Marbles on a staircase

Imagine a staircase with NS number of steps such that a single marble placed on step j has energy ϵj . I want to distribute N
identical marbles on this staircase such that the total energy is U . Also I want to study two cases separately (F) There can be
at most one marble per step. (B) There is no restriction. The question is how many ways can I do this ? The answer is denoted
by Ω(U, N ).

X
ΩF (U, N ) = δU,Pj ϵj nj δN,Pj nj
nj =0,1

and X
ΩB (U, N ) = δU,Pj ϵj nj δN,Pj nj
nj =0,1,2,...

These two quantities are not easy to calculate so we instead calculate the transformed quantities namely,
X
ΞF (w, z) = wU z N ΩF (U, N )
U,N

and X
ΞB (w, z) = wU z N ΩB (U, N )
U,N

These two may be easily calculated as follows.

Y
ΞF (w, z) = (1 + wϵj z)
j=1,2,..,NS

and Y
ΞB (w, z) = (1 − wϵj z)−1
j=1,2,...,NS

The quantities we are actually interested in viz. ΩF/B (U, N ) may be easily evaluated from these as follows:
 N
dU F/B

F/B 1 d
Ω (U, N ) = Ξ (w, z)
N !U ! dz N dwU w,z=0

This is useful only when U, N are small. But we want to reinterpret this system as corresponding to fermions or bosons trapped
in a box with energy levels ϵj where j ≡ (nx , ny , nz ) for particles in a box in 3 dimensions. In this case typically, N ∼ 1023 and
it is not practical to differentiate 1023 times to get Ω(U, N ). Note that according to Boltzmann, the entropy of the system is
simply given by S(U, N ) = log(Ω(U, N )).

Grand Ensemble: In fact the quantities Ξ are not merely auxiliary constructs made to finally obtain Ω. The quantities Ξ
themselves have their own physical meaning. To uncover this meaning we introduce the concept of grand ensemble. Imagine
there is a system described by Ω1 (U1 , N1 ) and another called “surroundings” described by Ω2 (U2 , N2 ). The idea is that the
system exchanges both energy and particles with the surroundings so that the net energy U and number of particles N are fixed.
This means U2 = U − U1 and N2 = N − N1 . Now we ask how many ways there are of rearranging the microstates of the system
and surroundings so that the total energy is U and total number of particles is N . The answer is clearly,
X
Ωjoint (U, N ) = Ω1 (U1 , N1 ) Ω2 (U − U1 , N − N1 )
U1 ,N1

1
I now introduce the concept of the most probable energy U1 = U1∗ and number of particles N1 = N1∗ of the system. This is
when the quantity Ω1 (U1 , N1 ) Ω2 (U − U1 , N − N1 ) becomes a maximum. Note that Ω = eS where S is the entropy. Hence the
maximum (extremum for now) is at,
 
∂
(S1 (U1 , N1 ) + S2 (U − U1 , N − N1 )) =0
∂U1 U1 =U ∗ ,N1 =N ∗ 1 1

and  
∂
(S1 (U1 , N1 ) + S2 (U − U1 , N − N1 )) =0
∂N1 U1 =U1∗ ,N1 =N1∗

This means
∂S(U1∗ , N1∗ ) ∂S(U2∗ , N2∗ ) 1
= =
∂U1∗ ∂U2∗ T
In other words the temperature of both the system and surroundings is the same. Also,

∂S(U1∗ , N1∗ ) ∂S(U2∗ , N2∗ ) µ

∗ = =−
∂N1 ∂N2∗ T

This means the chemical potentials are also the same.

Fluctuations: We may expand around these most probable values to see if other values - even though less likely - are also
sometimes seen. For this we write,

S1 (U1 , N1 ) + S2 (U − U1 , N − N1 ) ≈ S2 (U − U1 ∗, N − N1 ∗) + S1 (U1 ∗, N1 ∗)
 
(1,1) (1,1)
+(N1 − N1 ∗)(U1 − U1 ∗) S2 (U − U1 ∗, N − N1 ∗) + S1 (U1 ∗, N1 ∗)
1 
(0,2) (0,2)
 1 
(2,0) (2,0)

+ (N1 − N1∗ )2 S2 (U − U1 ∗, N − N1 ∗) + S1 (U1 ∗, N1 ∗) + (U1 − U1 ∗)2 S2 (U − U1 ∗, N − N1 ∗) + S1 (U1 ∗, N1 ∗)
2 2
but
(1,1) ∂ ∂S1 ∂ 1
S1 (U1 ∗, N1 ∗) = = =0
∂N1∗ ∂U1∗ ∂N1∗ T
as temperature is intensive and independent of everything else.

(2,0) ∂ ∂S1 ∂ 1 ∂U1∗ −1 d 1 1 1

S1 (U1 ∗, N1 ∗) = ∗ ∗ = ∗ = ( ) =− ∗
∂U1 ∂U1 ∂U1 T ∂T dT T N1 c1 T 2

where c1 > 0 is specific heat per particle. Now,

∂S1 µ
∗ =−
∂N1 T
and
(0,2) ∂ µ
S1 (U1 ∗, N1 ∗) = −
∂N1∗ T
Here µ can depend on the density of particles N1∗ /V1∗ so we must write,

(0,2) 1 ∂µ 1 q
S1 (U1 ∗, N1 ∗) = − ∗ ∗ =−
N1 ∂ log(ρ1 ) T T N1∗
∂µ
where q = ∂ log(ρ∗ > 0. Since the surroundings are much larger 1/N2 << 1/N1 and may be neglected. Therefore,
1)

1 q 1 1 1
S1 (U1 , N1 ) + S2 (U − U1 , N − N1 ) ≈ S2 (U − U1 ∗, N − N1 ∗) + S1 (U1 ∗, N1 ∗) − (N1 − N1∗ )2 ∗ − (U1 − U1 ∗)2 ∗
2 T N1 2 N1 c1 T 2

Thus,
X − 21 (N1 −N1∗ )2 T N
q 1 2 1
∗ − 2 (U1 −U1 ∗) N ∗ c
1
2
Ωjoint (U, N ) ≈ Ω1 (U1 ∗, N1 ∗) Ω2 (U − U1 ∗, N − N1 ∗) e 1 1 1 T

U1 ,N1

2
Note that
U1 − U1 ∗ = N1∗ (u1 − u∗1 )
where u1 is energy per particle. Thus,
2 q
1
+(u1 −u1 ∗)2 c1 1
e− 2 N1 ∗[(N1 /N1 ∗−1)
X ]
Ωjoint (U, N ) ≈ Ω1 (U1 ∗, N1 ∗) Ω2 (U − U1 ∗, N − N1 ∗) T 1 T2

U1 ,N1

Since N1 ∗ ∼ 1023 the above result shows that we may easily assume that N1 = N1 ∗ and U1 = U1∗ is always true.

Boltzmann weight: Now we show that there is an alternative expansion that we can make which is also illuminating.
X X ∂S2 (U,N ) ∂S2 (U,N )
Ωjoint (U, N ) = Ω1 (U1 , N1 ) eS2 (U −U1 ,N −N1 ) ≈ eS2 (U,N ) Ω1 (U1 , N1 ) e−U1 ∂U e−N1 ∂N

U1 ,N1 U1 ,N1

X U1 µ
= eS2 (U,N ) Ω1 (U1 , N1 ) e− T eN1 T
U1 ,N1
µ
− T1
Now if we set w = e and z = e T we get,
X
Ωjoint (U, N ) ≈ eS2 (U,N ) Ω1 (U1 , N1 ) wU1 z N1 = eS2 (U,N ) Ξ(w, z)
U1 ,N1

Thus the auxiliary function Ξ(w, z) has the physical meaning of joint number of micro-states of the system plus surroundings
when the two are freely exchanging energy and particles with each other and when the surroundings are much bigger than the
system.

Averages: We may calculate the average number of particles as follows:

P
U ,N U1 Ω1 (U1 , N1 )Ω2 (U − U1 , N − N1 )
< U1 >= P1 1
U1 ,N1 Ω1 (U1 , N1 )Ω2 (U − U1 , N − N1 )

and P
U1 ,N1 N1 Ω1 (U1 , N1 )Ω2 (U − U1 , N − N1 )
< N1 >= P
U1 ,N1 Ω1 (U1 , N1 )Ω2 (U − U1 , N − N1 )
But, X
Ξ(w, z) = Ω1 (U1 , N1 ) wU1 z N1
U1 ,N1

so that we may also write,

∂
z ∂z Ξ(w, z) w ∂ Ξ(w, z)
< N1 >= ; < U1 >= ∂w
Ξ(w, z) Ξ(w, z)
or,
∂ ∂
< N1 >= log(Ξ(w, z)); < U1 >= log(Ξ(w, z))
∂ log(z) ∂ log(w)
or,
∂ ∂
< N1 >= log(Ξ(e−β , es )); < U1 >= − log(Ξ(e−β , es ))
∂s ∂β
where s = −βµ independent of β.

For fermions: X
log(ΞF (e−β , es )) = log(1 + e−βϵj +s )
j

Hence,
∂ X ∂ X
< N1 >= log(1 + e−βϵj +s ); < U1 >= − log(1 + e−βϵj +s )
∂s j ∂β j

3
or,
X 1 X ϵj
< N1 >= −s
; < U1 >= log
j
(1 + e βϵ j ) j
(1 + eβϵj −s )
or,
X 1 X ϵj
< N1 >= ; < U1 >= log
j
(1 + eβ(ϵj −µ) ) j
(1 + eβ(ϵj −µ) )

The above is the famous FERMI-DIRAC DISTRIBUTION Similarly we may derive (Homework),

For bosons: X 1 X ϵj
< N1 >= ; < U1 >= log
j
(eβ(ϵj −µ) − 1) j
(e β(ϵ j −µ) − 1)

This is called the BOSE-EINSTEIN DISTRIBUTION.