4.

Classical phase space quantum states in the part of the space under
consideration.
4.1. Phase space and probability density The ensemble or statistical set consists, at a given
We consider a system of N particles in a d-dimensional moment, of all those phase space points which correspond
space. Canonical coordinates and momenta to a given macroscopic system.
Corresponding to a macro(scopic) state of the system
q = (q1 , . . . , qdN ) there are thus a set of micro(scopic) states which belong
p = (p1 , . . . , pdN ) to the ensemble with the probability ρ(P ) dΓ. ρ(P ) is the
probability density which is normalized to unity:
determine exactly the microscopic state of the system. Z
The phase space is the 2dN -dimensional space {(p, q)}, dΓ ρ(P ) = 1,
whose every point P = (p, q) corresponds to a possible
state of the system.
A trajectory is such a curve in the phase space along and it gives the local density of points in (q, p)-space at
which the point P (t) as a function of time moves. time t.
Trajectories are determined by the classical equations of The statistical average, or the ensemble expectation
motion value, of a measurable quantity f = f (P ) is
Z
dqi ∂H
= hf i = dΓ f (P )ρ(P ).
dt ∂pi
dpi ∂H
= − , We associate every phase space point with the velocity
dt ∂qi
field  
where ∂H ∂H
V = (q̇, ṗ) = ,− .
∂p ∂q
H = H(q1 , . . . , qdN , p1 , . . . , pdN , t)
The probability current is then Vρ. The probability
= H(q, p, t) = H(P, t) weight of an element Γ0 evolves then like
Z Z
is the Hamiltonian function of the system. ∂
The trajectory is stationary, if H does not depend on ρ dΓ = − Vρ · dS.
∂t Γ0 ∂Γ0
time: trajectories starting from the same initial point P
are identical. d S
Trajectories cannot cross: if trajectories meet at a point,
they must be identical!
Let F = F (q, p, t) be a property of the system. Now G 0

dF ∂F
= + {F, H},
dt ∂t
Because Z Z
where {F, G} stands for Poisson brackets
Vρ · dS = ∇ · (Vρ) dΓ,
∂Γ0 Γ0
X  ∂F ∂G ∂G ∂F 
{F, G} ≡ − . we get in the limit Γ0 → 0 the continuity equation
i
∂qi ∂pi ∂qi ∂pi
∂
We define the volume measure of the phase space ρ + ∇ · (Vρ) = 0.
∂t
dN
Y dqi dpi According to the equations of motion
dΓ = = h−dN dq1 · · · dqdN dp1 · · · dpdN .
i=1
h
∂H
q̇i =
−34
Here h = 6.62608 · 10 Js is the Planck constant. (Often ∂pi
dΓ has 1/N ! in front to remove degeneracy caused by ∂H
ṗi = −
identical particles.) ∂qi
Note: [dq dp] = Js, so dΓ is dimensionless.
Note: In classical dynamics the normalization is we have
∂ q̇i ∂ ṗi
irrelevant. In quantum mechanics, it is natural to choose + = 0,
∂qi ∂pi
h = Planck constant.
Note: ∆0 Γ = 1 corresponds to the smallest possible so we end up with the incompressibility condition
volume element of the phase space where a point (sourceless)
representing the system can be localized in accordance
X  ∂ q̇i ∂ ṗi 
with the
R QM uncertainty principle. The volume ∇·V = + = 0.
∆Γ = dΓ is then roughly equal to the number of i
∂qi ∂pi

1
From the continuity equation we get then Ergodic flow: Almost all points of the surface ΓE are
sometimes arbitrarily close to any point in ∆ΓE .
∂ρ
0 = + ∇ · (Vρ) ⇔
∂t
∂ρ The flow is ergodic if ∀f (P ), f (P ) ”smooth enough”,
= + ρ∇ · V + V · ∇ρ
∂t
∂ρ f¯ = hf iE
= + V · ∇ρ.
∂t holds. Here f¯ is the time average
When we employ the convective time derivative Z T
1
f¯ = lim dt f (P (t))
d ∂ T →∞ T 0
= +V·∇
dt ∂t
X ∂  and hf iE the energy surface expectation value
∂ ∂
= + q̇i + ṗi , Z
∂t i
∂qi ∂pi 1
hf iE = dΓE f (P ).
ΣE
the continuity equation can be written in the form known
as the Liouville theorem We define the microcanonical ensemble so that its density
distribution is
d
ρ(P (t), t) = 0.
dt 1
ρE (P ) = δ(H(P ) − E).
ΣE
Thus, the points in the phase space move like an
incompressible fluid, and the local probability density ρ Every point of the energy surface belongs with the same
remains constant during evolution. Different points probability to the microcanonical ensemble.
naturally can have different ρ. The microcanonical ensemble is stationary, i.e. ∂ρ ∂t = 0
E

and the expectation values over it temporal constants.

4.2. Flow in phase space The mixing flow is such an ergodic flow where the
The (constant) energy surface ΓE is the manifold points of an energy surface element dΓE disperse in the
determined by the equation course of time all over the energy surface. If ρ̂E (P, t) is an
arbitrary non stationary density distribution at the
H(q, p) = E. moment t = t0 , then
1
If the energy is a constant of motion, every phase point lim ρ̂E (P, t) = δ(H(P ) − E) = ρE (P )
t→∞ ΣE
P i (t) moves on a certain energy surface ΓEi , of dim.
(2N d − 1). and
The expectation value of the energy of the system Z
Z lim hf i = lim dΓ ρ̂E (P, t)f (P )
t→∞ t→∞
E = hHi = dΓ Hρ Z
= dΓ f (P )ρE (P )
is also a constant of motion.
= hf iE
The volume of the energy surface is
Z Z i.e. the density describing an arbitrary (non equilibrium)
ΣE = dΓE = dΓ δ(H(P ) − E). state evolves towards a microcanonical ensemble under
mixing flow.
The volume of the phase space is Liouville theorem (volume element(s) are conserved) +
mixing: a phase space infinitesimal volume element
Z Z ∞ becomes infinitely stretched and folded and is distributed
dΓ = dE ΣE . over the full energy surface.
−∞

Let us consider the time evolution of a surface element 4.3. Microcanonical ensemble and
∆ΓE of an energy surface. entropy
If the total energy of a macroscopic system is known
Non-ergodic flow: In the course of time the element exactly and it remains fixed, the equilibrium state can be
∆ΓE traverses only a part of the whole energy described by a microcanonical ensemble. The
surface ΓE . corresponding probability density is
Examples: periodic motion; presence of other 1
constants of motion. ρE (P ) = δ(H(P ) − E).
ΣE

2
For a convenience we allow the energy to have some A) Let us now show that the maximisation of the Gibbs
”tolerance” and define entropy gives us microcanonical ensemble, when the
energy of the system is restricted to (E, E + ∆E):
1
ρE,∆E (P ) = θ(E + ∆E − H(P ))θ(H(P ) − E). Z
ZE,∆E
δS = −kB dΓ(δρ ln ρ + ρ δ ln ρ)
∆ΓE
Here the normalization constant Z
Z = −kB dΓδρ(ln ρ + 1) = 0.
ZE,∆E = dΓ θ(E + ∆E − H(P ))θ(H(P ) − E) ∆ΓE
R
Because δ1 = dΓδρ = 0, above condition is satisfied if
is the microcanonical state sum or partition function. ln ρ =const. or ρ(P ) =const., when P ∈ ∆ΓE .
ZE,∆E is the number of states contained in the energy S is indeed maximum:
slice E < H < E + ∆E (see the volume measure of the Z
phase space). In the microcanonical ensemble the 2 1 ∂ 2 (ρ ln ρ)
δ S = −kB dΓ (δρ)2
probability is distributed evenly in every allowed part of 2 ∂ρ2
the phase space. Z∆ΓE
1 1
= −kB dΓ (δρ)2 ≤ 0.
∆ΓE 2 ρ
Entropy
We define statistical entropy as the quantity (depending B) Additive: if we consider 2 separate systems, the joint
on ρ) which is a) maximized for a physical ensemble and probability distribution is ρ12 (P1 , P2 ) = ρ1 (P1 )ρ2 (P2 ),
b) is an extensive quantity, in order to connect to the laws and dΓ12 = dΓ1 dΓ2 .
of thermodynamics. The maximization of the entropy Z
determines the physical distribution ρ. The Gibbs entropy S12 = −kB dΓ12 ρ12 ln ρ12
Z Z
S = −kB dΓ ρ(P ) ln ρ(P ) = −kB dΓ1 dΓ2 ρ1 ρ2 (ln ρ1 + ln ρ2 )
Z Z
= −kB ( dΓ1 ρ1 ln ρ1 + dΓ2 ρ2 ln ρ2 ) = S1 + S2
has these properties, as we shall show shortly.
Let ∆Γi the volume of the phase space element i and ρi
the average probability density in i. The state of the
system is, with the probability Entropy and disorder
Maximizing the entropy S ⇒ minimization of the
pi = ρi ∆Γi , information about the system. In a microcanonical
ensemble complete lack of information (besides the total
in the element i and energy) means that all states with the same total energy
X are equally probable.
pi = 1. Conversely, losing the information (= non-trivial
probability distribution) ⇒ maximizing the entropy.
We choose the sizes of all elements to be smallest Maximum of entropy ⇔ maximum of disorder.
possible, i.e. ∆Γi = 1. Then
X X
S = −kB ∆Γi ρi ln ρi = −kB ρi ∆Γi ln ρi ∆Γi
i i
X
= −kB pi ln pi ,
i

since ln ∆Γi = 0. If ρ is smooth in the range ∆Γ = W we

have
1
ρ= ,
W
so that Z
1 1
S = −kB ln dΓ.
W W
We end up with the Boltzmann entropy

S = kB ln W.

Here W is the thermodynamic probability: the number of

all those states that correspond to the macroscopic state
of the system.

3
5. Quantum mechanical ensembles • Unitary operator U : U U † = U † U = 1. Thus,
||U ψ|| = ||ψ|| and U correspond to a rotation in
5.1. Quantum mechanical concepts Hilbert space, |n′ i = U |ni. Unitary transformations
correspond to a change of eigenbasis.
States and Hilbert space
• Trace of an operator:
Quantum mechanical states (using Dirac notation, |ψi) of
a system form a Hilbert space H, which is a linear vector X X
Tr A = hn| A |ni = ak ,
space with inner (scalar) product and associated norm. n k
Linearity: if |ψi , |ψ ′ i ∈ H, then |cψ + c′ ψ ′ i ∈ H, for
c, c′ ∈ C. where the first form is independent of the choice of
The inner product h·|·i is a mapping H ⊗ H → C, with basis ketvn, and the second is obtained by choosing
eigenvectors of A.
hψ|φi = hφ|ψi∗
hψ|αφ + α′ φ′ i = αhψ|φi + α′ hψ|φ′ i Tr AB = Tr BA, Tr U † AU = Tr A
hφ|φi ≥ 0
• If the quantum mechanical state of the system is |ψi,
hφ|φi = 0 ⇔ φ = 0 the projection operator corresponding to the state is

ρ = |ψi hψ| .
The norm is ||φ|| = hφ|φi1/2 .
Time evolution is defined by Schrödinger equation: Now expectation values of observables are

d hAi = Tr ρA .
ih̄ |ψ(t)i = H |ψ(t)i .
dt
Thus, trace in quantum mechanics corresponds to
If H does not depend on t, the formal solution is phase space integral in classical mechanics:
Z
i
|ψ(t)i = exp[− H(t − t0 )] |ψ(t0 )i Tr A ⇐⇒ dΓA(q, p).
h̄

Operators, eigenvalues and trace

Systems of identical particles
• Physical observable → quantum mechanical operator Let H1 be a Hilbert space for one particle. Then the
A, |ψi → A |ψi. Hilbert space for N particles is

• Conjugated operator A† : HN = H1 ⊗ H1 ⊗ · · · ⊗ H1 .
| {z }
N copies
hψ|A† |φi = hφ|A|ψi∗
If, for example, |xi i ∈ H1 is a position eigenstate the
†
• Eigenvalue and -vector: A |ai = a |ai. If A = A N -particle state can be written as
(hermitean), a is real. Eigenvectors form a complete
(orthonormal) basis of the Hilbert space. |Ψi =
Z Z Z
• Basis of H: any vector can be written as a sum of ··· dx1 · · · dxN |x1 , . . . , xN i ψ(x1 , . . . , xN ),
eigenvectors of some operator:
Z where
X
|ψi = ψn |ni or |ψi = daψa |ai
n
|x1 , . . . , xN i = |x1 i ⊗ |x2 i ⊗ · · · ⊗ |xN i .

depending on whether the eigenvectors form a There are two kinds of particles:
discrete or continuous set. Here we shall use the
Bosons The wave function is symmetric with respect to
former notation.
the exchange of particles.
P
• Identity operator I = n |ni hn|.
Fermions The wave function is antisymmetric with
Here Pn = |ni hn| is the projection operator to vector respect to the exchange of particles.
|ni.
Note: If the number of translational degrees of freedom
• Spectral representation of operator: if A |ni = an |ni, is less than 3, e.g. the system is confined to a two
X dimensional plane, the phase gained by the many particle
A= |ni an hn| . wave function under the exchange of particles can be
n other than ±1. Those kind of particles are called anyons.

4
The Hilbert space of a many particle system is not the Occupation number operator of state |ℓi is
whole HN but its subspace:
 n̂ℓ = a†ℓ aℓ
SHN = S(H1 ⊗ · · · ⊗ H1 ) symm.
H= with eigenvalues 0, 1, 2, . . . for bosons and 0, 1 for
AHN = A(H1 ⊗ · · · ⊗ H1 ) antisymm.
fermions. Thus, antisymmetry of fermion states ⇒ Pauli
exclusion principle.
Fock space For non-interacting systems the Hamilton operator can be
Fock space enables us to describe many-particle quantum expressed as X
states with creation and annihilation operators. Formally, H= Eℓ n̂ℓ
Fock space is the direct sum of all (anti)symmetrized ℓ
N -particle spaces: when the 1-particle states are eigenstates of H.

F = H(0) (1) (N )
p ⊕ Hp ⊕ . . . Hp ⊕ ... 5.2. Density operator and entropy
Let H be the Hilbert space a many particle system.
with H(N
p
)
= SH(N ) for bosons (A for fermions). Fock The probability measure tells us the weight that a state
space wave function in coordinate representation is a |ψi ∈ H represents a system with given macroscopical
vector properties. The density operator ρ tells us the probability
(1) (2) (2)
Ψ = (C, ψ1 (ξ1 ), ψ2 (ξ1 , ξ2 ), . . .) of a given state, pφ = hφ| ρ |φi.
where C is a complex number, ξi = (xi , si ) (si labels the
Ensemble
spin and other internal degrees of freedom), and ψN is a
The quantum mechanical ensemble or statistical set can
fully (anti)symmetric wave function.
be defined in a similar fashion as in classical mechanics.
Fock space normalized N -particle states can be written as
Statistical macrostate is the state determined by
|n1 , n2 , . . . , nℓ , . . .i macroscopic parameters, microstate is a particular QM
state in Hilbert space.
where n1 , . . . are P 1-particle state occupation numbers. In
N -particle state ℓ nℓ = N . In coordinate space Pure state and mixed state
representation these states can be written as If the quantum mechanical state of a system is fully
known it is in a pure state. In this case the density
1 operator is a (pure) projection operator
hξ1 , . . . , ξN |n1 , . . .i = p Q
N ! i ni !
X ρ = |Ψi hΨ|
× ǫP hξ1 |ℓ1 i . . . hξN |ℓN i ,
P (ξ1 ,...ξN ) Statistical mechanics of a pure state reduces into normal
quantum mechanics; for example hAi = Tr ρA = hΨ| A |Ψi
where ℓ1 , ℓ2 , . . . ℓN contain the 1-particle indices of the Mixed state: only the probability pi that the system is
N -particle state; i.e. it contains n1 times 1, n2 times 2 in state ψi is known. Density operator is, in orthonormal
etc. Sum is over all permutations P of N coordinates, base, X
and ǫP is 1 for bosons, and ±1 for fermions for even/odd ρ= |ψn i pn hψn |
permutations (Slater determinant). n
We can define creation and annihilation operators: Now Tr ρ = 1. Ensembe expectation values are
√ X X
aℓ |n1 , . . . , nℓ , . . .i = (−1)Pℓ nℓ |n1 , . . . , nℓ − 1, . . .i hAi = Tr ρA = pi hψi | A |ψi i = pi hAi i .
√
a†ℓ |n1 , . . . , nℓ , . . .i = (−1)Pℓ 1 ± nℓ |n1 , . . . , nℓ + 1, . . .i i i

The a priori probability: if there is no knowledge of the

with upper/lower sign for bosons/fermions, and Pℓ = 0
actual state of the system every state in H can taken with
for bosons, and X equal weight. Then
Pℓ = nk
k<ℓ 1 X
ρ= |ni hn| ,
for fermions. N n
The symmetry properties and normalization imply for
where N = dim H.
bosons

[al , ak ] = [a†l , a†k ] = 0, [al , a†k ] = δl,k , Properties of the density operator
Density operator can be any operator with properties
and for fermions
ρ† = ρ
{al , ak } = {a†l , a†k } = 0, {al , a†k } = δl,k ,
hψ|ρ|ψi ≥ 0 ∀ |ψi ∈ H
([A, B] ≡ AB − BA, {A, B } = AB + BA). Tr ρ = 1.

5
The density operator associates with every normalized 1. S ≥ 0, because 0 ≤ pα ≤ 1.
|ψi ∈ H the probability
2. S = 0 corresponds to a pure state, i.e. ∃α : pα = 1
pψ = Tr ρPψ = hψ|ρ|ψi. and pα′ = 0 ∀α′ 6= α.

Since ρ is hermitean there exists an orthonormal basis 3. If the dimension N of the Hilbert space H is finite,
{|αi} for H, where ρ is diagonal the entropy has a maximum when
X 1
ρ= pα |αi hα| . ρ= I
N
α
1
or pα = N ∀ |αi ∈ H. Then
Here
0 ≤ pα ≤ 1 S = kB ln N .
and X Thus, in this case the entropy fully corresponds to
pα = 1. classical Boltzmann entropy S = kB ln W .
In this basis 4. The entropy is additive. If we have 2 (independent)
X systems, the total Hilbert space is
hAi = Tr ρA = pα hα|A|αi.
α
H1+2 = H1 ⊗ H2

The equation of motion

and correspondingly
Let us assume that the ensemble gives (and fixes) the ρ1+2 = ρ1 ⊗ ρ2 .
probabilities pα corresponding to the states |αi. Now
(i)
X If ρi α(i) = pα α(i) , then
ρ(t) = pα |α(t)i hα(t)| . E E
α (2)
ρ1+2 α(1) , β (2) = p(1)
α p β α(1)
, β (2)
.
Since the state vectors satisfy the Schrödinger equations
Now
d X
ih̄ |α(t)i = H |α(t)i Tr 1+2 A = hα(1) , β (2) |A|α(1) , β (2) i,
dt
d α,β
−ih̄ hα(t)| = hα(t)| H,
dt so that
we end up with the equation of motion
S1+2 = −kB Tr 1+2 ρ1+2 ln ρ1+2
X (2) (2)
d = −kB p(1) (1)
α pβ (ln pα + ln pβ )
ih̄ ρ(t) = [H, ρ(t)].
dt α,β
X X (2) (2)
= −kB p(1) (1)
α ln pα − kB pβ ln pβ
It is easy to show that
α β
 
d hAi ∂A 1 = S1 + S2 .
= + h[A, H]i .
dt ∂t ih̄
The properties of the statistical entropy above are
In a stationary ensemble the expectation values are equivalent to the thermodynamic entropy. This will be
independent on time, so ρ̇ = 0 or shown later.
[H, ρ] = 0. 5.3. Density of states
Let us denote
This is possible e.g. when ρ = ρ(H).
H |ni = En |ni ,
Entropy so that X
The entropy is defined by H= En |ni hn| .
n
S = −kB Tr ρ ln ρ. If the volume V of the system is finite the spectrum is
discrete and the states can be conveniently normalized
In a base where ρ is diagonal, like
X hn|mi = δn,m .
S = −kB pα ln pα .
α Thermodynamic limit:

Entropy has the properties V → ∞ and N → ∞

6
so that N/V remains constant. If the particle has spin S, it has g = 2S + 1 spin degrees
The cumulant function of states is defined as of freedom. Then
Z   Z p
X p2 V
J(E) = θ(E − En ), J1 (E) = g dNkθ E − = g 3 4π dp′ p2
n
2m h 0
V 4π 3
i.e. the value of J at the point E is the number of those = g 3 p .
h 3
states whose energy is less than E. The density of states
So we get
(function) can be defined as
2
dJ(E) X J1 (E) = C1 V E 3/2
ω(E) = = δ(E − En ), 3
dE n ω1 (E) = C1 V E 1/2
 3/2
since dθ(x)/dx = δ(x). However, in macroscopic systems 2m
C1 = 2πg .
the energy levels are extremely closely spaced, and a more h2
sensible definition for the density of states is Example: 2. Maxwell-Boltzmann gas
J(E + ∆E) − J(E) Let us consider N free particles. The total energy is
ω(E) = lim
∆E→0 ∆E X p2j
E=
i.e. the number of states in the interval (E, E + ∆E), j
2m
where we assume that we do not allow ∆E to become
and the cumulant function
smaller than the interval between energy levels. Density
of states becomes a smooth function when the intervals JN (E)
between energy levels → 0 (volume → ∞), as does J(E). Z Z 
p21 p2N

We can also write, using the Hamilton operator, = dNk1 · · · dNkN θ E − − ··· −
2m 2m
Z Z
J(E) = Tr θ(E − H) = dE1 · · · dEN ω1 (E1 ) · · · ω1 (EN )
ω(E) = Tr δ(H − E).
×θ(E − E1 − · · · − EN ).
ω(E) corresponds to the volume ΓE of the energy surface
of the classical phase space. Thus the corresponding density of states is
Example: 1. Free particle dJN (E)
ωN (E) =
Let us consider a free particle in a box of size V = L3
Z dE
with periodic boundary conditions. The Hamiltonian is
= dE1 · · · dEN ω1 (E1 ) · · · ω1 (EN )
p2
H= . ×δ(E − E1 − · · · − EN ).
2m
The eigenfunctions are the plane waves We define the Laplace transforms
Z ∞
1 ik·r Ω1 (s) = dE e−sE ω1 (E)
ψk(r) = √ e ,
V Z0 ∞
ΩN (s) = dE e−sE ωN (E).
where the wave vector k = p/h̄ can acquire the values 0
(periodicity)
Now
2π ΩN (s)
k= (nx , ny , nz ), ni ∈ Z, V = L3 .
L Z ∞
= dE1 · · · dEN ω1 (E1 ) · · · ω1 (EN )
The corresponding energy levels are 0
Z ∞
h̄2 k 2 p2 × dE e−sE δ(E − E1 − · · · − EN )
ǫk = = .
2m 2m Z ∞0
In the limit of large volume the summation can be = dE1 · · · dEN ω1 (E1 )e−sE1 · · · ω1 (EN )e−sEN
0
transformed to the integration over the wave vector, using N
Z = [Ω 1 (s)] .
X 2π X
dk = ∆k = . Since
L n Z ∞
3
Ω1 (s) = dE e−sE C1 V E 1/2 = C1 V Γ( ) s−3/2
Thus, 0 2
X Z Z Z 1 √ −3/2
V 3 V 3 = C1 V πs
= dNk = 3
d k= 3 d p. 2
(2π) h
k

7
we have
ΩN (s) = (C2 V )N s−3N/2 , s
where
 3/2
1√ 2πm
C2 = πC1 = g .
2 h2
residues
Performing the inverse Laplace transforms we get

1
ωN (E) = (C2 V )N E 3/2N −1 .
Γ( 23 N )

Note: We ignored the permutation symmetry! Thus, for γ

each N -particle states there are N ! permutations which
are physically equivalent (unless the particles are all
different). We can correct the density of states by
dividing this by N !, which gives us so-called Boltzmann
counting. Using this we obtain the classical ideal gas or
Maxwell-Boltzmann gas theory (will be discussed later):

1
ωN (E) = (C2 V )N E 3/2N −1 .
N !Γ( 23 N )

Note: This does not take into account the quantum 5.4. Energy, entropy and temperature
mechanical features of multiple occupation of 1-particle The density of states operator ρ enables us to calculate
states (bosons, fermions). all thermodynamical properties of the system: for
example the moments hE n i and entropy S. We can also
Sidebar: inverse Laplace transform define temperature T from properties of ρ. For this we
Obviously, if f (t) = tα , then Laplace transform is need to define microcanonical ensemble, in analogy with
the classical mechanics way done earlier:
Z ∞ Z ∞
−ts α −α−1
dt e t =s dye−y y α = s−α−1 Γ(α + 1) Microcanonical ensemble
0 0
We require that
Thus, if α = 3N/2 − 1, we obtain the result for ωN . a) energy is restricted between (E, E + ∆E), and
Many standard function (inverse) Laplace b) the entropy is maximized.
transformations can be found tabulated (Arfken). According to 5.2. this is satisfied when all states are
More generally, inverse Laplace transforms can be equally likely, thus, the density of states operator is
calculated using complex plane integral (see Arfken, for 1
example): ρE = θ(E + ∆E − H)θ(H − E),
Z γ+i∞ ZE
1
f (t) = dsest fˆ(s) where
2πi γ−i∞
where γ ∈ R is chosen so that the Laplace transform ZE = Tr θ(E + ∆E − H)θ(H − E)
Z ∞ = Tr [θ(E + ∆E − H) − θ(E − H)]
fˆ(s) = dte−st f (t) = J(E + ∆E) − J(E)
0
is microcanonical partition function or the number of
exists when s ≥ γ. It is easy to see that this is inverse
states between (E, E + ∆E). When ∆E is small, we have
Laplace:
Z γ+i∞ Z ∞ ZE ≈ ω(E) ∆E.
1 st ′ −st′ ′
dse dt e f (t )
2πi γ−i∞ 0 Entropy is
Z ∞ Z γ+i∞
1 ′
= dt′ f (t′ ) dses(t−t ) SE = −kB Tr ρ ln ρ = kB ln ZE .
0 2πi γ−i∞
Z ∞ Z ∞
′ ′ γ(t−t′ ) 1 ′ Since ZE is a positive integer, SE ≥ 0 holds. Furthermore
= dt f (t )e dyeiy(t−t ) = f (t) we get
0 2π −∞

The integral can be closed to the left on complex s-plane, SE = kB ln[ω(E) ∆E]
and calculated by finding residues within the contour. = kB ln ω(E) + S0 ,

8
and we can write 6. Equilibrium distributions
Microcanonical ensemble was discussed in the previous
SE = kB ln ω(E), section. That is obtained by maximising entropy with the
boundary condition H = E =constant, N constant. Now
because ln ∆E is non-extensive and negligible when V we shall discuss canonical and grand canonical ensembles,
large. (This expression is slightly incorrect dimensionally; which are obtained by maximizing entropy with boundary
we should use ln(Cω(E)), where C has dimensions of conditions hHi = E, N const. for canonical ensemble (i.e.
energy.) energy is allowed to fluctuate) and and hHi = E,
Note: As a matter of fact hN̂ i = N for grand canonical ensemble (i.e. both energy
and particle number are allowed to fluctuate).
ω = ω(E, V, N ).
6.1. Canonical ensemble
Temperature Let us now maximise the entropy under the constraints
According to thermodynamics we have
hHi = Tr ρH = E = constant
 
1 ∂S hIi = Tr ρ = 1.
= .
T ∂E V,N
Using Lagrange multipliers, we require that
In the microcanonical ensemble we define the
temperature T so that δ(S − λ hHi − λ′ hIi) = 0,

1 ∂ where λ are λ′ are multipliers. We get

= kB ln ω(E, V, N ).
T ∂E
δTr (−kB ρ ln ρ − λρH − λ′ ρ) =
Denoting Tr (−kB ln ρ − kB I − λH − λ′ I)δρ = 0.
1
β= ,
kB T Since δρ is an arbitrary variation, we can solve for ρ and
end up (after relabelling the constants) with the canonical
we have
∂ ln ω or Gibbs distribution
β= .
∂E 1 −βH
ρ= e ,
Example: Maxwell-Boltzmann gas Z
Now
ωN ∝ E 3/2N −1 , where Z is the canonical sum over states (or partition
function, which is determined from the condition Tr ρ = 1:
so Z
3 X
ln ωN = N ln E + · · · Z = Tr e−βH = e−βEn = dE ω(E)e−βE .
2
n
and
3N The constant β is yet undetermined! We shall show below
β=
2E that β = 1/kB T .
or we end up with the equation of state for 1-atomic ideal Note: In the canonical ensemble the number of particles
gas: is constant, i.e.
3
E = kB T N.
2 Z = Z(β, V, N, . . .).
The thermodynamics of a quantum mechanical system
can be derived from the density of states ω(E, V, N ). In The probability for the state ψ is
practice the density of states of a microcanonical 1
ensemble (E and N constant) is difficult to calculate, due pψ = Tr ρPψ = hψ|e−βH |ψi.
Z
to the constraint in total E.
Partcularly, in the case of an eigenstate of the
Hamiltonian,
H |ni = En |ni ,
we have
1 −βEn
pn = e .
Z
For one particle system we get Boltzmann distribution

1 −βǫν X
pν = e ; Z= e−βǫν .
Z ν

9
 R
Here ǫν is the one particle energy. This is normalized dEP (E) R = 1, and we get the correct
expectation values hH n i = dEE n P (E). Using canonical
Entropy and temperature density ρ, we obtain
Because in the canonical ensemble we have
1
P (E) = ω(E)e−βE .
ln ρ = −βH − ln Z, Z
When the volume is large, this typcially has a
the entropy will be well-defined maximum at E = Ē ≡ hHi.
S = −kB Tr ρ ln ρ = −kB hln ρi w (E )e - b E

= kB βE + kB ln Z. D E

Here E is the expectation value of the energy

_ E
1 E
E = hHi = Tr He−βH .
Z Let us write the sum over states as
Z Z
Let us now calculate ∂S/∂E. Note that the parameter β −βE
very well can depend on E! Thus, using the Z = dE ω(E)e = dE e−βE+ln ω(E) .
thermodynamical identity we can define T :
  Now we can use the saddle-point method to calculate the
1 ∂S ∂ ln Z sharply peaked integral: expand the exponent to second
= = kB β + kB Eβ ′ + kB
T ∂E V,N ∂E order
= kB (β + Eβ ′ − β ′ Tr He−βH /Z) ln ω(E) − βE =
′ ′
= kB (β + Eβ − β hHi) = kB β ln ω(Ē) − β Ē
=0, maximum
where β ′ = ∂β/∂E. Thus, we obtain the familiar relation z }| {
∂ ln ω
1 + − β (E − Ē)
β= . ∂E E=Ē
kB T 1 ∂ 2 ln ω
+ (E − Ē)2 + · · · .
2 ∂E 2 E=Ē
Free energy
The partition function is the central quantity, and all Linear term must vanish at E = Ē:
thermodynamic properties can be derived from it: ∂ ln ω 1 ∂S(E) 1
β= = =
∂E kB ∂E kB T (Ē)
∂Z E=Ē E=Ē
= −Tr e−βH H = −Z hHi = −ZE
∂β where we used the microcanonical entropy
or S(E) = kB ln ω(E) to obtain the microcanonical T (Ē).
∂ ∂ ln Z Thus, the temperature of the heat bath in canonical
E=− ln Z = kB T 2 .
∂β ∂T ensemble = T = T (Ē), the temperature of the
microcanonical ensemble at energy Ē = hHican. . Actually,
Using the expression above we can write
Z
∂ ρβ,can. = dE P (E, β) ρE,microcan.
S = kB (T ln Z) .
∂T
and canonical ensemble can be thought as an ensemble of
Using the definition for the Helmholtz free energy
microcanonical ensembles. The 2nd order term in the
F = E − T S, we get
Taylor series is
F = −kB T ln Z.  
∂2S ∂ 1 1 ∂T 1
2 = ∂E =− 2 =− 2 ,
∂E T T ∂E T CV
With the help of this the density operator takes the form
so Z
− 1
(E−Ē)2
ρ=e β(F −H)
. Z ≈ ω(Ē)e−β Ē dE e| 2kB T 2 CV
{z }.
normal distribution
Note that F is a number, H operator. Thus, we find the variance of the normal distribution:
Fluctuations (∆E)2 = kB T 2 CV
The probability distribution of the energy E is
or p √
P (E) = hδ(H − E)i = Tr ρδ(H − E). ∆E = kB T 2 CV = O( N ),

10
because CV , as well as E, is extensive (O(N )). Thus the ending up with the grand canonical distribution
fluctuation of the energy is
1 −β(H−µN̂ )
∆E 1 ρ= e
∝ √ . ZG
E N
Note: More straightforward way to calculate the width is (again relabeling constants). Here
to use
D
2
E ZG = Tr e−β(H−µN̂ )
(∆E)2 = (H − hHi) = hH 2 i − hHi2 ,
is the grand canonical partition function.
which is, using F = −kB T ln Tr e−βH , The trace can be split in N -sectors (effectively using
eigenstates of N̂ ):
∂2
(∆E)2 = − (βF ) X X
∂β 2 ZG = Tr N e−β(H−µN̂ ) = eβµN Tr N e−βH
(N )

∂ 2 ∂ F N N
= −kB T 2 T X
∂T
 2 ∂T T = N
z ZN ,
∂ F
= −kB T 3 = kB T 2 CV N
∂T 2 V,N
where ZN is the canonical partition function with N
Note: Alternative (and perhaps more rigorous) way to particles, and fugasity z ≡ eβµ , and H (N ) is a N -particle
derive the density operator for canonical ensemble would Hamiltonian.
be to start from a very large microcanonical system, and This directly gives the probability distribution of particle
divide this into 2 pieces: part I, “system” and II, “heat number:
bath”, where volume of II ≫ volume of I. Using only
1 N
microcanonical total density operator, it is P (N ) ≡ hδ(N − N̂ )i = Tr ρN̂ = z ZN
straightforward to derive the canonical density for I: ZG
i.e. the grand canonical ensemble is equivalent to the sum
ρI ∝ e−βII HI
of N -particle canonical ensembles where the weight of
where βII = 1/kB TII . each N sector is given by z N .
In the base where the Hamiltonian is diagonal the
6.2. Grand canonical ensemble partition function is
Let us consider a system where both the energy and the XX (N )
number of particles are allowed to fluctuate. The Hilbert ZG = e−β(En −µN ) ,
space of the system is then the Fock space, direct sum N n

H = H(0) ⊕ H(1) ⊕ · · · ⊕ H(N ) ⊕ · · · where

and the Hamiltonian operator the sum of 1,2,. . . -particle H |N ; ni = H (N ) |N ; ni = En(N ) |N ; ni ,

Hamiltonians:
when |N ; ni ∈ H(N ) is a state of N particles, i.e.
(0) (1) (N )
H =H +H + ··· + H + ···.
N̂ |N ; ni = N |N ; ni .
We define the (particle) number operator N̂ so that
N̂ |N i = N |N i Particle number and energy
Now
for eigenstates of N̂ .
Grand canonical ensemble can derived by allowing both ∂ ln ZG 1
= Tr e−β(H−µN̂ ) β N̂
E and N to fluctuate, and requiring that the expectation ∂µ ZG
D E
values are fixed:
= β N̂ = β N̄
hHi = Ē = given energy
D E and
N̂ = N̄ = given particle number
∂ ln ZG 1
Tr ρ = hIi = 1 probability normalization = − Tr e−β(H−µN̂ ) (H − µN̂ )
∂β ZG
D E
We can now demand that the entropy S = −kB ρ ln ρ is = − hHi + µ N̂ = −Ē + µN̄ ,
maximized with the above constraints. Thus, using
Lagrange multipliers we obtain so that, using β = 1/(kB T ),
D E
0 = δ(S + λ hHi + λ′ N̂ + λ′′ hIi) ∂ ln ZG
N̄ = kB T
∂µ
= δTr (kB ρ ln ρ − λρH − λ′ ρN̂ − λ′′ ρ) ∂ ln ZG ∂ ln ZG
= Tr δρ(kB ln ρ + kB − λH − λ′ N̂ − λ′′ ) Ē = kB T 2 + kB T µ .
∂T ∂µ

11
(this is assuming that β = 1/(kB T ), which we have not manner. Let us now study the relation closer. We shall
really yet shown!) assume that the Hamiltonian H and eigenstates |αi
depend on external parameters {xi }:
Entropy and Grand potential
According to the definition of entropy we have H(xi ) |α(xi )i = Eα (xi ) |α(xi )i .

S = −kB Tr ρ ln ρ = −kB hln ρi . Now the change in energy of |αi:

 
Now ∂Eα ∂ ∂H ∂
ln ρ = −βH + βµN̂ − ln ZG , = hα| H |αi = α α + Eα hα| αi
∂xi ∂xi ∂xi ∂xi
 
so that ∂H
1 1 = α α ,
S = Ē − µ N̄ + ln ZG . ∂xi
kB β β
In thermodynamics we defined the grand potential because hα| αi = 1 (Feynman-Hellman theorem).

Ω = E − T S − µN, Adiabatic variation

In quantum mechanics it can be shown that, if the
which agrees with the above expression if we identify variation of the parameters x(t) is slow enough and the
β = 1/(kB T ) (as expected) and initial state of the system is |αi, the system will remain in
the eigenstate |α(x(t))i and there are no transitions to
Ω = −kB T ln ZG . other Hamiltonian eigenstates (note that this is only true
for eigenstates of H).
Thus, the density operator can be written as E a ( x i)
a = 1
ρ = eβ(Ω−H+µN̂ ) .
a = 0

x
Note: The grand canonical partition function depends on
variables T , V and µ, i.e. i

ZG = ZG (T, V, µ). This implies that the probabilities for the states remain
constant and the change in the entropy
X
Fluctuations S = −kB pα ln pα
Now α

∂2 is zero.
Tr e−β(H−µN̂ ) = Tr e−β(H−µN̂ ) β 2 N̂ 2 Let us consider the density operator in an equilibrium
∂µ2
D E state ([H, ρ] = 0). We assume N is constant. In the base
= ZG β 2 N̂ 2 , {|αi}, where the Hamiltonian is diagonal,
so H |αi = Eα |αi ,
D E D E
(∆N )2 = (N̂ − N̄ )2 = N̂ 2 − N̄ 2
we have X
∂ 2 ln ZG ∂ N̄ ρ= pα Pα , Pα = |αi hα| .
= (kB T )2 2 = kB T = O(N̄ ),
∂µ ∂µ α

We divide the variation of the density operator into two

because only N̄ is extensive in the last expression. Thus parts:
the particle number fluctuates like
  adiabatic nonadiabatic
∆N 1 z }| { X z }| {
=O √ . X
N̄ N̄ δρ = pα δPα + δpα Pα
α α
A corresponding expression is valid also for the = δρ(1) + δρ(2) .
fluctuations of the energy. For a mole of matter the
fluctuations are ∝ 10−12 or the accuracy ≈ the accuracy The first part corresponds to adiabatic variation (in
of the microcanonical ensemble. statistical sense), because the probabilities of the
eigenstates remain fixed.
6.3. Relation with thermodynamics Let us define Fi , the generalized force conjugate to the
In thermodynamics the entropy is a state variable related generalized displacement xi as
to the exchange of the heat energy (dQ = T dS) between
 
the system and the environment. However, the statistical ∂H ∂H
Gibbs entropy −kB ρ ln ρ is constructed in a very different Fi = −Tr ρ =− .
∂xi ∂xi

12
For example, F = p, x = V , and F δx the work done by Isolated system and microcanonical ensemble
the system. If the system is microcanonical, then Eα = E for all
Then the change in energy is states. However, when we change x, E(x) can vary! In
this case
δ hHi = Tr δρ H + Tr ρ δH X X
X ∂H δ hHi = δE = δpα Eα − Fi δxi
= Tr δρ(1) H + Tr δρ(2) H + δxi Tr ρ α i
i
∂xi X
X X = − Fi δxi ,
= pα Tr H δPα + Tr δρ(2) H − Fi δxi . i
α i
which is equivalent to thermodynamical work done by an
The first term vanishes, because isolated system
X X
Tr H δPα = hβ| H (|αi hδα| + |δαi hα|) |βi δU = −δW = − Fi δxi
i
β
= Eα δ hα| αi = 0,
6.4. Einstein’s theory of fluctuations
so that the change in energy becomes Let us divide a large system into macroscopic parts with
X X X weak mutual interactions.
δ hHi = Tr δρ(2) H − Fi δxi = δpα Eα − Fi δxi . ⇒ ∃ operators {X̂i }, which correspond to the extensive
i α i properties of the partial systems so that
The definition of the statistical Gibbs entropy is [X̂i , X̂j ] ≈ 0
X [X̂i , H] ≈ 0.
S stat = −kb Tr ρ ln ρ = −kB pα ln pα ,
α ⇒ ∃ mutual eigenstates |E, X1 , . . . , Xn i, which are
macrostates of the system, i.e. for each set of the
and its variation is parameters (E, X1 , . . . , Xn ) there is a macroscopic
=0 number of microstates. Let Γ(E, X1 , . . . , Xn ) be the
X zX}| { number of the microstates corresponding to the state
δS stat = −kB δpα ln pα − kB δpα |E, X1 , . . . , Xn i (the volume of the phase space).
α α
X The total number of the states is
= −kB δpα ln pα X
α Γ(E) = Γ(E, X1 , . . . , Xn )
X {Xi }
= kB β δpα Eα ,
α and the relative probability of the state (E, {X})
where in the last stage we used the canonical ensemble Γ(E, X1 , . . . , Xn )
f (E, X1 , . . . , Xn ) = .
Γ(E)
1 −βEα
pα = e
Z The entropy of the state |E, X1 , . . . , Xn i is
Thus, if we denote β = 1/(kB T stat ), we obtain S(E, X1 , . . . , Xn ) = kB ln Γ(E, X1 , . . . , Xn )
X or
δ hHi = T stat δS stat − Fi δxi . 1 1
S(E,X1 ,...,Xn )
f (E, X1 , . . . , Xn ) = e kB .
i Γ(E)
This is the probability of state X with fixed energy E.
This is equivalent to the first law of the thermodynamics,
In thermodynamic equilibrium the entropy S is
maximized:
δU = T therm δS therm − δW,
(0)
S 0 = S(E, X1 , . . . , Xn(0) ).
provided we identify
Let us denote by
(0)
hHi = Ē = U = internal energy xi = Xi − Xi
T stat = T therm deviations from the equilibrium positions, and expand the
S stat = S therm entropy to 2nd order around extremum: the Taylor series
X
Fi δxi = δW = work. of the entropy will be
i 1 X
S = S 0 − kB gij xi xj + · · · ,
2 i,j

13
where   We can naturally also calculate correlations of other
1 ∂2S quantities, for example
gij = − .
kB ∂Xi ∂Xj {Xi
(0)
}
   
∂S ∂S
h∆S∆V i = h∆T ∆V i + h(∆V )2 i
We use vector-matrix notation ∂T V ∂V T
 

x1
 ∂p
= V kB T κT
∂T V
x =  ...  and g = (gij ).
 
"    #−1
xn ∂V ∂T
= − V kB T κT
∂p T ∂V p
Then the probability becomes
= kB V T αp
− 21 xT gx
f (x) = Ce , ∂S

 
∂p
where we used the Maxwell relation ∂V T
= ∂T V .
where p Note that if we were to origially take other set of
C = (2π)−n/2 detg. variables, say, S, V as the independent ones, the matrix
g = −1/kB ∂ 2 S/∂~x2 would look different (in this case
Correlation functions non-diagonal).
Correlation functions can be written as
Z 6.5. Reversible minimum work
hxp · · · xr i ≡ Dx f (x)xp · · · xr Another way to calculate the fluctuation probability is to
  consider the concept of reversible minimum work.
∂ ∂ Let x = X − X (0) be the fluctuation of the variable X.
= ··· F (h) ,
∂hp ∂hr h=0 For one variable we have
1 2
where f (x) ∝ e− 2 gx .
Dx = dx1 · · · dxn Let us assume that X is generalized displacement; thus
and the generating function is
dU = T dS − F dX − d̄Wother .
Z
1 T 1 T T 1 T −1
F (h) = C Dxe− 2 x gx+ 2 (h x+x h) = e 2 h g h . If we write S = S(U, X, . . .) we get the partial derivative
∂S F
= .
SVN-system ∂X T
When studying the stability conditions of matter we On the other hand we had
found out that the 1 X
S = S 0 − kB gij xi xj
1 X 2 i,j
∆S = − (∆Ti ∆Si − ∆pi ∆Vi + ∆µi ∆Ni ).
2T i 1
= S0 − kB gx2 ,
2
Let us consider now only one volume element: so
∂S
− 2k 1 (∆T ∆S−∆p ∆V +∆µ ∆N ) = −kB gx
f = Ce BT . ∂X
and in the limit of small displacement
We assume that the system is not allowed to exchange
particles, i.e. ∆N = 0. Taking now T and V as our F = −kB T gx.
independent variables we can employ the definitions of When there is no action on X from outside, the deviation
the heat capacity and compressibility: x fluctuates spontaneously. Let us give rise to the same
h
CV
i deviation x by applying reversible adiabatic external
− 12 (∆T )2 + V k 1
(∆V )
f (∆T, ∆V ) ∝ e kB T 2 B T κT
. work:
dU = −F dX = kB T g x dx.
We can now read out the matrix g: Integrating this we get
T V 1
! (∆U )rev ≡ ∆R = kB T gx2 ,
T CV
0 2
kB T 2
g= 1 . where ∆R is the minimum reversible work required for
V 0 V kB T κT
the fluctuation x = ∆X. We can write
The variances are then − k∆RT
f (∆X) ∝ e B .
2
kB T Note that this relates a fluctuation of X at constant
(∆T )2 = 2
CV U = E (f (x) ∼ eδ S/kB ) to the adiabatic work done to
(∆V )2 = V kB T κT . achieve same change of X (δS = 0).

14
  
7. Ideal equilibrium systems 1 1 1
− ( N − ν) ln( N − ν) − ( N − ν)
In ideal systems there are no interactions between the 2 2 2
particles/degrees of freedom. If we know the solution for 1 N2 1
N +ν
2
1-particle state, N-particle state can be solved by taking = N ln 1 2 2
− ν ln 1
2 4 N −ν 2 N −ν
into account the correct statistics.
1 1 1 + 2ν/N
= N ln 2 + N ln 2 2
− ν ln
7.1. System of free spins 2 1 − 4ν /N 1 − 2ν/N
Let us consider N particles with spin 12 . The ν2
≈ N ln 2 − 2
z-component of the spins are N

Siz = ± 12 h̄ i = 1, . . . , N. where in the last step we have taken the approximation

|ν| ≪ N . The maximum of W is clearly at ν = 0.
The z component of the total spin is Thus, the binomial distribution can be approximated by a
normal distribution:
X 1
Sz = Siz = h̄(N + − N − ), 2
2 W (ν) ≈ W (0)e−2ν /N
, W (0) ≈ 2N .
i
√
where The mean deviation ∆ν ≈ N /2 ≪ N , i.e. large values
of ν are very unlikely.
1
N+ = + h̄ spin count Note that the total number of states
2 X
1 Wtot = W (ν) = 2N
N− = − h̄ spin count.
2 ν

Sz determines the macrostate of the system. Thus, a correct normalization of the gaussian distribution
Denoting Sz = h̄ν we have would be
Z p
2
1 Wtot = C dνe−2ν /N = C N π/2 = 2N
N+ = N +ν
2
1 p
which implies C = 2N / N π/2. However, the correction
N− = N −ν
2 in this case is non-extensive and can be neglected in the
and large N limit:
1 1 1
ν = − N, − N + 1, . . . , N. non-extensive
2 2 2 extensive
z }| {
z }| { 1
Let W (ν) the number of those microstates for which ln C = N ln 2 − ln(N π/2)
2
Sz = h̄ν, i.e. W (ν) tells us, how many ways there are to
distribute N particles into groups of N + and N −
particles so that N + + N − = N and N + − N − = 2ν. Energy
From combinatorics we know that this is given by the Let us put the system in an external magnetic field
binomial distribution: H = H ẑ, with magnetic flux density
 
N N! B = µ0 H.
W (ν) = = + −
N+ N !N ! The potential energy is
N!
= . X X
( 2 N + ν)!( 12 N − ν)!
1 E = −µ0 µi · H = −µ0 H µiz ,
i i
W (ν) is the degeneracy (=number of states) of the
where µi is the magnetic moment of the particle i.
macrostate Sz = h̄ν.
The magnetic moment of a particle is proportional to its
The Boltzmann entropy is
spin:
S = kB ln W (ν). µ = γS,
where γ is the gyromagnetic ratio. In classical
Using Stirling’s formula electrodynamics, with a homogeneously charged solid
body of mass m, total charge q and angular momentum S
ln N ! ≈ N ln N − N , this has the value
q
γ0 = .
we get 2m
However, for elementary particles this is different, and for
ln W (ν) ≈ N ln N − N electrons we have
 
1 1 1 e
− ( N + ν) ln( N + ν) − ( N + ν) γ ≈ 2γ0 = − ,
2 2 2 m

15
Usually this is given in terms of Bohr magneton The magnetization or the magnetic polarization is the
magnetic moment per the volume element, i.e.
eh̄ eV
µB = = 5.79 · 10−5 .
2m T 1 X
M= µ.
The energy of the system is V i i
X
E = −µ0 H µiz = −µ0 γHSz = ǫν, The z component of the magnetization is

where 1 ǫν 1 h̄γµ0 Hν
Mz = − =
ǫ = −h̄γµ0 H V µ0 H V µ0 H
1
is the energy/particle. For electrons we have = γh̄ν.
V
ǫ = 2µ0 µB H. Now
E = −µ0 HV Mz ,
Now the change in energy is
so we get for our system as the equation of state
∆E = ǫ ∆ν,
 
1 µ0 h̄γH
Changing the variable ν ↔ E, and using the condition M= ρh̄γ tanh ,
2 2kB T
(Jacobi!)
ω(E) |∆E| = W (ν) |∆ν|
where ρ = N/V is the particle density. Note: The
we get as the density of states relations derived above
 
1 E E = E(T, H, N )
ω(E) = W .
|ǫ| ǫ M = M (T, H, N )

determine the thermodynamics of the system.

1) Microcanonical ensemble
Let us first consider the spin system at fixed energy. 2) Canonical ensemble
Denoting The canonical partition function is
1
E0 = ǫN, X
2 Z= e−βEn .
the total energy will lie between −E0 ≤ E ≤ E0 . n
With the help of the energy W can be written as
Here
1 4E 2 E E0 + E N
N ln 2 0 2 −
X
ln W (ν) = ln En = −µ0 H µiz
2 E0 − E ǫ E0 − E
i=1
= ln ω(E) + ln |ǫ|.
the energy of a single microstate.
As the entropy we get Denote
1
S(E) = kB ln ω(E) µiz = h̄γνi , νi = ± .
  2
1 4E02 E E0 + E Now
= N kB ln 2 − ln
2 E0 − E 2 2E0 E0 − E P
X
+non extensive term. Z = eβµ0 H i
µiz

all microstates
The temperature was defined like 1 1
2
X 2
X P
1 ∂S = ··· eβµ0 h̄γH i
νi

= , ν1 =− 21 νN =− 12
T ∂E
 1
N
so 2
X
1 N E0 + E =  eβµ0 Hγh̄ν  = Z1N ,
β(E) = =− ln .
kB T (E) 2E0 E0 − E ν=− 12
We can solve for the energy:
where Z1 the one particle state sum
βE0
E = −E0 tanh 1 1
N   Z1 = e− 2 βµ0 h̄Hγ + e 2 βµ0 h̄Hγ
1 µ0 h̄γH µ0 h̄Hγ
= − N µ0 h̄γH tanh . = 2 cosh .
2 2kB T 2kB T

16
The same result can be obtained using W (ν): Thus the differential of the free energy is
X
Z = W (ν)e−βE(ν) dG = −S dT − µ0 V M · dH,
ν
X so that the magnetization is
= W (ν)e−βǫν
ν
 
1 ∂G
X N

+
M = −
− 12 N ) µ0 V ∂H T
= e−βǫ(N
N+  
N+ 1 µ0 h̄γH

N = − ρh̄γ tanh .
1
= e− 2 βǫN 1 + e−βǫ . 2 2kB T

From the partition function we can determine This is identical with the result we obtained in the
thermodynamic potential, free energy. However, what is microcanonical ensemble.
the “correct” thermodynamic potential in this case? The microcanonical entropy = the canonical entropy + a
Naively we might identify non extensive term.
Energy
F (T, H) = −kB T ln Z
a) Energy E(T, H) (enthalpy) can be calculated directly
as the Helmholtz free energy (as is done in many from partition function
textbooks). This is (mostly) OK for thermodynamics, 1 ∂
and the simple choice, but it is not fully consistent: E = hE(ν)i = ǫν̄ = − Z
Z ∂β
Note that H is generalized force (“p”), M is displacement  
(“V ”). Thus, when our set of independent variables is (T , 1 1
= − N ǫ tanh βǫ
H), the right thermodynamic potential is the Gibbs free 2 2
energy = the energy of the microcanonical enesmble.

G = G(T, H) = −kB T ln Z b) Alternatively, using thermodynamic Legendre

  transformation from enthalpy to Gibbs potential
µ0 Hγh̄
= −kB T N ln 2 + ln cosh .
2kB T G = E − TS

Furthermore, in this case we should identify our energy E or

as the enthalpy of the system: ∂G
E = G + TS = G − T
E = −µ0 HV M = U − H × (µ0 V M ) ∂T
∂G ∂
which is related to the true internal energy U by a = G+β = (βF )
∂β ∂β
Legendre transform. In this case internal energy vanishes: ∂
= − ln Z
U = 0. ∂β

True internal energy U should not depend on external Susceptibility

force like H, only on internal properties like M . However, According to the definition the susceptibility is
in this case there is no H-independent energy!    2 
Thus, the entropy is ∂M 1 ∂ G
χ = =−
  ∂H T µ0 V ∂H 2
∂G
2
S = − µ0 ρ 1
h̄γ
∂T H = 2 .
" kB T cosh2 h̄γµ0 H
µ0 Hγh̄ 2KB T
= N kB ln 2 + ln cosh
2kB T When H → 0 we end up with Curie’s law
#
µ0 h̄γH µ0 h̄γH
− tanh . C
2kB T 2kB T χ= ,
T
Differentiating the free energy with respect to the field H where
we get µ0 ρ

1
2
  C= h̄γ .
∂G 1 ∂ X kB T 2
− = kB T W (ν)e−βǫν
∂Hz T Z ∂H ν
1 X Thermodynamic identifications
= µ0 γh̄ νW (ν)e−βǫν Earlier we identified
Z ν D E
= µ0 γh̄ hνi = µ0 V Mz . E stat ≡ E = Ĥ = enthalpy,

17
so that or
Tb Hb
= .
G = E − T S = Gtherm Ta Ha
= the Gibbs free energy Can be serialised to reach even smaller T .
= U − T S − µ0 V M · H. Negative temperature
The entropy of the spin system is
Now using the differentials we get
 
1 4E 2 E E0 + E
dU = dG + d(T S) + d(µ0 V M · H) S(E) = N kB ln 2 0 2 − ln ,
2 E0 − E 2E0 E0 − E
= T dS + µ0 V H · dM
= T dS − d̄W where

Thus, the work done by a magnetic system is E0 = µ0 µB HN ja − |E0 | < E < |E0 |.

S (E )
d̄W = −µ0 V H · dM. b = 0
(compare to pdV ). b > 0 b < 0
Example: Adiabatic demagnetization is used to T > 0 T < 0
- |E 0 | + |E 0 |
achieve cooling of atomic spins to nanokelvin
temperatures. This is due to very large degeneracy of
spin states, which means large entropy down to very Now
small temperatures. Removing entropy with magnetic 1 ∂S N E + E0
β(E) = =− ln .
fields can then reduce temperature very effectively. kB ∂E 2E0 E − E0

e
Now -b E
S
H -H
= ln 2 + ln cosh x − x tanh x,
N kB
where
µ0 h̄Hγ
-E + E + E -E
x= .
2kB T
0 0 0 0
When T → 0, then x → ∞, so that −βE
Originally the maximum of ω(E)e /Z is at a negative
1 x value E. Reversing the magnetic field abruptly E → −E
ln cosh x = ln e (1 + e−2x ) and correspondingly β → −β.
2
= x − ln 2 + e−2x + · · · The temperature can be negative if the energy is bounded
both above and below.
and
ex (1 − e−2x )
7.2. Classical ideal gas
tanh x =
ex (1 + e−2x ) (Maxwell-Boltzmann gas)
Kinetic gas theory, James Clerk Maxwell (1860): gas of
= 1 − 2e−2x + · · · . atoms/molecules. Explained many properties of real
Hence gases, supporting the atomic hypothesis of matter.
S At standard temperature and pressure (STP) most gases
→ 2xe−2x + · · · . are dilute:
N kB

2 Å
When T → ∞, then x → 0, and

r
S
→ ln 2. i
N kB
Let us cool down a system by some means at a large field
Ha , reaching point a:
S
We define ri so that the volume occupied by one molecule
N k
H < H < H
B =
1 2 3
ln 2 4 V 1
vi = πri3 = =
3 N ρ
b a
or r
3
T
ri = 3 .
4πρ
We decrease the field adiabatically within the interval Typically
a → b. Now S = S(H/T ), so that
    • the diameter of an atom or a molecule d ≈ 2Å.
Ha Hb
Sa = S = Sb = S • the range of the interaction 2–4Å.
Ta Tb

18
 • the mean free path (collision interval) l ≈ 600Å. From the relation
Z Z ∞
• at STP (T = 273K, p = 1atm) ri ≈ 20Å. d3 v = 4πv 2 dv
0
or
d ≪ ri ≪ l we can obtain for the speed (the absolute value of the
2 20 600 Å velocity v = |v|) the distribution function F (v)
The most important effect of collisions is that the system
F (v) = 4πv 2 f (v).
thermalizes i.e. attains an equilibrium, which corresponds
to a statistical ensemble. Otherwise we can forget the
F (v )
collisions and describe the gas as non-interacting
molecules. Let us consider a system of one molecule
which can exchange energy (heat) with its surroundings.
Then the suitable ensemble is the canonical ensemble and v
the distribution the Boltzmann distribution
1 −βǫl
pl = hl| ρ |li = e , • The most likely speed (maximum of the distribution)
Z
r
where the canonical partition function is 2kB T
vm = .
X m
Z= e−βǫl .
l • The average speed
r
We recall that in k-space the density of 1-particle states Z ∞
8kB T
is constant: in volume V = L3 the wave function is hvi = dv vF (v) = .
0 πm
ψk(x) ∝ eik·x
• The average of the square of the speed
where, using periodic b.c. k = 2πn/L, n ∈ Z 3 . Thus, in Z ∞
3kB T
volume element (k, k + ∆k) the number of states is v2 = dv v 2 F (v) = .
constant independent of k. 0 m
For convenience, we shall switch to velocity space, with
Note:
v = p/m = h̄/mk. However, everything would work also      
in k-space too. 1 1 1 1
mvx2 = mvy2 = mvz2 = kB T
The volume element in velocity space is 2 2 2 2
 3
1 h̄ and    
d3 v = 3 d3 p = d3 k, 1 1 3
m m mv 2 =3 mvx2 = kB T,
2 2 2
i.e. the density of states is also constant. The energy of i.e. the energy is evenly distributed among the 3
k-state is (translational) degrees of freedom: the equipartition of the
h̄2 k 2 1 energy.
ǫk = hk| H |ki = = mv 2 ,
2m 2 Partition function and thermodynamics
so that the velocity distribution becomes The single particle partition function is
Z
dE ω(E)e−βE
2
f (v) ∝ hk| ρ |ki = e
mv
− 2k T B
Z1 (β) =
Z
X h̄2 k2 V − p
2
or = g e−β 2m = g 3 d3 p e 2mkB T
mv
− 2k
2
h
f (v) = Ce BT
. k
V
C can be determined from the condition = g 3 (2πmkB T )3/2 .
Z ∞ 3 h
Z mv2
3 − 2k xT Here g is the spin degeneracy.
1 = f (v) d v = C dvx e B

−∞ When we denote the thermal de Broglie wave length by

 3/2
2πkB T s
= C . h2
m λT =
2πmkB T
Thus the velocity obeys Maxwell’s distribution
we can write the 1 body partition function as
 3/2
m mv2
− 2k V
f (v) = e BT . Z1 (β) = g .
2πkB T λ3T

19
In the N particle system the canonical partition function the the equation of state, we obtain µ in terms of the
takes the form natural variables for the Gibbs ensemble:
 
1 N X X −β(ǫk +···+ǫk ) 5 3 h2
ZN = g ··· e 1 1 µ(p, T ) = T ln p − ln kB T + ln − ln g .
N! 2 2 2πm
k1 kN
!N
1 N X −βǫ Grand canonical partition function
= g e k
N! According to the definition we have
k
1 N XX (N ) X
= Z . ZG = e−β(En −µN ) = z N ZN ,
N! 1
N n N
Here N ! takes care of the fact that each state where
z = eβµ
|k1 , . . . , kN i
is called the fugacity and ZN is the partition function of
is counted only once. Neither the multiple occupation nor N particles. Thus we get
the Pauli exclusion principle has been taken into account.  
X 1
Using Stirling’s formula ln N ! ≈ N ln N − N the free N N zZ1 βµ gV
ZG = z Z1 = e = exp e .
energy can be written as N! λ3T
N

FN = The grand potential is

= −kB T ln ZN

N
 gV
= N kB T ln − 1 − ln g + hλT3 Ω(T, V, µ) = −kB T ln ZG = −kB T eβµ 3 .
V λT
" #
N 3 3 h2 Because dΩ = −S dT − p dV − N̄ dµ, we get
= N kB T ln − ln kB T − 1 − ln g + ln .
V 2 2 2πm 
∂Ω

Ω g
p=− = − = kB T eβµ 3
∂V T,µ V λT
Because dF = −S dT − p dV + µ dN, the pressure will be
  and
∂F 1 ∂Ω gV pV
p=− = N kB T N̄ = − = eβµ 3 = ,
∂V T,N V ∂µ λ T kBT

and we end up again with the ideal gas equation of state

i.e. we end up with the ideal gas equation of state
pV = N̄ kB T.
pV = N kB T.
Here
With the help of the entropy P
N zN Z
N̄ = hN i = PN N N
N z ZN
 
∂F F 3
S=− = − + N kB 1 ∂ZG ∂ ln ZG
∂T V,N T 2 = z = .
ZG ∂z ∂ ln z
the internal energy is Another way
3 We distribute N particles among the 1 particle states so
U = F + TS = N kB T that in the state l there are nl particles.
2
ǫl
i.e. the ideal gas internal energy. 6
The heat capacity is

∂U

3
t nl = 1
CV = = N kB .
∂T V,N 2
t t t t nl = 4
Comparing this with

1 t t nl = 2
CV = f kB N
2
we see that the number of degrees of freedom is f = 3.
We can also calculate the Gibbs function
G = F + pV = µN , and substuting N/V = p/(kB T ) using

20
Now X X so we must have
N= nl and E = ǫ l nl . λT ≪ ri .
l l
Now s
The number of possible distributions is
h2
λT =
N! 2πmkB T
W = W (n1 , n2 , . . . , nl , . . .) = .
n1 !n2 ! · · · nl ! · · ·
is the minimum diameter of the wave packet of a particle
Since in every distribution (n1 , n2 , . . .) each of the N ! with the typical thermal energy (ǭl = kB T ) so in other
permutatations of the particles gives an identical state words:
the partition function is The Maxwell-Boltzmann approximation is valid when the
∞ X
∞
wave packets of individual particles do not overlap.
X 1
ZG = ··· W e−β(E−µN )
n1 =0 n2 =0
N! 7.3. Diatomic ideal gas
∞ X ∞ P
X 1
= ··· e−β l nl (ǫl −µ)
n1 =0 n2 =0
n1 !n2 ! · · ·

d
" ∞ #
Y X 1 −βnl (ǫl −µ)
= e
n
nl =0 l
!
l
Y h i
= exp e−β(ǫl−µ) We classify molecules of two atoms to
l
" # • homopolar molecules (identical atoms), e.g. H2 , N2 ,
X
= exp e −β(ǫl −µ) O2 , . . ., and
l
  • heteropolar molecules (different atoms), e.g. CO,
βµ
= exp e Z1 NO, HCl, . . .
or exactly as earlier. When the density of the gas is low the intermolecular
Validity range of MB gas law interactions are minimal and the ideal gas equation of
Now state holds. The internal degrees of freedom, however,
P change the thermal properties (like CV ). When we
∂ ln ZG −β ∞ 1 −βn(ǫl −µ)
n=0 n n! e assume that the modes corresponding to the internal
= h i
∂ǫl Q P∞ 1 −βnm (ǫm −µ) degrees of freedom are independent on each other, we can
m nm =0 nm ! e
write the total Hamiltonian of the molecule as the sum
= −β hnl i
H ≈ H tr + H rot + H vibr + H el + H ydin.
so the occupation number n̄l of the state l is
Here
1 ∂ ln ZG 1 ∂ −β(ǫl −µ) p2
n̄l = hnl i = − =− e H tr == kinetic energy
β ∂ǫl β ∂ǫl 2m
= e−β(ǫl −µ) . and m = mass of molecule.

The Boltzmann distribution gives a wrong result (for real L2 h̄2 l(l + 1)
gases) if 1-particle states are multiply occupied. Our H rot = = = rotational energy
2I 2I
approximation is therefore valid if with L = angular momentum and I = moment of inertia.
n̄l ≪ 1 ∀l If xi are distances of atomic nuclei from the center of
mass, X m1 m2 2
or I= mi x2i = d
eβµ ≪ eβǫl ∀l. i
m 1 + m2

Now min ǫl = 0, so that Example:

p H2 -molecule d = 0.75Å,
L = h̄ l(l + 1), l = 0, 1, 2, . . .. Now
eβµ ≪ 1.
h̄2
On the other hand = 85.41K
2IkB
N̄ 3
eβµ = λ , when g = 1 is the characteristic temperature of rotational d.o.f.
V T 2
Eigenvalues h̄2I l(l + 1) have (2l+1) -fold degeneracy.
and
N̄ 1 3 1
= = , H vibr = h̄ωv (n̂ + ) = vibration energy
V vi 4πri3 2

21
The vibrational degrees of freedom of the separation d of can be divided into terms
nuclei correspond at small amplitudes to a linear  
1
harmonic oscillator. n̂ = a† a = 0, 1, 2, . . . Each energy F tr
= −kB T ln tr N
(Z )
level is non-degenerate. N!
"  3 N#
1 2πmkB T 2
H el = electron orbital energy = −kB T ln V
N! h2
 
Corresponds to jumping of electrons from an orbital to V 3 3 2πm
= −kB T N ln + 1 + ln kB T + ln 2
another and ionization. Characteristic energy levels are N 2 2 h
>
∼1eV ≈ kB 104 K, thus, in normal circumstances these
(∞ )
X Tr
degrees of freedom are frozen and can be neglected. H nucl F rot = −N kB T ln (2l + 1)e− T l(l+1)
corresponds to energy related to nucleonic spin l=0
 
interactions. The spin degeneracy is vibr Tv
gy = (2I1 + 1)(2I2 + 1), where I1 and I2 are the spins of F = N kB T ln 2 sinh
2T
the nuclei
Fnucl = −N kB T ln gy .
Energy terms are effectively decoupled at low
temperatures, i.e. the energy Ei of the state i is The internal energy is

Ei ≈ Etr + Erot + Evibr , ∂F

U = F + TS = F − T
  ∂T
so the partition sum of one molecule is 2 ∂ F
= −T ,
∞ X
∞ ∂T T
XX
Z1 = gy (2l + 1) ×
so the internal energy corresponding to tranlational
p l=0 n=0
degrees of freedom is
p2 h̄ 2
l(l+1)−βh̄ωv (n+ 21 )
e−β 2m −β 2I  tr 
tr 2 ∂ F 3
= Z tr Z rot Z vibr Z nucl , U = −T = N kB T
∂T T 2
i.e. the state sum can be factorized. Above and
3
X p2 V N kBCVtr =
Z tr = e−β 2m = 2
λ3T
p so we end up with the ideal gas result.
s
Since only F tr depends on volume V the pressure is
h2
λT =
2πmkB T ∂F ∂F tr N kB T
∞
p=− =− = ,
X Tr
∂V ∂V V
Z rot = (2l + 1)e− T l(l+1)

l=0
i.e. we end up with the ideal gas equation of state
h̄2 pV = N kB T.
Tr =
2IkB
∞
X 1
Z vibr = e−βh̄ωv (n+ 2 )
n=0
Rotation
 −1 Typical rotational temperatures
Tv Gas Tr
= 2 sinh
2T H2 85.4
h̄ωv N2 2.9
Tv =
kB NO 2.4
Z nucl = gy = (2I1 + 1)(2I2 + 1). HCl 15.2
Cl2 0.36
Approximatively (neglecting the multiple occupation of We see that Tr ≪ the room temperature.
states) the state sum of N molecules is T ≪ Tr
Now
1 N ∞
ZN = Z , X
N! 1 Z rot = (2l + 1)e− T
Tr
l(l+1)
≈ 1 + 3e−2
Tr
T ,
l=0
where 1/N ! takes care of the identity of molecules. We
associate this factor with the tranlational sum. The free so the corresponding free energy is
energy
Tr
F = −kB T ln ZN F rot ≈ −3N kB T e−2 T

22
and the internal energy Now the free energy is
 
∂ F rot Tr Tv
U rot = −T 2 ≈ 6N kB Tr e−2 T . F vibr ≈ N kB T ln
∂T T T

Rotations contribute to the heat capacity like and the internal energy correspondingly

Tr
2
Tr U vibr ≈ N kB T,
CVrot ≈ 12N kB e−2 T → 0.
T T →0
so the heat capacity is
T ≫ Tr
Now CVvibr ≈ N kB .
Z ∞
Tr We see that in the limit T ≫ Tv two degrees of freedom
Z rot ≈ dl (2l + 1)e− T l(l+1)
0 are associated with vibrations like always with harmonic
,∞ oscillators (E = hT i + hV i = 2 hT i).
C Vr o t
T Tr T
= − e− T l(l+1)
= , N k B r o o m
Tr Tr te m p e r a tu r e
0 io n iz a tio n
d is s o c ia tio n
e tc .
so the free energy is
T T
F rot ≈ −N kB T ln T r T v
Tr
and the internal energy Rotation of homopolar molecules
The symmetry due to the identical nuclei must be taken
U rot ≈ N kB T. into account.
Example H2 -gas:
The contribution to the heat capacity is The nuclear spins are
1 1
CVrot ≈ N kB = f rot N kB , I1 = I2 = ,
2 2
or in the limit T ≫ Tr there are f rot = 2 rotational so the nuclei are fermions and the total wave function
degrees of freedom. must be antisymmetric. The total spin of the molecule is
Precisely: I = 0, 1. We consider these two cases:
C V
r o t

N k B
↑↑ ↑↓
I=1 I =0
Iz = −1, 0, 1 Iz = 0
triplet singlet
T
T r orthohydrogen parahydrogen
spin spin wave
Vibration wavefunctions function
Typical vibrational temperatures: symmetric: antisymmetric:
Gas Tv |1 1i = |↑↑i
H2 6100 |1 0i = √12 (|↑↓i + |↓↑i) |0 0i = √1 (|↑↓i − |↓↑i)
N2 3340 2
|1−1i = |↓↓i
NO 2690 Space wave Space wave
O2 2230 function function
HCl 4140 antisymmetric: symmetric:
We see that Tv ≫ the room temperature. (−1)l = −1 (−1)l = 1
T ≪ Tv
The free energy is The corresponding partition functions are
h Tv Tv
i X
F vibr = N kB T ln e 2T (1 − e− T )
Tr
Zortho = (2l + 1)e− T l(l+1)

1 Tv
l=1,3,5,...
≈ N kB Tv − N kB T e− T , X Tr
2 Zpara = (2l + 1)e− T l(l+1)

so l=0,2,4,...
 2
Tv − TTv
CVvibr ≈ N kB e . and the partition function associated with rotation is
T
T ≫ Tv Z rot = 3Zortho + Zpara .

23
When T ≫ Tr collisions cause conversions between ortho Bosons
and para states so the system is in an equilibrium. In For bosons the creation and annihilation operators obey
addition Zorto ≈ Zpara , so all 4 spin states are equally the commutation relations
probable.
When T ∼ <
Tr the gas may remain as an metastable mixture [ai , a†j ] = δij
of ortho and para hydrogens. In the mixture the ratio of [ai , aj ] = [a†i , a†j ] = 0.
the spin populations is 3 : 1. Then we must use the
partion sum It can be shown that
3N N
rot
ZN = Zorto
4
Zpara
4
. √
ai |n1 , . . . , ni , . . .i = ni |n1 , . . . , ni − 1, . . .i
The internal energy is now √
a†i |n1 , . . . , ni , . . .i = ni + 1 |n1 , . . . , ni + 1, . . .i .
3 orto 1 para
U rot = U + U The (occupation) number operator
4 4
and the heat capacity correspondingly n̂i = a†i ai
3 orto 1 para
C rot = C + C . obeys the relation
4 4
n̂i |n1 , . . . , ni , . . .i = a†i ai |n1 , . . . , ni , . . .i
7.4. Occupation number representation = ni |n1 , . . . , ni , . . .i
Let us recall the Fock space states, which are vectors in
and ni = 0, 1, 2, . . ..
Hilbert space
An arbitrary one particle operator, i.e. an operator O(1) ,
F = H(0) (1) (N ) which in the configuration space operates only on the
p ⊕ Hp ⊕ . . . Hp ⊕ ...
coordinates on one particle, can be written in the
occupation number representation as
with H(N
p
)
is N -particle Hilbert space. Here actually
(N )
Hp = SH(N ) is symmetrised for bosons and AH(N ) XD E
Ô(1) = i O(1) j a†i aj .
antisymmetrised for fermions.
i,j
Denote by
|n1 , n2 , . . . , ni , . . .i A two body operator O(2) can be written as
the quantum state where there are ni particles in the 1 XD E
particle state i. Let the energy of the state i be ǫi . Then Ô(2) = ij O(2) kl a†i a†j al ak .
ijkl
!
X
H |n1 , n2 , . . .i = ni ǫi |n1 , n2 , . . .i Example: Hamiltonian
i

N =
X
ni .
X h̄2 2 1 X
H= − ∇ + V (ri , rj )
i i
2m i 2
i6=j

We define the creation operator a†i so that takes in the occupation representation the form

a†i |n1 , n2 , . . . , ni , . . .i = C |n1 , n2 , . . . , ni + 1, . . .i X h̄2 2


H = i − ∇ j a†i aj
i,j
2m
i.e. a†i creates one particle into the state i.
1 X
Correspondingly the annihilation operator ai obeys: + hij| V |kli a†i a†j al ak ,
2
ijkl
ai |n1 , n2 , . . . , ni , . . .i = C ′ |n1 , n2 , . . . , ni − 1, . . .i ,
where
i.e. ai removes one particle from the state i.  
h̄2 2 h̄2 ∗
The basis {|n1 , n2 , . . .i} is complete, i.e. i − ∇ j =− φ (r)∇2 φj (r) d3 r
2m 2m i
X
|n1 , n2 , . . .i hn1 , n2 , . . .| = 1 and
{ni }

hij| V |kli =
and orthonormal or Z
φ∗i (r1 )φ∗j (r2 )V (r1 , r2 )φk (r2 )φl (r1 ) d3 r1 d3 r2 .
hn′1 , n′2 , . . . | n1 , n2 , . . .i = δn1 n′1 δn2 n′2 · · · .

24
 1
Fermions 2 -integer,
the particle is fermion with antisymmetric wave
The creation and annihilation operators of fermions function.
satisfy the anticommutation relations In bosonic systems the occupations of one particle states
are nl = 0, 1, 2, . . .. Only the grand canonical ensemble is
{ai , a†j } = ai a†j + a†j ai = δij simple to calculate. The grand canonical state sum is
{ai , aj } = {a†i , a†j } = 0.
ZG,BE = Tr e−β(Ĥ−µN̂ )
∞ D
It can be shown that X E
= n1 n2 · · · e−β(Ĥ−µN̂ ) n1 n2 · · ·
ai |n1 , . . . , ni , . . .i = {n}=0
 √ ∞
(−1)Si ni |n1 , . . . , ni − 1, . . .i , if ni = 1 X P
= e−β l
nl (ǫl −µ)
0, otherwise
{n}=0
a†i |n1 , . . . , ni , . . .i = "∞ #
 √ Y X
−βn(ǫl −µ)
(−1)Si ni + 1 |n1 , . . . , ni + 1, . . .i , if ni = 0 = e
0, otherwise l n=0
Y 1
Here = .
1 − e−β(ǫl −µ)
Si = n1 + n2 + · · · + ni−1 . l

Note: In Maxwell-Boltzmann we had

The number operator satisfies
W ({n}) = N1 ! n1 !n
N!
2 !...
; here we effectively have
n̂i |n1 , . . . , ni , . . .i = ni |n1 , . . . , ni , . . .i W ({n}) = 1, which is the correct way to symmetrise
multiple occupation numbers.
and ni = 0, 1. The grand potential is
One and two body operators take the same form as in the X h i
case of bosons. Note: Since ai and aj anticommute one ΩBE = kB T ln 1 − e−β(ǫl −µ) .
must be careful with the order of the creation and l

annihilation operators in O(2) . The occupation number of the state l is

In the case of non-interacting particles the Hamiltonian P
1 XX
operator in the configuration space is n̄l = hnl i = · · · nl e−β m nm (ǫm −µ)
ZG n n
X 1 2
H= H1 (ri ), 1 ∂ ∂Ω
i = − ln ZG = ,
β ∂ǫl ∂ǫl
where 1-particle Hamiltonian H1 is and for the Bose-Einstein occupation number we get
h̄2 2 1
H1 (ri ) = − ∇ + U (ri ). n̄l = .
2m i eβ(ǫl −µ) −1
Let φj be eigenfunctions of H1 i.e.
Entropy
H1 φj (r) = ǫj φ(r). Because dΩ = −S dT − p dV − N dµ we have
 
In the occupation space we have then ∂Ω
S =
X X ∂T µ,V
Ĥ = ǫj a†j aj = ǫj n̂j X h i
j j = −kB ln 1 − e−β(ǫl −µ)
l
and X X X 1 −1
N̂ = a†j aj = n̂j . −kB T (ǫl − µ)e−β(ǫl −µ) .
1− e−β(ǫl −µ) kB T 2
j j l

The grand canonical partition function is now Now

1
P eβ(ǫl −µ) = 1 +
XX n̄l
ZG = Tr e−β(Ĥ−µN̂ ) = · · · e−β l nl (ǫl −µ) .
n1 n2 and
β(ǫl − µ) = ln(1 + n̄l ) − ln n̄l ,
so
7.5. Bose-Einstein (BE) ideal gas X 
n̄l

In relativistic QM it can be shown that the spin of the S = −kB ln 1 −
particle is connected to the statistics: if the spin is n̄l + 1
l
X
integer, s = 0, 1, . . ., the particle is boson and the wave +kB n̄l [ln(n̄l + 1) − ln n̄l ]
function is symmetric wrt. permutations. If the spin is l

25
 n
or l
k BT
X 1
S = kB [(n̄l + 1) ln(n̄l + 1) − n̄l ln n̄l ] .
l

7.6. Fermi-Dirac ideal gas e l

m
The Hamiltonian operator is
X † The expectation value of the square of the occupation
Ĥ = ǫ l al al number will be
l
1
and the number operator n2l = Tr n̂2l e−β(Ĥ−µN̂ )
ZG,FD
X 1 1
N̂ = a†l al . 1 X X P
= · · · n2l e−β l′
nl′ (ǫl′ −µ)
l ZG,FD n1 =0 n2 =0
Now
1 1 ∂ 2 ZG,FD
{al , a†l′ } = δll′ =
β 2 ZG,FD ∂ǫl 2
and  
{al , al′ } = {a†l , a†l′ } = 0. 1 1 Yh i
= −  1 + e−β(ǫl′ −µ) 
The eigenvalues of the number operator related to the β ZG,FD ′ l 6=l
state l,
∂ −β(ǫl −µ)
n̂l = a†l al , × e
∂ǫl
are
e−β(ǫl −µ)
nl = 0, 1. = = n̄l .
1 + e−β(ǫl −µ)
The state sum in the grand canonical ensemble is
This is natural, since n2l = nl . For the variance we get
ZG,FD
2
= Tr e−β(Ĥ−µN̂ ) (∆nl )2 = n2l − hnl i = n̄l − n̄2l
1 1
X X D E = n̄l (1 − n̄l ).
= · · · n1 n2 · · · e−β(Ĥ−µN̂ ) n1 n2 · · ·
D n
k BT
n1 =0 n2 =0
l
1
X 1
X P
−β nl (ǫl −µ)
= ···e l

n1 =0 n2 =0
( 1
)
e
Y X
−βn(ǫl −µ) l
= e m
l n=0
Yh i There are fluctuations only in the vicinity of the chemical
= 1 + e−β(ǫl −µ) . potential µ.
l The entropy is
The grand potential is
∂Ω
X h i S = −
ΩFD = −kB T ln 1 + e−β(ǫl −µ) . ∂T h i
X
l = kB ln 1 + e−β(ǫl −µ)
l
The average occupation number of the state l is =n̄l
}| { z
1 1 X e−β(ǫl −µ)
n̄l = hnl i = Tr n̂l e−β(Ĥ−µN̂ ) + (ǫl − µ).
ZG,FD T 1 + e−β(ǫl −µ)
1 1 l
1 X X P
= · · · nl e−β l′
nl′ (ǫl′ −µ)
1−n̄l 1
ZG,FD Now β(ǫl − µ) = ln n̄l and 1 + e−β(ǫl −µ) = 1−n̄l , so
n1 =0 n2 =0
1 ∂ ln ZG,FD ∂ΩFD X
= − = S = −kB [(1 − n̄l ) ln(1 − n̄l ) + n̄l ln n̄l ] .
β ∂ǫl ∂ǫl
−β(ǫl −µ) l
e
= .
1 + e−β(ǫl−µ)
Thus the Fermi-Dirac occupation number can be written
as
1
n̄l = β(ǫ −µ) .
e l +1

26