degeneracy, i.e., spins all up or spins all down. We
notice that for Eq. (6.13) S
i
S
i
(global spin reversal) is a symmetry
of the Hamiltonian. However, global spin reversal does not leave the
ground state invariant. When such happens, the physical state has
less symmetry than the Hamiltonian we say that there is spontaneous
symmetry breaking.
Clearly the spins will become disordered as we raise the tempera-
ture. In particular when k
B
T >> J the interaction between neighbor-
ing spins can not compete with the entropic eect of disordering. As
the result we expect a paramagnetic to ferromagnetic phase transition
as the temperature is lowered.
Clearly it is extremely naive to expect a simple model like Eq. (6.13)
to describe the behavior of a real material (say Y FeO
3
). The reason
that such model description has any value at all is because the univer-
sality, i.e., it is plausible that the critical behavior of Eq. (6.13) (i.e.
the predicted values of critical exponents) will be the same as that of
real systems.
69
Unfortunately it is very dicult to compute quantities such as
Eq. (6.15), Eq. (6.19) and Eq. (6.21). The root of the diculty lies
in the fact that in Eq. (6.18) spins on dierent sites interact with each
other. Were it for such interaction : and f can be computed exactly.
Indeed setting J = 0 in Eq. (6.18) simple calculations give
: = [2 cosh h]
N
f = k
B
T ln [2 cosh h]. (6.22)
Indeed, the interaction between innite degrees of freedom is at the
heart of many-body problems.
6.5 The predictions of mean-eld theory
6.5.1 The spontaneous magnetization, the expo-
nent
In the absence of exact solution for Eq. (6.13) in d = 3 (or in d = 2 in
the presence of non-zero h), we have to resort to approximate methods.
Among various such methods the most important and the most physical
one is the mean-eld theory. The key intuition behind mean-eld theory
is the following. Through the spin-spin interaction, the neighbors of a
given spin exert an eective magnetic eld on it. This eective magnetic
eld polarize the spin so that it points in the same direction as the
average magnetization of its neighbors. Formally we replace Eq. (6.13)
by the following Hamiltonian
1
mf
=

i
h
eff,i
S
i
, (6.23)
where
h
eff,i
= h + J

j
C
ij
S
j
. (6.24)
In the above C
ij
= 1 if i, j are nearest neighbor and = 0 otherwise. To
determine S
j
we require the following self-consistent condition to be
70
met:
S
i
=
Tr[S
i
e
7
mf
]
Tr[e
7
mf
]
= tanh(h
eff,i
)
= tanh(h + J

j
C
ij
S
j
). (6.25)
In the case where the external magnetic eld is uniform, we expect S
i

to be independent of i. In that case Eq. (6.25) becomes quite simple:


S = tanh (h + JzS), (6.26)
where z is the number of nearest neighbors of a site, or the coordination
number. For the hypercubic lattice in d dimension z = 2d.
A simple way to solve Eq. (6.26) is by graphical means. Let us
rst concentrate on the case where h = 0. If we plot the straight line
y = x and the curve y = tanh (Jzx), it is simple to see that when
Jz < 1 the two graphs intersect each other only when x = 0, while
when Jz > 1 they intersect at three points. The temperature
T
c

Jz
k
B
, (6.27)
at which the transition between the previous two cases occurs, marks
the mean-eld phase transition. For T > T
c
the only solution of
Eq. (6.26) is S = 0, while for T < T
c
two extra solutions S = m
0
become allowed. We shall later prove that under the latter condition
the free energy is lower when S = m
0
. Since as T T

c
m
0
0, we
can expand tanhJzS on the right hand side of Eq. (6.26) in power
series of JzS when T T
c
S JzS
1
3
(JzS)
3
+ .... (6.28)
Thus m
0
satises
1 =
T
c
T

1
3
(
T
c
T
)
3
m
2
0
+ O(m
4
0
) + ..., (6.29)
71
or
m
0

_
3
T
3
T
3
c
(
T
c
T
1). (6.30)
Thus as T T

c
m
0
[t[
1/2
, (6.31)
which implies
=
1
2
. (6.32)
6.5.2 The exponent
When T = T
c
but h > 0 Eq. (6.26) read
S = tanh [S +
c
h], (6.33)
where
c

1
k
B
Tc
. If we again expand the right hand side of the above
equation in power series we obtain
S = S +
c
h
1
3
(S +
c
h)
3
+ ..., (6.34)
or

c
h =
1
3
(S +
c
h)
3
. (6.35)
Eq. (6.35) implies that as h 0
S h
1/3
, (6.36)
hence
= 3. (6.37)
72
6.5.3 The exponent
For the more general case where T is slightly higher than T
c
and h <<
1, Eq. (6.26) reduces to
S (h + JzS)
1
3
[(h + JzS)]
3
+ .... (6.38)
Thus for (h + JzS) << 1 we have
S =
h
k
B
(T T
c
)
. (6.39)
Consequently
= 1. (6.40)
6.5.4 The exponent
To compute the free energy f in mean-eld theory we have to be more
careful in writing down Eq. (6.23). The subtlety lies in a constant term
that corrects the doubling counting of the energy due to the interaction
between average magnetizations:
1
mf
=

i
h
eff,i
S
i
+ J

ij)
S
i
S
j
. (6.41)
While the last term of Eq. (6.41) does not aect the calculation of the
average magnetization, it enters into the free energy.
The mean-eld prediction of the free energy read
f
mf
=
k
B
T
N
ln Tr[e
7
mf
]
=
J
N

ij)
S
i
S
j

k
B
T
N

i
ln [2 coshh
eff,i
]. (6.42)
In the case where the external magnetic eld is uniform Eq. (6.42)
reduces to
f
mf
=
Jz
2
S
2
k
B
T ln 2 cosh [(h + JzS)]. (6.43)
73
It is important to realize that Eq. (6.26) is equivalent to
f
S
= 0, (6.44)
i.e. the self-consistent mean-eld solution is the one that minimizes
the free energy in Eq. (6.43). First, we use Eq. (6.43) to prove that
for T < T
c
and h = 0 the free energy associated with the S = m
0
solution of Eq. (6.26) is lower than that of the S = 0 solution. To do
that we plot the function
f
mf
=
Jz
2
S
2
k
B
T ln 2 cosh[(JzS)] (6.45)
versus S. For small S we can expand the ln and the cosh in
Eq. (6.45). The rst three terms read
f
mf
= k
B
T ln 2 +
T
c
2T
(T T
c
)S
2
+
T
12
(
T
c
T
)
4
S
4
. (6.46)
First we notice that f consists of even terms only, thus f(S) =
f(S). Second, it is clear that for T < T
c
the global minima of f
occurs at S , = 0. By nding the solution to f/S[
S)=m
= 0, and
substitute the solution S = m back into Eq. (6.46), we obtain for
T T
c
:
f
mf
= k
B
T ln 2 ; T > T
c
= k
B
T ln 2 +
3
4
T
T
2
c
(T T
c
)
2
; T < T
c
. (6.47)
As the result we nd
C
h=0
= T

2
f
T
2
= 0 ; T > T
c
= k
B
T
T
2
c
[
9
2
T 3T
c
] ; T < T
c
. (6.48)
Therefore at T = T
c
the value of C
h=0
is
3k
B
2
, thus the specic heat
is discontinuous across the transition in mean eld theory. If we in-
sist on putting this behavior in the framework described by algebraic
singularity, we conclude
= 0. (6.49)
74
Note that Eq. (6.48) is only valid for T T
c
hence one should not take
T 0 in that equation and concludes C
h=0
is negative.
So far by using mean-eld approximation we have obtained

Mean-eld exponents 0
1
2
1 3
Because the calculations of and is slightly more involved, I do
not want to get into those except just tell you the mean-eld results.
The prediction of mean-eld theory for is 1/2 and for is zero.
6.5.5 Scaling
Not only does the mean-eld theory predicts the exponents, it also
predict scaling. For example, Eq. (6.38) can be rewritten as
h S(1 Jz) +
1
3
(JzS)
3
+ ..., (6.50)
or
h k
B
T
c
S[t +
1
3
1
(1 + t)
2
S
2
]
k
B
T
c
S[t +
1
3
S
2
]. (6.51)
Therefore
h
[t[
3/2
k
B
T
c
S
[t[
1/2
[1 +
1
3
(
S
[t[
1/2
)
2
]. (6.52)
Eq. (6.52) can be inverted to yield
S
[t[
1/2
= S

(
h
[t[
3/2
). (6.53)
Similar scaling relations can be obtained for C
h
, , and even G(r)
within the mean-eld theory.
75
6.6 The Landau theory of phase transi-
tions
We see, in Eq. (6.46), that when the order parameter is small, we can
expand the free energy density as a power series in the order parameter,
i.e,
F =
_
d
d
xf
f =
A
2

2
+
B
4

4
+ .... (6.54)
In mean-eld theory we ignore the spatial dependence of the order
parameter, and the equilibrium value of the order parameter satises
f

= 0. (6.55)
In a continuous phase transition as we have discussed so far, the
order parameter is non-zero below the critical temperature and zero
above, and, more importantly, it vanishes continuously as T T

c
.
The root of this phenomenon is the fact that as T T
c
A = (T T
c
) (6.56)
as clearly demonstrated by Eq. (6.46), and B stays positive denite. It
turns out that these two assumption immediately implies all the critical
phenomena and scaling discussed above.
For example, using Eq. (6.56) in Eq. (6.1) and taking the derivatives
with respect to gives
(T T
c
) + B
3
= 0. (6.57)
The solution of the above equation is
=

(T
c
T)
B
T < T
c
= 0 T > T
c
. (6.58)
76
This implies = 1/2. For the specic heat we have
C = T

2
f
T
2
=
T
2
2B
, T < T
c
= 0 T > T
c
. (6.59)
In the presence of an applied magnetic eld the Landau free energy
includes an extra term
f =
A
2

2
+
B
4

4
+ .... h. (6.60)
Given Eq. (6.60) it is simple to derive the critical exponents and .
Here we show that Eq. (6.60) also implies scaling, i.e.,
(T, h) = [t[
1/2
S

_
h
[t[
3/2
_
. (6.61)
For T > T
c
, Eq. (6.61) is obviously satised with S
+
= 0. What is left
to show is the validity of Eq. (6.61) for T < T
c
. Using Eq. (6.56) the
equation of state read
(T
c
T) + B
3
= h. (6.62)
Divide the whole equation by [t[
3/2
, we have
h
[t[
3/2
= B
_

_
[t[
_
3
T
c

_
[t[
. (6.63)
The whole equation exhibit scaling which implies that its solution has
the form given by Eq. (6.61).
In the above we arrive at Eq. (6.1) by explicitly compute the mean-
eld free energy density of a specic problem. Lev Landaus great in-
sight is that Eq. (6.1) should be valid in general near a continuous phase
transition as long as is the appropriate order parameter. One might
wonder that why are the terms with odd power missing in Eq. (6.1).
The answer is that they are forbidden by symmetry. In the absence
of an applied magnetic eld, the free energy should be invariant under
. As a result, the odd power terms must vanish. Thus when
77
Figure 6.1: The evolution of the Landau free energy across a continuous
phase transition.
you try to analyze a continuous phase transition, you should (1) iden-
tify all symmetries of the problem, (2) identify the order parameter,
(3) gure out how does the order parameter transform under symme-
try operations, and (4) write down a free energy expansion consistent
with the symmetry. After all the above are done, let the coecient of
the quadratic term be proportional to T T
c
.
6.6.1 First order phase transition
A rst order phase transition diers from a continuous one in that the
order parameter discontinuously jump to nite value as the temperature
is lowered below T
c
. The evolution of the Landau free energy f across a
continuous phase transition is shown in Fig. (6.1). In the neighborhood
of T
c
, the order parameter is small, it is always permissible to expand
the free energy as in Eq. (6.1)
For a rst order transition, the free energy evolution is shown in
Fig. (6.2). In this case the global free energy minima changes from
occurring at zero order parameter to non-zero order parameter at T
c
.
78
Figure 6.2: The evolution of the Landau free energy across a rst order
phase transition.
In this case, the order parameter immediately below T
c
is not necessarily
small. An example of the free energy function that exhibits a rst order
behavior is given by
f =
A(T)
2

2

[B[
4

4
+
[C[
6

6
. (6.64)
Note that the coecient in front of
4
is negative. If A(T) monotoni-
cally decreases with T, a rst order transition occurs when
A(T) =
3
16
B
2
C
. (6.65)
Should we be fully satised with the mean-eld theory and declare
victory in the understanding of critical phenomena? The answer is
negative. The reason that we are not fully satised with the mean-
eld theory, is because it fails to predict the dependence of the critical
exponents on the space dimension d and the number of components of
the magnetic moments n(see table below).
79
Figure 6.3: Ken Wilson
(n,d)
(1,1) - - - - - -
(1,2) 0 1/8 7/4 15 1 1/4
(1,3) 0.11 0.32 1.24 4.80 0.65 0.05
(2,1) - - - - - -
(2,2) - - 15 1/4
(2,3) 0.02 0.34 1.32 4.7 0.66 0
(3,1) - - - - - -
(3,2) - - - - - -
(3,3) 0.14 0.36 1.37 4.6 0.7 0
The improvement beyond mean-eld theory took many decades to
achieve. It nally was accomplished by Wilsons renormalization group
treatment of critical phenomena. This however will take an advanced
course (212) to discuss.
80

Vous aimerez peut-être aussi