Physics 262

Topic 2: Foundations of Quantum Statistical Physics

Erel Levine
Sep 2015

1. QUANTUM-MECHANICAL ENSEMBLE THEORY: THE DENSITY MATRIX

 
Assume an ensemble of N systems of N particles. Let  k denote the (normalized) wavefunction characterizing the
state of system k at time t. Use a complete orthogonal set |ni to write
 k  X

 X k
 (t) =
|ni n| k (t)
an (t) |ni .
(1.1)
n

The wavefunctions akn evolve according to Schrodingers

equation,
X
i~a kn (t) =
Hnm akm (t) .
m


2
The number akn (t) represents the probability that a measurement at time t finds the kth system of the ensemble to
be in the particular state |ni. Clearly, we must have
X

akn (t)2 = 1

(for all k).

Define now the density operator

N
1 X k
am (t)ak
n (t)
N

mn (t) =

(1.2)

k=1

The quantity nn (t) represents the probability that a system, chosen at random from the ensemble is found to be in
the state |ni at time t,
nn (t) =

N
2
1 X 

n| k 
N

(1.3)

k=1

(both ensemble and QM averages), and of course

nn = Tr() = 1.

N


1 X
k
i~ a km (t)ak
k
n (t) + am (t)a
n (t)
N
k=1
"
!
!#
N
X
X
1 X
k
k
k
k
=
Hml al (t) an (t) am (t)
Hnl al (t)
N
k=1
l
l
X
=
[Hml ln ml Hln ) = (H H)mn

i~ mn =

i~ = [H, ]

(1.4)

2
We thus conclude that an equilibrium is diagonal in the energy representation. Otherwise is symmetric. Thermal
averages are calculated as
hGi =

N
1 X
k   k 
G
N
k=1
N

1 X X k k
an am Gnm
N
k=1 m,n
X
X
=
mn Gnm =
(G)mm
=

m,n

hGi = Tr(G)

(1.5)

 
[if the wavefunctions  k are not normalized, then hGi = Tr(G)/Tr()].

2. MICRO-CANONICAL ENSEMBLE

In the energy basis (where is diagonal) we postulate

nn = 1 En ,E0 .

(2.1)

Pure state - all systems are in the same state |i: = 1 and
= |i h| .
In another representation may not even be diagonal. However, it is straightforward to show that in any representation 2 = .
Example: Assume a two-state system with energy eigenstates |1i and |2i. In a pure state, all systems are in one state,
say state |1i. In the energy representation,


1 0
=
.
0 0
However if we choose an alternative representation |S i =

=

1
2
1
2

1
2
1
2
1
2

(|1i |2i) then


.

In this representation is not diagonal, but it is symmetrical and it obeys 2 = , as expected.

Mixed state - In this natural situation we must require
1. Equal a priory probability
 
2. Random a priory phases (equivalently: every  k is an incoherent superposition of the basis |ni.)
Then
mn =

N
N
D
E
k
k
1 X 2 i(km kn )
1 X k k
am an =
|a| e
= c ei(m n ) = cmn .
N
N
k=1

k=1

Example: assume that an energy level is doubly degenerate, with eigenstates |Ri and |Li. We construct the ensemble
by placing our systems in states
 k
 = sin |Ri + eik cos |Li .

3
Then
k =

cos2
eik cos sin
ik
2
e
cos sin
sin2


and

cos2
0
0
sin2

and therefore we must require cos2 = sin2 = 1 = 1/2 or = /4.

3. THE CANONICAL ENSEMBLE

Anticipating Pn exp(En ) the density matrix in the energy representation should take the form
1
eEn mn
ZN ()
X
X
1
1
eH
=
|ni
eEn hn| =
eH
|ni hn| =
ZN ()
ZN ()
Tr(eH )
n
n

mn =

(3.1)
(3.2)

In this ensemble the thermal quantum average is given by

hGiN =

Tr(GeH )
Tr(eH )

(3.3)

4. THE GRAND CANONICAL ENSEMBLE

Again using the analogous classical ensemble we can define the density operator

e(HN )

(4.1)

Tr(e(HN ) )

and the corresponding thermal (and quantum) average

Tr(Ge(HN ) )
=
hGi =
Z

hGiN ZN
NZ

N
N

N
P

(4.2)

5. SIMPLE EXAMPLES
5.1. Particle in a potential well

The problem of a single particle in a box will serve us soon as the initial step towards constructing a theory of
quantum ideal gases.
~2 2

2m
eikx
hx|ki =
V

H=

~2 k 2
E(k) =
2m




Z
X
~2 k 2
d3 k
~2 k 2
V
Z1 = Tr() =
exp
'V
exp

=
2m
(2)2
2m
vQ
k






Z
0
vQ
V d3 k eik(xx )
~2 k 2
1
m(x x0 )2
1
(x x0 )2
0
hx | |xi =
exp
= exp
= exp
V
(2)3
V
2m
V
2~2
V
vQ
1
P (x) = hx| |xi =
V

4
The transition from a sum to an integral corresponds to taking the classical limit. Note that the consistency with the
fundamental assumption of statistical mechanics.

5.2. Harmonic oscillator

One more important and recurring problem.



1

~
H= N+
2
X
X
e~/2
Tr(eH ) =
hn| eH |ni =
e~(n+1/2) =
1 e~
n
n

1
ln Tr eH = ~ coth(~/2)

2

1/2
h m

i
m tanh(~/2)
hq| |q 0 i =
exp
(q + q 0 )2 tanh(~/2) + (q q 0 )2 coth(~/2)
~
4~

1/2
h m

i
m tanh(~/2)
q 2 tanh(~/2)
P (q) = hq| |qi =
exp
~
~
~
Gaussian distribution, q2 =
.
2m tanh(~/2)
kB T
Thermal fluctuations
High T :
q2 '
m 2
~
Low T :
q2 '
Quantum fluctuations
2m
U = hHi = Tr H =

5
6. MANY-PARTICLE QUANTUM SYSTEMS

We denote by P a permutation operator, such that

P |(1, 2, ..., N )i = |(P 1, P 2, ..., P N )i .
Since the particles are identical we require
2

|| = |P |

and we therefore distinguish between two types of particles,

Bosons correspond to
P |i = |i
Ferminos correspond to
P |i = (1)P |i
where (1)P is +1 for an even permutation and (-1) for and odd permutation.
Now assume a single particle Hamiltonian Hi with eigenstates |ki and energies (k). The total Hamiltonian is given
by
X
H=
Hi .
i

We can construct the N -particle states as follows:

The product space is obtained from
|i0 = |k1 i |k2 i ... |kN i
but these do not obey any of the symmetry requirements.
Bosonic space obtained by
1 X
|S i = p
P |i0 .
N+ P

(6.1)

Here some (or all) ki may appear more than once, and one can specify the state by the numbers nk such that
P
nk = N .
Fermionic space obtained by
1 X
(1)P P |i0 ,
|A i = p
N P

(6.2)

where all the ki s are different.

Note that the |A i can be written in terms of a Slater determinant


 ui (1) ui (2) ui (N ) 


1  uj (1) uj (2) uj (N ) 

|A i = p

 .

N 


 u (1) u (2) u (N ) 
k
k
k

(6.3)

The numbers N+ and N should be chosen such that the state function is normalized. For
Q fermions this is simply
N = N ! since all particles must be at different states. For bosons it is given by N+ = N !/ k nk !.

6
7. QUANTUM IDEAL GAS

Well construct the theory of an ideal gas moving from a single particle, for which weve already seen that


m(x x0 )2
1
0
,
(single particle)
hx | |xi = exp
V
2~2
to two particles, and then generalize.

Two particles

Following the above discussion, the energy states of this ensemble are
1
|Ei = (|E1 E2 i |E2 E1 i)
2
so the density matrix in the space representation is
hx0 | |xi =

1 X (E1 +E2 ) 0
e
hx |Ei hE|xi
Z
E1 ,E2

with
hx|Ei =


1 1  i(k1 x1 +k2 x2 )
e
ei(k1 x2 +k2 x1 ) ,
V 2

or
hx0 | |xi =



0
0
1 1 1 X ~2 (k12 +k22 )  i(k1 x01 +k2 x02 )
e 2m
e
ei(k1 x2 +k2 x1 ) ei(k1 x1 +k2 x2 ) ei(k1 x2 +k2 x1 )
Z4V
k1 ,k2

We now use
1

L3/2

1
=
(2)3

d3 kek eikx =

2
1
e|x| /4
3/2
(4)

to get
hx0 | |xi =

i
0 2
0 2
0 2
0 2
m
m
1 1 h 2~
2~
2 (|x1 x1 | +|x2 x2 | )
2 (|x1 x2 | +|x2 x1 | )
e

e
,
2
Z 2vQ

so the diagonal elements are

P (x) = hx| |xi =
1
Z= 2
2vQ

d3 x1 d3 x2 [ |{z}
1 |e

2
m
~
2 (|x1 x2 | )

O(V 2 )

{z

O(V )

}] '

V2
2
2vQ

i
2
m
1 h
~
2 (|x1 x2 | )
1

e
V2

One could capture this behavior of a non-interacting

quantum gas in terms of an interacting classical gas.
In this case, the Boltzmann factor eu(x) should be
the same as the term in the square brackets in P (x)

(7.1)
1

uHxL

P (x) =

i
2
m
1 1 h
~
2 (|x1 x2 | )
1

e
2
Z 2vQ

0.5

FD
0.5

-0.5

BE

7
above; that is1
"
u(x) = kB T ln 1 exp

2x2

!#
(7.2)

2/3

vQ

The strong repulsive interaction at close distances between fermions is due to Pauli. Notice however that Bosons
experience statistical attraction. Both interactions vanish as the inter-particle distance becomes comparable with
the quantum wavelength, that is within a few nanometers.

N particles

We now write the N -particle matrix element

h1, 2, ..., N | eH |10 , .., N 0 i =

1 1 3N/2 X
P f (|xP 1 x01 |)f (|xP 2 x02 |) f (|xP N x0N |)
N ! vQ
P

with


m 2
x
f (x) = exp
2~2


.

In each term of the sum, particles that are not permuted contribute f (0) = 1. We can therefore order the permeations
in the sum by their order, and rewrite it as
X
X
X
P f (|xP 1 x01 |)f (|xP 2 x02 |) f (|xP N x0N |) = 1
fij fji +
fij fjk fki + ...
i<j

i<j<k

which can be viewed as an expansion in vQ /V :

ZN =

1
N!

V
vQ


N 
N 2 vQ
+
...
.
1
2 23/2 V

(7.3)

To lowest order, we need only to consider the non-permuted (Boltzmann) state, and get the classical partition
function. Note that the Gibbs correction factor N ! appears automatically. In the subleading term the factor
N 2 /2 ' N (N 1)/2 is just the number of such permutations.
F
We can now write F = kB T ln Z and P = V
to obtain

P = 2

vQ
25/2

(7.4)
v

Q
. Indeed,
which looks like a virial expansion, with a virial coefficient B2 = 25/2

vQ
1
B2 = 5/2 , B2 =
2V
2
u(q)

f12 = e

1 = exp

as seen above.

dq1 dq2 f12 = f12 = exp

But many authors have a strong objection to this idea.

2q 2
2/3

vQ

2q 2

2/3

vQ
"

= u(q) = kB T ln 1 exp

2q 2
2/3

vQ

!#

8
8. BOSE-EINSTEIN AND FERMI-DIRAC DISTRIBUTIONS

Quantum gases are much easier to handle in the grand canonical ensemble. We therefore write the grand partition
function
X
XY
Y
P
P
Z(T, V, ) =
W ({nk })e[ k nk k nk ] =
W (nk )e[nk k nk ] =
Zk (T, V, )
{nk }

where Zk (T, V, ) =

nk

{nk } k

W (nk )e[nk k nk ] =

nk

W (nk )z nk enk k is the k-state partition function.

The combinatorial prefactor W (nk ) dependes on the way we constructed the ensemble:

1
BE

W (nk ) = nk ,0 + nk ,1 FD

1/nk !
MB

(8.1)

and thus the partition function

[1 zek ]1 BE
Zk =
1 + zek
FD

exp(zek ) MB

(8.2)

We therefore obtain for the equation of state

P V = ln Z =

X1
ln(1 + azek )
a

(8.3)

where a = 1, 0, +1 for BE, MB and FD, resp.

1
hnk i = 1 
z e k +a
X
X
k
U=
k n k =
z 1 ek + a
k

8.1. Ideal Bose gas

Before we deal with the idealPbose gas, it is useful to introduce the density of state, a(). What we like to do is to
replace the sum over states k with an integral over a continuous state space, weighted appropriately by the
density of states in each volume. In other words we want to make the approximation
Z
Z
Z
X
d3 n 4 n2 dn a()d .
n

For example, for the energy levels of an free particle in a box

n =

h2 n2
2mL2

we have
a() = 4n()2

dn
d

hence
a()d = (2V /h3 )(2m)3/2 1/2 d .

9
With this density of state we can immediately find
P =

2(2mkB T )3/2
h3

N N0
2(2mkB T )
=
V
h3

x1/2 ln(1 zex )dx =

0
3/2

Z
0

1
g5/2 (z)
vQ

x1/2
1
dx =
g3/2 (z) .
z 1 ex 1
vQ

(8.4)
(8.5)

Here we took care not include the number of particles in the ground state in the integral, since we understand that
in moving from a sum to an integral we are missing this term. In general, this is just one state so we might this it
z
is negligible. But since NV0 = 1z
, and we expect that at high densities (or low temperatures) z should approach 1,
wed better not dismiss this term too quickly.
The Bose-Einstein function g (z) is defined as
1
g (z) =
()

Z
0

x1 dx
z3
z2
+
+
=
z
+
z 1 ex 1
2
3

and has the property

1
g1 (z) .
z

g0 (z) =

Use this in the expressions above to get a virial expansion

vQ
P = 2
4 2

2
1

9 3 8

2 3
vQ

`
a` `+1 vQ

(8.6)

`=0

Since g (z) is only defined for z 1 the number of particles in the excited states Ne is bound from above by vVQ g3/2 (1)
with g3/2 (1) = (3/2) ' 2.612. If N is smaller than this maximum, then practically Ne ' N . However at higher
densities
N0 = N

V
(3/2)
vQ

and therefore
z=

N0
1
'1
.
1 + N0
N0

This leads to a macroscopic occupation of the ground state, known as the Bose Einstein Condensation (BEC). The
condition for BEC is therefore
3/2

2mkB
3/2
(3/2)
(8.7)
N <VT
h2
which can be seen either as a condition on N (keeping V and T fixed) or a condition on the temperature (holding N
and V ).
As T Tc from below
No
=1
N

T
Tc

3/2
'

3
(1 t)
2

with the reduced temperature t T /Tc .

Below Tc , the pressure is just
P (T ) =

kB T
(5/2)
vQ

10
(the correction is order 1/N ), which is independent of V , implying infinite compressibility. From this expression we
find
The energy of the system is
U =

V
3

3
1
ln Z = (P V ) = V g5/2 (z) vQ
g5/2 (z) = P V
= V g5/2 (z) (2m/h2 )3/2 = kB T

2
vQ
2

(8.8)

From this, we are interested in calculating


cV =

u
T


.
V,N

At cold temperatures, T < Tc this is straightforward:

cV =

1 3
kB 15
V
kB V (5/2)T (T /vQ ) =
(5/2)
N2
N 4
vQ

(T Tc ) T 3/2 .

Were happy to find that cV 0 when the temperature approaches 0, as required by the third law of thermodynamics.
At high temperatures, we can take two alternative approaches. One is to use this expression for U (z), express z in
terms of , and then take the derivative:



X
3 X 2 3`
3
3
vQ
`
`
a` (vQ ) = kB
a` (vQ ) = kB 1 + + ,
= T kB T
cV = T
2
2
2
2
8 2
V,N

`=0
`
(8.9)
with the classical limit at small or high temperature (namely, small vQ ) and positive correction at higher densities.
Together with our result for low temperature, we conclude the cV must have a maximum at Tc .


U
N

3
= T
2

kB T P

The other approach is to keep working with z, but remember that z is in fact a function of T and N :


CV
3 g5/2 (z)

T
=
N kB
T 2 g3/2 (z)
Above Tc we have g3/2 (z) =


g3/2 (z)
T


=
v

Another way to get

1 N
vQ V

3
g3/2 (z)
2T

dz
dT

T 3/2 and therefore

and

g3/2 (z) = g1/2 (z)

z

is to realize that in the canonical ensemble

dN
dT

1
z

z
T


=
v

3 g3/2 (z)
2T g1/2 (z)

= 0, and therefore





d g3/2 (z)
V
3
dz
V
3
dz 1
dN
=0=V
=
g3/2 (z) +
g3/2 (Z) =
g3/2 (z) +
g1/2 (Z)
dT
dT vQ
vQ 2T
dT z
vQ 2T
dT z
CV
3 g5/2
9 g3/2 g5/2
15 g5/2 (z) 9 g3/2 (z)
=

N kB
2 g3/2
4 g1/2 z g3/2
4 g3/2 (z) 4 g1/2 (z)

(T Tc ).

1 dz
3 g3/
=
z dT
2T g1/

(8.10)

At z = 1 the expression here coincides with the T Tc expression at T = Tc , so CV is continuous at Tc , but its first
derivative is not.

According to London, the N0 particles that occupy a single, entropyless state (" = 0)
could be identified with the superfluid component of the liquid and the Ne particles
that occupy the excited states (" 6= 0) with the normal component. As required in the
1.925

~T

(Cv/Nk)

3/2

1.5

(T /Tc )
FIGURE 7.4 The specific heat of an ideal Bose gas as a function of the temperature parameter (T /Tc ).

11

12
8.2. Ideal Fermi gas

Following the same steps as for the Bose gas, we find

g
f5/2 (z)
vQ
N
g
=
f3/2 (z)
V
vQ

(8.11)

P =

(8.12)

The FD function
f (z) =

1
()

x1 dx
z3
z2
= z +
+1
2
3

z 1 ex

has the property zf 0 (z) = f1 (z). Thermodynamic properties are readily obtained
U =

f5/2 (z)

3
gV
3
ln Z = kT 2
ln Z = kB T
f5/2 (z) = N kB T

T
2
vQ
2
f3/2 (z)
2U
P =
3V
15 f5/2 (z) 9 f3/2 (z)

.
cV /kB =
4 f3/2 (z) 4 f1/2 (z)

(8.13)
(8.14)
(8.15)

The classical limit, corresponding to high-T or low density, is obtained through the limit z  1, where f (z) ' z. In
this case the results above yield as expected
U=

3
kB T,
2

cV =

15 9
3
kB = kB ,
4
2

P = N kB T /V .

We could of course take more terms in the expansion, the same as we did it for the Bose gas, and obtain a virial
expansion.
In the limit T 0 the occupation number
hns i0 = lim

1
= ( )
e(s ) + 1

where (x) is the Heaviside function, (x) = 1 for x > 0 and zero otherwise. In this limit
Z
N0 =
0

F

gV
a()d =
6 2

3/2
2m
3/2
F
=
~2
Z F
U0
1
3
=
a()d = F
N0
N 0
5
2U
2
P0 =
= nF n5/3 .
3V
5

F = (6 2 n/g)2/3

~2
2m

At low but finite temperature we take large but finite z. We use an asymptotic expansion of f (z) in terms of (ln z)1 :


(ln z)
2
7 4
2
4
1 + ( 1) (ln z) + ( 1)( 2)( 3)
(ln z) +
(8.16)
f (z) =
( + 1)
6
360
which yields (to first order)
N
4g
=
V
3

2m
h2

3/2
(kB T ln z)

3/2



2
2
1+
(ln z) +
8

(8.17)

13
and
"
= kB T ln z = F

2
1
12

kB T
F

2 #
(8.18)

With this results we can obtain

#
"

2
U
3
5 2 kB T
= F 1 +
+
N
5
12
F
#
"

2
2U
2
5 2 kB T
P =
= nF 1 +
+
3V
5
12
F
Thus cV grows linearly for T  TF before saturating at the classical limit cV /kB = 3/2.

8.3. Fermi gas in the presence of a magnetic field Pauli paramagnetism

Assuming spin-1/2 particles, the energy is given by

=

p2
B Bz
2m

with z = 1. At T = 0
M = B (N + N ) =

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

B)

(

B)
F
B
F
B
3h3

More generally,
E=


 2
  X 2
X  p2
p
p

B B n+
+
+

B
n
(n+
B
p
p =
p + np ) B B(N N )
2m
2m
2m
p
p
ZN =

N
X

eB B(2N

N )

(0)

(0)

ZN + ZN N +

N + =0

1
1 X
(0)
(0)
ln ZN = B B +
ln
exp[2B BN + FN + FN N + ]
N
N
N+


N + = 0
=
(0) ( N + ) (0) ( N N + ) = 2B B
And finally for small fields we expand


(0) ( N ) ' (0) (N = N/2)+N (0) (N = N/2)(N N/2) = (0) (N = N/2)N+ (0) (N = N/2)(N N/2)
to get


M = B N + N =

22B B


 +
+
N

(0) (N + )

N =N/2

We can now use the expansions given above for high and low temperatures, to find
"

2 #
M
3 n2B
2 kB T
=
1
low T
=
VB
2 F
12
F
n2B h
nvQ i
=
1 5/2
high T.
kB T
2