Statistical Physics

ETH Zürich – Herbstsemester 2014

Manfred Sigrist, HIT K23.8

Tel.: 3-2584, Email: sigrist@itp.phys.ethz.ch
Webpage: http://www.itp.phys.ethz.ch/research/condmat/strong/

Literature:

• Statistical Mechanics, K. Huang, John Wiley & Sons (1987).

• Introduction to Statistical Physics, K. Huang, Chapman & Hall Books (2010).

• Equilibrium Statistical Physics, M. Plischke and B. Bergersen, Prentice-Hall International

(1989).

• Theorie der Wärme, R. Becker, Springer (1978).

• Thermodynamik und Statistik, A. Sommerfeld, Harri Deutsch (1977).

• Statistische Physik, Vol. V, L.D. Landau and E.M Lifschitz, Akademie Verlag Berlin
(1975).

• Statistische Mechanik, F. Schwabl, Springer (2000).

• Elementary Statistical Physics, Charles Kittel,John Wiley & Sons (1967).

• Statistical Mechanics, R.P. Feynman, Advanced Book Classics, Perseus Books (1998).

• Statistische Mechanik, R. Hentschke, Wiley-VCH (2004).

• Introduction to Statistical Field Theory, E. Brézin, Cambridge (2010).

• Statistical Mechanics in a Nutshell, Luca Peliti, Princeton University Press (2011).

• Principle of condensed matter physics, P.M. Chaikin and T.C. Lubensky, Cambridge Uni-
versity Press (1995).

• Many other books and texts mentioned throughout the lecture

Lecture Webpage:

http://www.itp.phys.ethz.ch/education/hs14/StatPhys

1
Contents

1 Kinetic approach to statistical physics 5

1.1 Time evolution and Master equation . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.1.1 H-function and information . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.1.2 Simulation of a two-state system . . . . . . . . . . . . . . . . . . . . . . . 9
1.1.3 Equilibrium of a system . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.2 Analysis of a closed system . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.2.1 H and the equilibrium thermodynamics . . . . . . . . . . . . . . . . . . . 12
1.2.2 Master equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.2.3 Irreversible processes and increase of entropy . . . . . . . . . . . . . . . . 14

2 Classical statistical Physics 18

2.1 Gibbsian concept of ensembles . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.1.1 Liouville Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.1.2 Equilibrium system . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.2 Microcanonical ensemble . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.2.1 Entropy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.2.2 Relation to thermodynamics . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.2.3 Ideal gas - microcanonical treatment . . . . . . . . . . . . . . . . . . . . . 23
2.3 Canonical ensemble . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.3.1 Thermodynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.3.2 Equipartition law . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.3.3 Ideal gas - canonical treatment . . . . . . . . . . . . . . . . . . . . . . . . 28
2.4 Grand canonical ensemble . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.4.1 Relation to thermodynamics . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.4.2 Ideal gas - grand canonical treatment . . . . . . . . . . . . . . . . . . . . 30
2.4.3 Chemical potential in an external field . . . . . . . . . . . . . . . . . . . . 31
2.5 Fluctuations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.5.1 Energy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.5.2 Particle number . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
2.5.3 Magnetization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34

3 Quantum Statistical Physics 35

3.1 Basis of quantum statistical physics . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.2 Density matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
3.3 Ensembles in quantum statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
3.3.1 Microcanonical ensemble . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
3.3.2 Canonical ensemble . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
3.3.3 Grand canonical ensemble . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
3.4 Ideal quantum paramagnet - canonical ensemble . . . . . . . . . . . . . . . . . . 40
3.4.1 Spin 1/2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
3.4.2 Spin S - classical limit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
3.5 Ideal quantum gas - grand canonical ensemble . . . . . . . . . . . . . . . . . . . . 43

2
 3.6 Properties of Fermi gas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
3.6.1 High-temperature and low-density limit . . . . . . . . . . . . . . . . . . . 47
3.6.2 Low-temperature and high-density limit: degenerate Fermi gas . . . . . . 47
3.6.3 Spin-1/2 Fermions in a magnetic field . . . . . . . . . . . . . . . . . . . . 49
3.7 Bose gas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
3.7.1 Bosonic atoms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
3.7.2 High-temperature and low-density limit . . . . . . . . . . . . . . . . . . . 50
3.7.3 Low-temperature and high-density limit: Bose-Einstein condensation . . . 51
3.8 Photons and phonons . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
3.8.1 Blackbody radiation - photons . . . . . . . . . . . . . . . . . . . . . . . . 56
3.8.2 Phonons in a solid . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
3.9 Diatomic molecules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60

4 Identical Quantum Particles - Formalism of Second Quantization 64

4.1 Many-body wave functions and particle statistics . . . . . . . . . . . . . . . . . . 64
4.2 Independent, indistinguishable particles . . . . . . . . . . . . . . . . . . . . . . . 65
4.3 Second Quantization Formalism . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
4.3.1 Creation- and annihilation operators . . . . . . . . . . . . . . . . . . . . . 66
4.3.2 Field operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
4.4 Observables in second quantization . . . . . . . . . . . . . . . . . . . . . . . . . . 69
4.5 Equation of motion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
4.6 Correlation functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
4.6.1 Fermions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
4.6.2 Bosons . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
4.7 Selected applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
4.7.1 Spin susceptibility . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
4.7.2 Bose-Einstein condensate and coherent states . . . . . . . . . . . . . . . . 77
4.7.3 Phonons in an elastic medium . . . . . . . . . . . . . . . . . . . . . . . . . 81

5 One-dimensional systems of interacting degrees of freedom 84

5.1 Classical spin chain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
5.1.1 Thermodynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
5.1.2 Correlation function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
5.1.3 Susceptibility . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
5.2 Interacting lattice gas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
5.2.1 Transfer matrix method . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
5.2.2 Correlation function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
5.3 Long-range order versus disorder . . . . . . . . . . . . . . . . . . . . . . . . . . . 92

6 Phase transitions 93
6.1 Ehrenfest classification of phase transitions . . . . . . . . . . . . . . . . . . . . . 93
6.2 Phase transition in the Ising model . . . . . . . . . . . . . . . . . . . . . . . . . . 95
6.2.1 Mean field approximation . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
6.2.2 Instability of the paramagnetic phase . . . . . . . . . . . . . . . . . . . . 97
6.2.3 Phase diagram . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
6.3 Gaussian transformation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
6.3.1 Correlation function and susceptibility . . . . . . . . . . . . . . . . . . . . 103
6.4 Ginzburg-Landau theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
6.4.1 Ginzburg-Landau theory for the Ising model . . . . . . . . . . . . . . . . 106
6.4.2 Critical exponents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
6.4.3 Range of validity of the mean field theory - Ginzburg criterion . . . . . . 109
6.5 Self-consistent field approximation . . . . . . . . . . . . . . . . . . . . . . . . . . 110
6.5.1 Renormalization of the critical temperature . . . . . . . . . . . . . . . . . 110

3
 6.5.2 Renormalized critical exponents . . . . . . . . . . . . . . . . . . . . . . . . 112
6.6 Long-range order - Peierls’ argument . . . . . . . . . . . . . . . . . . . . . . . . . 113
6.6.1 Absence of finite-temperature phase transition in the 1D Ising model . . . 113
6.6.2 Long-range order in the 2D Ising model . . . . . . . . . . . . . . . . . . . 114

7 Superfluidity 116
7.1 Quantum liquid Helium . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
7.1.1 Superfluid phase . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
7.1.2 Collective excitations - Bogolyubov theory . . . . . . . . . . . . . . . . . . 119
7.1.3 Gross-Pitaevskii equations . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
7.2 Berezinskii-Kosterlitz-Thouless transition . . . . . . . . . . . . . . . . . . . . . . 127
7.2.1 Correlation function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
7.2.2 Topological excitations and BKT transition . . . . . . . . . . . . . . . . . 129

4
Chapter 1

Kinetic approach to statistical

physics

Thermodynamics deals with the behavior and relation of quantities of macroscopic systems
which are in equilibrium. The concrete time evolution does not enter the discussions in thermo-
dynamics, although systems may change - quasi-statically - between di↵erent equilibrium states.
A macroscopic system, however, consists of many microscopic entities, many degrees of freedom,
such as moving atoms in a gas or magnetic / electric dipoles in magnets or ferroelectrics. There-
fore it seems very natural to analyze the time evolution of these entities to get insight into the
behavior of macroscopic systems whereby a statistical approach is most adequate, for the simple
reason that we cannot track down the temporal behavior of each entity separately.
In this chapter we will discuss a conceptual version of a many-body system to understand
how one can view non-equilibrium behavior and how equilibrium is reached and eventually
characterized. Entropy will play an essential role to give us the connection to thermodynamics
of the equilibrium state. This discussion will also make clear how the immense number of degrees
of freedom, N , is essential for the quality of the statistical approach and the emergence of the
laws of thermodynamics. The kinetic approach find one important application in kinetic gas
theory and Boltzmann’s transport theory which will not be covered here (part of the lecture
”Theorie der Wärme” in Bachelor of Physics).

1.1 Time evolution and Master equation

We start by considering a model with N units (atoms, ...) which can be in z di↵erent micro-states
(Fig.1.1):
{s⌫i } with i = 1, . . . , N ; ⌫ = 1, . . . , z . (1.1)
We may define the variable s⌫i such that if the unit i is in the state ⌫ 0 then
8
< 1 ⌫ = ⌫0 ,
⌫
si = (1.2)
:
0 ⌫ 6= ⌫ 0 .

Thus ~si is a vector with one component (⌫ 0 ) being 1 and the others 0. For the moment, these
degrees of freedom are considered to be independent and their z micro-states have the same
energy, when we introduce the time evolution of this system. For the purpose of a simple
simulation of the time evolution of these units, we introduce discrete, equally spaced time steps
tn with tn+1 tn = t. During each time step the micro-state of each unit can change from ⌫
to ⌫ 0 with a probability p⌫⌫ 0 which is connected with the transition rate ⌫⌫ 0 by

p⌫⌫ 0 = ⌫⌫ 0 t. (1.3)

5
 1"
8" 2"

7" 3"

6" 4"
5"
si

Figure 1.1: A schematic system (box) of N units (cirlce) with micro-states ⌫ = 1, . . . , 8. Each
unit can be in an individual state at a given time.

We require that the reverse processes have equal probability due to time reversal symmetry:
p⌫⌫ 0 = p⌫ 0 ⌫ and, thus, ⌫⌫ 0 = ⌫ 0 ⌫ . At a given time, we find N⌫ units of the system are in the
micro-state ⌫ (Fig.1.2)
XN Xz
⌫
N⌫ = si with N⌫ = N. (1.4)
i=1 ⌫=1

Thus, picking at random a unit i we would find it in the micro-state ⌫ with the probability,
z
X
N⌫
w⌫ = with w⌫ = 1 . (1.5)
N
⌫=1

1" 2" ………." z"

detailed"balance"

1" 2" ………." z"

Figure 1.2: Distribution of the units on di↵erent micro-states: Evolution towards the fixed
point with detailed balance where all micro-states are occupied by the same number of units on
average.

Let us now discuss the ”budget” of each micro-state ⌫ during the evolution in time: (1) the
number N⌫ is reduced, because some P units in micro-state ⌫ will have a transition to another
micro-state (⌫ ! ⌫ 0 ) with the rate ⌫ 0 ⌫⌫ 0 N⌫ ; (2) the number N⌫ increases, since
P some units in
another micro-state undergo a transition to the micro-state ⌫ with the rate ⌫ 0 ⌫ 0 ⌫ N⌫ 0 . Note
that these rates are proportional to the number of units in the micro-state which is transformed

6
into another micro-state, because each unit changes independently with the rate ⌫⌫ 0 . The
corresponding budget equation is given by
X X
N⌫ (tn+1 ) = N⌫ (tn ) t ⌫⌫ 0 N⌫ (tn ) + t ⌫ 0 ⌫ N⌫ 0 (tn ) . (1.6)
⌫ 0 6=⌫ ⌫ 0 6=⌫
| {z } | {z }
leaving state ⌫ entering state ⌫

This set of z iterative equations describes the evolution of the N units in a statistical way,
whereby we do not keep track of the state of each individual unit, but only of the number of
units in each micro-state. As we will see below (H-theorem), starting from an arbitrary initial
configuration {N⌫ } of the units at t1 = 0, this iteration moves generally towards a fixed point,
N⌫ (tn+1 ) = N⌫ (tn ) which requires that
X X
0= ⌫⌫ 0 N⌫ (tn ) ⌫ 0 ⌫ N⌫ 0 (tn ) . (1.7)
⌫ 0 6=⌫ ⌫ 0 6=⌫

There is no time evolution anymore for N⌫ (t), although the units are still changing their states
in time. As this equation is true for all ⌫ we find that independently

0= ⌫⌫ 0 N⌫ (tn ) ⌫ 0 ⌫ N⌫ 0 (tn ) , (1.8)

which means that for any pair of micro-states ⌫ and ⌫ 0 the mutual transitions compensate each
other. Equation (1.8) is known as the condition of detailed balance.1 Due to ⌫⌫ 0 = ⌫ 0 ⌫ in our
conceptual model, we find the fixed point condition N⌫ = N⌫ 0 = N/z, i.e all micro-states are
equally occupied.
We now slightly reformulate the iterative equation (1.6) taking the limit of t ! 0 and dividing
it by N :
dw⌫ X X
= w⌫ ⌫⌫ 0 + ⌫ 0 ⌫ w⌫ 0 , (1.14)
dt 0 0 ⌫ 6=⌫ ⌫ 6=⌫

which turns into a di↵erential equation. This is the so-called master equation of the system. Also
here it is obvious that the detailed balance condition leads to a solution where all probabilities
w⌫ (t) are constant in time:
w⌫ ⌫0⌫ 1
w⌫ ⌫⌫ 0 = w⌫ 0 ⌫0⌫ ) = = 1 , w⌫ = . (1.15)
w⌫ 0 ⌫⌫ 0 z
1
We show here that the detailed balance condition follows from rather general arguments even if the condition
⌫⌫ 0 = ⌫ 0 ⌫ is not satisfied. First, we define for the general case

N⌫ = ⌫ ⌫ such that R⌫⌫ 0 = ⌫⌫ 0 ⌫ = ⌫0⌫ ⌫0 = R⌫ 0 ⌫ , (1.9)

which leads to the fixed point condition Eq.(1.7),

X X
0= (R⌫⌫ 0 ⌫ R⌫ 0 ⌫ ⌫0 ) = R⌫⌫ 0 ( ⌫ ⌫0 ) . (1.10)
⌫0 ⌫0

Note that the condition ⌫⌫ 0 = ⌫ 0 ⌫ would imply ⌫ = ⌫ 0 . Assuming R⌫⌫ 0 > 0 for all pairs (⌫, ⌫ 0 ) (more generally
ergodicity is necessary, i.e. the system cannot be decomposed into disconnected subsystems), this equation can
only be solved by ⌫ = = const for all ⌫.
Proof: Assume that

1 = 2 = ··· = ⌫1 > ⌫1 +1 = ⌫1 +2 = ··· = ⌫2 > ⌫2 +1 = ··· > ··· = z >0 (1.11)

then for ⌫  ⌫1 we find X X

R⌫⌫ 0 ( ⌫ ⌫0 ) = R⌫⌫ 0 ( ⌫ ⌫0 ) >0, (1.12)
⌫0 ⌫ 0 >⌫1

such that we cannot satisfy Eq.(1.10) and only the solution ⌫ = = const for all ⌫ is possible. Thus, using
Eq.(1.9) it follows that
R⌫⌫ 0 ⌫ R⌫ 0 ⌫ ⌫ 0 = 0 ) ⌫⌫ 0 N⌫ ⌫ 0 ⌫ N⌫ 0 =0, (1.13)
and the detailed balance condition is satisfied.

7
We can define a mean value of a property or quantity ”↵” associated with each micro-state ⌫,
which we call ↵⌫ : X
h↵i = ↵⌫ w ⌫ . (1.16)
⌫

Such mean values describe the macro-state of the system. The time evolution of this quantity
is easily derived,
2 3
d X dw⌫ X X X
h↵i = ↵⌫ = ↵⌫ 4 w⌫ ⌫⌫ 0 + ⌫ 0 ⌫ w⌫ 0
5
dt ⌫
dt ⌫ 0 0
⌫ 6=⌫ ⌫ 6=⌫
(1.17)
1 X
= {↵⌫ ↵⌫ 0 }(w⌫ w⌫ 0 ) ⌫⌫ 0 ,
2 0
⌫,⌫

where the last equality is obtained by symmetrization using ⌫⌫ 0 = ⌫0⌫ . Obviously the time
evolution of h↵i stops when detailed balance is reached.

1.1.1 H-function and information

We introduce the function H(t) which has the following form
X
H(t) = w⌫ ln w⌫ . (1.18)
⌫

This function can be viewed as a measure for the lack of our knowledge about the microscopic
state of our system. Concretely, if at the time t we pick one atom at random, we may ask how
well we can predict to find the atom in a given micro-state. Assume that w1 = 1 and w⌫ = 0
for ⌫ = 2, . . . , z. Then we can be sure to find the micro-state 1. In this case H(t) = 0. For
generic distributions of probabilities H(t) > 0. The larger H(t) the less certain the outcome of
our picking experiment is, i.e. the less information is available about the system.
It is interesting to follow the time dependence of H(t),

dH(t) X dw⌫ X
= (ln w⌫ + 1) = ⌫⌫ 0 (w⌫ w⌫ 0 )(ln w⌫ + 1)
dt dt
⌫ ⌫,⌫ 0
(1.19)
1X
= (ln w⌫ ln w⌫ 0 )(w⌫ w⌫ 0 ) ⌫⌫ 0 0.
2 0
⌫,⌫

Note that (x y)(ln x ln y) 0. This implies that H(t) evolves in a specific direction in time,
despite the assumption of time reversal symmetry, which was an essential part in reaching the
symmetrized form of Eq.(1.19). The condition dH/dt = 0 then requires w⌫ = w⌫ 0 , corresponding
to detailed balance. The inequality (1.19) suggests that detailed balance corresponds to the
macroscopic state of the system with maximal possible H and, thus, the maximal uncertainty
of knowledge.
Let us consider this idea from a slightly di↵erent view point. For given w⌫ = N⌫ /N in a system
of N units, there are
N!
= W (N⌫ ) (1.20)
N1 !N2 ! · · · Nz !
realizations among z N possible configurations. In the large-N limit we may apply Stirling
formula
1
ln n! ⇡ n ln n n + ln 2⇡n, (1.21)
2

8
which then leads to
X
ln W (N⌫ ) ⇡ N ln N N [N⌫ ln N⌫ N⌫ ]
⌫
X (1.22)
= N w⌫ ln w⌫ = N H(t) .
⌫

where we have ignored terms of the order ln N . Thus, H measures for given set {N⌫ } the number
of available configurations of the units.
Now we show the maximal H corresponds indeed to the detailed balance condition being then
equivalent to the set {N⌫ } with the largest possible number of configurations. For this purpose
we maximize H as a functional of w⌫ through the following extended form of H
( )
X X
0
H = w⌫ ln w⌫ + w⌫ 1 , (1.23)
⌫ ⌫

where we introduced the constraint of normalization of the probability distribution using a

Lagrange multiplier. Thus, searching the maximum with respect to w⌫ and , we get to the
equation
dH 0 X
= ln w⌫ 1 + = 0 and w⌫ = 1 , (1.24)
dw⌫ ⌫
which can only be solved by w⌫ = e 1 = 1/z and constitutes again the condition of detailed
balance.
The conclusion of this discussion is that the detailed balance describes the distribution w⌫ with
the highest probability, i.e. the largest number of di↵erent possible realizations yield the same
{N1 , . . . , Nz }. This is indeed the maximal value of W (N⌫ ) = N !/{(N/z)!}z , the number of
realizations. In its time evolution the system, following the master equation, moves towards the
fixed point with this property and H(t) grows monotonously until it reaches its maximal value.

1.1.2 Simulation of a two-state system

The master equation (1.14), introduced above, deals with a system of independent units in a
statistical way. It corresponds to a rate equation. We would like now to consider a ”concrete”
model system and analyze how it evolves in time both analytically by solving the master equation
and numerically by simulating a system with a not so large number of units. This will be
most illustrative in view of the apparent problem we encounter by the fact that our model
conserves time reversal symmetry. Nevertheless, there is an apparent preference in the direction
of evolution.
We consider a system of N units with two di↵erent micro-states (z = 2) to perform a computa-
tional simulation. We describe the micro-state of unit i by the quantity
8
< +1 ⌫ = 1
mi = (1.25)
:
1 ⌫=2.
This may be interpreted as a magnetic moment point up (+1) or down ( 1). The transition
between the two states corresponds simply to a sign change of mi , which happens with equal
probability in both directions (time reversal symmetry). The computational algorithm works in
the following way. Within a time step each unit experiences a transition with a probability p
(0 < p < 1), whereby the probabilistic behavior is realized by a random number generator. This
means generate a uniformly distributed random number R 2 [0, 1] and then do the following
process for all units: 8
< mi (tn ) if p < R < 1
mi (tn+1 ) = (1.26)
:
mi (tn ) if 0  R  p .

9
This time series corresponds to a so-called Markov chain which means that the transition at
each step is completely independent from previous steps.
After each time step we ”measure” the quantities
N 2
1 X X
M (t) = mi (t) and H(t) = w⌫ (t) ln w⌫ (t) (1.27)
N
i=1 ⌫=1

where
1 1
w1 (t) = {1 + M (t)} and w2 (t) = {1 M (t)} . (1.28)
2 2
Note that M (t) may be interpreted as a average magnetization per unit. The results of the
simulations for both quantities are shown in Fig.1.3. In the starting configuration all units have
the micro-state mi = +1 (completely polarized). There is only one such configuration, while the
fixed point distribution is characterized by w1 = w2 = 1/2 with a large number of realizations
r
N! N 2
W = ⇡2 for large N , (1.29)
{(N/2)!}2 ⇡N

following Stirling’s formula. This corresponds to M = 0. The closer to the fixed point distribu-
tion the system is the more realizations are available. We may estimate this by
N! N!
W (M ) = =
N1 !N2 ! {N/2(1 + M )}!{N/2(1 M )}!

N 1 2
) ln W (M ) ⇡ N ln 2 {(1 + M ) ln(1 + M ) + (1 M ) ln(1 M )} + ln
2 2 ⇡N (1 M 2 )

1 2 M 2N
⇡ N ln 2 + ln
2 ⇡N 2

r
N 2 M 2 N/2
) W (M ) ⇡ 2 e ,
⇡N
(1.30)
where we have used the Stirling formula for large N and the expansion for M ⌧ 1 (ln(1 + x) =
2
x x2 + · · · ). The number of realizations drop quickly from the maximal value at M = 0
p
with a width M = 2/N . The larger N the larger the fraction of realizations of the system
belonging to the macroscopic state with M ⇡ 0.
This explains the overall trend seen in the simulations. The ”measured” quantities approach a
fixed point, but show fluctuations. These fluctuations shrink with increasing system size N . The
relation found from (1.19) is not strictly true, when we consider a specific time series of a finite
system. The master equations leading to (1.19) are statistical and consider an averaged situation
for very large systems (N ! 1). The ”fixed point” reached after a long time corresponds to the
situation that each configuration of micro-states is accessed with the same probability. Macro-
scopic states deviating from the fixed point occupy less and less of the available configuration
space for increasing N .
The master equations of our two-state system are
dw1 dw2
= (w2 w1 ) and = (w1 w2 ) , (1.31)
dt dt
where = 12 = 21 . Adding and subtracting the two equations leads to
d d
(w1 + w2 ) = 0 and (w1 w2 ) = 2 (w1 w2 ) , (1.32)
dt dt

10
respectively. The first equation states the conservation of the probability w1 + w2 = 1. The
second leads to the exponential approach of M (t) to 0,
2t
M (t) = M0 e , (1.33)

with M0 = M (0) = 1, from which we also obtain

1 2t 2t 1 2t 2t 4t
H(t) = ln 2 (1 + e ) ln(1 + e ) (1 e ) ln(1 e ) ⇡ ln 2 e . (1.34)
2 2
where the last approximation holds for times t 1/4 and shows the relaxation to the unpo-
larized situation. We see that the relaxation times for M and H di↵er by a factor two
1 1
⌧M = and ⌧H = (1.35)
2 4
which can be observed also roughly by eye in the results of our simulation (Fig.1.3.).

0.75 N=100 N=1000 N=10000

0.5
M(t)

0.25

-0.25

0.75

0.5
H(t)

0.25

0
0 100 200 0 100 200 0 100 200
tn tn tn

Figure 1.3: Simulation of a two-state system for di↵erent numbers of units N : 1s t column,
N = 100; 2nd column, N = 1000; 3r d column, N = 10000. Note that for sufficiently large t
( /p), M (t) ! 0 and H(t) ! ln 2.

1.1.3 Equilibrium of a system

The statistical discussion using the master equation and our simulation of a system with many
degrees of freedom shows that the system relaxes towards a fixed point of the probability distribu-
tion w⌫ which describes its macroscopic state independent of time. This probability distribution
of the micro-states accounts for averaged properties, and allows us to calculate average values
of certain physical quantities related to the micro-states. This fixed point is equivalent to the

11
equilibrium of the system. Deviations from this equilibrium yield non-equilibrium states of the
system, which then relax towards the equilibrium, if we allow the system to do so. The equi-
librium state is characterized in a statistical sense as the state with the maximal number of
realizations in terms of configurations of micro-states of the units. We now aim at getting a
connection with macroscopic thermodynamic properties of the system in equilibrium.

1.2 Analysis of a closed system

We turn to a system of N units whose states possess di↵erent energies, ✏⌫ . Again we describe
the macro-state of the system by means of the probability distribution w⌫ = N⌫ /N . The system
shall be closed, such that no matter and no energy is exchanged with the environment. Thus,
the total energy and the number of units is conserved:
z
X z
X
h✏i = w⌫ ✏⌫ and 1= w⌫ , (1.36)
⌫=1 ⌫=1

which define the internal energy U = N h✏i = N u unchanged in a closed system, and the
normalization of the probability distribution.

1.2.1 H and the equilibrium thermodynamics

As before we define the function H(t), an increasing function of time, saturating at a maximal
value, which we consider as the condition for ”equilibrium”. Time evolution and aspects of
non-equilibrium will be addressed later. Rather we search now for the maximum of H with
respect to w⌫ under the condition of energy conservation. Thus, we introduce again Lagrange
multipliers for these constraints
X X 1 X
H(w⌫ ) = w⌫ ln w⌫ + { w⌫ 1} { w⌫ ✏⌫ h✏i} (1.37)
⌫ ⌫
✓ ⌫

and obtain the equation

dH ✏⌫
0= = ln w⌫ 1+ . (1.38)
dw⌫ ✓
This equation leads to
X
1 ✏⌫ /✓
w⌫ = e with e1 = e ✏⌫ /✓
=Z (1.39)
⌫

which satisfies the normalization condition for w⌫ .

We aim now at giving di↵erent quantities a thermodynamic meaning. We begin by multiplying
Eq.(1.38) by w⌫ and sum then over ⌫. This leads to
Xn ✏⌫ o h✏i
0= w⌫ ln w⌫ + w⌫ (1 ) + w⌫ =H 1+ (1.40)
⌫
✓ ✓

which we use to replace 1 in (1.38) and obtain

h✏i = ✓(ln w⌫ + H) + ✏⌫ . (1.41)

The di↵erential reads

✓ ◆
dw⌫
dh✏i = (H + ln w⌫ )d✓ + ✓ dH + + d✏⌫ . (1.42)
w⌫

12
After multiplying by w⌫ and a ⌫-summation the first term on the right hand side drops out and
we obtain X
dh✏i = ✓dH + w⌫ d✏⌫ . (1.43)
⌫
P ⌫
Now we view h✏i = u as the internal energy per unit and d✏⌫ = i @✏
@qi dqi where qi is a generalized
@✏⌫
coordinate, such as a volume, a magnetization etc. Thus @qi = Fi⌫ is a generalized force, such
as pressure, magnetic field etc. Therefore we write (1.43) as
X X du 1 X
du = ✓dH w⌫ Fi⌫ dqi = ✓dH hFi idqi ) dH = + hFi idqi . (1.44)
✓ ✓
⌫,i i i

This suggests now to make the following identifications:

s
✓ = kB T and H= , (1.45)
kB
where ✓ represents the temperature T and H the entropy density s. Naturally we identify the
relations, ✓ ◆ ✓ ◆ ✓ ◆
@S @s 1 @s
= = and = hFi i . (1.46)
@U qi @u qi T @qi u
In this way we have made a sensible connection to thermodynamics. Below we will use the
H-function as equivalent to the entropy.

1.2.2 Master equation

Because the micro-states have di↵erent energies, the time evolution as we have discussed by the
master equation for completely independent units has to be modified in a way so as to conserve
the energy in each time step. This can be guaranteed only by involving two micro-states of
di↵erent units to be transformed together under the constraint of overall energy conservation.
A master equation doing this job has the form
dw⌫ X
= { ⌫⌫1 ;⌫2 ⌫3 w⌫2 w⌫3 ⌫2 ⌫3 ;⌫⌫1 w⌫ w⌫1 } . (1.47)
dt ⌫ ,⌫ ,⌫
1 2 3

For this purpose we assume that there is no correlation between the states of di↵erent units.
Here ⌫⌫1 ;⌫2 ⌫3 denotes the rate for the transition of the pair of states (⌫2 , ⌫3 ) to (⌫, ⌫1 ) under the
condition that ✏⌫2 + ✏⌫3 = ✏⌫ + ✏⌫1 . Time reversal symmetry and the freedom to exchange two
states give rise to the following relations:

⌫⌫1 ;⌫2 ⌫3 = ⌫2 ⌫3 ;⌫⌫1 = ⌫1 ⌫;⌫2 ⌫3 = ⌫⌫1 ;⌫3 ⌫2 . (1.48)

This can be used in order to show that the H-function is only increasing with time,
dH X dw⌫
= {ln w⌫ + 1}
dt ⌫
dt
(1.49)
1 X
= ⌫,⌫1 ;⌫2 ⌫3 (w⌫ w⌫1 w⌫2 w⌫3 ) {ln(w⌫ w⌫1 ) ln(w⌫2 w⌫3 )} 0.
4 ⌫,⌫ ,⌫ ,⌫
1 2 3

The equilibrium is reached when dH/dt = 0 and using (1.39) leads to,

e ✏⌫ /kB T e ✏⌫1 /kB T e ✏⌫2 /kB T e ✏⌫3 /kB T

w ⌫ w ⌫1 = = = w ⌫2 w ⌫3 . (1.50)
Z2 Z2

13
We now revisit the condition of detailed balance under these new circumstances. On the one
hand, we may just set each term in the sum of the right hand side of (1.47) to zero to obtain a
detailed balance statement:

0= ⌫⌫1 ;⌫2 ⌫3 w⌫2 w⌫3 ⌫2 ⌫3 ;⌫⌫1 w⌫ w⌫1 = ⌫⌫1 ;⌫2 ⌫3 {w⌫2 w⌫3 w ⌫ w ⌫1 } , (1.51)

whereby the second equality is a consequence of time reversal symmetry. On the other hand,
we may compress the transition rates in order to reach at a similar form as we had it earlier in
(1.8),
0 0
⌫⌫ 0 w⌫ = ⌫ 0 ⌫ w⌫ 0 . (1.52)
where we define
X X
0 0
⌫⌫ 0 = ⌫⌫1 ;⌫ 0 ⌫2 w⌫1 and ⌫0⌫ = ⌫⌫1 ;⌫ 0 ⌫2 w⌫2 . (1.53)
⌫1 ,⌫2 ⌫1 ,⌫2

It is important now to notice that time reversal does not invoke 0 = 0 , but we find that
⌫⌫ 0 ⌫0⌫
at equilibrium
X e ✏⌫2 /kB T X e ✏⌫1 /kB T
0 (✏⌫ ✏⌫ 0 )/kB T (✏⌫ ✏⌫ 0 )/kB T 0
⌫0⌫ = ⌫⌫1 ;⌫ 0 ⌫2 =e ⌫⌫1 ;⌫ 0 ⌫2 =e ⌫⌫ 0 .
⌫1 ,⌫2
Z ⌫1 ,⌫2
Z
(1.54)
Thus detailed balance implies here di↵erent transition rates for the two directions of processes
depending on the energy di↵erence between the two micro-states, ✏⌫ ✏⌫ 0 , and the temperature
T,
0
w⌫ 0
= ⌫0 ⌫ = e (✏⌫ ✏⌫ 0 )/kB T . (1.55)
w⌫ 0 ⌫⌫ 0
The degrees of freedom of each unit fluctuate in the environment (heat reservoir) of all the other
units which can interact with it.

1.2.3 Irreversible processes and increase of entropy

Although thermodynamics shows its strength in the description of reversible processes, it pro-
vides also important insights into the irreversible changes. An example is the Gay-Lussac ex-
periment. However, the nature of the irreversible process and in particular its time evolution is
not part of thermodynamics usually. A process which considers the evolution of a system from
a non-equilibrium to its equilibrium state is irreversible generally.
Here we examine briefly some aspects of this evolution by analyzing a closed system consisting
of two subsystems (Fig.1.4) , 1 and 2, which are initially independent (inhibited equilibrium) but
then at t = 0 become very weakly coupled. Both systems shall have the same type of units with
(1) (1) (2) (2)
their micro-states which they occupy with probabilities w⌫ = N⌫ /N1 and w⌫ = N⌫ /N2 ,
respectively (Ni is the number of units in system i). Each subsystem is assumed to be in its
equilibrium internally which can be sustained approximatively as long as the coupling between
them is weak.
The number of configurations for the combined system is given by the product of the number
of configurations of the two subsystems:
Ni !
W = W1 W2 with Wi = (i) (i) (i)
(1.56)
N1 !N2 ! · · · Nz !
which then leads to the entropy
S = kB ln W = kB ln W1 + kB ln W2 = S1 + S2
X X (1.57)
= N1 kB w⌫(1) ln w⌫(1) N2 kB w⌫(2) ln w⌫(2) ,
⌫ ⌫

14
 1 ~ 2
U

T1 T2

Figure 1.4: A closed system consisting of two weakly coupled subsystems, 1 and 2, with initial
temperatures T1 and T2 respectively, evolves towards equilibrium through energy transfer from
the warmer to the colder subsystem.

using Eq.(1.22) and (1.45). In the complete system the internal energy is conserved, U = U1 +U2 .
We define U01 and U02 as the internal energy of the two subsystems, if they are in equilibrium
with each other. The non-equilibrium situation is then parametrized for the energy by the
deviation Ũ : U1 = U01 + Ũ and U2 = U02 Ũ . The entropy satisfies at equilibrium
S(Ũ ) = S1 (U01 + Ũ ) + S2 (U02 Ũ )
(1.58)
@S @S1 @S2 1 1
with 0= = = ) T1 = T2 = T0 ,
@ Ũ @U 1 @U2 T1 T2
for Ũ = 0.2 Thus, we find for small Ũ ,
1 @S1 1 @ 2 S1
= ⇡ + Ũ ,
T1 @U1 U1 =U01 +Ũ T0 @U12 U1 =U01
(1.60)
1 @S2 1 @ 2 S2
= ⇡ Ũ .
T2 @U2 U2 =U02 Ũ T0 @U22 U2 =U02

With (1.58) we then obtain through expansion in Ũ ,

!
Ũ 2 @ 2 S1 @ 2 S2
S(Ũ ) = S1 (U01 ) + S2 (U02 ) + + + O(Ũ 3 ) . (1.61)
2 @U12 U01 @U22 U02

so that the time evolution of the entropy is given by

! ✓ ◆
dS dŨ @ 2 S1 @ 2 S2 dŨ 1 1
= Ũ + = , (1.62)
dt dt @U12 U01 @U22 U02 dt T1 T2

where we used Eq.(1.60). Note that the only time dependence appears in Ũ . The derivative
dŨ /dt corresponds to the energy flow from one subsystem to the other. We express dŨ /dt in
terms of the distribution functions of one of the subsystems, say system 1.
X dw⌫ (1) X (1)
dŨ @w⌫ dT1 dT1 X @ e ✏⌫ /kB T1
= N1 ✏⌫ = N1 ✏⌫ ⇡ N1 ✏⌫ . (1.63)
dt ⌫
dt ⌫
@T1 dt dt ⌫ @T1 Z1
2
Note that the entropy as a thermodynamic potential obeys,
dU X Fi
S = S(U, qi ) with dS = dqi . (1.59)
T i
T

15
Note that here and in the following
e ✏⌫ /kB Tj X
w⌫(j) = with Zj = e ✏⌫ /kB Tj
. (1.64)
Zj ⌫

The derivative yields

2 ( )2 3
dŨ dT1 4X 2 (1) X
5 1 N1 dT1 dT1
⇡ N1 ✏⌫ w⌫ ✏⌫ w⌫(1) 2 = h✏2 i1 h✏i21 = C1 ,
dt dt ⌫ ⌫
k B T1 kB T12 dt dt
(1.65)
where C1 denotes the heat capacity of subsystem 1, since
dŨ dT1 @ Ũ dT1
= = C1 . (1.66)
dt dt @T1 dt
Thus, we have here derived the relation
kB T12 C1 = N1 h✏2 i1 h✏i21 , (1.67)
which connects the heat capacity with the fluctuations of the energy value. Note that the
internal energy of subsystem 1 is not conserved due to the coupling to subsystem 2.
In which direction does the energy flow? Because we know that dS/dt 0, from (1.62) follows
dS dT1 T2 T1
0 = C1 , (1.68)
dt dt T1 T2
which yields for T1 > T2 that
dT1 dŨ
<0 ) <0. (1.69)
dt dt
This means that the energy flow goes from the warmer to the colder subsystem, reducing the
temperature of the warmer system and increasing that of the colder subsystem as can be seen
by the analogous argument. The flow stops when T1 = T2 = T0 .
Let us now consider the situation again by means of the master equation. We may approximate
(a)
the equation of w⌫ by (a, b = 1, 2),
(a)
dw⌫1 X Xh (ab) (a) (b) (ab)
i
⇡ w w⌫ 0
⌫1 ,⌫2 ;⌫10 ⌫20 ⌫10
w(a) w(b)
⌫10 ,⌫20 ;⌫1 ⌫2 ⌫1 ⌫2
(1.70)
dt 0 0
2
⌫1 ,⌫2 ,⌫2 b

which describes the evolution of the probability distribution transferring in general energy be-
tween the two subsystems, as only the overall energy is conserved but not in each subsystem
separately,
✏⌫1 + ✏⌫2 = ✏⌫10 + ✏⌫20 . (1.71)
All processes inside each subsystem are considered at equilibrium and give no contribution by
the condition of detailed balance. This remains true as long as the subsystems are very large
and the coupling between them weak. For symmetry reasons
(12) (12) (12) (12)
⌫1 ⌫2 ;⌫10 ⌫20
= ⌫10 ⌫20 ;⌫1 ⌫2
= ⌫2 ⌫1 ;⌫10 ⌫20
= ⌫1 ⌫2 ;⌫20 ⌫10
>0. (1.72)

We use now this equation and estimate dŨ /dt

dŨ X h i
(12) (1) (2) (12)
= N1 ✏ ⌫1 w w⌫ 0
⌫1 ,⌫2 ;⌫10 ⌫20 ⌫10
w(1) w(2)
⌫10 ,⌫20 ;⌫1 ⌫2 ⌫1 ⌫2
dt 2
⌫1 ,⌫10 ,⌫2 ,⌫20
(1.73)
X ✏⌫ 0 /kB T1 ✏⌫ 0 /kB T2 ✏⌫1 /kB T1 ✏⌫2 /kB T2
N1 (12) e 1 2 e
= (✏⌫1 ✏⌫10 ) ⌫1 ⌫2 ;⌫10 ⌫20
.
2 Z1 Z2
⌫1 ,⌫10 ,⌫2 ,⌫20

16
Using (1.71) and assuming that T1 and T2 are close to each other (T1 ⇡ T2 ⇡ T0 ), we obtain
may expand in T1 T2 ,

dŨ N1 X (12) T2 T1
⇡ ⌫1 ⌫2 ;⌫10 ⌫20
e (✏⌫1 +✏⌫2 )/kB T0 (✏⌫1 ✏⌫10 )2 ⇡ K(T2 T1 ) (1.74)
dt 2Z 2 kB T02
⌫1 ,⌫10 ,⌫2 ,⌫20

P ✏⌫ /kB T0
with Z = ⌫ e ⇡ Z1 ⇡ Z2 and

N1 X (12) (✏⌫1 ✏⌫10 )2

K⇡ ⌫1 ⌫2 ;⌫10 ⌫20
e (✏⌫1 +✏⌫2 )/kB T >0. (1.75)
2Z 2 kB T02
⌫1 ,⌫10 ,⌫2 ,⌫20

The energy flow is proportional to the di↵erence of the temperatures and stops when the two
subsystems have the same temperature. It is also here obvious that the spontaneous energy
transfer goes from the warmer to the colder reservoir. This corresponds to Fourier’s law as dŨ /dt
can be considered as a heat current (energy current) from subsystem 1 to 2, JQ = dŨ /dt. This
heat transport is an irreversible process leading to an increase of the entropy,
✓ ◆ 2
dS dŨ 1 1 (T2 T1 ) 2 JQ
= ⇡K = 0 (1.76)
dt dt T1 T2 T02 KT02

This irreversible process leads us from the less likely system configuration of an inhibited equi-
librium (di↵erent temperature in both subsystems) to the most likely of complete equilibrium
(T1 = T2 = T0 ) with the entropy

X (1) (2)
N⌫ N⌫ N⌫
S= N kB w⌫ ln w⌫ with w⌫ = = w⌫(1) = = w⌫(2) = , (1.77)
⌫
N N1 N2

i.e. the probability distribution is uniform for the whole system with N = N1 + N2 units.

17
Chapter 2

Classical statistical Physics

Statistical physics deals with the equilibrium properties of matter and provides the microscopic
understanding and basis for thermodynamics. This chapter develops a new practical approach
to equilibrium state of macroscopic systems. Time is not a variable anymore and measurements
may be considered as averages over a large ensemble of identical systems in di↵erent possible
microscopic states under the same external parameters.

2.1 Gibbsian concept of ensembles

As a practical example, we consider the state of a gas of N classical particles, given by 3N canon-
ical coordinates q1 , . . . , q3N and by the corresponding 3N conjugate momenta p1 , . . . , p3N . These
define a 6N -dimensional space , where each point in represents a state of the microscopic
system. Considering the whole system of N particles under certain macroscopic conditions,
given external parameter such as temperature, pressure, volume, internal energy, .... , we find
that an infinite number of states in are compatible with the same external condition, and
would not be distinguishable by macroscopic measurements.

[q]
Γ
t=0

t>0

[p]

time evolution ensemble

Figure 2.1: Time averages are replaced by averages over an ensemble of systems in di↵erent
microscopic states, but with the same macroscopic conditions.

If we would like to calculate a certain macroscopic quantity we could perform temporal average
of the microscopic variables over a very long evolution time. Based on the states in this
is, however, not a practical method as it involves the discussion of the temporal evolution.
Gibbs introduced the concept of ensembles to circumvent this problem. Taking a large (infinite)
number of systems under identical macroscopic conditions, we can find macroscopic quantities
by averaging over an ensemble of states. That this scheme is equivalent to a temporal average
is a hypothesis, as it is assumed that the time evolution of the system would lead to all possible
states also represented in the ensemble. This is the ergodicity hypothesis.1 The set of states
1
Ergodicity hypothesis: We consider the subspace defined by H(p, q) = E in . Due to energy conservation a

18
for given macroscopic parameters is then represented by a distribution of points in the space .
This distribution is generally continuous for a gas of particles.
For the calculation of averages we introduce the density function ⇢(p, q) providing the measure
of the density of points in -space ((p, q) stands for the whole state (p1 , . . . , p3N ; q1 , . . . q3N )).
Then
⇢(p, q)d3N pd3N q (2.1)
gives the number of representative points contained in the small volume d3N pd3N q in , very
analogous to the distribution function in Boltzmann theory. This can now be used to calculate
averages of any desired quantity which can be expressed in the variable (p, q), A(p, q):
R
dpdqA(p, q)⇢(p, q)
hAi = R . (2.2)
dpdq⇢(p, q)

We will use from now on the short notation dpdq for d3N pd3N q where it is not misleading.

2.1.1 Liouville Theorem

The dynamics of the system of N particles shall be described by a Hamiltonian H(p, q) which
yields the equation of motion in the Hamiltonian formulation of classical mechanics,
@H @H
ṗi = and q̇i = (i = 1, . . . , 3N ) . (2.3)
@qi @pi
This equation describes the motion of the points in . If H does not depend on time derivatives
of pi and/or qi , then the equations of motion are time reversal invariant. They determine the
evolution of any point in uniquely.
Now let us consider the points in the space and note that their number does not change in
time, as if they form a fluid. Thus, they satisfy the continuity equation
@⇢ ~
+ r · (⇢~v ) = 0 (2.4)
@t
~ = (@/@p1 , . . . , @/@p3N ; @/@q1 , . . . , @/@q3N ). Intro-
where ~v = (ṗ1 , . . . , ṗ3N ; q̇1 , . . . , q̇3N ) and r
ducing the generalized substantial derivative we can rewrite this equation as
D⇢ ~ · ~v = 0 .
+ ⇢r (2.5)
Dt
The divergence of the ”velocity” ~v is
3N ⇢
X 3N ⇢
X
~ · ~v = @ q̇i @ ṗi @ @H @ @H
r + = = 0. (2.6)
@qi @pi @qi @pi @pi @qi
i=1 i=1 | {z }
=0
This means that the points in space evolve like an incompressible fluid. This property is
known as Liouville’s theorem. We may write this also in the form
3N ⇢ 3N ⇢
D⇢ @⇢ X @⇢ @⇢ @⇢ X @H @⇢ @H @⇢
0= = + q̇i + ṗi = + . (2.7)
Dt @t @qi @pi @t @pi @qi @qi @pi
i=1 i=1

point in this subspace, evolving in time, does not leave this subspace. The ergodicity hypothesis states, that a
point on H = E can reach in its temporal evolution every point of this subspace (Boltzmann, 1887). This is not
rigorously valid, however. Ehrenfest’s version of this hypothesis (1911) states: The trajectory of a point comes
any other point in the same subspace arbitrarily close in the course of time. Of course, it is possible to find
(singular) counter examples, for which only a tiny fraction of the subspace is accessed.

19
Using Poisson brackets2 this equation reads
@⇢
= {H, ⇢} . (2.9)
@t

2.1.2 Equilibrium system

A satisfactory representation of a system in equilibrium requires for an ensemble that the density
function does not depend on time t, i.e. @⇢/@t = 0. From Liouville’s theorem we get the
condition that
3N ⇢
X @⇢ @⇢ ~ = {H, ⇢} .
0= q̇i + ṗi = ~v · r⇢ (2.10)
@qi @pi
i=1

A general way to satisfy this is to take a density function which depends only on quantities
conserved during the motion, such as energy or particle number. Then the system would evolve
within a subspace where ⇢ is constant.
We may use this feature of ⇢ now to consider averages of certain quantities, based on the above
mentioned equivalence between the temporal and ensemble averages. Defining the temporal
average of A(p, q) as
Z
1 T
hAi = lim A(p(t), q(t))dt (2.11)
T !1 T 0

for any starting point (p(t = 0), q(t = 0)) in the space and (p(t), q(t)) obeying the equation
of motion (2.3). The hypothesis of ergodicity, even in its restricted sense, implies that this
average can be taken as an average of an ensemble of an infinite number of di↵erent microscopic
states (points in -space). As the evolution of (p, q) conserves the energy, this leads naturally
to consider an ensemble of states of a fixed energy (internal energy). We call such ensembles
microcanonical. Although the microcanonical ensemble is not so frequently used in practice
than other ensembles which we will consider later, it is very useful for illustrative purposes. It
describes an isolated closed system with no energy exchange with the environment.
We postulate that in equilibrium any state of a macroscopic system satisfying the external
conditions appears with equal probability. In our microcanonical description with fixed energy,
number of particles N and volume V , we postulate
8
< const. E  H(p, q)  E + E
⇢(p, q) = (2.12)
:
0 otherwise

where E is small. The average value of A is then given by (2.2). The validity of this approach
is based on the assumption of small mean square fluctuations (standard deviation)

h{A hAi}2 i
⌧1. (2.13)
hAi2

Such fluctuations should be suppressed by the order N 1 as we will see below.

2.2 Microcanonical ensemble

We consider a macroscopic system of N particles in a volume V which is isolated and closed.
The microcanonical ensemble for a given energy E consists of all systems of this kind, whose
2
The Poisson bracket is defined as
X ⇢ @u @v @u @v
{u, v} = = {v, u} . (2.8)
i
@qi @pi @pi @qi

20
energy lies in the range [E, E + E]. First we define the phase volume
Z
(E) = ⇤N dpdq , (2.14)
H(p,q)E

which contains all point in space with energy lower than or equal to E. Moreover, ⇤N is a
renormalization factor
1
⇤N = (2.15)
N !h3N
which compensates for the over-counting of the phase space in the integral by dividing by the
number of equivalent states reached by permutations of the particles, N !. The factor h3N corrects
for the dimension integral to produce a dimensionless (E). Thus, h has the units of action
([pq] = Js). While this could be Planck constant h, its magnitude is completely unimportant in
classical statistical physics. From this we obtain the volume of the microcanonical ensemble as
Z
d (E)
!(E) = ⇤N dpdq = (E + E) (E) = E. (2.16)
EH(p,q)E+ E dE

Thus we may renormalize the density function ⇢(p, q) with the condition,
Z Z
⇤N
1 = ⇤N dp dq ⇢(p, q) = dp dq (2.17)
!(E) EH(p,q)E+ E

such that 8
> 1
>
< E  H(p, q)  E + E
!(E)
⇢(p, q) = . (2.18)
>
>
: 0 otherwise
As postulated we choose ⇢(p, q) to be a constant in the energy range [E, E + E].

2.2.1 Entropy
We use !(E) to define the entropy

S(E, V, N ) = kB ln !(E) . (2.19)

We can consider !(E) or S as a measure of the imprecision of our knowledge of the state of the
system. The more states are available in microcanonical ensemble, the less we know in which
state the system is at a given time and the larger is the volume ! and the entropy.
We consider a composite system consisting of two subsystems, 1 and 2,

H(p, q) = H1 (p1 , q1 ) + H2 (p2 , q2 ) , (2.20)

with (N1 , V1 ) and (N2 , V2 ), resp., for the corresponding particle number and volume. Each of
the two systems is characterized by !1 (E1 ) and !2 (E2 ), respectively. The volume !(E) of the
microcanonical ensemble is the product of the subsystems under the conditions

E = E1 + E2 , N = N1 + N2 and V = V1 + V 2 . (2.21)

For simplicity we assume that the volumes and particle numbers of the subsystems may be fixed,
while they can exchange energy, such that E1 and E2 can fluctuate. Therefore the volume of
the microcanonical ensemble of the total system for given E reads,
X
!(E) = !1 (E 0 ) !2 (E E 0 ) , (2.22)
0E 0 E

21
where we assume for the sum a ”discrete” mesh of equally spaced E 0 -values of mesh spacing
E (⌧ E). We claim that this sum is well approximated by a single value E00 giving rise to
a sharp maximum among the summands (E1 = E00 and E2 = E E00 )3 . The largest term is
obtained by
@!1 (E 0 ) !2 (E E 0 )
=0 (2.25)
@E 0 E 0 =E 0 0

such that ⇢
1 @!1 (E 0 ) 1 @!2 (E E 0 )
0 = +
!1 (E 0 ) @E 0 !2 (E E 0 ) @E 0 E 0 =E00
(2.26)
@ ln !1 (E1 ) @ ln !2 (E2 )
= .
@E1 E1 =E00 @E2 E2 =E E00

From this we obtain with Ē1 = E00 and Ē2 = E E00

@S1 (E1 ) @S2 (E2 )

= (2.27)
@E1 E1 =Ē1 @E2 E2 =Ē2

which can be considered as the equilibrium condition. Note that this is equivalent to the state-
ment that the equilibrium state corresponds to the macrostate with the largest number of mi-
croscopic realizations.
Identifying E as the internal energy U we define the temperature
@S 1 1 1
= ) = , (2.28)
@U T T1 T2
leading to T as an equilibrium state variable. We have ignored the other variables V, N which
we will consider later.
Let us assume that there is some impediment which forces the two subsystems to specific vari-
ables such that each subsystem independently is at equilibrium, but not the total system. For
example, a big vessel of gas may be separated by a wall into two subvessels. Then we find for
the corresponding ! of the combined system,
8
< E = E1 + E2
!
˜ (E, V, N ) = !1 (E1 , V1 , N1 )!2 (E2 , V2 , N2 ) with V = V1 + V2 (2.29)
:
N = N1 + N2

such that the entropy is given by

S̃(E, V, N ) = S1 (E1 , V1 , N1 ) + S2 (E2 , V2 , N2 )  S(E, V, N ) , (2.30)

i.e. the entropy is concave. This means also the equilibrium is obtained by the maximal entropy,
which is a consequence of the second law of thermodynamics. In terms of the volume of the
microcanonical volume ! the equilibrium state assumes among all volumes the maximal one,
the most likely one in terms of probability.
3
Note that ln !i / Ni and Ei / Ni (i = 1, 2) as both quantities are extensive. We then find quickly the
following bounds:
E
!1 (E00 )!2 (E E00 )  !(E)  !1 (E00 )!2 (E E00 ) . (2.23)
E
Note that E/ E is the number of summands. Thus, we obtain for the entropy
E
kB ln !1 (E00 )!2 (E E00 )  S(E)  kB ln !1 (E00 )!2 (E E00 ) + kB ln . (2.24)
E
It is important to see that the last term only scales with system size as ln N (keeping E fixed), while the first
term scales with N . Therefore in the very large-N limit the last term is irrelevant. Which proves that S(E) is
given by the maximal term in the sum (2.22).

22
2.2.2 Relation to thermodynamics
With (2.19) we have a definition of the entropy which for the variables E(= U ), V and N is a
thermodynamic potential and allows us to calculate a variety of state variables and relations,
✓ ◆ ✓ ◆ ✓ ◆
@S @S @S 1 p µ
dS = dE + dV + dN = dE + dV dN . (2.31)
@E V,N @V E,N @N E,V T T T

This allows also to calculate the thermodynamic equation of state,

✓ ◆
@S
p=T (2.32)
@V E,N

and to determine other thermodynamic potentials. The caloric equation of state is obtained
from ✓ ◆
@S 1
= , (2.33)
@U V,N T
by using U = E as the internal energy. The derivative with respect to N yields finally the
chemical potential ✓ ◆
@S
µ= T , (2.34)
@N E,V
the energy in order to add a particle to the system.

2.2.3 Ideal gas - microcanonical treatment

We consider a classical gas of N independent mono-atomic particles in the fixed volume V ,
which is closed and isolated. The Hamiltonian is simply given by
N
X p~i2
H(p, q) = H(p) = . (2.35)
2m
i=1

Hence we obtain the volume

Z Z
N
(E) = ⇤N dpdq = ⇤N V dp . (2.36)
H(p)E H(p)E

This p-integral corresponds to the volume of a sphere of radius R in 3N -dimensional space.4

The integral is straightforward,

⇡ n/2
(E) = ⇤N V N C3N (2mE)3N/2 with Cn = n (2.39)
2 +1
4
Volume of sphere in ⌫-dimensional space: V(R) = CR⌫ . The volume enclosed in a thin shell of width R at
the radius R is given by  ✓ ◆⌫
R
vshell = V(R) V(R R) = CR⌫ 1 1 (2.37)
R
with R ⌧ R, see Fig.2.2 . In the limit ⌫ very large, ⌫ R R, we find

vshell ⇡ V(R) . (2.38)

Thus, the main part of the volume in the sphere is concentrated at the outermost shell.

23
 Rn vshell,1% /CRn
1 1.0
0.8 0.8
n=5 0.6
R 0.6 n = 10
0.4 n = 40 0.4
0.2 0.2
R 0 0
0 0.2 0.4 0.6 0.8 1 R 2 200 400 600 800 n

Figure 2.2: The volume of the sphere is located close to the surface in high dimensions, see
the central diagram. Thus, the fraction of the volume close to the shell and the total volume
converges towards one as n ! 1 (diagram on the right hand side).

where C3N is the proper prefactor for the volume of an 3N -dimensional sphere.5 This leads to
@ (E) 3N
!(E) = E = ⇤N C3N V N 2m (2mE)3N/2 1
E. (2.42)
@E 2
Remarkably, for very large N (⇠ 1023 ) we find that the following definitions for the entropy are
identical up to terms of order ln N and constants:

S! = kB ln !(E, V, N ) and S = kB ln (E, V, N ) (2.43)

leading to
✓ ◆ ✓ ◆
N 3N 3N
S! = kB ln(⇤N V C3N ) + kB 1 ln(2mE) + kB ln 2m E
2 2
(2.44)
3N
= kB ln(⇤N V N C3N ) + kB ln(2mE) + O(ln N ) = S + O(ln N ) .
2
Since we can drop terms of order ln N for the extensive entropy, we will continue, for convenience,
using S instead of S! ,
( ✓ ◆ )
2m⇡E 3/2 3N 3N 3N
S(E, V, N ) = N kB ln V 2
kB ln + kB N kB ln N + N kB (2.45)
h 2 2 2

where we used Stirling’s formula

1
ln n! ⇡ n ln n n+ ln(2⇡n) for n!1, (2.46)
2
and neglected all terms of order ln N . We then rewrite
( ✓ ◆ )
V 4⇡mE 3/2 5
S(E, V, N ) = N kB ln + N kB . (2.47)
N 3N h2 2

This equation may now be solved for E so that we obtain the internal energy as a thermodynamic
potential ⇢
3n2/3 h2 2S 5
U (S, V, N ) = E = N exp . (2.48)
4⇡m 3N kB 3
5
Prefactor Cn : Use the n-dimensional Gaussian integral in Cartesian coordinates
Z +1 Z +1 ✓Z +1 ◆n
2 2 2
I= dx1 · · · dxn e (x1 +···+xn ) = dxe x = ⇡ n/2 . (2.40)
1 1 1

The same integral in spherical coordinates is given by

Z 1 Z 1 ⇣n⌘ ⇣n ⌘
n 1 r2 n n 1 t n
I = nCn drr e = Cn dt t 2 e = Cn = Cn +1 (2.41)
0 2 0 2 2 2
such that we obtain Cn given in (2.39). Note, (n + 1) = n! for n 0 as an integer.

24
The thermodynamic quantities are obtained by derivation: the temperature
✓ ◆
@U 2U 3
T = = ) U = N kB T , (2.49)
@S V,N 3N kB 2

the pressure ✓ ◆
@U 2U N kB T
p= = = ) pV = N kB T , (2.50)
@V S,N 3V V
and the chemical potential
✓ ◆ ✓ ◆ ( ✓ ◆3/2 )
@U U 5 2 S V 2⇡mkB T
µ= = = kB T ln . (2.51)
@N S,V N 3 3 N kB N h2

Through further derivatives it is possible to obtain various response functions. The ideal gas is
readily described by means of the microcanonical ensemble.

2.3 Canonical ensemble

We change to a macroscopic system for which we control the temperature by connecting it to a
very large heat reservoir. The system together with the reservoir forms a closed system of given
total energy. Therefore we consider two subsystems, system 1 describing our system and system
2 being the heat reservoir (Fig.2.3),

H(p, q) = H1 (p1 , q1 ) + H2 (p2 , q2 ) . (2.52)

The heat reservoir is much larger than system 1, N2 N1 , such that energy transfer between
the two subsystems would be too small to change the temperature of the reservoir. Within
the microcanonical scheme we determine the phase space of the combined system in the energy
range
E  E1 + E2  E + E . (2.53)

E2 , S2 , T

E1 , S1 , T

Figure 2.3: The two systems 1 and 2 are coupled thermally. System 2 acts as a huge reservoir
fixing the temperature of the system 1 which we want to investigate.

Therefore the volume of the microcanonical ensemble of the total system is

X
!(E) = !1 (E1 )!2 (E E1 ) . (2.54)
0E1 E

Analogously to our previous discussion in section 2.2.1, there is one value Ē1 = E00 (Ē2 = E E00 )
which provides the by far dominant contribution. In addition here Ē2 Ē1 is valid. The
corresponding volumes in -space are !1 (Ē1 ) and !2 (Ē2 ) and !(E) ⇡ !1 (Ē1 )!2 (E Ē1 ). Due

25
to this simple product form we can determine the (unrenormalized) density function ⇢1 (p1 , q1 )
of the system 1, by considering the mean value of A(p, q) in system 1,
R R R
1 dpR1 dq1 A(p1 , q1 )⇢1 (p1 , q1 ) 1 dpR1 dq1 A(p1 R
, q1 ) 2 dp2 dq2 ⇢(p, q)
hAi1 = = . (2.55)
1 dp1 dq1 ⇢1 (p1 , q1 ) 1 dp1 dq1 2 dp2 dq2 ⇢(p, q)

Taking into account that ⇢(p, q) is constant in the range E  H1 (p1 , q1 ) + H2 (p2 , q2 )  E + E
we obtain6 R
dp1 dq1 A(p1 , q1 )!2 (E H1 (p1 , q1 ))
hAi1 = 1 R . (2.57)
1 dp1 dq1 !2 (E H1 (p1 , q1 ))
Using the assumption that Ē2 ⇡ E Ē1 we may expand !2 (E H1 (p1 , q1 )) in H1 (p1 , q1 )) ,

@S2 (Ē2 )
kB ln !2 (E H1 (p1 , q1 )) = S2 (E H1 (p1 , q1 )) = S2 (E) H1 (p1 , q1 ) + ···
@E2 Ē2 =E

H1 (p1 , q1 )
= S2 (E) + ···
T
(2.58)
from which we derive
H1 (p1 ,q1 )/kB T
!2 (E H1 (p1 , q1 )) = eS2 (E)/kB e . (2.59)

Here T is the temperature of both systems which are in equilibrium.

Within the canoncial ensemble, taking the temperature T as a given parameter, we write gener-
ally for the density function, the probability to find the system of N particles in the microstate
(p, q),
1
⇢(p, q) = e H(p,q)/kB T , (2.60)
Z
where we introduced the partition function Z
Z
1
Z = ⇤N dp dqe H(p,q) with = , (2.61)
kB T
which, up to prefactors, corresponds to the volume of the ensemble of system 1, called canonical
ensemble, again renormalized by ⇤N .7

2.3.1 Thermodynamics
The connection to thermodynamics is given by the relation
F (T,V,N )
Z=e , (2.62)

where F (T, V, N ) is the Helmholtz free energy, a thermodynamical potential. Note, F is an

extensive quantity, because obviously scaling the system by a factor would yield Z . Moreover,

F =U TS (2.63)

with ✓ ◆
@F
U = hHi and S= . (2.64)
@T V,N
6
We use Z Z
dp2 dq2 ⇢(p, q) / dp2 dq2 = !2 (E(p1 , q1 )) (2.56)
2 E(p1 ,q1 )H2 (p2 ,q2 )E(p1 ,q1 )+

where E(p1 , q1 ) = E H1 (p1 , q1 ).

7
Note that there is, rigorously speaking, the constraint H1 (p1 , q1 ) < E. However, ignoring this constraint is a
good approximation, as the main contribution is from the valid range.

26
This can be proven using the equation,
Z
(F H)
1 = ⇤N dp dqe , (2.65)

which through di↵erentiation with respect to on both sides gives,

Z ( ✓ ◆ )
(F H) @F
0 = ⇤N dp dqe F+ H
@ V,N
(2.66)
✓ ◆
@F
) F (T, V, N ) U (T, V, N ) T =0.
@T V,N

Using this formulation for the free energy we obtain for the pressure
✓ ◆
@F
p= , (2.67)
@V T,N

which in the case of a gas leads to the thermodynamic equation of state.

The internal energy is easily obtained from the partition function in the following way,
Z
⇤N 1 @Z @
U (T, V, N ) = hHi = dp dq He H = = ln Z . (2.68)
Z Z@ @
This is the caloric equation of state.

2.3.2 Equipartition law

We now consider a set of special average values which will lead us to the so-called equipartition
law, the equal distribution of energy on equivalent degrees of freedom. We examine the mean
value,
⌧ Z Z
@H ⇤N @H ⇤N @
qµ = dp dqqµ e H= dp dqqµ e H
@q⌫ Z @q⌫ Z @q⌫
Z 0 Z (2.69)
⇤N ⇤N µ⌫
= dp d0 q qµ e H
+ dpdqe H
= µ⌫ kB T ,
Z Z
| {z }
=0
where
R 0 we used integration by parts leading to the boundary terms in the q⌫ -coordinate (expressed
by d0 q...), which we assume to vanish. Analogously we find for the momentum
⌧
@H
pµ = µ⌫ kB T . (2.70)
@p⌫

If the Hamiltonian is separable into a p-dependent kinetic energy and a q-dependent potential
energy part and, moreover, if the following scaling behavior is valid
2 ↵
H(p, q) = Ekin (p) + V (q) with Ekin ( p) = Ekin (p) and V ( q) = V (q) (2.71)

then we can use the above relations and find for mono-atomic particles
3N 3N
hEkin i = kB T and hV i = kB T . (2.72)
2 ↵
The total energy is given by the sum of the two contributions.

27
2.3.3 Ideal gas - canonical treatment
Consider a gas of N particles without external potential and mutual interactions described by
the Hamiltonian
XN
p~i 2
H(p) = . (2.73)
2m
i=1
The partition function is given by
N Z
Y ⇢Z N
~i 2 /2mkB T ~ 2 /2mkB T
Z = ⇤N d 3 pi d 3 q i e p
= ⇤N d 3 p d3 q e p

i=1 (2.74)

= ⇤N V N {2⇡mkB T }3N/2 .
From this we obtain the free energy and the internal energy using Stirling’s formula,
( ✓ ◆ )
V 2⇡mkB T 3/2
F (T, V, N ) = kB T ln Z = N kB T ln N kB T ,
N h2
(2.75)
@ 3N
U (T, V, N ) = ln Z = kB T (caloric equation of state) .
@ 2
The entropy is given by
✓ ◆ ( ✓ ◆3/2 )
@F V 2⇡mkB T 5N
S(T, V, N ) = = N kB ln + kB (2.76)
@T V,N N h2 2

and the pressure by ✓ ◆

@F N kB T
p= = (2.77)
@V T,N V
which corresponds to the thermodynamic equation of state. Finally the chemical potential is
obtained as ✓ ◆ ( ✓ ◆ )
@F V 2⇡mkB T 3/2
µ= = kB T ln . (2.78)
@N T,V N h2
An important aspect for the ideal system is the fact that the partition function has a product
form because each particle is described independently. In this way it leads to an extensive free
energy and internal energy.

2.4 Grand canonical ensemble

We consider now a new situation by allowing beside the heat exchange also the exchange of
matter of our system with a very large reservoir, see Fig.2.4 . Thus we take the system 1 with
N1 particles in a volume V1 coupled to the reservoir 2 with N2 particles in the volume V2 with
N1 ⌧ N2 and V 1 ⌧ V2 , (2.79)
and N = N1 + N2 and V = V1 + V2 fixed.
The Hamiltonian can be decomposed into two parts
H(p, q, N ) = H(p1 , q1 , N1 ) + H(p2 , q2 , N2 ) (2.80)
such that the corresponding partition function for given temperature (everything is coupled to
an even larger heat reservoir) is given by
Z
1
ZN (V, T ) = 3N dp dq e H(p,q,N ) . (2.81)
h N!

28
 T

N2 , V2
T, µ N1 , V1
T, µ

Figure 2.4: TThe two systems 1 and 2 can exchange matter between each other while the
surrounding heat reservoir fixes the temperature of system 1 and 2. System 2 acts as a huge
particle reservoir fixing the chemical potential µ of the system 1 which we want to investigate.

The factor 1/N ! takes into account that all possible commutation of the particles give the same
states (distinguishable classical particles). Now we segregate into the subsystems fixing the
volumes and particle numbers (N2 = N N1 ),
N
X Z Z Z
1 N! {H(p1 ,q1 ,N1 )+H(p2 ,q2 ,N2 )}
ZN = dp1 dp2 dq1 dq2 e
h3N N ! N1 !N2 ! V1 V2
N1 =0
(2.82)
N
X Z Z
1 H(p1 ,q1 ,N1 ) 1 H(p2 ,q2 ,N2 )
= dp1 dq1 e dp2 dq2 e .
h3N1 N 1! V1 h3N2 N 2! V2
N1 =0

Note that the distribution of the particles into the two subsystems is not fixed yielding the
combinatorial factor of N !/N1 !N2 ! (number of configurations with fixed N1 and N2 by permuting
the particles in each subsystem). From this we define the probability ⇢(p1 , q1 , N1 ) that we can
find N1 particles in the volume V1 at the space coordinates (p1 , q1 ),
Z
e H(p1 ,q1 ,N1 )
⇢(p1 , q1 , N1 ) = dp2 dq2 e H(p2 ,q2 ,N2 ) (2.83)
ZN N1 !h3N2 N2 ! V2
which is renormalized as
N
X Z
1
dp1 dq1 ⇢(p1 , q1 , N1 ) = 1 . (2.84)
h3N1 V1
N1 =0

We may write
Z N2 1 H(p1 ,q1 ,N1 )
⇢(p1 , q1 , N1 ) = e , (2.85)
Z N N1 !
where we now use the relation
ZN2 (V2 , T ) {F (T,V V1 ,N N1 ) F (T,V,N )}
=e (2.86)
ZN (V, T )
with
✓ ◆ ✓ ◆
@F @F
F (T, V V1 , N N1 ) F (T, V, N ) ⇡ V1 N1 = µN1 + pV1 . (2.87)
@V T,N @N T,V

Thus we define
µ
z=e (2.88)
which we call fugacity. Thus within the grand canonical ensemble we write for the density
function
zN
⇢(p, q, N ) = e {pV +H(p,q,N )} . (2.89)
N!

29
µ is the chemical potential as introduced earlier. We now introduce the grand partition function
1
X
Z(T, V, z) = z N ZN (V, T ) , (2.90)
N =0

which incorporates all important information of a system of fixed volume, temperature and
chemical potential.

2.4.1 Relation to thermodynamics

We use now (2.89) and integrate both sides
1
X Z
pV zN dp dq H(p,q,N ) pV
1=e e =e Z(T, V, z) (2.91)
N! h3N
N =0

which leads to
⌦(T, V, µ) = pV = kB T ln Z(T, V, z) , (2.92)
the grand potential
d⌦ = SdT pdV N dµ . (2.93)
The average value of N is then given by
✓ ◆ 1
@⌦ @ @ 1 X
hN i = = kB T ln Z = z ln Z = N z N ZN . (2.94)
@µ T,V @µ @z Z
N =0

It is also convenient to derive again the internal energy

✓ ◆
@ @U
U= ln Z ) CV = . (2.95)
@ @T V,µ

2.4.2 Ideal gas - grand canonical treatment

For the ideal gas, it is easy to calculate the grand partition function (here for simplicity we set
h = 1),
1
X 1
X ⇢
N zN V N 3N/2 zV
Z(T, V, z) = z ZN (T, V ) = 3N
(2⇡mkB T ) = exp (2⇡mkB T )3/2 .
N! h h3
N =0 N =0
(2.96)
We can also derive the probability PN of finding the system with N particles. The average value
is given by
@ zV zV
hN i = z 3
(2⇡mkB T )3/2 = 3 (2⇡mkB T )3/2 ) Z = ehN i . (2.97)
@z h h
From this we conclude that the distribution function for the number of particles is given by
N
hN i hN i 1 (N hN i)2 /2hN i
PN = e ⇡p e (2.98)
N! 2⇡hN i

which is strongly peaked at N = hN i 1.8 The fluctuations are given by

@hN i v 1
hN 2 i hN i2 = z = hN i ) T = = . (2.99)
@z kB T p
8
Note that for µ > 0 it pays the energy µ to add a particle to the system. Therefore as T goes to 0 the average
particle number hN i increases (diverges). Oppositely, hN i decreases for T ! 0, if µ < 0 and energy has to be
paid to add a particle.

30
The grand potential is given by

µ V
⌦(T, V, µ) = kB T e (2⇡mkB T )3/2 = kB T hN i = pV . (2.100)
h3
The chemical potential is obtained by solving Eq.(2.97) for µ
!
V (2⇡mkB T )3/2
µ= kB T ln . (2.101)
hN ih3

2.4.3 Chemical potential in an external field

In order to get a better understanding of the role of the chemical potential, we now consider an
ideal gas in the gravitational field, i.e. the particles are subject to the potential (h̃) = mg h̃,
where h̃ and g denote the altitude and the earth accelation of gravity, respectively. We introduce
a reference chemical potential µ0 as a constant. Then we write for the chemical potential,
( )
(2⇡mkB T )3/2
µ = µ0 + mg h̃ kB T ln (2.102)
nh3

where we define n = hN i/V as the local number density of particles. In equilibrium the tem-
perature and the chemical potential shall be constant. We may determine µ by the condition
that at h̃ = 0 the density is n = n0 ,
( )
(2⇡mkB T )3/2 ⇣n ⌘
0
µ(T ) = µ0 + kB T ln ) mg h̃ = k B T ln . (2.103)
n 0 h3 n

We can now solve this equation for n = n(h̃),

mg h̃
n(h̃) = n0 e , (2.104)

and with the (local) equation of state we find

mg h̃
p(h̃) = n(h̃)kB T = p0 e . (2.105)

This is the famous barometer formula.

2.5 Fluctuations
Changing from one type of ensemble to the other we have seen that certain quantities which
have been strictly fixed in one ensemble are statistical variables of other ensembles. Obvious
examples are the internal energy which is fixed in the microcanonical ensemble but not in the
other two, or the particle number which is fixed in the microcanonical and canonical ensembles
but not in the grand canonical ensemble. The question arises how well the mean values of
these quantities are determined statistically, which is connected with the equivalence of di↵erent
ensembles. In this context we will also introduce the fluctuation-dissipation theorem which
connects the fluctuations (statistical variance) of a quantity with response functions.

2.5.1 Energy
In the canonical ensemble the internal energy is given as the average of the Hamiltonian U = hHi.
Therefore the following relation holds:
Z
⇤N dp dq [U H] e (F H) = 0 . (2.106)

31
 Microcanonical ensemble Canonical ensemble Grand canonical ensemble

(H(p,q) µN )
H(p,q)
(p, q) = e
density (p, q) = (E H(p, q)) (p, q) = e N H(p,q)
(unrenormalized) =z e

Z 1
X Z
Z Z= dpdq (p, q, N )
ZN = dpdq (p, q) N
N
Zm0 can = N dpdq (p, q) N =0
partition Z 1
function H(p,q) X
= (E) = N dpdq e = z N ZN
N =0

⇤N
Z
⇤N
Z X ⇤N Z
average value hAi = dpdq ⇢(p, q)A(p, q) hAi = dpdq ⇢(p, q)A(p, q) hAi = dpdq ⇢(p, q, N )A(p, q, N )
!(E) ZN Z
N =0

thermodyn. S(E, V, N ) = kB log (E, V, N ) F (T, V, N ) = kB T log ZN (T, V ) (T, V, µ) = kB T log Z(T, V, µ)
potential

Figure 2.5: Synopsis of the three ensembles for a gas.

Taking the derivative of this equation with respect to we obtain

Z 
@U @F @U
0= + ⇤N dp dq (U H) F T H e (F H) = + h(U H)2 i . (2.107)
@ @T @
This leads to the relation for the fluctuations of the energy around its average value U ,
@U @U
hH2 i hHi2 = h(U H)2 i = = kB T 2 = kB T 2 CV . (2.108)
@ @T
Because CV is an extensive quantity and therefore proportional to N , it follows that

hH2 i hHi2 1
/ (2.109)
hHi2 N
which is a sign of the equivalence of microcanonical and canonical ensembles. In the thermo-
dynamic limit N ! 1, the fluctuations of the energy vanish compared to the energy itself.
Therefore the internal energy as U = hHi = E is a well defined quantity in the canonical
ensemble as well as it was in the microcanonical ensemble.
We now consider the partition function
Z Z 1
Fcan (T ) H(p,q) E
Z =e = ⇤N dp dq e = dE !(E)e
0
Z Z (2.110)
1 1
E+ln !(E) (T Sm0 can (E) E)
= dE e = dE e
0 0

where the entropy S(E) = Sm0 can (E) is defined according to the microcanonical ensemble. The
maximum of the integrand at E = E0 is defined by the condition, see Fig.2.6,

@S @2S
T =1 and <0. (2.111)
@E E=E0 @E 2 E=E0

Note that
@2S @ 1 1 @T 1
= = = <0. (2.112)
@E 2 E=E0 @E T E=E0 T 2 @E E=E0 T 2 CV

32
If we now expand the exponent in the integrand, we obtain
1
T S(E) E = T S(E0 ) E0 (E E0 ) 2 + · · · (2.113)
2T CV
such that with U = E0 ,
Z 1 p
(E U )2 /2kB T 2 CV
Z ⇡ e (T S U ) dE e =e (T S U )
2⇡kB T 2 CV
0

1
) Fcan ⇡ U T Sm0 can kB T ln 2⇡kB T 2 CV =U T Sm0 can + O(ln N ) .
2
(2.114)
Since the free energy is extensive, the term of order ln N is in the large-N limit irrelevant.

Gaussian approximation

(T S(E) E)
e

U = E0
Figure 2.6: Saddle point approximation of the partition function: integrand in Eq.(2.110) and
the approxiation in Eq.(2.114). Solid line is the full integrand and dashed line its Gaussian
approximation.

2.5.2 Particle number

We consider now the fluctuations of the particle number. The following relation holds
( )2
@ @ @ 1 X 1 X 2 N 1 X
hN 2 i hN i2 = z z ln Z = z N z N ZN = N z ZN N z N ZN
@z @z @z Z Z Z
N N N
(2.115)
from which we derive
@2 @hN i
hN 2 i hN i2 = kB T 2
⌦(T, V, µ) = kB T . (2.116)
@µ @µ
We now relate the right hand side with the isothermal compressibility. Introducing the specific
volume v = V /N we consider µ = µ(v, T ) (note that µ is not extensive). Moreover we use the
Gibbs-Duhem relation
SdT V dp + N dµ = 0 (2.117)
which derives from considerations on the Gibbs free energy (see thermodynamics course) and
obtain ✓ ◆ ✓ ◆
S @µ @p
dµ = v dp dT ) =v . (2.118)
N @v T @v T

33
Since v = V /N depends on both V and N , the following derivatives are useful
✓ ◆ ✓ ◆ ✓ ◆ ✓ ◆
@ @N @ N2 @
= = ,
@v V,T @v V,T @N V,T V @N V,T
✓ ◆ ✓ ◆ ✓ ◆ ✓ ◆ (2.119)
@ @V @ @
= =N .
@v N,T @v N,T @V N,T @V N,T

From (2.118) then conclude

✓ ◆ ✓ ◆ ✓ ◆ ✓ ◆
N2 @µ @p 1 @N 1 @V T
=V ) = = . (2.120)
V @N V,T @V N,T N @µ V,T vV @p N,T v

Consequently the fluctuations of the particle number is connected with the response function,
T , the isothermal compressibility,
v
T = hN 2 i hN i2 . (2.121)
hN ikB T

2.5.3 Magnetization
Considering a system of N classical magnetic moments m
~ i which have fixed magnitude m and
can be fully rotated, we define the magnetization
N
X
~ =
M m
~i. (2.122)
i=1

The average value of the magnetization is

Z Y
N ✓ ◆
[F H(m
~ 1 ,...,m ~
~ N ;H)] @F
hMz i = N hm1,z i = d⌦i Mz e = . (2.123)
@Hz T,N
i=1

Note that the magnetic field couples to magnetic moments through Zeeman coupling,

H(m
~ 1, . . . , m ~ = H0 ( m
~ N ; H) ~ 1, . . . , m
~ N) ~ ···M
H ~ . (2.124)

Now we discuss the equation,

Z
(F H)
0= d⌦1 · · · d⌦N (hMz i Mz )e , (2.125)

taking the derivative with respect to the field Hz leads to

Z
@hMz i @ (F H)
0 = d⌦1 · · · d⌦N (hMz i Mz ) (F H) e
@H @Hz
Z (2.126)
= zz d⌦1 · · · d⌦N (hMz i Mz )2 e (F H) = zz h(hMz i Mz ) i . 2

This yields the convenient fluctuation-dissipation relation for magnetic susceptibility,

1
zz = hMz2 i hMz i2 . (2.127)
kB T

34
Chapter 3

Quantum Statistical Physics

Quantum statistical physics opens the door to new phenomena and also provides an under-
standing of the third law of thermodynamics, which we had found to be often violated within
classical statistical physics. Moreover, we will find that for ideal quantum gases the concept of
indistinguishable particles leads to the classification of particles into two categories: Bosons and
Fermions.

3.1 Basis of quantum statistical physics

Every state of a quantum mechanical system (with many degrees of freedom) can be described
by a superposition of stationary states | n i of the Hamiltonian H, which form a complete
orthonormal basis, X
| i= cn | n i (3.1)
n

where the complex coefficients cn represent the wavefunction whose time dependence is deter-
mined by the Hamiltonian H. For the stationary states the following holds:
X
H| n i = "n | n i with h n | n0 i = nn0 and 1 = | n ih n | . (3.2)
n

The renormalization of the state | i requires

X X
1=h | i= c⇤n cn0 h n | n0 i = |cn |2 . (3.3)
n,n0 n

The square modulus |cn |2 denotes the probability to find the state | n i in | i.
Observable quantities are represented by Hermitian operators Â, and their expectation value is
X
h |Â| i = c⇤n cn0 h n |Â| n0 i . (3.4)
n,n0

This expectation value is generally time dependent through the wavefunction cn = cn (t) =
cn (0)e i"n t/~ .
Using the viewpoint of time evolution of the system, let us now consider the measurement of a
macroscopic observable. This corresponds to an average (indicated here by over-bar) over a time
much longer than the microscopic time scales (time between collisions, inverse energy spacing,
etc.),
Z T X X
1
hÂi = h |Â| i = lim dt c⇤n (t)cn0 (t)h n |Â| n0 i = c⇤n cn0 h n |Â| n0 i . (3.5)
T !1 T 0 n,n0 n,n0

35
 P evolution) follows from the Hamiltonian of the systems, | i is a state of
As the dynamics (time
fixed energy, E = n |cn (t)|2 "n . Like in the classical case we may represent the average value
hÂi also by an average over an ensemble of equivalent systems,
X
hÂi = ⇢nn0 h n |Â| n0 i (3.6)
n,n0

with ⇢nn0 = c⇤n cn0 . The time dependence of cn yields

Z T
1
c⇤n cn0 = lim dt c⇤n (0)cn0 (0)ei("n "n0 )t/~
= |cn |2 nn0 (3.7)
T !1 T 0

such that all terms invoking coherent interference (c⇤n cn0 with n 6= n0 ) vanish such that
X
hÂi = |cn |2 h n |Â| n i (3.8)
n

where we note that |cn |2 is time independent. The same result in the ensemble picture requires
that
⇢nn0 = |cn |2 nn0 (3.9)
implying that there is no interference between di↵erent member systems of a given ensemble.

Postulates of quantum statistical physics

The quantum statistical physics is based on two key postulates:

• Equal probability: 8
< r , E  ✏n  E + E ,
c⇤n cn = (3.10)
:
0 , otherwise ,
where r is a real constant.

• Random phase: if n 6= n0 then

c⇤n cn0 = 0 , (3.11)
i.e. the phase of the wavefunction is a random variable. Such a behavior requires an at
least weak coupling to the environment in order to randomize the phases.

As a consequence we consider a system as described e↵ectively by a wavefunction bn ,

X
| ie↵ = bn | n i (3.12)
n

(unrenormalized state) with

8
< 1 E  "n  E + E
|bn |2 = (3.13)
:
0 otherwise

and bn have a random phase. Then average values can be considered as the expectation value
P
|bn |2 h n |Â| n i
hÂi = n P 2
(3.14)
n |bn |

whereby each eigenstate of the Hamiltonian with energy in the given energy interval [E, E + E]
appears in the state | ie↵ with the same probability.

36
3.2 Density matrix
Mixed states or incoherent superpositions of states are conveniently described by density matrices
⇢ˆ.1
We define first the density matrix as an operator through its matrix elements

h n |ˆ
⇢| n0 i = ⇢nn0 = |cn |2 nn0 . (3.20)

In the basis of stationary states the density matrix is diagonal. Therefore we may formally write
the density matrix in spectral form as
X
⇢ˆ = |cn |2 | n ih n | . (3.21)
n

We may also express the average value of  as a trace of the form,

P
h n |ˆ
⇢| n i tr(ˆ⇢)
hÂi = Pn = . (3.22)
h
n n |ˆ
⇢ | n i tr ⇢
ˆ

Note that the final expression is independent of the basis {| n i}, as known from standard linear
algebra. The cyclic permutation of matrices (operators) does not change the trace: tr(ABC) =
tr(BCA). Thus the basis transformation U yields A0 = U AU 1 , such that

tr(A0 ) = tr(U AU 1
) = tr(AU 1
U ) = tr(A) . (3.23)

In this sense the density matrix is a convenient tool to perform ensemble averages as it contains
basis-free informations. In order to describe equilibrium properties ⇢ˆ should not depend on
time t and therefore it should commute with the Hamiltonian,
@ ⇢ˆ
i~ = [H, ⇢ˆ] = 0 . (3.24)
@t
It is obvious that the density matrix defined in (3.21) commutes with H. Compare this with
Liouville’s theorem of classical mechanics.

1
We distinguish pure and mixed states. A pure state given as superposition, for instance, energy eigenstates
(Eq.(3.1)) X
| i= cn | n i (3.15)
n

leads to the density matrix of a pure state, if we assume phase coherence of the wavefunction cn :
X ⇤
⇢ˆpure = | ih | = cn cn0 | n ih n0 | (3.16)
n,n0

whereby the mixed terms c⇤n cn0 with n 6= n0 are interference terms. For mixed states we assume that there is no
phase coherence in the wavefunctions such that all mixed terms disappear. Such a state represented by density
matrix has the form X
⇢ˆmixed = |cn |2 | n ih n | . (3.17)
n

Note that 8
⇢2pure ) = 1 ,
< tr(ˆ
tr(ˆ
⇢pure ) = tr(ˆ
⇢mixed ) = 1 ) (3.18)
:
⇢2mixed )  1 .
tr(ˆ
Average (expectation) values are calculated in both cases the same way:
8 P ⇤
< n,n0 cn cn h n |Â| n i pure ,
0 0

hÂi = tr(ˆ
⇢Â) = (3.19)
: P 2
n |cn | < n |Â| n i mixed .

37
 Synopsis - Classical versus Quantum statistical physics.
Classical formulation for gas of N particles (⇤N = 1/N !h3N )

Classical Statistical Physics Quantum Statistical Physics

observable phase space function A(p, q) Hermitian operator Â

density density function ⇢(p, q) density matrix ⇢ˆ

sum / integral phase space integral trace over Hilbertspace

R
⇤N dp dq · · · tr(· · · )

R
renormalized density ⇤N dp dq ⇢(p, q) = 1 tr(ˆ
⇢) = 1

R
statistical average hAi = ⇤N dp dq ⇢(p, q)A(p, q) hÂi = tr(ˆ
⇢Â)

@⇢ @ ⇢ˆ i
Liouville’s theorem = {H, ⇢} = [H, ⇢ˆ]
@t @t ~

(Poisson bracket) (commutator)

stationary ensembles {H, ⇢} = 0 [H, ⇢ˆ] = 0

3.3 Ensembles in quantum statistics

We now formulate the quantum statistical physics in terms of the three types of ensembles which
we introduced for the classical statistical physics.

3.3.1 Microcanonical ensemble

The microcanonical ensemble is suitable for closed systems which do not exchange energy and
particles with the environment. We take the density matrix in its diagonal form ⇢nn0 = nn0 |bn |2 .
In this ensemble we consider an incoherent state which consists of the equally distributed set of
states within a certain narrow energy range between E and E + E. Thus,
8
< 1 E  ✏n  E + E
|bn |2 = (3.25)
:
0 otherwise .
The energies ✏n are the eigenenergies of the stationary states | n i for the Hamiltonian H.
Formally we then write the density matrix as
X X
⇢ˆ = | n ih n | with tr⇢ˆ = ⇢nn = !(E) . (3.26)
E✏n E+ E n

38
Here !(E) is the number of quantum states | n i with energy in the given energy range. Anal-
ogously to the classical case we use now !(E) to define the entropy

S(E, V ) = kB ln !(E) . (3.27)

From this we may derive the thermodynamics. Note that !(E) is connected with the density of
states d (E)/dE of the quantum system:
X X d (E)
(E) = 1 ) !(E) = 1= E. (3.28)
dE
"n E E"n E+ E

3.3.2 Canonical ensemble

For a system which is in thermal equilibrium with a reservoir of given temperature T we use the
canoncial ensemble formulation. Analogously to the classical statistical physics we define the
density matrix here as
H(p,q) ✏n
⇢(p, q) = e ! ⇢nn0 = nn0 e , (3.29)

and we define the partition function as

X
✏n F (T,V,N )
Z= e = tr⇢ˆ = e (3.30)
n

where F is the Helmholtz free energy. The density matrix can be written as an operator,
X X
⇢ˆ = | n ie ✏n h n | = e H | n ih n | = e H . (3.31)
n
|n {z }
=1

Thus the partition function can also be expressed as

H
Z = tre (3.32)

and average values of observables are given by

tr(Âe H) 1 ⇣ H
⌘
hÂi = H
= tr Âe . (3.33)
tre Z
The connection to thermodynamics is given via the Helmholtz free energy,

F (T, V, N ) = kB T ln Z . (3.34)

3.3.3 Grand canonical ensemble

Now we connect the system in addition to the heat reservoir also to the particle reservoir of
given chemical potential µ. In the same spirit as for the canonical ensemble we use the analogy
to the classical systems. The density matrix now reads,
b)
(H µN
⇢ˆ = e (3.35)

with N the particle number operator and the grandcanonical partition function is then
b X
Z = tre (H µN ) = z N ZN (3.36)
N

with z = e µ as the fugacity, leading to the grand potential

⌦(T, V, µ) = kB T ln Z . (3.37)

39
3.4 Ideal quantum paramagnet - canonical ensemble
The ideal paramagnet with quantum spins represents a simple example to illustrate some aspects
of quantum statistics before going to the more involved cases.

3.4.1 Spin 1/2

We consider N quantum spin 1/2 moments (S = ~/2) where for each spin the Hilbert space
contains two states, |s, sz i = {|1/2, ±1/2i} for a given quantization axis which we may choose
to be the z-axis. In a magnetic field these moments are described by the Hamiltonian,
N
X N N
µB g X ~ ~ˆ gµB X ~ ˆ
H= Hi = H · Si = H · ~i (3.38)
~ 2
i=1 i=1 i=1

with µB the Bohr magneton and g = 2 the gyromagnetic ratio. The Pauli matrices are defined
as ✓ ◆ ✓ ◆ ✓ ◆
x 0 1 y 0 i z 1 0
ˆ = , ˆ = and ˆ = , (3.39)
1 0 i 0 0 1
and ˆ 0 is the 2 ⇥ 2 unit matrix. The following property for Pauli matrices,

(~a · ~ˆ )(~b · ~ˆ ) = ~a · ~b ˆ 0 + i(~a ⇥ ~b) · ~ˆ (3.40)

yield the simple relation,

(~a · ~ˆ )2n = |~a|2n ˆ 0 and (~a · ~ˆ )2n+1 = |~a|2n~a · ~ˆ (3.41)

and tr~ˆ = 0 and trˆ 0 = 2 in order to determine the partition function2 ,

n o n P ~ ˆ o ⇣ n o⌘N
H µB ~ ~ˆ
Z = tr e = tr e i H·~ i = tr e µB H·

(3.45)
⇣ n o⌘N
~ˆ · ~ˆ sinh( µB H)
= tr ˆ 0 cosh( µB H) + H = (2 cosh( µB H))N ,
2
For independent spin degrees of freedom we use the product state

| s1 ,s2 ,...,sn i = |s1z i ⌦ |s2z i ⌦ · · · ⌦ |sN z i (3.42)

with the short notation | 12 , siz i = |siz i. If the Hamiltonian can be written in terms of independent degrees of
P
freedom, H(s1 , s2 , . . . , sN ) = N i=1 Hi (si ) (all Hi are identical), then the partition function can be split into a
product form
!
⇣ ⌘ Y N YN ⇣ ⌘ ⇣ ⌘N
H Hi
ZN = tr| i e = tr| i e = tr|si i e Hi = tr|s1 i {e H1 } = Z1N . (3.43)
i=1 i=1

Next we calculate the average value hŜ1z i of one single spins

⇣ Q ⌘
tr| i (Ŝ1z e H ) tr| i { i6=1 e Hi }Ŝ1z e H1 Z1N 1
tr|s1 i (Ŝjz e H1
) tr|s1 i Ŝ1z e H1
hŜ1z i = = ⇣ Q ⌘ = = (3.44)
tr| i (e H ) tr| i { i6=1 e Hi }e H1 Z1N Z1

40
where H = |H| ~ˆ = H/H.
~ and H ~ 3 We obtain the free energy

8
< N µB H kB T ⌧ µB H
F = N kB T ln[2 cosh( µB H)] = (3.47)
:
N kB T ln 2 kB T µB H

which is interpreted as F = U T S being dominated by the internal energy U at low and by

the entropy S = kB ln 2 in the high-temperature limit. Note that the entropy is given by
@F
S(T ) = = N kB ln[2 cosh( µB H)] N kB µB H tanh( µB H) (3.48)
@T
which goes to zero for T ! 0 satisfying the third law of thermodynamics. Now we turn to the
magnetization,
N
X
Mn = µ B hn̂ · ~ˆi i = µB N hn̂ · ~ˆ1 i
i=1
n h io
tr µB n̂ · ~ˆ ˆ 0 cosh( µB H) + Ĥ · ~ˆ sinh( µB H)
=N = N n̂ · ĤµB tanh( µB H)
2 cosh( µB H)
(3.49)
~
where n̂ is a normal vector. Let us calculate the magnetic susceptibility for the case that H k ẑ,
first for a moment along z,

zz = µ2B N hˆ z2 i hˆ z i2 = µ2B N hˆ 0 i hˆ z i2
(3.50)
µ2B N µ2B N 1
= 1 tanh2 ( µB H) = .
kB T kB T cosh2 ( µB H)

This yields a high-temperature behavior (kB T µB H)

µ2B N
zz (T ) = (3.51)
kB T
corresponding to a Curie-behavior.

3.4.2 Spin S - classical limit

We now turn to larger quantum spins with magnitude S = ~s where s is an integer or half-
integer. The Hilbert space of such a spin contains 2s + 1 basis states {|s, sz i} = {|s, si, |s, s +
~ˆ2 |s, sz i = ~2 s(s + 1)|s, sz i and Ŝz |s, sz i = ~sz |s, sz i. We apply a
1i, . . . , |s, s 1i, |s, si} with S
magnetic field parallel to the z-axis of the N independent spins, leading to the Hamiltonian,
N N
gµB X ~ˆ ~ gµB X z
H= S·H = Ŝi H , (3.52)
~ ~
i=1 i=1
3
Evaluation of exponential function: Expansion in a Tayler-series,
X1 1 
~ ~
µB H· ˆ ( µB )n ~ ˆ n X ( µB )2ñ 2ñ ( µB )2ñ+1 2ñ+1 ~ˆ ˆ
e = (H · ~ ) = H ˆ0 + H (H · ~ )
n=0
n! ñ=0
(2ñ)! (2ñ + 1)!
(3.46)

~ˆ · ~ˆ sinh( µB H) .
= ˆ 0 cosh( µB H) + H

where we used the relations (3.41).

41
where g = 2 again. Note that the field direction does not matter for the following discussion.
We calculate again the partition function of the canonical ensemble,
(N ) !N !N
n o Y s
X s
X
H gµB H Ŝiz /~ gµB H Ŝiz /~ gµB Hsz
Z = tr e = tr e = hs, sz |e |s, sz i = e
i=1 sz = s sz = s

2s
!N !N ✓ ◆
X gµB H(2s+1) sinh ( µB H(2s + 1)) N
gµB Hs gµB Hn gµB Hs e 1
= e e = e gµB H
= .
e 1 sinh ( µB H)
n=0
(3.53)
We obtain the free energy
⇢
sinh ( µB H(2s + 1))
F = N kB T ln (3.54)
sinh ( µB H)

and the internal energy

n o
U = N µB H (2s + 1) coth ( µB H(2s + 1))) coth ( µB H) = N µB H2sBs ( µB H) (3.55)

where Bs (x) is the ”Brillouin function”.

8 2
>
< (s + 1)x + · · · x ⌧ (2s + 1) 1 ,
2s + 1 1 3
Bs (x) = coth[(2s + 1)x) coth(x) = (3.56)
2s 2s >
:
1 1s e x + · · · x (2s + 1) 1 .

In Fig.3.1 the heat capacity is depicted, which is the derivative of U with respect to temperature.
Note the approach towards the classical limit with increasing spin s (see also discussion for F
below). Also the magnetization can be expressed by the Brillouin function,
N
X
Mz = gµB hŜiz i = N µB 2sBs ( µB H) , (3.57)
i=1

which also identifies the internal energy as U = Mz H. The other components vanish, Mx =
My = 0.

CV
classical
quantum
s = 80
s = 25
s = 10
/ 1/s

kB T
Figure 3.1: As the spin s increases, the ”quantum” heat capacity approaches the classical heat
capacity.

42
The limit of s 1 leads us to the classical limit4 . Let us examine this by simultaneously fixing
2sµB = m to be finite. The free energy and entropy take the limit for kB T mH,

F = N kB T ln 2s and S = N kB ln 2s (3.62)

where we may identify the value of S with S0 of the classical case in the footnote. In the
low-temperature limit we distinguish two regimes:
(1) 2skB T mH kB T : we find

F = N mH + N kB T ln( mH/s) and S = N kB ln 2s N kB ln(2 mH) + N kB (3.63)

which corresponds essentially to the low-temperature classical regime.

(2) mH 2skB T : we obtain
✓ ◆ ⇣ ⌘
1
F = N mH 1 + O T e 2 mH/s and S ! 0 (3.64)
2s
which corresponds to the quantum result. This quantum range shrinks for increasing s, see
Fig.3.1 .

3.5 Ideal quantum gas - grand canonical ensemble

A gas of independent particles in quantum mechanics has the additional key feature that indis-
tinguishable particles are classified in Fermions (particles with half-integer spins) and Bosons
(particles with integer spins). The former follow the Pauli principle, i.e. we cannot find two
Fermions in the same quantum state. For Bosons no such restriction exists. A many-particle
state of Bosons (Fermions) is described by a wavefunction which is completely symmetric (anti-
symmetric) under the pair-wise exchange of particle variables. We will discuss this in detail in
the next chapter when we introduce the technique of second quantization. The free particles in
quantum mechanics are represented by plane waves

1 p~ 2
~ (~
p r) = h~r| ~i
p = p ei~p·~r/~ with ✏p~ = . (3.65)
V 2m
4
Classical ideal paramagnet: We take a classical vector magnetic moment m
~ of length m and assume a magnetic
field along the z-direction. Thus, each moment possesses only the angular orientation as a degree of freedom.
This leads to the partition function,
N ⇢
Z Y ✓Z ◆N ✓ ◆N
d⌦i m ~
~ i ·H d⌦ mH cos ✓ sinh( mH)
Z= e = e = . (3.58)
i=1
4⇡ 4⇡ mH

The free energy is given by

8
>
> N mH + N kB T ln(2 mH) kB T ⌧ mH
<
F F0 = N kB T {ln (sinh( mH)) ln( mH)} = ✓ ◆2 (3.59)
>
> N kB T mH
: kB T mH
3! kB T
and the entropy in the limiting cases,
8
✓ ◆ >
> N kB ln(2 mH) + N kB kB T ⌧ mH
<
@F
S S0 = = ✓ ◆3 (3.60)
@T >
> N mH
H,N : kB T mH
3 kB T
where we observe that in the zero-temperature limit the classical entropy diverges logarithmically in temperature.
In the high-temperature limit the entropy should reach a value connected to the phase space of each magnetic
moment, such that we may choose

S0 = N kB ln(2s + 1) ⇡ N kB ln(2s) and F0 = N kB T ln(2s + 1) ⇡ N kB T ln(2s) . (3.61)

N
corresponding to the count of states in the quantum system: (2s + 1) .

43
Introducing periodic boundary conditions we quantize the momentum quantum number. For
this purpose we assume a cubic system with edge length L and V = L3 . The wavefunction
satisfies the boundary conditions

~ (~
p r + (L, 0, 0)) = ~ (~
p r + (0, L, 0)) = ~ (~
p r + (0, 0, L)) = ~ (~
p r) (3.66)

which is satisfied, if
2⇡~ h
p~ = (nx , ny , nz ) = (nx , ny , nz ) with nx , ny , nz 2 Z . (3.67)
L L
The grand partition function for a gas of such particles is given by
X (E µN ){np~ }
Z= g{np~ } e , (3.68)
{np~ }

with X X
E= ✏p~ np~ and N= np~ . (3.69)
p
~ p
~
P
Here np~ is the number of particles per state, the occupation number, whereby the sum {np~ }
runs over all allowed configurations of occupations.5 The factor g{np~ } is given by
8
>
> 1 indistinguishable particles (Fermions and Bosons) ,
>
<
g{np~ } = Y 1 (3.71)
>
> classical particles (Boltzmann) .
>
: n ! p
~
p
~

Each many-particle quantum state for indistinguishable particles includes all permutations
through the total symmetrization (Bosons) or antisymmetrization (Fermions) of the wavefunc-
tion. For Fermions we find that for each state | p~ i the occupation number is np~ = 0, 1 and for
Bosons np~ = 0, 1, 2, . . . . Thus, calculating the partition function we obtain for Fermions/Bosons,
X hn onp~ n
1
onp~
2
i
Z = ze ✏p~1 ze ✏p~2 ···
np~1 ,np~2 ,...

8 Y⇣ ⌘
>
> 1 + ze ✏p~
Fermions
>
> (3.72)
YX⇣ ⌘np~ >
< p~
✏p~
= ze =
>
> Y 1
p
~ np~ >
> Bosons
>
: ✏p~
p
~
1 ze

with z = e µ as the fugacity.6

5
Note that sums over p
~ are discrete sums over ~n due to Eq.(3.67), but for large L can be approximated well
by integrals
X X Z 3 L3
Z
3 V
Z
= ⇡ d n= d p = d3 p . (3.70)
(2⇡~)3 h3
p
~ ~
n

6
Classical particles: The partition function is given by
8 9N
 <X =
1 n onp~ 1 n onp~
X 1
X 1
X
✏p ✏p 1 (class)
zN ✏p
z N ZN
1 2
Z= ze ~ 1 ze ~ 2 ··· = e ~
= ,
np
np~1 ! np~2 ! N! : ;
~1 ,np
~2 ,... N =0 p
~ N =0
(3.73)
which corresponds to the usual partition function of independent classical particles within the grand canonical
ensemble.

44
From the partition function we arrive at the equation of state
8 X ⇣ ⌘
>
> ln 1 + ze ✏p~
Fermions ,
>
>
>
< p~
pV
= ⌦(T, V, µ) = (3.74)
kB T > X ⇣
>
⌘
>
> ln 1 ze ✏p~
Bosons ,
>
:
p
~

and the particle number

8 X
> 1
>
> Fermions ,
> ✏p~ z 1
@ < p~ e
> +1
N =z ln Z = (3.75)
@z >
> X 1
>
> Bosons .
>
: ✏p~ z 1
p
~
e 1

These equations can be used to determine the chemical potential when N is fixed. The occupa-
tion number of a state is also easily obtained,
8 9
1 <Y X⇣ ⌘np~0 = X
hnp~ i = ze ✏p~0 n (ze ✏p~ )np~
Z:0 n ; n p~
~ 6=p
p ~ ~0
p p
~

8
> 1 (3.76)
P >
> Fermions ,
np~ (ze ✏p~ )np~
@ < z 1 e ✏p~ +1
np~
= P ✏p~ )np~
= kB T ln Z =
np~ (ze @✏p~ >
> 1
Bosons . >
:
z 1 1 e ✏p~

These correspond to the Fermi-Dirac and the Bose-Einstein distribution, respectively.

Eventually we compare the grand potential and the internal energy. Using (3.74) we obtain
X ⇣ ⌘ Z ⇣ ⌘
✏p~
⌦ = ⌥kB T ln 1 ± ze = ⌥kB T d✏ g(✏) ln 1 ± ze ✏
p
~

✓ ◆3/2 Z 1
1 2m (3.77)
= ⌥V kB T 2 d✏ ✏1/2 ln(1 ± ze ✏
)
4⇡ ~2 0

✓ ◆3/2 Z 1
2 1 2m ✏3/2
= V 2 d✏ ,
3 4⇡ ~2 0 z 1e ✏ ± 1
where the plus (minus) sign corresponds to Fermions (Bosons). For the second equality we
performed an integration by parts. We have also introduced the density of states
X ✓ ◆
1 2m 3/2 1/2
g(✏) = (✏ ✏p~ ) = V 2 ✏ , (3.78)
4⇡ ~2
p
~

for the energy integration. Now we consider the internal energy

X Z 1 ✓ ◆ Z
✏ 1 2m 3/2 1 ✏3/2
U= hnp~ i✏p~ = d✏ g(✏) 1 ✏ =V 2 d✏ . (3.79)
0 z e ±1 4⇡ ~2 0 z 1e ✏ ± 1
p
~

Thus from ⌦ = pV we obtain the general relation

3
U = pV . (3.80)
2
Note that we did not so far include the spin s of the particles in our considerations. This gives
an additional factor 2s + 1 to ⌦ and U .

45
3.6 Properties of Fermi gas
We consider now Fermions of spin s, i.e. there are 2s + 1 di↵erent species. The equation of state
and the equation for the particle number of the Fermi gas are given by
Z 1 ⇣ ⌘
p 4⇡
= 3 (2s + 1) dp p2 ln 1 + ze ✏p~ ,
kB T h 0
Z 1 (3.81)
1 N 4⇡ 1
= = n = 3 (2s + 1) dp p2 1 ✏ ,
v V h 0 z e p~ + 1
P R
where we used the relation p~ = hV3 d3 p (see previous footnote). We rewrite these equations
introducing special functions of z,
p 2s + 1
= 3
f5/2 (z)
kB T
(3.82)
1 2s + 1
n= = 3
f3/2 (z)
v
p
where = h/ 2⇡mkB T is the thermal wavelength. These functions are defined as7
Z 1 X1
4 2 zl
f5/2 (z) = p dx x2 ln(1 + ze x ) = ( 1)l 5/2
⇡ 0 l l=1
(3.84)
1
X
@ zl
f3/2 (z) = z f (z) = ( 1)l .
@z 5/2 l3/2
l=1

For the following it will also be important to consider the asymptotic behavior of these functions.
For z ⌧ 1 we may use the expansion just given. For z 1 we find8

8 5/2 5⇡ 2
f5/2 (z) ⇡ p (ln z) 1+ + ···
15 ⇡ 8(ln z)2
 (3.88)
4 ⇡ 2
f3/2 (z) = p (ln z)3/2 1 + + ··· .
3 ⇡ 8(ln z)2
7
Expansion of f5/2 (z):
Z 1 Z 1 1
X l lx2 1
X
4 2 x2 4 2 lz e zl
p dx x ln(1 + ze )= p dx x ( 1) = ( 1)l . (3.83)
⇡ 0 ⇡ 0 l=1
l
l=1
l5/2

8
Derivation of the large-z-expansion: Use ⌫ = µ/kB T = ln z and rewrite
Z 1 Z 1 0
2 8 ey
f5/2 (z) = p dy y 1/2 ln(1 + e⌫ y ) = p dy 0 (y 0 + ⌫)5/2 , (3.85)
⇡ 0 15 ⇡ ⌫ (1 + ey0 )2
where the second equation is obtained by twice integrating by parts. More over we extend the lower integration
boundary to 1, at the expense of a small error of order e ⌫ = 1/z. We now expand (y 0 + ⌫)5/2 assuming ⌫
large and find
Z +1  0
8 5 15 1/2 0 2 ey
f5/2 (z) = p dy ⌫ 5/2 + ⌫ 3/2 y 0 +
0
⌫ y + ···
15 ⇡ 1 2 8 (1 + ey0 )2
(3.86)

8 5/2 5 3/2 15 1/2
= p I0 ⌫ + I1 ⌫ + I2 ⌫ + ···
15 ⇡ 2 8
where Z +1
y n ey ⇡2
In = dy ) I0 = 1, I1 = 0, I2 = ,... . (3.87)
1 (1 + ey )2 3
Note that all In = 0 for n odd. It is easy to obtain f3/2 (z) by taking the derivative.

46
3.6.1 High-temperature and low-density limit
Low density (high temperature) implies 3n ⌧ 1 and z = e µ ⌧ 1 . We can use now (3.82)
⇢
3 z2 3n ( 3 n)2
n = (2s + 1) z + ··· ) z= + + ··· . (3.89)
23/2 2s + 1 23/2 (2s + 1)2

This allows us to replace z in the equation of state 9 ,

⇢ ⇢
p 2s + 1 z2 N 3n
⇡ 3
z ⇡ 1 + . (3.91)
kB T 25/2 V 25/2 (2s + 1)

The second term represents the first quantum correction to the classical limit of the ideal gas.
This allows us also to calculate the isothermal compressibility,
✓ ◆
1 @V V 1
T = = . (3.92)
V @p T,N N kB T 1 + 3/2 3 n
2 (2s+1)

The quantum correction suppresses the compressibility, as a consequence of Pauli’s exclusion

principle, i.e. Fermions avoid each other. The occupation numbers is approximatively given by

1 3n 1 nh3
✏p~ ✏p~ ✏p~
hnp~ i = 1 e ✏p~
⇡ ze ⇡ e = e (3.93)
z +1 2s + 1 2s + 1 (2⇡mkB T )3/2

corresponding to the Maxwell-Boltzmann distribution per spin. With the general relation (3.80)
we obtain immediately the internal energy and the heat capacity
⇢ 3n
⇢
3 3 N 3
U ⇡ N kB T 1 + 5/2 and CV = N kB 1 , (3.94)
2 2 (2s + 1) 2 27/2 (2s + 1)V

including the first quantum corrections.

3.6.2 Low-temperature and high-density limit: degenerate Fermi gas

At low temperature we reach the ”quantum limit” 3 v = V /N , which implies large fugacity z.
First we consider the zero-temperature situation. The occupation numbers follow a step function
8
< 1 , ✏p~ < µ ,
hnp~ i = ⇥(µ ✏p~ ) = (3.95)
:
0 , ✏p~ > µ .

The Fermions occupy states within a sphere in momentum space, the Fermi sphere (Fig.3.2).
The particle density n is
Z
N 2s + 1 3 2s + 1 4⇡ 3 (2s + 1)kF3
n= = d phn p
~ i = p F = (3.96)
V h3 h3 3 6⇡ 2
where pF is the Fermi momentum (✏p~F = µ(T = 0) = ✏F ), isotropic, and kF = pF /~ is the Fermi
wavevector. The groundstate energy is
Z Z pF
2s + 1 3 2s + 1 2 p
2 2s + 1 4⇡V 3 p2F 3
U0 = 3
V d p✏ p
~ hn p
~ i = 3
4⇡V dpp = 3
pF = N ✏F (3.97)
h h 0 2m h 5 2m 5
9
Note that the lowest order expansion reproduces the chemical potential of the classical ideal gas, as given in
Eq.(2.51) and (2.78): ( ✓ ◆3/2 )
µ N 3 V 2⇡mkB T
z=e ⇡ ) µ = kB T ln . (3.90)
V N h2

47
where ✏F denotes the Fermi energy. The zero-point pressure is obtained through (3.74),
2 U0 2N
p0 = = ✏F . (3.98)
3V 5V
In contrast to the classical ideal gas, a Fermi gas has finite zero-point pressure which is again
a consequence of the Pauli principle and is responsible for the stability of metals, neutron stars
etc.

p
z

p
F py

px

Figure 3.2: Fermi sphere of occupied single particle states. Fermi radius pF .

Next we turn to finite temperatures for which the occupation number broadens the step at pF .
We use now (3.88, 3.96) to obtain the relation
✓ ◆ p ✓ ◆3/2 ✓ ◆ 1/2
✏F 3/2 3 ⇡ 3 N µ ⇡2 µ
= = + + ··· , (3.99)
kB T 4 2s + 1 V kB T 8 kB T
which at constant density n = N/V can be solved for the chemical potential,
✓ ◆ !
⇡ 2 kB T 2
µ(T ) = ✏F 1 + ··· , (3.100)
12 ✏F

and analogously we obtain for the pressure,

✓ ◆2 !
5⇡ 2 kB T
p(T ) = p0 1+ + ··· . (3.101)
12 ✏F

Again we derive the internal energy from the relation (3.74)

✓ ◆ !
3 5⇡ 2 kB T 2
U = pV = U0 1 + + ··· , (3.102)
2 12 ✏F

which also leads to the heat capacity for fixed N

✓ ◆ 2T
@U 5⇡ 2 kB ⇡2 2 N
CV = = U0 2 + · · · = k T + ··· , (3.103)
@T V,N 6 ✏F 2 B ✏F

see Fig.3.3. This is the famous linear temperature dependence of the heat capacity, which can be
well observed for electrons in simple metals: CV = T +· · · with as the Sommerfeld coefficient.
T !0
Obviously now the third law of thermodynamics is satisfied, CN ! 0. Also the entropy goes
to zero linearly in T .

48
 CV
U 3
2
N kB
3
2
N kB T

quantum quantum
3 classic classic
5
N ✏F
0 0
0 kB T 0 kB T

Figure 3.3: The internal energy U and the heat capacity CV in the quantum situation compared
with the classical situation.

3.6.3 Spin-1/2 Fermions in a magnetic field

We consider now the magnetic response of spin s = 1/2 Fermions in a magnetic field (ideal
paramagnetic gas). The Hamiltonian has to be extended by a Zeeman term. Taking the field
along the z-axis this reads,
N
gµB X z
HZ = si H (3.104)
~
i=1

with g = 2 as gyromagnetic ratio and µB the Bohr magneton. The energy di↵erence between
spins parallel and antiparallel to the applied field can be absorbed into a spin-dependent fugac-
ity,10
z± = e µ± µB H , (3.108)
such that the density of Fermions is given by
N 1 1
n= = = 3 f3/2 (z+ ) + f3/2 (z ) = n+ + n (3.109)
V v
and the magnetization
M µB
m= = µB (n+ n )= 3
f3/2 (z+ ) f3/2 (z ) . (3.110)
V
Let us now calculation the spin susceptibility for zero magnetic field, given by

@m µ2B @f3/2 (z) 2µ2B

= = 3k T
2z = f (z)
3 k T 1/2
(3.111)
@H H=0 B @z H=0 B
10
We calculate the grand canonical partition function
8 98 9
Y <X ⇣ ⌘np~ = <X ⇣ ⌘np~ = Y Y X⇣ ⌘np~
Z= ze ✏p~ + µB H ze ✏p~ µB H
= z e ✏p
~
(3.105)
: ;: ;
p
~ np
~ np
~ p
~ =+, np
~

where z is defined as in Eq.(3.108). The grand canonical potential is given by

kB T
⌦= kB T ln Z = 3
f5/2 (z+ ) + f5/2 (z ) (3.106)

from which we determine the magnetization

1 @⌦ kB T X @ kB T X @z @
m = = 3 f5/2 (z ) = 3 f5/2 (z)
V @H @H @H @z z=z
(3.107)
µB X @ µB X
= 3
z f5/2 (z) = 3
f3/2 (z )
@z z=z

corresponding to Eq.(3.110).

49
with z = e µ . We may now again consider limiting cases.
High-temperature limit: We replace z ⌧ 1 in Eq.(3.111) using Eq.(3.89) with n = 1/v and find
⇢
µ2 n 3n
= B 1 . (3.112)
kB T 25/2

The first term is the result for particles with spin and has a Curie like behavior and the second
term is the first quantum correction reducing the susceptibility.
Low-temperature limit: Taking only the lowest order for z 1 we obtain,
✓ ◆1/2
µ2 4 µ2B 4 ✏F 3n
= 3 B p (ln z)1/2 = 3k T
p = µ2B . (3.113)
kB T ⇡ B ⇡ kB T 2✏F

This is the famous Pauli spin susceptibility for a Fermi gas, which is temperature independent.

3.7 Bose gas

There are two situations for Bosons: (1) a system with well-defined particle number, e.g. bosonic
atoms, 4 He, ... ; (2) Bosons which results as modes of harmonic oscillators (no fixed particle
number), e.g. photons, phonons, magnons, etc..

3.7.1 Bosonic atoms

We consider Bosons without spin (S = 0) for which 4 He is a good example. Analogously to the
Fermions we introduce functions of z to express the equation of state and the particle number,
1
p 1 1 X zl
= 3 g5/2 (z) = 3
kB T
l=1
l5/2
(3.114)
1
1 N 1 1 X zl
= = 3 g3/2 (z) = 3 .
v V
l=1
l3/2

For small z both functions grow linearly from zero and g3/2 (z) has a divergent derivative for
z ! 1. We concentrate on the range 0 < z  1, such that µ(T )  0. For z = 1 we obtain
X 1 X 1
g3/2 (1) = = ⇣(3/2) ⇡ 2.612 and g5/2 (1) = = ⇣(5/2) ⇡ 1.342 (3.115)
l
l3/2 l
l5/2

where ⇣(x) is Riemann’s ⇣-function (see Fig.3.4).

3.7.2 High-temperature and low-density limit

It is easy to see that (like the Fermions) the Bosons behave in this limit as a classical ideal gas.
An intriguing aspect occurs, however, in the quantum corrections. For the pressure we find
⇢
N N 3
p(T ) = kB T 1 + ··· . (3.116)
V 25/2 V

The quantum correction reduces the classical ideal gas pressure and yields the compressibility
✓ ◆
1 @V V 1
T = = 3N . (3.117)
V @p T,N N kB T 1 3/2 2 V

50
 g 3/2

g 5/2

Figure 3.4: Functions g3/2 (z) and g5/2 (z).

In contrast to the Fermions where the quantum nature of the particles diminishes the compress-
ibility, here the compressibility is enhanced. Actually, in this approximation the compressibility
even diverges if
N 3
23/2 = , (3.118)
V
i.e. at low enough temperature or high enough density. We will now see that this indeed indicates
an instability of the Bose gas.

3.7.3 Low-temperature and high-density limit: Bose-Einstein condensation

Let us now consider Eq. (3.114) carefully. The function g3/2 (z) is monotonically increasing with
z. If T is lowered, / T 1/2 increases, such that z has to increase too in order to satisfy (3.114).
Therefore µ approaches the singular point at 0 (z = 1). The critical point is determined by
8
>
> h2
>
> T c = ,
N 3 < 2⇡kB m[⇣(3/2)V /N ]2/3
g3/2 (1) = ⇣(3/2) = ) (3.119)
V >
> 3
>
> N h
: Vc = .
⇣(3/2) (2⇡mkB T )3/2

This defines a critical temperature Tc and critical volume Vc below which a new state of the
Bose gas occurs. Note that this equation is qualitatively very similar to (3.118) and even
quantitatively not so far (⇣(3/2) ⇡ 2.612 $ 23/2 ⇡ 2.85) . The question arises what happens for
T < Tc or V < Vc . Actually the problem occurring in (3.114) and (3.119) arises in the step
X Z
1 V 1
N= (✏ µ)
! 3 d3 p (✏ µ) . (3.120)
e p
~ 1 h e p
~ 1
p
~

The integral does not count the occupation of the state p~ = 0, because the momentum distribu-
tion function entering the integral,

p2
⇢(p) = (✏p~ µ)
) ⇢(0) = 0 . (3.121)
e 1
This is fine as long as the occupation of the ’~
p = 0’-state (single-particle groundstate) is vanish-
ingly small compared to N . However, for T < Tc (V < Vc ) the occupation becomes macroscopic,

51
hnp~=0 i/N > 0 and we cannot neglect this contribution in the calculation of N (see Fig.3.5). Thus,
the correct density is
N 1
= 3 g3/2 (z) + n0 (T ) = nn (T ) + n0 (T ) (3.122)
V
with n0 (T ) denoting the density of Bosons in the single-particle groundstate (~ p = 0). These
particles form a condensate, the Bose-Einstein condensate. What happens at Tc is a phase
transition. We encounter here a ”two-fluid” system for which the total particle density split
into a condensed fraction n0 and a normal fraction nn . From (3.122) we find the temperature
dependence of n0 (see Fig.3.5),
" ✓ ◆3/2 #
N T
n0 (T ) = 1 . (3.123)
V Tc

n0
n
1

n0 nn

p Tc T

Figure 3.5: Occupation: Left panel: A macroscopic fraction of particle occupy the momentum
p = 0-state for T < Tc . Right panel: Temperature dependence of the condensate fraction.

Next we also determine the equation of state,

8
> kB T g5/2 (z) , V > Vc
>
>
< 3
p= . (3.124)
>
>
: kB T g5/2 (1) , V < Vc
>
3

We now consider the compressibility for V > Vc . For this purpose we first determine
0 (z)
g3/2
@V 3
= N , (3.125)
@z g3/2 (z)2

and consider
0 (z)
g3/2
@p kB T 0 @z N 6
= 3 g5/2 (z) ) T = 0 (z) , (3.126)
@V @V V kB T g3/2 (z)2 g5/2

where we use the notation gn0 (z) = dgn (z)/dz. As anticipated earlier the compressibility diverges
at the transition V ! Vc (or T ! Tc ), since the derivative g3/2 0 (z) ! 1 for z ! 1. In the

condensed phase the pressure is independent of V as is obvious from (3.124). Therefore the
condensed phase is infinitely compressible, i.e. it does not resist to compression.

52
Some further thermodynamic quantities can be derived. First we consider the entropy S from
the grand canonical potential
8 ✓ ◆
> 5v
>
> N kB g (z) ln z , T > Tc ,
✓ ◆ ✓ ◆ >
< 2 3 5/2
@⌦ @pV
S(T, V, µ) = = = ✓ ◆3/2 (3.127)
@T V,µ @T V,µ > > g (1)
>
> 5 5/2 T
: N kB , T < Tc ,
2 g3/2 (1) Tc

where we used (3.114).11 For the heat capacity at fixed particle number N we find from the
internal energy U = 32 pV ,
8 ✓ ◆
>
> 15v 9 g3/2 (z)
>
> N kB g (z) , T > Tc ,
✓ ◆ >
< 4 3 5/2 4 g1/2 (z)
@U
CV = = (3.131)
@T V,N > > ✓ ◆3/2
>
> 15 g5/2 (1) T
>
: N kB , T < Tc .
4 g3/2 (1) Tc

CV
Nk B

3/2

Tc
T

Figure 3.6: Heat capacity: CV has a cusp at the transition and vanishes as T 3/2 towards
zero-temperature. In the high-temperature limit Cv approaches 3N kB /2 which corresponds to
the equipartition law of a mono-atomic gas.

In accordance with the third law of thermodynamics both the entropy and the heat capacity go
to zero in the zero-temperature limit. The entropy for T < Tc can be viewed as
✓ ◆3/2
S T nn (T ) 5 g5/2 (1)
=s = s with s = kB (3.132)
N Tc n 2 g3/2 (1)
11
Calculation of the temperature derivatives: (1) Fixed chemical potential:
@ V kB T 5V kB V kB T g3/2 (z) @z
3
g5/2 (z) = 3
g5/2 (z) + 3
(3.128)
@T | {zz @T}
V
= kB 3
g3/2 (z) µ = N kB ln z

@
where we used @z g5/2 = g3/2 /z.
(2) Fixed particle number: we use
N 3 dg3/2 g1/2 (z) dz 3N 3
g3/2 (z) = ) = = (3.129)
V dT z dT 2V T
which leads to the relation
3
dg5/2 g3/2 (z) dz 9 g3/2 (z) N
= = . (3.130)
dT z dT 4 g1/2 (z) V
This leads to the expression for the heat capacity.

53
where s is the entropy per normal particle (specific entropy), i.e. a non-vanishing contribution
to the entropy is only provided by the normal fraction (two-fluid model). The heat capacity has
a cusp at T = Tc .

p transition line
p
transition line

isothermal
T

BEC

vc(T) v T

Figure 3.7: Phase diagrams; left panel: p-v-diagram; the isothermal lines reach the transition
line with zero-slope, i.e. the compressibility diverges. Right panel: p-T -diagram; the condensed
phase corresponds to the transition line, there is no accessible space above the transition line.

We consider now the phase diagram of various state variable.

(1) p-v-diagram: phase transition line

h2 g5/2 (1)
p0 v 5/3 = , (3.133)
2⇡m [g3/2 (1)]5/3

(2) p-T -diagram: phase transition line

kB T
p0 = 3
g5/2 (1) / T 5/2 (3.134)

which is the vapor pressure (constant for T < Tc ) (see Fig.3.7). We use this line to determine
the latent heat l per particle via the Clausius-Clapeyron relation,
dp0 l
= with l=T s. (3.135)
dT T v
The condensate takes no specific volume compared to the normal fraction. Thus, v = vc .
Therefore we obtain
dp0 5 kB g5/2 (1) 5 g5/2 (1)
l = T vc = T vc 3
= T kB (3.136)
dT 2 2 g3/2 (1)

where we used the relation 3 = vc g3/2 (1). Note that this is consistent with our result on
the specific entropy s. The condensed phase has no specific entropy such that s = s and,
consequently, l = T s using (3.132).

Examples of the Bose-Einstein condensates is the quantum fluid 4 He which shows a condensation
below T ⇡ 2.18K into a superfluid phase. We will discuss this in more detail in Chapt. 7. A
further very modern example are ultracold atoms in optical traps, e.g. 87 Rb (37 electrons +
87 nucleons = 124 Fermions ! Boson). For 2000 atoms in the trap the critical temperature to
Bose-Einstein condensation is as low as 170 nK (for the measured momentum distribution see
Fig.3.8).

54
Figure 3.8: Velocity distribution of Rb-atoms: Left panel: T > Tc ; middle panel: T ⇠ Tc ;
right panel T ⌧ Tc . A peak at the center develops corresponding to the fraction of particles
with ”zero-velocity” and at the same time the fraction of particles with finite velocity shrinks.
(Source: http://www.nist.gov/public a↵airs/gallery/bosein.htm)

3.8 Photons and phonons

We consider now classes of Bose gases whose particle numbers is not conserved. They are derived
as normal modes of harmonic oscillators. Thus we first consider the statistical physics of the
harmonic oscillator. The most simple example is the one-dimensional harmonic oscillator whose
spectrum is given by
✓ ◆
1
✏n = ~! n + with n = 0, 1, 2, . . . (3.137)
2

and the eigenstates |ni.12 The quantum number n is considered as the occupation number of
the oscillator mode. We analyze this within the canonical ensemble formulation with a given
temperature T . The partition function reads
1
X 1
X 1
X
H H ✏n ~!/2 ~!n e ~!/2
Z = tre = hn|e |ni = e =e e = . (3.141)
1 e ~!
n=0 n=0 n=0

The internal energy is obtained through

@ ln Z 1 ~!
U= = ~! + ~! . (3.142)
@ 2 e 1
12
Harmonic oscillator with the Hamiltonian
✓ ◆
P̂ 2 !2 2 † 1
H= + Q̂ = ~! a a + (3.138)
2 2 2

with a and a† as the lowering and raising operators, respectively,

r 9
~ >
>
Q̂ = (a + a† ) >
>
=
2!
r ) [a, a† ] = 1 , [Q̂, P̂ ] = i~ (3.139)
>
>
~ >
P̂ = i! (a a† ) >;
2!
The stationary states |ni are defined by H|ni = ✏n |ni and obey
p p
a|ni = n|n 1i and a† |ni = n + 1|n + 1i . (3.140)

We can interpret a† and a as creation and annihilation operator for a particle of energy ~!. In the language of
second quantization the commutation relation of a† and a corresponds to that of Bosons.

55
The heat capacity is
✓ ◆2
dU ~! 1
C= = kB 2 , (3.143)
dT 2kB T sinh ( ~!/2)
with the limiting properties
8
>
> kB kB T ~!
<
C= ✓ ◆2 (3.144)
>
> ~! ~!
: kB e kB T ⌧ ~! .
kB T

In the high-temperature limit the heat capacity approaches the equipartition law of a one-
dimensional classical harmonic oscillator. The mean quantum number is given by
1
1 X ✏n 1
hni = ne = ~!
. (3.145)
Z e 1
n=0

This corresponds to the Bose-Einstein distribution function. Thus we interpret n as a number

of Bosons occupying the mode !.

3.8.1 Blackbody radiation - photons

Electromagnetic radiation in a cavity is a good example of a discrete set of independent harmonic
oscillators. Consider a cubic cavity of edge length L. The wave equation is expressed in terms
of the vector potential ✓ ◆
1 @2 ~ 2 ~=0
r A (3.146)
c2 @t2
and the electric and magnetic field are then
~
1 @A
~ =
E and ~ =r
B ~ ⇥A
~, (3.147)
c @t
~ ·A
where we used the Coulomb gauge r ~ = 0 and = 0. This can be solved by a plane wave,
8
1 X n o < ! = !~k = c|~k| ,
~ i~k·~
r i!t ⇤ ⇤ i~k·~
r+i!t
A(~r, t) = p A~k ~e~k e + A~k ~e~k e with
V ~ :
k, ~e~k · ~k = 0 ,
(3.148)
i.e. a linear dispersion relation and a transverse polarization ~e~k ( : polarization index). As-
suming for simplicity periodic boundary conditions in the cube we obtain the quantization of
the wavevector,
~k = 2⇡ (nx , ny , nz ) with ni = 0, ±1, ±2, . . . (3.149)
L
Each of the parameter set (~k, ) denotes a mode representing an independent harmonic oscillator.
These oscillators can be quantized again in the standard way.13 The states of a mode di↵er by
energy quanta ~!~k . The occupation number n~k is interpreted as the number of photons in this
mode.
13
Canonical quantization of the radiation field: Introduce the variables
1 i!~
Q~k = p A~k + A~⇤k and P~k = p k A~k A~⇤k (3.150)
4⇡c 4⇡c
which leads to the following expression for the Hamiltonian
Z X !~2
E~2 + B~2 2 1X 2
H = d3 r = k
A~ = P~k + !~k2 Q~2k . (3.151)
8⇡ 2⇡c2 k 2
~
k, ~
k,

56
The partition function is then derived from that of a harmonic oscillator
!2
Y e ~!~k /2 Y e ~!~k /2
Z= = (3.153)
~
1 e ~!~k ~
1 e ~!~k
k, k

where the exponent 2 originates from the two polarization directions. The internal energy follows
from
X ~!~ Z Z
@ ln Z k ~!
U (T ) = =2 = d!D(!) ~! =V d!u(!, T ) (3.154)
@ e ~!~k 1
~k
e 1

where we have neglected the zero point motion term (irrelevant constant). The density of modes
in (3.154) is denoted as
X Z
2V 2 !2
D(!) = (! !~k ) = 4⇡ dk k (! ck) = V (3.155)
(2⇡)3 ⇡ 2 c3
~k,

which leads to the spectral energy density

!2 ~!
u(!, T ) = 2 3 ~!
, (3.156)
⇡ c e 1
which is the famous Planck formula (Fig.3.9). There are two limits
8
>
> !2
>
< k B T ~! ⌧ kB T Rayleigh-Jeans-law
⇡ 2 c3
u(!, T ) ⇡ (3.157)
>
>
: ~! e ~! ~!
3
>
kB T Wien’s law
⇡ 2 c3
whereby the Rayleigh-Jeans law corresponds to the classical limit. The maximum for given T
follows Wien’s displacement law,
~!0 = 2.82kB T . (3.158)
The total internal energy density leads to the Stefan-Boltzmann law
Z
U ⇡ 2 (kB T )4
= d!u(!, T ) = / T4 . (3.159)
V 15 (~c)3
The energy current density of a blackbody is defined as
U energy
c= . (3.160)
V area · time
Thus the emission power of electromagnetic radiation per unit area for the surface of a blackbody
is defined by
Z 0 ~k · ~n Z 0
U 1 U 1 Uc ⇡ 2 (kB T )4
Pem = c d⌦~k = c d⌦~k cos ✓ = = = T4 (3.161)
V 4⇡ ~
|k| V 4⇡ 4V 60 ~ 3 c2

where for the current density the component

R0 perpendicular to the surface counts (~n: surface
normal vector). Note that the integral d⌦~k only extends over the hemisphere with cos ✓ > 0.

This is the Hamiltonian of a harmonic oscillator for each mode which we can quantize and obtain the new form
X ✓ ◆ X ✓ ◆
1 1
H= ~!~k a~†k a~k + = ~!~k n~k + (3.152)
2 2
~
k, ~
k,

where A~k ! a~k annihilates and A~⇤k ! a~†k creates a photon in the mode (~k, ).

57
 Rayleigh−Jeans
Planck

Wien

h ω /kB T

Figure 3.9: Spectral density of black body radiation.

This blackbody radiation plays an important role for the energy budget of the earth. The sun
can be considered a blackbody emitting an energy current at the temperature of T ⇡ 6000K.
This delivers an energy supply of 1.37kW/m2 to the earth. The earth, on the other hand, has
to emit radiation back to the universe in order not to heat up arbitrarily. The earth is not a
black body but a ”gray” body as it is strongly reflecting in certain parts of the spectrum. A
further example of blackbody radiation is the cosmic background radiation at a temperature
2.73 K which originates from the big bang.

3.8.2 Phonons in a solid

We consider Debye’s theory of the lattice vibration and their influence on the thermodynamics of
a solid. A solid consists of atoms which form a lattice. They interact with each other through a
harmonic potential. Let us assume that the solid consists of NA atoms arranged in a cube of edge
length L, i.e. there are 3NA degrees of freedom of motion. For our purpose it is convenient and
sufficient to approximate this solid as a homogeneous isotropic elastic medium whose vibration
are described by the following equations of motion:
1 @ 2 ~u ~ r~ · ~u) = 0 longitudinal sound mode ,
r(
c2l @t2
(3.162)
1 @ 2 ~u ~ 2 ~u = 0
r transversal sound mode .
c2t @t2
There are two independent transversal (~k · ~u = 0) and one longitudinal (~k ⇥ ~u = 0) sound
mode. These equation can be solved by plane waves and yield linear dispersion analogous to
the electromagnetic waves:
(l) (t)
!~ = cl |~k| and !~ = ct |~k| . (3.163)
k k
The density of states is obtained analogously using periodic boundary conditions for the waves,
✓ ◆
V !2 1 2
D(!) = + . (3.164)
2⇡ 2 c3l c3t
A di↵erence occurs due to the finite number of degrees of freedom. In the end we get 3NA
modes. Thus there must be a maximal value of ! and |~k|. We take the sum
X Z kD 3 ✓ 2 ◆1/3
3V 2 V kD 6⇡ NA
3NA = 3= 4⇡ dk k = ) kD = (3.165)
(2⇡)3 0 2⇡ 2 V
|~k|kD

58
and define in this way the Debye wave vector kD and the Debye frequency !D = cef f kD where
✓ ◆
3 1 2
= + . (3.166)
c3ef f c3l c3t

The internal energy is obtained again in the same way as for the electromagnetic radiation apart
from the limit on the frequency integration,
Z !D
U (T )
= d! u(!, T ) . (3.167)
V 0

D( ω ) D( ω )

ω ω
ωD
Debye real materials

Figure 3.10: Density of states of phonons. Left panel: Debye model; right panel: more realistic
spectrum. Note that the low frequency part in both cases follows an ! 2 law and leads to the T 3 -
behavior of the heat capacity at low temperature. This is a consequence of the linear dispersion
which is almost independent of the lattice structure and coupling.

We consider first the limit of small temperatures kB T ⌧ kB ✓D = ~!D (✓D : Debye temperature).
The internal energy is given by
Z 1 ✓ ◆
(kB T )4 3 x3 ⇡ 2 (kB T )4 3⇡ 4 kB T T 3
U (T ) = V 4⇡ dx x =V = NA (3.168)
(2⇡~)3 c3ef f e 1 10~3 c3ef f 5 ✓D
|0 {z }
= ⇡ 4 /15

and correspondingly the low-temperature heat capacity is

✓ ◆3
12⇡ 4 T
CV = NA kB , (3.169)
5 ✓D

the famous Debye law. On the other hand, at high temperatures (T ✓D ) we use

1 1 1 ~!
~!
⇡ + . (3.170)
e 1 ~! 2 12
This yields for the internal energy
Z !D ✓ ◆
3V ~! 3 ~2 ! 4
U (T ) = d! ! 2 kB T + + ···
2⇡ 2 c3ef f 0 2 12kB T
( ✓ ◆2 ) (3.171)
3 ~!D 1 ~!D
= 3NA kB T 1 + + ···
8 kB T 20 kB T

59
and leads to the heat capacity
⇢ 2
1 ✓D
CV = 3NA kB 1 + ··· . (3.172)
20 T 2
In the high-temperature limit the heat capacity approaches the value of the equipartition law
for 3NA harmonic oscillators (Fig.3.11).

C
Dulong−Petit
3Nk B

3
T T
θD
Figure 3.11: Heat capacity: Low-temperature behavior follows a T 3 -law. At high-temperature
the universal Dulong-Petit law recovers, which is a classical result of the equipartition law for
particles in a harmonic potential.

The Debye temperature lies around room temperature usually. However, there also notable
exception such as lead (Pb) with ✓D = 88K or diamond with ✓D = 1860K.

3.9 Diatomic molecules

We now investigate the problem of the diatomic molecules
P (made out of N atoms) which are
bound by a two-particle interaction potential V (r) = ri ~rj |).
i<j Vij (r) with Vij (r) = v(|~
A good approximation for the binding potential is the Lenard-Jones potential composed of an
attractive and a ’hard-core’ repulsive part,
"✓ ◆ ✓ ◆6 #
r̃ 12 r̃
v(r) = 4✏ (3.173)
r r

with ✏ as the potential depth and r0 = 21/6 r̃ as the minimal point, see Fig.3.12 . Quantum
mechanical aspects appear in the low-temperature regime kB T ⌧ ✏ (✏: depth of the Lenard-Jones
potential). Under this condition we consider the Lenard-Jones potential around its minimum as
a harmonic potential. Ignoring for the moment the motion of the center of mass, the Hamiltonian
can be restricted to the relative coordinates r and relative momenta p~,
p~2 p~2 p2r ~2
L
H= + v(r) = + A(r r0 ) 2 ✏⇡ + + A(r r0 ) 2 ✏ (3.174)
2m⇤ 2m⇤ 2m⇤ 2m⇤ r02
where the reduced mass is given by m⇤ = m/2 and pr is radial momentum along the axis of
the molecule bond. The motion separates into radial vibrational part which corresponds to a
one-dimensional harmonic oscillator and the rotational motion. The two independent spectra
are given by ✓ ◆
vib 1 ~2 l(l + 1)
En = ~! n + and Elrot = (3.175)
2 2m⇤ r02

60
 V

r −12

r
r −6

harmonic

Figure 3.12: Lenard-Jones potential.

p
with ! = 2A/m⇤ .
Let us now analyze the partition function of the di↵erent degrees of freedom. The translational
degree of freedom (center of mass motion) corresponds to the motion of a particle of mass 2m
in three dimensions. At low temperatures this has to be treated as a bosonic quantum liquid in
the case that the atoms are identical (not only chemically but also as an isotope, same mass m),
Y 1
Ztrans = . (3.176)
p
~
1 ze p~2 /4m

Next we have the vibrational part and the rotational part which we want to consider more
carefully,
!N/2
e ~!/2
Zvib = (3.177)
1 e ~!
and !N/2
1
X
l(l+1)/Irot
Zrot = (2l + 1)e (3.178)
l=0

where Irot = 2m⇤ r02 /~2 . Note that per angular momentum quantum number l there are 2l + 1
degenerate states. Since we know the behavior of Ztrans and Zvib already from the previous
sections, we address now only the rotational part. The partition function cannot be expressed
in closed form. Thus we examine only the limiting behaviors. As a characteristic energy scale
we take kB ✓rot = 2/Irot . For T ✓rot we obtain
✓Z 1 ◆N/2 ✓ Z 1 ◆N/2
d
Zrot ⇡ dl (2l + 1)e l(l+1)/Irot = Irot kB T dl e l(l+1)/Irot
0 0 dl
(3.179)
✓ ◆N/2
T
= (Irot kB T )N/2 = 2 .
✓rot
We may expand further in /Irot 14 and find
✓ ◆N/2
T 1 ✓rot
Zrot ⇡ 2 + + + ··· . (3.181)
✓rot 3 30T
14
We use the Euler-MacLaurin sum,
1
X Z 1 1
X
1 ( 1)k Bk (2k 1)
f (l) = dlf (l) + f (0) + f (0) + R1 , (3.180)
0 2 (2k)!
l=0 k=1

with Bk the Bernoulli numbers B1 = 1/6, B2 = 1/30, .... and R1 is a small correction.

61
For T ⌧ ✓rot ,
⇣ ⌘N/2
2/Irot
Zrot ⇡ 1 + 3e + ··· . (3.182)

There is a hierarchy of temperatures in this system, Tc ⌧ ✓rot ⌧ ✓vib ⌧ Tdis , where Tc is

the critical temperature for the Bose-Einstein condensation of the molecules, kB ✓vib = ~! and
kB Tdis is the molecule dissociation temperature (⇠ binding energy). We consider in the di↵erent
regimes the behavior of the heat capacity per molecule, C(T ) = dU/dT (see Fig.3.13,
8 ✓ ◆
>
> 3 ✓rot 2 ✓rot /T
>
> kB + 3kB e Tc ⌧ T ⌧ ✓rot
>
> 2 T
>
>
>
>
>
> ✓ ◆ ✓ ◆
>
>
< 3 kB ✓rot 2 ✓vib 2 ✓vib /T
2C(T ) kB + kB + + kB e ✓rot ⌧ T ⌧ ✓vib
= 2 180 T 2T (3.183)
N >
>
>
>
>
> 3
>
> kB + kB + kB ✓vib ⌧ T ⌧ Tdis
>
> 2
>
>
>
>
:
3kB Tdis ⌧ T .

2C/Nk B

7/2
3
5/2

3/2

θ rot θvib Tdis T

Figure 3.13: Schematic behavior of heat capacity of a diatomic molecule.

Note that due to the vibrational modes the heat capacity goes through maxima when molecules
are formed. Also the rotational modes are responsible for a non-monotonic behavior. Upon
lowering temperature it looses gradually in magnitude due to the quantum mechanical freezing
out of degrees of freedom.
For the hydrogen molecule H2 the energy scales are Tdis ⇠ 50000K, ✓vib ⇠ 2000K and ✓rot ⇠
85K. 15 There is no Bose-Einstein condensation for H2 , because it solidifies at 14K.

15
Note that due to the fact that the nuclear spins of the H-atom is 1/2, there is a subtlety about the degeneracies.
Actually, as identical Fermions their wave function has to be totally antisymmetric under exchange. If the two
spins form a spin singlet then the orbital part has to have even parity, i.e. l = 0, 2, 4, . . . (called ”para-hydrogen”),
while for the spin triplet configuration odd parity is required, i.e. l = 1, 3, 5, . . . (called ”ortho-hydrogen”). As
a consequence in the partition function (3.178) the summands of odd l should have a prefactor 3. This does not
a↵ect the heat capacity at temperatures T ✓rot which is the classical limit of the rotator. But there is a (minor)

62
correction at lower temperature such that
✓ ◆2
2C 3 ✓rot ✓rot /T
⇡ kB + 9kB e (3.184)
N 2 T
for Tc ⌧ T ⌧ ✓rot .

63
Chapter 4

Identical Quantum Particles -

Formalism of Second Quantization

This chapter gives an introduction to the formalism of second quantization which is a convenient
technical tool discussing many-body quantum systems. It is indispensable in quantum field
theory as well as in solid state physics. We distinguish between Fermions (half-integer spins)
and Bosons (integer spins) which behave quite di↵erently, as we have seen in the previous chapter.
This behavior is implemented in their many-body wave functions. While in the previous chapter
we could circumvent to deal with this aspect as we considered independent indistinguishable
quantum particles, it is unavoidable to implement a more careful analysis once interactions
between the particles appear.

4.1 Many-body wave functions and particle statistics

The Hamiltonian describing the dynamics of a systems of many identical quantum particles
must be invariant under exchange (permutation) of particle degrees of freedom (coordinate,
momentum, spin etc). The identical quantum particles are indistinguishable, since in quantum
mechanics it is impossible to follow the trajectories of particles under general conditions, unlike
in classical mechanics. Permutations play indeed an important role in characterising quantum
particles. We introduce the many-body wave function of N particles,

(~r1 , s1 ; ~r2 , s2 ; . . . ; ~rN , sN ) (4.1)

where each particle is labeled by the coordinate ~r and spin s. In the following we will use for
this the short-hand notation (1, . . . , N ). Analogously we define many-body operators,

~ˆ1 ; . . . ; ~rˆN , p~ˆN , S

Â(1, . . . , N ) = A(~rˆ1 , p~ˆ1 , S ~ˆN ) (4.2)

~ˆj being the operators for position, momentum and spin of particle j. Note that
with ~rˆj , p~ˆj and S
the Hamiltonian H belongs to these operators too.
We introduce the transposition (exchange) operator P̂ij which is an element of the permutation
group of N elements and exchanges the particles i and j (1  i, j  N ),

P̂ij (1, . . . , i, . . . , j, . . . , N ) = (1, . . . , j, . . . , i, . . . , N ) ,

(4.3)
P̂ij Â(1, . . . , i, . . . , j, . . . , N ) = Â(1, . . . , j, . . . , i, . . . , N ) .

Note that (P̂ij ) 1 = P̂ij . As it is invariant under particle exchange, the Hamiltonian commutes
with P̂ij ,
[H, P̂ij ] = 0 (4.4)

64
and, consequently, any combination of several transpositions, i.e. all elements of the permutation
group SN , commute with H. Hence, eigenstates of H have the property

H| i = E| i ) HP̂ij | i = P̂ij H| i = E P̂ij | i , (4.5)

where we define the wave function as

(1, . . . , N ) = h1, . . . , N | i . (4.6)

We distinguish now between Fermions and Bosons through their behavior under transpositions
P̂ij , 8
< + (1, . . . , j, . . . , i, . . . , N ) Bosons ,
(1, . . . , i, . . . , j, . . . , N ) = (4.7)
:
(1, . . . , j, . . . , i, . . . , N ) Fermions .
This means that bosonic wave functions are completely symmetric under exchange of particles,
while fermionic wave functions are completely antisymmetric1 . Note that the antisymmetric
wave functions prevents two Fermions from having the same quantum numbers. If (~ri , si ) and
(~rj , sj ) are identical, then we find

(1, . . . , i, . . . , i, . . . , N ) = (1, . . . , i, . . . , i, . . . , N ) = 0 , (4.8)

which implies the Pauli exclusion principle.

4.2 Independent, indistinguishable particles

We consider N identical particles in a potential V which are not interacting among each other.
The Hamiltonian is then given by
N
X p~ˆ2i
H= Hi with Hi = + V (~rˆi ) . (4.9)
2m
i=1

The states of each particle form an independent Hilbert space { ⌫} and we can find the stationary
states
Hi ⌫ (~ri , si ) = ✏⌫ ⌫ (~ri , si ) . (4.10)
These single-particle wave functions are renormalised, i.e.
XZ
d3 r | ⌫ (~r, s)|2 = 1 . (4.11)
s

We may now construct a many-body wave function as a product wave function with the corre-
sponding exchange property.
For Bosons we write
X
h~r1 , s1 ; . . . , ~rN , sN | B i = B (1, . . . , N ) = P̂ ⌫1 (~r1 , s1 ) · · · ⌫N (~rN , sN ) (4.12)
P̂ 2SN

and for Fermions

X
h~r1 , s1 ; . . . , ~rN , sN | Fi = F (1, . . . , N ) = sgn(P̂ )P̂ r1 , s1 ) · · ·
⌫1 (~ , ⌫N (~
rN , s N ) (4.13)
P̂ 2SN

1
Composite particles: Note that this property is valid also for composite particles. Any particle composed of
an even number of particles would be a Boson, e.g. 4 He which contains 2 protons + 2 neutrons + 2 electrons = 6
Fermions, as exchange two such particles leaves the sign of wave function unchanged. In the same way a particle
with an odd number of Fermions is a Fermions, e.g. 3 He with 2 protons + 1 neutron + 2 electrons = 5 Fermions.

65
where the operator P̂ permutes the state indices ⌫i of the wave functions and sgn(P̂ ) is the
sign of the permutation P̂ which is +1 ( 1) if P̂ is composed of an even (odd) number of
transpositions. Interestingly the fermionic wave function can be represented as a determinant,
the so-called Slater determinant,
2 3
µ1 (1) · · · µ1 (N )
6 .. .. 7
F (1, . . . , N ) = Det 4 . . 5 . (4.14)
µN (1) ··· µN (N )

Obviously the determinant vanishes if two rows or columns are identical, enforcing the Pauli
principle. These wave functions are not renormalized so that

h B| Bi = N !n⌫1 ! · · · n⌫N ! ,
(4.15)
h F| F i = N! ,

where n⌫j denotes the number of particles in the stationary single particle state labeled by ⌫j .
For Fermions it is n⌫j = 0, 1 only.

4.3 Second Quantization Formalism

It is in principle possible to investigate many-body states using many-body wave functions.
However, we will introduce here a formalism which is in many respects much more convenient
and efficient. It is based on the operators which ”create” or ”annihilate” particles and act on
states in the Fock space F which is an extended space of states combining Hilbert space Qn of
di↵erent particle numbers n,
M1
F= Qn . (4.16)
n=0
Note that the name ”second quantization” does not imply a new quantum mechanics.
We can express a many-body state of independent particles in the particle occupation number
representations,
|n⌫1 , n⌫2 , . . .i (4.17)
which is a state in F whose particle number is given by N = n⌫1 + n⌫2 + · · · .

4.3.1 Creation- and annihilation operators

We define operators â⌫ and â†⌫ which connect Hilbertspaces of di↵erent particle number,

â⌫ : Qn ! Qn 1 and â†⌫ : Qn ! Qn+1 . (4.18)

The first we call annihilation and the second creation operator whose action is best understood
in the particle number or occupation representation.

Bosons: Let us first consider Bosons which, for simplicity, do not possess a spin. The two
operators have the following property,
p
â⌫ |n⌫1 , n⌫2 , . . . , n⌫ , . . .i = n⌫ |n⌫1 , n⌫2 , . . . , n⌫ 1, . . .i ,
(4.19)
p
â†⌫ |n⌫1 , n⌫2 , . . . , n⌫ , . . .i = n⌫ + 1|n⌫1 , n⌫2 , . . . , n⌫ + 1, . . .i ,

and p
hn⌫1 , n⌫2 , . . . , n⌫ , . . . |â†⌫ = n⌫ hn⌫1 , n⌫2 , . . . , n⌫ 1, . . . | ,
(4.20)
p
hn⌫1 , n⌫2 , . . . , n⌫ , . . . |â⌫ = n⌫ + 1hn⌫1 , n⌫2 , . . . , n⌫ + 1, . . . | .

66
It is obvious that

â⌫ |n⌫1 , n⌫2 , . . . , n⌫ = 0, . . .i = 0 and hn⌫1 , n⌫2 , . . . , n⌫ = 0, . . . |â†⌫ = 0 . (4.21)

The operators satisfy the following commutation relations,

[â⌫ , â†⌫ 0 ] = ⌫⌫ 0 and [â⌫ , â⌫ 0 ] = [â†⌫ , â†⌫ 0 ] = 0 . (4.22)

Note that these relations correspond to those of the lowering and raising operators of a harmonic
oscillator. Indeed we have seen previously that the excitation spectrum of a harmonic oscillator
obeys bosonic statistics.
The creation operators can also be used to construct a state from the vacuum, denoted as
|0i, where there are no particles, such that â⌫ |0i = 0. A general state in occupation number
representation can be written as,

· · · (â†⌫ )n⌫ · · · (â†⌫2 )n⌫2 (â†⌫1 )n⌫1

|n⌫1 , n⌫2 , . . . , n⌫ , . . .i = p |0i . (4.23)
n⌫1 !n⌫2 ! · · ·
The number operator is defined as

n̂⌫ = â†⌫ â⌫ with n̂⌫ |n⌫1 , n⌫2 , . . . , n⌫ , . . .i = n⌫ |n⌫1 , n⌫2 , . . . , n⌫ , . . .i (4.24)

and the total number of particles is obtained through the operator

X
Nb= n̂⌫i . (4.25)
i

Knowing the spectrum of the Hamiltonian of independent particles as given in Eq.(4.10) we may
express the Hamiltonian as X X
H= ✏⌫ â†⌫ â⌫ = ✏⌫ n̂⌫ . (4.26)
⌫ ⌫

Fermions: Now we turn to Fermions with spin 1/2 (half-integer spin). Again the single-particle
state shall be labelled by ⌫ including the spin index for " and #. Analogously to the case of
Bosons we introduce operators â†⌫ and â⌫ which obey anti-commutation rules,

{â⌫ , â†⌫ 0 } = ⌫⌫ 0 and {â⌫ , â⌫ 0 } = {â†⌫ , â†⌫ 0 } = 0 , (4.27)

where {. . .} is defined as {Â, B̂} = ÂB̂ + B̂ Â. In particular this implies that

â†⌫ â†⌫ = 0 and â⌫ â⌫ = 0 (4.28)

such that n⌫ = 0, 1, i.e. each single-particle state labelled by ⌫ can be occupied by at most one
particle, because
â†⌫ | . . . , n⌫ = 1, . . .i = â†⌫ â†⌫ | . . . , n⌫ = 0, . . .i = 0 . (4.29)
A general state may be written as

|n⌫1 , n⌫2 , . . . , n⌫ , . . .i = · · · (â†⌫ )n⌫ · · · (â†⌫2 )n⌫2 (â†⌫1 )n⌫1 |0i (4.30)

which restricts n⌫ to 0 or 1. The order of the creation operators plays an important role as the
exchange of two operators yields a minus sign. We consider an example here,

|n1 , n2 , n3 , n4 i = |1, 1, 1, 1i = â†4 â†3 â†2 â†1 |0i . (4.31)

Removing now one particle yields

â2 |1, 1, 1, 1i = â2 [â†4 â†3 â†2 â†1 ] |0i = â†4 â†3 â2 â†2 â†1 |0i = â†4 â†3 (1 â†2 â2 )â†1 |0i
(4.32)
= â†4 â†3 â†1 |0i = |1, 0, 1, 1i

67
and now analogously

â3 |1, 1, 1, 1i = â3 [â†4 â†3 â†2 â†1 ] |0i = â†4 â3 â†3 â†2 â†1 |0i = â†4 (1 â†3 â3 )â†2 â†1 |0i
(4.33)
= â†4 â†2 â†1 |0i = |1, 1, 0, 1i .

Obviously, the order of the operators is important and should not be ignored when dealing with
Fermions.

4.3.2 Field operators

We consider now independent free particles whose states are characterized by momentum p~ = ~~k
and spin s with an energy ✏~k = ~2~k 2 /2m. The wave function has a plane wave shape,

1 ~ ~k = 2⇡ (nx , ny , nz ) ,
~k = p eik·~r with (4.34)
V L

where we used periodic boundary conditions in a cube of edge length L (volume V = L3 ). On

this basis we write field operators
X ~ X
b s (~r) = p1 eik·~r â~ks and b †s (~r) = p1 e i~k·~
r †
â~ (4.35)
V ~ V ~ ks
k k

and the inverse,

Z ~ Z ~
eik·~r e ik·~r
â~† = d3 r p b † (~r) und â~k = d3 r p b (~r) (4.36)
k V V

Also these operators b s (~r) and b †s (~r) act as annihilation or creation operators, respectively, in
the sense,
b s (~r)† |0i = |~r, si and r) = h~r, s| i = h0| b s (~r)| i .
s (~ (4.37)
Moreover we have the condition
b s (~r)|0i = 0 and h0| b †s (~r) = 0 . (4.38)

The field operators also satisfy (anti-)commuation relations,

X ~
b s (~r) b †0 (~r 0 ) ⌥ b †0 (~r 0 ) b s (~r) = 1 eik·~r i~k 0 ·~
r0
(â~ks â~† 0 0 ⌥ â~† 0 0 â~ks ) = (~r ~r 0 ) ss0 (4.39)
s s V k s
~k,~k 0 | {z k s }
= ~k~k 0 ss0

and analogously
b s (~r) b s0 (~r 0 ) ⌥ b s0 (~r 0 ) b s (~r) = 0 and b †s (~r) b †0 (~r 0 ) ⌥ b †0 (~r 0 ) b †s (~r) = 0 (4.40)
s s

for Bosons ( ) and Fermions (+). Taking these relations it becomes also clear that

h~r 0 , s0 |~r, si = h0| b s0 (~r 0 ) b s (~r)† |0i

(4.41)
= h0| (~r ~r 0 ) ss0 |0i ⌥ h0| b s (~r)† b s0 (~r 0 )|0i = (~r ~r 0 ) ss0 .
| {z }
=0

Applying a field-operator to a N -particle state yields,

p
b † (~r)|~r1 , s1 ; . . . ; ~rN , sN i = N + 1|~r1 , s1 ; . . . ; ~rN , sN ; ~r, si , (4.42)
s

68
such that
1
|~r1 , s1 ; ~r2 , s2 ; . . . ; ~rN , sN i = p b †sN (~rN ) · · · b †s1 (~r1 )|0i . (4.43)
N!
Note that particle statistics leads to the following relation under particle exchange,

|~r1 , s1 ; ~r2 , s2 ; . . . ; ~rN , sN i = ±|~r2 , s2 ; ~r1 , s1 ; . . . ; ~rN , sN i (4.44)

where + is for Bosons and is for Fermions. The renormalisation of the real space states have
to be understood within the projection to occupation number states, yielding many-body wave
functions analogous to those introduced Eqs.(4.12, 4.13),

(1, . . . , N ) = h~r1 , s1 ; . . . , ~rN , sN |n~k1 ,s0 , n~k2 ,s0 , . . .i . (4.45)

1 2
P
Note that N = ~k,s n~k,s . Taking care of the symmetry / antisymmetry of the many-body wave
function we recover the renormalization behavior in Eqs.(4.42, 4.43).

4.4 Observables in second quantization

It is possible to express Hermitian operators in the second quantization language. We will show
this explicitly for the density operator by calculating matrix elements. The particle density
operator is given by
N
X
⇢b(~r) = (~r b~ri ) . (4.46)
i=1

Now we take two states | i, | 0i

2 QN with the fixed particle number N and examine the matrix
element
Z X
0
h |b⇢(~r)| i = d3 r1 · · · d3 rN h 0 |~r1 , . . . , ~rN ih~r1 , . . . , ~rN | (~r b
~ri )| i
i
Z X
= d 3 r1 · · · d 3 rN (~r ~ri )h 0 |~r1 , . . . , ~rN ih~r1 , . . . , ~rN | i (4.47)
i
Z
0
=N d 3 r 1 · · · d 3 rN 1h |~r1 , . . . , ~rN rih~r1 , . . . , ~rN 1 , ~r|
1, ~ i,

where we suppress spin indices for the time being. Here we used in the last equality that we
can relabel the coordinate variables and permute the particles. Since we have the product of
two states under the same perturbation, Fermion sign changes do not appear and N identical
integrals follow. We claim now that the density operator can also be written as

⇢b(~r) = b † (~r) b (~r), (4.48)

which leads to
Z
h 0 | b † (~
r) b (~r)| i = d 3 r1 · · · d 3 rN 0
| b † (~r)|~r1 , . . . , ~rN b (~r)| i
1h 1 ih~
r1 , . . . , ~rN 1 |

Z (4.49)
0
=N d 3 r1 · · · d 3 rN 1h |~r1 , . . . , ~rN rih~r1 , . . . , ~rN 1 , ~r|
1, ~ i

which is obviously identical to Eq.(4.47).

According to Eq.(4.26) the kinetic energy can be expressed as
X ~2~k 2 X ~2~k 2
Hkin = â~† â~k = n̂~k (4.50)
2m k 2m
~k ~k

69
which, using Eq.(7.19), may also be expressed in field operator language as
Z ⇣ ⌘⇣ ⌘
1 X ~ i~k·~r ~r~ 0 e i~k·~r 0 b † (~r) b (~r 0 )
Hkin = d3 rd3 r0 ~re
2mV
~k
(4.51)
Z
~2 ~ b † (~r)) · (r
~ b (~r)).
= d3 r ( r
2m
Note the formal similarity
R 3 with the expectation value of the kinetic energy using single-particle
~2 ~ ⇤ (~r)· r'(~
~ r) . In an analogous way we represent the potential energy,
wave functions, 2m d r r'
Z Z
b b † b
Hpot = d rU (~r) (~r) (~r) = d3 rU (~r)b
3
⇢(~r) . (4.52)

Beside the particle density operator ⇢ˆ(~r) also the current density operators can be expressed by
field operators,
b~ ~ ⇣b† ~ b (~r)) (r
~ b † (~r)) b (~r)
⌘
J(~r) = (~r)(r (4.53)
2mi
and the spin density operator for spin-1/2 Fermions (writing spin indices again),
X
~ r) = ~
b
S(~ b † (~r)~ ss0 b s0 (~r), (4.54)
2 0 s
ss

where ~ ss0 are the Pauli matrices. In momentum space the operators read,
Z X †
⇢bq~ = d3 re i~q·~r ⇢b(~r) = â~ â~k+~q,s (4.55)
k,s
~k,s

b
~ q~ = ~ X †
S â~ ~ ss0 â~k+~q,s0 (4.56)
2 k,s
~k,s,s0
✓ ◆
b ~ X ~ ~q
J~q~ = k+ â~† â~k+~q,s . (4.57)
m 2 k,s
~k,s

Finally we turn to the genuine many-body feature of particle-particle interaction,

Z
1X 1 XX
b
Hint = d3 r d3 r0 b †s (~r) b †s0 (~r 0 )V (~r ~r 0 ) b s0 (~r 0 ) b s (~r) = Vq~ â~† â~† 0 â 0 0 â ,
2 0 2V k+~
q ,s k q~,s0 ~k ,s ~k,s
s,s 0~k,~k 0 q~ s,s
(4.58)
where the factor 1/2 corrects for double counting and
1 X
V (~r) = Vq~ ei~q·~r . (4.59)
V
q~

Note that the momentum space representation has the simple straightforward interpretation that
two particles with momentum ~~k and ~~k 0 are scattered into states with momentum ~(~k + ~q )
and ~(~k 0 ~q ), respectively, by transferring the momentum ~~q .

4.5 Equation of motion

For simplicity we discuss here a system of independent free quantum particles described by the
Hamiltonian X
H= ✏~k â~† â~k (4.60)
k
~k

70
where we suppress the spin index. We turn now to the Heisenberg representation of time
dependent operators,
â~k (t) = eiHt/~ â~k e iHt/~ . (4.61)
Thus, we formulate the equation of motion for this operator,
d X
i~ â~ = [H, â~k ] = ✏~k 0 [â~† 0 â~k 0 , â~k ]
dt k k
~k 0

8 †
>
> â~ 0 [â~k 0 , â~k ] + [â~† 0 , â~k ] â~k 0 for Bosons
>
> k | {z }
| k{z }
>
>
>
< =0 = ~k,~k 0
X
= ✏~k 0 (4.62)
>
>
~k 0 >
>
> â~† 0 {â~k 0 , â~k } {â~† 0 , â~k } â~k 0 for Fermions
> k | {z } k
>
: | {z }
=0 = ~ k0
k,~

X
= ✏~k 0 â~k 0 ~k,~k 0 = ✏~k â~k ,
~k 0

and analogously
d †
â = [H, â~† ] = ✏~k â~† .
i~ (4.63)
dt ~k k k

A further important relation in the context of statistical physics is

H † H ✏~k †
e â~ e =e â~ . (4.64)
k k

b = P~ ↠â~ ,
Analogously we find for the number operator N k ~ k k

b † b
e µN
â~ e µN
= e µ â~† . (4.65)
k k

Both relations are easily proven by examining the action of this operator on a eigenstate of the
Hamiltonian | i = |n~k1 , . . . , n~k , . . .i,

H ↠e H | Ee H ↠|
p Ee H |n , . . . , n
e ~k i =e ~k i= n~k + 1e ~k1 ~k + 1, . . .i
(4.66)
p (E E 0 ) |n , . . . , n (E E 0 ) ↠|
= n~k + 1e ~k1 ~k + 1, . . .i = e ~k i
P P
where E = ~k 0 ✏~k 0 n~k 0 and E 0 = ~k 0 ✏~k 0 n~k 0 +✏~k such that E E 0 = ✏~k . Note that for Fermions
the operation of â~† on | i is only finite, if n~k = 0 otherwise we have a zero. Still the relation
k
remains true for both types of quantum particles. The analogous proof applies to Eq.(4.65).

Fermi-Dirac and Bose-Einstein distribution: Let us look at the thermal average,

tr{e H0 ↠â }
~k ~k
hn̂~k i = hâ~† â~k i = H 0 , (4.67)
k tre

where we use the Hamiltonian H0 = H b . We can rearrange the numerator of (4.67) using
µN
Eqs.(4.64) and (4.65),

tr{e H0 ↠â } = tr{e H0 ↠e H0 e H0 â } =e (✏~k µ)

tr{â~† e H0 â }
~k ~k ~k ~k k ~k
(4.68)
=e (✏~k µ)
tr{e H0 â ↠} =e (✏~k µ)
tr{e H0 [1 ± â~† â~k ]} ,
~k ~k k

71
where ’+’ and ’ ’ stand for Bosons and Fermions, respectively. Inserting this into Eq.(4.67) we
find,
8 1
>
> e (✏~k µ) 1 Bosons
>
<
hn̂~k i = e (✏~k µ) (1 ± hn̂~k i) ) hn̂~k i = (4.69)
>
> 1
>
: Fermions
e (✏~k µ) + 1
which corresponds to the standard Bose-Einstein and Fermi-Dirac distribution.

4.6 Correlation functions

Independent classical particles do not have any correlation among each other. This is di↵erent
for quantum particles. The second quantization language is very suitable for the formulation of
correlation functions and to show that Fermion and bose gases behave rather di↵erently.

4.6.1 Fermions
First let us write the ground state of a free Fermi gas of spin-1/2 Fermions as we have introduced
it already in Sect.3.6.2. Starting from the vacuum |0i we fill successively all low lying states
with a Fermion of both spins s until all Fermions are placed. This defines the Fermi sphere in
k-space with the radius kF , the Fermi wave vector. The ground state is then,
|~k|kF
Y Y
| 0i = â~† |0i (4.70)
ks
~k s=",#
P
and n~k = h 0 |n̂~k | 0 i = ⇥(kF |~k|) is a step function with n̂~k = s â~† â~ks .
ks
First we formulate the one-particle correlation function in real space using field operators,2
n
gs (~r ~r 0 ) = h b †s (~r) b s (~r 0 )i (4.72)
2
which measure the probability amplitude to be able to insert a Fermion at place ~r after having
removed one at ~r 0 with the same spin s. We evaluate this expression by going to k-space,
n 1 X i~k·~r+i~k 0 ·~r 0
gs (~r ~r 0 ) = e hba~† b
a 0 i . (4.73)
2 V ks ~k s
~k,~k 0 | {z }
= hn̂~ks i ~k,~k 0

At T = 0 we obtain
Z Z kF Z +1
n 0 d3 k i~k·(~ r 0)
r ~ 1 2 r 0 | cos ✓
gs (~r ~r ) = 3
e = dk k d cos ✓ eik|~r ~
2 |~k|kF (2⇡) (2⇡)2 0 1
(4.74)
Z kF
1 3n sin x x cos x
= dk k sin(k|~r ~r 0 |) = .
2⇡ 2 |~r ~r 0 | 0 2 x3 x=kF |~ r 0|
r ~

Note the limits: gs (~r ! 0) = 1 and gs (~r ! 1) = 0 where gs (~r ~r 0 ) corresponds to the overlap
of the two states r r
2b 2b
r)| 0 i and
s (~ r 0 )|
s (~ 0i . (4.75)
n n
2
The correlation function at zero-temperature is given by the expectation value for the ground state,
n
gs (~r ~r 0 ) = h 0 | b †s (~r) b s (~r 0 )| 0 i , (4.71)
2
which is just the zero-temperature limit of Eq.(4.72).

72
Analogous results can be calculated for finite temperatures (here T TF ), where for the
”classical” limit an analytical result can be found based on the Maxwell-Boltzmann distribution:
n (2⇡)3 k2 /A2 2mkB T 4⇡
hn̂~ks i = p e with A2 = = 2 , (4.76)
2 ( ⇡A)3 ~ 2

leading to
Z
n n i~k·(~ r 0) k2 /A2 n A2 (~ r 0 )2 /4 n r 0 )2 / 2
gs (~r ~r 0 ) = d3 k e r ~
e = e r ~
= e ⇡(~
r ~
(4.77)
2 2⇡ 3/2 A3 2 2

Next we turn to the pair correlation function which we define as

⇣ n ⌘2
gss0 (~r ~r 0 ) = h b †s (~r) b †s0 (~r 0 ) b s0 (~r 0 ) b s (~r)i (4.78)
2
being the probability to be able to pick two Fermions at the di↵erent places, ~r and ~r 0 , with the
spins s and s0 , respectively. Again we switch to the more convenient k-space,
⇣ n ⌘2 1 X ~ ~0 0 0
gss0 (~r ~r 0 ) = 2 e i(k k )·~r e i(~q q~ )·~r h b a~† b a† 0 b a 0 i.
a 0 0b (4.79)
2 V ks q~s q~ s ~k s
~k,~k 0 ,~ q0
q ,~

In order to evaluate the mean value h ba~† b

a† 0 b a 0 i we use the same technique as presented
a 0 0b
ks q~s q~ s ~k s
in Sect.4.5. 3

From this it follows straightforwardly for s = s0 ,

⇣ n ⌘2 1 X⇣ ⌘ ⇣ n ⌘2
~ 0
gss (~r ~r 0 ) = 2 1 e i(k q~)·(~r ~r ) hb
n~ks ihb
nq~s i = [1 gs (~r ~r 0 )2 ], (4.84)
2 V 2
~k,~
q

and we can write,

8
>
> 9(sin x x cos x)2
>
< 1 T =0,
0 x6 x=kF |~ r0 |
r ~
gss (~r ~r ) = (4.85)
>
>
>
: 1 2⇡(~ r 0 )2 /
r ~ 2
e T TF .
3
Evaluation of h b a~†ks b
a†q~s0 b a~k 0 s i: We separate the task into two cases: (1) s = s0 and (2) s 6= s0 .
aq~ 0 s0 b
0
(1) s = s :
1 h 0
i 1 h 0 0 0
i
a~†ks b
hb a†q~s b a~k 0 s i = tr e H b
aq~ 0 s b a~†ks b
a†q~s b a~k 0 s = tr e H b
aq~ 0 s b a~†ks e H e H ba†q~s b
aq~ 0 s b
a~k 0 s
Z Z

e (✏~ µ) h i e (✏~ µ) h i
k H0 † k H0 †
= tr e b
aq~s b
aq~ 0 s b a~†ks
a~k 0 s b = tr e b
aq~s b
aq~ 0 s ( ~k,~k0 a~†ks b
b a~k 0 s )
Z Z

e (✏~ µ) h i e (✏~ µ) h i
k H0 † k H0 †
= tr e b
aq~s b
aq~ 0 s ~
k,~
k0 tr e b
aq~s b a~†ks b
aq~ 0 s b a~k 0 s )
Z Z

e (✏~ µ) h i
k H0 †
=e (✏~
k
µ)
hb
nq~,s i q q0 ~
~,~ k,~
k0 tr e b
aq~s ( q~0 ,~k a~†ks b
b aq~ 0 s )b
a~k 0 s )
Z

=e (✏~
k
µ)
hb
nq~,s i( q q0 ~
~,~ k,~
k0 ~0 ,~
q k0 )
~,~
k q +e (✏~
k
µ)
a~†ks b
hb a†q~s b a~k 0 s i .
aq~ 0 s b
(4.80)
From this we find
1
a~†ks b
hb a†q~s b a~k 0 s i =
aq~ 0 s b (✏~ µ)
hb
nq~,s i( q q0 ~
~,~ k,~
k0 ~0 ,~
q k0 )
~,~
k q = hb
n~k,s ihb
nq~,s i( q q0 ~
~,~ k,~
k0 ~0 ,~
q k0 )
~,~
k q . (4.81)
e k +1
(2) s 6= s0 : n o
a~†ks b
hb a†q~s0 b a~k 0 s i = e
aq~ 0 s0 b (✏~
k
µ)
hb
nq~,s i q q0 ~
~,~ k,~
k0 a~†ks b
+hb a†q~s0 b a~k 0 s i
aq~ 0 s0 b (4.82)
analogous to case (1), such that
a~†ks b
hb a†q~s0 b a~k 0 s i = hb
aq~ 0 s0 b n~k,s ihb
nq~,s i q q0 ~
~,~ k,~
k0 . (4.83)

73
 gss (r) (a)$ gss gss (r) (b)$
1.0$
1.0
1.0
0.8$
0.8
0.8
1.0 .
0.6$
0.6
0.6
0.4$
0.4 0.4 0.99

0.2$ 0.2
0.2
4 6 8 10 12
0.0$0.0 0.0
0.0$
0.5 1.0
1.0$
1.5 2.0
2.0$ r/ 0 2 4 6 8 kF r

Figure 4.1: The equal-spin pair correlation function for Fermions: (a) T TF and (b) T = 0.

Fig.4.1 shows the equal-spin correlation function which goes to zero for ~r ! 0, since Fermions
avoid eachother. This dip in the correlation is called exchange hole and has a radius of order
(thermal wave length) for T TF and kF 1 in the groundstate. Note that in the latter case the
correlation function shows a weak oscillator with wavevector kF .
The case of s 6= s0 leads to gss0 (~r) = 1, i.e. there is no correlation between s = 1/2-Fermions of
opposite spin. The probability to find another Fermion around the position of a Fermion at ~r
corresponds to
1
g(~r) = [g"" (~r) + g"# (~r)]. (4.86)
2
The density depletion around such a Fermion is then,
Z Z
R 3 n 3 2 2 1 X ~ ~0
n d r (g(~r) 1) = d r {gs (~r)} = d3 r 2 hb n~k 0 s iei(k k )·~r
n~ks ihb
2 n V
~k,~k 0

8
> 1 T =0, (4.87)
>
<
2 X
= n~ks i2 =
hb 3n
nV >
>
~k : T TF .
25/2
which means that the exchange hole expels one Fermion such that each Fermion ”defends” a
given volume against other Fermions of the same spin for T = 0, while the exchange hole shrinks
like n3 for T TF .

4.6.2 Bosons
Analogous to the Fermions we consider first the single-particle correlation function for s = 0
Bosons,
1 X i~k·~
r+i~k 0 ·~
r0 1 X i~k·(~ r 0)
g1 (~r ~r 0 ) = h b † (~r) b (~r 0 ) i = e a~† b
hb a~k 0 i = hb
n~k ie r ~
, (4.88)
V k V
~k,~k 0 ~k

which in the limit ~r 0 ! ~r approaches the constant density n and vanishes at very large distances.
For T = 0 we consider the groundstate, the Bose-Einstein condenstate, hb n~k i = N ~k,0 and for
T Tc we use the classical distribution (Eq.(4.76)) where Tc is the critical temperature for
Bose-Einstein condensation.
8
< n T =0,
0
g1 (~r ~r ) = (4.89)
: ⇡(~
r ~
r 0 )/ 2
ne T Tc .

74
The pair correlation functions reads,

g2 (~r ~r 0 ) = h b † (~r) b † (~r 0 ) b (~r 0 ) b (~r) i

1 X (4.90)
i(~k ~k 0 )·~ q q~ 0 )·~
r0
= e r i(~
a~† b
hb a†q~b a~k 0 i ,
aq~ 0 b
V2 k
~k,~k 0 ,~ q0
q ,~

Analogous to the Fermions we evaluate4

n o
a~† b
hb a†q~b a~k 0 i = (1
aq~ 0 b ~k~
q
) ~k~k 0 q~q~ 0 + ~k~
q 0 q~~k 0
hb
n~k ihb
nq~i + ~k~
q ~k~k 0 q~q~
n~2k i
0 (hb hb
n~k i) (4.92)
k

This leads to
2 3
1 4 X ⇣ ⌘ X
i(~k q~)·(~ r 0)
g2 (~r ~r 0 ) = (1 ~k~
q
) 1+e r ~
hb
n~k ihb
nq~i + n~2k i
(hb n~k i)5
hb
V2
~k,~
q ~k

2 23
1 4 X Xn o X
i~k·(~ r 0)
= hb
n~k ihb
nq~i + n~2k i
hb n~k i2
hb hb
n~k i(hb
n~k i + 1) + e r ~
n~k i 5
hb
V2
~k,~
q ~k ~k

2 0 2 1 Xn 2 2
o
= n + g1 (~r ~r ) + 2 hb
n~k i hb
n~k i hb
n~k i(hb
n~k i + 1)
V
~k
(4.93)
For T = 0 with hb
n~k i = N ~k,0 n~2 i = N 2
(hb ~k,0 ) we obtain
k

1 N (N 1)
g2 (~r ~r 0 ) = 2n2 2
N (N + 1) = , (4.94)
V V2
so no correlation is observed. The probability to pick the first particle is n = N/V and a second
one (N 1)/V (⇡ n for large N ). For the high-temperature limit, T Tc , the correlation
function is given
⇣ 0 2 2
⌘
g2 (~r ~r 0 ) = n2 + g1 (~r ~r0 )2 = n2 1 + e 2⇡(~r ~r ) / . (4.95)

The probability of finding two Bosons at the same position is twice as large as for long distances,
see Fig. 4.2 . Thus, in contrast to Fermions (Fig.4.1), Bosons like to cluster together.
The radius of bunching of the Bosons in the limit T Tc is of order and shrinks with increasing
T (classical limit).

4.7 Selected applications

We consider here three examples applying second quantization to statistical physics systems.

4.7.1 Spin susceptibility

We calculate the spin susceptibility of spin-1/2 Fermions using the fluctuation-dissipation rela-
tion.
1 n 2 o
= hM̂z i hM̂z i2 , (4.96)
V kB T
4
a~†k b
Special case of h b a†q~b a~k 0 i for
aq~ 0 b ~
k~q ~
k~k0 q
~q~0:

a~†k b
hb a~†k b
a~k b a~†k {b
a~k i = h b a~†k
a~k b a~†k b
a~k i = h b
1}b a~†k b
a~k b a~k i a~†k b
hb b~2k i
a~k i = h n hb
n~k i (4.91)

75
 g2"
2.0

1.5

1.0

0.5

0.0
0 2 4
r

Figure 4.2:p The pair correlation function for Bosons. g2 (r) is given in units of n 2 and r in
units of / 2⇡.

where Z
gµB XX X
M̂z = d3 r Sbz (~r) = µB â~† z
0 â 0 = µB sâ~† â~ks (4.97)
~ ks ss ~ks ks
~k ss0 ~k,s

using Sect.4.4. Moreover, g = 2 and s = ±1. First we calculate the magnetization in zero
magnetic field, X X
hM̂z i = µB shâ~† â~ks i = µB sn~k = 0 . (4.98)
ks
~k,s ~k,s

Now we turn to XX
hM̂z2 i = µ2B ss0 hâ~† â~ks â~† 0 0 â~k 0 s0 i , (4.99)
ks k s
~k,s ~k 0 ,s0

which we determine like in Sect.4.6.1,

1 n H0 †
o e (✏~k µ) n
H0
o
hâ~† â~ks â~† 0 0 â~k 0 s0 i = tr e â~ â~ks â~† 0 0 â~k 0 s0 = tr e â~ks â~† 0 0 â~k 0 s0 â~†
ks k s Z ks k s Z k s ks

e (✏~k µ) h n o n oi
H0 H0
= tr e â~ks â~† 0 ~k~k 0 ss0 + tr e â~ks â~† â~† 0 0 â~k 0 s0
Z k s0 ks k s
n⇣ ⌘ o
=e (✏~k µ)
1 hâ~† â~ks i ~k~k 0 ss0 + hâ~† 0 0 â~k 0 s0 i e (✏~k µ)
hâ~† â~ks â~† 0 0 â~k 0 s0 i
ks k s ks k s
(4.100)
which leads straightforwardly to
⇣ ⌘
hâ~† â~ks â~† 0 0 â~k 0 s0 i = hâ~† â~ks i 1 hâ~† â~ks i ~k~k 0 ss0 + hâ~† â~ks ihâ~† 0 0 â~k 0 s0 i . (4.101)
ks k s ks ks ks k s

We now insert this result into Eq.(4.99) and obtain

X X 1
hM̂z2 i = 2µ2B hb
n~ks i(1 n~ks i) = 2µ2B
hb , (4.102)
~k ~k
4 cosh2 ( (✏~k µ)/2)

where the second term cancels due to the spin summation. In the low-temperature limit this is
confined to a narrow region (⇠ kB T ) around the Fermi energy, such that we approximate
Z +1
2 2 1
hM̂z i ⇡ µB V d✏N (✏F ) 2 = V µ2B kB T N (✏F ) , (4.103)
1 4 cosh ( ✏/2)

76
where the density of states is defined as
1 X
N (✏) = (✏ ✏~k ) . (4.104)
V
~k,s

Then the spin susceptibility is given as the Pauli susceptibility,

3n
= µ2B N (✏F ) = µ2B , (4.105)
2✏F
where the expression with the density of states at ✏F is general and the second equality is valid for
free Fermions. The Pauli susceptibility is independent of temperature, because only N (✏F )kB T
Fermions can be spin polarised (thermal softening of the Fermi sea). Thus, the factor (kB T ) 1
is compensated by the shrinking density of polarisable spins as temperature decreases.5

4.7.2 Bose-Einstein condensate and coherent states

Our aim here is to characterise the Bose-Einstein condensate further beyond what we did in
the last chapter. Here, we consider the concepts of both the o↵-diagonal long-range order and
the order parameter for the condensate. We start with the discussion of the single-particle
correlation function for a homogeneous gas of spin-0 Bosons in more detail than in Sect.4.6.2,
1 X † ~0 0 ~ 1 X ~ 0
g(~r ~r 0 ) = h b † (~r) b (~r 0 )i = hâ~ â~k 0 iei(k ·~r k·~r) = hn̂~k ie ik·(~r ~r ) , (4.109)
V k V
~k,~k 0 ~k

where hn̂~k i is the Bose-Einstein distribution. For independent free Bosons we may write

1 X e i~k·R
~
~ =
g(R) , (4.110)
V (✏ µ)
~
e ~k 1
k

with ✏~k = ~2~k 2 /2m and R

~ = ~r ~ ! 0 and R
~r 0 . Let us look at the two limits R ~ ! 1. For the
first limit we may expand
1 X 1 i~k · R
~ (~k · R)
~ 2 /2 + · · · ~2
R
~ =
g(R) (✏~k µ)
=n h~k 2 i + · · · , (4.111)
V e 1 6
~k

where n = N/V is the particle density6 and

8
2m U < 6⇡n 2 T Tc
~ 2
hk i = 2 = (4.113)
~ V : 2 (T /T
3.08⇡n c )3/2 T < Tc
5
Classical limit: The classical limit can be discussed using the Maxwell-Boltzmann distribution function,
n 3 k2 2 /4⇡
n~ks =
e , (4.106)
2
with as the thermal wavelength. Inserting into Eq.(4.102) we obtain
Z Z ⇢ 3 ✓ ◆
d3 k d3 k n 2 2 n2 6 k2 2 /2⇡
3
n
hM̂z2 i = 2µ2B V 3
n~ks (1 n~ks ) = 2µ 2
B V 3
e k /4⇡ e = µ2B V n 1 ,
(2⇡) (2⇡) 2 4 25/2
(4.107)
which using Eq.(4.96) leads to the susceptibility found earlier in Eq.(3.112).
✓ ◆
µ2 n 3
n
= B 1 (4.108)
kB T 25/2
The factor 1 n~ks in Eq.(4.102) introduces the quantum correction in the second term.
6
Average for an isotropic momentum distribution function hb
n~k i:
X h~k2 i h~k2 i ~ 2
~ · ~k)2 i =
h(R Rj2 hkj2 i = (Rx2 + Ry2 + Rz2 ) = R , (4.112)
j=x,y,z
3 3

because hkx2 i = hky2 i = hkz2 i = h~k2 i/3 and hkx ky i = hky kz i = hkz kx i = 0.

77
where U is the internal energy of the Bose gas. The correlation falls o↵ quadratically for finite,
but small R~ (|R|
~ ⌧ ). Note that the in the T ! 0 limit the correlation function does not
drop in agreement with Eq.(4.89). For the long-distance limit we note that only the small wave
vectors contribute to the ~k-sum so that we may expand the integrand in the following way
( (✏~k µ) ⌧ 1),
Z ~ ~ Z ~ ~
~ ⇡ d3 k e ik·R 2mkB T d3 k e ik·R
g(R) = (4.114)
(2⇡)3 (✏~k µ) ~2 (2⇡)3 ~k 2 + k02
2mµ
where k02 = ~2
> 0. This form we know from the Yukawa potential,
~ ~
~ ⇡ mkB T e k0 |R| e k0 |R|
g(R) = . (4.115)
(2⇡)4 ~2 |R|
~ (2⇡)3 2 |R|~

The single-particle correlation function decays exponentially for large distances (Fig.4.3). This
behavior is valid for T > Tc where µ < 0.

~
g(R)
n
T < Tc

n0
T > Tc

~
R

Figure 4.3: Schematic behavior of the single-particle correlation function in the normal(T > Tc )
and the Bose-Einstein condensed phase (T < Tc ). n is the overall particle density and n0 the
density of condensed particles.

For T < Tc the chemical potential lies at the lowest single-particle state, i.e. µ = 0 for free
Bosons, such that k0 = 0. The short-distance behavior is still described by Eq.(4.111). For
the long-distance behavior we conclude from Eq.(4.115) that the correlation function goes to
~ 1 . However, this is not true, since our integral approach neglects the macroscopic
zero like |R|
occupation of the ~k = 0 state. Thus, we should use
1 1
hn̂~k i = n0 (~k) + , (4.116)
(2⇡)3 e (✏~k µ)
1
~ ! 1,
such that for |R|
~ = n0 + 1
g(R) . (4.117)
(2⇡)3 ~
2 |R|

The correlation function approaches a finite value on long distances in the presence of a Bose-
Einstein condensate (Fig.4.3).

Bogolyubov approximation:
We consider this now from the view point of the field operator for free Bosons,
X 1 X
b (~r) = p1 ~ â0
â~k eik·~r = p + p
~
â~k eik·~r . (4.118)
V ~ V V ~
k k6=0

78
 p
The correlation function in Eq.(4.117) suggests the following approximation: â0 ! a0 = N0 .
For a Bose-Einstein condensate we may replace the operator â0 simply by a complex number,
such that
b (~r) ! b (~r) = 0 (~r) + b (~r) , (4.119)
p i
with 0 (~r) = n0 e , where is an arbitrary phase and n0 = N0 /V . In a uniform system
this phase does not a↵ect the physical properties. This so-called Bogolyubov approximation is,
of course, incompatible with the occupation number representation. On the other hand, it is
possible for a condensate state whose particle number is not fixed. Indeed a state incorporating
this property is a coherent state.

Coherent state:
We introduce a coherent state as an eigenstate of the annihilation operator â⌫ of a Bosonic state
of energy ✏⌫ . Let us call this state | ↵ i with the property,

â⌫ | ↵i = ↵| ↵i , (4.120)

where ↵ is a complex number. Such a state is given by

X ↵ N⌫
|↵|2 /2
| ↵i =e p |N⌫ i; , (4.121)
N
N ⌫ !
⌫

p
with â⌫ |N⌫ i = N⌫ |N⌫ 1i. The expectation value for n̂⌫ = â†⌫ â⌫ is
† ⇤
hn̂⌫ i = h ↵ |â⌫ â⌫ | ↵i =h ↵ |↵ ↵| ↵i = ↵⇤ ↵ = |↵|2 (4.122)

and the variance is

hn̂2⌫ i hn̂⌫ i2 = hâ†⌫ â⌫ â†⌫ â⌫ i |↵|4 = hâ†⌫ â⌫ i + hâ†⌫ â†⌫ â⌫ â⌫ i |↵|4 = |↵|2 + |↵|4 |↵|4 = |↵|2 (4.123)

such that
hn̂2⌫ i hn̂⌫ i2 1 1
2
= 2
= . (4.124)
hn̂⌫ i |↵| hn̂⌫ i
Taking now the ~k = 0 state as coherent we identify
p
â0 | i = ↵0 | i = N0 ei | i . (4.125)

In this spirit we find that the mean value is

h b (~r)i = 0 (~
r) , (4.126)

which does not vanish for the condensed state. Note, however, hâ~k i = 0, if ~k 6= 0. The finite
value of hâ0 i requires states of di↵erent number of particles in the ~k = 0 state for the matrix
elements making up this mean value. This is an element of spontaneous symmetry breaking.
The condensate can be considered as a reservoir with on average N0 particles (N0 1), to
which we can add or from which we can remove particles without changing the properties of
the system. The coherent state satisfies this condition. We also can define an order parameter
characterizing the condensate (see also Chapter 6), the condensate wavefunction of Eq.(4.126),
p
r) = | 0 (~r)|ei (~r) = n0 ei .
0 (~ (4.127)

Spontaneous symmetry breaking occurs via the (arbitrary) choice of the phase of the condensate
wave function.
The number of particles and the phase are conjugate in the sense that a state with fixed
particle number has no definite phase (also no order parameter like Eq.(4.127)) and a state with

79
fixed phase has no definite particle number.7 First we consider the wave function of the coherent
state in the number representation,

|↵|2 /2 ↵N
N = hN | ↵i =e p (4.132)
N!
p
with ↵ = N0 ei 0 . Thus, the probability for the particle number N is given by

2 N0N 1 (N N0 )2 /2N0
PN = | N| = e N0 ⇡p e (4.133)
N! 2⇡N0
for large N0 . On the other hand, projecting into the phase representation,
1
X |↵|2 /2X1
e ↵N e i N
=h | ↵i = h |N ihN | ↵i = p p
N =0
2⇡ N =0 N!
(4.134)
Z 1 (N N0 )2 /4N
✓ ◆1/4
1 e 0
iN ( 0)
N0 ( 0)
2N
0 /4
⇡p dN e = e
2⇡ 0 (2⇡N0 )1/4 2⇡

such that r
N0 ( 2 2
0 ) N0 /2
P = |h | e ↵ i| . ⇡ (4.135)
2⇡
The Gaussian approximation is in both representations only valid, if N0 1. The coherent state
b i b
is neither an eigenstate of N nor e . But for both the distributions are well localized around
the corresponding mean values, N0 and 0 . The uncertainty relation is obtained by considering
the deviations from the mean values,
9
1 >
2
= h ↵ |( b 0 ) 2
| ↵ i = >
=
N0
) N =1, (4.136)
>
>
b N0 ) 2 | ↵ i = N0 ;
N 2 = h ↵ |(N

b , b] = i.
compatible with a commutation relation of the form [N
7
Phase and number operator eigenstates: The define the number operator and the phase operator and their
corresponding eigenstates.
b
b |N i = N |N i
N and ei | i = ei | i (4.128)
where the two states are connected by the Fourier transform
1
1 X iN eiN
| i= p e |N i with hN | i = p (4.129)
2⇡ N =0 2⇡

analogous to the relation between real and momentum space states. In this context care has to be taken to ensure
that the states | i form an orthogonal complete set of the Hilbert space. A way to construct this is to start with
an finite Hilbert space {|N i} assuming that 0  N  L 1 1. Then we can restrict ourselves to a discrete
set of phases = k = 2⇡k/L with k = 0, . . . , L 1 (analog to wave vectors in a finite system with periodic
boundary conditions). Now it is easy to see that

h k | k0 i = k,k0 . (4.130)

Keeping this in mind we take the limit L ! 1.

Based on this above operators can be represented as
1
X 1
X
b
b=
N N |N ihN | and ei = |N ihN + 1| ; . (4.131)
N =0 N =0

b and ei b the coherent state does not represent an eigenstate, but rather the best localized in
Thus for both N
either basis.

80
4.7.3 Phonons in an elastic medium
We consider here vibrations of an elastic media using a simplified model of longitudinal waves
only. As in Sect.3.8.2 we describe deformation of the elastic medium by means of the displace-
ment field ~u(~r, t). The kinetic and elastic energy are then given by
Z ✓ ◆2 Z
⇢m 3 @~u(~r, t) e ~ · ~u(~r, t))2 ,
Ekin = d r and Eel = d3 r(r (4.137)
2 @t 2

where ⇢m is the mass density of the medium and e denotes the elastic modulus. Note that we
use a simplified elastic term which involves density fluctuations only, corresponding to r ~ · ~u,
and ignores the contributions of shear distortion. This two energies are now combined to the
Lagrange functional L[~u] = Ekin Eel , whose variation with respect to ~u(~r, t) yields the wave
equation,
1 @2 ~ r~ · ~u) = 0 ,
~u r( (4.138)
c2l @t2
p
for longitudinal waves with the sound velocity cl = e /⇢m (Eq.(3.162). The general solution
can be represented as a superposition of plane waves,
1 X ⇣ ~ i~k·~
⌘
~u(~r, t) = p ~e~k q~k (t)eik·~r + q~k (t)⇤ e r
, (4.139)
V ~
k

with polarization vector ~e~k = ~k/|~k| and the amplitudes q~k (t) satisfy the equation,

d2
q~ + !~k2 q~k = 0 , (4.140)
dt2 k

with the frequency !~k = cl |~k| = cl k. We may rewrite the energy, E = Ekin + Eel , in terms of q~k ,
X h i
E= ⇢m !~k2 q~k (t)q~k⇤ (t) + q~k (t)⇤ q~k (t) . (4.141)
~k

which we express in a symmetrized form, for future convenience. Now we introduce new variables
p d p
Q~k = ⇢m (q~k + q~k⇤ ) and P~k = Q~ = i!~k ⇢m (q~k q~k⇤ ) , (4.142)
dt k
leading to the energy
1 X⇣ 2 2 2
⌘
E= P~k + !~k Q~k . (4.143)
2
~k

This corresponds to a set of independent harmonic oscillators labelled by the wave vectors ~k,
as we have seen in Sect.3.8. We now turn to the step of canonical quantization replacing the
variables (P~k , Q~k ) ! (Pb~k , Q
b~ ) which satisfy the standard commutation relation,
k

b~ , Pb~ 0 ] = i~
[Q ~k,~k 0 . (4.144)
k k

This can be reexpressed in terms of lowering and raising operators,

1 ⇣ ⌘ 1 ⇣ ⌘
b̂~k = p b~ + iPb~ ,
!~k Q k k
b̂~† = p b~
!~k Q k
iPb~k , (4.145)
2~!~k k 2~!~k

which obey the following commutation relations due to Eq.(4.144),

[b̂~k , b̂~† 0 ] = ~k,~k 0 , [b̂~k , b̂~k 0 ] = [b̂~† , b̂~† 0 ] = 0 . (4.146)

k k k

81
Therefore b̂~† and b̂~k can be viewed as creation and annihilation operators, respectively, for
k
bosonic particles, called phonons. The Hamiltonian can be now written as
X ✓ ◆ X ✓ ◆
† 1 1
H= ~!~k b̂~ b̂~k + = ~!~k n̂~k + (4.147)
k 2 2
~k ~k

whose eigenstates are given in the occupation number representation, |n~k1 , n~k2 , . . .i.
We can now also introduce the corresponding field operator using Eq.(4.139),
s
b 1 X ~ h i~k·~r † i~k·~
r
i
~u(~r) = p ~e~ b̂~ e + b̂~ e (4.148)
V ~ k 2⇢m !~k k k
k

which is not an eigen operator for the occupation number states. Actually the thermal mean
b(~r)i = 0.
value of the field vanishes h~u

Correlation function:
The correlation function is given by

g(~r b(~r) · ~u
~r 0 ) = h~u b(~r 0 )i b(~r)i · h~u
h~u b(~r 0 )i = h~u(~r) · ~u(~r 0 )i

1 X ~~e~k · ~e~k 0 Dh i~k·~r i~k·~

ih
~0 0 i~k 0 ·~
r0
iE (4.149)
= p b̂~k e + b̂~† e r
b̂~k 0 eik ·~r + b̂~† 0 e .
V 2⇢m !~k !~k 0 k k
~k,~k 0

Note that

hb̂~† b̂~k 0 i = hn̂~k i ~k,~k 0 , hb̂~k b̂~† 0 i = 1 + hn̂~k i ~k,~k 0 , hb̂~† b̂~† 0 i = hb̂~k b̂~k 0 i = 0 , (4.150)
k k k k

such that
~ X 1 n ~ r 0) i~k·(~ r 0)
o
g(~r ~r 0 ) = 1 + hn̂~k i eik·(~r ~
+ hn̂~k ie r ~
(4.151)
2⇢m V !~k
~k

Melting:
Instead of calculating the correlation function ~r 6= ~r 0 we now analyze the local (onsite) fluctua-
tion, i.e ~r = ~r 0 ,
✓ ◆
2 ~ X 1 ~ X 1 ~!~k
h~u(~r) i = 1 + 2hn̂~k i = coth . (4.152)
2⇢m V !~k 2⇢m V !~k 2
~k ~k

With !~k = cl k and the fact the number of degrees of freedom are limited, as described in
Sect.3.8.2 (k  kD ), we find
8
> kD kB T
Z kD ✓ ◆ >>
< 2⇡ 2 e
T ⇥D ,
2 ~ ~cl k
h~u(~r) i = dkk coth = (4.153)
(2⇡)2 ⇢m cl 0 2 >
> kD kB ⇥D
>
: T ⌧ ⇥D ,
8⇡ 2 e
which are at high (low) temperature thermal (quantum) fluctuations. As ~u denotes the deviation
of the position of an atom from its equilibrium position, we can apply Lindemann’s criterion for
melting of the systems. We introduce the lattice constant a with kD ⇡ ⇡/a. If h~u2 i is a sizeable
fraction of a2 then a crystal would melt. Thus we define the Lindemann number Lm with the

82
condition that the lattice is stable for h~u2 i < L2m a2 . Thus we obtain a melting temperture Tm
with

h~u2 i k B Tm
L2m = = ) kB Tm = 2⇡ ea
3 2
Lm = 2⇡⇢m a3 c2l L2m = 2⇡Mi c2l L2m , (4.154)
a2 2⇡ e a3

where Mi = ⇢m a3 is the atomic mass per unit cell. Note that usually Lm ⇡ 0.1 give a reasonable
estimate for Tm .
At sufficiently low temperature we can also observe quantum melting which occurs due to quan-
tum fluctuations, the zero-point motion of the atoms in a lattice. We consider T ⌧ ⇥D and fix
the temperature,
kB ⇥D ~ ~
L2m = 3
= 4 ) clm = (4.155)
8⇡a e 8a ⇢m a4 cl 8⇢m a4 L2m

which defines a critical value for the sound velocity, clm , which here is temperature independent.
For cl > clm the fluctuations are small enough that lattice stable, because it sti↵ enough, while
for cl < clm the lattice is ”soft” such that the zero-point motion destroys the lattice. We will
see in Chapt.7 that the 4 He shows such a quantum melting transition at very low temperature
under pressure (pressure increases elastic constant and sound velocity), where the solid phase is
stable at high pressure and turns into a liquid under decreasing the pressure.

Lower dimensions:
We consider the elastic medium at lower dimensions. For two dimensions, we rewrite Eq.(4.153),
Z kD ✓ ◆
2 ~ ~cl k
h~u(~r) i = dk coth (4.156)
4⇡cl ⇢m 0 2

and find that for all temperatures T > 0 the integral diverges at lower integral boundary
(”infrared divergence”). Only at T = 0 we find

~kD kB ⇥ D kB ⇥D
h~u(~r)2 i = = 2 = (4.157)
4⇡cl ⇢m 4⇡cl ⇢m 4⇡ e

finite. Thus in two dimension the lattice forming an elastic medium is only stable at zero
temperature. But still we can have quantum melting, if the lattice becomes sufficiently soft.
In one dimension Eq.(4.153) turns into
Z kD ✓ ◆
~ 1 ~cl k
h~u(~r)2 i = dk coth , (4.158)
2⇡cl ⇢m 0 k 2

which (infrared) diverges at all temperatures including T = 0. Quantum and thermal fluctua-
tions are strong enough in one dimension to destabilize any lattice.

83
Chapter 5

One-dimensional systems of
interacting degrees of freedom

So far we considered almost exclusively ideal systems of non-interacting degrees of freedom,

which we could treat with reasonable e↵ort. Systems with interactions are considerably more
complex to handle and in most cases no exact analytical discussion is possible. Notably in one
spatial dimension we find a number of models which can be treated exactly and give us some
interesting new aspects. Here we consider two classical systems, the spin chain and the lattice
gas in one-dimension. Note, however, that in principle there are exact solutions also for quantum
spin chains and other one-dimensional quantum models whose statistical physics, nevertheless,
is rather complex to address.

5.1 Classical spin chain

We consider here a chain of spins (magnetic moments) {~si } which are represented by freely
rotating vectors of fixed length S. They interact with each other via nearest-neighbor coupling
in the following way:
N
X
H(~si ) = J ~si · ~si+1 (5.1)
i=1

where J is the coupling strength. Note that this model has a global spin rotation symmetry,
i.e. the Hamiltonian remains invariant, if we rotate all the spin in the same way. It is known as
the Heisenberg Hamiltonian. For J < 0 the spins tend to align parallel (ferromagnetic) and for
J > 0 they tend to be antiparallel, alternating (antiferromagnetic). We require that the chain
is open and occupied by N + 1 spins (N bonds).

5.1.1 Thermodynamics
First we tackle the thermodynamics of the system using the canonical ensemble. Interestingly
the topology of the coupling allows to calculate the partition function exactly.1 We consider the
spins as vectors of fixed length whose degree of freedom is the angular orientation and we write
the partition function as
Z P
d⌦1 d⌦N +1
Z= ··· e J i ~si ·~si+1 , (5.2)
4⇡ 4⇡

which decays into a product form, if we consider the reference polar axis (”z-axis”) for the spin ~si
given by the direction of the spin ~si+1 (✓i is defined as the angle between ~si and ~si+1 ). Therefore
1
We follow here M.E. Fisher, American Journal of Physics 32, 343 (1964).

84
 J

J<0 J>0

ferromagnetic antiferromagnetic

Figure 5.1: Spin chain: Coupling J between spins of arbitrary orientation. J < 0: ferromagnetic
alignment preferred; J > 0: antiferromagnetic alignment preferred.

we may write
N ⇢Z
Y Z N ⇢Z Z
J~si ·~si+1 d⌦i d⌦N +1 Y JS 2 cos ✓i d⌦i d⌦N +1
Z = e = e
4⇡ 4⇡ 4⇡ 4⇡
i=1 i=1
(5.3)
⇢ Z +1 N ⇢ N
1 JS 2 x sinh( JS 2 )
= dxe =
2 1 JS 2

which looks similar to the expression obtained for the classical ideal paramagnet (see Sect.3.4.2).
We may actually consider each spin being subject to a Zeeman field induced via the coupling
by the neighboring spins, a feature which we will use later in the context of the mean field
approximation. It is interesting to see that Z does not depend on the sign of J. Thus, the
thermodynamics is the same for a ferromagnetic and an antiferromagnetic chain.
Easily we can determine the free and the internal energy,

F (T, N ) = N kB T ln( JS 2 ) N kB T ln sinh( JS 2 ) ,

(5.4)
U (T, N ) = N kB T N JS 2 coth( JS 2 ) .

In the low-temperature limit, JS 2 1, the internal energy approaches the ground state energy,
2
U ! N |J|S , i.e. all spins are parallel (J < 0) or alternating (J > 0). The heat capacity is
obtained through the derivative of U with respect to T ,
( ✓ ◆2 )
JS 2
C = N kB 1 (5.5)
sinh( JS 2 )

with a similar form as for the ideal paramagnet. Note, that C ! N kB in the zero-temperature
limit and, consequently, the third law of thermodynamic is not satisfied (see Fig.5.2).

5.1.2 Correlation function

We now discuss the correlation among the spins along the chain. Due to the spin-spin interaction,
a spin influences the orientation of other spins over larger distances. We define the correlation
function
l = h~si · ~si+l i = hsxi sxi+l i + hsyi syi+l i + hszi szi+l i = 3hszi szi+l i
Z P (5.6)
3 JS 2
= d⌦1 · · · d⌦N +1 S 2 cos ⇥i cos ⇥i+l e i cos ✓i
.
Z(4⇡)N +1

85
where ⇥i is the angle of spin i with respect to the general z-axis (e.g. along the chain). We can
perform all integrations for the spin ~sj with j < i and j > i + l which can be decomposed into
the same product form as discussed above. The remaining integrals are now
✓ ◆l+1 Z
2 JS 2 d⌦i 2
l = 3S 2
cos ⇥i e JS cos ✓i
sinh( JS ) 4⇡
(5.7)
Z Z
d⌦i+1 2
⇥ e JS cos ✓i+1 · · · d⌦i+l cos ⇥i+l
4⇡
Taking again the direction of spin ~si+1 as the reference for the spin ~si we find the relation2
cos ⇥i = cos ⇥i+1 cos ✓i + sin ⇥i+1 sin ✓i cos i . (5.10)
Inserting this we notice that the averages hcos ✓i i =
6 0 and hsin ✓i cos i i = 0. Thus
✓ ◆ l+1 Z
2 JS 2 d⌦i 2
l = 3S 2
cos ✓i e JS cos ✓i
sinh( JS ) 4⇡
Z Z
d⌦i+1 JS 2 cos ✓i+1 d⌦i+l
⇥ cos ⇥i+1 e ··· cos ⇥i+l (5.11)
4⇡ 4⇡
Z
JS 2 d⌦i JS 2 cos ✓i
= l 1 cos ✓i e = l 1 u( JS 2 )
sinh( JS 2 ) 4⇡
with 8
> 1
>
< sign(x) x 1,
1 x
u(x) = coth x ⇡ . (5.12)
x >
> x
: x⌧1.
3
If we take into account that l=0 = S 2 we find
2
⇥ ⇤
2 l 2 l/⇠
l = S u( JS ) = S e ( sign(J))l with ⇠ 1 = ln{|u( JS 2 )|} . (5.13)
Note that 0 < |u| < 1 for all finite temperatures and couplings. The correlation function decays
exponentially for all finite temperature and allows us to introduce a characteristic length scale,
the correlation length ⇠:
8
>
> [ln(3kB T /|J|S 2 )] 1 kB T |J|S 2
<
⇠(T ) = (5.14)
>
> |J|S 2
: kB T ⌧ |J|S 2
kB T
For large temperature the correlation length shrinks rather slowly to zero and in the zero-
temperature limit it diverges indicating an infinitely extending correlation at T = 0. Indeed we
find for T = 0 that u ! 1 such that

l (T = 0) = S 2 ( signJ)l (5.15)
for all l. This means that the spins order either ferromagnetically for J < 0 or antiferromag-
netically for J > 0 at T = 0. However, for any finite temperature we find l ! 0 for l ! 1.

2
Consider ~s1 and ~s2 and take ŷ 0 ? ẑ, ~s2 with |ŷ 0 | = |ẑ| = 1:
~s1 = ~s2 cos ✓1 + (ŷ 0 ⇥ ~s2 ) sin ✓1 cos 1 + ŷ 0 S sin ✓1 sin 1 (5.8)
and multiply by ẑ:
ẑ · ~s1 = cos ✓1 ẑ · ~s2 +ŷ 0 · (~s2 ⇥ ẑ) sin ✓1 cos 1 . (5.9)
| {z } | {z } | {z }
=S cos ⇥1 =S cos ⇥2 =ŷ 0 S sin ⇥2

86
 a) b)
antiferromagnetic

1/ χ
ferromagnetic

Figure 5.2: Thermodynamic properties of spin chain: a) heat capacity: saturates at low temper-
ature to N kB like the potential energy of a two-dimensional harmonic oscillator; b) Susceptibility
plotted inversely: ferromagnetic case show divergence at T = 0 and antiferromagnetic case ap-
proaches non-monotonically a finite value at T = 0. Extrapolations from high-temperature
region cut horizontal axis at positive (ferromagnetic) and negative (antiferromagnetic) axis,
indicating e↵ective energy scales for the corresponding correlations.

5.1.3 Susceptibility
Finally we want to study the susceptibility by assuming a coupling to the magnetic field of the
form
N
X +1
H0 = H g ~ .
~si · H (5.16)
i=1
Since it is impossible to use the above scheme to calculate the partition function for H0 with a
finite magnetic field, we will use the relation derived earlier (2.127) based on the fluctuations
of the magnetization, which does not require the knowledge of the partition function at finite
fields. In this way we obtain the susceptibility
PN +1 in the zero-field limit (linear response). The
magnetic moment is give by m ~ = g i=1 ~si . Thus, the susceptibility per spin is given by
N +1 N +1
1 g2 X X
zz = hm2z i hmz i 2
= hszi szj i hszi ihszj i . (5.17)
N kB T N kB T
i=1 j=1

The second term on the right hand side vanishes, since hszi i = 0 for all i in zero magnetic field.
For the first term we can use our result for the correlation function. In the limit of very large
N we obtain
N +1 N +1 N 1
!
g2 X X g2 X X
2
zz = |i j| ⇡ l=0 + 2S ul
3N kB T 3N kB T
i=1 j=1 i=1 l=1
(5.18)
2
✓ ◆ 2 2 2
g 2u g S 1 + u( JS )
= S2 1 + = .
3kB T 1 u 3kB T 1 u( JS 2 )
Here we use that for large systems (N ! 1) the corrections due to boundary contributions is
negligible. This susceptibility at high temperature (kB T |J|S 2 ) follows the Curie behavior
g2S 2
zz = (5.19)
3kB T

87
irrespective of the sign of J. For the antiferromagnetic coupling zz goes through a maximum at
kB Tmax ⇡ 0.238JS 2 to a finite zero-temperature value. On the other hand, in the ferromagnetic
case zz diverges at low temperatures

g 2 S 2 2|J|S 2
zz = , (5.20)
3 (kB T )2

more strongly than the Curie 1/T behavior.

Let us now also look at the first lowest order correction in the high-temperature limit,
✓ ◆
1 3kB T 2JS 2 3kB C
⇡ 2 2 1+ = 2 2 (T ⇥W ) ) zz = (5.21)
zz g S 3kB T g S T ⇥W

where ⇥W = 2JS 2 /3kB defines the so-called Weiss temperature. Thus the extrapolation
from the high-temperature limit of 1/ zz allows to determine ⇥W from the intersection on the
temperature axis and consequently the sign and magnitude of the J (see dashed lines in Fig.
5.2).
The Weiss temperature indicates the relevant energy scale of interaction among the spins. How-
ever, it does not imply that we would find a phase transition to a long-range ordered state (see
Sect.5.3). As our discussion in this section shows, there is no ordering at any finite temperature.

5.2 Interacting lattice gas

We consider a simple model of a gas of hard-core particles which occupy sites on a lattice.3
”Hard-core” means that it is not allowed to place two or more particles on the same site, i.e.
there is an infinite repulsive onsite interaction We introduce nearest-neighbor interaction and
neglect the kinetic energy. Then the Hamiltonian can be formulated as
L
X L ✓
X ◆
1
H=V ni ni+1 µ ni (5.22)
2
i=1 i=1

where the lattice has L sites and we assume periodic boundary conditions such that i = L + 1 !
i = 1. The coupling strength V for the interaction can be attractive (V < 0) or repulsive
(V > 0). The chemical potential is introduced and leave the particle number free, i.e. we work
with the grand-canonical ensemble which makes the problem more easy to solve. The variables
ni can only take the values 0 and 1 due to the hard-core constraint.

5.2.1 Transfer matrix method

Let us immediately determine the partition function,
X
Z= e H. (5.23)
{ni }

We introduce the 2 ⇥ 2-matrix (transfer matrix) for the bonds (i, i + 1),
[V ni ni+1 µ(ni +ni+1 1)/2]
Pni ,ni+1 = e
0 1 0 µ/2
1
P00 P01 e 1 (5.24)
) P̂ = @ A=@ A .
P10 P11 1 e V + µ/2

3
In the exercise we will show the connection of the lattice gas model to the Ising spin model. The corresponding
discussion will make the similarity of the two systems clear.

88
When we have one bond with periodic boundary condition it is obvious that the partition
function is given by four terms
X X
Z2 = e H(n1 ,n2 ) = Pn1 ,n2 Pn2 ,n1 = tr{P̂ 2 }
n1 ,n2 n1 ,n2
(5.25)
µ 2 V+ µ
= P00 P00 + P01 P10 + P10 P01 + P11 P11 = e +1+1+e

This allows us to rewrite the sum in the partition function as

X X X X
Z= ··· Pn1 ,n2 Pn2 ,n3 · · · PNL 1 ,nL PnL ,n1 = tr{P̂ L } = PnL1 ,n1 , (5.26)
n1 =0,1 n2 =0,1 nL =0,1 n1 =0,1

where we performed the matrix multiplication. Interestingly, we have now to analyse P L which
can easily be done in the eigenbasis of the matrix P̂ . For the trace we actually only need the
eigenvalues,
✓ ◆ s ✓ ◆
V /2 V µ V sinh2
V µ
± = e cosh ± e +1 (5.27)
2 2
and
L L L
Z= + + ⇡ + for L 1. (5.28)
Then the grand canonical potential and internal energy are given by
" ✓ ◆ s ✓ ◆ #
L V /2 V µ 2 V µ
⌦(T, µ) = kB T ln + = LkB T ln e cosh + e V sinh +1 .
2 2
(5.29)
The particle number of the system can then be calculated easily,
✓ ◆
L @⌦
N = Lhni i = (5.30)
2 @µ T

and the results are given in Figs.5.3 and 5.4 where N (µ) is plotted for di↵erent temperatures.
The two cases show distinct behaviors which is connected with the ground state properties.
For attractive interaction N changes in a single step, while for repulsive interactions two steps
appear with an intermediate plateaux at N = L/2.
The zero-temperature limit yields,
8
V |µ V | V < |µ V |
L<
⌦(T, µ) = (5.31)
2:
0 V > |µ V |

which leads to 8
1 + sign(µ V ) V < |µ V|
L<
N= (5.32)
2:
1 V > |µ V|
The condition V > |µ V | is never satisfied for V < 0, while it is valid in the range 0 < µ < 2V for
V > 0. Thus, the case V < 0 has two (Fig.5.3) and the case V > 0 has three regimes (Fig.5.4),
indicating that the two cases are di↵erent. Indeed there are two ground states for V < 0
correspond to a completely filled or empty lattice, the latter being favored by the attractive
interaction. On the other hand, the repulsive interaction V > 0 favors an intermediate state
where only every second lattice site is occupied.

89
 N
L
V
kB T =
10
kB T = V

L/2
kB T = 0.5V

1 0 µ/|V |

Figure 5.3: Particle number as a function of the chemical potential for V < 0 (attractive
interaction) for three di↵erent temperatures. The ground state corresponds to a completely
empty (full) lattice for µ < |V | (µ > |V |).

5.2.2 Correlation function

The transfer matrix method is also very convenient to obtain the correlation function,

l = hni ni+l i hni ihni+l i . (5.33)

We consider first the average particle density,

1 X
hni i = Pn1 ,n2 Pn2 ,n3 · · · Pni 1 ,ni ni Pni ,ni+1 · PnL ,n1
Z n1 ,...,nL
(5.34)
1 n o 1 n o
= tr P̂ i 1
ŵP̂ L i
= tr ŵP̂ L
Z Z
with ŵ defined as wn,n0 = n n,n0 . Let us transform this to the basis in which P̂ is diagonal
through the unitary transformation
✓ ◆ ✓ 0 0
◆
+ 0 0 w00 w01
1
Û P̂ Û = = P̂ and 1
Û ŵÛ = 0 0 = ŵ0 . (5.35)
0 w10 w11

This leads to
1 n 0 0 o w000 L
+
0
+ w11 L
0
hni i = tr ŵ P̂ = L
! w00 . (5.36)
Z + + L
In the same way we treat now
1 n o 1 n o 1 n o
hni ni+l i = tr P̂ i 1 ŵP̂ l ŵP̂ L i+1 l
= tr ŵP̂ l ŵP̂ L l = tr ŵ0 (P̂ 0 )l ŵ0 (P̂ 0 )L l
Z Z Z
✓ ◆l
w0 200 L
+
0 w0 (
+ w01 10
l L l+
+
l L l
+ ) + w0 211 L
2
= L + L
! w0 00 + 0
w01 0
w10 .
+ +
(5.37)
The correlation function is given by
✓ ◆l
0 0
l = w01 w10 = n2 e l/⇠
[sign( / + )]
l
(5.38)
+

90
 N
L V
kB T =
10
V
kB T =
100

L/2
kB T = V

2 0 +2 µ/V

Figure 5.4: Particle number as a function of the chemical potential for V > 0 (repulsive
interaction) for three di↵erent temperatures. The ground state corresponds to a completely
empty (full) lattice for µ < 0 (µ > 2|V |) and to a state with every second site only filled for
) < µ < 2|V |.

where we identify l=0 with n2 and the correlation length

1
⇠= . (5.39)
ln | +/ |

Now let us interpret the correlation function by writing

ni = hni i + ni (5.40)

where by definition, hni i = hhni i + ni i = hni i + h ni i yields h ni i = 0. Thus, the correlation

function (5.33) reads also
l = h ni ni+l i (5.41)
and describes the correlation from the deviation from the average value.
For V < 0 the correlation function decays monotonically with the same sign, because the
particles rather occupy neighboring sites to benefit from the attractive interaction. If there is a
particle on site i we then find, ni+l > 0.
In contrast, for V > 0 the correlation function shows alternating sign due to the trend to form
a particle density modulation, i.e. the particles prefer to occupy only every second lattice site
only in order to avoid the energy cost of the repulsive nearest-neighbor interaction. Thus, if
there is a particle on site i the neighboring site is more likely unoccupied, i.e. ni±1 < 0 while
the next-nearest neighbors are again more likely occupied with ni±2 > 0 and so on.
In the special case µ = V > 0 (intermediate range in Fig.5.4) we obtain
1 1
l = ( tanh( V /4))l and ⇠= . (5.42)
2 ln | coth( V /4)|

In the zero-temperature limit ⇠ diverges, but is finite for any T > 0. At T = 0 the state is a
perfectly alternating particle arrangement with degeneracy 2, as the particles may occupy either
all even- or all odd-numbered sites only.
It is important that both the thermodynamics as well as the correlation functions are determined
through the eigenvalues of the transfer matrix P̂ .

91
5.3 Long-range order versus disorder
We find in the one-dimensional spin chain no phase transition to any long-range ordered state at
finite temperatures. The term ”long-range order” is easiest understood through the correlation
function. We consider here again the example of a spin chain. The correlation function has the
property,
˜ l = h~si · ~si+l i h~si i · h~si+l i = h ~si · ~si+l i l!1
! 0 (5.43)
which is true for a spin system with or without order ( ~si = ~si h~si i). By rewriting

lim h~si · ~si+l i = h~si i · h~si+l i , (5.44)

l!1

we see that the left hand side is finite only with a finite mean value of h~si i. Note that for the
classical spin chain only at T = 0 long-range order is realized. All spins are parallel for J < 0
(ferromagnet) or alternating for J > 0 (antiferromagnet) such that
8
< ẑS J <0
h~si i = (5.45)
: i
ẑ( 1) S J > 0

assuming that the spin align parallel to the z-axis (ẑ is the unit vector along the z-axis). Natu-
rally the limit in Eq.(5.44) is then finite. 4 Also the lattice gas in one dimension does not show
long-range order except at T = 0.
Interestingly, in the case of quantum systems quantum phase fluctuations can destroy long-
range order even at T = 0. An important example is the antiferromagnetic spin chain with a
Heisenberg Hamiltonian. In contrast the ferromagnetic Heisenberg chain has an ordered ground
state with all spins aligned, because the state with all spins parallel is an eigenstate of the total
spin operator, S ~ˆtot = PN +1 ~sˆi and S
~ˆtot commutes with the Hamiltonian, since it has full spin
i=1
rotation symmetry as in the classical case (N + 1: number of spins on the chain),
z
| FM i = | """ · · · i ) Ŝtot | FM i = (N + 1)~s| FM i . (5.48)

In case of the antiferromagnetic chain the (classical) state with alternating spins | cAF i =
| "#"# · · · i, is not an eigenstate of the quantum Heisenberg Hamiltonian and consequently
not the ground state of the quantum antiferromagnetic chain. Our discussions of the melting
transition in Sect. 4.7.3 have shed some light on the role of thermal and quantum fluctuation in
this context, and indicate that dimensionality of a system is crucial for the stability of ordered
phases.

4
Previously we encountered long-range order in a Bose-Einstein condensate looking at the correlation function
in Eqs.(4.109) and (4.117). Extending the correlation function to
~ = h b † (~r) b (~r + R)i
g̃(R) ~ h b † (~r)ih b (~r + R)i
~ , (5.46)

we find that always limR!1

~
~ = 0, since with the Bogolyubov approximation,
g̃(R)

lim h b † (~r) b (~r + R)i

~ = lim h b † (~r)ih b (~r + R)i
~ = lim ⇤
0 (~
r) 0 (~
r ~ = n0
+ R) (5.47)
~
R!1 ~
R!1 ~
R!1

for T < Tc and a constant phase of 0 (~

r).

92
Chapter 6

Phase transitions

Phase transitions in macroscopic systems are ubiquitous in nature and represent a highly impor-
tant topic in statistical physics and thermodynamics. Phase transitions define a change of state
of a system upon changing external parameters. In many cases this kind of change is obvious,
e.g. transition between liquid and gas or between paramagnetic and ferromagnetic phase, and
in most cases it is accompanied by anomalies in measurable macroscopic quantities.
In a previous chapter we have seen a phase transition, the Bose-Einstein condensation. This
transition is special in the sense that it occurs for non-interacting particles. Generally, phase
transitions require an interaction favoring an ordered phase. Then the phase transition occurs
as a competition between the internal energy (or enthalpy) which is lowered by the order and
the entropy which at finite temperature favors disorder. The relevant thermodynamic potentials
to describe phase transitions are the Helmholtz free energy F (T, V, N, . . . ) and the Gibbs free
energy G(T, p, N, . . . ),
F = U TS and G = H TS . (6.1)
These potentials show anomalies (singularities) at the phase transition.
In this chapter we will consider the Ising model as one of the simplest models to describe phase
transitions from a statistical physics point of view. Based on this model we will introduce various
ways and concepts of addressing phase transitions.

6.1 Ehrenfest classification of phase transitions

The type of singularity in the thermodynamic potential defines the order of the phase transition.
According to Ehrenfest classification we call a phase transition occurring at a critical temperature
Tc (di↵erent phase for T > Tc and T < Tc ) to be of nth order, if the following properties hold:
✓ ◆ ✓ ◆ ✓ ◆ ✓ ◆
@mG @mG @mG @mG
= and = (6.2)
@T m p T =T @T m p T =T @pm T =Tc+ @pm T =Tc
c+ c

for m  n 1, and
✓ ◆ ✓ ◆ ✓ ◆ ✓ ◆
@nG @nG @nG @nG
6= and 6= (6.3)
@T n p T =T @T n p T =T @pn T =Tc+ @pn T =Tc
c+ c

The same definition is used for the free energy. In practice this classification is rarely used
beyond n = 2.
n = 1: A discontinuity is found in the entropy and in the volume:
✓ ◆ ✓ ◆
@G @G
S= and V = (6.4)
@T p @p T

93
The discontinuity of the entropy experimentally corresponds to the latent heat. The change in
volume is connected with the di↵erence in the density of the substance. A well-known example
is the transition between the liquid and the gas phase, for which the former is much denser than
the latter and accordingly takes a much smaller volume.
n = 2: The discontinuities are not in the first derivatives but in the second derivatives of the
Helmholtz free energy or Gibbs free energy, i.e. in the response functions. Some such quantities
are the heat capacity, the compressibility or the thermal expansion coefficient:
✓ 2 ◆ ✓ ◆ ✓ 2 ◆
@ G 1 @2G 1 @ G
Cp = T , T = , ↵ = (6.5)
@T 2 p V @p2 T V @T @p
As we will see later, second order phase transitions are usually connected with spontaneous
symmetry breaking and can be associated with the continuous growth of an order parameter.
Such transitions show also interesting fluctuation features which lead to the so-called critical
phenomena and universal behavior at the phase transition.
Ehrenfest relations: Interesting relations between various discontinuities at the phase transition
exist. They are generally known at Ehrenfest relations. We consider first a first-order transition
such as the gas-liquid transition. The phase boundary line in the p-T -phase diagram describes
the relation between temperature and vapor pressure in the case of liquid-gas transition. For
the di↵erentials of the Gibbs free energy in the two phases, the following equality holds:
dGl = dGg ) Sl dT + Vl dp = Sg dT + Vg dp . (6.6)
This allows us to get from the vapor pressure curve (p(T ) at the phase boundary in the p-T -
plane) the relation
dp S g Sl L
= = (6.7)
dT V g Vl T V
where L = T (Sg Sl ) is the latent heat and V = Vg Vl is the change of the volume. This
relation is known as the Clausius-Clapeyron equation.
If the transition is of second order then the both the entropy and the volume are continuous
through the transition between two phase A and B:
SA (T, p) = SB (T, p) and VA (T, p) = VB (T, p) , (6.8)
which yields the relations through the equality of their di↵erentials,
✓ ◆ ✓ ◆ ✓ ◆ ✓ ◆
@SA @SA @SB @SB
dSA = dT + dp = dT + dp = dSB ,
@T p @p T @T p @p T
✓ ◆ ✓ ◆ ✓ ◆ ✓ ◆ (6.9)
@VA @VA @VB @VB
dVA = dT + dp = dT + dp = dVB .
@T p @p T @T p @p T

We now use the Maxwell relation

✓ ◆ ✓ ◆
@S @V
= = V↵ (6.10)
@p T @T p

and obtain ⇣ ⌘ ⇣ ⌘
@SB @SA
dp @T @T Cp
= ⇣ ⌘p ⇣ ⌘p = (6.11)
dT @SB @SA TV ↵
@p T @p T
and analogously ⇣ ⌘ ⇣ ⌘
@VB @VA
dp @T @T ↵
= ⇣ ⌘p ⇣ ⌘p = . (6.12)
dT @VB @VA T
@p T @p T
Various other relations exist and are of experimental importance.

94
6.2 Phase transition in the Ising model
To discuss a concrete system we introduce here the Ising model is the simplest model of a
magnetic system. We consider magnetic moments or spins with two possible states, si = ±s
(Ising spins). Sitting on a lattice they interact with their nearest neighbors (analogously to the
spin chain in the Chapter 5). We write the model Hamiltonian as
X X
H= J si sj si H . (6.13)
hi,ji i
P
The sum hi,ji denotes summation over nearest neighbors on the lattice, counting each bond
only once. J is the coupling constant which we assume to be positive. The second term
corresponds to a Zeeman term due to an external magnetic field. The Ising spins are classical
variables, unlike quantum spins ~s whose di↵erent components do not commute with each other.
Ising spins represent only one component of a quantum spin and the Ising model may also be
considered as an extreme limit of anisotropic spin-spin interaction, where only one component
of the spins couples to neighboring spins.
The interaction favors the parallel alignment of all spins such that this model describes a fer-
romagnetic system. The ferromagnetic phase is characterized by a finite uniform mean value
hsi i = m 6= 0, the magnetization, even in the absence of an external magnetic field.

6.2.1 Mean field approximation

The analysis of many coupled degrees of freedom is in general not simple. For the Ising model
we have exact solutions for the one- and two-dimensional (quadratic lattice) case. For three
dimensions only numerical simulations or approximative calculations are possible. One rather
frequently used approximative method to study the phase transition and the magnetically or-
dered phase is the so-called mean field approximation. In the lattice model we can consider each
spin si as being coupled to a reservoir consisting of its neighboring spins. These neighboring
spins act then like a fluctuating field on the spin. In case of order they form a net directed field
on si , a mean field. Since this conditions applies to all spins equally, the scheme can be closed
by having a self-consistency in the mean field, as we will show below.
Let us now tackle the problem in a more systematic way. We rewrite the spin for each site

si = hsi i + (si hsi i) = m + (si m) = m + si (6.14)

and insert it into the Hamiltonian, where we approximate hsi i = m uniformly.

X X
H = J {m + (si m)} {m + (sj m)} si H
hi,ji i

X X
= J m2 + m(si m) + m(sj m) + si sj si H
(6.15)
hi,ji i

X⇣ z 2⌘ X X
= J zmsi m si H J si sj .
2
i i hi,ji

Here z is the number of nearest neighbors (for a hypercubic lattice in d dimensions z = 2d). In
the mean field approximation we neglect the last term assuming that it is small. This means
that the fluctuations around the mean value would be small,

h si sj i h si sj i
Eij = = ⌧1, (6.16)
hsi ihsj i m2

95
to guarantee the validity of this approximation. We will see later that this condition is not
satisfied very near the phase transition and that its violation will be important in the context
of the critical phenomena. We now write the mean field Hamiltonian
X z
Hmf = si he↵ + N J m2 with he↵ = Jzm + H , (6.17)
2
i

which has the form of an ideal paramagnet in a magnetic field he↵ . It is easy to calculate the
partition function and the free energy as a function of the parameter m,
( )N
Y X X
Jzm2 /2+ he↵ si Jzm2 /2+ he↵ s̃
ZN (T, m, H) = e = e
i si =±s s̃=±s (6.18)
Jzm2 N/2
=e {2 cosh( she↵ )}N

and
z
F (T, H, m) = kB T ln ZN = N J m2 N kB T ln {2 cosh( she↵ )} . (6.19)
2
The equilibrium condition is reached when we find the minimum of F for given T and H. To
this end we minimize F with respect to m as the only free variable,
@F
0= = N Jzm N Jzs tanh( she↵ ) . (6.20)
@m
This equation is equivalent to the self-consistence equation for the mean value of si :
P ✓ ◆
e Jzm N/2 X PN0 si0 hef f
2
s =±s si e si he↵ 1 @F
m = hsi i = si e i =1 = P i
si he↵
= s tanh( she↵ ) =
ZN si =±s e N @H T,m
{sj }
(6.21)
This is a non-linear equation whose solution determines m and eventually through the free
energy all thermodynamic properties. In zero magnetic field this equation is given as
✓ ◆
Jzsm
m = s tanh (6.22)
kB T

which yields Fig.6.1 for a graphical solution. 1

1
Variational approach: Consider the Ising model (6.13) without magnetic field. We now determine the free
energy on a variational level assuming a distribution of N independent spin with a net magnetization M = N+ N
(N = N+ + N ). The probability that a certain spin is ”+s” or ”-s” is given by w± = 12 (1 ± m̃) with m̃ = M/N .
There are
N!
!(M ) = ⇥ 1 ⇤ ⇥1 ⇤ . (6.23)
2
(N + M ) ! 2 (N M ) !
configurations corresponding to a given M . We may now determine the free energy as F = U T S = hHiM
kB T ln !(M ) in the following way:
X JN zs2 JN zs2 2
U = hHiM = J hsi sj i = (w+ w+ + w w w+ w w w+ ) = m̃ (6.24)
2 2
hi,ji

where we use for hsi sj i simply the configurational average for pairs of completely independent spins for given
M , i.e. w+ w+ + w w (w+ w + w+ w ) is the probability that neighboring spins are parallel (antiparallel).
In this approach there is no correlation between the neighboring spins. For the entropy term we use Stirling
approximation and keep only extensive terms,

1 1
T S = kB T ln !(M ) ⇡ N kB T ln 2 (1 + m̃) ln(1 + m̃) (1 m̃) ln(1 m̃) . (6.25)
2 2
Thus, we have expressed the free energy by a variational phase represented by independent spins whose variational

96
6.2.2 Instability of the paramagnetic phase
The disordered phase above a certain critical temperature Tc is called paramagnetic phase. For
this phase we consider first the magnetic susceptibility (T ) at zero magnetic field, which is
obtained from
⇢
dhsi i d2 F d @F @F @m d @F
(T ) = N = = + = (6.27)
dH H=0 dH 2 H=0 dH @H @m @H H=0 dH @H H=0

where we used the equilibrium condition (6.20). Thus we obtain

dm d
(T ) = N = Ns tanh [ (Jzsm(H) + sH)]
dH H=0 dH H=0
⇢ (6.28)
Ns dm s N s2
= Jzs +s = Jzs (T ) + .
kB T dH H=0 kB T kB T

where we used that for a paramagnet m(H = 0) = 0. This leads to the susceptibility

T = Tc
m
T > Tc
T < Tc

Figure 6.1: Graphical solution of the self-consistence equation (6.22). The crossing points of the
straight line and the step-like function gives the solution. There is only one solution at m = 0
for T Tc and three solutions for T < Tc .

N s2
(T ) = (6.29)
kB T Jzs2
which is modified compared to that of the ideal paramagnet. If kB T ! Jzs2 from above (T )
is singular. We define this as the critical temperature

Jzs2
Tc = . (6.30)
kB
parameter is the mean moment m = hsi = s(w+ w ) = sm̃. We minimize F with respect to m̃,
✓ ◆ ✓ ◆
@F N kB T 1 + m̃ Jzs2 m̃
0= = Jzs2 N m̃ + ln ) m̃ = tanh (6.26)
@ m̃ 2 1 m̃ kB T
which corresponds to Eq.(6.22) in the absence of a magnetic field. Thus, this variational approach and mean field
are equivalent and give the same thermodynamic properties. As we will see, the mean field approach is more
easily improved. In many cases similar variational calculations of the free energy based on independent degrees
of freedom yield thermodynamics equivalent to a mean field approach.

97
As the system approaches T = Tc it becomes more and more easy to polarize its spin by a tiny
magnetic field. This indicates an instability of the system which we now analyze in terms of
the self-consistence equation (6.22) in the absence of a magnetic field. Looking at Fig. 6.1 we
find that indeed the critical temperature Tc plays an important role in separating two types of
solutions of Eq.(6.22). For T Tc there is one single solution at m = 0 and for T < Tc there are
three solutions including m = 0, m(T ), m(T ). The physically relevant solution is then the one
with finite value of m, as we will show below. It is obvious that below T = Tc the mean field m
grows continuously from zero to a finite value. From Fig.6.1 we see that the critical temperature
can be derived from Eq.(6.22) by requiring that the slope of both sides at m = 0 is identical.
Thus taking the derivative we find
✓ ◆
Jzs2 2 Jzsm Jzs2 Jzs2
1= cosh = ) Tc = (6.31)
kB T kB T m=0 kB T kB
which agrees with Eq.(6.30).
In order to see which of the solutions is a minimum of the free energy we expand F in m assuming
that m and H are small.
 2 ⇢
m kB T ( she↵ )2 ( she↵ )4
F (T, H, m) ⇡ N Jz N kB T ln 2 (6.32)
2 Jz 2 12
For H = 0 we obtain
"✓ ◆ ✓ ◆3 #
Tc m2 1 Tc 4
F (T, H = 0, m) ⇡ F0 (T ) + N Jz 1 + m
T 2 12s2 T
(6.33)
✓ ◆ 2
T m m4
⇡ F0 (T ) + N Jz 1 +
Tc 2 12s2
where for the last step we took into account that our expansion is only valid for T ⇡ Tc .
Moreover, F0 = N kB T ln 2. This form of the free energy expansion is the famous Landau
theory of a continuous phase transition (for more details see later in Sect.6.4).

F T = Tc

T > Tc

T < Tc

Figure 6.2: Landau free energy: T > Tc : 2nd -order term is positive and minimum of F at
m = 0; T = Tc , 2nd vanishes and free energy minimum at m = 0 becomes very shallow; T < Tc :
2nd -order term is negative and minimum of F is at finite value of m, bounded by the 4th -order
term.

It is obvious that for T > Tc the minimum lies at m = 0. For T < Tc the coefficient of the
m2 -term (2nd -order) changes sign and a finite value of m minimizes F (see Fig. 6.2). The
minimization leads to 8 p
< ±s 3⌧ T < Tc
m(T ) = (6.34)
:
0 T Tc

98
with ⌧ = 1 T /Tc as a short-hand notation. There are two degenerate minima and the system
chooses spontaneously one of the two (spontaneous symmetry breaking).
Next we analyze the behavior of the free energy and other thermodynamic quantities around
the phase transition. The temperature dependence of the free energy and the entropy is given
by

3N kB Tc ⌧ 2 @F (T ) 3N kB ⌧
F (T ) = F0 (T ) ⇥(⌧ ) and S(T ) = = N kB ln 2 ⇥(⌧ ) ,
4 @T 2
(6.35)
and eventually we obtain for the heat capacity,
C @S 3N kB
= = ⇥(⌧ ) + C0 (6.36)
T @T 2Tc
where C0 is zero in the present approximation for H = 0. While the free energy and the entropy
are continuous through the transition, the heat capacity shows a jump indicating the release of
entropy through ordering. Thus, we conclude that this phase transition is of second order.

-0.5
F/NkBTc

-0.75

0.75
-1

0.5
S/NkB

0.25

0
2
C Tc / NkBT

1.5

0.5

0
0 0.5 1
T / Tc

Figure 6.3: Thermodynamic quantities within mean field theory. Free energy, entropy and heat
capacity.

Within mean field approximation the region close to Tc is described by the Landau expansion.
However, taking the solution of the complete mean field equations leads to the thermodynamic
behavior for the whole temperature range as shown in Fig. 6.3. Note that in this mean field
approximation the entropy is N kB ln 2 in the paramagnetic phase, the maximal value the entropy
can reach.

6.2.3 Phase diagram

So far we have concentrated on the situation without magnetic field. In this case the phase
transition leads to one of two degenerate ordered phases, m = +|m| or m = |m|. An applied
magnetic field lifts this degeneracy by introducing a bias for one of the two states, favoring the
order with m parallel to the field. In a finite field the transition turns into a crossover, since
there is already a driven moment m for temperatures above Tc . This is also reflected in the
thermodynamic properties which show broad features around Tc and not anymore the sharp
transition, e.g. the heat capacity is turned into a broadened anomaly (see Fig. 6.4).
Next we turn to the behavior of the magnetization m as a function of the magnetic field and
temperature (illustrated in Fig. 6.5 and 6.6). At H = 0 going from high to low temperatures
the slope of m(H)|H=0 is linear and diverges as we approach Tc . This reflects the diverging
susceptibility as a critical behavior.

99
 1.5 1.5

m(T,H) / s

CTc / NkBT
1 1

0.5 0.5

0 0
0 0.5 1 1.5 0 0.5 1 1.5
T / Tc T / Tc

Figure 6.4: Ising system in a finite magnetic field: (left panel) Magnetization as a function of
temperature in a fixed magnetic field (solid line) and in zero field (dashed line); (right panel)
heat capacity for a fixed magnetic field. In a magnetic field no sharp transition exists.

stable
m

metastable

instable

H
T > Tc

T = Tc
T < Tc

Figure 6.5: Magnetization as a function of magnetic field for di↵erent temperatures.

For all temperatures T > Tc , the magnetization m(H) is a single-valued function in Fig. 6.5 .
Below Tc , however, m(H) is triply valued as a solution of the self-consistence equation. The part
with dm/dH > 0 is stable or metastable representing local minima of the free energy. The part
of dm/dH < 0 corresponds to a local maximum of the free energy and is unstable.2 Considering
for T < Tc the magnetic field running from negative to positive values, we find for H < 0 the
state with m < 0 has lowest free energy. At H = 0 there is a degeneracy between +|m| and |m|
as both minima have the same energy and for H > 0, m > 0 is the lowest energy phase. Thus we
2
Metastable and unstable states: We consider the uniform free energy of the magnetization m including the
Zeeman coupling to the external field H, F 0 (m, H) = F (m) mH. The (local) minimum of the free energy is
determined by
@F 0 @F @2F 0 @2F
= H=0 and = >0. (6.37)
@m @m @m2 @m2
while a local maximum has a negative second derivative. Considering the derivative with respect to H at the
extrema defined by Eq.(6.37), for m = m(H),

d @F 0 @ 2 F dm
= 1=0. (6.38)
dH @m m=m(H) @m2 dH m=m(H)

Therefore, 8
>
> @2F 1
< @m2 )
> ( >0 stable / metastable
dm
= = (6.39)
dH >
> 2
: (@ F )
> 1
<0 unstable .
@m2

100
encounter a level-crossing of the free energy for the two states at H = 0 and the magnetization
jumps from negative to positive direction. This represents a first-order transition, since the free
energy as a function of H is singular, i.e. it has a sharp slope change (a discontinuity in the first
derivative of F with respect to H).3
This is visualized in a three-dimensional graph in m, H and T (Fig. 6.6). The shaded region
appearing for T < Tc is the region where the two degenerate phases coexist. As m is changed in
this region, the fraction of the two degenerate finite magnetization phases is changed continuously
following a Maxwell construction.

m(q) = q|m(H = 0, T )| + (1 q){ |m(H = 0, T )|} = (2q 1)|m(H = 0, T )| (6.40)

where q can change continuously with 0  q  1. This may be viewed as domains of the two
states changing in size.

H
p
isothermal

T isothermal

m
liquid
gas

Figure 6.6: Left panel: phase diagram of ferromagnetic Ising model in (H, m, T ); right panel:
phase diagram of the liquid-gas transition in (p, V, T ). In both cases the shaded region represents
a coexistence region.

In the H-T -plane this shaded region is simply a line for the first order transition of a discontin-
uously changing magnetization.

6.3 Gaussian transformation

We analyze the mean field approximation in a di↵erent formulation, using the Gaussian transfor-
mation. The partition function of the Ising model can be rewritten by introducing a continuous
3
Note that in reality ramping H through 0 does not necessarily lead to a sharp transition. Flipping all
magnetic moments at once is not a trivial matter. Thus the system may stay trapped in the metastable free
energy minimum for a certain range of positive H. The transition may happen through the nucleation of reversed
magnetic bubbles (domains) and the subsequent expansion of the reversed region by means of domain wall motion.
(Similar to the nucleation and expansion of droplets in undercooled vapor.) There is a certain field where the
metastable minimum becomes a saddle point and thus unstable for infinitesimally higher fields. Varying the
magnetic field periodically can lead to hysteretic loops.

101
auxiliary field i:
X P P
Jij si sj + i si H i
Z = e 2 i,j

{si }

Z !
1 +1 Y P 1) Y X
= p d e2 i,j (J ij ( i Hi )( j Hj )
e i si
i0
(2⇡kB T )N/2 detJ 1 i0 i si =±s

Z !
1 +1 Y P 1)
P
= p d e2 i,j (J ij ( i Hi )( j Hj )+ i ln[2 cosh( s i )]
i0
(2⇡kB T )N/2 detJ 1 i0
(6.41)
where we use the N ⇥ N -matrix
8
< J (i, j) nearest neighbors
Jij = (6.42)
:
0 otherwise

and (J 1 )ij is the inverse of Jij . We assume the magnetic field to be site dependent, which will
be convenient later. We used the identity
Z +1 2
Z +1 p
a 2 1 2 a 2
+s s
d e 2a =e 2 d e 2a ( sa) = 2⇡ae 2 s
1 1

Z ! (6.43)
+1 Y 1 P 1)
P p 1 P
i (A ij j + i si N/2 si Aij sj
) d i e 2 i,j i = (2⇡) detA e 2 ij

1 i

with A being a positive definite N ⇥ N -matrix. This exact rewriting of the partition function
is called Gaussian transformation (also known as Hubbard-Stratonovich transformation).4 We
replaced here the discrete variable si by a continuous field i .
We introduce the potential S( i , Hi ) and write
Z +1 Y !
Z=C d i0 e S( i ,Hi ) = e F (6.44)
1 i0
p
with C = 1/(2⇡kB T )N/2 detJ and
1X 1 1X
S( i , Hi ) = (J )ij ( i Hi )( j Hj ) ln[2 cosh( s i )] . (6.45)
2
i,j i

The evaluation of these integrals in Eq.(6.44) is, however, as difficult as it was to determine
the partition function for the original formulation with Ising spins. One approximation is based
on the proposal that the integral is dominated by the maximum of the integrand. The type of
evaluation is known as saddle point approximation (sometimes also called method of steepest
descent). Thus, one replaces the auxiliary field by the value ¯i which dominates the integral
for the partition function.5 This method is similar to discussions we had earlier when we tested
the equivalence of di↵erent ensembles, although statistically less reliable as we will see later. It
relies on the assumption that the fluctuations of the field i around ¯i are small - something
4
Reference: ”Field Theory, the Renormalization Group, and Critical Phenomena”, Daniel J. Amit, World
Scientific.
5
Method of steepest descent: We consider the integral
Z b
I= eN g(x) dx (6.46)
a

102
which has to be tested. Therefore we look for the minimum of S now and approximate Z then
by

@S X
S( ¯i ,Hi )
Z ⇡ Ce with 0= = (J 1
)ij ( ¯j Hj ) s tanh( s ¯i ) (6.49)
@ i ¯
i= i j

which leads to the saddle-point equation

X
¯i = Hi s Jij tanh( s ¯j ) (6.50)
j

For Hi = 0 the saddle point is given by the uniform solution ¯i = ¯, satisfying

X
¯= s Jij tanh( s ¯) = Jzs tanh( s ¯) . (6.51)
j

This yields the same critical temperature for the onset of a finite solution for ¯ as the mean field
solution, following the same scheme of discussion as above. The relation to m of the mean field
approximation is given by the condition

@ ln Z dS( ¯i , Hi ) @S X
hsi i = kB T = = = (J 1
)ij ( ¯j Hj ) = s tanh( s ¯i ) (6.52)
@Hi dHi @Hi
j

such that
m = s tanh( s ¯) ) ¯ = Jzm . (6.53)

The discussion of the thermodynamic properties are in this formulation analogous to the ordinary
mean field treatment and give exactly the same behavior. We are, however, now in a position
to go beyond the mean field level and, in particular, to test the validity of the mean field
approximation.

6.3.1 Correlation function and susceptibility

We consider first the correlation function
@ 2 ln Z d2 S( ¯i , Hi )
ij = hsi sj i hsi ihsj i = (kB T )2 ⇡ kB T . (6.54)
@Hi @Hj dHi dHj

With (6.52) we obtain

d d ¯i
ij = s tanh( s ¯i ) = s2 cosh 2
( s ¯) (6.55)
dHj dHj

which if inverted yields

⇢
1 dHj 1 s2 Jij
ij
1
= cosh2 ( s ¯) ¯ = 2 cosh2 ( s ¯) ij + (6.56)
s2 d i s cosh2 ( s ¯)
and N 1 and g(x) a function with a unique maximum at x = x̄ in the interval [a, b], i.e.
1 00
g(x) = g(x̄) + g 0 (x̄)(x x̄) + g (x̄)(x x̄)2 + · · · . (6.47)
2
For the maximum, the conditions g 0 (x̄) = 0 and g 00 (x̄) < 0 holds. We approximate the integral
Z b Z +1 ✓ ◆1/2
1 00 2 1 00 2 2⇡
I ⇡ eN g(x̄) dx e 2 N |g (x̄)|(x x̄) ⇡ eN g(x̄) dx e 2 N |g (x̄)|(x x̄) = eN g(x̄) (6.48)
a 1 N |g 00 (x̄)|

which is exact in the limit N ! 1. Considering ln I we find that ln I ⇡ N g(x̄) + O(ln N ).

103
where we use (6.50)
X dHj s2 Jij
Hj = ¯j + s Jjj 0 tanh( s ¯j 0 ) ) = ij + . (6.57)
d ¯i cosh2 ( s ¯)
j0

The following Fourier-transformations lead to a simple form for (6.56), 6

1 X 1 X 1 X i~q·(~ri
Jij = J(~q )ei~q·(~ri ~
rj )
, ij = (~q )ei~q·(~ri ~
rj )
, ij = e ~
rj )
(6.62)
N N N
q~ q~ q~

with
kB T 0 s2
(~q ) = with 0 = = s2 1 tanh2 ( s ¯) = (s2 m2 ) ,
1 + 0 J(~q ) 2 ¯
cosh ( s )
(6.63)
using (6.53). On a d-dimensional hypercubic lattice with only nearest-neighbor coupling, we
obtain for the Fourier transformed coupling strength,
d
1 X q ·(~
i~ ri ~
rj )
X
J(~q ) = Jij e = 2J cos q↵ a (6.64)
N
i,j ↵=1

with the lattice constant a. As we will see below, if we focus on the long-distance correlations
only, we can restrict ourselves to the small ~q range and we expand J(~q ) as

J(~q ) ⇡ Jz + Jq 2 a2

kB T kB T s2 (6.65)
) (~q ) ⇡ kB T
⇡
Jz + Jq 2 a2 kB (T Tc ) + Js2 q 2 a2 + kB T m2 /s2
s2 m 2

where for the last approximation, we assumed m ⌧ s as is the case in ordered phase close to Tc
and for T > Tc . This correlation function has then the famous Ornstein-Zernike form,
A
(~q ) = (6.66)
1 + ⇠2q2
where ⇠ is the correlation length, as introduced in (5.13).
First let us use this result to determine the susceptibility. For this purpose we take the earlier
derived connection of the susceptibility with the fluctuations.
X X N s2
= {hsi sj i hsi ihsj i} = ij =N (~q = 0) = (6.67)
kB (T Tc ) + kB T m2 /s2
i,j i,j
6
For the inversion of ij it is important to realize that ij = (~ri ~rj ) is translation invariant. We use now
the linear equation X X
1
ai = ij bj ) ij aj = bi . (6.58)
j j

We perform now the Fourier transform of the second equation,

1 XX q ·~ 1 XX q ·(~ 1 X
ij aj e
i~ ri
= ij aj e
i~ ri ~
rj )
e i~q·~rj = (~
q )a(~
q) = bi e ~·~
q ri
= b(~
q) . (6.59)
N i j N i j N i

On the other hand, we find

1 XX 1 1 XX 1 X
bj e i~q·~ri = 1
bj e q ·(~
i~ ri ~
rj )
e q ·~
i~ rj
= 1
(~
q )b(~
q) = ai e ~·~
q ri
= a(~
q) , (6.60)
N i j ij N i j ij
N i

leading to the relation

1 1
(~
q) = , (6.61)
(~
q)
1
which is a convenient way to handle the inverted matrix ij .

104
We use now the earlier mean field result for m2 = 3s2 ⌧ in (6.34) and obtain for the regime very
close to Tc the behavior, 8
>
> N s2
>
> T > Tc ,
< kB (T Tc )
(T ) = (6.68)
>
> 2
>
> N s
: T < Tc ,
2kB |T Tc |
showing that the susceptibility is singular approaching Tc from both sides of the phase transition
with the same exponent for |⌧ |.
Now return to the correlation function in the paramagnetic phase (T > Tc ) and represent the
correlation function in real space, where we restrict ourselves to the long-distance behavior
r ! 1. Thus we have the Fourier-transform (~q ). For large |~r| the phase factor ei~q·~r rapidly
oscillates for for large ~q such that in (~q ) only small ~q are important.
For simplicity we consider only the three-dimensional case,
Z Z 1 Z Z 1
d3 q q ·~
i~ r A 2 eiqr cos ✓ A eiqr e iqr
~
r = (~
q )e ⇡ dq q d✓ sin ✓ = dq q
(2⇡)3 4⇡ 2 0 1 + ⇠2q2 4⇡ 2 ir 0 1 + ⇠2q2
Z +1
A eiqr A e r/⇠ kB T e r/⇠
= 2 dq q = =
4⇡ ir 1 1 + ⇠2q2 4⇡ r⇠ 2 4⇡J r
(6.69)
where we used residue calculation for the last integral and introduced the parametrization,
s2 kB T ⇠ 2 Js2 a2
A= = and ⇠2 = . (6.70)
1 Tc /T Ja2 kB (T Tc )
The general form of the correlation function for other dimensions d is
e r/⇠
~
r / (6.71)
r(d 1)/2

if T > Tc .7 In all cases there is a correlation length which diverges as we approach T ! Tc+ .
At Tc we find 8
Z < ln r d = 2
kB T dd q ei~q·~r
~
r = / (6.75)
Ja2 (2⇡)d q 2 : 2 d
r d 6= 2
This suggests that for T ! Tc+ the correlation function should rather behave as
8
>
> ln r e r/⇠ d = 2
<
~r / (6.76)
> r/⇠
: e
>
d 6= 2
r 2
d

7
Note that the correlation function in ~
q -space
A
(~
q) = , (1 + ⇠ 2 ~
q ) (~
q) = A (6.72)
1 + ⇠2~
q2
is the Fourier transform of the di↵erential equation
⇢
~ 2} ⇠2 @ d @
A (~r) = {1 ⇠2r ~
r = 1 r 1
~
r . (6.73)
rd 1 @r @r
With the ansatz ~r = rb e r/⇠ we obtain easily, b = (1 d)/2 as given in (6.71). At the transition point we have
⇠ 1 = 0 such that the di↵erential equation has then the form,
8
< r d=1
r~ ~r = Ã (~r)
2
) r /
~ ln r d=2 , (6.74)
: 2 d
r d 3
as given in (6.75).

105
which we will encounter later in the context of scaling again.
Eventually we may characterize the ordered phase (T < Tc ), also through the behavior of the
correlation function, which has the property

lim ~
r =0 (6.77)
r!1

for T > Tc as well as T < Tc . In the latter case ⇠ plays the role of ”healing length” over which
any distortion of the mean field recovers to its uniform value hsi i = m. This means now that
8
< 0 T > Tc ,
lim hsi sj i = hsi ihsj i = (6.78)
|~
ri ~rj |!1 : 2
m T < Tc .

Thus in the ordered phase spins are correlated of ”infinite” distances (defined only in the ther-
modynamic limit of infinitely large systems) and are called long range ordered in contrast to
short range order which means correlations over short distances only.

6.4 Ginzburg-Landau theory

We have used the Landau expansion of the free energy above to discuss phase transitions in
the vicinity of the critical temperature where m was small. This can be extended to a highly
convenient method which allows us to discuss phase transition more generally, in particular, those
of second order. Landau’s concept of the disorder-to-order transition is based on symmetry
and spontaneous symmetry breaking. The disordered high-temperature phase has a certain
symmetry which is characterized by a group G of all symmetry operations leaving the system
invariant. At the second order phase transition a form of order appears reducing this symmetry
such that the low-temperature phase obeys a symmetry group G 0 which is a subgroup of G. This
change of symmetry is called spontaneous symmetry breaking whereby the system chooses one
among several degenerate states. The number of choices corresponds to the order of G/G 0 , i.e.
order of G devided by order of G 0 . This symmetry lowering is described by the appearance of
an order parameter, a quantity which does not possess all symmetries of the group G.
A further important aspect emerges when long-length scale variations of the order parameter
are taken into account. This can be easily incorporated in the Ginzburg-Landau theory and
allows to discuss spatial variations of the ordered phase as well as fluctuations.

6.4.1 Ginzburg-Landau theory for the Ising model

For the Ising model of the previous section we can identify m as the order parameter. The order
parameter m is not invariant under time reversal symmetry K̂,

K̂m = m. (6.79)

The two states with positive and negative m are degenerate. The relevant symmetry group
above the phase transition is
G =G⇥K (6.80)
with G as the space group of the lattice (simple cubic) and K, the group {E, K̂} (E denotes the
identity operation). As for the space group we consider the magnetic moment here independent
of the crystal lattice rotations such that G remains untouched through the transition so that
the corresponding subgroup is
G0 = G ⇢ G (6.81)
The degeneracy of the ordered phase is given by the order of G/G 0 which is 2 in our case.

106
The Ginzburg-Landau free energy functional has in d dimensions the general form
Z ⇢
d A B  ~
F [m; H, T ] = F0 (H, T ) + d r m(~r)2 + m(~r)4 H(~r)m(~r) + [rm(~ r)]2
2 4 2
Z (6.82)
= F0 (H, T ) + ~
dd r f (m, rm; H, T )

where we choose the coefficients according to the expansion done in (6.33) as

✓ ◆
Jz T Jz Jz
A= d 1 = ⌧ and B = 2 d . (6.83)
a Tc ad 3s a
Here a is the lattice constant. We have introduced the spatial continuum limit for the order
parameter m which is based on the procedure of coarse graining. We take a block of sites with
the volume Ldb with Lb much larger than the lattice constant a and define

1 X ad
m(~r) = hsi i with Nb = (6.84)
Nb Ldb
i2⇤b (~
r)

and ⇤b (~r) is the set of sites in the block around the center position ~r. Here we assume that hsi i
is changing slowly in space on the length scale Lb .
Under this condition we can now also determine  from the model Hamiltonian using the fol-
lowing consideration. The variational equation of the free energy functional is given by
F @f ~ · @f = ~ 2 m + Am + Bm3
0= ) 0= r r H (6.85)
m @m ~
@ rm
Let us compare this equation with the saddle-point equation (6.50,6.51) assuming H = 0. It is
sufficient to keep only the terms of first order in m. We then take the equations (6.50,6.53) and
expand the self-consistence equation to linear order
X X X
¯i = ¯(~ri ) = s Jij tanh( s ¯(~rj )) ⇡ s2 Jij ¯(~rj ) = s2 J ¯(~ri + ~a)
j j {~a}n.n.

2 3 (6.86)
X X X @2 ¯
= s2 J 4z ¯(~ri ) + ~ ¯(~ri ) + 1
~a · r aµ a⌫ (~ri )5 .
2 µ,⌫=x,y...
@rµ @r⌫
{~a}n.n. {~a}n.n.
P
The sum {~a}n.n. runs over nearest-neighbor sites. Note that the second term in the bracket
[. . . ] vanishes due to symmetry. Now using the coarse graining procedure we may replace
¯(~ri ) = Jzm(~r) and obtain
✓ ◆
T ~ 2 m(~r) ,
0 = Jz 1 m(~r) Ja2 r (6.87)
Tc
and the comparison of coefficients leads to

 = Ja2 d
. (6.88)

We may rewrite the equation (6.87) as

~ 2m a 2 k B Tc Js2 a2
0=m ⇠2r with ⇠2 = = (6.89)
zkB (T Tc ) kB (T Tc )
where we introduced the length ⇠ which is exactly equal to the correlation length for T > Tc in
(6.70).

107
6.4.2 Critical exponents
Close to the phase transition at Tc various quantities have a specific temperature or field de-
pendence which follows powerlaws in ⌧ = 1 T /Tc or H with characteristic exponents, called
critical exponents. We introduce here the exponents relevant for a magnetic system like the Ising
model. The heat capacity C and the susceptibility follow the behavior
↵
C(T ) / |⌧ | and (T ) / |⌧ | (6.90)
for both ⌧ > 0 and ⌧ < 0. Also the correlation length displays a powerlaw
⌫
⇠(T ) / |⌧ | . (6.91)
For ⌧ > 0 (ordered phase) the magnetization grows as
m(T ) / |⌧ | . (6.92)
At T = Tc (⌧ = 0) the magnetization has the field dependence
m / H 1/ (6.93)
and the correlation function has a powerlaw dependence on the distance r
1
.
~
r / (6.94)
rd 2+⌘
These exponents are not completely independent but are related by means of so-called scaling
laws:
• Rushbrooke scaling: ↵ + 2 + =2
• Widom scaling: = ( 1)
• Fisher scaling: = (2 ⌘)⌫
• Josephson scaling: ⌫d = 2 ↵
We do not derive all relations, but restrict to Fisher’s result in order to show the basic idea. We
consider the correlation function for ⌧ < 0 but very close to Tc . Then using (6.94) we assume
that we can write ~r as
1
~r / d 2+⌘ g(r/⇠) . (6.95)
r
According to our previous discussion the susceptibility is given by the integral of ~r over space
Z Z Z
d d 1 2 ⌘ 1
/ d r ~r / d r d 2+⌘ g(r/⇠) / ⇠ dd y d 2+⌘ g(y) / |⌧ | ⌫(2 ⌘) / |⌧ | (6.96)
r y
which leads to = ⌫(2 ⌘).
Let us now determine the exponents within mean field theory. The only one we have not found
so far is . Using the Ginzburg-Landau equations for ⌧ = 0 leads to
Bm3 = H ) =3 (6.97)
Thus the list of exponents is
1 1
↵=0, =, =1 =3, ⌫= ⌘=0 (6.98)
2 2
These exponents satisfy the scaling relations apart from the Josephson scaling which depends
on the dimension d.
The critical exponents arise from the specific fluctuation (critical) behavior around a second-
order phase transition. They are determined by dimension, structure of order parameter and
coupling topology, and are consequently identical for equivalent phase transitions. Therefore,
the critical exponents incorporate universal properties.

108
6.4.3 Range of validity of the mean field theory - Ginzburg criterion
In Eq. (6.16) we gave a condition for the validity of the mean field approximation. The fluc-
tuations around the mean field should be small. We formulate this statement more concretely
here. In previous chapters we have found that for large systems the fluctuations are very small
compared to the mean value (e.g. energy, particle number, ...). Looking at the fluctuations of
the magnetic moments, the question arises what is the natural ”system size”. The natural length
scale for fluctuations is the correlation length which we know from the correlation function ij .
Thus, the relevant ”system size” corresponds to the volume V⇠ = ⇠ d . Looking at the ratio

h(si hsi i)(sj hsj i)i

Eij = (6.99)
hsi ihsj i

we estimate denominator and numerator in the same volume V⇠ in the following way,
P
j2⇤⇠ 0j
EGL = P , (6.100)
j2⇤⇠ hs0 ihsj i

where ⇤⇠ denotes the set of lattice points around 0 covered by the volume V⇠ . We first look at the
numerator and use the fact that ~r decays on the length ⇠ quickly. Thus we use the fluctuation-
dissipation theorem relating the correlation function and the susceptibility and obtain,
Z
(T )
dd r ~r = f kB T . (6.101)
V⇠ N

where the factor f is independent of ⇠ and temperature and P gives the fraction to which integral
limited to V⇠ corresponds to the susceptibility kB T = i,j ij . We can estimate it by using
the scaling form of ~r ,
Z ⇠ Z 1 Z 1 Z 1
rd 1 rd 1
dr d 2 g(r/⇠) = f dr d 2 g(r/⇠) ) ⇠2 dx xg(x) = f ⇠ 2 dx xg(x) ,
0 r 0 r 0 0
(6.102)
which leads indeed to f independent of ⇠ and of order one. Next we consider the denominator
of EGL and we restrict to T < Tc so that we immediately find
X
hs0 ihsj i = m(T )2 N⇠ (6.103)
j2V⇠

with N⇠ = (⇠/a)d as the number of lattice sites within the volume V⇠ . The criterion for the
mean field theory to be applicable is then given by
✓ ◆d ✓ ◆d
(T ) ⇠ 2 f s2 2 ⇠0
f kB T c ⌧ m(T ) ) ⌧ 3s ⌧ ⌧ d/2 (6.104)
N a 2⌧ a
with ⇠0 as the correlation length at zero-temperature. Since we are testing the mean field theory
we have used here the mean field exponents of Eq.(6.98). From these we obtain the range of
validity
✓ ◆
4 d f a d
⌧ 2 (6.105)
6 ⇠0
which excludes obviously a narrow range of temperature close to Tc for dimensions d  3. For
dimensions d > 4 there is no limitations and the mean field theory including the Ginzburg-
Landau theory is valid also close to Tc . The temperature region
( ✓ ◆ ) 2
f a d 4 d
T = Tc T ⇠ Tc (6.106)
6 ⇠0

109
is called the critical region and grows with shrinking ⇠0 as the number of spins in the correlation
volume V⇠ becomes smaller.
One important result is the fact that the dimension dc = 4 has been identified as an upper
critical dimension. For d > dc the mean field exponents are valid. This does, however, not mean
that the mean field approximation delivers the exact thermodynamics.

6.5 Self-consistent field approximation

Fluctuation e↵ects a↵ect not only the critical exponents, they also alter the condition for the
instability. The mean field approximation overestimates the tendency to order by ignoring
fluctuations. The functional integral formulation based on the Gaussian transformation allows
us to tackle the discussion of fluctuation e↵ects in a systematic way. A good insight into some
aspects of the fluctuation e↵ects is obtained by the so-called self-consistent field approximation
which is used to analyze the instability condition. We consider the fluctuations around the mean
order parameter value by looking at the partition function in Eq.(6.44)
Z +1 Y !
Z=C d i e S( i ,Hi ) . (6.107)
1 i

Our aim in the following will be to examine the instability condition of the paramagnetic state
in the absence of a magnetic field, i.e. Hi = 0. Including thermal fluctuations we find that the
mean square of the field i is finite above the transition temperature, while the simple mean
value vanishes, i.e. h 2i i =
6 0 while h i i = 0.

6.5.1 Renormalization of the critical temperature

We take now advantage of the fact that in the paramagnetic phase only small values of i are
important. Therefore, we expand to S( i ) to second order in the fields i in the following way,
8 9
1 <X X ⌧ @ 2 ln[2 cosh( s i ] =
S( ) ⇡ S0 (J 1 )ij i j + kB T i j
2: @ i@ j ;
i,j i,j

⇢ ⌧
1X 1 2 1
= S0 (J )ij + s ij i j (6.108)
2 2
i,j
cosh ( s i )

1X 1
⇡ S0 (J )ij + s2 ij (1
2 2
s h 2
ii i j .
2
i,j

We take h 2i =h 2i independent of position. Now we turn to Fourier space,

i
1 X q ·~
i~ ri ⇤
i = q~ e with q~ = q~ , (6.109)
N
q~

and we obtain
⇢
1 X 1
S( ) = S0 + s2 (1 2 2
s h 2
i) q~ q~
2N J(~q )
q~

⇢
1 X q 2 a2
⇡ S0 + +1 Jzs2 + Jzs4 3
h 2
i q~ q~ (6.110)
2JzN z
q~

ad X
= S0 + q 2 + A + 3JzBs6 3
h 2
i q~ q~
2J 2 z 2 N
q~

110
A slightly more compact form is reached by taking the relation q~ = Jzmq~ following Eq.(6.53)
and kB Tc = Jzs2 such that
ad X 1X
S = S0 + q 2 + A + 3Bhm2 i mq~m q~ = S0 + G 1
(~q )mq~m q~ (6.111)
2N 2
q~ q~

taking the coefficients from the above Ginzburg-Landau formulation.

Now we may rewrite the partition function in this approximation as
Y0 Z
0
Z = Z0 dmq~ dm q~ exp G 1 (~q )mq~m q~/2
q~
(6.112)
Y0 Z n
2 2
o
= Z0 dm0q~ dm00q~ exp G 1
(~q )(m0q~ + m00 q~ )/2
q~
Q
where we used the parametrization m±~q = m0q~ ± im00q~ .QThe product 0 runs only over the half-
space of ~q , e.g. {~q | qz 0},
Q because the full product q~ dmq~dm q~ would constitute a doubling
of integrals compared to i d i . Using this we calculate the mean value
Z
1 1 X
hm i = d dd r hm(~r)2 i = 2
2 hmq~m q~i
L N
q~
(6.113)
1 X 02 2 kB T X 1 X kB T
= 2 hmq~ + m00q~ i = 2
G(~q ) = d
N N L A + 3Bhm2 i + q 2
q~ q~ q~

which represents a self-consistent equation for hm2 i. Note that the G(~q ) corresponds to a
renormalized correlation function, analogous to q~ in (6.65).
On the other hand, we can determine the susceptibility using the fluctuations,
Z
1
(T ) = dd r dd r0 hm(~r)m(~r 0 )i hm(~r)ihm(~r 0 )i
Ld
(6.114)
1
= hm2q~=0 i = G(~q = 0) = .
A + 3Bhm2 i
The fact that the susceptibility diverges at the phase transition can now be used to determine
the instability temperature Tc⇤ which is di↵erent from the ”bare” Tc defining the zero of A. Thus
we analyze the equation

1 3BkB T X 1
= [A + 3Bhm2 i] = A + d 1
, (6.115)
L + q 2
q~

where 1 = 0 determines T ⇤ . For the sum over ~q we have to be careful because the form we
c
have chosen for the expansion in small powers of q is not valid at large q. We correct for this
by introducing an upper cuto↵ ⇤Pfor q, Rwhich is of the order a 1 , the inverse lattice constant.
Then going to integral form L d q~ ! dd q/(2⇡)d and setting 1 = 0 we reach at equation

✓ ⇤ ◆ Z
d Tc Jza d Cd kB Tc⇤ ⇤ qd 1
Ac = Jza 1 = dq (6.116)
Tc s2 (2⇡)d 0 q 2
R
where Cd = d⌦ is the surface area of a unit-sphere in d dimensions as defined in Eq.(2.39).
Thus the renormalized transition temperature is
Tc
Tc⇤ = < Tc . (6.117)
Cd z (⇤a)d 2
1+
(2⇡)d d 2

111
Therefore the transition temperature is reduced by the fluctuations. The dimension d = dL = 2
appears as a lower critical dimension. For dimension d  dL the integral diverges at the lower
boundary (infrared divergence) such that no finite transition temperature is possible. The
dimension dL is more subtle. Below we will comment from a di↵erent point of view on this
result.8
The renormalization of Tc in Eq.(6.117) contains the fluctuation-induced correction term,
(⇤a)d 2 2d⇡ d/2 d 1 1 d!1
f racCd z(2⇡)d ⇡ ! ! 0 (6.118)
d 2 (2⇡)d (1 + d/2)(d 2) 2d ⇡ d/2 (d/2)!
such that Tc⇤ ! Tc for infinity dimensions. This means that fluctuations do not a↵ect the mean
field solution at all in infinite dimension. Each site is coupled to a reservoir of infinite neighbours,
making fluctuation negligible.9

6.5.2 Renormalized critical exponents

Now we turn to the behavior of the susceptibility at the new critical temperature. For this
purpose we rewrite (6.115) with the help of (6.116) and obtain for T ! Tc+⇤

Z 
1 = (A 3BCd ⇤ d 1 kB T kB Tc⇤
Ac ) + d
dq q 1 2
(2⇡) 0 + q q 2
Z ⇤
3BCd kB Tc⇤ qd 3
⇡ (A Ac ) dq (6.119)
(2⇡)d  0 1 + q 2
Z p
⇤ 
3BCd kB Tc⇤ xd 3
= (A Ac ) { }(2 d)/2
dx .
(2⇡)d  0 1 + x2
Note that  = ⇠ 2 . We distinguish now two cases. First choose the dimension larger than the
upper critical dimension d > dc = 4 and obtain
3BCd kB Tc⇤ d)/2 {⇤( )1/2 }d 4
1 ⇡ (A Ac ) { }(2
(2⇡)d  d 4

3BCd kB Tc⇤ ⇤d 4 kB Cd z 2 Tc⇤ (⇤a)d 4

= (A Ac ) 1
= (T Tc⇤ ) 1
(2⇡)d 2 d 4 ad s2 2(2⇡)d Tc d 4 (6.120)

⇢ 1
ad s 2 Cd z 2 Tc⇤ (⇤a)d 4
) (T ) = 1+ / |T Tc⇤ | 1
.
kB (T Tc⇤ ) 2(2⇡)d Tc d 4
We observe that this corresponds to the behavior found in mean field calculation. Indeed above
the critical dimension dc the mean field exponents are correct, and the only e↵ect of fluctuations
is to renormalize certain quantities such as the critical temperature. Obviously for d ! 1 we
recover the mean field result as exact with the same arguments as in the last section.
Next we consider d < dc = 4. In this case the integral over x in (6.119) converges and the upper
bound does not enter in the limit T ! Tc⇤ . The integral only depends on d and is given by
Kd = [(d 2)/2] [(4 d)/2]. Therefore, we obtain
1 3BCd kB Tc⇤ (2 d)/2
= (A Ac ) Kd . (6.121)
(2⇡)d d/2
8
Actually for the Ising model this approach overestimates the e↵ect of fluctuations. As we will show in
Sect.6.6.2 there is long-range order at finite temperatures for d = 2. However, we have seen in Sect.4.7.3 that
long-range order of a solid (crystalline lattice) is destroyed at any finite temperature for d = 1 and 2. The same
is true for quantum spin systems (Heisenberg model).
9
Note that for practical reason one assumes that the coupling scales so that Jz = const in order to obtain a
non-divergent mean field Tc .

112
 χ

−γ
|T − Tc|
γ=1
mean field
mean field

γ=1 γ=1
critical

Tc T
Figure 6.7: Critical region.

This equation has two regimes. For large the second term on the right hand side is dominating
over 1 . On the other hand, if is small, 1 is dominant. The first case applies for T very
⇤
close to Tc such that we arrive at the behavior
2
(T ) / |T Tc⇤ | with = . (6.122)
d 2
Away from Tc⇤ the second case holds and we return back to the usual mean field behavior,

(T ) / |T Tc⇤ | 1
. (6.123)

The crossover happens roughly at the boundary defined by the Ginzburg criterion. It is now
also obvious that the critical exponent depends on the dimension. The critical dimensions
dL = 2 and dc = 4 need a special care in this discussion as the expression we have obtained do
not straightforwardly apply to them.
We can now also use this result to determine the exponent of the correlation length ⌫. We may
use ⇠ 2 =  such that we find immediately

⌫= . (6.124)
2

6.6 Long-range order - Peierls’ argument

We have seen in the previous section that the critical temperature is reduced from its mean field
value by thermal fluctuations. The lower the dimension (the coordination number) the more
severe thermal fluctuations act against order. We consider here the case of the Ising model in one
and two dimensions. In one dimension no magnetic order is possible at any finite temperature,
while in two dimensions there is a finite-temperature phase transition.

6.6.1 Absence of finite-temperature phase transition in the 1D Ising model

We consider the ground state of the ferromagnetic Ising chain. All spins are aligned. The lowest
energy excitation is a domain wall for which the spins point up on the left and down on the
right hand side of the domain wall. The energy for such a wall is 2Js2 . With N spins on the
Ising chain, there would be N 1 positions possible for this wall and would yield an entropy
S = kB ln(N 1) accounting for the uncertainty of the configuration of the state. Therefore a
simple estimate of the free energy for the presence of a domain wall leads to

F = 2Js2 kB T ln(N 1) . (6.125)

113
Taking now the thermodynamic limit N ! 1 would yield for any finite temperature a negative
F such that the ordered phase (without domain wall) would be unfavorable.

domain
wall

Figure 6.8: Domain wall as the lowest excitation.

In the one-dimensional Ising chain long-range order exists only at zero temperature analogously
to the classical spin chain discussed earlier.

6.6.2 Long-range order in the 2D Ising model

The argument given by Peierls to proof long-range order for the 2D Ising model is also based on
domains. We consider a square lattice and assume that all spins at the boundary are pointing
up. This requirement is not relevant, if the thermodynamic limit is considered, i.e. the number
of site or spins N ! 1. However, it implies that all domain walls on the lattice are closed. The
energy cost for a domain wall of the length L compared to the completely aligned configuration
is
E(L) = 2Js2 L . (6.126)
Now choose a site j somewhere in the interior of the system. We define P± as the probability
that this spin sj is ±s. Thus the mean value of the spin is

hsj i = s(P+ P ) (6.127)

where in the case of long range order hsj i > 0 due to the boundary condition. Moreover the
correlation function satisfies for |i j| ! 1,

hsi sj i ! hsi ihsj i = s2 (P+ P )2 (6.128)

which is finite for the long-range ordered system, if P+ 6= P . Thus our proof has to address
this point.
We consider the system at low temperature such that the correlation length is long and di↵erent
states are most e↵ectively described by configurations of domain walls which are rather dilute.
The partition function is then given by
X
Z= e KLc K = 2 Js2 (6.129)
c2

where is the set of configurations and Lc is the total length of all domain walls of configuration
c.If the spin sj = +s ( s) then there is an even (odd) number of domain walls encircling the
site j. Thus the probability that the spin sj = ±s is given by

1 X KLc
P± = e (6.130)
Z ±
c2 j

114
Now we consider P whose sum over c includes one innermost domain wall encircling site j.
Thus we may rewrite
1 X KL
X
KLc
X
KL
P = g(L)e e = g(L)e P+ (6.131)
Z +
L c2 L
j

where g(L) is the number of shapes of the innermost domain wall of length L. Note that + j
contains all configurations of domain walls including an even number of walls encircling j and
all others not encircling j.

Figure 6.9: Domain wall configurations.

We now try to find an upper bound for g(L). The domain wall should not cut itself. If we ignore
this condition and in addition the requirement that the domain wall is closed, we can derive a
simple upper bound for g(L),
✓ ◆2
1 1 L L2 L ln 3
g(L) < 4 ⇥ 3L 1
⇥ ⇥ ⇥ = e , (6.132)
L 2 4 24L

which can be understood as a walk of length L through the lattice. From the starting point,
we may start in 4 directions and every following step has 3 possible directions. On the contour
there are L equivalent starting points (assuming close contour) and 2 directions. Then there are
maximally (L/4)2 sites enclosed by the domain and the starting point may be shift on any of
them (for given circumference the square has the largest area among all rectangles). Therefore
we can write
X L2 X L2
P < eL(ln 3 K) P+ < eL(ln 3 K) (6.133)
24L 96
L 4 L=4,6,8,...

since P+ < 1. For sufficiently low temperature ln 3 K = ln 3 2Js2 < 0 so that the sum
converges very well. By making the temperature small enough, also the condition P < 12 can
be reached. With the condition P+ + P = 1 it follows that

P+ P >0 ) lim hsi sj i = const. (6.134)

|i j|!1

which means that we have long- range order at a finite temperature.

115
Chapter 7

Superfluidity

Only the nobel gas Helium remains a liquid down to zero temperature at ambient pressure. After
liquefaction it is a quantum liquid which involves two isotopes, with bosonic (fermionic) character
for 4 He (3 He). Note that the nucleus of 4 He consists of two protons and two neutrons, resulting
in a Boson, while 3 He has only one neutron and forms a fermionic atom. Only under pressure
eventually solidification occurs (Fig. 7.1). Both quantum liquids undergo a phase transition to
the miraculous state of superfluidity, characterized by the frictionless or dissipationless flow of
the fluid through constrictions and narrow capillaries, beside a few other amazing properties.
Considering their phase diagrams in Fig.7.1, we recognize a profound di↵erence for the superfluid
phase between 4 He (left panel) and 3 He (right panel). The bosonic atoms undergo a Bose-
Einstein condensation to become a superfluid below T ⇡ 2.18K at ambient pressure. The
fermionic atoms, on the other hand, become superfluid only around 10 3 K, when they form so-
called Cooper pairs, in many aspects resembling the superconducting phase of electrons. Here
we will discuss only the case of superfluid 4 He.
40$ 40%
solid%
solid$
30$ 30% A#phase%
Pressure%[bar]%
pressure$[bar]$

20$ normal$$ 20% B#phase% normal%

λ>line$

liquid$ Fermi%liquid%
superfluid%
10$ superfluid$ 10% liquid%

gas$ 0%
gas%
0$
0$ 1.0$ 2.0$ 3.0$ 4.0$ 5.0$ 10#4% 10#3% 10#2% 10#1% 1% 3.3% 10%
Temperature$[K]$ Temperature%[K]%

Figure 7.1: Phase diagrams of Helium. Left panel: bosonic 4 He. Right panel: fermionic 3 He.
Note the logarithmic temperature scale in the right panel. While 4 He has one superfluid phase,
there are two, the A- and B-phase, for 3 He.

7.1 Quantum liquid Helium

We consider now the quantum liquid 4 He.1 The first question we ask is why Helium remains
liquid even in its ground state. A classical system of interacting particles would form a solid
1
The quantum liquid 3 He is a so-called Fermi liquid well described by Landau’s phenomenological theory based
on the concept of fermionic quasiparticles. This concept can be extended to other interacting Fermions, most
notably to electrons in metals, as will be discussed in the course ”Solid State Theory”.

116
lattice in order to minimize the potential energy, as the kinetic energy vanishes at zero temper-
ature. This is not the case when we consider quantum particles which are subject to zero-point
motion. Let us consider here a simple argument on why Helium is liquid while other noble gases
solidify.
The interaction between Helium atoms consists of two components. There is a strongly repulsive
interaction on short range, since the closed electron shells (two 1s electrons) repel each other
when the atoms come close. On long distances the atoms attract each other through the rather
weak van der Waals force. A good fit of the potential is given by2
⇣ r ⌘6
2
V (r) = Ae r/r1 B (7.1)
r
with A = 489eV, B = 9.3 ⇥ 10 5 eV, r1 = 0.22 Å and r2 = 4.64 Å. Qualitatively this resembles
the Lenard-Jones potential we introduced earlier, with a negative minimum and a vanishing
potential for large distances r. It has a minimum at r0 ⇡ 3Å with a potential energy V =
7.8 ⇥ 10 4 eV. The mean distance between Helium atoms towards zero temperature at ambient
pressure is d ⇡ 4.4Å. Assuming that Helium is a solid with this lattice constant, the condition
for melting is that position fluctuations of the atoms acquire a ”considerable” fraction of the
lattice spacing d. The Lindemann criterion for melting request the mean displacement of the
atom from its equilibrium position,
r
 Lm with Lm ⇡ 0.1 , (7.2)
d
such that for the solid the atom would have to be confined with a range of linear size r ⇡ 0.5Å.
Taking Heisenberg’s uncertainty principle, r ⇥ p ⇡ ~ we obtain for the zero-point energy,
✓ ◆
p2 1 ~ 2
E= ⇡ ⇡ 8.6 ⇥ 10 4 eV , (7.3)
2m 2m r
which is slightly larger than the potential depth. Thus, the Helium atoms dissociate due to
the zero-point motion. Only pressure beyond 25 bar opens the way to solidifications at low
temperatures (T < 5K). We encounter ”quantum melting” at very low temperatures, when we
lower the pressure isothermally through the critical pressure pc ⇠ 25bar (see also in Sect. 4.7.3).
Classical thermally driven melting is observed for isobar increase of temperature from the solid
phase with the pressure clearly above 25 bar.

7.1.1 Superfluid phase

Around T ⇡ 2.17K the quantum liquid undergoes a phase transition between a normal and
superfluid phase. This transition is called ” -transition” due to the peculiar shape of the specific
heat as a function of temperature. (Fig.7.2). There is an obvious similarity with the specific heat
of Bosons as shown in Fig.3.6, which suggest that the superfluid phase is based on Bose-Einstein
condensation. There are, however, also clear di↵erences such as the low-temperature behavior
of the specific heat, which has a power law T 3/2 for the ideal Bose gas and T 3 for 4 He. This is
connected with the di↵erence in the excitation spectrum.
We examine here the property of a superfluid to flow frictionless. First we ask the question
whether the ideal Bose-Einstein condensate is a superfluid. For this purpose we study an ideal
Bose gas of particles flowing in a capillary tube with a velocity ~v relative to the capillary
walls. We can view the same situation in the rest frame of the superfluid, such that the tube is
moving with velocity ~v . The former inertial system we call K 0 , the latter K. Then the following
relations for the momentum and energy of the superfluid hold,
K: P~ E
(7.4)
M
K0 : P~ 0 = P~ M~v E0 = E P~ · ~v + ~v 2
2
2
Reference: J.C. Slater and J.G. Kirkwood, Phys. Rev. 37, 682 (1931).

117
 C"

3kB"
2"

T"3"
0"
0" Tλ" T"

Figure 7.2: Specific heat at the superfluid transition T .

where M is the total mass of all particles of the superfluid. Initially all particles have the same
velocity ~v such that P~ = 0 and E = 0 in system K (rest frame of superfluid). Let us assume one
particle scatters with the wall and transfers the momentum p~ to the wall. Such an excitation of
the superfluid costs the energy ✏(~
p), viewed in system K. For this situation we obtain E = ✏(~ p)
and P~ = p~. Thus, the energy di↵erence between the initial and the final state in the rest frame
of the tube, K 0 , is given by
E 0 = ✏(~
p) + p~ · ~v . (7.5)
The minimal energy drop is obtained by choosing p~ antiparallel to ~v , such that the superfluid
overall slows down. The scattering process can only happen, if the energy of the superfluid is
reduced, E 0 < 0. For an ideal Bose gas the energy of an excitation is given by ✏(~ p) = p~2 /2m.
There is always a momentum p~ with

p~2 p2
E0 = + p~ · ~v = vp < 0 ) p < 2mv . (7.6)
2m 2m
Consequently the condensed ideal Bose gas is not a superfluid and is subject to friction, as we
can always find a scattering channel which slows the superfluid down by transferring momentum
to the capillary.
Real Bose gases have interaction. This leads in the superfluid phase to a di↵erent form of
elementary excitations with a linear dispersion,

p) = cs |~
✏(~ p| = c s p , (7.7)

corresponding to a sound-like excitation. Inserting this in Eq.(7.6) we find that the inequality
can only be satisfied for v cs . Therefore there is a critical velocity vc below which the fluid
flows frictionless (dissipationless).
The real excitation spectrum of the superfluid has a more complex structure as shown in Fig. 7.3.
At small energy and momentum the spectrum is linear as assumed above. However, with growing
momentum it turns non-monotonic with a local minimum around p0 . This part of the spectrum
is called roton. For these two parts of ✏(p) we write
8
>
> c p p ⌧ p0
< s
✏(p) = (7.8)
>
> (p p0 )2
: + p ⇡ p0
2m⇤
where for 4 He the parameters are
p0 1
cs = 240 ms 1
, = 1.9 Å , = 8.7 K , m⇤ = 0.16 mHe . (7.9)
~ kB

118
The critical velocity is now given by the lowest possible excitation within Eq.(7.5), corresponding
to the dashed line in Fig.7.3,
✏(p0 )
vc ⇡ = = 60 ms 1 . (7.10)
p0 p0
Note that the roton minimum is a precursor of solid He. Under pressure shrinks and eventually

✏(p)
roton%

phonon% vc p

p0 p
Figure 7.3: Spectrum of a real superfluid with a linear sound-like behavior at low energy and
a local minimum around p0 , the rotons. The dashed line (vc p) indicates the maximal velocity
of the superfluid to sustain frictionless flow, i.e. flow without generating elementary excitations
which would lead to a slow down.

reaches zero at a critical pressure where the rotons condense forming a coherent state with specific
momentum corresponding to a crystalline (hexagonal closed packed) lattice.

7.1.2 Collective excitations - Bogolyubov theory

In this section we discuss the low-energy excitations of a Bose gas of weakly interacting particles.
For this purpose we follow the theory developed by Bogolyubov. We write the Hamiltonian in
the second-quantization language for spinless Bosons,
H = Hkin + Hint (7.11)
with X
Hkin = (✏~k µ)â~† â~k
k
~k
(7.12)
Z
1
Hint = d3 r d3 r0 b † (~r) b † (~r 0 )V (~r ~r 0 ) b (~r 0 ) b (~r)
2
where ✏~k = ~2~k 2 /2m and V (~r ~r 0 ) = U (~r ~r 0 ) denotes a contact interaction, i.e. the particles
repel each other when they are on the same position. We transform the interaction part into
the occupation number representations using Eq.(4.35), using the relation,
X
b (~r) = p1 ~
â~ eik·~r (7.13)
V ~ k
k

and obtain,
U X †
Hint = â~ â~† 0 â 0 â . (7.14)
2V k+~
q k q~ ~k ~k
~k,~k 0 ,~
q

119
 †
Now we turn p to the Bogolyubov approximation introduced in Sect. 4.7.2, by identifying the â0
and â0 by N0 (phase = 0 for simplicity). This is justified when N0 N N0 . We can
then approximate the Hamiltonian by keeping only the leading terms in N0 obtained through
the following combinations of momenta:

~k, ~k 0 , ~q â~† ↠â 0 â ~k, ~k 0 , ~q â~† ↠â 0 â

q ~k 0 q~ ~k ~k
k+~ q ~k 0 q~ ~k ~k
k+~

~k = ~k 0 = ~q = 0 N02 ~k = ~k 0 = 0, ~q 6= 0 N0 â†q~↠q~

~q = ~k 6= 0, ~k 0 = 0 N0 â~† â~k ~q = ~k 0 6= 0, ~k = 0 N0 â~† 0 â~k 0

k k

~k = ~k 0 = ~q 6= 0 N0 â~k â ~k ~q = ~k = 0, ~k 0 6= 0 N0 â~† 0 â~k 0

k

~q = ~k 0 = 0, ~k 6= 0 N0 â~† â~k
k

3/2 1/2
Terms of order N0 do not exist and terms of order N0 and lower in N0 are neglected.
Collecting these terms we obtain to this order,
⇢
U N02 X U N0
H0 = µN0 + + ✏~k µ + â~† â~k
2V V k
~k6=0
(7.15)
U N0 X n † † o
+ â~ â ~ + â ~k â~k + â~† â~k + ↠~ â ~k .
2V k k k k
~k6=0

The requirement that the total particle number remains constant on average at lowest temper-
ature leads to µ = U n0 (derivation is somewhat tedious and omitted here),
U n0 1 X h i
H 0 = N0 + ✏~k + U n0 (â~† â~k + ↠~ â ~k ) + U n0 (â~† ↠~ + â ~k â~k ) . (7.16)
2 2 k k k k
~k6=0

This Hamiltonian has a single-particle form, but does not conserve the number of particles, as
terms like â ~k â~k and â~† ↠~ describe how particles from excited states are absorbed by the con-
k k
densate and emitted from the condensate, respectively. We search a set of bosonic annihilation
and creation operators which diagonalize this Hamiltonian into a form
1X
H0 = E~k (ˆ~† ˆ~k + ˆ † ~ ˆ ~k ) + E0 µN0 . (7.17)
2 k k
~k6=0

where the operators ˆ~† (ˆ~k ) create (annihilate) independent Bogolyubov quasiparticles. The
k
following Bogolyubov transformation allows us to reach this form,3

â~k = ˆ~k u~k ˆ † ~ v~k and â ~k =ˆ ~k u~k ˆ~† v~k , (7.23)

k k
3
Bogolyubov transformation: We insert the transformation (7.23) into the second term of the Hamiltonian
(7.16) and obtain for a fixed ~k,
n o
(✏~k + U n0 ) (ˆ~k† u~k ˆ ~k v~k )(ˆ~k u~k ˆ † ~k v~k ) + (ˆ † ~k u~k ˆ~k v~k )(ˆ ~k u~k ˆ~k† v~k )
n o (7.18)
+U n0 (ˆ~k† u~k ˆ k v~
~
†
k )(ˆ ~ u~
k k
ˆ~k v~k ) + (ˆ k u~
~ k ˆ~k† v~k )(ˆ~k u~k ˆ † ~k v~k )

120
where the new operators satisfy the relation,

[ˆ~k , ˆ~† ] = [ˆ †
~k , ˆ ~k ] =1 ) [â~k , â~† ] = u~2k v~k2 = 1 . (7.24)
k k

The diagonalization leads to

q 1 X⇥ ⇤ U n0
E~k = ✏~2 + 2U n0 ✏~k and E0 = E~k ✏~k U n0 N0 (7.25)
k 2 2
~k6=0

and the transformation coefficients,

9 8 s✓
1 > ◆2
u~k = q >
> >
> ✏~ ✏~
2 >
> >
> =1+ k 1+ k 1
1 ~k >
= >
< ~k
U n0 U n0
) (7.26)
>
> >
>
v~k = q
~k >
> >
> ✏~k E~k
2 >
> >
: =1+ .
1 ; U n0 U n0
~k

The spectrum of the elementary excitations is linear in the small momentum limit, ~k ! 0,
r
U n0
E~k ⇡ cs p = ~k , (7.27)
m

corresponding to a phonon spectrum of compressive waves. Thus, for small ~k the Bogolyubov
quasiparticles represent phonons. In the large momentum region, ✏~k U n0 the spectrum merges
gradually with ✏~k ,
E~k ⇡ ✏~k + U n0 (7.28)
with a shift of U n0 (see Fig.7.4). In E~k rotons are not included as they are beyond the simple
approximation applied here. Therefore, within the Bogolyubov theory the critical velocity is
determined by the slope of the phonon spectrum, i.e. by the speed of sound,
r
U n0
vc = cs = (7.29)
m
which is higher than the real vc discussed above. The constant E0 is the ground state energy.
where we assume that u~k and v~k are real. We now separate the terms to compare with Eq.(7.17),

(ˆ~k† ˆ~k + ˆ † ~k ˆ k )[(✏~

~ k + U n0 )(u~2k + v~k2 ) 2U n0 u~k v~k ] + 2(✏~k + U n0 )v~k2 2U n0 u~k v~k
(7.19)
+(ˆ~k† ˆ † ~k + ˆ k ˆ~
~
2
k )[U n0 (u~
k
+ v~k2 ) 2(✏~k + U n0 )u~k v~k ] .

With the condition that u~2k v~k2 = 1 from Eq.(7.24) we parametrize u~k = cosh ✓~k and v~k = sinh ✓~k . In order to
get the terms in the second line of Eq.(7.19) to vanish we find then
U n0
0 = U n0 cosh 2✓~k (✏~k + U n0 ) sinh 2✓~k ) tanh 2✓~k = (7.20)
✏~k + U n0

which leads after using some identities of hyperbolic functions to u~k and v~k in Eq.(7.26). Moreover by comparison
with Eq.(7.17) we find
✏~k + U n0 U n0 tanh 2✓~k q
E~k = (✏~k + U n0 ) cosh 2✓~k U n0 sinh 2✓~k = q q = ✏~2k + 2✏~k U n0 . (7.21)
1 tanh2 2✓~k 1 tanh2 2✓~k

Finally the constant term

2(✏~k + U n0 )v~k2 2U n0 u~k v~k = (✏~k + U n0 ) cosh 2✓~k + U n0 sinh 2✓~k ✏~k + U n0 = E~k ✏~k U n0 (7.22)

which yields the term E0 in Eq.(7.17).

121
 5

energy&
U n0
4

3
Ek

phonon& ✏k
1

0.5 1.0 1.5 2.0

|~k|
Figure 7.4: Spectrum of the Bogolyubov quasiparticles in Eqs.(7.17) and (7.25). The solid line
displays E~k with a linear dispersion for small momenta p~ = ~~k and the dashed line shows ✏~k .
There are no rotons in this simple approach.

Note that in the approximation done here the sum in E0 diverges at large ~k. This is an artefact
of taking the potential as an infinitely sharp contact interaction, i.e. U is independent of ~k. 4
Now we turn to the condensate fraction of the interacting system. How many particles condense?
This can be straightforwardly calculated,
1 X †
n0 = n hâ~ â~k i . (7.32)
V k
~k6=0

This is advantageous when we want to calculate now mean values for which we insert the
Bogolyubov transformation in Eq.(7.23),

hâ~† â~k i = h(ˆ~† u~⇤k ˆ ⇤

~k v~k )(ˆ~k u~k ˆ † ~ v~k )i
k k k
(7.33)
= |u~k | 2
hˆ~† ˆ~k i + |v~k | hˆ 2 †
~k ˆ ~k i u~⇤k v~k hˆ~† ˆ † ~ i u~k v~k⇤ hˆ ~k ˆ~k i .
k k k

Then we use the fact that the Bogolyubov quasiparticles are independent, described by the
Hamiltonian in Eq.(7.17),
1 1
hˆ~† ˆ~k i = E~k
, hˆ †
~k ˆ ~k i = 1 + hˆ † ~ ˆ ~k i =1+ E~k
,
k e 1 k e 1
(7.34)
hˆ ~k ˆ~k i = hˆ~† ˆ † ~ i =0.
k k
4
Groundstate energy: It is rather easy to see that in the large-~k limit the integrand in E0 (Eq.(7.25)) behaves
like
m(U n0 )2
E~k , ✏~k U n0 ! (7.30)
p2
and leads to a divergence of the sum. A careful and more extensive analysis of the problem shows that this
singular part can be extracted such that the renormalised ground state energy is given by
" ◆3/2 #
0
X m(U n0 )2 U n0 U n0 16
✓
U n0 m
E0 = E~k ✏~k U n0 + + N0 = N0 1+ . (7.31)
p2 2 2 15⇡ 2 n0 ~2
~
k6=0

The approximation for the ground state energy used here relies on the assumption that we are dealing with a
Bose gas. This is, however, not guaranteed even for weak interactions.

122
Note that E~k = E ~k . The momentum distribution of the Bosons for ~k 6= 0 is obviously,

1+ 2 2
1 ~k 1 ~k
n~k = hâ~† â~k i 2
= (|u~k | + |v~k | )2
E~k
2
+ |v~k | = 2 E~k
+ 2 . (7.35)
k e 1 1 ~k e 1 1 ~k

The first term vanishes at T = 0 for all ~k 6= 0, while the second remains finite. Thus, unlike
for the ideal Bose gas for which all particles condense into the ~k = 0 state, here we have an
interaction induced reduction. Let us discuss this aspect. The zero-temperture distribution
function has the following behavior,
8 p
>
> mU n0 p
>
> ~k ⌧ 2mU n0
2 < 2~k
~k
hâ~† â~k iT =0 = 2 ⇡ (7.36)
k 1 >
> (mU n ) 2 p
~k >
> 0
: ~k 2mU n0
4(~k)4
which represents the depletion of the condensate due to particle-particle interaction. We calcu-
late now the total reduction of n0 at T = 0 integrating over all ~k,
2 Z 2 ✓ ◆
1 X ~k d3 k ~k 1 mU n0 3/2
n0 = n = n = n (7.37)
V 1 2
~ (2⇡)3 1 2
~ 3⇡ 2 ~2
~k6=0 k k

which is valid only for sufficient weak interactions. This can be measured in terms of the
characteristic length ⇠, defined as
~2 k 02 ~2
= U n0 with k 0 ⇠ = 1 ) ⇠2 = (7.38)
2m 2mU n0
which leads to the condition
✓ ◆3/2
2 mU n0
3⇡ n ) n⇠ 3 1. (7.39)
~2
The characteristic length is much larger than the mean distance between particles.
If we apply the same scheme to a one-dimensional interacting Bose gas, we find that the integral
Z 2
dk ~k
2 ! 1 (7.40)
2⇡ 1 ~ k

diverges due to the k 1 dependence of the integrand for small k (see Eq.(7.36). This leads to the
conclusion that in one dimension quantum fluctuations destroy the Bose-Einstein condensate,
as soon as interaction is present.
The phonon spectrum determines the low-temperature behavior of the superfluid. Let us con-
sider the heat capacity as an example. For this purpose we calculate the low-temperature internal
energy through the Hamiltonian H0 in Eq.(7.17),
X X ✓ ◆3/2
E~k ⇡2 m
U = hH0 i = E0 µN0 + E~k hˆ~† ˆ~k i = E0 µN0 + ⇡ N (kB T )4
k e E~k
1 30n U n 0 ~ 2
~ k6=0 ~ k6=0
(7.41)
for kB T ⌧ U n0 . Note that in this limit we look at Bosons with a linear energy-momentum dis-
persion, like photons or phonons which in three dimensions yield U / T 4 . The low-temperature
heat capacity is then accordingly
✓ ◆3/2
dU 2⇡ 2 m
C= =N kB (kB T )3 (7.42)
dT 15n U n0 ~2
which is di↵erent from the Bose-Einstein condensate with C / T 3/2 .

123
7.1.3 Gross-Pitaevskii equations
We would like to introduce now a description of the Bose-Einstein condensate which allows
us to take also spatial variations into account. For this purpose we write the Hamiltonian of
interacting Bosons entirely in the language of field operators,
Z ⇢ 2
3 ~ ~ b† ~ b (~r)) + [V (~r) µ] b † (~r) b (~r)
H = d r (r (~r))(r
2m
Z (7.43)
1
+ d3 r d3 r0 b † (~r) b † (~r 0 ))U (~r ~r 0 ) b (~r 0 ) b (~r)
2

where V (~r) is a potential (shifted by the chemical potential µ) acting on the Bosons which are
interacting by a contact interaction. We derive now the equation of motion for the field operator,
" #
@ b ~ ~2
2r
i~ (~r, t) = [ b (~r, t), H] = + V (~r) µ + U b † (~r, t) b (~r, t) b (~r, t) . (7.44)
@t 2m

If the external potential varies very slowly or is even constant we will replace the field operator
by the Bogolyubov approximation discussed in Sect. 4.7.2 and obtain the equation
!
@ ~2
~2 r
i~ 0 (~
r, t) = + V (~r) µ + U | 0 (~r, t)|2 0 (~
r, t) , (7.45)
@t 2m

which is called Gross-Pitaevskii equation. This equation is obtained by variation from the energy
functional Z  2
3 ~ ~ U
E= d r |r 0 (~r)|2 + {V (~r) µ}| 0 (~r)|2 + | 0 (~r)|4 (7.46)
2m 2
through
@ E
i~ 0 (~
r, t) = . (7.47)
@t 0 (~
r, t)
We have also introduced the chemical potential µ which can be absorbed in the potential V (~r).
p
In the homogeneous equilibrium (time independent) phase we require that 0 (~r) = n0 and
fix V (~r) = 0. In order to get the right magnitude for | 0 | we have to choose µ = U n0 as in
Eq.(7.15).

Spatial variation:
Let us consider here some properties. First we address the characteristic length scale for the
recovery of the equilibrium phase near a local perturbation. For this purpose we assume a small
deviation,
p
0 (~
r) = n0 + ⌘(~r) , (7.48)
with |⌘(~r)|2 ⌧ n0 . Inserted into the Gross-Pitaevskii equation (Eq.(7.45)) we obtain to linear
order in ⌘(~r),
~2 ~ 2 p
r ⌘(~r) + 2U n0 ⌘(~r) = V (~r)( n0 + ⌘(~r)) . (7.49)
2m
We now consider point-like defect, V (~r) = V0 (~r) which yields the following long distance (r ⇠)
behavior, p
e r/ 2⇠ ~2
⌘(~r) = ⌘0 with ⇠2 = (7.50)
r 2mU n0
p
the same length as introduced in Eq.(7.38) and ⌘0 ⇡ mV0 n0 /2⇡~2 .

124
 (x)
V (x)

µ0

x
Figure 7.5: Bose-Einstein condensate trapped in a harmonic potential V (x) / x2 . The Thomas-
Fermi approximation yields a condensate concentrated around the center of the potential well.
Note that the approximation is only justified for large condensate densities (n0 ⇠ 3 1). Thus,
near the region where the condensate density vanishes a more accurate analysis shows a smoother
vanishing than the Thomas-Fermi approximation.

For very slow variations of the potential where the kinetic part (di↵erential part) of the Gross-
Pitaevskii equation can be neglected, the solution is rather simple,
✓ ◆1/2
µ0 V (~r)
(~r) = (7.51)
U

with µ0 = U n0 the chemical potential where V (~r) = 0. This is called the Thomas-Fermi
approximation 5 . The approximation breaks down when | (~r)| approaches zero. Then the full
problem has to be solved.

Continuity equation and currents:

We multiply now the Gross-Pitaevskii equation by ⇤ r , t)
0 (~ and subtract the complex conjugate
of the resulting equation. From this we obtain,

@ ~ ⇤ ~ 2 0 (~r, t) ~2 ⇤
r, t)|2
@t | 0 (~ = [ r, t)r
0 (~ r, t)r
0 (~ 0 (~
r, t)]
2mi
(7.55)
= ~ · ~ [
r ⇤ ~ 0 (~r, t)
r, t)r r, t)r~ ⇤
0 (~ 0 (~ 0 (~
r, t)]
2mi
5
Spatial modulation: To illustrate the e↵ect of slow variation we assume a weak periodic potential V (x) =
V0 cos(qx). Inserting this potential into Eq.(7.49) we obtain the e↵ectively one-dimensional di↵erential equation,

~2 00 p
⌘ (x) + 2U n0 ⌘(x) = V0 n0 cos(qx) (7.52)
2m
where we neglect the last term on the right-hand side. We then obtain
p
V0 n0 cos(qx)
⌘(x) = . (7.53)
U n0 (⇠ 2 q 2 + 2)
”Slow variation” corresponds to ⇠q ⌧ 1 which we may now compare with the Thomas-Fermi approximation in
Eq.(7.51) with µ0 = U n0 ,
 1/2 p
V0 p V0 n 0 p
(x) = n0 cos(qx) ⇡ n0 cos(qx) = n0 + ⌘(x) (7.54)
U 2U n0
as obtained above in this limit.

125
which we may rewrite as
8 2
@⇢(~r, t) ~ ~ < ⇢(~r, t) = |
> 0 (~
r, t)|
+ r · j(~r, t) = 0 with
@t : ~j(~r, t) = ~ [
> ⇤ ~ 0 (~r, t)
r, t)r ~ 0⇤ (~r, t)]
r, t)r
0 (~ 0 (~ .
2mi
(7.56)
The current density is connected with the phase of the condensate wave function, (~r) =
p
n0 ei (~r) when the condensate density is constant.

~j(~r) = ~ n0 r
~ (~r) = n0~vs (~r) (7.57)
m
which we use also to define the superfluid velocity ~vs . The energy of the homogeneous system
(V (~r) = 0) is then given by 
m 2 U n20
E(~vs ) = V ~vs n0 (7.58)
2 2
with which we determine the critical velocity by E(vc ) = 0, i.e.
p there is no energy gain in
forming a condensate. The critical current is obtained as vc = U n0 /m = cs as found in the
Bogolyubov theory in Eq.(7.29).

Frictionless current and quantized vortices:

An important aspect of the condensate wave function is its property to be single-valued. This
gives rise to spectacular properties, for example, for the current flow in a torus. In this multiply
connected system the phase of the wave function can show a winding of the phase under the
constraint that 0 (~r) has to match on a path around the torus. Thus the phase can only
change by an integer multiple of 2⇡ (de Broglie condition). If we assume a regular torus the
current flow would be homogeneous with a constant velocity |~vs |. We find that the circular
integral of the velocity around the torus is quantized, because
I I
~ ~ · d~s = ~ 2⇡n
~vs · d~s = r (7.59)
m m
with n being an integer. Thus the current j = n0 vs can only take discrete values in the torus,
such that a decay of current would mean an unwinding of the phase . This constitutes a
frictionless flow of the superfluid as it will flow eternally.
A phase winding can occur within the superfluid also when the superfluid density forms a
singularity, a zero along a line. The solution of the Gross-Pitaevskii equations then has form
p
0 (r? , ✓, z) = n0 f (r? )ei✓n (7.60)

for a line along the z-axis with (r? , ✓) being the cylindrical coordinates perpendicular to z. The
n
function f (r? ) goes to zero on the line r? = 0 like f (r? ) / r? and recovers to 1 with the
length scale ⇠. This structure represents a vortex with winding number n and is a topological
defect in the superfluid. This means that it cannot be simply created or destroyed, but can only
enter or leave the superfluid at the boundary. Two vortices with opposite winding numbers can
annihilate each other.
The energy of a vortex consists of the core energy due to the local depletion of the condensate
and the contribution of the circular flow of superfluid. The latter, however, is larger. To be
concrete we place the vortex along the rotation axis of a cylindrical vessel of radius R. Then

126
 | (r )|2
⇠

current'

n0
vortex'core'

Figure 7.6: Phase winding and vortex: Left panel: phase winding (red line) around the torus
leads to a current. As the phase has close the current is quantized. Right panel: Vortex structure
of | (r? )|2 with a core of linear extension ⇠ ⇠.

the energy per unit length is given by

Z Z
2
Evortex = d rE[ (~r)] + d2 rE[ (~r)]
r⇠ rg⇠

Z R Z 2⇡ 2 ✓ ◆
~2 1 @ (r? , ✓) R ~2 n0
⇡ Ecore + dr? r? d✓ 2 = Ecore + , 2⇡n2 ln
⇠ 0 2m r? @✓ ⇠ 2m
(7.61)
The first part is the energy expense due to the vortex core, where the depletion of the condensate
leads to a loss of condensation energy
U n0 2
Ecore ⇡ ⇡⇠ . (7.62)
2
The second part is due to the currents and dominates the vortex energy. The amazing result is
that this latter energy scales logarithmically with the system size, i.e. radius of the cylinder in
our example, although the condensate density is only damaged on a comparatively small volume.
A further important point is that this energy grows with n2 indicating that vortex of lowest
energy has n = ±1. Therefore vortices of higher winding numbers would would separate into
such vortices to reduce the energy.

7.2 Berezinskii-Kosterlitz-Thouless transition

A peculiar behavior can be found in the two-dimensional superfluids, e.g. a 4 He film on a
substrate. How does superfluid order establish in this case? We will see that due to phase
fluctuations of the condensate wave function here a new type of phase transition appears.

7.2.1 Correlation function

First let us examine the correlation functions of the two-dimensional film. For the bosonic
system we can do this the same way as we did previously in Sect.4.7.2. The correlation function
is given
~ = h b † (~r) b (~r + R)i
g(R) ~ . (7.63)

127
In the high-temperature limit we have approximated the long-distance behavior by
Z ~ ~ Z ~ ~
~ = d2 k eik·R 2mkB T d2 k eik·R
g(R) ⇡
(2⇡)2 e (✏~k µ) 1 ~2 (2⇡)2 k 2 + k02
(7.64)
r
2 1 2⇡ k0 R
= 2
K0 (k0 R) ⇡ 2
e
k0 R
p
with k02 = 2mµ/~2 , = h/ 2⇡mkB T (thermal wavelength) and K0 (x) is the modified Bessel
function (also known as MacDonalds function). Thus, we find exponential decay of the correla-
tion.
The low-temperature limit can be conveniently discussed by taking advantage of the simple
formulation of the energy in Eq.(7.46), where we keep n0 constant and restrict to the phase
fluctuations. Then we obtain
2 3
Z  X
~ ) 2 1 = U n2 4
E[ ] = U n20 d2 r ⇠ 2 (r ⇠ 2 q 2 q~ q~
V5
(7.65)
0
2 2
q~

with the Fourier transform

1 X q ·~
i~ r
(~r) = p q~ e (7.66)
V q~
and V here as the area of the system. Now turn to the calculation of the correlation function
✓ ◆
~ ~
i( (0) (R)) ~ 2i
h( (0) (R)) ⇡R ⌘(T )
g(R) = n0 he i = n0 e = (7.67)
r0
where the last equality is connected with the fact that with the above energy the averaging is
Gaussian. 6 Here the correlation function still decays, but now algebraically with ⌘(T ) = 2/n0 2
6
Averaging: In order to calculate the correlation function in Eq.(7.67) we turn the Fourier space for the phase:
1 X q ·~
i~ r
(~r) = p ~e
q (7.68)
V q~

and rewrite the energy X X

E[ ] = U n20 ⇠ 2 q2 q
~ q
~ = U n20 ⇠ 2 q2 ( 2
1,~
q + 2
q)
2,~ (7.69)
q
~ q
~

ignoring the constant term and separating real and imaginary part, q~ = 1,~
q +i q.
2,~ The partition function is
given by Z Y
2 2P 2 2 2
Z= d 1,~q d 2,~q e U n0 ⇠ q~ q ( 1,~q + q)
2,~ (7.70)
q
~

and the correlation function can be calculated by

8 9
Z Y < =
~ = 1 i X i~ ~
q ·R
X
g(R) d 1,~q d 2,~q exp p ( 1,~q + i q )(1
2,~ e ) U n20 ⇠ 2 2
q ( 2
1,~
q + 2
q)
2,~ (7.71)
Z : V ;
q
~ q
~ q
~

which can be evaluated by completing the square.

8 9 ( )
< X ~ = Z ~
~ k B T 1 1 cos ~
q · R kB T 1 q·R
cos ~
g(R) = exp = exp d2 q . (7.72)
: U n20 ⇠ 2 V q2 ; U n20 ⇠ 2 (2⇡)2 q2
q
~

We can again use the relation connection with the Bessel function used above and obtain
( Z ⇡/r0 ) ( Z ⇡R/r0 )
~ kB T 1 J0 (qR) 1 J0 (x)
g(R) = exp dq = exp ⌘(T ) dx
U n20 ⇠ 2 2⇡ 0 q 0 x
(7.73)
⇢ ✓ ◆ ✓ ◆ ⌘(T )
⇡R ⇡R
⇡ exp ⌘(T ) ln =
r0 r0
1/3
where we introduce a cuto↵ for large q with r0 ⇡ n the mean distance between the Bosons.

128
being a function linear in T and r0 3 = n being a characteristic length. At all temperatures there
is no real long-range order, but it looks like we are for all low enough temperatures always on a
critical point, since the correlation length is obviously infinitely larger.

7.2.2 Topological excitations and BKT transition

This qualitative change of the correlation function indicates a phase transition which does not
fit into the picture of suppression of order due to collective modes of the phonon type. Phonons
are, however, not the only excitations possible. Kosterlitz and Thoulouss realized that in a two
dimensional superfluid (and also in other phases mentioned below) vortices play an important
role for this change of behavior. Vortices constitute a topological excitation.
A conceptual argument for the occurrence of transition can be derived through the free energy
of a single vortex. As we have seen above the energy of a vortex with n = 1 is given by
✓ ◆
~2 n0 R
E= 2⇡ ln (7.74)
2m ⇠
where R a linear extension of the of the two-dimensional system. The entropy of the vortex is
given through the phase space of the vortex which is given by number of distinct vortex positions
in the system ⇠ R2 /⇠ 2 (number of cells of size ⇠ 2 where we can place the vortex), leading to
S = kB ln(R/⇠)2 . Thus the free energy of the vortex is
⇢ 2 ✓ ◆
~ n0 R
F = E TS = 2⇡ 2kB T ln (7.75)
2m ⇠
which suggests a phase transition at
~2 n0
kB TKT = ⇡ (7.76)
2m
where the prefactor changes sign. Above TKT vortices are favored below they are suppressed.
Note that this argument does not depend on the system size as both contributions in F scale
the same way.
The transition can be understood in terms of vortex-anti-vortex pairs being bound or dissociated
(anti-vortex: has the same structure as the vortex but with opposite winding of the phase). The
high temperature phase corresponds to unbound free vortices destroying the phase coherence,
while at low temperatures vortices bind into pairs of opposite winding number such that there is
no phase winding of bound pairs. A vortex and anti-vortex pair behave like particle and antipar-
ticle. They can be created spontaneously at the expense of the energy of two vortices, without
changing the overall phase winding of the system. For the same reason they can annihilate each
other releasing the vortex energies. Thus the number of vortex pairs is not conserved. Only the
totel winding number of all vortices is conserved,
(tot)
X (i)
n = n . (7.77)
i

We can discuss the gas of vortices in terms of the grand canonical ensemble.
Two vortices interact with each other by an energy
~2 (i) (j) ~ri ~rj
Eij = 2 2⇡n n ln , (7.78)
2m ⇠
which is formally equivalent to the Coulomb energy of two charged particles (winding number
as charge). Note this expression is only true if the vortex cores do not overlap, i.e. |~ri ~rj | ⇠.
The grand canonical partition function is given by
2 3
1
X Z
zN 1 X ~ 2 X (i) (j) ~
ri ~
r j 5
Zv = d2 r1 · · · d2 rN exp 42 2⇡ n n ln (7.79)
[(N/2)!]2 ⇠ 2N 2m ⇠
N =0 {n } i<j

129
 P
where the first sum runs only over even numbers N (vortex pairs) such that {n } restricts to
(tot)
winding numbers adding up to n = 0. The fugacity z includes the energy of each vortex.
Considering the very dilute vortex system with z ! 0 we can restrict to N = 0, 2 and discuss
(1) (2)
only a single vortex-anti-vortex pair (n = n = 1). In view of the transition of the vortex
system it is now interesting to examine the vortex spacing. The mean square distance is given
by
Z 2⇡ ~2
1 ~
r1 ~
r 2 m
h|~r1 ~r2 |2 i = d 2
r 1 d 2
r 2 |~
r1 ~
r 2 | 2
(7.80)
2⇡V ⇠ 2 Zv |~r1 ~r2 | ⇠ ⇠
where we renormalize by the volume (area) V of the system. The partition function can be
approximated by Zv ⇡ 1. Then decomposing the two positions ~r1 and ~r2 into center of mass
and relative coordinates, ~rvp = (~r1 + ~r2 )/2 and ~r = ~r1 ~r2 we obtain
Z 1 ✓ ◆ 1
2 2⇡ ~2
2 3 2⇡ ~2 ⇠2 ⇠2 TKT
h|~r| i = ⇠ 2m dr r 2m = 2⇡~2
= 1 (7.81)
⇠ 4 4 T
mkB T

which shows that the vortex-anti-vortex separation is finite for T < TKT but diverges as T !
TKT . Note that here the same transition temperature TKT is found as in Eq.(7.76).
The dissociated vortices move freely and destroy the phase coherence, while the moving bound
pair is not detrimental for the phase, as it does not carry a net phase winding. There are
similar two-dimensional systems which contain topological excitations. In a crystal lattice the
topological excitations are dislocations which have also a topological nature characterized by
means of the Burger’s vector. Another example is the two-dimensions XY-spin system with the
Hamiltonian, X X
HXY = J (Six Sj+ Siy Sjy ) = JS 2 cos(✓i ✓j ) (7.82)
hi,ji hi,ji

~i = S(cos ✓i , sin ✓i ) where the in-plane angle plays a similar role

considering classical spins with S
as the phase of the condensate wave function discussed above.

130