Lecture 3: Identical particles,

quantum statistics again (9/27/2005)

Today, we start with Blochs theory that we did not finish last time. And
then we will have a look at the Bose-Einstein and Fermi-Dirac distributions
from a different (combinatorial) perspective than the quick approach in the
first class. Our discussion will include the Bose-Einstein condensation and
Plancks blackbody formula. Sadly, I must ask you to invest your time and
read the part of the notes that we wont finish today later (or in the class if
you can do multitasking).
Additional reading if you wish: Griffiths ch. 5.4.

General principles of statistical physics

Each distinct state with the same energy must have the same probability.
However, the number of states depends on the statistics of your particles.

An example: 3 particles in an infinite well

This example is similar to the first homework problem, but its not quite the
same thing. Imagine three non-interacting particles in the one-dimensional
infinite potential well. The total energy is
E=

2h
2 2
(n + n22 + n23 ).
2ma2 1

Imagine that the total energy is

E=

2h
2
243.
2ma2

There are several ways how can we get 243 by summing three squares. The
triplets of positive integers (n1 , n2 , n3 ) are
(9, 9, 9)
(3, 3, 15),
(5, 7, 13),

(5, 13, 7),

(3, 15, 3),

(13, 5, 7),

(15, 3, 3)

(13, 7, 5),
1

(7, 5, 13),

(7, 13, 5)

If the particles are distinguishable, there is one state from the first line, three
different states from the second line, and six different states from the third
line.
If they are indistinguishable bosons, there is one state coming from each
line because the permutations do not matter.
If they are indistinguishable fermions, the Pauli principle dictates that
there are no states coming from the first two lines, and only one state from
the last line (5, 7, 13). You can see that statistics matters, indeed.

General case
Imagine that the particles can be in states whose energies are
E1 , E2 , E3 , . . .
and whose degeneracies are
d1 , d2 , d3 , . . .
and we label the numbers of particles in these states as
N1 , N2 , N3 , . . .
You can see that the total number of particles and the total energy are
N=

Ni ,

E=

Ni Ei .

How many ways there are to arrange the particles to these boxes with fixed
values of Ni ?
For distinct particles, the total number of arrangements is
Qdistinct = N!

Y
i

i
dN
i
Ni !

This is because every permutation of the N particles gives a new state (thus
N!) except for the permutations of the particles inside the groups which
Q
reduces it by the product i (Ni !) and every particle gives an extra factor
di coming from the degeneracy of its state whose energy is Ei .
For identical fermions we only get a nonzero number of possibilities if
the degeneracies di are enough to host the particles, di Ni , and the total
2

number of possibilities is the product of the combination numbers (di choose

Ni ):
Y
di !
Qf ermions =
i Ni !(di Ni )!

For bosons, we need a similar but slightly different product. What are the
factors? We need the combinatorial number that counts the ways in which
Ni identical particles can be separated into di boxes. This problem can be
solved exactly. Imagine that you describe the problem by Ni +di 1 slots, and
each slot can be either filled with the word particle (or an actual particle)
or the words new box (or a separator). Moreover, you require that the
number of particles is Ni and the number of the words new box that
act as separators between the different di boxes must be di 1. Obviously,
such arrangements of the words are in one-to-one correspondence with the
configurations of bosons, and therefore the total number of the states for
bosons is the product of the combinatorial numbers (Ni + di 1 choose Ni
that determine which of the slots are particles (and which of them are
separators) which means
Qbosons =

Y
i

(Ni + di 1)!
.
Ni !(di 1)!

Adding the Lagrange multipliers

We want to find such numbers Ni in the case of each statistics that maximize
the number Q of different multiparticle states. But we only want to do so
given the assumption that the total number of particles and total energy are
fixed:
X
X
N=
Ni ,
E=
Ni Ei .
i

How can one find such a constrained maximum? First of all, a maximum
of Q occurs at the same place as the maximum of ln Q (log is an increasing
function) and we will choose the latter because it is convenient. We define
G = ln Q + N

X
i

Ni + E

Ni Ei

and require that

G
= 0,
Ni

G
= 0,

G
= 0.

The last two equations impose the previous conditions for the total number
of particles and total energy, and because , can be arbitrary, the first
equation (actually a set of equations) only imposes the constrained maximality of Q. For large degeneracies di , we may simplify and approximate the
formulae for Q using the Stirling formula. For large x, the factorial can be
approximated by
 x

x
x! 2x
e
which means that the limit of the ratio of LHS/RHS goes to one as x .
This can be proved from the Euler representation of the factorial
x! =

dt et tx

by identifying the maximum of the integrand near t = x and using the

saddle point approximation (of the integrand by a Gaussian). We also have
the Stirling formula for the logarithms
ln(N!) N(ln N 1)

where even the 2N is negligible and similar formulae for Ni ! etc. Combining these insights, the previous formula for Qdistinct reduces to
ln Qdistinct = ln(N!) +

X
i

[Ni ln di ln(Ni !)]

and thus
Gdistinct = ln(N!) +

X
i

+ N

X
i

[Ni ln di Ni ln Ni + Ni ] +
!

Ni + E

X
i

Ni Ei .

Its derivatives must vanish:

G
=0
Ni

ln di ln Ni Ei = 0

We see that
Ni = di eEi .
The normalization e is determined from the total number of particles
and the coefficient is identified with 1/kB T . Thats the usual MaxwellBoltzmann result.
4

Fermi-Dirac distribution
For fermions, some details in Gf ermions must be modified. We obtain
Gf ermions =

X
i

{ln(di !) ln(Ni !) ln(di Ni )!} +

+ N

X
i

X
i

Ni + E

Ni Ei

[ln(di !) Ni ln(Ni ) + Ni (di Ni )(ln(di Ni ) 1)]

+ N

X
i

Ni + E

Ni Ei

The derivatives must vanish again:

0=

Gf ermions
= ln(Ni ) + ln(di Ni ) Ei
Ni

which means, for identical fermions, that

Ni =

di
+E
i
e

+1

Note that unlike the Maxwell-Boltzmann case, affects not only the overall
normalization. You should notice that we were not cheating: Ni was large
even in the case of fermions because these Ni fermions have di (degeneracy)
available slots.

Bose-Einstein distribution: the result

You are positively invited to repeat the very same calculation for the bosons,
using Qbosons . What you get at the end only differs by a sign from the FermiDirac distribution above:
Ni =

di
+E
i
e

For every physical system, we must then determine , from the conditions
that the total number and total energy are what they should be, namely
N, E. Lets finish this task for the three-dimensional box or, equivalently,
the three-dimensional fermion gas in all three cases of different statistics.
5

Requiring the right total number and the total energy

For distinct particles, we have
N=

di e

Ei

(2s + 1)V
=
e
2 2

or
N = (2s + 1)V e

eh

m
2
h2

2 2
k /2m

k 2 dk

!3/2

where we inserted the spin degeneracy (2s + 1) and replaced the summation
by the integration using the approximate formula for the degeneracy:
dk =

(2s + 1)V k 2 dk
(2s + 1)V 4k 2 dk
=
8
3 /V
2 2

This means that

(2s + 1)V
e =
N

m
2
h2

!3/2

A similar constraint is derived from the total energy condition by inserting

an extra factor of Ei :
E=

di Ei eEi =

2
(2s + 1)V h
e
2 2
2m

or
m
(2s + 1)3V
e
E=
2
2
h2

eh

!3/2

2 2
k /2m

k 4 dk

Substitute our previous result for e to see that

E=

E
3
=
.
N
2

3N
,
2

This nicely agrees with the usual definition = 1/kB T because of the
equipartition theorem that implies
3
E
= kB T.
N
2

In the case of distinguishable particles, only determines the normalization.

But in the cases of all statistics, we usually write down the exponent in a
new form
(T )
=
kB T
and interpret (T ) as the chemical potential: it is the energy needed to add
a particle for constant S, V . Note that in the case of all three statistics, Ni
was proportional to di , so if we evaluate the total number of particles per
one (non-degenerate) state, we simply divide the result by di to obtain
n() =

1
e()/kB T

where = 1 for bosons, 1 for fermions, and 0 for distinguishable particles. As long as the occupation numbers n are low, the distibutions coincide
because the exponential dominates over in the denominator.

Temperatures near zero: fermions

What happens with the Fermi-Dirac distribution for tiny T 0? You can
see that
(
1,  <
n()
0,  >
which simply means that (T = 0) may be identified with the Fermi energy
EF and the fermions occupy all states up to EF .

Temperatures near zero: bosons

For bosons, note that n() must be positive which really means that < 0 (if
we label the minimal allowed energy as zero). The total number of particles
must satisfy
Z
(2s + 1)V
k 2 dk
N=
2
2 2
0
e[h k2 /2m(T )]/kB T 1
Look at this equation and start to reduce (in your brain) the temperature
T , keeping N fixed. You must also allow the chemical potential to vary
to make it work. If you think about which ways various things go, you will
agree that the negative must actually go towards zero as T drops. At some
critical temperature T0 , you hit (T ) = 0. If you lower temperature below
7

T = T0 , we have said that (T ) is not allowed to become positive. Instead,

there will be a macroscopic number of particles (bosons) that occupy exactly
the single state with E = 0. The previous calculations required a negative
(T ) and = 0 is the marginally plausible case. In this case, the number of
bosons in the E = 0 state approaches 1/0 and must be treated separately.
The fraction of the particles in this particular (lowest-energy) state may thus
stay finite. This is called the Bose-Einstein condensation. At T = 0, all of the
bosons will be found in the lowest energy state. Note that it is a momentum
(or energy) eigenstate; its position is not well-defined.

Bose-Einstein condensation: density and numerology

At (T0 ) = 0 and an arbitrary T < T0 (cold temperatures where the condensate is expected), we may calculate
(2s + 1)
N
=
V
2 2

(2s + 1)
N
=
V
2 2

2mkB T
h
2

!3/2

1
2

eh

Choose a better variable x:

2mkB T 1/2
k=
x ,
h

The identity above will become

k 2 dk

2 2
k /2mk

dk =

BT

2mkB T 1/2
x
dx.
2
h

mkB T
xdx
2.61(2s + 1)
x
e 1
2
h2

!3/2

Note the power law dependence of N on T . The integral above only gives
the number of particles with positive energies
NE>0 = N

T
T0

3/2

where the normalization was obtained from N = N at T = T0 while the

rest NE=0 = N NE>0 = N[1 (T /T0 )3/2 ] of the particles is found in the
condensate.
You can actually try to use these equations. For example, for Helium 4
you may use He = 0.15 g/cm3 . This will give you T0 = 3.1 kelvins, very
close to the experimentally determined temperature T0 = 2.17 kelvins where
Helium 4 becomes a superfluid. Recently, the Bose-Einstein condensation
was observed on many real gases with low and correspondingly low T0 .
8

Plancks blackbody formula

Photons are a special case of bosons. However, we must
set 2S +1 2 even though the spin is S = 1 because photons only have
two (transverse) polarizations and the third, longitudinal polarization
is unphysical
we must set = 0 because the number of photons may change thermalization may involve emission and absorption of photons and we
should have avoided the Lagrange multiplier i.e. completely
realize that E = h
for photons, which allows us to choose the angular
frequency
appreciate that E = c|p| i.e. = c|k|
The energy is conserved and enters as usual. The occupation numbers
are
N =

dk
h

/k
BT
e

where

2f rom spin V 2
V 2
k
dk
=
d
2 2
2 c3
The contribution of a frequency interval (, + d) to the energy density
can be expressed in two related ways:
dk =

N h
/V = ()d
Finally we obtain, using more deductive approach than Planck himself in
1900, his formula
h
3
.
() = 2 3 h/k T
B
c [e
1]

Unlike Max Planck 105 years ago, you should actually try to believe that
this formula and the derivation are true.