Solid State Theory

Spring Semester 2014

Manfred Sigrist
Institut f
ur Theoretische Physik HIT K23.8
Tel.: 044-633-2584
Email: [email protected]
Website: http://www.itp.phys.ethz.ch/research/condmat/strong/

Lecture Website:
http://www.itp.phys.ethz.ch/education/fs14/sst

Literature:
N.W. Ashcroft and N.D. Mermin: Solid State Physics, HRW International Editions, 1976.
C. Kittel: Einf
uhrung in die Festk
orperphysik, R. Oldenburg Verlag, 1983.
C. Kittel: Quantentheorie der Festk
orper, R. Oldenburg, 1970.
O. Madelung: Introduction to solid-state theory, Springer 1981; auch in Deutsch in drei
Banden: Festk
operphysik I-III, Springer.
J.M. Ziman: Principles of the Theory of Solids, Cambridge University Press, London,
1972.
M.P. Marder: Condensed Matter Physics, John Wiley & Sons, 2000.
G. Grosso & G.P. Parravicini: Solid State Physics, Academic Press, 2000.
G. Czychol: Theoretische Festk
orperphysik, Springer 2004.
P.L. Taylor & O. Heinonen, A Quantum Approach to Condensed Matter Physics, Cambridge Press 2002.
G.D. Mahan, Condensed Matter in a Nutshell, Princeton University Press 2011.
numerous specialized books.

Contents
Introduction

1 Electrons in the periodic crystal - band structure

1.1 Symmetries of crystals . . . . . . . . . . . . . . . . . . . . . . .
1.1.1 Space groups of crystals . . . . . . . . . . . . . . . . . .
1.1.2 Reciprocal lattice . . . . . . . . . . . . . . . . . . . . . .
1.2 Blochs theorem and Bloch functions . . . . . . . . . . . . . . .
1.3 Nearly free electron approximation . . . . . . . . . . . . . . . .
1.4 Tight-binding approximation . . . . . . . . . . . . . . . . . . .
1.4.1 Linear combination of atomic orbitals - LCAO . . . . .
1.4.2 Band structure of s-orbitals . . . . . . . . . . . . . . . .
1.4.3 Band structure of p-orbitals . . . . . . . . . . . . . . . .
1.4.4 Wannier functions . . . . . . . . . . . . . . . . . . . . .
1.4.5 Tight binding model in second quantization formulation
1.5 Symmetry properties of the band structure . . . . . . . . . . .
1.6 Band-filling and materials properties . . . . . . . . . . . . . . .
1.6.1 Electron count and band filling . . . . . . . . . . . . . .
1.6.2 Metals, semiconductors and insulators . . . . . . . . . .
1.7 Semi-classical description . . . . . . . . . . . . . . . . . . . . .
1.7.1 Equations of motion . . . . . . . . . . . . . . . . . . . .
1.7.2 Bloch oscillations . . . . . . . . . . . . . . . . . . . . . .
1.7.3 Current densities . . . . . . . . . . . . . . . . . . . . . .
1.8 Appendix: Approximative band structure calcuations . . . . . .
1.8.1 Pseudo-potential . . . . . . . . . . . . . . . . . . . . . .
1.8.2 Augmented plane wave . . . . . . . . . . . . . . . . . .
2 Semiconductors
2.1 The band structure in group IV . .
2.1.1 Crystal and band structure
2.2 Elementary excitations . . . . . . .
2.2.1 Electron-hole excitations . .
2.2.2 Excitons . . . . . . . . . . .
2.2.3 Optical properties . . . . .
2.3 Doping semiconductors . . . . . . .
2.3.1 Impurity state . . . . . . .
2.3.2 Carrier concentration . . .
2.4 Semiconductor devices . . . . . . .
2.4.1 pn-contacts . . . . . . . . .
2.4.2 Diodes . . . . . . . . . . . .
2.4.3 MOSFET . . . . . . . . . .

.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

8
8
8
10
11
12
14
16
17
17
19
21
21
24
24
25
27
27
28
29
30
30
33

.
.
.
.
.
.
.
.
.
.
.
.
.

35
36
36
38
39
40
43
44
44
46
46
46
48
49

3 Metals
3.1 The Jellium model . . . . . . . . . . . . . . . . . . . . .
3.1.1 Theory of metals - Sommerfeld and Pauli . . . .
3.1.2 Stability of metals - a Hartree-Fock approach . .
3.2 Charge excitations . . . . . . . . . . . . . . . . . . . . .
3.2.1 Dielectric response and Lindhard function . . . .
3.2.2 Electron-hole excitation . . . . . . . . . . . . . .
3.2.3 Collective excitation - Plasmon . . . . . . . . . .
3.2.4 Screening . . . . . . . . . . . . . . . . . . . . . .
3.3 Phonons . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.3.1 Vibration of a isotropic continuous medium . . .
3.3.2 Phonons in metals . . . . . . . . . . . . . . . . .
3.3.3 Peierls instability in one dimension . . . . . . . .
3.3.4 Dynamics of phonons and the dielectric function

.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.

51
51
52
54
57
58
60
60
63
66
66
68
69
74

4 Itinerant electrons in a magnetic field

4.1 The de Haas-van Alphen effect . . . . . . . . . .
4.1.1 Landau levels . . . . . . . . . . . . . . . .
4.1.2 Oscillatory behavior of the magnetization
4.1.3 Onsager equation . . . . . . . . . . . . . .
4.2 Quantum Hall Effect . . . . . . . . . . . . . . . .
4.2.1 Hall effect of the two-dimensional electron
4.2.2 Integer Quantum Hall Effect . . . . . . .
4.2.3 Fractional Quantum Hall Effect . . . . . .

. . .
. . .
. . .
. . .
. . .
gas .
. . .
. . .

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

76
76
76
78
79
80
82
83
89

5 Landaus Theory of Fermi Liquids

5.1 Lifetime of quasiparticles . . . . . . . . . .
5.2 Phenomenological Theory of Fermi Liquids
5.2.1 Specific heat . . . . . . . . . . . . .
5.2.2 Compressibility . . . . . . . . . . . .
5.2.3 Spin susceptibility . . . . . . . . . .
5.2.4 Galilei invariance . . . . . . . . . . .
5.2.5 Stability of the Fermi liquid . . . . .
5.3 Microscopic considerations . . . . . . . . . .
5.3.1 Landau parameters . . . . . . . . . .
5.3.2 Distribution function . . . . . . . . .
5.3.3 Fermi liquid in one dimension? . . .

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

93
93
96
98
100
101
102
103
105
106
108
109

.
.
.
.
.
.
.
.
.
.
.
.
.
.

111
111
113
113
115
118
119
119
121
122
125
127
128
128
130

6 Transport properties of metals

6.1 Electrical conductivity . . . . . . . . . . . .
6.2 Transport equations and relaxation time . .
6.2.1 The Boltzmann equation . . . . . .
6.2.2 The Drude form . . . . . . . . . . .
6.2.3 The relaxation time . . . . . . . . .
6.3 Impurity scattering . . . . . . . . . . . . . .
6.3.1 Potential scattering . . . . . . . . .
6.3.2 Kondo effect . . . . . . . . . . . . .
6.4 Electron-phonon interaction . . . . . . . . .
6.5 Electron-electron scattering . . . . . . . . .
6.6 Matthiessens rule and the Ioffe-Regel limit
6.7 General transport coefficients . . . . . . . .
6.7.1 Generalized Boltzmann equation . .
6.7.2 Thermoelectric effect . . . . . . . . .
3

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.

6.8

Anderson localization . . . . . . . . . . . . . .
6.8.1 Landauer Formula for a single impurity
6.8.2 Scattering at two impurities . . . . . . .
6.8.3 Anderson localization . . . . . . . . . .

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

133
133
135
136

7 Magnetism in metals
7.1 Stoner instability . . . . . . . . . . . . . . . . . . . . . . .
7.1.1 Stoner model within the mean field approximation
7.1.2 Stoner criterion . . . . . . . . . . . . . . . . . . . .
7.1.3 Spin susceptibility for T > TC . . . . . . . . . . . .
7.2 General spin susceptibility and magnetic instabilities . . .
7.2.1 General dynamic spin susceptibility . . . . . . . .
7.2.2 Instability with finite wave vector Q . . . . . . . .
7.2.3 Influence of the band structure . . . . . . . . . . .
7.3 Stoner excitations . . . . . . . . . . . . . . . . . . . . . .

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

138
139
139
140
143
144
144
147
148
150

8 Magnetism of localized moments

8.1 Mott transition . . . . . . . . . . . . . . . . . . . . . . . .
8.1.1 Hubbard model . . . . . . . . . . . . . . . . . . . .
8.1.2 Insulating state . . . . . . . . . . . . . . . . . . . .
8.1.3 The metallic state . . . . . . . . . . . . . . . . . .
8.1.4 Fermi liquid properties of the metallic state . . . .
8.2 The Mott insulator as a quantum spin system . . . . . . .
8.2.1 The effective Hamiltonian . . . . . . . . . . . . . .
8.2.2 Mean field approximation of the anti-ferromagnet .
8.3 Collective modes spin wave excitations . . . . . . . . . .

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

153
154
154
155
156
158
160
160
161
163

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

Introduction
Solid state physics (or condensed matter physics) is one of the most active and versatile branches
of modern physics that have developed in the wake of the discovery of quantum mechanics. It
deals with problems concerning the properties of materials and, more generally, systems with
many degrees of freedom, ranging from fundamental questions to technological applications. This
richness of topics has turned solid state physics into the largest subfield of physics; furthermore,
it has arguably contributed most to technological development in industrialized countries.

Figure 1: Atom cores and the surrounding electrons.

Condensed matter (solid bodies) consists of atomic nuclei (ions), usually arranged in a regular
(elastic) lattice, and of electrons (see Figure 1). As the macroscopic behavior of a solid is
determined by the dynamics of these constituents, the description of the system requires the use
of quantum mechanics. Thus, we introduce the Hamiltonian describing nuclei and electrons,
b =H
be + H
bn + H
b ne ,
H

(1)

with
be =
H

X p
b2i
1X
e2
+
,
2m 2 0 |r i r i0 |
i

bn =
H

i6=i

b2
P
j

2Mj

b ne =
H

X
i,j

1 X Zj Zj 0 e2
,
2
|Rj Rj 0 |
0

(2)

j6=j

Zj e2
,
|r i Rj |

b e (H
b n ) describes the dynamics of the electrons (nuclei) and their mutual interaction and
where H
b ne includes the interaction between ions and electrons. The parameters appearing are
H
m
e
Mj
Zj

free electron mass

elementary charge
mass of j-th nucleus
atomic (charge) number of j-th nucleus

9.1094 1031 kg
1.6022 1019 As
103 104 m

The characteristic scales known from atomic and molecular systems are
5

Length: Bohr radius

Energy: Hartree

aB = ~2 /me2 0.5 1010 m

e2 /aB = me4 /~2 = mc2 2 27eV = 2Ry

with the fine structure constant = e2 /~c = 1/137. The energy scale of one Hartree is much
less than the (relativistic) rest mass of an electron ( 0.5MeV), which in turn is considered small
in particle physics. In fact, in high-energy physics even physics at the Planck scale is considered,
at least theoretically. The Planck scale is an energy scale so large that even gravity is thought
to be affected by quantum effects, as
r
r
~c
~G
2
19
EPlanck = c
1.6 1035 m,
(3)
10 GeV,
lPlanck =
G
c3
where G = 6.673 1011 m3 kg1 s2 is the gravitational constant. This is the realm of the GUT
(grand unified theory) and string theory. The goal is not to provide a better description of
electrons or atomic cores, but to find the most fundamental theory of physics.
high-energy physics
astrophysics and cosmology

solid state physics

10 meV
metals
semiconductors
magnets
superconductors
ferroelectrics
......

10 eV

1 MeV

electrons, cores
atom

phenomenological
particle physics
standard model

effective
models
known and established

GUT
string theory
M-theory

most fundamental
theory

Figure 2: Energy scales in physics.

In contrast, in solid state physics we are dealing with phenomena occurring at room temperature
(T 300K) or below, i.e., at characteristic energies of about E kB T 0.03eV = 30meV,
which is even much smaller than the energy scale of one Hartree. Correspondingly, the important length scales are given by the extension of the system or of the electronic wave functions.
The focus is thus quite different from the one of high-energy physics.
There, a highly successful phenomenological theory for low energies, the so-called standard
model, exists, whereas the underlying theory for higher energies is unknown. In solid state
physics, the situation is reversed. The Hamiltonian (1) describes the known high-energy physics
(on the energy scale of Hartree), and one aims at describing the low-energy properties using reduced (effective, phenomenological) theories. Both tasks are far from trivial.
Among the various states of condensed matter that solid state theory seeks to describe are
metals, semiconductors, and insulators. Furthermore, there are phenomena such as magnetism,
superconductivity, ferroelectricity, charge ordering, and the quantum Hall effect. All of these
states share a common origin: Electrons interacting among themselves and with the ions through
the Coulomb interaction. More often than not, the microscopic formulation in (1) is too complicated to allow an understanding of the low-energy behavior from first principles. Consequently,
the formulation of effective (reduced) theories is an important step in condensed matter theory.
On the one hand, characterizing the ground state of a system is an important goal in itself. However, measurable quantities are determined by excited states, so that the concept of elementary
excitations takes on a central role. Some celebrated examples are Landaus quasiparticles for
6

Fermi liquids, the phonons connected to lattice vibrations, and magnons in ferromagnets. The
idea is to treat the ground state as an effective vacuum in the sense of second quantization,
with the elementary excitations as particles on that vacuum. Depending on the system, the
vacuum may be the Fermi sea or some state with a broken symmetry, like a ferromagnet, a
superconductor, or the crystal lattice itself.
According to P. W. Anderson,1 the description of the properties of materials rests on two principles: The principle of adiabatic continuity and the principle of spontaneously broken symmetry.
By adiabatic continuity we mean that complicated systems may be replaced by simpler systems
that have the same essential properties in the sense that the two systems may be adiabatically
deformed into each other without changing qualitative properties. Arguably the most impressive example is Landaus Fermi liquid theory mentioned above. The low-energy properties of
strongly interacting electrons are the same as those of non-interacting fermions with renormalized parameters. On the other hand, phase transitions into states with qualitatively different
properties can often be characterized by broken symmetries. In magnetically ordered states the
rotational symmetry and the time-reversal invariance are broken, whereas in the superconducting state the global gauge symmetry is. In many cases the violation of a symmetry is a guiding
principle which helps to simplify the theoretical description considerably. Moreover, in recent
years some systems have been recognized as having topological order which may be considered
as a further principle to characterize low-energy states of matter. A famous example for this is
found in the context of the Quantum Hall effect.
The goal of these lectures is to introduce these basic concepts on which virtually all more elaborate methods are building up. In the course of this, we will cover a wide range of frequently
encountered ground states, starting with the theory of metals and semiconductors, proceeding
with magnets, Mott insulators, and finally superconductors.

P.W. Anderson: Basic Notions of Condensed Matter Physics, Frontiers in Physics Lecture Notes Series,
Addison-Wesley (1984).

Chapter 1

Electrons in the periodic crystal band structure

One of the characteristic features of many solids is the regular arrangement of their atoms
forming a crystal. Electrons moving in such a crystal are subject to a periodic potential which
originates from the lattice of ions and an averaged electron-electron interaction (like HartreeFock approximation). The spectrum of extended electronic states, i.e. delocalized eigenstates of
the Schrodinger equation, form bands of allowed energies and gaps of forbidden energies.
There are two limiting starting points towards the understanding of the band formation: (1)
the free electron gas whose continuous spectrum is broken up into bands under the influence of
a periodic potential (electrons undergo Bragg scattering); (2) independent atoms are brought
together into a lattice until the outer-most electronic states overlap and lead to delocalized
states turning a discrete set of states into continua of electronic energies - bands. In this
chapter we will address the emergence of band structures from these two limiting cases. The
band structure of electrons is essential for the basic classification of materials into metals and
insulators (semiconductors).

1.1
1.1.1

Symmetries of crystals
Space groups of crystals

Most solids consist of a regular lattice of atoms with perfectly repeating structures. The minimal
repeating unit of such a lattice is the unit cell. The symmetries of a crystal are contained in
the space group R, a group of symmetry operations (translations, rotations, the inversion or
combinations) under which the crystal is left invariant. In three dimensions, there are 230
different space groups1 (cf. Table 1.1).
1

All symmetry transformations form together a set which has the properties of a group. A group G combined
with a multipliation has the following properties:
the product of two elements of G is also in G: a, b G

a b = c G.

multiplications are associative: a (b c) = (a b) c.

a unit element e G exists with: e a = a e = a for all a G.
for every element a G there is an inverse a1 G with a1 a = a a1 = e.
A group with a b = b a for all pairs of element is called Abelian group, otherwise it is non-Abelian. A subset
G 0 G is called a subgroup of G, if it is a group as well.
Guides to group theory in the context of condensed matter physics can be found in the textbooks
- Mildred S. Dresselhaus, Gene Dresselhaus and Ado Jorio: Group Theory - Application to the Physics of
Condensed Matter
- Peter Y. Yu and Manuel Cardona: Fundamentals of Semiconductors, Springer.

We consider here a crystal displayed as a point lattice, each point either symbolizing an atom or
a whole unit cell (Fig.1.1 for a square lattice). Translations in the space group are represented by
linear combinations of a basic set of translation vectors {ai } (primitive lattice vectors) connecting
lattice points. Any translation that maps the lattice onto itself is a linear combination of the
{ai } with integer coefficients,
Rn = n1 a1 + n2 a2 + n3 a3 .

a2

(1.1)

R
a1

Figure 1.1: Crystal point lattice in two dimensions: the vectors a1 and a2 form the basic set
of translations and R = 2a1 + a2 in the figure. The shaded area is the Wigner-Seitz cell which
is obtained by drawing perpendicular lines (planes in three dimensions) through the center of
all lines connecting neighboring lattice points. The Wigner-Seitz cell also constitutes a unit cell
of the lattice.
General symmetry transformations including general elements of the space group may be written
in the notation due to Wigner,
r 0 = gr + a = {g|a}r,

(1.2)

where g represents a rotation, reflection or inversion with respect to lattice points, axes or
planes. The elements g form the generating point group P. In three dimensions there are 32
point groups. We distinguish the following basic symmetry operations:
basic translations

{E|a},

rotations, reflections, inversions

{g|0},

screw axes, glide planes

{g|a},

where E is the unit element (identity) of P. A screw axis is a symmetry operation of a rotation
followed by a translation along the rotation axis. A glide plane is a symmetry operation with
reflection at a plane followed by a translation along the same plane. The symmetry operations
{g|a}, together with the associative multiplication
{g|a}{g 0 |a0 } = {gg 0 |ga0 + a}

(1.3)

form a group with unit element {E|0}. In general, these groups are non-Abelian, i.e., the group
elements do not commute with each other. However, there is always an Abelian subgroup of
R, the group of translations {E|a}. The elements g P do not necessarily form a subgroup,
9

because some of these elements (e.g., screw axes or glide planes) leave the lattice invariant only
in combination with a translation. Nevertheless, the relation
{g|a}{E|a0 }{g|a}1 = {E|ga0 }

(1.4)

{g|a}1 {E|a0 }{g|a} = {E|g 1 a0 }

(1.5)

holds generally. If P is a subgroup of R, then R is said to be symmorphic. In this case, the

space group contains only primitive translations {E|a} and neither screw axes nor glide planes.
The 14 Bravais lattices2 are symmorphic. Among the 230 space groups 73 are symmorphic and
157 are non-symmorphic.
crystal system

point groups

space group numbers

(# point groups, # space groups)

Schonflies symbols

international tables

triclinic (2,2)

C1 , C1

1-2

monoclinic (3,13)

C2 , Cs , C2h

3-15

orthorhombic (3,59)

D2 , C2v , D2h

16-74

tetragonal (7,68)

C4 , S4 , C4h , D4 , C4v , D2d , D4h

75-142

trigonal (5,25)

C3 , S6 , D3 , C3v , D3d

143-167

hexagonal (7,27)

C6 , C3h , C6h , D6 , C6v , D3h , D6h

168-194

cubic (5, 36)

T, Th , O, Td , Oh

195-230

Table 1.1: List of the point and space groups for each crystal system in three dimensions.

1.1.2

Reciprocal lattice

We define now the reciprocal lattice which is of importance for the electron band structure and
x-ray diffraction on a periodic lattice. The reciprocal lattice is also perfectly periodic with a
translation symmetry with a basic set {bi } defining arbitrary reciprocal lattice vectors as
Gm = m1 b1 + m2 b2 + m3 b3 ,

(1.6)

where mi are integers and

ai bj = 2ij ,
such that
bi = 2

aj ak
ai (aj ak )

i, j = 1, 2, 3 ,

and ai = 2

bj bk
.
bi (bj bk )

(1.7)
(1.8)

The reciprocal lattice of a simple cubic lattice is simple cubic. However, a body centered cubic
(bcc) lattice has a face centered cubic (fcc) reciprocal lattice and vice versa (see slides).
It follows that any real space lattice vector Rn and any reciprocal lattice vector Gm have the
property that
Gm Rn = 2(m1 n1 + m2 n2 + m3 n3 ) = 2N
(1.9)
2

Crystal systems, crystals, Bravais lattices are discussed in more detail in

- Czycholl, Theoretische Festk
orperphysik, Springer

10

with N being an integer. This allows us to expand any function f (r) periodically in the real
lattice as
X
fG eiGr
(1.10)
f (r) =
G

with the obvious property: f (r + R) = f (r). The coefficients are given by

Z
1
fG =
d3 rf (r)eiGr
UC UC

(1.11)

where the integral runs over the unit cell of the periodic lattice with the volume UC . Finally, we
define the (first) Brillouin zone as the Wigner-Seitz cell constructed in the reciprocal lattice
(see Fig.1.1 and 1.2).

1.2

Blochs theorem and Bloch functions

We consider a Hamiltonian H of electrons invariant under a discrete set of lattice translations

{{E|a}}, a symmetry introduced by a periodic potential. This implies that the corresponding
translation operator Tba on the Hilbert space commutes with the Hamiltonian H = He + Hie
(purely electronic Hamiltonian He , interaction between electrons and ions Hie ),
[Tba , H] = 0.

(1.12)

This translation operator is defined through Tba |ri = |r + ai and hr|Tba = hr a|. Neglecting the
interactions among electrons, which would be contained in He , we are left with a single particle
problem
H H0 =

b2
p
+ V (b
r ),
2m

b and p
b are position and momentum operators, and
where r
X
V (r) =
Vion (r Rj ),

(1.13)

(1.14)

describes the potential landscape of the single particle in the ionic background. With Rj being
the position of the j-th ion, the potential V (r) is by construction periodic, with V (r +a) = V (r)
for all lattice vectors a, and represents Hie . Therefore, H0 commutes with Tba . For a Hamiltonian H0 commuting with the translation operator Tba , the eigenstates of H0 are simultaneously
eigenstates of Tba .
Blochs theorem states that the eigenvalues of Tba lie on the unit circle of the complex plane,
which ensures that these states are extended. This means3
Tba (r) = (r a) = a (r) ,

Tbla (r) = (r la) = Tbal (r) = la (r)

(1.16)

with l an integer (positive or negative). In order to be bounded (renormalizable) and delocalized

in the periodic potential the wave function satisfies,
|(r)|2 = |(r la)|2 = |la |2 |(r)|2 = |a |2l |(r)|2 ,

(1.17)

requiring
|a | = 1
3

a = eia .

(1.18)

Transformation of wave function:

| 0 i = Tba |i

0 (r) = hr| 0 i = hr|Tba |i = hr a|i = (r a) .

11

(1.15)

This condition is satisfied if we express the wave function as product of a plane wave eikr and
a periodic Bloch function uk (r)
1
n,k (r) = eikr un,k (r) .

(1.19)

with
Tba un,k (r) = un,k (r a) = un,k (r),

(1.20)

Tba n,k (r) = n,k (r a) = eika n,k (r),

(1.21)

H0 n,k (r) = n,k n,k (r).

(1.22)

The integer n is a quantum number called band index, k is the pseudo-momentum (wave vector)
and represents the volume of the system. Note that the eigenvalue of n,k (r) with respect
to Tba , eika , implies periodicity in the reciprocal space, the k-space, because ei(k+G)a = eika
for all reciprocal lattice vectors G. We may, therefore, restrict k to the first Brillouin zone and
n,k+G = n,k .
Blochs theorem simplifies the initial problem to the so-called Bloch equation for the periodic
function uk ,


(b
p + ~k)2
+ V (b
r ) uk (r) = k uk (r),
(1.23)
2m
where we suppress the band index to simplify the notation. This equation follows from the
relation
b eikr uk (r) = eikr (b
p
p + ~k)uk (r),

(1.24)

which can be used for more complex forms of the Hamiltonian as well. There are various
numerical methods which allow to compute rather efficiently the band energies k for a given
Hamiltonian H.

1.3

Nearly free electron approximation

We start here from the limit of free electrons assuming the periodic potential V (r) is weak.
Using Eqs.(1.10) and (1.11) we expand the periodic potential,
X
V (r) =
VG eiGr ,
(1.25)
G

VG

1
=
UC

d3 r V (r)eiGr .

(1.26)

UC

The potential is real and we assume it also to be invariant under inversion (V (r) = V (r))
= V
with r = 0 being an inversion center of the crystal lattice, leading to VG = VG
G . Note
that the uniform component V0 corresponds to an irrelevant energy shift and may be set to zero.
Because of its periodicity, the Bloch function uk (r) is expressed in the same way,
X
uk (r) =
cG eiGr ,
(1.27)
G

where the coefficients cG = cG (k) are functions of k, in general. Inserting this Ansatz and the
expansion (1.25) into the Bloch equation, (1.23), we obtain a linear eigenvalue problem for the
band energies k ,
 2

X
~
(k G)2 k cG +
VG0 G cG0 = 0.
(1.28)
2m
0
G

12

This represents an eigenvalue problem in infinite dimensions with eigenvectors cG (k) and eigenvalues k as band energies. These k include corrections to the bare parabolic dispersion,
(0)
(0)
k = ~2 k2 /2m, due to the potential V (r). Obviously, the dispersion k is naturally parabolic
in absence of the potential V (r) whereby the eigenstates would be simply plane waves. As a
lowest order approach we obtain the approximative energy spectrum within the first Brillouin
zone, considering only all parabolic bands of the type k (G) = ~2 (k G)2 /2m centered around
the reciprocal wave vectors G (see dashed line in Fig.1.2)).

Example: 1-dimensional system

We illustrate here the nearly free electron method using the case of a one-dimensional lattice.
Assuming that the periodic modulation of the potential is weak, say |VG |  ~2 G2 /2m (taking G
as a characteristic wavevector in the periodic system), the problem (1.28) can be simplified. Let
us start with the lowest energy values around the center of the first Brillouin zone, i.e. k 0
(|k|  /a). For the lowest energy eigenvalue we solve Eq.(1.28) by

1
for G = 0

cG
(1.29)
2mVG


1
for
G
=
6
0
~2 {(k G)2 k 2 }
leading to the energy eigenvalue
k

~2 k 2 X

2m

~2

[(k
G6=0 2m

|VG |2

G)2 k 2 ]

with
E0 =
and

X
G6=0

|G |2

~2 k 2
+ E0
2m

~2 2
G
2m

X
1
1
2
=
1

4
|
|
G

m
m

(1.30)

(1.31)

(1.32)

G6=0

with G = VG /{~2 G2 /2m} (|G |  1). We observe that k is parabolic with a slightly modified
(effective) mass, m m > m. Note that this result resembles the lowest order corrections in the
Rayleigh-Schr
odinger perturbation theory for a non-degenerate state. This solution corresponds
to the lowest branch of the band structure within this approach (see Fig.1.2). The parabolic
approximation of the band structure at a symmetry point with an effective mass m , is a standard
way to approximate band tops or bottoms. It is called k p-approximation (k-dot-p).
We stay at the zone center and address the next eigenstates which are dominated by the parabola
which cross for k = 0 at a value ~2 G
2 /2m. Restricting
originating from G = 2/a = G
ourselves to these two components we obtain the two-dimensional eigenvalue equation system,
2

~
2
c

V2G
2m (k G) k
G = 0 .
(1.33)
2
~
cG
2 k
V
(k + G)
2G

2m

The eigenvalues are obtained through the secular equation,

2

~
2

(k G) k
V2G
=0,
det 2m
~2
2 k
V2G
(k
+
G)
2m

13

(1.34)

leading to

k =

~2

1
2 2m

n
o
2 + (k G)
2
(k + G)

s

~2
2m

n
o2
2 (k G)
2
(k + G)
+ 4|V2G |2

~2 2
~2 2
G |V2G | +
k
2m
2m

with the effective mass

(1.35)


1
1

1 2|G |1

m
m

(1.36)

where m+ > 0 and m < 0 as |G |  1. We observe a energy band gap separating two bands
with opposite curvature (see Fig.1.2). Note that the curvature diverges for V2G 0 as m 0.
The wavefunctions at k = 0 are given by
()
cG

V2G
|V2G |

()

()

c+G = c+G

where we have chosen V2G to be real and positive. Thus,

for k=0,+
sin Gx
uk=0 (x) =

cos Gx
for k=0,

(1.37)

(1.38)

one being even and the other odd under parity operation x x.
A similar analysis can be done at the boundary of the first Brillouin zone where two energy
parabolas cross. For example at k = /a we find the two dominant contributions originate
from G = +2/a and 2/a, respectively, together with G = 0. Also here the energy eigenvalues
show a band gap with parabolic bands centered at k = /a (boundary of the first Brillouin
zone in one-dimension) with positive and negative effective mass (see Fig.1.2). Analogous as
for the band center we can distinguish the wavefunction with even and odd parity for the two
bands at the Brillouin zone boundary. Indeed every crossing energy parabola centered around
different reciprocal lattice points contributes to a band gap. By construction we can extend the
band structure beyond the first Brillouin zone and find a periodic energy spectrum with
k+G = k

(1.39)

where G is a reciprocal lattice vector. Moreover, we find in Fig.1.2 that k = k due to parity
as well as time reversal symmetry, like for free electrons.

1.4

Tight-binding approximation

We consider now a regular lattice of atoms which are well separated such that their atomic
orbitals have small overlaps only. Therefore, in a good approximation the electronic states are
rather well represented by localized atomic orbitals, n (r). The discrete spectrum of the atoms
is obtained with the atomic Hamiltonian,
Ha (R)n (r R) = n n (r R) ,

(1.40)

for an atom located at position R, so that

Ha (R) =

2
p
+ Va (r R)
2m
14

(1.41)

band gap

1st Brillouin zone

2/
a

/a

/
a

2/
a

Figure 1.2: Band structure obtained by the nearly free electron approximation for a regular
one-dimensional lattice.

with Va (r) as the rotation symmetric atomic potential as shown in Fig.1.3 a). The index n shall
include all necessary quantum numbers, besides the principal quantum number also angular
momentum (l, m) and spin. The single-particle Hamiltonian combines all the potentials of the
atoms on the regular lattice (see Fig.1.3 b)),
H=

X
2
p
Va (r Rj ) = Ha (Rj ) + VRj (r)
+
2m

(1.42)

Rj

where we single out one atomic potential (the choice of Rj is arbitrary) and introduce the
correction
X
VRj (r) =
Va (r Rj 0 ) .
(1.43)
Rj 0 6=Rj

a)

b)

V
extended states

localized
atomic orbitals

Figure 1.3: Potential landscape: a) a single atomic Coulomb potential yields a discrete spectrum
electronic states; b) atoms arranged in a regular lattice give rise to a periodic potential which
close to the atom sites look much like the attractive Coulomb-like potential. Electron states
of low energy can be considered as practically localized at the atom sites, as the extension of
their wave functions is very small. The higher energy states, however, extend further and can
delocalize to form itinerant electron states which form bands.

15

1.4.1

Linear combination of atomic orbitals - LCAO

We use here a linear combination of atomic orbitals (LCAO) to approximate the extended Bloch
states
1 X ikRj
e
n (r Rj ) ,
(1.44)
n k (r) =
N R
j

where N denotes the number of lattice sites. This superposition has obviously the properties of
a Bloch function through n k (r a) = eika n k (r) for all lattice vectors a.4 Note that this is
similar to the Hund-Mullikan ansatz for molecular orbitals.
First we determine the norm of this Bloch function,
Z
Z
1 X
3

h1in n 0 (k) = d rn k (r) n 0 k (r) =

d3 reik(Rj 0 Rj ) n (r Rj )n 0 (r Rj 0 )
N
Rj ,Rj 0

XZ

d3 reikRj n (r Rj )n 0 (r)

Rj

= n n 0 +

eikRj n n 0 (Rj )

Rj 6=0

(1.46)
where due to translational invariance in the lattice we may set Rj 0 = 0 eliminating the sum over
Rj 0 and dropping the factor 1/N . To estimate the energy we calculate,
Z
1 X
d3 reik(Rj 0 Rj ) n (r Rj ){Ha (Rj 0 ) + VRj 0 (r)}n 0 (r Rj 0 )
hHin n 0 (k) =
N
Rj ,Rj 0

= En 0 h1in n 0 (k) +

Z
1 X
d3 reik(Rj 0 Rj ) n (r Rj )VRj 0 (r)n 0 (r Rj 0 )
N
Rj ,Rj 0

= En 0 h1in n 0 (k) + En n 0 +
where

Z
En n 0 =

and

Z
n n 0 (Rj ) =

eikRj n n 0 (Rj )

Rj 6=0

(1.47)

d3 rn (r)VRj 0 =0 (r)n 0 (r)

d3 rn (r Rj )VRj 0 =0 (r)n 0 (r) .

(1.48)

(1.49)

From this we can now calculate the band energies through the secular equation,
det [hHin n 0 (k) k h1in n 0 (k)] = 0.

(1.50)

The merit of the approach is that the tightly bound atomic orbitals have only weak overlap such
that both n n 0 (Rj ) and n n 0 (Rj ) fall off very quickly with growing Rj . Mostly it is sufficient to
4

We apply the translation operation to the wave function (1.44),

Tba n k (r)

1 X ikRj
1 X ik(R0j a)
= n k (r a) =
e
n (r a Rj ) =
e
n (r R0j )
| {z }
N Rj
N Rj
R0j

(1.45)
=e

ika

1 X ikR0j

e
n (r R0j ) = eika n k (r) .
N R0
j

16

take Rj connecting nearest-neighbor and sometimes next-nearest-neighbor lattice sites. This is

for example fine for bands derived from 3d-orbitals among the transition metals such as Mn, Fe
or Co etc.. Also transition metal oxides are well represented in the tight-binding formulation.
Alkali metals in the first row of the periodic table, Li, Na, K etc. are not suitable because
their outermost s-orbitals have generally a large overlap. Note that the construction of the
Hamiltonian matrix ensures that k k + G does not change k , if G is a reciprocal lattice
vector.

1.4.2

Band structure of s-orbitals

The most simple case of a non-degenerate atomic orbital is the s-orbital with vanishing angular
momentum (` = 0). Since these orbitals have rotation symmetric wavefunctions, s (r) = s (|r|),
the matrix elements only depend on the distance between sites, |Rj |. As an example we consider
a simple cubic lattice taking nearest-neighbor (Rj = (a, 0, 0), (0, a, 0) and (0, 0, a)) and
next-nearest-neighbor coupling (Rj = (a, a, 0), (a, 0, a), (0, a, a)) into account. For
simplicity we will neglect the overlap integrals ss (Rj ), as they are not important to describe
the essential feature of the band structure.

t Rj connects nearest neighbors

ss (Rj ) =
(1.51)

t0 Rj connects next nearest neighbors

which leads immediately to
k = Es + Es t

n.n.
X
Rj

ikRj

n.n.n.
X

eikRj

Rj

= Es + Es 2t{cos(kx a) + cos(ky a) + cos(kz a)}

(1.52)

4t0 [cos(kx a) cos(ky a) + cos(ky a) cos(kz a) + cos(kz a) cos(kx a)}

Note that VRj (r) 0 in most cases due to the attractive ionic potentials. Therefore t, t0 > 0.
There is a single band resulting from this s-orbital, as shown in Fig.1.4. We may also consider
the k p-approximation at k = 0 which yields an effective mass
k = Es + Es + 6t + 12t0 +

~2 2
k +
2m

(1.53)

with

2
1
= 2 (t + 4t0 ) .
(1.54)
m
~
Note that t and t0 shrink quickly, if with growing lattice constant a the overlap of atomic orbitals
decreases.

1.4.3

Band structure of p-orbitals

We turn to the case of degenerate orbitals. The most simple case is the p-orbital with angular
momentum l = 1 which is three-fold degenerate, represented by the atomic orbital wavefunctions
of the form,
x (r) = x(r), y (r) = y(r), z (r) = z(r) ,
(1.55)
with (r) being a rotation symmetric function. Note that {x, y, z} can be represented by spherical harmonics Y1,m . We assume again a simple cubic lattice such that these atomic orbitals
remain degenerate. Analyzing the properties of the integrals by symmetry, we find,
Ex = Ey = Ez = Ep

and
17

En n 0 = Ep n n 0 .

(1.56)

s-orbitals

p-orbitals

Figure 1.4: Band structures derived from atomic orbitals with s- (one band, upper panel) and
p-symmetry (three bands, lower panel) in a simple cubic lattice. Left side: First Brillouin zone
of the simple cubic lattice. Dispersion given along the k-line connecting X R M . We
choose the parameters: t0 = 0.2t for the s-orbitals; t0 = 0.2t, t = 0.1t, t0 = 0.05t and t00 = 0.15t.
For the band derived from atomic p-orbitals, the irreducible representations of the bands are

given at the symmetry points:

15 (d = 3); X2 (d = 1) , X5 (d = 2); R15 (d = 3); M2 (d = 1),
M5 (d = 2) where d is the dimension of the representation showing the degeneracy.

The overlaps Eq.(1.49) for nearest neighbors,

Rj = (a, 0, 0) k x
( bonding)
t
xx (Rj ) =

t0 Rj = (0, a, 0), (0, 0, a) x

( bonding)

(1.57)

and analogous for yy and zz , while n n 0 = 0, if n

6= n
0 . For next-nearest neighbors by symmetry
we obtain,

Rj = (a, a, 0), (a, 0, a)

t
xx (Rj ) =
(1.58)
0
t Rj = (0, a, a)
and analogous for yy and zz . Next-nearest neighbor coupling also allows for inter-orbital
matrix elements, e.g.
xy (Rj ) = yx (Rj ) = t00 sign(Rnx Rny )
(1.59)
for Rj = (a, a, 0) and analogous for yz (Rj ) and zx (Rj ). The different configuration of
nearest- and next-nearest-neighbor coupling is shown in Fig.1.5.
Now we may setup the coupling matrix,

Ex (k)
4t00 sin(kx a) sin(ky a) 4t00 sin(kx a) sin(kz a)
Ey (k)
4t00 sin(ky a) sin(kz a) (1.60)
hHin n 0 = 4t00 sin(kx a) sin(ky a)
4t00 sin(kx a) sin(kz a) 4t00 sin(ky a) sin(kz a)
Ez (k)

18

next- nearest neighbors

nearest neighbors

" - bonding

! - bonding

no coupling

Figure 1.5: The configurations for nearest- and next-nearest-neighbor coupling between porbitals on different sites. The p-orbitals are depicted by the dumb-bell structured wavefunction
with positve (blue) and negative (red) lobes. For nearest-neighbor couplings we distinguish here
-bonding (full rotation symmetry around connecting axis) and -bonding (two-fold rotation
symmetry around connecting axis). Generally the coupling is weaker for - than for -bonding.
No coupling for symmetry reasons are obtained between orbitals in the lower panel.

with
Ex (k) = Ep + Ep + 2t cos(kx a) 2t0 (cos(ky a) + cos(kz a))

(1.61)

+4tcos(kx a)(cos(ky a) + cos(kz a)) 4t0 cos(ky a) cos(kz a)

and analogous for Ey (k), Ez (k). The three bands derived from the atomic p-orbitals are obtained
by solving the secular equation of the type Eq.(1.50) and shown in Fig.1.4.
Also in this case we may consider a kp-approximation around a symmetry point in the Brillouin
zone. For the -point we find the expansion around k = 0:

Akx2 + B(ky2 + kz2 )

Ckx ky
Ckx kz

Ckx ky
Aky2 + B(kz2 + kx2 )
Cky kz
hHin n 0 = E +
(1.62)
2
2
2
Ckx kz
Cky kz
Akz + B(kx + ky )
with E = Ep + Ep + 2t 4t0 + 4t 4t0 , A = a2 (t + 4t), B = a2 (t0 2t + 2t0 ) and C = 4t00 .
These band energies have to be determined through the secular equations and lead to three
bands with anisotropic effective masses.

1.4.4

Wannier functions

An alternative approach to the tight-binding approximation is through Wannier functions. These

are defined as the Fourier transformation of the Bloch wave functions,
1 X ikR
k (r) =
e
w(r R)
N R

1 X ikR
w(r R) =
e
k (r)
N k
19

(1.63)

where the Wannier function w(r R)5 is centered on the real-space lattice site R. We consider
here the situation of a non-degenerate band analogous to the atomic s-orbital case, such that
there is only one Wannier function per site. Wannier functions obey the orthogonality relation
Z
Z
1 X ikRik0 R0
3

0
d r w (r R )w(r R) =
e
d3 rk 0 (r)k (r)
N
0
k,k

(1.65)
1 X ikRik0 R0
e
kk0 = RR0 .
=
N
0
k,k

We consider the one-particle Hamiltonian to be of the form H = ~2 2 /2m + V (r), with a

periodic potential V (r). Then, k can be expressed through
Z
Z
1 X ik(R0 R)
k = d3 r k (r)Hk (r) =
d3 r w (r R0 )Hw(r R)
e
N
0
R,R

(1.66)
=

X
R

ikR

d r w (r R)Hw(r) ,

where we took translational invariance of the lattice into account. With the definitions
Z
0 = d3 r w (r)Hw(r),
(1.67)
Z
t(R) = d3 r w (r R)Hw(r) for R =
6 0
(1.68)
the band energy can be written as
k = 0 +

t(R)eikR .

(1.69)

This is the same type of tight-binding band structure as we have derived above from the LCAO
view point.
We can extend the Wannier function to the case of several bands, like the p-orbital bands. Then
we define
1 X ikR
nk (r) =
e
cnn (k)wn (r R)
(1.70)
N R,n
where for all k,
X

cnn (k)cn n0 (k) = nn0 .

(1.71)

The matrix cnn (k) rotates the Wannier function from the band basis into the atomic orbital
basis, i.e. for p-bands into the three Wannier function with symmetry like {x, y, z}.
5

Ambiguity of the Wannier functions: The Wannier function is not uniquely defined, because there is a gauge
freedom for the Bloch function which can be multiplied by a phase factor
k (r) ei(k) k (r)

(1.64)

where (k) is an arbitrary real function. In particular, we find different degrees of localization of w(r R) around
its center R depending on the choice of (k).

20

Again we can express the band energy in terms of a tight-binding Hamiltonian,

Z

nk = d3 rnk
(r)Hnk (r)
=

XX

cn 0 n (k)cnn (k)eikR

R n
,
n0

XX

d3 rwn 0 (r R)Hwn (r)

(1.72)

cn 0 n (k)cnn (k)eikR tn 0 n (R) .

R n
,
n0

1.4.5

Tight binding model in second quantization formulation

The tight-binding formulation of band electrons can also be implemented very easily in second
quantization language and provides a rather intuitive interpretation. For simplicity we restrict
ourselves to the single-orbital case and define the following Fermionic operators,
cj,s creates an electron of spin s on lattice site Rj ,

(1.73)

cj,s annihilates an electron of spin s on lattice site Rj ,

in the corresponding Wannier states. We introduce the following Hamiltonian,
X
X
H=
0 cj,s cj,s +
tij ci,s cj,s
j,s

(1.74)

i,j

with tij = tji real. These coefficients tij are called hopping matrix elements, since ci,s cj,s
annihilates an electron on site Rj and creates one on site Ri , in this way an electron moves
(hops) from Rj to Ri . Thus, this Hamiltonian represents the kinetic energy of the electron.
Let us now diagonalize this Hamiltonian by following Fourier transformation, equivalent to the
transformation between Bloch and Wannier functions,
1 X ikRj
1 X
cj,s =
and
cj,s =
(1.75)
a
ks e
a
ks eikRj
N k
N k
aks ) creates (annihilates) an electron in the Bloch state with pseudo-momentum k
where a
ks (
and spin s. Inserting Eq.(1.75) into the Hamiltonian (1.74) leads to

X 1 X
X
X
0
0
1
H=
k0 s a
ks =
k a
ks a
ks
(1.76)
0 ei(kk )Ri +
tij eikRj ik Ri a

N
N
0
k,k ,s

i,j

k,s

where a
ks a
ks = n
ks constitutes the number operator for electrons. The band energy is the same
as obtained above from the tight-binding approach. The Hamiltonian (1.74) will be used later
for the Hubbard model where a real-space formulation is helpful.
The real-space formulation of the kinetic energy allows also for the introduction of disorder, nonperiodicity which can be most straightforwardly implemented by site dependent potentials 0
0i and to spatially (bond) dependent hopping matrix elements tij = t(Ri , Rj ) 6= t(Ri Rj ).

1.5

Symmetry properties of the band structure

The symmetry properties of crystals are a helpful tool for the analysis of their band structure.
They emerge from the symmetry group (space and point group) of the crystal lattice. Consider
the action Sb{g|a} of an element {g|a} of the space group on a Bloch wave function k (r) 6
6

In Dirac notation we write for the Bloch state with pseudo-momentum k as

k (r) = hr|k i.

21

(1.77)

Sb{g|a} k (r) = k ({g|a}1 r) = k (g 1 r g 1 a).

(1.81)

Because {g|a} belongs to the space group of the crystal, we have [Sb{g|a} , H0 ] = 0. Applying a
pure translation Tba0 = Sb{E|a0 } to this new wave function and using Eq.(1.5)
1 0
Tba0 Sb{g|a} k (r) = Sb{g|a} Tbg1 a0 k (r) = Sb{g|a} eik(g a ) k (r)
0
= Sb{g|a} ei(gk)a k (r)
0

= ei(gk)a Sb{g|a} k (r),

(1.82)

0
the latter is found to be an eigenfunction of Tba0 with eigenvalue ei(gk)a . Remember, that,
according to the Bloch theorem, we chose a basis {k } diagonalizing both Tba and H0 . Thus,
apart from a phase factor, the action of a symmetry transformation {g|a} on the wave function7
corresponds to a rotation from k to gk.

Sb{g|a} k (r) = {g|a} gk (r),

(1.84)

Sb{g|a} |ki = {g|a} |gki.

(1.85)

with |{g|a} |2 = 1, or

in Dirac notation. Then it is easy to see that

1
H0 Sb{g|a} |ki = hk|H0 |ki = k .
gk = hgk|H0 |gki = hk|Sb{g|a}

(1.86)

Consequently, there is a star-like structure of equivalent points gk with the same band energy
( degeneracy) for each k in the Brillouin zone (cf. Fig. 1.6).
For a general point k the number of equivalent points in the star equals the number of point
The action of the operator Sb{g|a} on the state |ri is given by
Sb{g|a} |ri = |gr + ai

and

hr|Sb{g|a} = hg 1 r g 1 a|,

(1.78)

such that
hr|Sb{g|a} |k i = k (g 1 r g 1 a).

(1.79)

The same holds for pure translations.

Note that this definition has also implications on the sequential application of transformation operators such
as
Sb{g1 |a1 } Sb{g2 |a2 } k (r) = hr|Sb{g1 |a1 } Sb{g2 |a2 } |k i = h{g1 |a1 }1 r|Sb{g2 |a2 } |k i
(1.80)


= h{g2 |a2 }1 {g1 |a1 }1 r|k i = k {g2 |a2 }1 {g1 |a1 }1 r .
This is important in the context of Eq.(1.82).
7
Symmetry behavior of the wave function.
Sb{g|a} k (r)

X
X
1
1
1
1
1
1
= Sb{g|a} eikr
cG (k) eiGr = eik(g rg a)
cG (k) eiG(g rg a)

G
G
X
X
1
1
= ei(gk)a ei(gk)r
cG (k) ei(gG)r = ei(gk)a ei(gk)r
cg1 G (k) eiGr

G
G

(1.83)

X
1
= ei(gk)a ei(gk)r
cG (gk) eiGr = {g|a} gk (r),

G
where we use the fact that cG = cG (k) is a function of k with the property cg1 G (k) = cG (gk) i.e. Sb{g|a} uk (r) =
ugk (r).

22

Figure 1.6: Star of k-points in the Brillouin zone with degenerate band energies: Left panel:
Star of k; Right panel: contour plot of a two dimensional band k = 2t{cos(kx a) + cos(ky a)} +
4t0 cos(kx a) cos(ky a). The dots correspond to the star of k with degenerate energy values, demonstrating nk = n,gk .

group elements for this k (without inversion). If k lies on points or lines of higher symmetry, it
is left invariant under a subgroup of the point group. Consequently, the number of beams of the
star is smaller. The subgroup of the point group leaving k unchanged is called little group of k.
If inversion is part of the point group, k is always contained in the star of k. In summary, we
have the simple relations
nk = n,gk ,

nk = n,k ,

nk = n,k+G .

(1.87)

We can also use symmetries to characterize Bloch states for given pseudo-momentum k. Let us
take a set of degenerate Bloch states belonging to the band n, {|; n, ki} satisfying the eigenvalue
equation,
H|; n, ki = nk |; n, ki
(1.88)
For given k we consider the little group operations. Operating an element g of the little group
on a state |; n, ki we obtain again an eigenstate of the Hamiltonian H, as Sb{g,0} commutes with
H,
X
Sb{g,0} |, nki =
| 0 , nki h 0 , nk|Sb{g,0} |, nki .
(1.89)
|
{z
}
0
=M 0 , (g)

We transform only within the subspace of degenerate states |, nki and the matrix M 0 , (g) is
a representation of the group element g on the vector space of eigenstates {|, nki}. If this representation is irreducible then its dimension corresponds to the degeneracy of the corresponding
set of Bloch states.
Looking back to the example of tight-binding bands derived from atomic p-orbitals (Fig.1.4).
The symmetry at the -point (k = 0) is the full crystal point group Oh (simple cubic lattice).
The representation
15 is three-dimensional corresponding to a basis set {x, y, z} (p-orbital).
At the X-point (symmetry point on the Brillouin zone boundary) the group is reduced to D4h
(tetragonal) and the representations appearing are X2 (one-dimensional corresponding to z)
and X5 (two-dimensional corresponding to {x, y}). Note that generally the little group of k has
lower symmetry and leads to splitting of degeneracies as can be seen on the line M where
the bands, degenerate at the -point split up into three and combine again at the M -point
into two level of degeneracy one and two, respectively. The symmetry group of the M -point
is D4h while for arbitrary k k [110] it is C2v containing only four elements leaving k invariant:
C2v D4h Oh .
23

1.6

Band-filling and materials properties

Due to the fermionic character of electrons each of the band states |n, k, si can be occupied
with one electron taking also the spin quantum number into account with spin s = and
(Pauli exclusion principle). The count of electrons has profound implications on the properties
of materials. Here we would like to look at the most simple classification of materials based on
independent electrons.

1.6.1

Electron count and band filling

We consider here a most simple band structure in the one-dimensional tight-binding model with
nearest-neighbor coupling. The lattice has N sites (N even) and we assume periodic boundary
conditions. The Hamiltonian is given by
H = t

N X
X

{
cj+1,s cj,s + cj,s cj+1,s }

(1.90)

j=1 s=,

where we impose the equivalence j + N = j (periodic boundary conditions). This Hamiltonian

can be diagonalized by the Fourier transform
1 X
cj,s =
a
k,s eiRj k
N k

(1.91)

leading to
H=

X
k,s

k,s
k a
k,s a

with k = 2t cos ka .

(1.92)

Now we request
eiRj k = ei(Rj +L)k

Lk = N ak = 2n

k=

2
2 n
n=
L
a N

(1.93)

with the pseudo-momentum k within the first Brillouin zone ( a < k < a ) and n being an
integer. On the real-space lattice an electron can take 2N different states. Thus, for k we find
that n should take the values, n + N/2 = 1, 2, . . . , N 1, N . Note that k = /a and k = +/a
differ by a reciprocal lattice vector G = 2/a and are therefore identical. This provides the
same number of states (2N ), since per k we have two spins (see Fig.1.7).
We can fill these states with electrons following the Pauli exclusion principle. In Fig.1.7 we show
the two typical situations: (1) N electrons corresponding to half of the possible electrons which
can be accommodated leading to a half-filled band and (2) 2N electrons exhausting all possible
states representing a completely filled band. In the case of half-filling we define the Fermi energy
as the energy F of the highest occupied state, here F = 0. This corresponds to the chemical
potential, the energy necessary to add an electron to the system at T = 0K.
An important difference between (1) and (2) is that the former allows for many different states
which may be separated from each other by a very small energy. For example, considering
the ground state (as in Fig.1.7) and the excited states obtained by moving one electron from
k = /2a (n = N/4) to k 0 = 2(1 + N/4)/N a (n = 1 + N/4), we find the energy difference
E = k0 k 2t

2
1

N
N

(1.94)

which shrinks to zero for N . On the other hand, for case (2) there is only one electron
configuration possible and no excitations within the one-band picture.

24

half filled band

-3

-2

-1

completely filled band

Figure 1.7: One-dimensional tight-binding model with N = 8 and periodic boundary conditions.
The dispersion has eight different k-levels whereby it has to be noticed that +/a and /a
are equivalent. The condition of half-filling and complete filling are shown, where for half-filling
a ground state configuration is shown (note there are 4 degenerate states). For the completely
filled band all k levels are occupied by two electrons of opposite spin. This means in real space
that also all sites are occupied by two electrons. This is a non-degenerate state.

1.6.2

Metals, semiconductors and insulators

The two situations depicted in Fig.1.7 are typical as each atom (site) in a lattice contributes an
integer number of electrons to the system. So we distinguish the case that there is an odd or
an even number of electrons per unit cell. Note that the unit cell may contain more than one
atom, unlike the situation shown in our tight-binding example.
The bands can be either completely filled or empty when the number of electrons per
atom (unit cell) is even. Thus taking the complete set of energy bands into account, the
chemical potential cannot be identified with a Fermi energy but lies within the energy
gap separating highest filled and the lowest empty band (see Fig.1.8). There is a finite
energy needed to add, to remove or to excite an electron. If the band gap Eg is much
smaller than the bandwidth, we call the material a semiconductor. for Eg of the order of
the bandwidth, it is an (band) insulator. In both cases, for temperatures T  Eg /kB the
application of a small electric voltage will not produce an electric transport. The highest
filled band is called valence band, whereas the lowest empty band is termed conduction
band. Examples for insulators are C as diamond and for semiconductors Si and Ge. They
have diamond lattice structure with two atoms per unit cell. As these atoms belong to
the group IV in the periodic table, each provides an even number of electrons suitable for
completely filling bands. Note that we will encounter another form of an insulator, the
Mott insulator, whose insulating behavior is not governed by a band structure effect, but
by a correlation effect through strong Coulomb interaction.
If the number of electrons per unit cell is odd, the uppermost non-empty band is half
filled (see Figure 1.8). Then the system is a metal, in which electrons can move and
25

insulator
semiconductor
E

metal
semimetal

metal
E

EF

EF
EF
filled

filled

filled

Figure 1.8: Material classes according to band filling: left panel: insulator or semiconductor
(partially filled bands with the Fermi level in band gap); center panel: metal (Fermi level inside
band); right panel: metal or semimetal (Fermi level inside two overlapping bands).

excitations with arbitrarily small energies are possible. The electrons remain mobile down
to arbitrarily low temperatures. The standard example of a metal are the Alkali metals
in the first column of the periodic table (Li, Na, K, Rb, Cs), as all of them have the
configuration [noble gas] (ns)1 , i.e., one mobile electron per ion.
In general, band structures are more complex. Different bands need not to be separated
by energy gaps, but can overlap instead. In particular, this happens, if different orbitals
are involved in the structure of the bands. In these systems, bands can have any fractional
filling (not just filled or half-filled). The earth alkaline metals are an example for this
(second column of the periodic table, Be, Mg, Ca, Sr, Ba), which are metallic despite
having two (n, s)-electrons per unit cell. Systems, where two bands overlap at the Fermi
energy but the overlap is small, are termed semi-metals. The extreme case, where valence
and conduction band touch in isolated points so that there are no electrons at the Fermi
energy and still the band gap is zero, is realized in graphene.
The electronic structure is also responsible for the cohesive forces necessary for the formation
of a regular crystal. We may also classify materials according to relevant forces. We distinguish
four major types of crystals:
Molecular crystals are formed from atoms or molecules with closed-shell atomic structures
such as the noble gases He, Ne etc. which become solid under pressure. Here the van der
Waals forces generate the binding interactions.
Ionic crystals combine different atoms, A and B, where A has a small ionization energy while
B has a large electron affinity. Thus, electrons are transferred from A to B giving a
positively charged A+ and a negatively charge B . In a regular (alternating) lattice the
energy gained through Coulomb interaction can overcome the energy expense for the charge
transfer stabilizing the crystal. A famous example is NaCl where one electron leaves Na
([Ne] 3s) and is added to Cl ([Ne] 3s2 3p5 ) as to bring both atoms to closed-shell electronic
configuration.
Covalent-bonded crystals form through chemical binding, like in the case of the H2 , where
neighboring atoms share electrons through the large overlap of the electron orbital wavefunction. Insulators like diamond C or semiconductors like Si or GaAs are important
26

examples of this type as we will discuss later. Note that electrons of covalent bonds are
localized between the atoms.
Metallic bonding is based on delocalized electrons (in contrast to the covalent bonds) stripped
from their atoms. The stability of simple metals like the alkaly metals Li, Na, K etc will
be discussed later. Note that many metals, such as the noble metals Au or Pt, can also
involve aspects of covalent or molecular bonding through overlapping but more localized
electronic orbitals.

1.7

Dynamics of band electrons - semiclassical approach

In quantum mechanics, the Ehrenfest theorem shows that the expectation values of the position
and momentum operators obey equations similar to the equation of motion in Newtonian mechanics.8 An analogous formulation holds for electrons in a periodic potential, where we assume
that the electron may be described as a wave packet of the form
X
0
k (r, t) =
gk (k0 )eik rik0 t ,
(1.98)
k0

where gk (k0 ) is centered around k in reciprocal space and has a width of k. k should be
much smaller than the size of the Brillouin zone for this Ansatz to make sense, i.e., k  2/a.
Therefore, the wave packet is spread over many unit cells of the lattice since Heisenbergs
uncertainty principle (k)(x) > 1 implies x  a/2. In this way, the pseudo-momentum k
of the wave packet remains well defined. Furthermore, the applied electric and magnetic fields
have to be small enough not to induce transitions between different bands. The latter condition
is not very restrictive in practice.

1.7.1

Semi-classical equations of motion

We introduce the rules of the semi-classical motion of electrons with applied electric and magnetic
fields without proof:
The band index of an electron is conserved, i.e., there are no transitions between the bands.
8

Ehrenfest theorem for free electrons with the Hamiltonian,

H=

b2
p
+ V (b
r)
2m

(1.95)

states for the expectation values for a particle represented by a wave packet,
hb
pi
d
i
b]i =
hb
r i = h[H, r
dt
~
m

(1.96)

and

i
d
b]i = hV (b
hb
pi = h[H, p
r )i
dt
~
which has a form similar to Newtons equations with the restriction hV (b
r )i 6= V (hb
r i).

27

(1.97)

 The equations of motion read9

nk
,
~k
e
~k = eE(r, t) v n (k) H(r, t).
c
r = v n (k) =

(1.101)
(1.102)

All electronic states have a wave vector that lies in the first Brillouin zone, as k and k + G
label the same state for all reciprocal lattice vectors G.
In thermal equilibrium, the electron density per spin in the n-th band in the volume
element d3 k/(2)3 around k is given by
nF [n (k)] =

1
e[n (k)]/kB T

+1

(1.103)

Each state of given k and spin can be occupied only once (Pauli principle).
Note that ~k is not the momentum of the electron, but the so-called lattice momentum or pseudo
momentum in the Bloch theory of bands. It is connected with the eigenvalue of the translation
operator on the state. Consequently, the right-hand side of the equation (1.102) is not the force
that acts on the electron, as the forces exerted by the periodic lattice potential is not included.
The latter effect is contained implicitly through the form of the band energy (k), which governs
the first equation.

1.7.2

Bloch oscillations

The fact that the band energy is a periodic function of k leads to a strange oscillatory behavior of
the electron motion in a static electric field. For illustration, consider a one-dimensional system
where the band energy k = 2 cos ka leads to the solution of the semi-classical equations
(1.101,1.102)
~k = eE
eEt
k=
~


2a
eEat
x = sin
,
~
~

(1.104)
(1.105)
(1.106)

in the presence of a homogenous electric field E. It follows immediately, that the position x of
the electron oscillates like


2
eEat
x(t) =
cos
.
(1.107)
eE
~
This behavior is called Bloch oscillation and means that the electron oscillates around its initial
position rather than moving in one direction when subjected to a static electric field. This effect
can only be observed under very special conditions where the probe is absolutely clean. The
effect is easily destroyed by damping or scattering.
9

A plausibility argument concerning the conservation of energy leading to the equation (1.102) is given here.
The time derivative of the energy (kinetic and potential)
E = n (k(t)) e(r(t))

(1.99)

has to vanish, i.e.,

0=



n (k)
dE
=
k e r = v n (k) ~k e .
dt
k

(1.100)

From this, equation (1.102) follows directly for the electric field E = and the Lorentz force is allowed
because the force is always perpendicular to the velocity v n .

28

time
Figure 1.9: Experimental observation of Bloch oscillation for accelerated cesium atoms trapped
in an periodic optical lattice generated by standing waves of laser light: mean velocity hvi versus
time [M.B. Dahan et al, Phys. Rev. Lett. 76, 4508 (1996)].

1.7.3

Current densities

We will see in chapter 6 that homogenous steady current carrying states of electron systems can
be described by the momentum distribution n(k). Assuming this property, the current density
follows from
Z
Z
d3 k
d3 k
(k)
j = 2e
v(k)n(k)
=
2e
n(k)
,
(1.108)
3
(2)
(2)3
~k
BZ

BZ

where the integral extends over all k in the Brillouin zone (BZ) and the factor 2 originates
from the two possible spin states of the electrons. Note that for a finite current density j, the
momentum distribution n(k) has to deviate from the equilibrium Fermi-Dirac distribution in
equation (1.103). It is straight forward to show that the current density vanishes for an empty
band. The same holds true for a completely filled band (n(k) = 1) where equation (1.101)
implies
Z
d3 k 1 (k)
j = 2e
=0
(1.109)
(2)3 ~ k
BZ

because (k) is periodic in the Brillouin zone, i.e., (k + G) = (k) when G is a reciprocal lattice
vector. Thus, neither empty nor completely filled bands can carry currents.
An interesting aspect of band theory is the picture of holes. We compute the current density
for a partially filled band in the framework of the semi-classical approximation,
Z
d3 k
n(k)v n (k)
(1.110)
j = 2e
(2)3
BZ
Z 3
h Z d3 k
i
d k
= 2e
v(k)

[1

n(k)]v(k)
(1.111)
(2)3
4 3
BZ
BZ
Z
d3 k
= +2e
[1 n(k)]v(k).
(1.112)
(2)3
BZ

This suggests that the current density comes either from electrons in filled states with charge
e or from holes, missing electrons carrying positive charge and sitting in the unoccupied
electronic states. In band theory, both descriptions are equivalent. However, it is usually easier
to work with holes, if a band is almost filled, and with electrons if the filling of an energy band
is small.
29

1.8

Appendix: Approximative band structure calcuations

While the approximation of nearly free electrons gives a qualitative picture of the band structure,
it rests on the assumption that the periodic potential is weak, and, thus, may be treated as
a small perturbation. Only few states connected with different reciprocal lattice vectors are
sufficient within this approximation. However, in reality the ionic potential is strong compared
to the electrons kinetic energy. This leads to strong modulations of the wave function around
the ions, which is not well described by slightly perturbed plane waves.

1.8.1

Pseudo-potential

In order to overcome this weakness of the plane wave solution, we would have to superpose a
very large number of plane waves, which is not an easy task to put into practice. Alternatively,
we can divide the electronic states into the ones corresponding to filled low-lying energy states,
which are concentrated around the ionic core (core states), and into extended (and more weakly
modulated) states, which form the valence and conduction bands. The core electron states
may be approximated by atomic orbitals of isolated atoms. For a metal such as aluminum (Al:
1s2 2s2 2p6 3s2 3p) the core electrons correspond to the 1s-, 2s-, and 2p-orbitals, whereas the 3sand 3p-orbitals contribute dominantly to the extended states of the valence- and conduction
bands. We will focus on the latter, as they determine the low-energy physics of the electrons.
The core electrons are deeply bound and can be considered inert.

extended
conduction/valence states

3s23p

Al
1s2 2s2 2p63s23p
localized core states

1s2 2s2 2p6

Figure 1.10: Separation into extended and core electronic states (example Aluminium).
We introduce the core electron states as |j i, with H|j i = Ej |j i where H is the Hamiltonian
of the single atom. The remaining states have to be orthogonal to these core states, so that we
make the Ansatz
X
|n,k i = |nk i
|j ihj |n,k i,
(1.113)
j

with |n,k i an orthonormal set of states. Then, hn,k |j i = 0 holds for all j. If we choose plane
waves for the |nk i, the resulting |n,k i are so-called orthogonalized plane waves (OPW). The
Bloch functions are superpositions of these OPW,
X
|n,k i =
bk+G |n,k+G i,
(1.114)
G

where the coefficients bk+G converge rapidly, such that, hopefully, only a small number of OPWs
is needed for a good description.
30

First, we again consider an arbitrary |nk i and insert it into the eigenvalue equation H|nk i =
Enk |nk i,


X
X
H|nk i
H|j ihj |n,k i = Enk |nk i
|j ihj |n,k i
(1.115)
j

or
H|nk i +

X
j

(Enk Ej ) |j ihj |n,k i = Enk |nk i.

(1.116)

P
We introduce the integral operator in real space Vb 0 = j (Enk Ej )|j ihj |, describing a nonlocal and energy-dependent potential. With this operator we can rewrite the eigenvalue equation
in the form
(H + Vb 0 )|n,k i = (H0 + Vb + Vb 0 )|n,k i = (H + Vbps )|n,k i = Enk |nk i.

(1.117)

This is an eigenvalue equation for the so-called pseudo-wave function (or pseudo-state) |nk i,
instead of the Bloch state |nk i, where the modified potential Vbps = Vb + Vb 0 is called pseudopotential. The attractive core potential Vb = V (b
r ) is always negative. On the other hand,
Enk > Ej , such that Vb 0 is positive.
It
follows
that
Vbps is weaker than both Vb and Vb 0 . An
P
arbitrary number of core states j aj |j i may be added to |nk i without violating the orthogonality condition (1.113). Consequently, neither the pseudo-potential nor the pseudo-states are
uniquely determined and may be optimized variationally with respect to the set {aj } in order
to optimally reduce the spatial modulation of either the pseudo-potential or the wave-function.
potenial

pseudo-potenial

plane-wave approximation
wave function
Figure 1.11: Illustration of the pseudo-potential.

If we are only interested in states inside a small energy window, the energy dependence of the
pseudo-potential can be neglected, and Vps may be approximated by a standard potential (see
Figure 1.11). Such a simple Ansatz is exemplified by the atomic pseudo-potential, proposed by
Ashcroft, Heine and Abarenkov (AHA). The potential of a single ion is assumed to be of the
form
(
V0
r < Rc ,
vps (r) =
(1.118)
Zion e2
r
r > Rc ,
where Zion is the charge of the ionic core and Rc its effective radius (determined by the core
electrons). The constants Rc and V0 are chosen such that the energy levels of the outermost
electrons are reproduced correctly for the single-atom calculations. For example, the 1s-, 2s-,
and 2p-electrons of Na form the ionic core. Rc and V0 are adjusted such that the one-particle
problem p2 /2m + vps (r) leads to the correct ionization energy of the 3s-electron. More flexible
31

Rc

Figure 1.12: Pseudopotential: Ashcroft, Heine and Abarenkov form.

approaches allow for the incorporation of more experimental input into the pseudo-potential.
The full pseudo-potential of the lattice can be constructed from the contribution of the individual
atoms,
X
Vps (r) =
vps (|r Rj |),
(1.119)
n

where Rj is the lattice vector. For the method of nearly free electrons we need the Fourier
transform of the potential evaluated at the reciprocal lattice vectors,
Z
Z
1
N
Vps,G =
d3 r Vps (r)eiGr =
d3 r vps (r)eiGr .
(1.120)

For the AHA form (1.118), this is given by

Vps,G =

4Zion e2
G2


cos(GRc )



V0
2 2
(Rc G 2) cos(GRc ) + 2 2Rc G sin(GRc ) . (1.121)
+
Zion e2 G

For small reciprocal lattice vectors, the zeroes of the trigonometric functions on the right-hand
side of (1.121) reduce the strength of the potential. For large G, the pseudo-potential decays
1/G2 . It is thus clear that the pseudo-potential is always weaker than the original potential.
Extending this theory for complex unit cells containing more than one atom, the pseudo-potential
may be written as
X

Vps (r) =
vps
(|r (Rj + R )|),
(1.122)
n,

is the pseudowhere R denotes the position of the -th base atom in the unit cell. Here, vps
potential of the -th ion. In reciprocal space,
Z
N X iGR

Vps,G =
e
d3 r vps
(|r|)eiGr
(1.123)

X
=
eiGR F,G .
(1.124)

The form factor F,G contains the information of the base atoms and may be calculated or
obtained by fitting experimental data.
32

1.8.2

Augmented plane wave

We now consider a method introduced by Slater in 1937. It is an extension of the so-called

Wigner-Seitz cell method (1933) and consists of approximating the crystal potential by a socalled muffin-tin potential. The latter is a periodic potential, which is taken to be spherically
symmetric and position dependent around each atom up to a distance rs , and constant for larger
distances. The spheres of radius rs are taken to be non-overlapping and are contained completely
in the Wigner-Seitz cell10 (Figure 1.13).

Figure 1.13: Muffin-tin potential.

The advantage of this decomposition is that the problem can be solved using a divide-andconquer strategy. Inside the muffin-tin radius we solve the spherically symmetric problem, while
the solutions on the outside are given by plane waves; the remaining task is to match the
solutions at the boundaries.
The spherically symmetric problem for |r| < rs is solved with the standard Ansatz
(r) =

ul (r)
Ylm (, ),
r

(1.125)

where (r, , ) are the spherical coordinates or r and the radial part ul (r) of the wave function
obeys the differential equation


~2 d2
~2 l(l + 1)

+
+
V
(r)

E
ul (r, E) = 0.
(1.126)
2m dr2
2mr2
We define an augmented plane wave (APW) A(k, r, E), which is a pure plane wave with wave
vector k outside the Muffin-tin sphere. For this, we employ the representation of plane waves
by spherical harmonics,
eikr = 4

X
l
X

il jl (kr)Ylm
(k)Ylm (
r ),

(1.127)

l=0 m=l
10
The Wigner-Seitz cell is the analogue of the Brillouin zone in real space. One draws planes cutting each
line joining two atoms in the middle, and orthogonal to them. The smallest cell bounded by these planes is the
Wigner-Seitz cell.

33

where jl (x) is the l-th spherical Bessel function. We parametrize

A(k, r, E) =

r u (r, E)
4 X l

p
Ylm (k)Ylm (
r ), r < rs ,
i jl (krs ) s l

ru
(r
,
E)

l
s
UC

l,m

4 X l

p
i jl (kr)Ylm
(k)Ylm (
r ),

UC l,m

(1.128)

r > rs ,

where UC is the volume of the unit cell. Note that the wave function is always continuous at
r = rs , but that its derivatives are in general not continuous. We can use an expansion of the
wave function k (r) similar to the one in the nearly free electron approximation (see equations
(1.19) and (1.27)),
X
cG (k)A(k + G, r, E),
(1.129)
k (r) =
G

where the G are reciprocal lattice vectors. The unknown coefficients can be determined variationally by solving the system of equations
X
hAk (E)|H E|Ak+G (E)icG (k) = 0,
(1.130)
G

where

hAk (E)|H E|Ak0 (E)i =


~2 k k0
E UC k,k0 + Vk,k0
2m

(1.131)

with
"
Vk,k0 =

4rs2

~2 k k0
E
2m

j1 (|k k0 |rs )
|k k0 |

0 (r , E)
X
u
~2
0
k
)j (kr )j (k 0 r ) l s
(2l + 1)Pl (k
. (1.132)
+
l
s l
s
2m
ul (rs , E)
l=0

Here, Pl (z) is the l-th Legendre polynomial and u0 = du/dr. The solution of (1.130) yields
the energy bands. The most difficult parts are the approximation of the crystal potential by
the muffin-tin potential and the computation of the matrix elements in (1.130). The rapid
convergence of the method is its big advantage: just a few dozens of G-vectors are needed and
the largest angular momentum needed is about l = 5. Another positive aspect is the fact that
the APW-method allows the interpolation between the two extremes of extended, weakly bound
electronic states and tightly bound states.

34

Chapter 2

Semiconductors
The technological relevance of semiconductors can hardly be overstated. In this chapter, we
review some of their basic properties. Regarding the electric conductivity, semiconductors are
placed in between metals and insulators. Normal metals are good conductors at all temperatures,
and the conductivity usually increases with decreasing temperature. On the other hand, for
semiconductors and insulators the conductivity decreases upon cooling (see Figure 2.1).

T
semiconductor/isolator

T
metal

Figure 2.1: Schematic temperature dependence of the electric conductivity for semiconductors
and metals.
We will see that the conductivity may be written in the Drude form as
=

ne2
,
m

(2.1)

where n is the density of mobile electrons, is the average time between two scattering events of
the electrons (transport life time of electrons) and m and e are the electronic mass and charge,
respectively. In metals, n is essentially independent of temperature, whereas the scattering
time decreases with increasing temperature. Thus, determines mainly the temperature
dependence of the conductivity in metals. On the other hand, insulators and semiconductors
have no mobile charges at T = 0. At finite temperature, charges are induced by thermal
excitations which have to overcome the band gap1 Eg between the valence and the conduction
1

Actually, one has to count both the excited electrons in the conduction band and the resulting holes in the
valence band, as both contribute to the current,
j = (+ + )E,

with

n e2
,
m

(2.2)

where + and stand for holes and electrons, respectively, and n+ = n holds for thermal excitation. Note that,
in general, the effective masses and scattering times are not the same for the valence and conduction bands.

35

band, yielding

n n0

T
T0

3/2

eEg /2kB T ,

(2.3)

where T0 = 300K and the electron density in the material n0 is typically 1020 cm3 . For
insulators, the energy gap is huge, e.g., 5.5 eV for diamond. Consequently, the charge carrier
density at room temperature T = 300K is around n 1027 cm3 . For a higher charge carrier
density n 103 1011 cm3 , smaller gaps Eg 0.5 1eV are necessary. Materials with
a band gap in this regime are not fully isolating and, therefore, are termed semiconductors.
However, the carrier densities of both insulators and semiconductors are dwarfed by the electron
density in metals contributing to current transport (nmetal 1023 1024 cm3 ). Adding a
small amount of impurities in semiconductors, a process called doping with acceptors or donors,
their conductivity can be engineered in various ways, rendering them useful as components in
innumerable applications.

2.1
2.1.1

The band structure of the elements in group IV

Crystal and band structure

The most important semiconductor for technological applications is silicon (Si) that like carbon
(C), germanium (Ge) and tin (Sn) belongs to the group IV of the periodic table. These
elements have four electrons in their outermost shell in the configuration (ns)2 (np)2 (n=2 for
C, n=3 for Si, n=4 for Ge, and n=5 for Sn). All four elements form crystals with a diamond
structure (cf. Figure 2.2), i.e., a face-centered cubic lattice with a unit cell containing two atoms
located at (0, 0, 0) and ( 41 , 14 , 41 ) (for Sn this is called -Sn). The crystal structure is stabilized
by hybridization of the four valence electrons, leading to covalent bonding of oriented orbitals,
|1 i = |nsi + |npx i + |npy i + |npz i,

|3 i = |nsi |npx i + |npy i |npz i,

|2 i = |nsi + |npx i |npy i |npz i,

|4 i = |nsi |npx i |npy i + |npz i.

(2.4)

Locally, the nearest neighbors of each atom form a tetrahedron around it, which leads to the
diamond structure of the lattice.
a
(1, 1, 1)
4

a
Figure 2.2: The crystal structure of diamond corresponds to two face-centered cubic latices
shifted by a quarter of lattice spacing along the (1,1,1) direction (2 sublattices blue and orange).

The electron band structure of this system is represented in the Brillouin zone of the fcc lattice
(Fig. 2.3). In Fig. 2.4 the approximate band structures of both carbon C and silicon Si are
displayed. They are rather similar. Focussing on the -point we find the lowest band belongs
to the trivial representation 1 of the cubic point group and is, consequently non-degenerate.
36

Next we find the three-fold degenerate bands belonging to the three-dimensional representation
25 . Here the -point is the band top. There is a finite energy gap to the next higher bands
which at the -point are also three-fold degenerate belonging to 15 . This is, however, neither a
band bottom nor top. The lowest point of these bands lies between the -point and the Brillouin
zone boundary at the X-point (Fig. 2.3) with k0 (100)k0 and five other equivalent directions.
kz

ky

kx
Figure 2.3: The Brillouin zone of a face-centered cubic crystal (in real space) is embedded in a
bcc lattice.

Let us look now at the electron count. The two atoms per unit cell provide together eight valence
electrons. With these we can fill four bands completely (Sect. 1.6). Thus, in the ground state
the lowest four bands constitute the completely filled valence bands, while all the upper ones
are empty and represent conduction bands. The energy gap between the valence and conduction
bands (top of valence band at k = 0 and bottom of conduction bands at k0 6= 0) is smallest for
a non-vanishing connecting k-vector. We call this an indirect gap. This is the case for both C
and Si. A typical example for a direct gap semiconductor is GaAs where the two sub lattices of
the crystal structure in Fig. 2.2 are occupied by an atom of the group III and V, respectively,
in the periodic table.

Si

"
L

#
!

"

#
!

Figure 2.4: Band structure of C and Si.

37

Note that generally energy gaps in semiconductors and insulators are classified to be direct if the
wave-vector connecting the maximum of the valence band and the minimum of the conduction
band vanishes. Otherwise a gap is called indirect (see Fig. 2.5).
E

direct energy gap

indirect energy gap

Figure 2.5: Illustration of direct and indirect band gaps.

List of some facts about these materials of the group IV and compounds combining group III
and V:
Carbon has an energy gap of around 5.5eV in the diamond structure. Thus, in this
configuration it is not a semiconductor but an insulator. The large energy gap causes the
transparency of diamond in the visible range (1.5 - 3.5eV), as the electromagnetic energy
in this range cannot be absorbed by the electrons.
The energy gap of silicon is 1.12eV and thus much smaller; furthermore, it is indirect.
Germanium has an indirect gap of 0.67eV.
GaAs and GaN have a direct energy gap of 1.43eV and 3.5eV , respectively.

2.2

Elementary excitations in semiconductors

We consider a simple two-band model to illustrate the most basic properties of the excitation
spectrum of a semiconductor. The Hamiltonian is given by
X
X
H=
V,k b
cV,k,s b
cV,k,s +
C,k b
cC,k,s b
cC,k,s ,
(2.5)
k,s

k,s

where V,k and C,k are the band energies of the valence band and conduction band, respectively
(Fig. 2.6).
The operator cnks (cnks ) creates (annihilates) an electron with (pseudo-)momentum k and spin
s in the band n, n {V, C}. In the ground state |0 i,
Y
|0 i =
b
cV,k,s |0i,
(2.6)
k,s

the valence band is completely filled, whereas the conduction band is empty. The product on the
right-hand side runs over all wave vectors in the first Brillouin zone. The ground state energy
is given by
X
E0 = 2
V,k .
(2.7)
k

38

kp

conduc,on'band'

C,k

Eg

V,k

valence'band'
kp

/a

+/a

Figure 2.6: Schematic band structure of a direct-gap semiconductor with a k p-approximation

around the top of the valence and the bottom of the conduction band.

The total momentum and spin of the ground state vanish. We next consider single electron
excitations from that ground state.2

2.2.1

Electron-hole excitations

A simple excitation of the system consists of removing an electron (i.e., creating a hole) from
the valence band and inserting it into the conduction band. We write such an excitation as
|k + q, s; k, s0 i = b
cC,k+q,s b
cV,k,s0 |0 i = bk,q,ss0 |0 i,

(2.12)

where we remove an electron with pseudo-momentum k and spin s0 in the valence band and
replace it by an electron with k + q and s in the conduction band. The possibility of changing
the spin from s0 to s and of shifting the wave vector of conduction electrons by q is included.
2

Where is the chemical potential? We assume a simple band structure for a direct-gap semiconductor based
on the k p approximation (Fig. 2.6):
V,k =

~2 k2
2mV

and

C,k = Eg +

~2 k2
2mC

Now we calculate the electron distribution. The overall electron density is n:

Z
X X
d3 k
n=
1=2
1.
3
BZ (2)
s

(2.8)

(2.9)

kBZ

The density of electrons in the conduction and valence band for small finite temperatue are given by
Z
Z
Z
d3 k
d3 k
1
d3 k (C,k )
e(Eg )
nC
=2
n
(
)
=

e
=
(2mC kB T )3/2 ,
F C,k
3
3
3
(
)
C,k
4 3
+1
BZ (2)
BZ 4 e
BZ 4
d3 k (V,k )
e
e
=
(2mV kB T )3/2 .
3
4 3
BZ
BZ
BZ 4
(2.10)
Here we assumed that both and Eg are much larger than kB T ( = 1/kB T ). Now we see that electron
count gives nC + nV = n such that we can set the two lines of the above equation equal to obtain for the chemical
potential,


Eg
3
mV
=
+ kB T ln
.
(2.11)
2
4
mC
Z

n nV

d3 k
(1 nF (V,k )) =
4 3

d3 k
1

4 3 e(V,k ) + 1

At T = 0 the chemical potential is exactly in the center of the band gap and then moves with increasing T , if the
effective masses are different. Note also that for small temperatures the thermally activated carriers, electrons
and holes, are well described by the classical Maxwell-Boltzmann distribution, as was used in the approximation
of the integrals.

39

Furthermore |k + q, s; k, s0 i is assumed to be normalized. The electron-hole pair may either be

in a spin-singlet (pure charge excitation s = s0 ) or a spin-triplet state (spin excitation s 6= s0 ).
Apart from spin, the state is characterized by the wave vectors k and q. The excitation energy
can be obtained by
Ek,q,ss0 |k + q, s; k, s0 i = [H, bk,q,ss0 ]|0 i = H|k + q, s; k, s0 i bk,q,ss0 H|0 i ,
|
{z
}

(2.13)

E0 |k+q,s;k,s0 i

such that,
Ek,q,ss0 = C,k+q V,k .

(2.14)

The spectrum of the electron-hole excitations with fixed q is determined by the spectral function
X
I(q, E) =
|hk + q, s; k, s0 |b
cC,k+q,s b
cV,ks0 |0 i|2 (E (C,k+q V,k )).
(2.15)
k,s,s0

Excitations exist for all pairs E and q for which I(q, E) does not vanish, thus, only above a qdependent threshold, which is minimal for q = k0 , where k0 = 0 (k0 6= 0) for a direct (indirect)
energy gap. As k is arbitrary, there is a continuum of excited states above the threshold for
each q (see Figure 2.7).

E
electron(hole*
con+nuum*

electron(hole*
con+nuum*

Eg

Eg

indirect*gap*

direct*gap*

Figure 2.7: Electron-hole excitation spectrum for direct and indirect gap semiconductors, excitation energy E versus momentum transfer q. Excitations exist in the shaded region, where
I(q, E) 6= 0.
For the electron-hole excitations considered here, interactions among them was assumed to
be irrelevant, and the electrons involved are treated as non-interacting particles. Note the
analogy with the Dirac-sea in relativistic quantum mechanics: The electron-hole excitations of
a semiconductor correspond to electron-positron pair creation in the Dirac theory.

2.2.2

Excitons

Taking into account the Coulomb interaction between the electrons, there is another class of
excitations called excitons. In order to discuss them, we extend the Hamiltonian (2.5) by the
Coulomb interaction,
2
XZ
b (r)
b 0 (r 0 ) e
b (r 0 )
b (r),
Vb =
d3 r d3 r0

(2.16)
s
s
s
0 | s0
|r

r
0
s,s

40

where the field operators are defined by

X X
b (r) = 1

un,k (r)eikr b
cn,ks ,
s
n=V,C k

(2.17)

where un,k (r) are the Bloch functions of the band n = C, V . Now, we consider a general
particle-hole state,
X
X
|q i =
A(k)b
cC,k+q,s b
cV,k,s0 |0 i =
A(k)|k + q, s; k, s0 i,
(2.18)
k

and demand that it satisfies the stationary Schrodinger equation (H + Vb )|q i = (E0 + Eq )|q i.
This two-body problem can be expressed as
X
hk + q, s; k, s0 |H + Vb |k0 + q, s; k0 , s0 iA(k0 ) = (E0 + Eq )A(k).
(2.19)
k0

The matrix elements are given by

hk + q, s; k, s0 |H|k0 + q, s; k0 , s0 i = k,k0 {E0 + C,k+q V,k }

(2.20)

and
hk + q, s; k, s0 |Vb |k0 + q, s; k0 , s0 i =
Z
2ss0
e2
0
d3 r d3 r0 uC,k+q (r)uV,k (r)uC,k0 +q (r 0 )uV,k0 (r 0 )eiq(rr )
2

|r r 0 |
Z
0
e2
1
0
, (2.21)
2 d3 r d3 r0 uC,k+q (r)uV,k (r 0 )uC,k0 +q (r)uV,k0 (r 0 )ei(k k)(rr )

|r r 0 |
The first term is the exchange term, and the second term the direct term of the Coulomb
interaction. Now we consider a semiconductor with a direct energy gap at the -point. Thus,
the most important wave vectors are those around k = 0. We approximate
Z
1
1
un,k0 (r)un,k (r)
(2.22)
d3 run,k0 (r)un,k (r) = hun,k0 |un,k i 1,

which is reasonable for k k0 (= k + q). In the same manner, we see that

uCk+q (r)uV,k (r)

1
1
huC,k+q |uV,k i huC,k |uV,k i = 0.

(2.23)

Note that the semiconductor is a dielectric medium with a dielectric constant (D = E).
Classical electrodynamics states that
E =

4
,

(2.24)

i.e., the Coulomb potential is partially screened due to dielectric polarization. Including this
effect in the Schr
odinger equation phenomenologically, the matrix element (2.21) takes on the
form
4e2
hk + q, s; k, s0 |Vb |k0 + q, s; k0 , s0 i =
.
(2.25)
|k k0 |2

Thus, we can write the stationary equation (2.19) as




1 X 4e2
C,k+q V,k E A(k)
A(k0 ) = 0.
0 |k k0 |2
k

41

(2.26)

We include the band structure using the k p - approximation which, for a direct energy gap,
leads to
~2 k2
~2 k2
C,k =
+ Eg and V,k =
.
(2.27)
2mC
2mV
For simplicity we assume now that mV = mC = m , We define a so-called envelope function
F (r) by
1 X
F (r) =
A(k)eikr .
(2.28)
k
This function satisfies the differential equation



 2 2
~2 q 2
~
e2
F (r) = E Eg
F (r),

2ex
|r|
2Mex

(2.29)

1 = 2/m and total mass M = 2m .

where ex is the reduced mass, i.e., ex
ex
The stationary equation (2.29) is equivalent to the Schrodinger equation of a hydrogen atom.
The energy levels then are given by

Eq = Eg

ex e4
~2 q 2
+
,
22 ~2 n2 2Mex

(2.30)

which implies that there are excitations below the particle-hole continuum, corresponding to
particle-hole bound states. This excitation spectrum is discrete and there is a well-defined
relation between energy and momentum (q), which is the wave vector corresponding to the
center of mass of the particle-hole pair. This non-trivial quasiparticle is called exciton. In the
present approximation it takes on the form of a simple two-particle state. In fact, however, it
may be viewed as a collective excitation, as the dielectric constant includes the polarization by
all electrons. When the screening is neglected, the excitonic states would not make sense as
their energies would not be within the band gap but much below. For the case of weak binding
considered above, the excitation is called a Wannier exciton. The typical binding energy is
Eb

ex
Ry.
m2

(2.31)

Typical values of the constants on the right-hand side are 10 and ex m/10, so that the
binding energy is in the meV range. This energy is much smaller than the energy gap, such that
the excitons are inside the gap, as shown schematically in Figure 2.8.

E
electron,hole)
con.nuum)
excitons)

Eg

Figure 2.8: Qualitative form of the exciton spectrum below the electron-hole continuum.
The exciton levels are dispersive and their spectrum becomes increasingly dense with increasing
energy, similar to the hydrogen atom. When they merge with the particle-hole continuum the
42

bound state is ionized, i.e., the electron and the hole dissociate and behave like independent
particles.
Strongly bound excitons are called Frenkel excitons. In the limit of strong binding, the pair is
almost local, so that the excitation is restricted to a single atom rather than involving the whole
semiconductor band structure.
Excitons are mobile, but they carry no charge, as they consist of an electron and a hole with
opposite charges. Their spin quantum number depends on s and s0 . If s = s0 the exciton is a
spin singlet, while for s 6= s0 it has spin triplet character, both corresponding to integer spin
quasiparticles. For small densities they approximately obey Bose-Einstein statistics, as they
are made from two fermions. In special cases, Bose-Einstein condensation of excitons can be
observed experimentally.

2.2.3

Optical properties

Excitation in semiconductors can occur via the absorption of electromagnetic radiation. The
energy and momentum transferred by a photon is ~ and ~q, respectively. With the linear light
dispersion relation = c|q| and the approximation Eg 1eV e2 /a, we can estimate this
momentum transfer in a semiconductor

~
e2 2
2
2
q= =
2
=

,
(2.32)
c
hc
hc a
a
a
where c denotes the speed of light, a the lattice constant, and 1/137 the fine structure
constant. With this, the momentum transfer from a photon to the excited electron can be
ignored. In other words, pure electromagnetic excitations lead only to direct excitations. For
semiconductors with a direct energy gap (e.g., GaAs) the photo-induced electron-hole excitation
is most easy and yields absorption rates with the characteristics

(~ Eg )1/2 , dipole-allowed,
abs ()
(2.33)
(~ E )3/2 , dipole-forbidden.
g
Here, the terms dipole-allowed and dipole-forbidden have a similar meaning as in the excitation of atoms regarding whether matrix elements of the type huV,k |r|uC,k i are finite or vanish,
respectively. Obviously, dipole-allowed transitions occur at a higher rate for photon energies
immediately above the energy gap Eg , than for dipole-forbidden transitions.
For semiconductors with indirect energy gap (e.g., Si and Ge), the lowest energy transition connecting the top of the valence band to the bottom of the conduction band is not allowed without
the help of phonons (lattice vibrations), which contribute little energy but much momentum
transfer, as ~Q  ~ with Q = cs |Q| and the sound velocity cs  c. The requirement of a
phonon assisting in the transition reduces the transition rate to
abs () c+ (~ + ~Q Eg )2 + c (~ ~Q Eg )2 ,

(2.34)

where c are constants and Q corresponds to the wave vector of the phonon connecting the top
of the valence band and the bottom of the conduction band. There are two relevant processes:
either the phonon is absorbed (c+ -process) or it is emitted (c -process) (see Figure 2.9).
In addition to the absorption into the particle-hole spectrum, absorption processes inducing
exciton states exist. They lead to discrete absorption peaks below the absorption continuum.
In Figure 2.10, we show the situation for a direct-gap semiconductor.
Naturally, the recombination of electrons and holes is important as well; in particular, if it is
a radiative recombination, i.e., leads to the emission of a photon. Additionally, other recombination channels such as recombination at impurities, interfaces and through Auger processes
are possible. The radiative recombination for the direct-gap semiconductors is most relevant for
applications. The photon emission rate follows the approximate law
em () [N () + 1](~ Eg )1/2 e~/kB T ,
43

(2.35)

~Q

~Q
Phonon

Phonon

Photon

k
~ Photon

k
Eg

Eg

E = ~ ~Q

E = ~ + ~Q

Figure 2.9: Phonon-assisted photon absorption in a semiconductor with indirect gap: phonon
absorption (left panel) and phonon emission (right panel).
excitons

n=1
n=2

electron-hole
continuum

n=3

Eg

Figure 2.10: Absorption spectrum including the exciton states for a direct-gap semiconductor
with dipole-allowed transitions. The exciton states appear as sharp lines below the electron-hole
continuum starting at ~ = Eg .

with the photon density N (). This yields the dominant rate for ~ very close to Eg .

2.3

Doping semiconductors

Let us replace a Si atom in a Si semiconductor by aluminum Al (group III) or phosphorus P

(group V), which then act as impurities in the crystal lattice. Both Al and P are in the same
row of the periodic table, and their electron configurations are given by
Al :

[(1s)2 (2s)2 (2p)6 ] (3s)2 (3p),

P:

[(1s)2 (2s)2 (2p)6 ] (3s)2 (3p)3 .

The compound Al (P) has one electron less (more) than Si.

2.3.1

Impurity state

We consider the case of a P-impurity contributing an additional electron whose dynamics is

governed by the conduction band of the semiconductor. For the sake of simplicity, we describe
44

the conduction band by a single isotropic band with effective mass mC ,

k =

~2 k2
+ Eg .
2mC

(2.36)

In the neutral Si background, the phosphorus (P) ion represents a positively charged center,
which attracts its additional electron. In the simplest model, this situation is described by the
so-called Wannier equation

 2 2
e2
~
F (r) = EF (r),
(2.37)

2mC
|~r|
which is nothing else than the static Schrodinger equation for the hydrogen atom, where the
dielectric constant measures the screening of the ionic potential by the surrounding electrons.
Analogous to the discussion of the exciton states, F (r) is an envelope wave function of the
electron. Therefore, the low energy states of the additional electron are bound states around
the P ion. The electron may become mobile when this reduced hydrogen atom is ionized. The
binding energy relative to the minimum of the conduction band given by
En Eg =

mC e4
m
= 2C 2 Ry,
2
2
2
2~ n
m n

(2.38)

for n N and the effective radius (corresponding to the renormalized Bohr radius in the material)
of the lowest bound state reads
r1 =

~2
m
=
a ,
m e2
mC B

(2.39)

where aB = 0.53
A is the Bohr radius for the hydrogen atom. For Si we find mC 0.2m and
12, such that
E1 20meV

(2.40)

r1 30
A.

(2.41)

and

Thus, the resulting states are weakly bound, with energies inside the band gap. We conclude
that the net effect of the P-impurities is to introduce additional electrons into the crystal, whose
energies lie just below the conduction band (Eg 1eV while Eg E1 10meV). Therefore,
they can easily be transferred to the conduction band by thermal excitation (ionization). One
speaks of an n-doped semiconductor (n: negative charge). In full analogy one can consider
Al-impurities, thereby replacing electrons with holes: An Al-atom introduces an additional hole
into the lattice which is weakly bound to the Al-ion (its energy is slightly above the band edge
of the valence band) and may dissociate from the impurity by thermal excitation. This case is
called p-doping (p: positive charge). In both cases, the chemical potential is tied to the dopand
levels, i.e., it lies between the dopand level and the valence band for p-doping and between the
dopand level and the conduction band in case of n-doping (Figure 2.11).
The electric conductivity of semiconductors (in particular at room temperature) can be tuned
strongly by doping with so-called donors (n-doping) and acceptors (p-doping). Practically
all dopand atoms are ionized, with the electrons/holes becoming mobile. Combining differently
doped semiconductors, the possibility to engineer electronic properties is enhanced even more.
This is the basic reason for the semiconductors being ubiquitous in modern electronics.

45

conduction band

conduction band

conduction band

impurity levels

impurity levels

valence band

valence band

valence band

no doping

n-doped

p-doped

Figure 2.11: Position of the chemical potential in semiconductors.

2.3.2

Carrier concentration

Let us briefly compare the carrier concentration in doped and undoped semiconductors at room
temperature. Carriers are always created in form of electron hole pairs, following the reaction
formula
e + h ,

(2.42)

where denotes a photon which is absorbed (e-h-creation) or emitted (e-h-recombination) and

accounts for the energy balance. The carrier concentration is described by a mass action law of
the form,
 3
T
2
ne nh = n0
eEg /kB T = n2 (T ),
(2.43)
T0
where T0 , n0 and Eg are parameters specific to the semiconductor. In the case of undoped

2
silicon at T = 300K, ne nh 1020 cm3 . Thus, for the undoped semiconductor we find
ne = nh 1010 cm3 . On the other hand, for n-doped Si with a typical donor concentration of
nD 1017 cm3 we can assume that most of the donors are ionized at room temperature such
that
1017 cm3

(2.44)

n2 (T )
103 cm3 .
ne

(2.45)

ne nD
and
nh =

We conclude, that in n-doped superconductors the vast majority of mobile carriers are electrons,
while the hole carriers are negligible. The opposite is true for p-doped Si.

2.4

Semiconductor devices

Semiconductors are among the most important components of current high-technology. In this
section, we consider a few basic examples of semiconductor devices.

2.4.1

pn-contacts

The so-called pn-junctions, made by bringing in contact a p-doped and an n-doped version
of the same semiconductor, are used as rectifiers.3 When contacting the two types of doped
3

dt. Gleichrichter

46

semiconductors the chemical potential, which is pinned by the dopand (impurity) levels, determines the behavior of the electrons at the interface. In electrostatic equilibrium, the chemical
potential is constant across the interface. This is accompanied by a band bending leading
to the ionization of the impurity levels in the interface region (see Figure 2.12). Consequently,
these ions produce an electric dipole layer which induces an electrostatic potential shift across
the interface. Additionally, the carrier concentration is strongly reduced in the interface region
(depletion layer).
ionized

electric
dipole

p-doped

n-doped

Figure 2.12: Occupation of the impurity levels of a pn-junction.

In the absence of a voltage U over the junction, the net current flow vanishes because the dipole
is in electrostatic equilibrium. This can also be interpreted as the equilibrium of two oppositely
directed currents, called the drift current Jdrift and the diffusion current Jdiff . From the point
of view of the electrons, the dipole field exerts a force pulling the electrons from the p-side to
the n-side. This leads to the drift current Jdrift . On the other hand, the electron concentration
gradient leads to the diffusion current Jdiff from the n-side to the p-side. The diffusion current
is directed against the potential gradient, so that the diffusing electrons have to overcome a
potential step. The equilibrium condition for U = 0 is given by
0 = Jtot (U = 0) = Jdiff + Jdrift C1 (T )eEg /kB T C2 (T )eEg /kB T = 0,

(2.46)

where C1 = C2 = C. Both currents are essentially determined by the factor C(T )eEg /kB T . For
the drift current, the exponential behavior eEg /kB T stems from the dependence of the current
on the concentration of mobile charge carriers (electrons and holes on the p-side and n-side,
respectively), which are created by thermal excitation (Boltzmann factor). Applying a voltage
does not change this contribution significantly. For the diffusion current however, the factor
C(T )eEg /kB T describes the thermal activation over the dipole barrier, which in turn strongly
depends on the applied voltage U . For zero voltage, the height of the barrier Eb is essentially
given by the energy gap Eb Eg . With an applied voltage, this is modified according to
Eb Eg eU , where eU = n p is the difference of the chemical potentials between the
n-side and the p-side. From these considerations, the well-known current-voltage characteristic
of the pn-junctions follows directly as


Jtot (U ) = C(T )eEg /kB T eeU/kB T 1 .
(2.47)
For U > 0, the current is rapidly enhanced with increasing voltage. This is called forward bias.
By contrast, charge transport is suppressed for U < 0 (reverse bias), leading to small currents
only. The current-voltage characteristics J(U ) (see Figure 2.13) shows a clearly asymmetric
behavior, which can be used to rectify ac-currents. Rectifiers (or diodes) are an important
component of many integrated circuits.
47

J
eU

p-doped

eU

eU

eU

p-doped

n-doped

reverse bias

forward bias
U

n-doped

Figure 2.13: The pn-junction with an applied voltage and the resulting J-U characteristics.

2.4.2

Semiconductor diodes

Light emitting diode

As mentioned above, the recombination of electrons and holes can lead to the emission of photons
(radiative recombination) with a rather well-defined frequency essentially corresponding to the
energy gap Eg . An excess of electron-hole pairs can be produced in pn-diodes by running a
current in forward direction. Using different semiconductors with different energy gaps allows
to tune the color of the emitted light. Direct-gap semiconductors are most suitable for this kind
of devices. Well-know are the semiconductors of the GaAs-GaN series (see table 2.1). These
techniques are commonly used in LED (light emitting diode) lamps.
There appear efficiency problems concerning the emission of light by semiconductors. In
semiconductor
wave length (nm)
color

GaAs
940
infrared

GaAs0.6 P0.4
660
red

GaAs0.4 P0.6
620
yellow

GaP
550
green

GaN
340
ultraviolet

Table 2.1: Materials commonly used for LEDs and their light emitting properties.
particular, the difference in refractive indices inside nSC 3 and outside nair 1 the device
leads to large reflective losses. Thus, the efficacy of diode light sources, defined as the number
of photons emitted per created particle-hole pair, is small, but still larger than the efficiency of
conventional dissipative light bulbs.
Solar cell
Inversely to the previous consideration, the population of charge carriers can be changed by the
absorption of light. Suppose that the n-side of a diode is exposed to irradiation by light, which
leads to an excess of hole carriers (minority charge carriers). Some of these holes will diffuse
towards the pn-interface and will be drawn to the p-side by the dipole field. In this way, they
induce an additional current JL modifying the current-voltage characteristics to
Jtot = Jpn JL = Js (eeU/kB T 1) JL .

(2.48)

It is important for the successful migration of the holes to the interface dipole that they do not
recombine too quickly. When Jtot = 0, the voltage drop across the diode is


kB T
JL
UL =
ln
+1 .
(2.49)
e
Js
The maximum efficiency is reached by applying an external voltage Uc < UL such that the
product Jc Uc is maximized, where Jc = Jtot (U = Uc ) (cf. Figure 2.14).

48

contacts

non-reflecting layer

J
U
UL

n
p

maximal
power
rectangle

JL

Figure 2.14: Solar cell design and shifted current-voltage characteristics. The efficiency is maximal for a maximal area of the power rectangle.

2.4.3

MOSFET

The arguably most important application of semiconductors is the transistor, an element existing with different architectures. Here we shortly introduce the MOSFET (Metal-OxideSemiconductor-Field-Effect-Transistor). A transistor is a switch allowing to control the current
through the device by switching a small control voltage. In the MOSFET, this is achieved by
changing the charge carrier concentration in a p-doped semiconductor using a metallic gate.
The basic design of a MOSFET is as follows (see Figure 2.15): A thin layer of SiO2 is deposited
on the surface of a p-type semiconductor. SiO2 is a good insulator that is compatible with the
lattice structure of Si. Next, a metallic layer, used as a gate electrode, is deposited on top of
the insulating layer.
z

p-type Si

x, y
source
n-type Si

SiO2 , insulator

metal gate

drain

n-type Si

Figure 2.15: Schematic design of a MOSFET device.

The voltage between the Si semiconductor and the metal electrode is called gate voltage UG .
The insulating SiO2 layer ensures that no currents flow between the electrode and the semiconductor when a gate voltage is applied. The switchable currents in the MOSFET flow between
the source and the drain which are heavily n-doped semiconductor regions.
Depending on the applied gate voltage UG three different regimes can be realized:
1. UG = 0
Virtually no current flows, as the conduction band of the p-doped semiconductor is empty. The
doping states (acceptor levels) are occupied by thermal excitations.
2. 0 <

e UG
Eg

<1

In this case, the energy of the Si bands is lowered, such that in a narrow region within the
p-doped Si the acceptor levels drops below the chemical potential and the states are filled with
electrons (or, equivalently, holes are removed). This depletion layer has the extension d measured
from the Si-SiO2 interface. The negative charge of the acceptors leads to a position-dependent
49

p-type Si
conduction band

p-type Si
conduction band

SiO2

SiO2

valence band
depleted
layer

valence band
depleted
layer

inversion
layer

UG small
z=0

UG large

d
z=0

Figure 2.16: Depletion layer at the SiO2 -Si interface for 0 < e UG < Eg (left panel) and the
inversion layer Eg < e UG (right panel).

potential (z), where z is the distance from the boundary between SiO2 and Si. This potential
(z) satisfies the simple one-dimensional Poisson equation
d2
4(z)
(z) =
,
dz 2

(2.50)

where the charge density originates in the occupied acceptor levels,

(
enA , z < d,
(z) =
0, z > d,

(2.51)

and nA is the density of acceptors. The boundary conditions are given by

(z = 0) = UG

and

(z = d) = 0.

(2.52)

The solution for 0 z d then reads

(z) =

2enA
(z d)2 ,

with

d2 =

UG
.
2enA

(2.53)

The thickness of the depletion layer increases with increasing gate voltage d2 UG .
3. 1 <

e UG
Eg

When the applied gate voltage is sufficiently large, a so-called inversion layer is created (cf.
Figure 2.16). Close to the boundary, the conduction band is bent down so that its lower edge
lies below the chemical potential. The electrons accumulating in this inversion layer providing
carriers connecting the n-type source and drain electrodes and producing a large, nearly metallic,
current between source and drain. Conduction band electrons accumulating in the inversion layer
behave like a two-dimensional electron gas. In such a system, the quantum Hall effect (QHE),
which is characterized by highly unusual charge transport properties in the presence of a large
magnetic field, can occur.

50

Chapter 3

Metals
The electronic states in a periodic atomic lattice are extended and have an energy spectrum
forming energy bands. In the ground state these energy states are filled successively starting
at the bottom of the electronic spectrum until the number of electrons is exhausted. Metallic
behavior occurs whenever in this way a band is only partially filled. The fundamental difference
that distinguishes metals from insulators and semiconductors is the absence of a gap for electronhole excitations. In metals, the ground state can be excited at arbitrarily small energies which
has profound phenomenological consequences.
We will consider a basic model suitable for the description of simple metals like the Alkali metals
Li, Na, or K, where the (atomic) electron configuration consists of closed shell cores and one
single valence electron in an ns-orbital. Neglecting the core electrons (completely filled bands),
we consider the valence electrons only and apply the approximation of nearly free electrons. The
lowest band around the -point is then half-filled. First, we will also neglect the influence of
the periodic lattice potential and consider the problem of a free electron gas subject to mutual
(repulsive) Coulomb interaction.

3.1

The Jellium model of the metallic state

The Jellium1 model is the probably simplest possible model of a metal that is able to describe
qualitative and to some extend even quantitative aspects of simple metals. The main simplification made is to replace the ionic lattice by a homogeneous positively charged background
(Jellium). The uniform charge density enion is chosen such that the whole system electrons
and ionic background is charge neutral, i.e. nion = n, where n is the electron density. In this
fully translational invariant system, the plane waves
1
k,s (r) = eikr

(3.1)

represent the single-particle wave functions of the free electrons. Here is the volume of the
system, k and s {, } denote the wave vector and spin, respectively. Assuming a cubic
system of side length L and volume = L3 we impose periodic boundary conditions for the
wave function
k,s (r + (L, 0, 0)) = k,s (r + (0, L, 0)) = k,s (r + (0, 0, L)) = k,s (r)

(3.2)

such that the reciprocal space is discretized as

k=
1

2
(n , n , n )
L x y z

Jellium originates form the word jelly (gelatin) and was first introduced by Conyers Herring.

51

(3.3)

where (nx , ny , nz ) Z3 . The energy of a single-electron state is given by k = ~2 k2 /2m (free

particle). The ground state of non-interacting electrons is obtained by filling all single particle
states up to the Fermi energy with two electrons. In the language of second quantization the
ground state is, thus, given by
Y Y
b
ck,s |0i
(3.4)
|0 i =
|k|kF

where the operators b

ck,s (b
ck,s ) create (annihilate) an electron with wave vector k and spin s.
The Fermi wave vector kF with the corresponding Fermi energy F = ~2 kF2 /2m is determined
by equating the filled electronic states with the electron density n. We have
Z
1 X
d3 k
4 kF3
n=
1=2
1
=
2
,
(3.5)

(2)3
3 (2)3
|k|kF ,s

which results in
kF = (3 2 n)1/3 .

(3.6)

Note kF is the radius of the Fermi sphere in k-space around k = 0.2

3.1.1

Theory of metals - Sommerfeld and Pauli

In a first step we neglect the interaction among the electrons and consider the electrons in the
metal simply as a Fermi gas. Then thermodynamic properties can be described by using the
Fermi-Dirac distribution function,
f (k ) =

1
e(k )/kB T

+1

(3.7)

and the density of states

N (E) =

X
k,s

(E k ) = 2

= 2
2

2m
~2

3/2
E

1/2



Z
2mE
d3 k
~2 k2

m
E
= 3 dk dk k 2 2 k
(2)3
2m
4
~ k
~

3N
=
2 F

E
F

1/2
,

(3.8)
for E > 0 with N = n as the total number of electrons. We first address the temperature
dependence of the chemical potential up second power in T for fixed electron number N , by
using the equation
Z +
Z
X
2
N=
f (k ) =
dE f (E)N (E) =
dEN (E) + (kB T )2 N 0 () + ,
(3.9)
6
0
0
k,s

where we used the Sommerfeld expansion assuming T  TF = F /kB .3 We now use

Z
Z F
dEN (E)
dEN (E) + ( F )N (F ) = N + ( F )N (F )
0

(3.14)

Note that the function kF (n) n(1/d) depends on the dimensionality d of the system.
Sommerfeld expansion: In the limit kB T  F the derivative f (E)/E is well concentrated around E = .
We consider

 Z +



Z +
f (E)
(E )2 00
f (E)
dE g(E)
=
dE g() + (E )g 0 () +
g () +

E
2
E

(3.10)


Z
g 00 () +
f (E)
2 00
2
2
= g() +
dE(E )
+ = g() +
g ()(kB T ) +
2
E
6

52

leading to
N N + ( F )N (F ) +

2
(k T )2 N 0 (F )
6 B

(T ) = F

N 0 (F )
2
(kB T )2
+ (3.15)
6
N (F )

with N 0 (F )/N (F ) = 1/2F . Now we also determine the internal energy
Z
U (T ) =

F

dE EN (E)f (E)

dE EN (E) +
0



2
(kB T )2 N 0 () + N ()
6


2
2
2 0
dE EN (E) + F ( F )N (F ) + (kB T ) N (F ) + (kB T )2 N (F )
6
6
{z
}
|


=0

= U0 +

2
(k T )2 N (F ) ,
6 B

(3.16)

where we used Eq.3.15. The specific heat is then given by

C=

1 U
2 2
=
k T N (F ) = T
T
3 B

(3.17)

and shows a T -linear behavior where is the Sommerfeld coefficient, proportional to the density
of states at the Fermi energy.
Next we consider the effect of a magnetic field coupling to the electron spin, so that k k,s =
k B sH with B the Bohr magneton and s = 1. We consider the magnetization due to the
spin polarization of the electrons,
Z

Z
B
M = B (N+ N ) =
dEN (E)f (E B H)
dEN (E)f (E + B H)
2
0
0




Z
f (E)
f (E)
2
dE
dEN (E)
2B H B HN (F )
= 2B HN (F ) .
E
E
0
0
(3.18)
By taking the derivative with respect to H we find for the susceptibility,

B
2

p =

M
= 2B N (F ) .
H

(3.19)

This is the Pauli paramagnetic susceptibility which is to lowest order temperature independent
and proportional to the density of states at the Fermi energy, like 4 .
and analogous
Z


dE

With the definition

Z E
g(E) =
dE 0 (E 0 )

g(E)
E


f (E) = g() +

dE (E)f (E) =

dE (E) +

Note that
Z

dx

2 00
g ()(kB T )2 + .
6

x2 ex
2
=
.
x
2
(e + 1)
3 2

2
(kB T )2 0 () + .
6

(3.11)

(3.12)

(3.13)

Temperature dependence of the p : In Eq.(3.18) we go beyond the lowest order approximation using Eq.(3.10),




Z
f (E)
2
M 2B H
dEN (E)
= 2B H N () +
(kB T )2 N 00 ()
(3.20)
E
6
0

53

3.1.2

Stability of metals - a Hartree-Fock approach

Now we would like to examine the stability of the Jelium model. For this purpose, we compute
the ground state energy of the Jellium system variationally, using the density n as a variational
parameter, which is equivalent to the variation of the lattice constant. In this way, we will
obtain an understanding of the stability of a metal, i.e. the cohesion of the ion lattice through
the itinerant electrons (in contrast to semiconductors where the stability was due to covalent
chemical bonding). The variational ground state shall be |0 i from Eq.(3.4) for given kF . The
Hamiltonian splits into four terms
H = Hkin + Hee + Hei + Hii

(3.23)

with
Hkin =
Hee

k b
cks b
cks

(3.24)

k,s

1X
=
2 0

b (r)
b 0 (r 0 )
d3 r d3 r0
s
s

s,s

Hei =
1
Hii =
2

XZ

d3 r d3 r0

d3 r d3 r0

e2 b
b (r)
0 (r 0 )
s
|r r 0 | s

ne2 b
b (r)
(r)
s
|r r 0 | s

(3.25)
(3.26)

n2 e 2
,
|r r 0 |

(3.27)

where we have used in second quantization language the electron field operators
X
b (r) = 1

b
c eikr
s
k k,s
X
b (r) = 1

b
ck,s eikr
s
k

(3.28)
(3.29)

The variational energy which we want to minimize with respect to n can be computed from
Eg = h0 |H|0 i and consists of four different contributions:
First we have the kinetic energy
X

Ekin = h0 |Hkin |0 i =
Z
= 2

k,s

k h0 |b
cks b
c | i
|
{zks 0}
= nks

d3 k
3
k nks = N F
3
(2)
5

(3.30)

(3.31)

where we used N = n the number of valence electrons and

1 |k| kF
nks =
.

0 |k| > kF

(3.32)

With Eq.(3.15) we write


N () N

N 0 (F )
2
F
(kB T )2
6
N (F )

which leads to

"
M

2B HN (F )

2
1
(kB T )2
6

(


N (F )

N 0 (F )
N (F )

2

N 0 (F )2
2
(kB T )2
,
6
N (F )

N 00 (F )

N (F )

(3.21)

)#
= p (T )H ,

(3.22)

and defines the temperature dependent spin susceptibility, which depends on details of the density of states.

54

Secondly, there is the energy resulting from the Coulomb repulsion between the electrons,
Z
1
e2 X
b (r)
b 0 (r 0 )
b 0 (r 0 )
b (r)| i
Eee =
d3 r d3 r0
h |
(3.33)
s
s
0
s
2
|r r 0 | 0 0 s
s,s
Z


1
e2
n2 G(r r 0 ) = EHartree + EFock .
(3.34)
=
d3 r d3 r0
0
2
|r r |
For this contribution we used the fact, that the two-particle correlation function from equation
(3.33) may be expressed5 as
X
b (r)
b 0 (r 0 )
b 0 (r 0 )
b (r)| i = n2 G(r r 0 )
h0 |
(3.42)
s
s
s
0
s
s,s0

where
9n2
G(r) =
2

kF |r| cos kF |r| sin kF |r|

(kF |r|)3

2
.

(3.43)

The Coulomb repulsion Hee between the electrons leads to two terms, called the direct or Hartree
term describing the Coulomb energy of a uniformly spread charge distribution, and the exchange
or Fock term resulting from the exchange hole that follows from the Fermi-Dirac statistics (Pauli
exclusion principle).
The third contribution originates in the attractive interaction between the (uniform) ionic background and the electrons,
Z
X
e2
b (r)| i
b (r)
(3.44)
Eei = d3 r d3 r0
n
h0 |
s
s
0
0
|r r | s
Z
e2
= d3 r d3 r0
n2 .
(3.45)
|r r 0 |
5

We shortly sketch the derivation of the pair correlation function. Using equations (3.28) and (3.29) we find
X
0
0
0
b s (r)
b 0 (r 0 )
b s0 (r 0 )
b s (r)|0 i = 1
cqs0 b
c q 0 s0 b
ck0 s |0 i.
(3.35)
cks b
h0 |
ei(kk )r ei(qq )r h0 |b
s
2

0
0
k,k ,q,q

We distinguish two cases:

First, consider s 6= s0 ,
cqs0 b
c q 0 s0 b
ck0 s |0 i = kk0 qq0 nks nqs0
h0 |b
cks b

(3.36)

leading to
b s (r)
b 0 (r 0 )
b s0 (r 0 )
b s (r)|0 i =
h0 |
s

1 X
n2
nks nq,s0 =
.
2

(3.37)

k,q

Secondly, assume s = s0 such that

h0 |b
cks b
cqs b
cq0 s b
ck0 s |0 i = (kk0 qq0 kq0 qk0 )nks nqs ,

(3.38)


X
i(qk)(rr 0 )
b s (r)
b s (r 0 )
b s (r 0 )
b s (r)|0 i = 1
h0 |
1

e
nks nq,s .
2

(3.39)

which in turn leads to

k,q

Both cases eventually lead to the result in equation (3.42) with

2
!2
Z
3
d
k
1 X ikr

G(r) = 2
e
nks
= 2
eikr

(2)3
k

1
= 2 2
2 r

(3.40)

|k|kF

ZkF

2
dk k sin kr = 2

and n = kF3 /3 2 (k = |k| and r = |r|).

55

1 sin kF r kF r cos kF r
2 2
r3

2
(3.41)

b (r) (r)| i corresponds to the uniform density n, as is

Were the expectation value h0 |
s
s
0
easily calculated from the definitions (3.28) and (3.29).
Finally we have the repulsive ion-ion interaction
Z
1
n2 e 2
Eii = h0 |Hii |0 i =
d3 r d3 r0
.
(3.46)
2
|r r 0 |

n2

1.2

G(r)

n21.0
0.8

n2 0.6
2 0.4

exchange""
hole"

0.2

0"

0"

2"

6"

4"

8"

10

kF |r|

b 0 (r 0 )
b (r)| i.
b (r)
b 0 (r 0 )
Figure 3.1: Pair correlation function n2 G(r r 0 ) = h0 |
s
s
0
s
s
It is easy to verify that the three contributions EHartree , Eei , and Eii compensate each other
to exactly zero. Note that these three terms are the only ones that would arise in a classical
electrostatic calculation, implying that the stability of metals relies purely on quantum effect.
The remaining terms are the kinetic energy and the Fock term. The latter is negative and reads


Z
2
sin kF |r| kF |r| cos kF |r| 2
9n2
3e2
3 e
EFock =
d r
=
N
k .
(3.47)
4
|r|
(kF |r|)3
4 F
Eventually, the total energy per electron is given by


Eg
3 ~2 kF2
3e2
2.21 0.916
=

k =

Ry
N
5 2m
4 F
rs2
rs

(3.48)

where 1Ry = e2 /2aB and the dimensionless quantity rs is defined via

n=

3
4d3

(3.49)

and
d
rs =
=
aB

9
4

1/3

me2
.
~2 kF

(3.50)

The length d is the average radius of the volume occupied by one electron. Minimizing the
energy per electron with respect to n is equivalent to minimize it with respect to rs , yielding
rs,min = 4.83, d 2.5
A (see Fig.3.2) and a density of n0 1.5 1024 cm3 . This corresponds
to a lattice constant of a = (4/3)1/3 d 4
A. This estimate is roughly in agreement with the
lattice constants of the Alkali metals : rs,Li = 3.22, rs,Na = 3.96, rs,K = 4.86. Note that in metals
the delocalized electrons are responsible for the cohesion of the positive background yielding a
stable solid.
56

Eg
[Ry]
N
0.3

rs,min = 4.83

0.2

0.1

10

rs

-0.1

-0.2

Figure 3.2: Total ground state Hartree-Fock energy as a function of rs . A minimum exists at
rs,min = 4.83.

The good agreement of this simple estimate with the experimental values is due to the fact that
the Alkali metals have only one valence electron in an s-orbital that is delocalized, whereas the
the core electrons are in a noble gas configuration and, thus, relatively inert. In the variational
approach outlined above correlation effects among the electrons due to the Coulomb repulsion
have been neglected. In particular, electrons can be expected to avoid each other not just
because of the Pauli principle, but also as a result of the repulsive interaction. However, for the
problem under consideration the correlation corrections turn out to be small for rs rs,min :
Etot
=
N

2.21 0.916

+ 0.062lnrs 0.096 + . . .
rs2
rs


Ry

(3.51)

which can be obtained from a more sophisticated quantum field theoretical analysis.

3.2

Charge excitations and the dielectric function

In analogy to semiconductors, the elementary excitations of metallic systems are the electronhole excitations, which for metals, however, can have arbitrarily small energies. One particularly
drastic consequence of this behavior is the strong screening of the long-ranged Coulomb potential.
As we will see, a negative test charge in a metal reduces the electron density in its vicinity, and
the induced cloud of positive charges, relative to the uniform charge density, weaken the Coulomb
potential as,
V (r)

1
r

V 0 (r)

er/l
r

(3.52)

i.e. the Coulomb potential is modified into the short-ranged Yukawa potential with screening
length l. In contrast to metals, the finite energy gap for electron-hole excitations the charge
distribution in semiconductors reduces the adaption of the system to perturbations, so that the
screened Coulomb potential remains long-ranged,
V (r)

1
r

V 0 (r)

1
.
r

(3.53)

As mentioned earlier, the semiconductor acts as a dielectric medium and its screening effects are
accounted for by the polarization of localized electric dipoles, i.e., the Coulomb potential inside
a semiconductor is renormalized by the dielectric constant .

57

3.2.1

Dielectric response and Lindhard function

We will now investigate the response of an electron gas to a time- and position-dependent weak
external potential Va (r, t) in more detail based on the equation of motion. We introduce the
Hamiltonian
X
XZ
b (r)
b (r)
H = Hkin + HV =
k b
cks b
cks +
d3 r Va (r, t)
(3.54)
s
s
s

k,s

where the second term is considered as a small perturbation. In a first step we consider the
linear response of the system to the external potential. On this level we restrict ourself to one
Fourier component in the spatial and time dependence of the potential,
Va (r, t) = Va (q, )eiqrit et ,

(3.55)

where 0+ includes the adiabatic switching on of the potential. To linear response this
potential induces a small modulation of the electron density of the form nind (r, t) = n0 +
nind (r, t) with
nind (r, t) = nind (q, )eiqrit .

(3.56)

Using equations (3.28) and (3.29) we obtain for the density operator in momentum space,
X
XZ
1X
b (r)eiqr = 1
b (r)
ck+qs =
b
cks b
bk,q,s ,
(3.57)
bq =
d3 r
s
s

s
k,s

k,s

ck+qs . The perturbation term HV now reads

where we define bk,q,s = b
cks b
Z
X
X
b (r)
b (r) = b V (q, )eit et = 1
HV =
d3 r Va (r, t)
bk,q,s Va (q, )eit et .
s
s
q a

s
k,s

(3.58)
The density operator bq (t) in Heisenberg representation is the relevant quantity needed to describe the electron density in the metal.
Linear response
We introduce the equation of motion for bk,q,s (t):
i~


 

d
bk,q,s = bk,q,s , H = bk,q,s , Hkin + HV
dt




= k+q k bk,q,s + b
cks b
cks b
ck+qs b
ck+qs Va (q, )eit et .

(3.59)
(3.60)

b = Tr[Ae
b H ]/Tr[eH ] and follow the linear response
We now take the thermal average hAi
scheme by assuming the same time dependence for hb
k,q,s (t)i eit+t as for the potential ,
so that the equation of motion reads,




(~ + i~)hb
k,q,s i = k+q k hb
k,q,s i + n0k,s n0k+q,s Va (q, )
(3.61)
where n0k,s = hb
cks b
cks i and, therefore,
nind (q, ) =

n0k+q,s n0k,s
1X
1X
hb
k,q,s i =
V (q, ).

k+q k ~ i~ a
k,s

(3.62)

k,s

With this, we define the dynamical linear response function as

n0k+q,s n0k,s
1X
0 (q, ) =

k+q k ~ i~

(3.63)

k,s

such that nind (q, ) = 0 (q, )Va (q, ), where 0 (q, ) is known to be the Lindhard function.
58

Coulomb interaction - Random phase approximation

So far we treated the linear response of the system to an external perturbation without considering feedback effects due to the interaction among electrons. In fact, the density fluctuation
n(r, t) can be thought as a source for an additional Coulomb potential Vn which can be determined by means of the Poisson equation,
2 Vn (r, t) = 4e2 n(r, t)

(3.64)

4e2
n(q, ).
q2

(3.65)

or in Fourier space
Vn (q, ) =

If we allow feedback effects in our system with external perturbation Va (q, ), the effective
potential V felt by the electrons is determined self-consistently via
V (q, ) = Va (q, ) + Vn (q, )
= Va (q, ) +

4e2
q2

(3.66)

n(q, ),

(3.67)

where
n(q, ) = 0 (q, )V (q, ).

(3.68)

The relation between V and Va may then be written as

V (q, ) =

Va (q, )
(q, )

(3.69)

with
(q, ) = 1

4e2
(q, ),
q2 0

(3.70)

where (q, ) is termed the dynamical dielectric function and describes the renormalization of
the external potential due to the dynamical response of the electrons in the metal. Extending
Eq.(3.68) to
n(q, ) = 0 (q, )V (q, ) = (q, )Va (q, ).
we define the response function (q, ) within random phase approximation
(q, ) =

0 (q, )
=
(q, )

0 (q, )
.
4e2
1 2 0 (q, )
q

(3.71)
6

to be
(3.73)

This response function (q, ) contains also effects of electron-electron interaction and comprises
information not only about the renormalization of potentials, but also on the excitation spectrum
of the metal.
6

The equation (3.73) can be written in the form of a geometric series,

"
#

2
4e2
4e2
(q, ) = 0 (q, ) 1 + 2 0 (q, ) +
0 (q, ) + .
q
q2

(3.72)

From the point of view of perturbation theory, this series corresponds to summing a limited subset of perturbative
terms to infinite order. This approximation is called Random Phase Approximation (RPA) and is based on the
assumption the phase relation between different particle-hole excitations entering the perturbation series are
random such that interference terms vanish on the average. This approximation is used quite frequently, in
particular, in the discussion of instabilities of a system towards an ordered phase.

59

3.2.2

Electron-hole excitation

For simple particle-hole excitations in metals, neglecting Coulomb interaction between the electrons, it is sufficient to study the bare response function 0 (q, ). We may separate 0 into its
real and imaginary part, 0 (q, ) = 01 (q, ) + i02 (q, ). Using the relation
 
1
1
+ i(z)
(3.74)
lim
=P
0+ z i
z
where the Cauchy principal value P of the first term has to be taken, we separate the Lindhard
function (3.63) into
!
n0,k+q n0,k
1X
01 (q, ) =
(3.75)
P

k+q k ~
k,s
X
02 (q, ) =
(n0,k+q n0,k )(k+q k ~)
(3.76)

k,s

The real part will be important later in the context of instabilities of metals. The excitation
spectrum is visible in the imaginary part which relates to the absorption of energy by the
electrons subject to a time-dependent external perturbation.7 Note that 02 (q, ) corresponds
to Fermis golden rule known from time-dependent perturbation theory, i.e. the transition rate
from the ground state to an excited state of energy ~ and momentum q.

Fermi&
sea&

Fermi&
sea&

k+q

Figure 3.3: Electron-hole excitation: hole with momentum k and electron with momentum k+q.

The relevant excitations originating from the Lindhard function are particle-hole excitations.
Starting from the ground state of a completely filled Fermi sea, one electron with momentum k
is removed and inserted again outside the Fermi sea in some state with momentum k + q (see
Figure 3.3). The energy difference is then given by
Ek,q = k+q k > 0.

(3.77)

In analogy to the semiconducting case, there is a continuum of particle-hole excitation spectrum

in the energy-momentum plane sketched in Figure 3.4. Note the absence of an energy gap for
excitations.

3.2.3

Collective excitation

For the electron-hole excitations the Coulomb interaction was ignored (by using 0 (q, ) instead
of (q, )), such that the bare Lindhard function provides information about the single particle
7

See Chapter 6 Linear response theory of the course Statistical Physics FS09.

60

E
electron(hole*
con+nuum*

plasma*
resonance*
~!p

q
Figure 3.4: Excitation spectrum in the E-q-plane. The large shaded region corresponds to
the electron-hole continuum and the sharp line outside the continuum represents the plasma
resonance which is damped when entering the continuum.

spectrum. Including the Coulomb interaction a new collective excitation will arise, the so-called
plasma resonance. For a long-ranged interaction like the Coulomb interaction this resonance
appears at finite frequency for small momenta q. We derive it here using the response function
(q, ). Assuming |q|  kF we expand 0 (q, ) in q, starting with
k+q = k + q k k +

(3.78)

n0
q k k +
(3.79)

Note that n0 /k = (k F ) at T = 0 and k k = ~v k is the velocity. Since we will deal
with states located at the Fermi energy here, v k = vF k/|k| is the Fermi velocity. This leads to
the approximation
Z
d3 k q v F (k )
0 (q, ) 2
(2)3 q v F i
"
#




Z +1
kF2 qvF cos
qvF cos 2
qvF cos 3
2
d cos
=
+
+
+
(3.80)
(2)2 1
~vF
+ i
+ i
+ i


k3 q2
3 vF2 q 2
1
+
(3.81)
2 F
3 m( + i)2
5 ( + i)2


n0 q 2
3 vF2 q 2
=
1+
.
(3.82)
m( + i)2
5 ( + i)2
n0,k+q = n0,k +

According to equation (3.70), we find

lim (q, ) = 1

|q|0

p2
2

(3.83)

for the dielectric function in the long wavelength limit (|q| 0), with
4e2 n0
.
m
We now use the result in Eq.(3.82) to approximate (q, ),
p2 =

n0 q 2 R(q, )2
m( + i)2 4e2 n20 R(q, )2
(
)
n0 q 2 R(q, )
1
1
=

2mp
+ i p R(q, ) + i + p R(q, )

(q, )

61

(3.84)

(3.85)
(3.86)

where we introduced


3vF2 q 2
.
R(q, ) = 1 +
5 2
2

(3.87)

Applying the relation (3.74) in Eq.(3.82) we obtain

i

 n0 q 2 R(q, p ) h
Im (q, )
( p R(q, p )) ( + p R(q, p ))
p

(3.88)

which yields a sharp excitation mode,

(
(q) = p R(q, p ) = p

3v 2 q 2
1 + F 2 +
10p

)
,

(3.89)

which is called plasma resonance with p as the plasma frequency. Similar to the exciton,
the plasma excitation has a well-defined energy-momentum relation and may consequently be
viewed as a quasiparticle (plasmon) which has bosonic character. When the plasmon dispersion
merges with the electron-hole continuum it is damped (Landau damping) because of the allowed
decay into electron-hole excitations. This results in a finite life-time of the plasmons within
the electron-hole continuum corresponding to a finite width of the resonance of the collective
excitation.
metal
Li
Na
K
Mg
Al

p(exp) [eV]
7.1
5.7
3.7
10.6
15.3

p(theo) [eV]
8.5
6.2
4.6
-

Table 3.1: Experimental values of the plasma frequency for different compounds. For the alkali
metals a theoretically determined p is given for comparison, using equation (3.84) with m the
free electron mass and n determined through rs,Li = 3.22, rs,Na = 3.96 and rs,K = 4.86.

It is possible to understand the plasma excitation in a classical picture. Consider negatively

charged electrons in a positively charged ionic background. When the electrons are shifted
uniformly by r with respect to the ions, a polarization P = n0 er results. The polarization
causes an electric field E = 4P which acts as a restoring force. The equation of motion for
an individual electron describes harmonic oscillations
m

d2
r = eE = 4e2 n0 r.
dt2

(3.90)

with the same oscillation frequency as in Eq.(3.84), the plasma frequency,

p2 =

4e2 n0
.
m

(3.91)

Classically, the plasma resonance can therefore be thought as an oscillation of the whole electron
gas cloud on top of a positively charged background.

62

+
Figure 3.5: Classical understanding of the plasma excitation.

3.2.4

Screening

Thomas-Fermi screening
Next, we analyze the potential V felt by the electrons exposed to a static field ( 0). Using
the expansion (3.79) we obtain
0 (q, 0) =

3n
1 k2
1X
(k F ) = 2 F = 0

~vF
2F

(3.92)

kT2 F
q2

(3.93)

k,s

and thus
(q, 0) = 1 +

with the so-called Thomas-Fermi wave vector kT2 F = 6e2 n0 /F . The effect of the renormalized
q-dependence of the dielectric function can best be understood by considering a bare point
charge Va (r) = e2 /r (or Va (q) = 4e2 /q 2 ) and its renormalization in momentum space
V (q) =

Va (q)
4e2
= 2
(q, 0)
q + kT2 F

(3.94)

or in real space
e2 kT F r
e
.
(3.95)
r
The potential is screened by a rearrangement of the electrons and this turns the long-ranged
Coulomb potential into a Yukawa potential with exponential decay. The new length scale is kT1
,
F
the so-called Thomas-Fermi screening length. In ordinary metals kT F is typically of the same
order of magnitude as kF , i.e. the screening length is of order 5
A comparable to the distance
8
between neighboring atoms. As a consequence also external electric fields cannot penetrate a
metal, but are screened on this length 1/kT F . This legitimates one of the basic assumptions
used in electrostatics with metals.
V (r) =

8
The Thomas-Fermi approach for electron gas is sketched in the following. The Thomas-Fermi theory for
the charge distributions slowly varying in space is based on the approximation that locally the electrons form a
Fermi gas with Fermi energy F and electron density ne (F ) neutralizing the ionic background. The electrostatic
potential (r) of an external charge distribution ex (r) induces a charge redistribution ind (r) relative to ne (F ).
Within Thomas-Fermi approximation the induced charge distribution can then be written as


ind (r) = e ne (F + e(~r)) ne (F )
(3.96)

with
ne (F ) =

kF3
1
=
(2mF )3/2
3 2
3 2 ~2

(3.97)

where F = ~2 kF2 /2m. This approach is justified, if the spacial change of the potential (r) is slow compared to
kF1 , so that locally we may describe the electron gas as a filled Fermi sphere of corresponding electron density.

63

Friedel oscillations
The static dielectric function can be evaluated exactly for a system of free electrons, resulting
for 3 dimensions in



4e2 mkF 1 4kF2 q 2  2kF + q 
.
(3.102)
(q, 0) = 1 +
+
ln 
q 2
2
8kF q
2kF q 
Noticeably the dielectric function varies little for small q  kF . At q = 2kF there is, however,
a logarithmic singularity. This is a consequence of the sharpness of the Fermi surface in k-space.
Consider the induced charge of a point charge at the origin: ena (r) = ena0 (r) which is Fourier
transformed is na (q) = na0 .9
Z
n(r) =

d3 q
(2)3


Z
1
1
iqr
1 na (q)e
=
g(q)na (q) sin qr dq
(q)
r

(3.104)

with
g(q) =

q (q) 1
.
2 2 (q)

(3.105)

Note that g(q) vanishes for both q 0 and q . Using partial integration twice, we find
n
n(r) = a0
r3

g 00 (q) sin qrdq

(3.106)

where
g 0 (q) A ln|q 2kF |

(3.107)

and
g 00 (q)

A
q 2kF

(3.108)

The Poisson equation may now be formulated as

"

#

ne () 
ex (r)
(r) = 4[ind (r) + ex (r)] 4 e (r)
 =F
2

= kT2 F (r) 4ex (r)

(3.98)
(3.99)

with the Thomas-Fermi momentum kT F defined as,

2

kT F


ne () 
6e2 ne
= 4e
=
.

 =F
F
2

(3.100)

and ne = ne (F ). For a point charge Q located at the origin we obtain,

(r) = Q

erkT F
.
r

(3.101)

This is the Yukawa potential as obtained above.

9
The charge distribution can be deduced from the Poisson equation (3.65):
en(q) =

Va (q)
1 (q, 0)
q2
Vn (q) = 0 (q, 0)V (q) = 0 (q, 0)
=
na (q, 0)
4e2
(q, 0)
(q, 0)

The charge distribution in real space can be obtained by Fourier transformation.

64

(3.103)

dominate around q 2kF . Hence, for kF r  1,

Ana0
n(r)
r3

2kZF +
2kF

sin[(q 2kF )r] cos 2kF r + cos[(q 2kF )r]sin2kF r

dq
q 2kF

(3.109)

cos 2kF r
.
(3.110)
r3
with a cutoff . The induced charge distribution exhibits so-called Friedel oscillations.
Finally we may ask what is the total electron charge displaced around the point charge ena0 (r).
We take Eq.(charge-displace) and integrate over r.


Z
1
Q = en = e d3 r n(r) = lim
1 na (q) = ena0
(3.111)
q0 (q)
Ana0

where we used 1/(q) 0 for q 0. The charge displacement corresponds to the exact opposite amount of charge of of the point charge. Thus we find a perfect compensation which
corresponds to perfect screening.

ni

standard
ThomasFermi

Lindhard form
(Friedel oscillation)

r
Figure 3.6: Friedel oscillations of the charge distribution.

Dielectric function in various dimensions

Above we have treated the dielectric function for a three-dimensional parabolic band. Similar
calculations can be performed for one- and two-dimensional systems. In general, the static
susceptibility is given by



s + 2
1

ln 
1D

2q  s 2 

(
)




1
4 1/2
0 (q, = 0) =
(3.112)

1 1 2
(s 2) ,
2D

2
s




 

s + 2
kF
s
4


 , 3D
1 2 ln 
2 1
2
4
s
s 2

where s = q/kF . Interestingly 0 (q, 0) has a singularity at q = 2kF in all dimensions. The
singularity becomes weaker as the dimensionality is increased. In one dimension, there is a
logarithmic divergence, in two dimensions there is a kink, and in three dimensions only the
derivative diverges. Later we will see that these singularities may lead to instabilities of the
metallic state, in particular for the one-dimensional case.

65

(q ,0)
(0,0)

1D

2D
3D

0
0

2k F

Figure 3.7: Lindhard functions for different dimensions. The lower the dimension the stronger
the singularity at q = 2kF .

3.3

Lattice vibrations - Phonons

The atoms in a lattice of a solid are not immobile but vibrate around their equilibrium positions.
We will describe this new degree of freedom by treating the lattice as a continuous elastic medium
(Jellium with elastic modulus ). This approximation is sufficient to obtain some essential
features of the interaction between lattice vibrations and electrons. In particular, renormalized
screening effects will be found. Our approach here is, however, limited to mono-atomic unit cells
because the internal structure of a unit cell is neglected.

3.3.1

Vibration of a isotropic continuous medium

The deformation of an elastic medium can be described by the displacement of the infinitesimal
volume element d3 r around a point r to a different point r 0 (r). We can introduce here the
so-called displacement field u(r) = r 0 (r) r as function of r. In general, u is also a function of
time. In the simplest form of an isotropic medium the elastic energy for small deformations is
given by
Z

Eel =
d3 r ( u(r, t))2
(3.113)
2
where is the elastic modulus (note that there is no deformation energy, if the medium is
just shifted uniformly). This energy term produces a restoring force trying to bring the system
back to the undeformed state. In this model we are neglecting the shear contributions.10 The
continuum form above is valid for deformation wavelengths that are much longer than the lattice
constant, so that details of the arrangement of atoms in the lattice can be neglected. The kinetic
energy of the motion of the medium is given by


Z

u(r, t) 2
(3.115)
Ekin = 0 d3 r
2
t
10

Note that the most general form of the elastic energy of an isotropic medium takes the form

Z
X 
Eel = d3 r
( u )( u ) + ( u )( u ) ,
2

(3.114)

,=x,y,z

where = /r . The Lame coefficients and characterize the elastic properties. The elastic constant
describes density modulations leading to longitudinal elastic waves, whereas corresponds to shear deformations
connected with transversely polarized elastic waves. Note that transverse elastic waves are not important for the
coupling of electrons and lattice vibrations.

66

where 0 = Mi ni is the mass density with the ionic mass Mi and the ionic density ni . Variation
of the Lagrangian functional L[u] = Ekin Eel with respect to u(r, t) leads to the equation of
motion
1 2
u(r, t) ( u(r, t)) = 0,
c2s t2

(3.116)

which is a wave equation with sound velocity c2s = /0 . The resulting displacement field can
be expanded into normal modes,

1 X 
u(r, t) =
ek qk (t)eikr + qk (t) eikr
(3.117)
k
where every qk (t) satisfies the equation
d2
q + k2 qk = 0,
dt2 k

(3.118)

with the frequency k = cs |k| = cs k and the polarization vector ek has unit length. Note that
within our simplification for the elastic energy (3.114), all modes correspond to longitudinal
waves, i.e. u(r, t) = 0 and ek k k. The total energy expressed in terms of the normal
modes reads
X
(3.119)
E=
0 k2 [qk (t)qk (t) + qk (t)qk (t)] .
k

Next, we switch from a Lagrangian to a Hamiltonian description by defining the new variables
Qk =

Pk =

Qk = ik 0 (qk qk )
dt

0 (qk + qk )

(3.120)
(3.121)

in terms of which the energy is given by

E=

1X 2
Pk + k2 Q2k .
2

(3.122)

Thus, the system is equivalent to an ensemble of independent harmonic oscillators, one for each
normal mode k. Consequently, the system may be quantized by defining the canonical conjugate
b which obey, by definition, the commutation relation,
operators Pk Pbk and Qk Q
k
b , Pb 0 ] = i~ 0 .
[Q
k
k
k,k

(3.123)

As it is usually done for quantum harmonic oscillators, we define the raising and lowering
operators


bb = p 1
b + iPb

Q
(3.124)
k
k k
k
2~k


bb = p 1
b iPb ,

Q
(3.125)
k k
k
k
2~k
satisfying the commutation relations
[bbk , bbk0 ] = k,k0 ,

(3.126)

[bbk , bbk0 ] = 0,
[bb , bb 0 ] = 0.

(3.127)

67

(3.128)

These relations can be interpreted in a way that these operators create and annihilate quasiparticles following the Bose-Einstein statistics. According to the correspondence principle, the
quantum mechanical Hamiltonian corresponding to the energy (3.122) is


X
1
H=
~k bbkbbk +
.
(3.129)
2
k

In analogy to the treatment of the electrons in second quantization we say that the operators bbk
(bbk ) create (annihilate) a phonon, a quasiparticle with well-defined energy-momentum relation,
b (r) can now
k = cs |k|. Using Eqs.(3.118, 3.121, and 3.125) the displacement field operator u
be defined as
s
1 X
~ hb ikr b ikr i
b (r) =
u
ek
b e
+ bk e
.
(3.130)
20 k k

As mentioned above, the continuum approximation is valid for long wavelengths (small k) only.
For wavevectors with k /a the discreteness of the lattice appears in the form of corrections
to the linear dispersion k |k|. Since the number of degrees of freedom is limited to 3Ni (Ni
number of atoms), there is a maximal wave vector called the Debye wavevector11 kD . We can now
define the corresponding Debye frequency D = cs kD and the Debye temperature D = ~D /kB .
In the continuous medium approximation there are only acoustic phonons. For the inclusion of
optical phonons, the arrangement of the atoms within a unit cell has to be considered, which
goes beyond this simple picture.

3.3.2

Phonons in metals

The consideration above is certainly valid for semiconductors, where ionic interactions are mediated via covalent chemical bonds and oscillations around the equilibrium position may be
approximated by a harmonic potential, so that the form of the elastic energy above is well motivated. The situation is more subtle for metals, where the ions interact through the long-ranged
Coulomb interaction and are held to together through an intricate interplay with the mobile
conduction electrons.
First, neglecting the gluey effect of the electrons, the positively charged background can itself be treated as an ionic gas. Similar to the electronic gas (3.84), the background exhibits a
well-defined collective plasma excitation at the ionic plasma frequency
2p =

4ni (Zi e)2

,
Mi

(3.131)

For equation (3.131) we used the formula (3.84) with n0 ni = n0 /Zi the density of ions with
charge number Zi , e Zi e, and m Mi the atomic mass. Apparently the excitation energy
does not vanish as k 0. So far, the background of the metallic system can not be described
as an elastic medium where the excitation spectrum is expected to be linear in k, k |k|.
The shortcoming in this discussion is that we neglected the feedback effects of the electrons that
react nearly instantaneously to the slow ionic motion, due to their much smaller mass. The
finite plasma frequency is a consequence of the long-range nature of the Coulomb potential (as
mentioned earlier), but as we have seen above the electrons tend to screen these potentials, in
particular for small wavevectors k. The bare ionic plasma frequency p is thus renormalized
to
k2 =
11

2p
(k, 0)

k 2 2p
k 2 + kT2 F

See course of Statistical Physics HS09.

68

(cs k)2 ,

(3.132)

where the presence of the electrons leads to a renormalization of the Coulomb potential by a
factor 1/(k, ). Having included the back-reaction of the electrons, a linear dispersion of a
sound wave (k = cs |k|) is finally recovered, and the renormalized velocity of sound cs reads
c2s

2p
k2

TF

Zmp2
Mi

k2

TF

1 m 2
= Z
v .
3 Mi F

For the comparison of the energy scales we find,

s
cs
D
1 m

=
Z
 1,
TF
vF
3 Mi

(3.133)

(3.134)

where we used kB TF = F and kD kF .

Kohn anomaly
Notice that phonon frequencies are much smaller than the (electronic) plasma frequency, so that
the approximation
2p

k2 =

(3.135)

(k, 0)

is valid even for larger wavevectors. Employing the Lindhard form of (k, 0), we deduce that
the phonon frequency is singular at |k| = 2kF . More explicitly we find
k

(3.136)

in the limit k 2kF . This behavior is called the Kohn anomaly and results from the interaction
between electrons and phonons. This effect is not contained in the previous elastic medium
model that neglected ion-electron interactions.

3.3.3

Peierls instability in one dimension

The Kohn anomaly has particularly drastic effects in (quasi) one-dimensional electron systems,
where the electron-phonon coupling leads to an instability of the metallic state. We consider a
one-dimensional Jellium model where the ionic background is treated as an elastic medium with a
displacement field u along the extended direction (x-axis). We neglect both the electron-electron
interaction and the slow time evolution of the background modulation so that the Hamiltonian
reads,
H = Hisol + Hint ,

(3.137)

where contributions of the isolated electronic and ionic systems are included in
Hisol =

X ~2 k 2
k,s

2m

ck s ck s

+
2

Z
dx

du
(x)
dx

2
(3.138)

whereas the interactions between the system comes in via the coupling
Hint = n0

XZ
s

dx dx0 V (x x0 )

d
b (x0 )
b (x0 )
u(x)
s
s
dx

(3.139)

In the general theory of elastic media u = n/n0 describes density modulations, so that
the second term in (3.137) models the coupling of the electrons to charge density fluctuations of
69

the positively charged background12 mediated by the screened Coulomb interaction V (x x0 )

(contact interaction). Consider the ground state of N electrons in a system of length L, leading to
an electronic density n0 = N/L. For a uniform background u(x) = const, the Fermi wavevector
of free electrons is readily determined to be
+kF
X L Z
L
N=
dk 1 = 2 2kF
2
2
s

(3.140)

kF

leading to
kF =

n .
2 0

(3.141)

Perturbative approach - instability

Now we consider the Kohn anomaly of this system. For a small background modulation u(x) 6=
const, the interaction term Hint can be treated perturbatively and will lead to a renormalization
of the elastic modulus in (3.137). For that it will be useful to express the full Hamiltonian in
momentum space,
Hisol =
Hint

X ~2 k 2
k,s

2m

ck s ck s +

0 X 2
u u
2 q q q q


i X 
=
q Vq uq b
ck+q,s b
ck,s Vq uq b
ck,s b
ck+q,s ,
2

(3.142)
(3.143)

k,q,s

where we used from previous considerations q 2 = 0 q2 . Furthermore we defined

1 X
u(x) =
uq eiqx ,
L q
1 X iqx
V (x) =
Vq e ,
L q

(3.144)
(3.145)

with Vq = 4e2 /q 2 (q, 0) const. - we consider an effectively short-ranged potential. We

compute the second order correction to the ground state energy using Rayleigh-Schrodinger
perturbation theory (note that the linear energy shift vanishes)
E (2) =

X q 2 |Vq |2
4

k,q,s

uq uq

X |h0 |b
ck,s b
ck+q,s |ni|2 + |h0 |b
ck+q,s b
ck,s |ni|2
E0 En

X q 2 |Vq |2
X nk+q nk
|
uq uq
4
k+q k
q

(3.146)

(3.147)

X
4

|Vq |2 q 2 0 (q, 0)uq uq

(3.148)

where the virtual states |ni are electron-hole excitations of the filled Fermi sea. This term gives
a correction to the elastic term in (3.142). In other words, the elastic modulus and, thus, the
phonon frequency q 2 = /0 q 2 = c2s q 2 is renormalized according to

2
qren

q2

|Vq |2 q 2
40

0 (q, 0) =

70

q2

|Vq |2 q



 q + 2kF 



ln
80  q 2kF 

(3.149)

q
q

(ren)

2k F

Figure 3.8: Kohn anomaly for the one-dimensional system with electron-phonon coupling. The
renormalization of the phonon frequency is divergent at q = 2kF .

From the behavior for q 0 we infer that the velocity of sound is renormalized. However, a much
more drastic modification occurs at q = 2kF . Here the phonon spectrum is softened, i.e. the
frequency vanishes and even becomes negative. The latter effect is an artifact of the perturbation
theory.13 This hints at an instability triggered by the Bose-Einstein condensation of phonons
with a wave vector of q = 2kF . This coherent superposition14 of many phonons corresponds
classically to a static periodic deformation of the ionic background with wave vector 2kF . The
unphysical behavior of the frequency q indicates that in the vicinity of 2kF , the current problem
can not be treated with the help of perturbation theory around the uniform state.
Peierls instability at Q = 2kF
Instead of the perturbative approach, we assume that the background shows a periodic density
modulation (coherent phonon state)
u(x) = u0 cos(Qx)

(3.154)

12

Note that only phonon modes with a finite value of u couple in lowest order to the electrons. This is only
possible of longitudinal modes. Transverse modes are defined by the condition u = 0 and do not couple to
electrons in lowest order.
13
Note that indeed the expression
q2 =

2p
(q, 0)

(3.150)

in (3.135) does not yield negative energies but gives a zero of q at q = 2kF .
14
We introduce the coherent state
2

||
|coh
Q i = e

/2

X
bQ )n n
(b
|0i
n!
n=0

(3.151)

which does not have a definite phonon number for the mode of wave vector Q. On the other hand, this mode is
macroscopically occupied, since
coh
2
b b
nQ = hcoh
Q |bQ bQ |Q i = ||

(3.152)

and, moreover, we find

hcoh
u(x)|coh
Q |b
Q i =

i
1 ~ h iQx
e
+ eiQx = u0 cos(Qx)
L 20 Q

with u0 = ~/0 LQ , assuming being real.

71

(3.153)

where Q = 2kF and u0 remains to be determined variationally. We investigate the effect of

this modulation on the electron-phonon system. To this end we show that such a modulation
lowers the energy of the electrons. Assuming that u0 is small we can evaluate the electronic
energy using the approximation of nearly free electrons, where Q appears as a reciprocal lattice
vector. The electronic spectrum for 0 k Q is then approximately determined by the secular
equation
!
~2 k2

2m
det
=0
(3.155)
~2 (kQ)2
E

2m
where derives from the Fourier transform of the potential V (x),
= iQu0 nVQ

(3.156)

with
VQ =

dx eiQx V (x).

(3.157)

The equation (3.155) leads to the energy eigenstates

Ek =

i
p
~2 h
(k Q)2 + k 2 {(k Q)2 k 2 }2 + 16m2 ||2 /~4 .
4m

(3.158)

The total energy of the electronic and ionic system is then given by
Etot (u0 ) = 2

Ek +

0k<Q

LQ2 2
u0
4

(3.159)

where all electronic states of the lower band (Ek ) are occupied and all states of the upper band
(Ek+ ) are empty. The amplitude u0 of the modulation is found by minimizing Etot with respect
to u0 :
0=

1 dEtot
L du0

(3.160)

~2 32Q2 m2 n2 VQ2
=
u0
2m
~4
4Qmn2 VQ2
~2

ZQ
0

dk
1

q
+ Q2 u0
2 {(k Q)2 k 2 }2 + 16m2 Q2 n2 V 2 u2 /~4
2
Q

+k
ZF

1
dq q
+ Q2 u0
2
q 2 + 4m2 n2 VQ2 u20 /~4
kF
!
8Qmn2 VQ2
~2 kF

= u0
arsinh
+ Q2 u0 .
2

~
2
2mnVQ u0
= u0

(3.161)

(3.162)

(3.163)

We solve this equation for u0 using arsinh(x) ln(2x) when x  1.

"
#
~2 kF
~2 kF
2 F 1/N (0)g
u0 =
exp
=
e
2
2

kF nV
mnV
8mn V
Q

(3.164)

where F = ~2 kF2 /2m is the Fermi energy and N (0) = 2m/~2 kF is the density of states at the
Fermi energy. We introduced the coupling constant g = 4n2 VQ2 / that describes the phononinduced effective electron-electron interaction. The coupling is the stronger the more polarizable

72

(softer) ionic background, i.e. when the elastic modulus is small. Note that the static displacement u0 depends exponentially on the coupling and on the density of states. The underlying
reason for this so-called Peierls instability to happen lies in the opening of an energy gap,


1
+

E = EkF EkF = 2|| = 8F exp

,
(3.165)
N (0)g
at k = kF , i.e. at the Fermi energy. The gap is associated with a lowering of the energy of the
electron states in the lower band in the vicinity of the Fermi energy. For this reason this kind
of instability is called a Fermi surface instability. Due to the gap the metal has turned into a
semiconductor with a finite energy gap for all electron-hole excitations.

+Q k

Figure 3.9: Change of the electron spectrum. The modulation of the ionic background yields
gaps at the Fermi points and the system becomes an insulator.
The modulation of the electron density follows the charge modulation due to the ionic lattice
deformation, which can be seen by expressing the wave function of the electronic states,
1 eikx + (Ek k )ei(kQ)x
p
k0 (x) =
,
(Ek k )2 + ||2

(3.166)

which is a superposition of two plane waves with wave vectors k and k Q, respectively. Hence
the charge density reads


2(k Ek )||
e
2
0
k (x) = e|k (x)| =
1
sin Qx
(3.167)

(Ek k )2 + ||2
and its modulation from the homogeneous distribution en is given by
ZkF 0
e
dk
m|| sin Qx
q
(x) =
k (x) (en) =
2
2 ~4 k 2 k 0 2 + m2 ||2
k
F
0


en||  2F 
=
ln
sin(2kF x).
16F  || 
X

(3.168)

(3.169)

Such a state, with a spatially modulated electronic charge density, is called a charge density
wave (CDW) state. This instability is important in quasi-one-dimensional metals which are for
example realized in organic conductors such as TTFTCNQ (tetrathiafulvalene tetracyanoquinomethane). In higher dimensions the effect of the Kohn anomaly is generally less pronounced, so
that in this case spontaneous deformations rarely occur. As we will see later, a charge density
wave instability can nevertheless be observed in multi-dimensional (d > 1) systems with a socalled nested Fermi surface. These systems resemble in some respects one-dimensional systems.
Finally, notice that the electron-phonon interaction strongly contributes to another kind of Fermi
surface instability, when metals exhibit superconductivity.
73

3.3.4

Dynamics of phonons and the dielectric function

We have seen that an external potential Va is screened by the polarization of the electrons.
As the positively charged ionic background is also polarizable, it should be included in the
renormalization of the external potential. In general, the fully renormalized potential Vren may
be expressed via
Vren = Va ,

(3.170)

with the full dielectric function . In order to determine Vren and , we define the bare (unrenormalized) electronic (ionic) dielectric function el (ion ). The renormalized potential in (3.170)
can be expressed considering three other points of view. First, if the ionic potential Vion is added
to the external potential Va , the remaining screening is due to the electrons only, i.e.,
el Vren = Va + Vion .

(3.171)

Secondly, the electronic potential Vel may be added to the external potential Va , so that the ions
exclusively renormalize the new potential Vel + Va , resulting in
ion Vren = Va + Vel .

(3.172)

Note that in (3.172) all effects of electron polarization are included in Vel , so that the dielectric
function results from the bare ions. Finally we use the fact that Vren may be expressed as
Vren = Va + Vel + Vion .

(3.173)

Adding (3.171) to (3.172) and subtracting (3.170), we obtain

(el + ion )Vren = Va + Vel + Vion

(3.174)

which simplifies with (3.173) to

= el + ion 1

(3.175)

In order to find an alternative expression relating the renormalized potential Vren to the external
potential Va , we make the Ansatz
1
1 1
Vren = Va = ion el Va

eff

(3.176)

i.e. the potential Va /el that results from bare screening of the polarizable electrons is additionally screened by an effective ionic dielectric function ion
eff which includes electron-phonon
ion
interactions. Using equation (3.175) and the definition of eff via (3.176) we obtain
ion
eff = 1 +

1 ion
( 1),
el

(3.177)

or using the definition (3.70)

ion
0,eff =

ion
0
.
el

(3.178)

Taking into account the discussion of the plasma excitation of the bare ions in Eqs.(3.83, 3.84,
and 3.131), and considering the long wave-length excitations (k 0), we approximate
ion = 1

2p

,
2
k2
el = 1 + T2F .
k
74

(3.179)
(3.180)

k,

k+q, +

q,

k,

kq,

Figure 3.10: Diagram for the electron-electron interaction involving also electron-phonon coupling.

For the electrons we used the result from the quasi-static limit in (3.93). The full dielectric
function now reads



2p
k2
kT2 F
kT2 F
=1+ 2 2 = 1+ 2
1 2 .
(3.181)
k

The time-independent Coulomb interaction

Va =

4e2
q2

(3.182)

between the electrons is replaced in a metal by an effective interaction

Vren (q, ) =

4e2
q 2 (q, )

4e2
= 2
kT F + q 2

(3.183)
2
2 q2

!
.

(3.184)

This interaction corresponds to the matrix element for a scattering process of two electrons
with momentum exchange q and energy exchange . The phonon frequency q is always less
than the Debye frequency D . Hence the effect of the phonons is almost irrelevant for energy
exchanges that are much larger than D . The time scale for such energies would be too short
for the slow ions to move and influence the interaction. Interestingly, the repulsive bare Coulomb
potential is renormalized to an interaction with an attractive channel for < D because of
overcompensation by the ions. This aspect of the electron-phonon interaction is most important
for superconductivity.

75

Chapter 4

Itinerant electrons in a magnetic

field
Electrons couple through their orbital motion and their spin to external magnetic fields. In
this chapter we focus on the case of orbital coupling which can be also used as a diagnostic
tool to observe the presence of a Fermi surface in a metallic system and to map out the Fermi
surface topology. A further most intriguing feature of electrons moving in a magnetic field is
the Quantum Hall effect of a two-dimensional electronic system. In both case the Landau levels
with play an important role and will be introduced here in a first step.

4.1

The de Haas-van Alphen effect

The ground state of a metal is characterized by the existence of a discontinuity of the occupation
number in momentum space - the Fermi surface. The de Haas-van Alphen experiment is one
of the best methods to verify its existence and to determine the shape of a Fermi surface. It is
based on the behavior of electrons at low temperatures in a strong magnetic field.

4.1.1

Landau levels

Consider a free electron gas subject to a uniform magnetic field B = (0, 0, B). The one-particle
Hamiltonian for an electron is given by
1 
e 2 g
H=
i~ A B Sbz B.
(4.1)
2m
c
~
We fix the gauge freedom of the vector potential A by working in the Landau gauge, A =
(0, Bx, 0), satisfying B = A. Hence the Hamiltonian (4.1) simplifies to
"
#

2
2
2
g
1

e
2
2
~
Bx ~
H=
+ i~
B Sbz B.
(4.2)
2m
x2
y c
z 2
~
In this gauge, the vector potential acts like a confining harmonic potential along the x-axis. As
translational invariance in the y- and z-directions is preserved, the eigenfunctions separate in
the three spacial components and take the form
(r) = eikz z eiky y (x)s

(4.3)

where s is the spin wave function. The states (x) are found to be the eigenstates of the
harmonic oscillator problem, so that we have
n,ky (x) =

1
2n n! 2 `2

2 )2 /2`2

Hn [(x ky `2 )/`]e(xky `
76

(4.4)

where Hn (x) are the Hermite polynomials and ` represents the magnetic length defined via
`2 = ~c/|eB|. The eigenenergies of the Hamiltonian (4.2) read


g
~2 kz2
1
+ ~c n +
B Bs
(4.5)
En,kz ,s =
2m
2
~
where s = ~/2, n N0 and we have introduced the cyclotron frequency c = |eB|/mc. Note
that the energy (4.5) does not depend on ky . The apparent differences in the spatial dependence
of the wave functions for the x- and y-directions are merely a consequence of the chosen gauge.1
The fact that the energy does not depend on ky in the chosen gauge indicates a huge degeneracy
of the eigenstates. To obtain the number of degenerate states we concentrate for simplicity on
kz = 0 and neglect the electron spin. We take the electrons to be confined to a cube of volume
L3 with periodic boundary conditions, i.e., ky = 2ny /L with ny N0 . As the wave function
n,ky (x) is centered around ky `2 , the condition
0 < ky `2 < L

(4.8)

fixes the maximal number Ndeg of degenerate states

0 < ny < Ndeg =

L2
L2 B

=
=
.
2
2`
hc/e
0

(4.9)

where 0 = hc/e is the magnetic flux quantum (e.g. Aharonov-Bohm interference effect). Thus
the degeneracy Ndeg corresponds to the number of flux quanta included in the total magnetic
flux threading the system.
The energies correspond to a discrete set of one-dimensional systems, so that the density of states
is determined by the structure of the one-dimensional dispersion (with square root singularities
at the band edges) along the z-direction:
N0 (E, n, s) =

Ndeg X

1
=
2`2
=

kz

(E En,kz ,s )



dkz
~2 kz2
1
gB
E
~c n +
+
Bs
2
2m
2
~

(2m)3/2 c
1
p
2
2
8 ~
E ~c (n + 1/2) + gB Bs/~

(4.10)
(4.11)
(4.12)

The total density of states N0 (E) for a given energy E is obtained by summing over n N0 and
s = ~/2. This should be compared to the density of states without the magnetic field,


1X
~2 k2
(2m)3/2
N0 (E, B = 0) =
E
=
E.

2m
2 2 ~3

(4.13)

k,s

The density of states for finite applied field is shown in Fig. 4.1 for one spin-component.
1

Like the vector potential, the wave function is a gauge dependent quantity. To see this, observe that under a
gauge transformation
A(r, t) A0 (r, t) = A(r, t) + (r, t)

(4.6)

the wave function undergoes a position dependent phase shift

(r, t) 0 (r, t) = (r, t)ei~c(r,t)/e .

77

(4.7)

~!c
20

N (E)

B=0

15

2.5

2.0
10

1.5

B 6= 0

1.0
5

0.5

22

44

66

Figure 4.1: Density of states for electrons in a magnetic field due to Landau levels. The dashed
line shows the density of states in the absence of a magnetic field.

4.1.2

Oscillatory behavior of the magnetization

In the presence of a magnetic field, the smooth density of states of the three-dimensional metal
is replaced by a discontinuous form dominated by square root singularities. The position of the
singularities depends on the strength of the magnetic field. In order to understand the resulting
effect on the magnetization, we consider the free energy


X
(4.14)
F = N T S = N kB T
ln 1 + e(En,kz ,s )/kB T
kz ,ky ,n,s

and use the general thermodynamic relation M = F/B. For the details of the somewhat
tedious calculation, we refer to J. M. Ziman, Principles of the Theory of Solids 2 and merely
present the result



F

sin
X
4
B B

F
k T
1
 2
 .

M = N P B 1 + L + B
(4.15)

k
P
B B B B
sinh
BT
=1

B B

Here P is the Pauli-spin susceptibility originating from the Zeeman-term and the second term
L = P /3 is the diamagnetic Landau susceptibility which is due to induced orbital currents
(the Landau levels). For sufficiently low temperatures, kB T < B B ~c , the magnetization as
a function of the applied field exhibits oscillatory behavior. The dominant contribution comes
from the summand with = 1. The oscillations are a consequence of the singularities in the
density of states that influence the magnetic moment upon successively passing through the
Fermi energy as the magnetic field is varied. The period in 1/B of the oscillations of the term
= 1 is easily found to be
 
F
1

= 2
(4.16)
B
B
or


1
B


=

2e 1
~c A(kF )

(4.17)

where we used that B = ~e/2mc and defined the cross sectional area A(kF ) = kF2 of the Fermi
sphere perpendicular to the magnetic field.
2

German title : Prinzipien der Festk

orpertheorie

78

4.1.3

Onsager equation

The behavior we have found above for a free electron gas, generalizes to systems with arbitrary
band structures. In these cases there are usually no exact solutions available. Instead of generalizing the above treatment to such band systems, we discuss the behavior of electrons within
the semiclassical approximation, as introduced in Section 1.7, and consider the closed orbits of
a wave packet subject to a magnetic field. The semiclassical equations of motion for the center
of mass of the wave packet (1.101, 1.102) simplify in the absence of an electric field to

r = v k = k
~k
e

~k = v k B.
c

(4.18)
(4.19)

This defines a closed path in a plane perpendicular to the applied uniform field. Hence, we can
apply the Bohr-Sommerfeld quantization scheme yielding quantized closed paths Pn ,
I
p dr = h(n + )
(4.20)
Pn

with n being an integer and a system specific shift, irrelevant for the final result and h = 2~.
The momentum within the semiclassical approach is expressed as
e
e
p = ~k A = (r B + A) ,
c
c

(4.21)

where we used Eq.(4.19) integrated over time. Thus, Eq.(4.20) can be now evaluated as
I
I
I
e
e
e
p dr =
{r B + A} dr = n + B
r dr = h(n + ) ,
(4.22)
c Pn
c
c
Pn
Pn
|
{z
}
=2eB Sn /c=2en /c

where Sn is the area encircled by the path Pn and n = B Sn the magnetic flux threading.
With this we find
n = (n + )0 .
(4.23)
Now we compare the areas of a given trajectory in real and reciprocal space, Sn and An ,respectively.
Considering ~k = e(r B)/c yields
0
|dr| =
|dk|
2B

Sn =

0
2B

2
An

(4.24)

With the flux quantization in Eq.(4.23) we obtain for a given magnetic field the area An ,
B
0

(4.25)

Bn
4 2
= (n0 + )
B 0.
0
0 n

(4.26)

An = (n + )4 2
while for a given area A we obtain,
A = (n + )4 2

Thus changing the field B by a certain amount Bn Bn+1 would yield the same A and one
period of chaning density of states, i.e. magnetization. From this we conclude that there is a
regular oscillation as a function of 1/B and the period is connected with the area A,
 
1
1
1
4 2
=

=
.
(4.27)

B
Bn+1 Bn
A0
79

band botton

extremal
areas

extremal
area

Fermi
surface

band top

band botton

Figure 4.2: Concentric tubes represent the quantized orbits in the reciprocal space corresponding
to the areas An embedded in the original Fermi surface which limits the height of the tubes,
occupation along the kk -direction (kk k B). Left panel: 3-dimensional view of system with a
simple Fermi surface. Right panel: cross section including kk . Extremal areas represent the
band edges (bottom: maximal A; and top: minimal A).

As we have seen above (Eq.(4.12)) the Fermi energy passes through a singularity in the density
of states at the band bottom of the one-dimensional motion of the electrons along kz or kk k B.
This corresponds the extremal area A(kk ). Note, also a band top yields the same type of
singularity and corresponds to a minimum of A(kk ). Therefore, in experiment the magnetization
maxima oscillate with a period in the inverse magnetic field as given by Eq.(4.27).
The oscillations in the magnetization, thus, allow to measure the cross sectional area of the
Fermi sphere. By varying the orientation of the field the topology of the Fermi surface can
be mapped. As an alternative to the measurement of magnetization oscillations one can also
measure resistivity oscillations known under the name Schubnikov-de Haas effect. For both
methods it is crucial that the Landau levels are sufficiently clearly recognizable. Apart from
low temperatures this necessitates sufficiently clean samples. In this context, sufficiently clean
means that the average life-time (average time between two scattering events) has to be much
larger than the period of the cyclotron orbits, i.e. c  1. This condition follows from the
uncertainty relation


4.2

~
 ~c .

(4.28)

Quantum Hall Effect

Classical Hall Effect

The Hall effect, discovered by Edwin Hall in 1879, originates from the Lorentz force exerted by
a magnetic field on a moving charge. This force is perpendicular both to the velocity of the
charged particle and to the magnetic field. In the presence of an electrical current, a Lorentz
force produces a transverse voltage, whenever the applied magnetic field points in a direction
non-collinear to the current in the conductor. This so-called Hall effect can be used to investigate
some properties of the charge carriers. Before treating the quantum version, we briefly review
80

the original Hall effect. To this end we consider the classical equation of motion of an electron,
subject to an electric and a magnetic field


dv
v
m
= e E + B ,
(4.29)
dt
c
where m is the effective electron mass. For this classical system, the steady state equation
reads


v
E + B = 0.
(4.30)
c
For the Hall geometry shown in Fig. 4.3 with fixed current j = (0, jy , 0) = (0, n0 ev, 0) and
magnetic field B = Bz , the steady state condition (4.30) simplifies to
vBz
= 0.
c

Ex +

(4.31)

The solution Ex = vBz /c yields the Hall voltage that compensates the Lorentz force. The Hall
conductivity H is defined as the ratio between the longitudinal current jy and the transverse
electric field Ex , leading to
H = yx =

jy
Ex

n0 ec
e2
= ,
Bz
h

(4.32)

where = n0 hc/Be.3 We infer from Eq.(4.32), that the measurements of the Hall conductivity
can be used to determine both the charge density n0 and the sign of the charge carriers, i.e.
whether the Fermi surfaces encloses the -point for electron-like, negative charges or a point on
the boundary of the Brillouin zone for hole-like, positive charges.
Vy
Vx
y

Bz

Figure 4.3: Schematic view of a Hall bar. The current runs a long the y-direction and the
magnetic field is applied along z-direction. The voltage Vy determines the conductance along
the Hall bar, while Vx corresponds to the transverse Hall voltage.

Discovery of the Quantum Hall Effects

When measuring the Hall effect in a special two-dimensional electron system, Klaus von Klitzing
and his collaborators made 1980 an astonishing discovery.4 The system was the inversion layer
3

Note that the conductivity is a tensor,

ja =

ab Eb .

(4.33)

b=x,y,z

Following Eq.(4.30) yielding E j B, we conclude that the off-diagonal elements due to the Hall effect satisfy
the relation ab = ba .
4
See [von Klitzing, Dorda, and Pepper, Phys. Rev. Lett. 45, 494 (1980)] for the original paper.

81

of GaAs-MOSFET device with a sufficiently high gate voltage (see Section 2.4.3) which behaves
like a two-dimensional electron gas with a high mobility e /m due to the mean free path
l 10
A and low density (n0 1011 cm2 . The two extended dimensions correspond to the
interface of the MOSFET, whereas the electrons are confined in the third dimension like in a
potential well (cf. Section 2.4.3). In high magnetic fields between 1 30T and at sufficiently
low temperatures (T < 4K), von Klitzing and coworkers observed a quantization of the Hall
conductivity corresponding to exact integer multiples of e2 /h
H = n

e2
h

(4.34)

xy / (e2 / h )

where n N. By now, the integer quantization is so widely verified, that the von Klitzing
constant (resistance quantum named after the discoverer of the Quantum Hall effect) RK =
h/e2 = 25812.807557 is used in resistance calibrations. In the field range where the transverse
conductivity shows integer plateaus in 1/B, the longitudinal conductivity yy vanishes and
takes finite values only when H crossed over from one quantized value to the next (see Fig.
4.4).
3
2

yy

Figure 4.4: Integer Quantum Hall effect: As a function of the filling factor plateaus in xy
appear at multiples of e2 /h. The longitudinal conductance yy is only finite for fillings where
xy changes between plateaus.
In 1982, Tsui, St
ormer, and Gossard5 discovered an additional quantization of H , corresponding
to certain rational multiples of e2 /h. Correspondingly, one now distinguishes between the integer
quantum Hall effect (IQHE) and the fractional quantum Hall effect (FQHE). These discoveries
marked the beginning of a whole new field in solid state physics that continues to produce
interesting results.

4.2.1

Hall effect of the two-dimensional electron gas

Here we first discuss the Hall effect in the quantum mechanical treatment. For this purpose
we start with the Hamilton operator (4.1) and neglect the electron spin. Working again in
the Landau gauge, A = (0, Bx, 0), and confining the electronic system to two dimensions, the
Hamiltonian reduces to
"

2 #
2
1

e
2
.
(4.35)
H=
~
+ i~
Bx
2m
x2
y c
5

See [Tsui, St
ormer and Gossard Phys. Rev. Lett. 48, 1559 (1982)] for the original paper

82

For the two-dimensional gas there is no motion in the z-direction, so that the highly degenerate
energy eigenvalues are given by the spectrum of a one-dimensional harmonic oscillator En =
~c (n + 1/2), where again c = |eB|/mc. Here, we will concentrate on the lowest Landau level
(n = 0) with the wave function
0,ky =

1
2`2

2 /2`2

e(xx0 )

eiky y .

(4.36)

p
where the magnetic length ` = ~c/|eB| gives the extension of the wave function in the presence
of the magnetic field. In x-direction, the wave function is localized around x0 = ky `2 , whereas
it takes the form of a plane wave in y-direction. As discussed previously, the energy does not
depend on ky .
We now introduce an electric field Ex along the x-direction. The Hamilton operator (4.35) is
then modified by an additional potential U (r) = eEx x. This term can easily be absorbed into
the harmonic potential and leads to a shift of the center of the wave function,
x0 (ky ) x00 (ky ) = ky `2 +

eEx
.
mc2

(4.37)

Moreover the degeneracy of the Landau level is lifted since the energy becomes ky -dependent
and (after completing the square) takes the form
~c
m
En=0 (ky ) =
eEx x00 (ky ) +
2
2

cEx
B

2
.

(4.38)

The energy (4.38) corresponds to the wave function 0ky from (4.36) where x0 is replaced by x00 .
The velocity of the electrons is then given by
vy (ky ) =

eE `2
cE
1 dEn=0 (ky )
= x = x,
~
dky
~
B

(4.39)

and from this, we determine the current density,

jy = en0 vy (ky ) = en0

cEx
e cEx
e2
=
=

E = H Ex
B
2`2 B
h x

(4.40)

where = n0 2`2 is the filling of the Landau level.6 The Hall conductivity is then identical to
the result (4.32) derived previously based on the quasiclassical approximation. There is a linear
relation between the Hall conductivity H and the index B 1 .

4.2.2

Integer Quantum Hall Effect

The plateaus observed by von Klitzing in the Hall conductivity H of the two-dimensional
electron gas as a function of the magnetic field correspond to the values H = n e2 /h, as if
= n N was restricted to be an integer. Meanwhile, the longitudinal conductivity of the
electron gas vanishes when a plateau of H is realized
yy =

jy
Ey

= 0,

(4.41)

and only becomes finite at the transition points of H between two plateaus (cf. Fig. 4.4). This
fact seems to be in contradiction with the results from the consideration above. The solution
to this mysterious behavior lies in the fact that disorder, which is always present in a real
Note that 1 = B/n0 0 where 0 = hc/e represents the flux quantum, i.e. 1 B is the number of flux
quanta 0 per electron.
6

83

inversion layer, plays a crucial role and should not be neglected. In fact, due to the disorder, the
electrons move in a randomly modulated potential landscape U (x, y). As we will find out, even
small amounts of disorder lead to the localization of electronic states in this two-dimensional
system. To illustrate this new aspect we focus on the lowest Landau level in the symmetric
gauge A = (y, x, 0)B/2. The Schr
odinger equation in polar coordinates is given by
"

2 #
~2
1
1
e

i
Br
(r, ) + U (x, y)(r, ) = E(r, ) .
(4.42)
2m
r r r
r
2~c
Without the external potential U (x, y) we find the ground state solutions
n=0,m (r, ) =

 r m
`

2`2 2m m!

eim er

2 /4`2

(4.43)

where all values of m N0 correspond to the same energy En=0 = ~c /2. One easily verifies,
that the wave functions |n=0,m (r, )| are peaked on circles of radius rm = 2m ` (see Fig.4.5).
Note that the magnetic flux threading such a circle is given by
2
= B2m`2 = 2mB
Brm

~c
hc
=m
= m0 ,
eB
e

(4.44)

which is an integer multiple of the flux quantum 0 = hc/e.

|| 2

Figure 4.5: Wavefunction of a Landau level state in the symmetric gauge.

Now we consider the effect of the disorder potential. The gauge can be adjusted to the potential
landscape. If, for simplicity, we assume the potential to be rotationally invariant around the
origin, the symmetric gauge is already optimal. For the potential
U (x, y) = U (r) =

C1
+ C2 r2 + C3 ,
r2

(4.45)

the exact expression of all eigenstates of equation (4.42) in the lowest Landau level is obtained
using the Ansatz
0,m (r, ) = p

 r 

2`2 2 ( + 1)

eim er

2 /4`2

(4.46)

After introducing the dimensionless parameters C1 = 2m C1 /~2 and C2 = 8`4 m C2 /~2 , the
quantities and ` from equation (4.46) can be expressed via
2 = m2 + C1 ,

2

2 p
`
= `
1 + C2 .

84

(4.47)
(4.48)

Indeed, the Ansatz (4.46) describes eigenstates of the disordered problem (4.42). The degeneracy
of the ground state energy (the lowest Landau level) is now lifted,


~c `2
E0,m =
( + 1) m + C3 .
(4.49)
2 `2

The wave functions are concentrated around the radii rm = 2` . For weak potentials
C1 , C2  1 and m  1 the energy is approximatively given by
E0,m

~c
C
2
+ C3 + . . . ,
+ 21 + C2 rm
2
rm

(4.50)

i.e. the wave function adjusts itself to the potential landscape. It turns out that the same is true
for arbitrarily structured weak potential landscapes. The wave function describes electrons on
quasi-classical trajectories that trace the equipotential lines of the underlying disorder potential.
Consequently the states described here are localized in the sense that they are attached to the
structure of the potential. The application of an electric field cannot set the electrons in the
concentric rings in motion. Therefore, the electrons are localized and do not contribute to
electric transport.
Picture of the potential landscape
When the magnetic field is varied the filling = n0 2`2 of the Landau level is adjusted accordingly. While all states of a given level are degenerate in the transitionally invariant case, now,
these states are spread over a certain energy range due to the disorder. In the quasi-classical approximation, these states correspond to equipotential trajectories that are either filled or empty
depending on the strength of the magnetic field, i.e. they are either below or above the chemical
potential. These considerations lead to an intuitive picture on localized (closed trajectories) and
extended (percolating trajectories) states. We may consider the potential landscape like a real
landscape where the the trajectories are contour lines. Assume that we fill now water into such
a landscape. The trajectories of the particles is restricted to the shore line. For small filling, we
find lakes whose shores are closed and correspond to contour lines. They correspond to closed
electron trajectories and represent localized electronic states. At very high water level, only
the large mountains of the potential landscape would reach out of the water, forming islands in
the sea. The coastlines again represent closed trajectories corresponding to localized electronic
states. At some intermediate filling, a boundary between the lake and the island topology,
there is a water level at which the coast lines become arbitrarily long and percolate through
the whole landscape. Only these contour lines correspond to extended (non-localized) electron
states. From this picture we conclude that when a Landau level of a system subject to a random
potential is gradually filled, first all occupied state are localized (low filling). At some special intermediate filling level, the extended states are filled and contribute to the current transport. At
higher chemical potential (filling) the states would be localized again. In the following argument,
going back to Robert B. Laughlin, the presence of filled extended states plays an important role.

Laughlins gauge argument

We consider a long rectangular Hall element that is deformed into a so-called Corbino disc, i.e.
a circular disc with a hole in the middle as shown in Fig.4.7. The Hall element is threaded
by a constant and uniform magnetic field B along the z-axis. In addition we can introduce an
arbitrary flux through the hole without influencing the uniform field in the disc. This flux
is irrelevant for all localized electron trajectories since only extended (percolating) trajectories
wind around the hole of the disc and by doing so receive an Aharonov-Bohm phase. When the

85

closed

extended

Figure 4.6: Contour plot of potential landscape. There are closed trajectories and extended
percolating trajectories.

flux is increased adiabatically by , the vector potential is changed according to

A A + A = A + ,
which in our case means
(A) =

.
2r

(4.51)

At the same time, the wave function acquires a phase factor

eie/~c = ei(/0 ) .

(4.52)

If the disc was translationally invariant, meaning that disorder is neglected and only extended
2 = m + . The
states exist, we could use the wave functions 0,m from (4.43), so that Brm
0
single-valuedness of the wave function implies that m has to be adjusted, m m /0 .
This guarantees, that increasing by one flux quantum leads to a decrease of m by 1. Hence,
gauge invariance implies that the wave functions are shifted in their radius. This argument is
also applicable to higher Landau levels.

Iy
Figure 4.7: Corbino disk for Laughlins argument. According to the Hall bar in Fig. 4.3, the
radial (transverse) component of the Corbino disc is denoted by x, while the angular (longitudinal) component is termed as y. Both the homogeneous magnetic field B and the flux point
along the z-axis perpendicular to the plane of the disc.
Since this argument is topological in nature, it will not break down for independent electrons
86

when disorder is introduced. The transfer of one electron between neighboring extended states
due to the change of by 0 leads to a net shift of one electron from the outer to the inner
boundary. If an electric field Ex is applied in the radial direction (here denoted by x-direction,
see Fig. 4.7), the transfer of this electron results in the energy change
V = eEx L

(4.53)

where L is the distance between the inner and the outer boundary of the Corbino disc. A further
change in the electromagnetic energy
I =

Iy
c

(4.54)

is caused by the constant current Iy (here the angular component is denoted by y, see Fig.4.7)
in the disc when the magnetic flux is increased by . Following the Aharonov-Bohm argument
that the energy of the system is invariant under a flux change by integer multiples of 0 ,
the two energies should compensate each other. Thus, setting = 0 and demanding that
V + I = 0 leads to
H =

jy
Ex

Iy
LEx

e2
.
h

(4.55)

We conclude from this argument, that each filled Landau level containing percolating states
will contribute e2 /h to the total Hall conductivity. Hence, for n N0 filled levels the Hall
conductivity is given by H = n e2 /h. Note the importance of the topological nature of the Hall
conductivity ensuring the universal character of the quantization.
Localized and extended states
The density of states of the two-dimensional electron gas (2DEG) in absence of an external
magnetic field is given by
!
X
~2 (kx2 + ky2 )
Lx Ly m
N2DEG (E) = 2
E
,
(4.56)
=
2m
2
kx ,ky

with twice degenerate energy states for the spins, whereas for the Landau levels in a clean
sample, we have
NL (E) =

Lx Ly X
2`2

n,s

(E En,s ).

(4.57)

Here the prefactor is given by the large degeneracy (4.9) of each Landau level.
According to our previous discussion, the main effect of a potential is to lift the degeneracy of the
states comprising a Landau level. This remains true for random potential landscapes. Most of
the states are then localized and do not contribute to electric transport. Only the few extended
states contribute to the transport if they are filled (see Fig. 4.8). For partially filled extended
states the Hall conductivity H is not an integer multiple of e2 /h, since not all percolating states
necessary for transferring one electron from one edge to the other, when the flux is changed
by 0 (in Laughlins argument) are occupied. Thus, the charge transferred does not amount
to a complete e. The appearance of partially filled extended states marks the transition from
one plateau to the next and is accompanied by a finite longitudinal conductivity yy . When all
extended states of a Landau level are occupied, the contribution to the longitudinal transport
stops, i.e., in the range of a plateau yy vanishes. Because of thermal occupation, the plateaus
quickly shrink when the temperature of the system is increased. This is the reason why the
Quantum Hall Effect is only observable for sufficiently low temperatures (T < 4K).
87

N (E)

delocalized)

B 6= 0

N (E)

B =0

clean)

localized)

disordered)

Figure 4.8: Density of states for the two-dimensional electron gas in three different cases. On
the left panel, the system without and with applied magnetic field. Without magnetic field
the density of states is flat (dashed line). If an external magnetic field is applied to a clean
system, the Landau levels appears as infinitely sharp strongly degenerate peaks. The right
panel visualizes the effect of disorder in the two-dimensional system with magnetic field; the
Landau levels are spread and the density of state shows broadened peaks where most of the
states are localized and only few states in the center percolate.

Edge states and B

uttikers argument
A confining potential prevents the electrons from leaving the metal. This potential at the edges
of the sample has also to be considered in the potential landscape. Interestingly, equi-potential
trajectories of states close to the edge are always extended and percolate along the edge. The
corresponding wave functions have already been discussed in Section 4.2.1. From Eq.(4.38), we
find that the energy is not symmetric in ky , the wave vector along the edge, i.e. E(ky ) 6= E(ky ).
This implies that the states are chiral and can move in one direction only for a given energy.
The edge states on the opposing edges move in opposite directions, a fact that can be readily
verified by inspection of (4.38) based on Fig.4.9. The total current flowing along the edge for a
given Landau level is
X e
I=
v ,
(4.58)
Ly y
ky

meaning that one state per ky extends over the whole length Ly of the Hall element. Thus,
the density is given by 1/Ly . The wave vector is quantized according to the periodic boundary
conditions; ky = 2ny /Ly with ny Z. The velocity vy is given by equation (4.39). In summary,
we have
Z
Z
dEn (ky )
e
e
dE
e
I=
dky
=
dx0
= ( En(0) )
(4.59)
2~
dky
h
dx0
h
occupied

occupied

where x0 = ky `2 is the transversal position of the wave function and is the chemical potential.
Sufficiently far away from the boundary En is independent of x0 and approaches the value
En(0) = ~c (1/2 + n) of a translationally invariant electron gas. The potential difference between
the two opposing edges leads to a net current along the edge direction of the Hall bar,
A B = eVH = eEx Lx =

h
h
(IA + IB ) = IH ,
e
e

(4.60)

and with that

H =

IH
e2
= ,
Ex Lx
h
88

(4.61)

where for A = B we have IA = IB . Note that IH = IA + IB is only valid when no currents

are present in the bulk of the system. The latter condition is ensured by the localization of the
states at the chemical potential. This argument by B
uttiker leads to the same quantization as
derived before, namely every Landau level contributes one edge state and thus H = n e2 /h (n
is the number of occupied Landau levels). Note that this argument is independent of the precise
shape of the confining edge potential.
E

n=2
n=1
n=0

IA
A

IB

x
E
A

Figure 4.9: Edge state picture by B

uttiker. The left panel gives a top view of the two dimensional
system (L = Lx ) where chiral edge states exist on both edges of the Hall bar with opposite
chirality. On the middle panel, a single Landau level without and with transverse potential
difference is shown, where the latter yields a finite net current due to current imbalance between
left and right edge. The right panel visualizes the two lowest Landau levels which are occupied.
All higher Landau levels are empty. This fixes the Hall conductance value n to 2.
Within this argument we are able to understand why the longitudinal conductivity has to vanish
at the Hall plateaus. For that we express Ohms law However it is simpler to discuss the
resistivity. Like the conductivity
b the resistivity b is a tensor:
j=
bE

(4.62)

E = bj

(4.63)

where the conductivity

b and the resistivity b are both 2 2 tensors with xx = yy (xx = yy )
and xy = yx (xy = yx ). Therefore, we find
yy =
xy =

yy
2yy

+ 2xy
xy

2yy + 2xy

(4.64)

(4.65)

In the following argument, we explain why the longitudinal resistivity yy in two dimensions
has to vanish in the presence of a finite Hall resistivity xy . Since the edge state electrons with
a given energy can only move in one direction, there is no backward scattering by obstacles as
long as the edges are far apart from each other. No scattering between the two edges implies
yy = 0 and hence yy = 0. A finite resistivity can only occur when extended states are present
in the bulk, such that the edge states on opposite edges are no longer spatially separated from
each other.

4.2.3

Fractional Quantum Hall Effect

Only two years after the discovery of the Integer Quantum Hall Effect, Stormer, Tsui and
Gossard observed further series of plateaus of the Hall resistivity in a 2DEG realized with very
high quality MOSFET inversion layers at low temperatures. The most pronounced of these

89

plateaus is observed at a filling of = 1/3 (xy = e2 /h). Later, an entire hierarchy of plateaus
at fractional values = p,m = p/m with p, m N has been discovered,

p,m

1 2 2 3 3
, , , , , ...
3 3 5 5 7


.

(4.66)

The emergence of these new plateaus is a clear evidence of the so-called Fractional Quantum
Hall Effect (FQHE).

Figure 4.10: Experimental evidence of the Fractional Quantum Hall Effect.

It was again Laughlin who found the key concept to explain the FQHE. Unlike the IQHE, this
new quantization feature can not be understood from a single-electron picture, but is based on
the Coulomb repulsion among the electrons and the accompanying correlation effects. Laughlin
specially investigated the case p,m = 1/3 and made the Ansatz
Y
X |z |2
i
1/m (z1 , . . . , zN )
(zi zj )m exp
4`2
i<j

!
(4.67)

for the N -body wave function, where z = xiy is a complex number representing the coordinates
of the two-dimensional system. Limiting ourselves to the consideration of the lowest Landau
level, this state gives a stable plateau with H = (1/3)e2 /h, when m = 3.
A heuristic interpretation of the Laughlin state was proposed by J. K. Jain and it is based on
the concept of so-called composite fermions. In fact, Laughlins state (4.67) can be written as
Y
1/m =
(zi zj )m1 S
(4.68)
i<j

where S is the Slater determinant7 describing the completely filled lowest Landau level. We
see that the prefactor of S in equation (4.68) acts as a so-called Jastrow factor that introduces
7
The Slater determinant of the lowest Landau level is obtained from the states of the independent electrons.
In symmetric gauge, the states are labelled by the quantum number m
N0 and apart from the normalization
(given in equation (4.43)) they are given by
m
|z|
m
(z) = z e

90

/4`2

(4.69)

AharanovBohmPhase

Austausch

Figure 4.11: Exchange of two particles in two dimensions involves the motion of the particles
around each other. There are two topologically distinct paths.

correlation effects into the wave function, since only the correlations due to the Pauli exclusion
principle are contained in S . The Jastrow factor treats the Coulomb repulsion among the
electrons and consequently leads to an additional suppression of the wave function whenever
two electrons approach each other. In the form introduced above, it produces an phase factor
for the electrons encircling each other. In particular, exchanging two electrons (see Fig. 4.11)
leads to a phase


e m1
exp(i(m 1)) = exp i
0 ,
(4.73)
~c 2
since 0 = 2~c/e. This phase has to be unity since the Slater state S is odd under exchange
of two electrons. Therefore m is restricted to odd integer values. This guarantees that the total
wave function 1/m still changes sign when two electrons are exchanged.
According to the Footnote 6 (see equation (4.40)), the case p,m = 1/3 implies that there are
three flux quanta 0 per electron. In order to understand the FQHE, one constructs so-called
composite fermions which do not interact with each other. Here, a composite fermion consists of
an electron that has two (in fact m 1) negative flux quanta attached to it. These objects may
be considered as independent fermions since the attached flux quanta compensate the Jastrow
factor in equation 4.68 through factors of the type (zi zj )(m1) . The exchange of two such
composite fermions in two dimensions leads to an Aharanov-Bohm phase that is just opposed
to that in equation (4.73). Due to the presence of the flux 20 per electron, the composite
fermions are subject to an effective field composed of the external field and the attached flux
where z = x iy. The Slater determinant for N independent electrons is

0 (z1 ) N 1 (z1 )
1

..
..
S (z1 , . . . , zN ) =
det

.
.
N!
0 (zN ) N 1 (zN )

det
=

N!

1
1
..
.
1

z1
z2
..
.
zN

z12
z22
..
.
2
zN

z1N 1
z2N 1
..
.
N 1
zN

(4.70)

X |zi |2

exp
4`2

!
.

(4.71)

The remaining determinant is a so-called Vandermonde determinant, which can be reexpressed in the form of a
product, such that
!
Y
X |zi |2
S =
(zi zj ) exp
.
(4.72)
4`2
i<j
i
The prefactor is a homogenous polynomial with roots whenever zi = zj , which is a manifestation of the Pauli
exclusion principle. We also see that the state S has a well defined total angular momentum Lz = N ~.

91

composite
Fermion

composite Fermion s
im externen Feld

Bex

+
Beff = B ex 20 n 0

20

FeldKompensation

Beff

IQHE

Figure 4.12: Sketch of the composite Fermion concept. Electrons with attached magnetic flux
lines, here for the state of = 1/3.

quanta:
Beff = B

20 (zi )

(4.74)

1
= B
3

X
i

2
20 (zi ) B
3

!
(4.75)

For an external field B = 3n0 0 , the expression in the brackets of equation (4.75) vanishes and
the composite fermions feel an effective field Beff = n0 0 (Fig. (4.12)). Thus, these fermions
form an Integer Quantum Hall state with = 1 (for B = 3n0 0 ), as discussed previously. This
way of interpretation is applicable to other Fractional Quantum Hall states, too, since for n
filled Landau levels with composite fermions consisting of an electron with an attached flux of
2k0 , the effective field reads
Beff = n0

2kn0 0 = n0 0 .
p,m
n

(4.76)

1
1
+ 2k =
n
p,m

(4.77)

From this we infer, that

or equivalently
p,m =

p
n
=
.
m
2kn + 1

(4.78)

Despite the apparent simplicity of the treatment in terms of independent composite fermions, one
should keep in mind that one is dealing with a strongly correlated electron system. The structure
of the composite fermions is a manifestation of the fact that the fermions are not independent
electrons. No composite fermions can exist in the vacuum, they can only arise within a certain
many-body state. The Fractional Quantum Hall state also exhibits unconventional excitations
with fractional charges. For example in the case p,m = 1/3, there are excitations with effective
charge e = e/3. These are so-called topological excitations, that can only exist in correlated
systems. The Fractional Quantum Hall system is a very peculiar ordered state of a twodimensional electron system that has many interesting and complex properties.8
8

Additional literature on the quantum Hall effects. For the Integer quantum Hall effect consult
- K. von Klitzing et al., Physik Journal 4 (6), 37 (2005)

while detailed literature on the fractional quantum Hall effect is found in

- R. Morf, Physik in unserer Zeit 33, 21 (2002)
- J.K. Jain, Advances in Physics 41, 105 (1992).

92

Chapter 5

Landaus Theory of Fermi Liquids

In the previous chapters, we considered the electrons of the system as more or less independent
particles. The effect of their mutual interactions only entered via the renormalization of potentials and collective excitations. The underlying assumptions of our earlier discussions were that
electrons in the presence of interactions can still be described as particles with a well-defined
energy-momentum relation, and that their groundstate is a filled Fermi sea with a sharp Fermi
surface. Since there is no guarantee that this hypothesis holds in general (and in fact they do
not), we have to show that in metals the description of electrons as quasiparticles is justified.
This quasiparticle picture will lead us to Landaus phenomenological theory of Fermi liquids.

5.1

Lifetime of quasiparticles

We first consider the lifetime of a state consisting of a filled Fermi sea to which one electron
is added. Let k with |k| > kF (k = ~2 k2 /2m with k > F ) be the momentum (energy) of
the additional electron. Due to interactions between the electrons, this state will decay into a
many-body state. In momentum space such an interaction takes the form
X X
ck0 ,s0 b
ck,s ,
(5.1)
ck0 +q,s0 b
Hee =
V (q)b
ckq,s b
k,k0 ,q s,s0

where V (q) represents the electron-electron interaction in momentum space while q indicates
the momentum transfer in the scattering process. Below, the short-ranged Yukawa potential
V (q) =

4e2
4e2
=
q 2 (q, 0)
q 2 + kT2 F

(5.2)

from equation (3.94) will be used. As we are only interested in very small energy transfers
~  F , the static approximation is admissible.
In a perturbative treatment, the lowest order effect of the interaction is the creation of a particlehole excitation in addition to the single electron above the Fermi energy. As the additional
electron changes its momentum from k to k q, a hole appears at k0 and a second electron with
wavevector k0 + q is created outside the Fermi sea. The transition is allowed whenever both
energy and momentum are conserved, meaning
k = (k q) k0 + (k0 + q),

(5.3)

k = kq k0 + k0 +q .

(5.4)

and

We calculate the lifetime k of the initial state with momentum k using Fermis golden rule,
yielding the transition rate from the initial state of a filled Fermi sea and one particle with
93

k0

k0
k+q

Figure 5.1: The decay of an electron state above the Fermi energy happens through scattering
by creating particle-hole excitations.

momentum k to a state with two electrons above the Fermi sea, with momenta k q and k0 + q,
and a hole with k0 , as shown in Fig. 5.1. Since neither the momenta k0 and q, nor the spin of
the created electron are fixed, a summation over the possible configuration has to be performed,
leading to
1
2 1 X X
=
|V (q)|2 n0,k0 (1 n0,kq )(1 n0,k0 +q )(kq k (k0 k0 +q )).
k
~ 2 0
0

(5.5)

k ,q s

Note that the term n0,k0 (1 n0,kq )(1 n0,k0 +q ) takes care of the Pauli principle, by ensuring
that the final state after the scattering process exists, i.e. the hole state k0 lies inside and the
two particle states k q and k0 + q lie outside the Fermi sea. First the integral over k0 is
performed under the condition that the energy k0 +q k0 of the excitation is small. With that,
the integral reduces to
1X
n 0 (1 n0,k0 +q )(kq k (k0 k0 +q ))
0 0,k
k
Z
1
=
d3 k 0 n0,k0 (1 n0,k0 +q )(k0 +q k0 ~q,k )
(2)3
N (F ) q,k
=
4 qvF

S(q,k , q) =

(5.6)
(5.7)
(5.8)

where N (F ) = mkF / 2 ~2 is the density of states of the electrons at the Fermi surface and
~q,k = ~2 (2k q q 2 )/2m > 0 is the energy loss of the decaying electron.1 In order to compute
the remaining integral over q, we assume that the matrix element |V (q)|2 depends only weakly
on q when q  kF . This is especially true when the interaction is short-ranged. In spherical
1
Small are justified, because ~ (2kF q q 2 )/2m for most allowed . The integral may be computed using
cylindrical coordinates, where the vector q points along the axis of the cylinder. It results in

1
S(q, ) =
(2)2

Zk1
k2

0 0
dk
k

ZkF

dkk0

~2 qkk0
~2 q 2
+
~
2m
m

!
(5.9)

m
k12 k22 ,
2
2
4 ~ q

(5.10)

2
with k12 = kF2 kk,0
and k22 = kF2 (kk,0 + q)2 , where kk,0 = (2m ~q 2 )/2~q is enforced by the delta function.

94

coordinates, the integral reads

q,k
1
2 N (F ) X

=
|V (q)|2
k
~ 4vF
q
q,s0
Z
q,k
N (F )
=
d3 q |V (q)|2
2
(2) 2~vF
q
N (F )
=
(2)4mvF

N (F )
=
(2)4mvF

2 2

dq |V (q)| q

Z2

d sin (2k cos q)

(5.11)
(5.12)

(5.13)


2
1
2
.
dq |V (q)| q (2k cos q)
4k
1
2 2

(5.14)

The restriction of the domain of integration of follows from the two conditions k 2 (k q)2
kF2 and (k q)2 = k 2 2kq cos + q 2 . From the first condition, cos 2 = q/2k, and from the
second, cos 1 = (k 2 kF2 + q 2 )/2kq. Thus,
N (F )
1
=
k
(2)4mvF

2
1
k 2 kF2
4k
Z
N (F ) m 1
2

( F )
dq |V (q)|2
(2)4vF kF ~4 k
Z
1 N (F )
2
=
(k F )
dq |V (q)|2 .
8~3 vF2
Z

dq |V (q)|2

(5.15)
(5.16)
(5.17)

Note that convergence of the last integral over q requires that the integrand does not diverge
stronger than q ( < 1) for q 0. With the dielectric constant obtained in the previous
chapter, this condition is certainly fulfilled. Essentially, the result states that
1
(k F )2
k

(5.18)

for k slightly above the Fermi surface. This implies that the state |ksi occurs as a resonance of
width ~/k and features a quasiparticle, which can be observed in the spectral function A(E, k)
as depicted in Fig.5.2.2 The quasiparticle (coherent) part of the spectral function has a weight
reduced from one (corresponding to the quasiparticle weight Zk ). The remaining weight is
shifted to higher energies as a so-called incoherent part (continuum without clear momentumenergy relation).

kF
k1
q
k ||
kF

k2

The wave vectors k2 and k1 are the upper and lower limits of integration determined from the condition n0,k0 (1
n0,k0 +q ) > 0 and can be obtained by simple geometric considerations. equation (5.8) follows immediately.
2
The spectral function is defined as
X
A(E, k)
|hn |b
cks |0 i|2 (E En )
(5.19)
n

95

Quasiparticle

incoherent part

A(E,k)
k

kF

EF

Figure 5.2: Quasiparticle spectrum: Quasiparticle peaks appear the sharper the closer the
energy lies to the Fermi energy. The area under the sharp quasiparticle peak corresponds to
the quasiparticle weight. The missing quasiparticle weight is transferred to higher energies.

The resonance becomes arbitrarily sharp as the Fermi surface is approached

lim

kkF

~/k
= 0,
k F

(5.21)

so that the quasiparticle concept is asymptotically valid. The equation (5.21) can also be seen as
a verification of Heisenbergs uncertainty principle. Consequently, the momentum of an electron
is a good quantum number in the vicinity of the Fermi surface. Underlying this result is the
Pauli exclusion principle, which restricts the phase space for decay processes of single particle
states close to the Fermi surface. In addition, the assumption of short ranged interactions is
crucial. Long ranged interactions can change the behavior drastically due to the larger number
of decay channels.

5.2

Phenomenological Theory of Fermi Liquids

The existence of well-defined fermionic quasiparticles in spite of the underlying complex manybody physics inspired Landau to the following phenomenological theory. Just like the states of
independent electrons, quasiparticle states shall be characterized by their momentum k and spin
. In fact, there is a one-to-one mapping between the free electrons and the quasiparticles. Consequently, the number of quasiparticles and the number of electrons coincide. The momentum
distribution function of quasiparticles, defined as n (k), is subject to the condition
X
N=
n (k).
(5.22)
k,

In analogy to the Fermi-Dirac distribution of free electrons, one demands, that the ground state
distribution function n(0)
(k) for the quasiparticles is described by a simple step function
n(0)
(k) = (kF |k|).

(5.23)

where |0 i is the exact (renormalized) ground state and |n i are the corresponding exact excited states. The
coherent part of the spectral function can be represented as a Lorentzian form
Acoh (E, k) =

Zk ~
1
k (E k )2 +

where Zk is the quasiparticle weight, smaller than 1.

96

~2
k2

(5.20)

n (k)

n0ks

kF

kF

Figure 5.3: Schematic picture of the distribution function: Left panel: modified distribution
function of the original electron states; right panel: distribution function of quasiparticle states
making a simple step function.

For a spherically symmetric electron system, the quasiparticle Fermi surface is a sphere with
the same radius as the one for free electrons of the same density. For a general point group
symmetry, the Fermi surface may be deformed by the interactions without changing the underlying symmetry. The volume enclosed by the Fermi surface is always conserved despite the
deformation.3 Note that the distribution n(0)
(k) of the quasiparticles in the ground state and

ck i of the real electrons in the ground state are not identical (Figure 5.3).
that n0ks = hb
ck b
Interestingly, n0ks is still discontinuous at the Fermi surface, but the height of the jump is, in
general, smaller than unity. The modification of the electron distribution function from a step
function to a smoother Fermi surface indicates the involvement of electron-hole excitations
and the renormalization of the electronic properties, which deplete the Fermi sea and populate
the states above the Fermi level. The reduced jump in n0ks is a measure for the quasiparticle
weight at the Fermi surface, ZkF , i.e. the amplitude of the corresponding free electron state in
the quasiparticle state.
In Landaus theory of Fermi liquids, the essential information on the low-energy physics of the
system shall be contained in the deviation of the quasiparticle distribution n (k) from its ground
state distribution n(0)
(k),
n (k) = n (k) n(0)
(k).

(5.24)

The symbol is generally used in literature to denote this difference. Unfortunately this may
suggest that the term n (k) is small, which is not true in general. Indeed, n (k) is concentrated on momenta k very close to the Fermi energy only, where the quasiparticle concept
is valid. This distribution function, describing the deviation from the ground state, enters a
phenomenological energy functional of the form
E = E0 +

 (k)n (k) +

1 XX
f 0 (k, k0 )n (k)n0 (k0 ) + O(n3 )
2 0 0

(5.25)

k,k ,

k,

where E0 denotes the energy of the ground state. Moreover, the phenomenological parameters
 (k) and f0 (k, k0 ) have to be determined by experiments or by means of a microscopic theory.
The variational derivative
 (k) =

E
1 X
=  (k) +
f 0 (k, k0 )n0 (k0 )
n (k)
0 0

(5.26)

k ,

yields an effective energy-momentum relation  (k), whose second term depends on the distribution of all quasiparticles. A quasiparticle moves in the mean-field of all other quasiparticles, so
3

This is the content of the Luttinger theorem [J.M. Luttinger, Phys. Rev. 119, 1153 (1960)].

97

that changes n (k) in the distribution affect  (k). The second variational derivative describes
the coupling between the quasiparticles
1
2E
= f0 (k, k0 ).

n (k)n0 (k0 )

(5.27)

We introduce a parametrization for these couplings f0 (k, k0 ) by assuming spherical symmetry

of the system. Furthermore, the radial dependence is ignored, as we only consider quasiparticles
in the vicinity of the Fermi surface where |k|, |k0 | kF . Therefore the dependence of f0 (k, k0 )
on k, k0 can be reduced to the relative angle k,
k
0
k0 ) + 0 f a (k,
k0 )
f0 (k, k0 ) = f s (k,

(5.28)

where k = k/|k|. The symmetric (s) and antisymmetric (a) part of f0 (k, k0 ) can be expanded
in Legendre-polynomials Pl (z) 4 , leading to
f

s,a

k0 ) =
(k,

fls,a Pl (cos k,
k
0 ).

(5.31)

l=0

The density of states at the Fermi surface is defined as

N (F ) =

m k
k2
2X
((k) F ) = 2 F = 2 2F

~vF
~

(5.32)

and follows from the dispersion (k) of the bare quasiparticle energy
k (k)|kF = vF =

~kF
m

(5.33)

where for a fully rotation symmetric system we may write (k) = ~2 k2 /2m with m as an
effective mass, although we will be only interested at the spectrum in the immediate vicinity
of the Fermi energy. With this definition, we also introduce the so-called Landau parameters
Fls = N (F )fls ,

(5.34)

N (F )fla ,

(5.35)

Fla

commonly used in the literature.5 In the following, we want to study the relation between the different phenomenological parameters of Landaus theory of Fermi liquids and the experimentally
accessible quantities of a real system, such as specific heat, compressibility, spin susceptibility
among others.

5.2.1

Specific heat - Density of states

Since the quasiparticles are fermions, they obey Fermi-Dirac statistics

n(0)
(T, k) =
4

1
e[(k)]/kB T

(5.36)

+1

Legendre-Polynomials: The Legendre polynomials satisfy the orthogonality relation

Z
Z +1
dz
ll0
dk
Pl (cos k )Pl0 (cos k ) =
Pl (z)Pl0 (z) =
4
2
2l
+1
1

(5.29)

for all l = 0, 1, 2, .... The first few polynomials are

P0 (z) = 1 ,
5

P1 (z) = z ,

P2 (z) =

1
(3z 2 1) ,
2

P3 (z) =

1
(5z 3 3z) .
2

(5.30)

Another frequently used notation for the Landau parameters in the literature is Fl = Fls and Zl = Fla .

98

with the chemical potential . For low temperatures, we find

(0)
n (k) = n(0)
(T, k) n (0, k)

(5.37)

which, using Eq.(5.26) leads to (k) = (k), because here

1 X
f 0 (k, k0 )n0 (k0 ) = 0 .
0 0

(5.38)

k ,

In n0 (k0 ) there are as many particles (positive) as holes (negative) such that they cancel in
the correction term. Therefore we rewrite the Fermi-Dirac distribution (5.36) as
n (k) =

1
e[(k)F ]/kB T + 1

(5.39)

with the bare quasiparticle energy  (k) in place of the renormalized dispersion  (k). Furthermore, we replaced = F + O(T 2 ) by F . When focussing on leading terms, which are usually
of the order T 0 and T 1 , corrections of higher order may be neglected in the low-temperature
regime. In order to discuss the specific heat, we employ the expression for the entropy of a
fermion gas. For each quasiparticles with a given spin there is one state labelled by k. The
entropy density may be computed from the distribution function
S=


kB X 
n (k) ln (n (k)) + (1 n (k)) ln (1 n (k)) .

(5.40)

k,

Taking the derivative of the entropy S with respect to T , the specific heat
S
T


n (k)
kB T X
e(k)/kB T
(k)
=
ln

1 n (k)
( e(k)/kB T + 1 )2 kB T 2
k,

C(T ) = T

kB T X
1
(k) (k)
2

4 cosh ((k)/2kB T ) kB T 2 kB T
k,

(5.41)
(5.42)
(5.43)

is obtained, where we introduced (k) = (k) F . In the limit T 0 we find

N (F )
C(T )

T
4kB T 3

Z
d

k 2 N (F )
B
4

2
cosh2 (/2kB T )

+
Z
dy

2 kB2 N (F )
=
,
3

y2
cosh2 (y/2)

(5.44)

(5.45)

(5.46)

which is the well-known linear behavior C(T ) = T for the specific heat at low temperatures,
with = 2 kB2 N (F )/3. Since N (F ) = m kF / 2 ~2 , the effective mass m of the quasiparticles
can directly determined by measuring the specific heat of the system.

99

5.2.2

Compressibility - Landau parameter Fs0

A Fermi gas has a finite compressibility because each fermion occupies a finite amount of space
due to the Pauli principle. The compressibility is defined as6


1
=
(5.48)
p T,N
where p is the uniform hydrostatic pressure. The indices T, N mean, that the temperature T
and the particle number N are kept fixed. We consider the response of the Fermi liquid upon
application of uniform pressure p. The shift of the quasiparticle energies is given by
(k) =

(k)
(k) k
(0)
p =

p =
~v k k p = k (0) p
p
k
p
3

(5.49)

with k = ~v k k/3 = 2 (k)/3. Analogous we introduce the shift of the renormalized quasiparticle energies,

 (k) = k p = k (0) p +

1 X
f 0 (k, k0 )n0 (k0 )
0 0 ,
k ,

= k (0) p +

n 0 (k0 )
1 X
 0 (k0 )
f,0 (k, k0 ) 0
0 0

0 (k )

(5.50)

k ,

= k (0) p

1 X
f 0 (k, k0 )(
0 (k0 ) F )k0 p
0 0 ,
k ,

Changes are concentrated on the Fermi surface such that we can replace k = 2F /3 so that
Z
dk0 s 0
(0)
(5.51)
f (k, k ) = (0) F0s .
= N (F )
4
Therefore we find
=

(0)
.
1 + F0s

(5.52)

Now we determine (0) from the volume dependence of the energy

E

(0)

X
k,

3 ~2 kF2
3 ~2 N
3
=
 (k) = N F = N
5
5 2m
10 m


3

2N

2/3
.

(5.53)

Then we determine the pressure

and



N 2/3 1
3 2

(5.54)

1
p
1 ~2 N
2
=

=
(3 2 n)2/3 = nF .

(0)

3m
3

(5.55)

p=

E (0)

=
N

1 ~2 N
5 m

An alternative definition considers the change of particle number upon change of the chemical potential,
 
1 n
= 2
(5.47)
n
T,

with n = N/.

100

kF

kF

k F

k F

k F

Figure 5.4: Deviations of the distribution functions: Left panel: isotropic increase of the Fermi
surface as used for the uniform compressibility; right panel: spin dependent change of size of
the Fermi surface as used for the uniform spin susceptibility.

5.2.3

Spin susceptibility - Landau parameter Fa0

In a magnetic field H coupling to the electron spins the bare quasiparticle energy is supplemented
by the Zeeman term,
~2 k2

 (k) =
gB H
(5.56)
2m
2
where = 1 denotes the spin component parallel to the applied field. The shift of the
renormalized quasiparticle energy due to the applied field is

 (k) =  (H, k) (H = 0, k)

1 X
= gB H +
f 0 (k, k0 )n0 (k0 )
2 0 0 ,

(5.57)

k ,

=
g B H .
2
Note that by symmetry, n (k) = n (k). Due to interactions, the renormalized gyromagnetic factor g differs from the value of g = 2 for free electrons. We focus on the second term in
Eq.(5.57), which can be reexpressed as
n 0 (k0 )
1 X
1 X
f0 (k, k0 )n0 (k0 ) =
f0 (k, k0 ) 0
 0 (k0 )
0 0
0 0

0 (k )
k ,

k ,

1 X
0
f0 (k, k0 )(
0 (k0 ) F )
g B H
0 0
2

(5.58)

k ,

Combining this result with the Eqs. (5.57) and (5.58), we derive
Z
dk0 a 0
g = g gN (F )
f (k, k ) = g gF0a ,
4

(5.59)

or equivalently
g =

g
.
1 + F0a

(5.60)

The magnetization of the system can be computed from the distribution function,
M = gB

X
k,

= gB

X
k,

n (k) = gB

X n (k)

 (k)
2
 (k)
k,

(
 (k) F )
g B H
101

(5.61)

from which the susceptibility is immediately found to be

=

2 N (F )
M
= B
.
H
1 + F0a

(5.62)

The changes in the distribution function induced by the magnetic field feed back into the susceptibility, so that the latter may be either weakened (F0a > 0) or enhanced (F0a < 0). For the
magnetic susceptibility, the Landau parameter F0a and the effective mass m (through N (F ))
lead to a renormalization compared to the free electron susceptibility.

5.2.4

Galilei invariance - effective mass and Fs1

We initially introduced by hand the effective mass of quasiparticles in  (k). In this section
we show, that overall consistency of the phenomenological theory requires a relation between
the effective mass and one Landau parameter (F1s ). The reason is, that the effective mass
is the result of the interactions among the electrons. This self-consistency is connected with
the Galilean invariance of the system. When the momenta of all particles are shifted by ~q
(|q| shall be very small compared to the Fermi momentum kF in order to remain within the
assumption-range of the Fermi liquid theory) the distribution function given by
(0)
(0)
n (k) = n(0)
(k + q) n (k) q k n (k).

(5.63)

This function is strongly concentrated around the Fermi energy (see Figure 5.5).
n=+1

n=1
kF

Figure 5.5: Distribution function due to a Fermi surface shift (Galilei transformation).
The current density can now be calculated, using the distribution function n (k) = n(0)
(k) +
n (k). Within the Fermi liquid theory this yields,
jq =

1X
1X
v(k)n (k) =
v(k)n (k)

k,

(5.64)

k,

with
1
 (k)
~k

1
1 X
0
0
=
k  (k) +
f 0 (k, k )n (k ) .
~
0 0 k

v(k) =

k ,

102

(5.65)

Thus we obtain for the current density,

1 X ~k
1 X X (0)
1
jq =
n
(k)
+
[n (k) + n (k)] k f0 (k, k0 )n (k0 )

~
0 0
k,

k, k ,

1 X ~k
1 XX 1
0
0
2
=
n
(k)

[k n(0)
(k)]f 0 (k, k )n (k ) + O(q )

m
2
~
0 0
k,

(5.66)

k, k ,

1 XX
~k0
1 X ~k
0
0
n
(k)
+
f
(k,
k
)(
(k
)


)
n (k) + O(q 2 ) = j 1 + j 2 .
0

m
2
m
0 0
k,

k, k ,

where, for the second line, we performed an integration by parts and neglect terms quadratic in
n and, in the third line, used f0 (k, k0 ) = f0 (k0 , k) and
k n(0)
(k) =

n(0)
~2 k
(k)
k  (k) = ( (k) F )k  (k) = ( (k) F ) .
 (k)
m

(5.67)

The first term of equation (5.67) denotes quasiparticle current, j 1 , while the second term can
be interpreted as a drag current, j 2 , an induced motion (backflow) of the other particles due to
interactions.
From a different viewpoint, we consider the system as being in the inertial frame with a velocity
~q/m, as all particles received the same momentum. Here m is the bare electron mass. The
current density is then given by
N ~q
1 X ~k
1 X ~k
jq =
=
n (k) =
n (k).
(5.68)
m

m
k,

k,

Since these two viewpoints have to be equivalent, the resulting currents should be the same.
Thus, we compare equation (5.67) and (5.68) and obtain the equation,
~k
~k
1 X
~k0
= +
f0 (k, k0 )( (k0 ) F )
(5.69)
m
m
0 0
m
k ,

which with k = k/kF then leads to

Z
Z
dk0 s 0 k k0
dk0
1
1
1
1
k0 )
= + N (F )
f (k, k ) = +
N ( )f s (k,
m
m
4
m
m
m
4 P
| F {z
}

l=0

Fls Pl (cos k0 )

cos k0
| {z }

(5.70)

P1 (cos k0 )

or by using the orthogonality of the Legendre polynomials,

m
1
(5.71)
= 1 + F1s .
m
3
where 1/3 = 1/(2l + 1) for l = 1 originates from the orthogonality relation of Legendrepolynomials, as shown above. Therefore, the relation (5.71) has to couple m to F1s in order for
Landaus theory of Fermi liquids to be self-consistent. Generally, we find that F1s > 0 so that
quasiparticles in a Fermi liquid are effectively heavier than bare electrons.

5.2.5

Stability of the Fermi liquid

Upon inspection of the renormalization of the quantities treated previously

m
=
,
0
m

m 1
=
,
0
m 1 + F0s

m 1
=
0
m 1 + F0a
103

(5.72)
(5.73)
(5.74)

with

m
1
= 1 + F1s ,
m
3
and the the response functions of the non-interacting system are given by
0 =

2 mk
mk
kB
3m
F
and 0 = 2B 2 F2
, 0 =
2
2
2
3~
~
n~ kF

(5.75)

(5.76)

one notes that the compressibility (susceptibility ) diverges for F0s 1 (F0a 1),
indicating an instability of the system. A diverging spin susceptibility for example leads to a
ferromagnetic state with a split Fermi surface, one for each spin direction. On the other hand, a
diverging compressibility leads to a spontaneous contraction of the system. More generally, the
deformation of the quasiparticle distribution function may vary over the Fermi surface, so that
arbitrary deviations of the Fermi liquid ground state may be classified by the deformation
=
n (k)

X
+l
X

n,l,m Ylm (k , k )

(5.77)

l=0 m=l

Note that we allow here formally for complex distribution functions. For pure charge density de = n

formations we have n+,l,m (k)

,l,m (k), while pure spin density deformations are described
= n

by n+,l,m (k)
,l,m (k). The general response function for a redistribution n (k) with the
7
anisotropy Ylm (k , k ) is given by
(0)

l,m =

l,m
1+

Fls,a
2l+1

(5.85)

7
General response and distribution deformations: We consider a force field F with conjugate polarisation P
which yields a modification of the quasiparticle dispersion,

 (k) = (k)F

and

 (k) =
(k)F

(5.78)

(k , k ) without spin dependence. Then we can write

where we assume that (k) = Yl,m (k , k ) = (1)m Yl,m

 (k) =  (k) +

X
1 X
k
0 )n0 (k0 ) =  (k) + 1
k
0 ) n0
0 (k0 ) .
f0 (k,
f0 (k,
0 0
0 0
0 (k0 )
k ,

(5.79)

k ,

In the last step we take for n0 (k0 ) the self-consistent value taking the feedback of the quasiparticle coupling
into account. We now use the relation
k
0 ) =
f s,a (k,

k
0 ) = 4
fls,a Pl (k

l=0

+l

X
fls,a X

Ylm (k , k )Ylm
(k0 , k0 )
2l + 1
l=0

(5.80)

m=l

and insert Eq.(5.78) to find

(k)F = (k)F N (F )

fls
Ylm (k , k )

2l + 1

dk0 Ylm
(k0 , k0 ) (k0 )F

(5.81)

which leads straightforwardly to

Fls
2l + 1

=
1+

Fls
2l+1

(5.82)

Now the polarisation is calculated which we may define as

P =

1 X
1 X
n0
(k)n (k) =
(k)

 (k) = N
(F )F

 (k)
k,

(5.83)

k,

such that the linear response is given by

=

2 N (F )
P
=
.
Fls
F
1 + 2l+1

104

(5.84)

Stability of the Fermi liquid against any of these deformations requires

1+

Fls,a
> 0.
2l + 1

(5.86)

If for any deformation channel l this conditions is violated one talks about a Pomeranchuck
instability.8 Generally, the renormalization of the Fermi liquid leads to a change in the Wilson
ratio, defined as
R
0
1
=
=
R0
0
1 + F0a

(5.87)

where R0 = 0 /0 = 62B / 2 kB2 . Note that the Wilson ratio does not depend on the effective
mass. A remarkable feature of the Fermi liquid theory is that even very strongly interacting
Fermions remain Fermi liquids, notably the quantum liquid 3 He and so-called heavy Fermion
systems, which are compounds of transition metals and rare earths. Both are strongly renormalized Fermi liquids. For 3 He we give some of the parameters in Table 5.1 both for zero pressure
and for pressures just below the critical pressure at which He solidifies (pc 2.5MPa = 25bar).
pressure
p=0
p < pc

m /m
3.0
6.2

F0s
10.1
94

F0a
-0.52
-0.74

F1s
6.0
15.7

/0
0.27
0.065

/0
6.3
24

Table 5.1: List of the parameters of the Fermi liquid theory for 3 He at zero pressure and at a
pressure just below solidification.
The trends show obviously, that the higher the applied pressure is, the denser the liquid becomes
and the stronger the quasiparticles interact. Approaching the solidification the compressibility
is reduced, the quasiparticles become heavier (slower) and the magnetic response increases drastically. Finally the heavy fermion systems are characterized by the extraordinary enhancements
of the effective mass which for many of these compounds lie between 100 and 1000 times higher
than the bare electron mass (e.g. CeAl3 , UBe13 , etc.). This large masses lead the notion of
almost localized Fermi liquids, since the large effective mass is induced by the hybridization of
itinerant conduction electrons with strongly interacting (localized) electron states in partially
filled 4f - or 5f -orbitals of Lanthanide and Actinide atoms, respectively.

5.3

Microscopic considerations

A rigorous derivation of Landaus Fermi liquid theory requires methods of quantum field theory
and would go beyond the scope of these lectures. However, plain Rayleigh-Schrodinger theory
applied to a simple model allows to gain some insights into the microscopic fundament of this
phenomenologically based theory. In the following, we consider a model of fermions with contact
interaction U (r r 0 ), described by the Hamiltonian
Z
X

b (r)
b (r 0 ) U (r r 0 )
b (r 0 )
b (r)
H=
k b
cks b
cks + d3 r d3 r0
(5.88)

k,s

X
k,s

k b
cks b
cks +

U X
b
c
b
c 0
b
c 0 b
c .
0 k+q k q k k

(5.89)

k,k ,q

where k = ~2 k2 /2m is a parabolic dispersion of non-interacting electrons. We previously

noticed that, in order to find well-defined quasiparticles, the interaction between the Fermions
has to be short ranged. This specially holds for the contact interaction.
8

I.J. Pomeranchuk, JETP 8, 361 (1958)

105

5.3.1

Landau parameters

Starting form the Hamiltonian (5.88), we will determine Landau parameters for a corresponding
(0)
Fermi liquid theory. For a given momentum distribution nks = hcks cks i = nks +nks , we can expand the energy resulting form equation (5.89) following the Rayleigh-Schrodinger perturbation
method,
E = E (0) + E (1) + E (2) +

(5.90)

with
E (0) =

k nks ,

(5.91)

UX
n n 0 ,
0 k k

(5.92)

U 2 X nk nk0 (1 nk+q )(1 nk0 q )

.
2
k + k0 k+q k0 q
0

(5.93)

k,s

E (1) =

k,k

E (2) =

k,k ,q

The second order term E (2) describes virtual processes corresponding to a pair of particle-hole
excitations. The numerator of this term can be split into four different contributions.
We first consider the term quadratic in nk and combine it with the first order term E (1) , which
has the same structure,
2 X
X
nk nk0
U
(1) = E (1) + U
E

n n 0 .
2
k + k0 k+q k0 q
0 k k
0
k,k ,q

(5.94)

k,k

through,
In the last step, we defined the renormalized interaction U
2 X
1
=U+U
U
.
q k + k0 k+q k0 q

(5.95)

depends on the wave vectors k and k0 . However, when the wave vectors are
In principle, U
restricted to the Fermi surface (|k| = |k0 | = kF ), and if the range of the interaction ` is small
compared to the mean electron spacing, i.e., kF `  1,9 this dependency may be neglected.
Since the term quartic in nk vanishes due to symmetry, the remaining contribution to E (2) is
9

We should be careful with our choice of a contact interaction, since it would lead to a divergence in the large-q
range. A cutoff for q of order Qc `1 would regularize the integral which is dominated by the large-q part.
Thus we may use the following expansion,
ZQc
Z
1 X
1
1
m
2
=
dq
q
dq 0
q k + k0 k+q k0 q
(2)3
(k k) q q 2

(5.96)



ZQc
q K 
d cos
m


dq q
=
dq
q
ln
q + K 
K cos q
(2)2
1
0



m
K 2 Q2c  Qc K 
=
Qc +
ln 
(2)2
2K
Qc + K 



2
4
K
2mQc
K

1 2 +O
,
(2)2
Qc
Q4c
m
=
(2)2

Z+1

(5.97)

(5.98)
(5.99)

is indeed
where we use K = |k0 k| 2kF  Qc . From this we conclude that the momentum dependence of U
weak.

106

cubic in nk and reads

2 X nk nk0 (nk+q + nk0 q )
(2) = U
.
E
2
k + k0 k+q k0 q
0

(5.100)

k,k ,q

2 , which is admissible at this order of the perturbative expansion. The

We replaced U 2 by U
variation of the energy E in Eq.(5.90) with respect to nk can easily be calculated,
 (k) = k +

X
2 X nk0 (nk+q + nk0 q ) nk+q nk0 q
U
U
nk 0 2
,
0
0
k + k0 k+q k0 q

(5.101)

k ,q

and an analogous expression is found for  (k). The coupling parameters f0 (k, k0 ) may be
determined using the definition (5.26). Starting with f (kF , k0F ) with wave-vectors on the
Fermi surface (|kF | = |k0F | = kF ), the terms contributing to the coupling can be written as

(0)
(0) 
2
2
0
X
X
X
n

n
0
0
0
0
k+qkF 1
U
U
k q
k
k q
k 
n

n
0
k+q
k

F
2 0
k + k0 k+q k0 q
0
0 k0 k0 q 
k ,q
k
k
0
q=kF kF

(5.102)

2
U
1X
nk0 0 (k0F kF ),
F
0
2

(5.103)

kF

(0)

(0)

where we consider nk0 = nk0 + nk0 . Note that the part in this term which depends on nk0
F
F
F
F
will contribute the ground state energy in Landaus energy functional. Here, 0 (q) is the static
Lindhard susceptibility as it was defined in (3.63). With the help of equation (5.26), it follows
immediately, that
2
U
(k k0F ).
2 0 F
The other couplings are obtained in a similar way, resulting in
f (kF , k0F ) = f (kF , k0F ) =

f (kF , k0F ) = f (kF , k0F ) = U

2 

U
2
0 (kF + k0F ) 0 (kF k0F ) ,
2

(5.104)

(5.105)

where the function

0 (q) is defined as
(0)

(0)

1 X nk0 +q + nk0

0 (q) =
0 2F k0 +q k0

(5.106)

If the couplings are be parametrized by the angle between kF and k0F , they can be expressed
as
"
!
N ( ) 

U
U
cos

1
+
sin(/2)
F
f0 () =
1+
2+
ln
0
(5.107)
2
4
2 sin(/2)
1 sin(/2)
#
!
N ( ) 
U
sin(/2)
1
+
sin(/2)
F
1+
1
ln
0 .
(5.108)
4
2
1 sin(/2)
Finally, we are in the position to determine the most important Landau parameters by matching
the expressions (5.104) and (5.105) to the parametrization (5.107),
h

i
1
F0s = u
1+u
1 + (2 + ln(2)) = u
+ 1.449 u
2 ,
(5.109)
6
h

i
2
F0a =
u 1+u
1 (1 ln(2)) =
u 0.895 u
2 ,
(5.110)
3
2
F1s = u
2 (7 ln(2) 1)

0.514 u
2 ,
(5.111)
15
107

 N ( )/2 has been introduced for better readability. Since the Landau parameter
where u
=U
F
s
F1 is responsible for the modification of the effective mass m compared to the bare mass m,
m is enhanced compared to m for both attractive (U < 0) and repulsive (U > 0) interactions.
Obviously, the sign of the interaction U does not affect the renormalization of the effective mass
m . This is so, because the existence of an interaction (whatever sign it has) between the particles
enforces the motion of many particles whenever one is moved. The behavior of the susceptibility
and the compressibility depends on the sign of the interaction. If the interaction is repulsive
(
u > 0), the compressibility decreases (F0s > 0), implying that it is harder to compress the
Fermi liquid. The susceptibility is enhanced (F0a < 0) in this case, so that it is easier to polarize
the spins of the electrons. Conversely, for attractive interactions (
u < 0), the compressibility is
s
enhanced due to a negative Landau parameter F0 , whereas the susceptibility is suppressed with
a factor 1/(1 + F0a ), with F0a > 0. The attractive case is more subtle because the Fermi liquid
becomes unstable at low temperatures, turning into a superfluid or superconductor, by forming
so-called Cooper pairs. This represents another non-trivial Fermi surface instability.

5.3.2

Distribution function

Finally, we examine the effect of interactions on the ground state properties, using again
Rayleigh-Schr
odinger perturbation theory. The calculation of the corrections to the ground
state |0 i, the filled Fermi sea can be expressed as
|i = |(0) i + |(1) i +

(5.112)

where
|(0) i = |0 i
|(1) i =

(5.113)
ck0 ,s0 b
ck,s
ck0 q,s0 b
b
ck+q,s b

U X X
| i.
0
k + k0 k+q k0 q 0
0

(5.114)

k,k ,q s,s

The state |0 i represents the ground state of non-interacting fermions. The lowest order correction involves particle-hole excitations, depleting the Fermi sea by lifting particles virtually
above the Fermi energy. How the correction (5.114) affects the distribution function, will be
cks i is obtain as the expectation value,
discussed next. The momentum distribution nks = hb
cks b
nks =

h|b
cks b
cks |i
(0)
(2)
= nks + nks +
h|i

(5.115)

(0)

where nks is the unperturbed distribution (kF |k|), and

U 2 X (1 nk1 )(1 nk2 )nk3

2
2 k+k3 ,k1 +k2

k1 ,k2 ,k3 (k + k3 k1 k2 )

(2)
nks =

nk1 nk2 (1 nk3 )

U2 X

2
(k1 + k2 k k3 )2 k+k3 ,k1 +k2
k1 ,k2 ,k3

|k| < kF
.

(5.116)

|k| > kF

This yields the modification of the distribution functions as shown in Figure 5.6. It allows us
also to determine the size of the discontinuity of the distribution function at the Fermi surface,


U N (F ) 2
nk F nk F + = 1
ln(2),
(5.117)
2
where
nk F =

lim

|k|kF 0

108

(0)

(2)

nk + nk

(5.118)

The jump of nk at the Fermi surface is reduced independently of the sign of the interaction.
The reduction is quadratic in the perturbation parameter U N (F ). This jump is also a measure
for the weight of the quasiparticle state at the Fermi surface.

nk

nk

1D

3D
kF

kF

Figure 5.6: Momentum distribution functions of electrons for a three-dimensional (left panel)
and one-dimensional (right panel) Fermion system.

5.3.3

Fermi liquid in one dimension?

Within a perturbative approach the Fermi liquid theory can be justified for a three-dimensional
system and we recognize the one-to-one correspondence between bare electrons and quasiparticles renormalized by (short-ranged) interactions. Now we would like to show that within the
same approach problems appear in one-dimensional systems, which are conceptional nature and
hint that interaction Fermions in one dimension would not form a Fermi liquid, but a Luttinger
liquid, as we will motivate briefly below.
The Landau parameters have been expressed above in terms of the response functions 0 (q =
kF k0F ) and
0 (q = kF k0F ). For the one-dimensional system, as given in Eqs.(5.104 - 5.106),
the relevant contributions come from two configurations, since there are two Fermi points only
(instead of a two-dimensional Fermi surface),
(kF , kF0 )

q = kF kF0 = 0, 2kF .

(5.119)

We find that the response functions show singularities for some of these momenta,10 and we
obtain
f (kF , kF ) = f (kF , kF )
(5.122)
as well as
f (kF , kF ) = f (kF , kF )

(5.123)

giving rise to the divergence of all Landau parameters. Therefore the perturbative approach to
a Landau Fermi liquid is not allowed for the one-dimensional Fermi system.
10
While 0 (q) is the Lindhard function given in Eq.(3.112) which diverges logarithmically at q = 2kF , we
obtain for the other response function



2m
2kF + q 2kF q

ln
|q| < 2kF

2
2
2kF + q + 2kF q

~ 4kF q

0 (q) =
(5.120)
s
s
!

2m
q
+
2k
q

2k
F
F

arctan
+ arctan
|q| > 2kF
2p 2
q 2kF
q + 2kF
~ q 4kF2

which diverges as
lim
0 (q) =

q0

m
q
ln(
),
~2 kF
2kF

lim

q2kF

0 (q) =

m
~2 kF

109

and

lim

q2kF +

0 (q) =

m
1

. (5.121)
~2 2kF q 2kF

The same message is obtained when looking at the momentum distribution form which had in
three dimensions a step giving a measure for the (reduced but finite) quasiparticle weight. The
analogous calculation as in Sect.5.3.2 leads here to

k+
1 U2

ln
k > kF

2
2
2
8 ~ vF
k kF
(2)
.
(5.124)
nks

k
U
1

k < kF
2 2 2 ln
8 ~ vF
kF k
Here, k are cutoff parameters of the order of the Fermi wave vector kF . Apparently the
quality of the perturbative calculation deteriorates as k kF , since we encounter a logarithmic
divergence from both sides.

Ladung

q=e

Spin

S=0

S=1/2

q=0

Figure 5.7: Visualization of spin-charge separation. The dominant anti-ferromagnetic spin correlation is staggered. A charge excitation is a vacancy which can move, while spin excitation
may be considered as domain wall. Both excitations move independently.
Indeed, a more elaborated approach shows that the distribution function is continuous at k = kF
in one dimension, without any jump. Correspondingly, the quasiparticle weight vanishes and the
elementary excitations cannot be described by Fermionic quasiparticles but rather by collective
modes. Landaus Theory of Fermi liquids is inappropriate for such systems. This kind of
behavior, where the quasiparticle weight vanishes, can be described by the so-called bosonization
of fermions in one dimension, a topic that is beyond the scope of these lectures. However,
a result worth mentioning, shows that the fermionic excitations in one dimensions decay into
independent charge and spin excitations, the so-called spin-charge separation. This behavior can
be understood with the naive picture of a half-filled lattice with predominantly antiferromagnetic
spin correlations. In this case both charge excitations (empty or doubly occupied lattice site)
and spin excitations (two parallel neighboring spins) represent different kinds of domain walls,
and are free to move at different velocities.

110

Chapter 6

Transport properties of metals

The ability to transport electrical current is one of the most remarkable and characteristic
properties of metals. At zero temperature, a ideal pure metal is a perfect electrical conductor,
i.e., its resistivity is zero. However, disorder due to impurities and lattice defects influence the
transport and yield a finite residual resistivity, as found in real materials. At finite temperature,
electron-electron and electron-phonon scattering lead to a temperature-dependent resistivity.
Furthermore, an external magnetic field may influence the resistivity, a phenomenon called
magnetoresistance, and also leads to the previously studied Hall effect. In this chapter, the
effects of a magnetic field will not be considered. Finally, heat transport, which is also mostly
mediated by electrons in metals, is going hand in hand with the electric transport. In this
context, transport phenomena such as thermoelectricity (Seebeck and Peltier effect) will be
analyzed here.

6.1

Electrical conductivity

In a normal metal, an electrical current density j(q, ) (in q, -space) is induced by an applied
electrical field E(q, ). For a homogeneous isotropic metal, we define the scalar1 electrical
conductivity (q, ) within linear response, through
j(q, ) = (q, )E(q, ).

(6.1)

The current density j(q, ) is related to the charge density (r, t) = en(r, t), via the continuity
equation

(r, t) + j(r, t) = 0,
t

(6.2)

(q, ) q j(q, ) = 0.

(6.3)

or, in Fourier transformed,

It is interesting to see that a relation between the conductivity (q, ) and the dynamical
dielectric susceptibility 0 (q, ) defined in equation (3.68) of chapter 3 arises from the equations
(6.1) and (6.3). For this, we can calculate
(q, )
q j(q, )
=
eV (q, )
eV (q, )
(q, ) q E(q, )
(q, ) [iq 2 V (q, )]
=
=
.
e
V (q, )

e2 V (q, )

0 (q, ) =

In an anisotropic material, the conductivity

would be a full 3 3 tensor.

111

(6.4)

In the first line, we used the definition (3.68) of 0 (q, ) and the continuity equation (6.3). To
the second line, we made use of the definition (6.1) of (q, ) and then replaced E(q, ) by
iqV (q, ), which is nothing else than the Fourier transform of the equation
eE(r, t) = r V (r, t).

(6.5)

From this calculations we conclude that

0 (q, ) =

iq 2
(q, ),
e2

(6.6)

and thus
(q, ) = 1

4i
4e2
0 (q, ) = 1 +
(q, ).
2
q

(6.7)

In the limit of large wavelengths q  kF , we know from previous discussions2 that (0, ) =
1 p2 / 2 . Then the conductivity simplifies to
() =

ip2
4

(6.8)

One might conclude from this result that the conductivity is purely imaginary in the small-q
limit. However, this conclusion is wrong, since the real part of () is related to its imaginary
part via the Kramers-Kronig relation. Defining 1 (2 ) as the real (imaginary) part of , this
relation states that
+

Z
1
1
1 () = P
d 0
( 0 )
(6.9)

0 2

and
+

Z
1
1
2 () = P
d 0
( 0 ) .

0 1

(6.10)

A simple calculation with 2 from equation (6.8), yields

p2

(),
4
p2
2 () =
.
4

1 () =

(6.11)
(6.12)

Obviously this metal is perfectly conducting ( for 0), which comes from the fact
that we considered systems without dissipation so far.
An additional important property coming from complex analysis, is the existence of the so-called
f -sum rule,
Z
0

1
d 1 ( ) =
2
0

+
Z
p2
e2 n
d 0 1 ( 0 ) =
=
.
8
2m

This relation is valid for all electronic systems (including semiconductors).

2

In the small-q limit we approximate 0 (q, ) nq 2 /m 2 from equation (3.82).

112

(6.13)

6.2

Transport equations and relaxation time

We introduce here Boltzmanns transport theory as as rather simple and efficient way to deal
with dissipation and momentum relaxation of non-stationary electronic states in metals.

6.2.1

The Boltzmann equation

In order to tackle the problem of a finite conductivity, we use a formalism similar to Landaus
Fermi liquid theory, based on a distribution function of quasiparticles. In transport theory, the
distribution function can be used to describe the deviation of the system from an equilibrium.
If the system is isolated from external influence, equilibrium is reached through relaxation after
some time, a process which is accompanied with an increase of entropy as discussed in statistical
physics. Analogously to the theory of transport phenomena, let us introduce the distribution
function3 f (k, r, t), where
f (k, r, t)

d3 k 3
d r
(2)3

(6.14)

is the number of particles in the infinitesimal phase space volume d3 rd3 k/(2)3 centered at (k, r),
at time t. Such a description is only applicable if the temporal and spacial variations occur at
long wavelengths and small frequencies, respectively, i.e., if typically q  kF and ~  F . The
total number of particles N is given by
Z
d3 k 3
N =2
d rf (k, r, t).
(6.15)
(2)3
The equilibrium distribution f0 for the fermionic quasiparticles is given by the Fermi-Dirac
distribution,
f0 (k, r, t) =

1
,
e(k )/kB T + 1

(6.16)

and is independent of space r and time t. The general distribution function f (k, r, t) obeys the
Boltzmann equation


 

f
D
,
(6.17)
f (k, r, t) =
+ r r + k k f (k, r, t) =
Dt
t
t coll
where the substantial derivative in phase space D/Dt is defined as the total temporal derivative
in a frame moving with the phase-space volume. The right-hand side is called collision integral
and describes the rate of change in f due to collision processes. Without scattering, the equation
(6.17) would represent a continuity equation for f . Now, consider the temporal derivatives of r
and k from a quasi-classical viewpoint. In absence of a magnetic field, we find
~k
,
m
~k = eE,
r =

(6.18)
(6.19)

which is our central interest, originates from the electric field. The collision
i.e., the force ~k,
integral may be expressed via the probability W (k, k0 ) to scatter a quasiparticle with wave
vector k to k0 . For simple scattering on static potentials, the collision integral is given by
 
Z
f
d3 k 0 
=
W (k, k0 )f (k, r, t)(1 f (k0 , r, t))
(6.20)
t coll
(2)3

W (k0 , k)f (k0 , r, t)(1 f (k, r, t)) .
(6.21)
3

For simplicity we neglect spin the electron spin. In general there would be a distribution function f (k, r, t)
for each spin species .

113

The first term, describing the scattering4 from k to k0 , requires a quasiparticle at k, hence
the factor f (k, r, t), and the absence of a particle at k0 , therefore the factor 1 f (k0 , r, t).
This process describes the scattering out of the phase space volume d3 k/(2)3 , i.e., reduces the
number of particles in it. Therefore, it enters the collision integral with negative sign. The
second term describes the opposite process and, according to its positive sign, increases the
number of particles in the phase space volume d3 k/(2)3 . For a system with time inversion
symmetry, we have W (k, k0 ) = W (k0 , k). Assuming this, we can combine both terms and end
up with
 
Z


f
d3 k 0
W (k, k0 ) f (k0 , r, t) f (k, r, t) .
(6.22)
=
3
t coll
(2)
The Boltzmann equation is a complicated integro-differential equation and suitable approximations are required. Usually, we study processes close to equilibrium, where the deviation
f (k, r, t) f0 (k, r, t) is small compared to f (k, r, t). Here, to generalize we assume f0 (k, r, t)
to be a local equilibrium distribution for which the temperature T = T (r, t) and the chemical
potential = (r, t) vary slowly in r and t, such that f0 (k, r, t) can still be expressed via the local Fermi-Dirac distribution (6.16). At small deviations from equilibrium (or local equilibrium),
we can approximate the collision integral by the relaxation-time approximation. For simplicity,
we assume that the system is isotropic, such that the quasiparticle dispersion k only depends
on |k| and, furthermore, that the scattering probabilities are elastic and depend on the angle
between k and k0 . Then, we make the Ansatz
 
f (k, r, t) f0 (k, r, t)
f
=
.
(6.23)
t coll
(k )
The time scale (k ) is called relaxation time and gives the characteristic time within which the
system relaxes to equilibrium.
Consider the simplest case of a system at constant temperature subject to a small uniform electric
field E(t). With f (k, r, t) = f0 (k, r, t) + f (k, r, t), we can calculate the Fourier-transform of
Boltzmann equation (6.17) in relaxation-time approximation and find, after linearizing in f ,
if (k, )

eE()
f (k, )
k f0 (k) =
~
(k )

(6.24)

with
Z

f (k, t) =

d
f (k, )eit
2

and E(k, t) =

d
E(k, )eit .
2

(6.25)

In order to come to this expression, we used that f (k, r, t) = f (k, t) for E = E(t) are spacially
uniform, and assumed for linearizing equation (6.24) that f |E|. Thus, the equation (6.24)
is consistent to linear order in |E| and can be easily solved as
f (k, ) =

e E()
e E() f0 ()
f (k) =
k k .
~(1 i ) k 0
~(1 i ) 

This result leads straightforwardly to the quasiparticle current j(),

Z
Z
( )[E() v k ]v k f0 (k )
d3 k
e2
j() = 2e
v f (k, ) = 3 d3 k k
,
(2)3 k
4
1 i (k )
k

(6.26)

(6.27)

4
Note that if spin was considered here, this collision term would account for scattering processes where spin is
conserved. However, there are in principle also scattering process where the electron spin can be transferred to
the lattice (spin-orbit coupling) or an impurity (Kondo effect) and would not be conserved independently.

114

with ~v k = k k , which in turn can be simplified to

X
j () =
()E (),

(6.28)

where the conductivity tensor reads

Z
Z
vk vk
f0 ()
()
e2
dk k 2
= 3 d
.
4
 1 i ()
~|v k |

(6.29)

This corresponds to the Ohmic law. Note that = in isotropic systems. We recover in
this case the expression (6.1) for the conductivity, which we introduced at the beginning of this
chapter. It is also important to notice that f0 ()/ concentrates around  = for T  TF .
In the following, we consider the result (6.29) for an isotropic system in different limiting cases.

6.2.2

The Drude form

For  1 equation (6.29) becomes independent on the relaxation time. In an isotropic system
= at low temperatures T  TF , this leads to
e2 m2 v
() i 3 3 F
4 ~

dk vF2 z = i

p2
e2 n
=i
,
m
4

(6.30)

which reproduces the result from equation (6.8). However now, this does not mean that our
system is a perfect conductor, as becomes clear when we turn towards small (  1). We are
actually interested in the static limit, where the dc conductivity ( = 0; dc means direct
current) reduces to
Z
p2
f
e2 n
e2 n

=
d 0 () =
=
.
(6.31)
m

m
4
Since the function f0R/ is strongly peaked around the Fermi energy F , we introduced a mean
relaxation time = d ()f0 /. In the form (6.31), the result recovers the well-known
Drude5 form of the conductivity.
If the relaxation time depends only weakly on energy, we can simply calculate the optical
conductivity at finite frequency,
!
2
p2
p2

i
() =
= 1 + i2 .
(6.32)
=
+
4 1 i

4 1 + 2 2 1 + 2 2
Note that the real part satisfies the f -sum rule,
Z

Z
d 1 () =

p2

p2

=
4 1 + 2 2
8

(6.33)

and that () recovers the behavior of equation (6.12) in the limit .6 This form of the
conductivity yields the dielectric function
() = 1

p2
(i +
)

=1

p2 2
1 + 2 2

2
i p
,
1 + 2 2

(6.35)

Note, that the phenomenological Drude theory of electron transport can be deduced from purely classical
considerations.
6
Note that 1 () has Lorentzian form and naturally leads to a delta function:
1 () =

p2

4 1 + 2 2

115

p2
() .
4

(6.34)

which can be used to discuss the optical properties of metals. The complex index of refraction
n + ik is given through (n + i)2 = . Next, we discuss three important regimes of frequency.
To probe optical properties we consider the reflectivity of light hitting a metal surface perpendicularly. Then the reflectivity is given by
R=

(n 1)2 + 2
.
(n + 1)2 + 2

(6.36)

Inside the metal the light propagation is renormalised by k k(n + i) with k = /c such that
eik(n+i)z = eiknz ekz = eiknz ez/

defining the penetration length

=

c
.

(6.37)
(6.38)

Relaxation-free regime (
 1  p )
In this limit, the real (1 ) and imaginary (2 ) part of the dielectric function (6.35) read
1 () p2 2 ,
2 ()

(6.39)

p2

.
(6.40)

The real part 1 is constant and negative, whereas the imaginary part 2 becomes singular in
the limit 0. Thus, the refractive index turns out to be dominated by 2
s
r
p2
2 ()
n() ()

 1,
(6.41)
2
2
As a result, the reflectivity R is practically 100%. The absorption index () determines the
penetration depth through
r
c
2
c

.
(6.42)
() =
()
p

With this, the skin depth of a metal with the famous relation () 1/2 is reproduced
within the relaxation time approximation of the Boltzmann equation. While length () is in
the centimeter range for frequencies of the order of 10 100Hz, the Debye length c/p , is only
of the order of 100
A for ~p = 10 eV. (cf. Figure 6.1).
Relaxation regime (1 
 p )
Here, we can expand the dielectric function (6.35) in (
)1 , yielding
() = 1

p2
2

+i

p2
3

(6.43)

The real part 1 p2 / 2 is large and negative and dominates in magnitude over 2 . For the
optical properties, we obtain
p
(6.44)
()

p
.
(6.45)
n()
2 2
We find ()  n()  1, which implies a large reflectivity of metals in this frequency range as
well. Note that visible frequencies are part of this regime (see Figures 6.2 and 6.3). The frequency
dependence of the penetration depth becomes weak, and its magnitude is approximately given
by the Debye length, c/p .
116

Figure 6.1: The frequency dependent reflectivity and penetration depth for p = 500.

Ultraviolet regime ( p and > p )

In this regime, the imaginary part of is approximately zero and the real part has the well
known form
1 () = 1

p2
2

(6.46)

such that the reflectivity drops drastically, from close to unity towards zero (cf. Figure 6.1).
Metals become nearly transparent in the range > p . In Figure 6.1, one also notices the rapid
increase in the penetration depth , showing the transparency of the metal.
In all these considerations, we have neglected the contributions to the dielectric function due to
the ion cores (core electrons and nuclei). This may be incorporated in Eq.(6.46) in the following
approximate way:
1 () =

p2
2

(6.47)

This influences the reflecting properties of metals; particularly, the value of p is reduced to

p0 = p / , where is the frequency-independent part of the dielectric function. With this,

the reflectivity for frequencies above p0 approaches R = ( 1)2 /( + 1)2 , and 0 < R < 1
(see Figure 6.2 and 6.3).
Color of metals
The practically full reflectance for frequencies below p is a typical feature of metals. Since for
most metals, the plasma frequency lies well above the range of visible light (~ = 1.5 3.5eV ),
they appear shiny to our eye. While most polished metal surfaces appear shiny white, like
silver, there are some metals with a color, like gold which is yellow and copper which is reddish.
White shininess results from reflectance on the whole visible frequency range, while for colored
metals there is a certain threshold above which the reflectance drops and frequencies towards
blue are not or much weaker reflected. In most cases this drop is not connected with the plasma
frequency, but with light absorption due to interband transitions. Note that the single band
metal which was used for the Drude theory does not allow for optical absorption apart from the
plasma excitation. Interband transition play a particularly important role for the noble metals,
Cu, Ag and Au. For these metals, the reflectance drop is caused by the transition from the
117

polelnrler '(Ae 6 - dcD4)3y pue (na 8'0t :'totl)

nJ ot Surpuodserrot serSreua
'y
uoloqd etuseld eqJ 're^lrs pue
eJt3eds{1r,tr1ceger

pa>lretuueaqe^eq
Puu (nJ) ,Le Z
't) uorlenbg ,(q

001

0y

l uleldxe urlceds
rqlIA\ pelelJossu
el qJIq^\ 'se8pe

OL

---l,, I

l a $= "4! t

t oz{leue oI

J d' ulnutulnle JO

ul uI suolllsueJl
,rt8 'lE '{8leue
g'lepotu JoIrIuJ
Ae E'l lnoqu l

L'0
\v

9Z

0z

9t

lo,{ltutctn eql ut
l dlp eql 'snqJ
ul uI luelJgJeoJ

nc

slels 3o {lrsuaP
soLule le JnJJO
uJec eql
^\oleq
tndpelleJ-os eql

=.

OL
n
o
=
o
001 o
L

t' 0

A ' 8'0t - " rr 4

n J e JJe f g
ur drP
r,r.t1ceger

100

9202910190

'l ]e ueql Je^\ol

ql se 'JeAe,^AoH
e /Y\o[eqspueq

(1e) {6reu3
'Q961'ddrpq4pue
qJIaJuoJqE
qlr,,r,r
ruor; uoISSrr-urad
pecnpo:de:)esecqlee ur peluJrpursr ,{cuanbe:.;
eusuld eql ol
Surpuodse"rroc
,{3reueuoto{d aq1 ':eddocpue Je^lrs;o erlceds,(1r,trlcegar
aq1 91'p arn8rg

6.2.3

The relaxation time

By replacing the collision integral by a relaxation time approximation, we implicitly introduced

a connection between the scattering rate W (k, k0 ) and the relaxation time . This relation,
Z
f (k) f0 (k)
d3 k 0
=
W (k, k0 ){f (k) f (k0 )},
(6.48)
(k )
(2)3
will be studied for an isotropic system, with elastic scattering, and a small external field E. The
solution of equation (6.24) is of the form
f (k) = f0 (k) + A(k)k E,

(6.49)

f (k) f (k0 ) = A(k)(k k0 ) E.

(6.50)

such that

118

llclpoJd eqt ot

L0'0 +

completely occupied d-band to the partially filled s-band, 3d 4s in case of Cu. For copper,
this drop appears below 2.5eV so that predominantly red light is reflected (see Figure 6.2). For
gold, this threshold frequency is slightly higher, but still in the visible, while for silver, it lies
beyond the visible range (see Figure 6.2). For all these cases, the plasma frequency is not so
easily recognizable in the reflectance. On the other hand, aluminum shows a reflectance rather
close to the expected behavior (see. Figure 6.3). Also here, there is a small reduction of the
reflection due to interband absorption. However, this effect is weak and the strong drop occurs
at the plasma frequency of ~p = 15.8eV . Like silver also polished aluminum is white shiny.

ql JoJsJunocJu

0t

Figure 6.2: Reflectance spectra for silver and copper. In both cases the drop of reflectance is due
to optical transitional between the completely filled d-band and the partially filled s-band. Note
the logarithmic scale for the reflectivity. (Source: An introduction to the optical spectroscopy of
inorganic solids, J. Garca Sole, L.E. Bausa and D. Jaque, Wiley (2005))

AND INSULATORS
SEMICONDUCTORS
::-: . :.lletillS
- .. : . . h or t er
. , :: - i r t ha n

1.0
0.8

. r rf tl g
- :. L lp Io
. t. 1nc es
_ _.

r27

a 0 .6
o

g
o
E

0.4
0.2

1 -l -),

0.0

10

15

20

25

hot(eV)
*. -- i

withthosepredicted
spectrum
of Aluminum(full line)compared
Figure4.5 Thereflectivity

f :
anda damped
oscillator
ltato: 15.8eV (dotted
fromtheideal
metalmodelwith
Figure 6.3: Reflectance
spectrum
of aluminum.
Theline)
slight
reduction
ofwith
reflectivity
below p is
withpermission
fromEhrenreich
line)(experimental
datareproduced
1.25x 16tar-t (dashed
-:.1+
due to interbandet al.,1962).
transitions. The thin solid line is the theoretical behavior for = 0 and the
dashed line for finite . (Source: An introduction to the optical spectroscopy of inorganic solids,
J. Garca Sole, L.E. Baus
a and D. Jaque, Wiley (2005))
In Figure4.5,theexperimentalreflectivityspectrumof aluminumis comparedwith
thosepredictedby the ideal metal andthe dampedmetal models.Al hasa free electron
densityof ly' : 18.I x 1922.*-3 (threevalenceelectronsper atom)andso,according
l5.SeV.Thus,thereflectivityspectrum
Without loss of toEquation(4.20),itsplasmaenergyisltc,;o:
generality, we define z k k, and introduce
the parametrization of the angles ,
for the ideal
metal
can be now calculated.Compared0to the experimentalspectrum,
0
0
polar angle of Ethe
and

(
)
polar
(azimuth)
angle
of
k
),
leading
to into account
ideal metal model spectrumis only slightly improvedwhen taking
the dampingterrn,with f : I.25 x 10la s-1, & valuededucedfrom DC conductivity
kThe
Emain
= kE
cos , between the two calculated spectraare that
(6.51)
differences
measurements.
0reflectivity
0 slightly
0
the
ultraviolet
below
op
and
less
one
a
than
dampingproduces
k k = kk cos ,
(6.52)
transmissionedgeis slightly smoothedout.
0
0
0
0
0
k Ebe=mentioned
k E(cos
cos
the
+ sin
metal
sin model
cos nor).the damped
(6.53)
neither
ideal
that
Finally, it should
metal model are able to explain why the actualreflectivity of aluminum is lower than
For elastic scattering,
k = one
k 0 , (R
weryobtain
1) at frequencieslower thanrt r. Also, thesesimple models
the calculated
around 1.5eV. In order
do not reproducefeaturessuchas the
 reflectivity dip observed

0
0
0
(k) for
f (k
)=
A(k)kE
cos
cos understanding
) sin sin
cos
0 .
(6.54)
of real
metals,
have
and then
to (1
a better
to faccount
these
aspects,
the band structuremust be taken into account.This will be discussedat the end of
0
Inserting this into
right-hand
in Sectionside
4.8. of equation (6.48), the -dependent part of the integration
thisthe
chapter,

vanishes for an isotropic system, and we are left with

Z
f (k) f0 (k)
4.5 SEMICONDUCTORS
= dk0AND
[f (k)INSULATORS
f (k0 )]W (k, k0 )
(k )
Z
Unlike metals,semiconductorsand insulatorshave bound valenceelectrons.This
0 next section
and
transitions.
aspectgivesrise to interband
= A(k)kE
cos The
objective
dk0 (1ofthis
cos
the
)W (k, k0 ) is
Z
= [f (k) f0 (k)] dk0 (1 cos 0 )W (k, k0 ).
The factor [f (k) f0 (k)] can be dropped on both sides of the equation, resulting in
Z
d3 k 0
1
=
W (k, k0 )(1 cos 0 ),
(k )
(2)3

(6.55)
(6.56)
(6.57)

(6.58)

where one should remember that, for elastic scattering, the quasiparticle energy k = k0 is
conserved in the collision process. The scattering probability W (k, k0 ) accounts for this restriction. In the next few sections we discuss different scattering processes, looking at collision
probabilities, relaxation times and the resulting conductivity and resistivity contributions.

6.3
6.3.1

Impurity scattering
Potential scattering

Every deviation from the perfect periodicity of the ionic lattice is a source of quasiparticle
scattering, leading to the loss of their original momentum. Without translational invariance,
119

the conservation of momentum is lost, the energy, however, is still conserved. Possible static
scatterers are among others vacancies, dislocations, and impurity atoms. The scattering rate
W (k, k0 ) for a potential Vb can be determined applying Fermis golden rule,7
W (k, k0 ) =

2
n |hk0 |Vb |ki|2 (k k0 ).
~ imp

(6.59)

By nimp we denote the density of impurities, assuming only one species of them. For small densities nimp , it is reasonable to neglect interference effects between different impurities. According
to equation (6.58), the relaxation time of a quasiparticle with momentum ~k is given by
Z
1
2
d3 k 0
|hk0 |Vb |ki|2 (1 k k0 )(k k0 )
(6.60)
=
nimp
(k )
~
(2)3
Z
d 0
d
(k, k0 )(1 k k0 ) k ,
(6.61)
= nimp (k v k )
d
4
with the differential scattering cross section d/d and k = k/|k|. Here, we used the connection8
between Fermis golden rule and the Born approximation. Note the difference in the expressions
for the relaxation time in equation (6.58) and for the lifetime ,
Z
d3 k
1
=
W (k, k0 ),
(6.65)

(2)3
given by Fermis golden rule. The factor (1 cos 0 ) in equation (6.58) gives more weight to
backscattering (0 ) compared to forward scattering (0 0), since the former has more
influence in impeding transport. This explains why is also termed transport lifetime.
Assuming defects in the form of point charges Ze, whose screened potential is
hk0 |Vb |ki =

4Ze2
2 .
|k k0 |2 + kTF

(6.66)

In the limit of very strong screening, kTF  kF , the differential cross section becomes independent of the deviation (k k0 ), the transport and the usual lifetime become equal, = , 9
and

2
1

4Ze2
N (F )nimp
.
(6.67)
2

~
kTF
With this, we are now able to determine the conductivity for scattering on Coulomb defects,
assuming s-wave scattering only. Then, since () depends weakly on energy, equation (6.31)
yields
=

e2 n (F )
,
m

(6.68)

7
This corresponds to the first Born approximation in scattering theory. Note, that this approximation is
insufficient to describe resonant scattering.
8
The scattering of particles with momentum ~k into the solid angle dk0 around k0 yields
2 X
W (k, k0 )dk0 =
|hk0 |Vb |ki|2 (k k0 )
(6.62)
~ 0
k dk0
Z
2
d3 k 0 0 b
2
=
dk0
hk |V |ki|2 (k k0 ) =
dk0 N ()|hk0 |Vb |ki|2 .
(6.63)
~
(2)3
~
k0 dk0

The scattering per incoming particle current jin d(k, k0 ) = W (k, k0 )dk0 determines the differential cross section
v k d (k, k0 ) = 2 N () |hk0 |Vb |ki|2 .
k
d
~ 4
leading to equation (6.61).
9
In the context of partial wave expansion, one speaks of s-wave scattering, i.e., l>0 0.

120

(6.64)

or, for the specific resistivity = 1/,

=

m
e2 n (

(6.69)

Both and are independent of temperature. This contribution is called the residual resistivity of a metal, which approaches zero for a perfect material. The temperature dependence
of the resistivity is induced in other scattering processes like electron-phonon scattering and
electron-electron scattering, which will be considered below. The so-called residual resistance
ratio RRR = R(T = 300K)/R(T = 0) is an often used quantity to benchmark the quality of
a material. It is defined as the ratio between the resistance R at room temperature and the
resistance at zero temperature. The bigger the RRR, the better the quality of the material.
The typical value of RRR for common copper is 40-50, while the RRR for very clean aluminum
reaches values up to 20000.

6.3.2

Kondo effect

There are impurity atoms inducing so-called resonant scattering. If the resonance occurs close to
the Fermi energy, the scattering rate is strongly energy dependent, inducing a more pronounced
temperature dependence of the resistivity. An important example is the scattering off magnetic
impurities with a spin degree of freedom, yielding a dramatic energy dependence of the scattering
rate. This problem was first studied by Kondo in 1964 in order to explain the peculiar minima
in resistivity in some materials. The coupling between the local spin impurities S i at Ri and
the quasiparticle spin s has the exchange form
X
X


b b
VbK =
VbK i = J
S
(6.70)
i s(r)(r Ri )
i


X
1
1
=J
Sbiz sbz (r) + Sbi+ sb (r) + Sbi sb+ (r) (r Ri )
2
2
i
i
0
J~ X h bz
ck0 + Si b
ck b
ck0 ei(kk )Ri .
ck0 ) + Si+ b
ck b
ck0 b
ck b
=
ck b
Si (b
2 0

(6.71)
(6.72)

k,k ,i

Here, it becomes important that spin flip processes, which change the spin state of the impurity
and that of the scattered electron, are enabled. The results for the scattering rate are presented
here without derivation,




D
0
2
W (k, k ) J S(S + 1) 1 + 2JN (F ) ln
,
(6.73)
|k F |
where D is the bandwidth and we have assumed that JN (F )  1. The relaxation time is found
to be




J 2 S(S + 1)
D
1

N () 1 + 2JN (F ) ln

.
(6.74)
(k )
~
|k F |
Note that W (k, k0 ) does not depend on angle, meaning that the process is described by s-wave
scattering. The energy dependence is singular at the Fermi energy, indicating that we are not
dealing with simple resonant potential scattering, but with a much more subtle many-particle
effect involving the electrons very near the Fermi surface. The fact that the local spins S i
can flip, makes the scattering center dynamical, because the scatterer is constantly changing.
The scattering process of an electron is influenced by previous scattering events, leading to the
singularity at F . This cannot be described within the first Born approximation, but requires
at least the second approximation or the full solution.10 As mentioned before, the resonant
10

We refer to J.M. Ziman, Principles of the Theory of Solids, and A.C. Hewson, The Kondo Problem to Heavy
Fermions for more details.

121

behavior induces a strong temperature dependence of the conductivity. Indeed,

Z
e2 k 3
1
(T ) = 2 F
d
()
2
6 m
4kB T cosh ( F )/2kB T )
Z
J 2 S(S + 1)[1 2JN (F ) ln(D/
)]
e2 n
d


.
2
8mkB T
cosh (
/2kB T )

(6.75)
(6.76)

A simple substitution in the integral leads to

(T )



e2 n 2
D

.
J S(S + 1) 1 2JN (F ) ln
2m
kB T

(6.77)

Usual contributions to the resistance, like electron-phonon scattering discussed below, typically
decrease with temperature. The contribution (6.77) to the conductivity is strongly increasing, inducing a minimum in the resistance, when we crossover from the decreasing behavior
at high temperatures to the low-temperature increase of (T ). At even lower temperatures,
the conductivity would decrease and eventually even turn negative which is an artifact of our
approximation. In reality, the conductivity saturates at a finite value when the temperature is
lowered below a characteristic Kondo temperature TK ,
kB TK = De1/JN (F ) ,

(6.78)

a characteristic energy scale of this system. The real behavior of the conductivity at temperatures
T  TK is not accessible by simple perturbation theory. This regime, known as the Kondo
problem, represents one of the most interesting correlation phenomena of many-particle physics.

6.4

Electron-phonon interaction

Even in perfect metals, the conductivity becomes non-zero at finite temperature. The thermally
induced distortions of the lattice, phonons, act as fluctuating scattering centers. In the language
of electron-phonon interaction, electrons are scattered via absorption and emission of phonons,
which induce local fluctuations in volume (cf. Chapter 3). The corresponding coupling term
was given in equation (3.143) and simplifies with the definition (3.130) to
s
X
~
Hint = 2i
Vq
cks .
(6.79)
|q|(bbq bbq )b
ck+q,s b
20 q
k,q,s

The interaction is formally similar to the coupling between electrons and photons. The dominant
processes consist of single-phonon processes, i.e., the absorption or emission of one phonon. Energy and momentum are conserved, such that, for the scattering of an electron from momentum
k to k0 due to the emission of a phonon with momentum q, we have
k = k0 + q + G,
k = ~q + k0 ,

(6.80)
(6.81)

Here, q = cs q is the phonon spectrum, while the reciprocal lattice vector G allows for scattering11 in nearby Brillouin zones. By this, the phase space available for scattering is strongly
reduced, especially near the Fermi energy. Note that ~q ~D  F . In order to calculate
11

This so-called Umklapp phenomenon will be discussed in some more detail later in this chapter.

122

the scattering rates, the matrix elements12 of the available processes,

q

0
b
b
hk + q; Nq0 |(bq bq )b
ck+q,s b
cks |k; Nq0 i = hk + q|b
ck+q,s b
cks |ki
Nq0 0 N

0
q 0 ,Nq 0 1

q,q0

q
Nq0 0 + 1 N

0
q 0 ,Nq 0 +1


q,q0 ,
(6.82)

need to be calculated. From Fermis golden rule we obtain


 

f
2 X
2
|g(q)| f (k)(1 f (k + q))(Nq + 1)
=
t coll
~ q

f (k + q)(1 f (k))Nq (k+q k + ~q )

f (k + q)(1 f (k))(Nq + 1)


f (k)(1 f (k + q))Nq (k+q k ~q ) ,

(6.83)

q
where g(q) = Vq |q| 2~/0 q . Each of these four terms describes one of the single phonon
scattering processes depicted in Figure 6.4.
k+q

q
k

k+q

q
k+q

q
k+q

q
k

Figure 6.4: The four single-phonon electron-phonon scattering processes.

The collision integral leads to a complicated integro-differential equation, whose solution is tedious and would involve the solution of the non-equilibrium phonon-problem as well. Instead of
a full rigorous calculation including the non-equilibrium redistribution of phonons, we will consider the behavior in various temperature regimes by an approximate treatment of the phonons.
The characteristic temperature of phonons, the Debye temperature D  TF , is much smaller
than the Fermi temperature. Hence, the phonon energy is virtually unimportant for the energy conservation, k0 =k+q k . Therefore we are allowed to impose momentum conservation
k+q = k and consider the lattice distortion as being essentially static, in the sense of an
adiabatic Born-Oppenheimer approximation. The approximate collision integral then reads
 
f
2 X
=
|g(q)|2 2N (q )[f (k + q) f (k)](k+q k ),
(6.84)
t coll
~ q
where we assume the occupation of phonon states according to the equilibrium distribution for
bosons,
N (q ) =

1
e~q /kB T

(6.85)

This approximation includes all important aspects of the electron-phonon scattering we need to
derive the temperature dependence of (T ).
12

In analogy to the discussion on electromagnetic radiation, the phenomenon of spontaneous phonon emission
due to zero-point fluctuations appears. It is formally visible in the additional +1 in the factors (Nq ) + 1.

123

In analogy to previous approaches, we obtain with relaxation-time Ansatz

Z
d3 q
1
2
=
~ N (q )(1 cos )(k+q k ),
(k )
~ N (F )
(2)3 q

(6.86)

where |k| = |k + q| = kF , meaning that only the electrons in a thin shell close to the Fermi
surface are relevant. Furthermore, we parametrized g(q) according to
|g(q)|2 =

~ ,
2N (F ) q

(6.87)

where is a dimensionless electron-phonon coupling constant. In usual metals < 1. As in the

case of defect scattering, the relaxation time depends only weakly on the electron energy. But,
unlike previously, the direct temperature dependence enters via the dependence on temperature
of the phonon occupation N (q ).

k+q
q

kF

Figure 6.5: The geometry of electron-phonon scattering.

In order to perform the integration in equation (6.86), we have to re-express (k+q k ) by
writing
 2



~
m
q
(k+q k ) =
(q 2 2kF q cos ) = 2

cos ,
(6.88)
2m
~ kF q
2kF
where is defined in Figure 6.5. From there, we also see that 2 + = , and thus, find the
relation
1 cos = 1 + cos(2) = 2 cos2 ().

(6.89)

Obviously, we have to integrate q over the range [0, 2kF ] on the right-hand side of equation
(6.86), which can be reformulated to
1

m
=
2
(F , T )
N (F ) ~ kF
=

2kF
Z

mcs

4N (F ) ~2 kF3

mcs kF2

=
4N (F ) ~2

Z/2
dq qq N (q )
d sin cos2 ()

0
2kF
Z
0

q 4 dq
e~cs q/kB T 1

T
D

5 2ZD /T
0

y 4 dy
,
ey 1

124

q
cos
2kF

(6.90)

(6.91)

(6.92)

where we have approximated the Debye temperature by kB D ~cs kF . We notice the two
distinct characteristic temperature regimes,



kB D
T 5

6(5)
, T  D ,

~
D
1
(6.93)
=



k
T

,
T  D .
B D
~
D
The prefactors depend on the details of the approximation, whereas the qualitative temperature
dependence does not. We finally obtain the conductivity and resistivity from equation (6.29),
e2 n
(T ),
m
m 1
= 2
,
e n (T )

(6.94)
(6.95)

where we used the weak energy dependence of ( F , T ). With this, we obtain the well-known
Bloch-Gr
uneisen form
5
T , T  D ,
(6.96)
(T )

T, T  D .
At high temperatures, is determined by the occupation of phonon states
N (q )

kB T
~q

(6.97)

which change the scattering strength (amplitude) of the lattice modulation linear in T . At low
temperature only the lowest phonon states are occupied ~q < kB T yield q < kB T /~cs . Thus, at
low temperatures only long-wave length modulations of the lattice generate a scattering potentials which deflects electrons only slightly from their trajectories (forward scattering dominates).
This represents a restriction of the scattering phase space becoming ever smaller with decreasing
temperature.

6.5

Electron-electron scattering

In Chapter 5 we have learned, that, taking a short-ranged electron-electron interaction into

account, scattering rate for electrons decreases strongly close to the Fermi surface. The basic
reason for this lies in the constraint of the scattering phase space imposed by the Pauli principle.
The lifetime, which we identify with the relaxation time here, has the form

2
1
1

=
1
(6.98)
()
e F
where energy and momentum conservation is taken into account and e is a constant of unit time.
We may now calculate the mean relaxation time as introduce in Sect.6.2.2. To regularize the
integrals we take here also impurity scattering into account by adding a constant to the scattering
rate in Eq.(6.98) and obtain for the resistivity13


kB T 2
2 m
(T ) = 0 +
(6.102)
3 ne2 e
F
Electron-electron scattering introduces a quadratic temperature dependence to resistivity. This
is a key property of a Fermi liquid and is often considered an identifying criterion.
13

Calculation of mean scattering time : We use Matthiessens rule to add the scattering rates of electronelectron and impurity scattering to the form

2
1
1

1
=
1 +
.
(6.99)
()
e F
0

125

Umklapp process
An important point, kept quiet so far, requires some explanation. One could argue, that the
momentum of the Fermi liquid is conserved upon the collision of two electrons. With this
argument, it is not quite clear what causes a finite resistance. This argument, however, is based
on translational invariance and ignores the existence of the underlying lattice. In the sense that
the kinematics (momentum conservation) is also satisfied for electrons being scattered from the
Fermi surface of one Brillouin zone to the one of another Brillouin zone, while incorporating a
reciprocal lattice vector. The equation of momentum conservation,
k = k0 + G,

(6.103)

where G is a reciprocal vector of the lattice, allows for scattering to other Brillouin zones
(G 6= 0). By this, the momentum is transferred to a static deformation of the lattice. Such processes are termed Umklapp processes and play an important role in electron-phonon scattering
as well (see equation (6.80)).

ordinary$
sca4ering$

~k 0

Umklapp$$
sca4ering$

~k 00
~k 0

~k 00

~
q

~
q

~k + ~
q

1st$Brillouin$zone$

~k

Figure 6.6: Considering energy and momentum conservation in two-electron collision there are
ordinary scattering processes (k, k0 ) (k + q, k0 q) which does not lead to a momentum
relaxation, and there are Umklapp scattering processes (k, k00 ) (k + q, k00 q) leading to a
neighbouring Brillouin zone, here visualized by an extended Brillouin zone scheme.

Restricting to the small temperature limit, we determine by



Z
Z +
Z +
(E)
1
02 E 2
= d ()f0 / =
dE

dE

0
e 2F
4kB T cosh2 (E/2kB T )
4kB T cosh2 (E/2kB T )

= 0

2 02
3 e

kB T
F

2
(6.100)

where E =  F . The resistivity is given by

m
m
(T ) =

ne2 (T )
ne2 0

2 0
1+
3 e

kB T
F

2 )

and with 0 = m/ne2 0 as the residual resistivity leads to (T ) in Eq.(6.102).

126

(6.101)

6.6

Matthiessens rule and the Ioffe-Regel limit

Matthiessens rule states, that the scattering rates of different scattering processes can simply
be added, leading to
W (k, k0 ) = W1 (k, k0 ) + W2 (k, k0 ),

(6.104)

or, expressed in the relaxation time approximation,

1
1
1
=
+ ,

1 2

(6.105)

and
m
m
= 2 = 2
ne
ne

1
1
+
1 2


= 1 + 2 .

(6.106)

This rule is not a theorem and corresponds effectively to a serial coupling of resistors in a classical circuit. It is applicable, if the different scattering processes are independent. Actually,
the assumption that the impurity scattering rate depends linearly on the impurity density nimp
is already an application of Matthiessens rule. Mutual influences of impurities, e.g., through
interference effects due to the coherent scattering of an electron at different impurities, would invalidate this simplification. An example where Matthiessens rule is violated is a one-dimensional
system, where a single scatterer i induces a finite resistance Ri . Two serial scatterers then lead
to a total resistance
R = R1 + R2 +

2e2
R R R1 + R2 .
h 1 2

(6.107)

The reason is, that in one-dimensional systems, the interference of backscattered waves is unavoidable and no impurity can be treated as isolated. Furthermore, every particle traversing the
whole system has to pass all scatterers. The more general Matthiessens rule,
1 + 2 ,

(6.108)

is still valid. Another source of deviation from Matthiessens rule arises, if the relaxation time
depends on k, since then the averaging is not the same for all scattering processes. The electronphonon coupling can be modified by the scattering on impurities, most importantly in the
presence of anisotropic Fermi surfaces. For the analysis of resistance data of simple metals, we
often assume the validity of Matthiessens rule. A typical example is the resistance minimum
explained by Kondo, where
(T ) = 0 + ep (T ) + K (T ) + ee (T )
5

(6.109)
2

= 0 + T + (1 + 2JN (F ) ln(D/kB T )) + T ,

(6.110)

where , , and are numerical constants. Upon decreasing temperature, the Kondo term is
increasing, whereas the electron-phonon and electron-electron contributions decrease. Consequently, there is a minimum.
We now turn to the discussion of resistivity in the high-temperature limit. Believing the previous
considerations entirely, the electrical resistivity would grow indefinitely with temperature. In
most cases, however, the resistivity will saturate at a finite limiting value. We can understand
this from simple considerations writing the mean free path ` = vF (F ) as the mean distance an
electron travels freely between two collisions. The lattice constant a is a natural lower boundary
to ` in the crystal lattice. Furthermore, we assumed so far that scattering occurs between two
states with sharp momenta k and k0 . If the de Broglie wavelength becomes comparable to the
mean free path, the framework becomes unfounded and kF1 would become a boundary for `. In
127

most systems a and kF1 are comparable lengths. Empirically, the resistivity is described via the
formula
1
1
1
=
+
,
(T )
BT (T ) max

(6.111)

corresponding to the parallel addition of two resistivities; on one hand, BT (T ), which we have
investigated using the Boltzmann transport theory, and on the other hand the limiting value
max . This is in clear disagreement to Matthiessens rule, which is to be expected, since for
kF ` 1, complex interference effects will arise. The saturated resistivity max can be estimated
from the Jellium model,
m
3 2 m
h 3
=
= 2 2
2
2
3
e n (F )
e kF (F )
e 2kF `
h 3
,
2
e 2kF

max =

(6.112)
(6.113)

where we used `1 kF . For a typical value kF 108 cm1 of the Fermi wave vector, we find
max 1mcm, which is called the Ioffe-Regel14 limit. Establishing a quantitative estimate of
max for a given material is often difficult. There are even materials whose resistivity surpasses
the Ioffe-Regel limit.

6.7

General transport coefficients

Simultaneously with charge, electrons will also transport energy, i.e., heat and entropy. This is
why charge and heat transport are naturally interconnected. In the following, we generalize the
transport theory set up above to include this interplay.

6.7.1

Generalized Boltzmann equation

We consider a metal with weakly space-dependent temperature T (r) and chemical potential
(r). The distribution function then reads
f (k; r) = f (k; r) f0 (k, T (r), (r)),

(6.114)

where
f0 (k, T (r), (r)) =

1
e(k (r))/kB T (r) + 1

(6.115)

In addition, we require the charge density to remain constant in space, i.e.,

Z
d3 k
f (k; r) = n0
(2)3

(6.116)

for all r. In this section, we introduce the electrochemical potential (r) = e(r) + (r)
generating the general force field E = (e + ), where (r) denotes the electrostatic potential which produces the electric field E = . With this, the Boltzmann equation for the
14

The saturated resistivity max 1mcm= 1000cm should be compared to the room-temperature resistivity
of good conductors,
metal
[cm]

Cu
1.7

Au
2.2

Ag
1.6

Pt
10.5

which are all well below max .

128

Al
2.7

Sn
11

Na
4.6

Fe
9.8

Ni
7

Pb
21

stationary situation reads



f
t


coll

= v k r f + k k f


k
f
=
r T E .
v
k k
T

a)

(6.117)
(6.118)

b)

ky

ky

fk
kx

f k
kx

Figure 6.7: Schematic view of the distribution functions f (k) on a slice cut through the k-space
(kz = 0) with a circular Fermi surface for two situations. On the left panel (a), for an applied
electric field along the negative x-direction, on the right panel (b) for a temperature gradient in
x-direction.
In the relaxation time approximation for the collision integral, we obtain the solution


f0
k
f (k) =
(k )v k E
r T ,
k
T
which allows us to calculate the charge and heat currents,
Z
d3 k
J e = 2e
v f (k),
(2)3 k
Z
d3 k
( )v k f (k),
Jq = 2
(2)3 k

(6.119)

(6.120)
(6.121)

respectively. Inserting the solution (6.119) into the two definitions above yields
(0) E e K
(1) (T ) ,
J e = eK
T
(1) E + 1 K
(2) (T ) ,
J q = K
T
(n) (n N ) are defined as
where should be understood as r and the tensors K
0
Z
Z
vF vF 2
f
1
(n)
K = 3 d 0 ()( )n dk
k .
4

~|v F | F

(6.122)
(6.123)

(6.124)

For an isotropic system these transport coefficients are no longer tensors but represented by
scalars,
Z
n0
f
(n)
K =
d 0 ()( )n
(6.125)
m


129

In the case T  TF we can calculate the coefficients,15

n0 (F ) (F )
m

2


2
(0)
(1)
(kB T )
K ()
,
K (F ) =
3

=F

K (0) (F ) =

K (2) (F ) =

2
(k T )2 K (0) (F ).
3 B

(6.128)
(6.129)
(6.130)

We measure the electrical resistivity assuming thermal equilibrium, T = 0 for all r. With
this, we find the expression
= e2 K (0) .

(6.131)

In order to determine the thermal conductivity

relating the heat current J q to T when no
external electric field E is applied, we set J e = 0 as for an open circuit. Then, the equations
(6.122) and (6.123) reveal the appearance of an general force field
E=

K (1)
T.
T K (0)

(6.132)

Thus, the heat current is given by

1
Jq =
T

(2)

K (1)2
(0)
K

!
T =
T.

(6.133)

In simple metals, the second term in (6.133) is often negligible as compared to the first one and
we obtain in this case

1 (2) 2 kB2 (0) 2 kB2

K =
TK =
T
,
T
3
3 e2

(6.134)

which is the well-known Wiedemann-Franz law. Note, that we can write the thermal conductivity
in the form

,
e2 N (F )

(6.135)

where C = 2 N (F )kB2 T /3 denotes the electronic specific heat.

6.7.2

Thermoelectric effect

Equation (6.132) shows, that a temperature gradient induces an electric field. For a simple,
isotropic system, this relation reduces to
E = QT =

K (1)
T,
T K (0)

(6.136)

15

If a function g() depends only weakly on  in the vicinity of F , we can use the Sommerfeld (Taylor) expansion
to derive a general approximation for following integrals

Z
2

f0
2
2 g() 
= g(F ) +
(kB T )
+ ...
(6.126)
dg()

6
2 =F
and
Z

dg()( F )


g() 
f0
2
=
(kB T )2
+ ...,

3
 =F

in the limit T 0. We used that F in that asymptotic case.

130

(6.127)

with the Seebeck coefficient


2 kB2 T 0 () 
.
Q=
3 e () =F

(6.137)

This is the so-called Mott formula which looses its validity at high-temperatures or very anisotropic
scattering. Using () = n()e2 ()/m, we investigate 0 (),
0 () =

0 ()
n0 ()
0 ()
N ()
() +
() =
() +
(),
()
n()
()
n()

(6.138)

and obtain an additional contribution to Q, if the relaxation time depends strongly on energy.
This is most prominent in collision processes in which resonant scattering is involved (e.g., the
Kondo effect). In the opposite situation, namely, when the first term is irrelevant, the Seebeck
coefficient
Q=

2 kB2 T N (F )
S
=
3 e n(F )
ne

(6.139)

is simply reduced to the entropy per electron. For simple metals such as the alkali metals we
may estimate the low-temperature values using equation (6.139)
Q=

2T
2 kB T
2 kB
=
2 eF
2 eTF

(6.140)

which for TF (N a, K) 3 104 K leads to Q = 14nV K 1 T [K]. A comparison with

experiments in Fig.6.8 shows that the order of magnitude works reasonably well for Na and K.
However, for Li and Cs even the sign is different. Differences occur through phonon effects, such
as the so-called phonon drag which we have neglected here.
In the following, we consider two different types of thermoelectric effects.
Seebeck effect
The first is the Seebeck effect, where a thermoelectric voltage appears in a bi-metallic system
(cf. Figure 6.9). With equation (6.136), a temperature gradient across metal B induces an
electromotive force16
Z
UEMF = dl E
(6.141)
ZT1
= QA

dl T + QB

T0

ZT2
T1

= (QB QA )(T2 T1 ).

dl T + QA

ZT0

dl T

(6.142)

T2

(6.143)

The resulting voltage Vtherm = UEMF appears between the two ends of a second metal A, whose
contacts are kept at the same temperature T0 . Here, a bi-metallic configuration was chosen to
reveal voltage differences across the contacts which are absent in a single metal.
Peltier effect
The second phenomenon, termed Peltier effect, emerges in a system kept at the same temperature everywhere. Here, an electric current Je between the two contacts of the metal A (see
Figure 6.9) induces a heat current in the bi-metallic system, such that heat is transferred from
16

The term electromotive force, first introduced by Alessandro Volta, is misleading in the sense, that it measures
a voltage instead of a force.

131

Cs

100

Li

Q [nV/K]

50

Na

50
0

T [K]

Figure 6.8: Seebeck coefficient for the Alkali metals Li, Na, K, and Cs at low temperatures. The
dashed line represents the estimate for Na and K following Eq.(6.140). (adapted from D.K.C.
MacDonald, Thermoelectricity: an introduction to the principles, Dover (2006).)

one reservoir (top) to another (bottom). This follows from the equations (6.122) and (6.123) by
assuming T = 0, where

2 (0)
Je = e K E
(6.144)

(1)
Jq = K E
implies
Jq =

K (1)
J = Je = QT Je .
K (0) e

(6.145)

The coupling = T Q between Jq and Je is called Peltier coefficient. According to Figure 6.9,
a contribution to the heat current is to be expected from both metals A and B,
Jq = (A B )Je = T0 (QA QB )Je .

(6.146)

This means, that the heat transfer between reservoirs can be controlled by electrical current.
It has to be emphasized here that the bi-metal design of the devices in Fig. 6.9 serves the
observation of the two effects which both represent bulk effects of the two metals A and B. By
no means, it should be mistaken as an effect originating from the inter-metal contacts.

132

Jq"

T2
Je"

A"
T0
T0

B"

Je"

B"

V"

T0
T0

A"

Je"

T1

A"

T0

A"

Je"

Je"
Jq"

Figure 6.9: Schematic design of bi-metallic device to measure the Seebeck and Peltier effect: On
the left panel (a) a representation of Seebeck effect is given, where the symbol E is used instead
of E. On the right panel (b) the Peltier effect is represented. In our analysis, both systems were
effectively one-dimensional.

6.8

Anderson localization in one-dimensional systems

Transport in one spatial dimension is very special, since there are only two different directions
to go: forward and backward. We introduce the transfer matrix formalism and use it to express
the conductivity through the Landauer formula. We will then investigate the effects of multiple
scattering at different obstacles, leading to the so-called Anderson localization, which turns a
metal into an insulator.

6.8.1

Landauer Formula for a single impurity

The transmission and reflection at an arbitrary potential with finite support17 in one dimension
can be described by a transfer matrix T .
V
a2+

a1+
I1

I2

T
a1

a2

x
Figure 6.10: Transfer matrices are sufficient to describe potential scattering in one dimension.
In this situation, a suitable choice for a basis of the electron states is the set of plane waves
{eikx } (cf. Figure 6.10) moving in the positive (negative) x-direction with wave vector +k
(k). Only plane waves with the same |k| on the left (I1 ) and right (I2 ) side of the scatterer
are interconnected. Therefore, we write

17

1 (x) = a1+ eikx + a1 eikx ,

(6.147)

2 (x) = a2+ eikx + a2 eikx ,

(6.148)

The support (dt. Tr

ager) of a function is the set of all points, where the function takes non zero values.

133

where 1 (2 ) is defined in the area I1 (I2 ). The vectors ai = (ai+ , ai ) i {1, 2} are connected
via the linear relation,


T11 T12

a2 = T a1 =
a1 ,
(6.149)
T21 T22
with the 2 2 transfer matrix T. The conservation of current (J1 = J2 ) requires that T is
unimodular, i.e., det T = 1. Here,


i~ d (x)
d(x)

J=
,
(6.150)
(x) (x)
2m
dx
dx

such that, for a plane wave (x) = (1/ L)eikx in a system of length L, the current results in
J = v/L

(6.151)

with the velocity v defined as v = ~k/m. Time reversal symmetry implies that, simultaneously
with (x), the complex conjugate (x) is a solution of the stationary Schrodinger equation.
and T

From this, we find T11 = T22

12 = T21 , such that


T11 T12
T =
.
(6.152)

T12
T11
It is easily shown that a shift of the scattering potential by a distance x0 changes the coefficient
T12 of T by a phase factor ei2kx0 . Meanwhile, the coefficient T11 remains unchanged.
With the Ansatz for a right moving incoming wave ( eikx ), producing a reflected ( reikx )
and transmitted ( teikx ) part, the wave functions on both sides of the scatterer read


1
1 (x) = eikx + reikx ,
L


1
2 (x) = teikx .
L
The coefficients of T can be determined explicitly in this situation, resulting in
!
r
1

t
t
T =
.
1
rt
t

(6.153)
(6.154)

(6.155)

Here, the conservation of currents imposes the condition 1 = |r|2 +|t|2 . Furthermore, we can find
a relation between the parameters (r, t) of the potential barrier and the electric resistivity. For
this, we notice that the incoming current density J0 is split into a reflected Jr and transmitted
Jt part, all given by
1
J0 = ve = n0 ve,
L
|r|2
Jr =
ve = nr ve,
L
|t|2
Jt =
ve = nt ve,
L

(6.156)
(6.157)
(6.158)

with the velocity v = ~k/m, the electron charge e, and the particle densities n0 , nr , and nt
corresponding to the incoming, reflected and transmitted particles respectively. The electron
density on the two sides of the barrier is given by
n1 = n0 + nr =
n2 = nt =

|t|2
.
L
134

1 + |r|2
,
L

(6.159)
(6.160)

From this consideration, a density difference n = n1 n2 = (1 + |r|2 |t|2 )/L = 2|r|2 /L results
between the left and the right side of the scatterer. The resistance R of the barrier is defined
by the ratio between the voltage drop over the resistor V and the transmitted current Jt , i.e.
R=

V
Jt

(6.161)

Consequently, a relation between V and the electron density n remains to be established to

determine R. The connection is easily found via the existing energy difference E = eV
between the two sides of the resistor, such that the expression
dn
E
dE
dn
=
(eV )
dE

n =

(6.162)
(6.163)

dn
produces the wished relation. Here, dE
dE is the number of states per unit length in the energy
interval [E, E + dE] and we find




Z
1X
~2 k 2
dk
~2 k 2
1
dn
=
E
=2
E
=
.
(6.164)
dE
L
2m
2
2m
~v(E)
k,s

The resistance R is finally obtained from the equations (6.161), (6.162), and (6.164), leading to
R=

h |r|2
,
e2 |t|2

(6.165)

The Klitzing constant RK = h/e2 25.8k is a resistance quantum named after the discoverer
of the Quantum Hall Effect. The result (6.165) is the famous Landauer formula, which is
valid for all one-dimensional systems and whose application often extends to the description of
mesoscopic systems and quantum wires.

6.8.2

Scattering at two impurities

We consider now two spatially separated scattering potentials, represented by T1 and T2 each
determined by r1 , t1 and r2 , t2 respectively.

T2

T1

Figure 6.11: Two spatially separated scattering potentials with transmission matrices T1 and T2
respectively.
The particles are multiply scattered at these potentials in a unknown manner, but the global
result can again be expressed via a simple transfer matrix T = T1 T2 , given by the matrix
multiplication of each transfer matrix. All previously found properties remain valid for the new
matrix T, given by

!
r1 r2
r2
r1
1
1
r

t + t t
t t t

t
t
t
t
1 2
1 2
1 2
=
.
(6.166)
T = 1 2
r1 r2
r
1
r1
r2
1

t1 t t1 t2
t
t
t1 t2 + t1 t
2

135

For the ratio between reflection and transmission probability we find

|r|2
1
= 2 1
|t|2
|t|

(6.167)

1
=
|t1 |2 |t2 |2
1
=
2
|t1 | |t2 |2



t 2

r
r
1
2
2
1 +
 1

t2 


r1 r2 t2 r1 r2 t2
2
2
1 + |r1 | |r2 | +
1.
+
t2
t2

(6.168)
(6.169)

Assuming a (random) distance d = x2 x1 between the two potential barriers, we may average
over this distance. Note, that for x1 = 0, we find r2 e2ikd , while r1 , t1 , and t2 are independent
on d. Consequently, all terms containing an odd power of r2 or r2 vanish after averaging over
d. The remainders of equation (6.169) can be collected to



|r|2 
1
2
2
1
+
|r
|
|r
|
1
(6.170)
=
1
2
|t|2 avg |t1 |2 |t2 |2
=

|r1 |2 |r2 |2
|r1 |2 |r2 |2
+
+
2
.
|t1 |2
|t2 |2
|t1 |2 |t2 |2

(6.171)

Even though two scattering potentials are added in series, an additional non-linear combination
emerges beside the sum of the two ratios |ri |2 /|ti |2 . It results from the Landauer formula applied
to two scatterers, that resistances do not add linearly to the total resistance. Adding R1 and
R2 serially, the total resistance is not given by R = R1 + R2 , but by
R = R1 + R2 + 2

R1 R2
> R1 + R2 ,
RK

(6.172)

with RK = h/e2 This result is a consequence of the unavoidable multiple scattering in one
dimensions. This effect is particularly prominent if Ri  h/e2 for i {1, 2}, where resistances
are then multiplied instead of summed.

6.8.3

Anderson localization

Let us consider a system with many arbitrarily distributed scatterers, and let be a mean
resistance per unit length. R(`) shall be the resistance between points 0 and `. The change in
resistance by advancing an infinitesimal ` is found from equation (6.172), resulting in
R(`)
d`,
RK

(6.173)

dR
,
1 + 2R/RK

(6.174)

dR = d` + 2
which yields
R(`)
Z

Z`
d` =
0

and thus,
` =

h
ln
1
+
2R(`)/R
.
K
2e2

(6.175)

Finally, solving this equation for R(`), we find

R(`) =


RK  2`/RK
e
1 .
2
136

(6.176)

Obviously, R grows almost exponentially fast for increasing `. This means, that for large `, the
system is an insulator for arbitrarily small but finite > 0. The reason for this is that, in one
dimension, all states are bound states in the presence of disorder. This phenomenon is called
Anderson localization. Even though all states are localized, the energy spectrum is continuous,
as infinitely many bound states with different energy exist. The mean localization length of
individual states, related to the mean extension of a wave function, is found from equation (6.176)
to be = /RK . The transmission amplitude is reduced18 on this length scale, since |t| 2e`/
for `  . In one dimension, there is no linearly increasing electric resistance, R(`) `.
For non-interacting particles, only two extreme situations are possible. Either, the potential
is perfectly periodic and the states correspond to Bloch waves. Then, coherent constructive
interference produces extended states19 that propagate freely throughout the system, resulting
in a perfect conductor without resistance. On the other hand, if the scattering potential is
disordered, all states are strictly localized. In this case, there is no propagation and the system
is an insulator. In three-dimensional systems, the effects of multiple scattering are far less drastic
and the Ohmic law is applicable. Localization effects in two dimensions is a very subtle topic
and part of todays research in solid state physics.

18

For an expanded discussion of this topic, the article [P.W. Anderson, D.J. Thouless, E. Abrahams, and D.S.
Fisher, New method for a scaling theory of localization, Physical Review B 22, 3519 (1980)] is recommended.
19
We have also seen in the context of chiral edge states in the Quantum Hall state, that perfect conductance in
a one-dimensional channel is obtained if there is no backscattering due to the lack of states moving in the opposite
direction. In chiral states, particle move only in one direction.

137

Chapter 7

Magnetism in metals
Magnetic ordering in metals can be viewed as an instability of the Fermi liquid state. We introduce this new phase of metals through the description of the Stoner ferromagnetism. The
discussion of antiferromagnetism and spin density wave phases will be only brief in this chapter.
In Stoner ferromagnets the magnetic moment is provided by the spin of itinerant electrons. Magnetism due to localized magnetic moments will be considered in the context of Mott insulators
which are subject of the next chapter.
Well-known examples of elemental ferromagnetic metals are iron (Fe), cobalt (Co) and nickel
(Ni) belonging to the 3d transition metals, where the 3d-orbital character is dominant for the
conduction electrons at the Fermi energy. These orbitals are rather tightly bound to the atomic
cores such that the electron mobility is reduced, enhancing the effect of interaction which is
essential for the formation of a magnetic state. Other forms of magnetism, such as antiferromagnetism and the spin density wave state are found in the 3d transition metals Cr and Mn.
Note, 4d and 5d transition metals within the same columns of the periodic system are not
magnetic. Their d-orbitals are more extended, leading to a higher mobility of the electrons,
such that the mutual interaction is insufficient to trigger magnetism. It is, however, possible
to find ferromagnetism in ZrZn2 where zink (Zn) may act as a spacer reducing the mobility
of the 4d-electrons of zirconium (Zr). The 4d-elements Pd and Rh and the 5d-element Pt are,
however, nearly ferromagnetic. Going further in the periodic table, the 4f -orbitals appearing
in the lanthanides are nearly localized and can lead to ferromagnetism, as illustrated by the
elements going from Gd through Tm in the periodic system.
Magnetism appears through a phase transition, meaning that the metal is non-magnetic at
temperatures above a critical temperature Tc , the Curie-temperature (cf. Table 7.1). In many
cases, magnetism appears at Tc as a continuous, second order phase transition involving the
spontaneous violation of symmetry. This transition is lacking latent heat (no discontinuity in
entropy and volume) but instead features a discontinuity in the specific heat.
element
Fe
Co
Ni
ZrZn2
Pd
HfZn2

Tc (K)
1043
1388
627
22

type
ferromagnet (3d)
ferromagnet (3d)
ferromagnet (3d)
ferromagnet
paramagnet
paramagnet

element
Gd
Dy
Cr
-Mn
Pt

Tc (K)
293
85
312
100

type
ferromagnet (4f)
ferromagnet (4f)
spin density wave (3d)
antiferromagnet
paramagnet

Table 7.1: Selection of (ferro)magnetic materials with their respective form of magnetism and
the critical temperature Tc .

138

7.1

Stoner instability

In the following section, we study the emergence of the metallic ferromagnetism originating from
the Stoner mechanism. In close analogy to the first Hunds rule, the exchange interaction among
the electrons plays a crucial role here. The alignment of the electronic spins in a favored direction
allows the system to reduce the energy contribution due to Coulomb repulsion. According
to Landaus theory of Fermi liquids, the interaction between electrons renormalizes the spin
susceptibility 0 to
=

m 0
,
m 1 + F0a

(7.1)

which obviously diverges for F0a 1 and leads to a ferromagnetic instability of the Fermi liquid.
Using Eq.(5.110), F0a U N (F )/2 provides a critical value for the interaction Uc = 2/N (F )
such that F0a = 1 and diverges. We will see below that this corresponds to a value we will
derive also by a mean field theory.

7.1.1

Stoner model within the mean field approximation

Consider the following model for conduction electrons with a repulsive contact interaction,
Z
X

H=
k b
cks b
cks + U d3 r d3 r0 b (r)(r r 0 )b
(r 0 ),
(7.2)
k,s

b (r)
b (r) and the field operator
b (
b ) follows
where we use the electron density bs (r) =
s
s
s
s
from the definition (3.28) [(3.29)]. The contact interaction is an approximation of the screened
Coulomb interaction. Due to the Pauli exclusion principle, the contact interaction is only active
between electrons with opposite spins. This is a consequence of the exchange hole in the twoparticle correlation between electrons of identical spin (cf. Figure 3.1). We obtain a useful
insight into mechanisms leading to ferromagnetism by means of a mean field approximation1 .
We rewrite,
bs (r) = ns + [b
s (r) ns ],

(7.3)

ns = hb
s (r)i,

(7.4)

where

and h i represents the thermal average. We stipulate that the deviation from the mean value ns
is small in the sense that
h[b
(r) n ][b
(r) n ]i  n n .
Inserting equation (7.3) into the Hamiltonian (7.2), we obtain
Z
X

Hmf =
k b
cks b
cks + U d3 r [b
(r)n + b (r)n n n ]

(7.5)

(7.6)

k,s

X
k,s

(k + U ns) b
cks b
cks U n n ,

(7.7)

the mean field Hamiltonian, describing electrons which move in the uniform background of
electrons of opposite spin coupling via the spin dependent exchange interaction (
s denotes to
opposite spin s). Fluctuations of the form h[b
(r) n ][b
(r) n ]i are neglected here. The
1

Note that the following mean field calculation is equivalent to a variational approach using simple many-body
wavefunction (Slater determinant) with different concentrations of up and down spins.

139

advantage of this approximation is, that the many-body problem is now reduced to an effective
one-particle problem, where only the mean electron interaction is taken into account. This
is equivalent to a generalized Hartree-Fock approximation and enables us to calculate certain
expectation values, such as the density of one spin species, e.g.;
1X
1X
n =
hb
ck b
ck i =
f (k + U n )
(7.8)

k
k
Z
1X
= d
( k U n )f ()
(7.9)

k
Z
1
= d N ( U n )f (),
(7.10)
2
where f () is the Fermi-Dirac distribution function. An analogous result is found for the opposite
spin direction. These mean densities are determined self-consistently, namely such that the
insertion of ns in into the mean field Hamiltonian (7.7) provides the correct output according
to the expectation values given in equation (7.8). Furthermore, the constraint that the total
number of electrons is conserved, must be implemented. The real magnetization M = B m is
proportional to m which is defined via
 n + sm
1
ns = (n + n ) + s(n n ) = 0
,
(7.11)
2
2
where n0 is the total particle density. This leads to the two coupled equations
Z


1
n0 =
d N ( U n ) + N ( U n ) f (),
(7.12)
2
Z


1
m=
d N ( U n ) N ( U n ) f (),
(7.13)
2
or equivalently
n0 =
m=

1X
2 s
1X
2



U n0
Um
dN 
s
f (),
2
2


Z
U n0
Um
s
dN 
s
f (),
2
2
Z

(7.14)
(7.15)

which usually can not be solved analytically and must be treated numerically.

7.1.2

Stoner criterion

An approximate solution can be found if m  n0 . Eqs.(7.14) and (7.15) are solved by adapting
the chemical potential . For low temperatures and small magnetization we can expand as
(m, T ) = F + (m, T ).

(7.16)

The constant energy shift U n0 /2 appearing in Eqs.(7.14) and (7.15) can be absorbed into F .
The Fermi-Dirac distribution takes the form
1
f () = [(m,T )]
,
(7.17)
e
+1
where = (kB T )1 . After expanding equation (7.14) for small m, one obtains using the Sommerfeld expansion,
"
#


Z
1 U m 2 00
n0 df () N () +
N ()
(7.18)
2
2

ZF



2
1
d N () + N (F ) + (kB T )2 N 0 (F ) +
6
2

140

Um
2

2

N 0 (F ),

(7.19)

where we introduced the abbreviations N 0 () = dN ()/d and N 00 () = d2 N ()/d2 . Since the
first term on the right side of Eq.(7.19) is identical to n0 , (m, T ) is immediately found to be
given by
"

2 #
N 0 (F ) 2
U
m
1
(m, T )
(k T )2 +
.
(7.20)
N (F ) 6 B
2
2
Analogously, the expansion of Eq.(7.15) in m and T , results in
"
 #

Z
Um
1 000
Um 3
0
m df () N ()
+ N ()
2
3!
2
#
"

2

U
m
Um
2
1
00
0
2 00
N (F ) + N (F )
,
N (F ) + (kB T ) N (F ) +
6
3!
2
2
and, finally, inserting the result for of Eq.(7.20) into (7.22), we find





2
Um 3
Um
2
2 2
m = N (F ) 1 (kB T ) 1 (F )
N (F )2 (F )
,
6
2
2

(7.21)
(7.22)

(7.23)

where
12 (F ) =

N 0 (F )
N (F )

2

N 00 (F )
,
N (F )

and

22 (F ) =

1
2

N 0 (F )
N (F )

2

N 00 (F )
.
3!N (F )

(7.24)

The structure of Eq.(7.23) is m = am + bm3 , where b is assumed to be negative. Thus, two

types of solutions emerge

a < 1,

0,
2
m =
(7.25)

1 a , a 1.
b
With this, a = 1 corresponds to a critical value.
E
m U N (F ) > 2

Um

U N (F ) < 2
am + bm2

N ()

N ()

Figure 7.1: Graphical solution of equation (7.23) and the resulting magnetization. The Fermi
sea of each spin configuration is shifted by U m/2, resulting in a finite total magnetization.
Here, this condition corresponds to


1
2
2 2
1 = U N (F ) 1 (kB TC ) 1 (F ) ,
2
6

(7.26)

yielding

6
kB TC =
1 (F )

2
1

U N (F )
141

r
1

Uc
U

(7.27)

for U > Uc = 2/N (F ). This is an instability condition for the paramagnetic Fermi liquid state
with m = 0, and TC is the Curie temperature, below which the ferromagnetic state appears (see
Figure 7.1). The temperature dependence of the magnetization M of the ferromagnetic state
(T < TC ) is given by
p
M (T ) = B m(T ) TC T ,
(7.28)
close to the phase transition (TC T  TC ). Note that the Curie temperature TC is nonzero
for U N (F ) > 2, and TC 0 in the limit U N (F ) 2+ . For U N (F ) < 2 no phase transition
occurs. This condition for a finite transition temperature TC is known as the Stoner criterion.
This simple model also describes a so-called quantum phase transition, i.e., a phase transition
that appears at T = 0 as a function of system parameters, which in our case are the density
of states N (F ) and the Coulomb repulsion U . While thermal fluctuations destroy the ordered
state at finite temperature via entropy increase, entropy is irrelevant at T = 0. Here, the order
is suppressed by quantum fluctuations (Heisenbergs uncertainty principle).
T

Paramagnet
Ferromagnet
Paramagnet

Ferromagnet
UN( F )

Druck

Figure 7.2: Phase diagram of a Stoner ferromagnet in the T -U N (F ) and T -p plane, respectively.

The density of states as an internal parameter can, for example, be changed by applying a
pressure. By reducing the lattice constant, pressure may facilitate the motion of the conduction
electrons and increase the Fermi velocity. Consequently, the density of states is reduced (cf.
Figure 7.2). Indeed, pressure is able to destroy ferromagnetism in weakly ferromagnetic materials as ZrZn2 , MnSi, and UGe2 . In other materials, the Curie temperature is high enough, such
that the technologically applicable pressure is insufficient to suppress magnetism. It is, however,
possible, that pressure leads to other transitions, such as structural phase transitions, that eventually destroy magnetism. This is seen in iron (Fe), where a pressure of about 12 GPa induces
a transition from magnetic iron with body-centered crystal (bcc) structure to a nonmagnetic,
hexagonal close packed (hcp) structure (cf. Figure 7.3).
While this structural transition is a quantum phase transition as well, it appears as a discontinT(K)

T(K)
50

UGe2

1000

FM

0.5

1.5

Fe
fcc

Fe
bcc
FM

10

p (GPa)

Fe
Fe
hpc

20

Figure 7.3: Phase diagrams of UGe2 and Fe.

142

p (GPa)

uous, first order2 transition. In some cases, pressure can also induce an increase in N (F ), for
example in metals with multiple bands, where compression leads to a redistribution of charge.
One example is the ruthenate Sr3 Ru2 O7 for which uniaxial pressure along the z-axis leads to
magnetism.

N( )
3d

4s
F
Ni

F
Cu

Figure 7.4: The position of the Fermi energy of Cu and Ni, respectively.
Finally, let us turn to the question, why Cu, being a direct neighbor of Ni in the 3d-row of
the periodic table, is not ferromagnetic, even though both elemental metals share the same fcc
crystal structure. The answer is given by the Stoner criterion U N (F ) = 2. While the conduction electrons at the Fermi level of Ni have 3d-character and belong to a narrow band with a
large density of states, the Fermi energy of Cu is situated in the broad 4s-band and constitutes
a much smaller density of states (cf. Figure 7.4). With this, the Cu conduction electrons are
much less localized and feature a weaker tendency towards ferromagnetic order.

7.1.3

Spin susceptibility for T > TC

Next we study the response of the metallic system in the paramagnetic state when we apply a
small magnetic field H along the z-axis, which induces a spin polarization due to the Zeeman
coupling,
Z
o
1n
HZ = gB d3 r H
b (r) b (r) .
(7.29)
2
From the self-consistency equations (7.14) and (7.15) we obtain


Z
X
1
Um
m=
df ()
s N  B sH s
(7.30)
2
2
s


Z
Um
0
df ()N ()
+ B H
(7.31)
2



2
Um
2
2
+ B H
(7.32)
= N (F ) 1 (kB T ) 1 (F )
6
2
to lowest order in m and H. Solving this equation for m yields
M = B m =

0 (T )
H,
1 U 0 (T )/22B

(7.33)

2
The Stoner instability is a simplification of the quantum phase transition. In most cases, a discontinuous
phase transition originates in the band structure or in fluctuation effects, which were ignored here, For more
details consult [D. Belitz and T.R. Kirkpatrick, Phys. Rev. Lett. 89, 247202 (2002)].

143

and, consequently, the magnetic susceptibility reads

=

0 (T )
M
=
,
H
1 U 0 (T )/22B

(7.34)

where the bare susceptibility 0 is given by


2
2
2
0 (T ) = B N (F ) 1 (kB T ) 1 (F ) .
6
2

(7.35)

We see, that the denominator of the susceptibility (T ) vanishes exactly when the Stoner instability criterion for finite temperatures is fulfilled (see Eq.(7.26)). The susceptibility
(T )

0 (TC )
2
TC
T2

(7.36)

diverging at Tc defined through Eq.(7.26) indicates the instability. Note that for T TC from
the paramagnetic side, the susceptibility diverges like (T ) |TC T |1 corresponding to the
mean field behavior, since the mean field critical exponent for the susceptibility takes the value
= 1.

7.2

General spin susceptibility and magnetic instabilities

The ferromagnetic state is characterized by a uniform magnetization. There are, however,

magnetically ordered states which do not feature a nonzero net magnetization but spacially
modulated magnetic moments.. Examples are spin density wave (SDW) states, antiferromagnets
and spin spiral states. In this section, we analyze general instability conditions for metallic
systems to some magnetically ordering.

7.2.1

General dynamic spin susceptibility

We consider a magnetic field, oscillating in time and with spacial modulation

H(r, t) = H 0 eiqrit et ,

(7.37)

where 0+ yields an adiabatic switching on the field. We calculate the resulting magnetization, for the corresponding Fourier component. For that, we proceed analogous to the discussion
b
in Chapter 3 and define the spin density operator S(r)
in real space,

b (r)
b (r) +
b (r)
b (r)

~ X b
b
b 0 (r) = ~
b (r)
b (r) + i
b (r)
b (r)
S(r)
=
(r) ss0
(7.38)
i

2 0 s
2

b (r)
b (r)
b (r)
b (r)
s,s

with momentum space representation

Z
1 Xb
~ X
iqr
b
b
ck,s ss0 ck+q,s0 =
S k,q ,
S q = d3 rS(r)e
=
2

0
k,s,s

(7.39)

b
where S
k,q = (~/2)ck,s ss0 ck+q,s0 . The Hamiltonian of the electronic system with contact
interaction is given by

H = H0 + HZ + Hint ,

144

(7.40)

where
H0 =

k b
cks b
cks ,

(7.41)

k,s

Z
gB
b
d3 rH(r, t) S(r),
~
Z
= U d3 rb
(r)b
(r).

HZ =
Hint

(7.42)
(7.43)

The operator HZ describes the Zeeman coupling between the electrons of the metal and the
perturbing field. We investigate a magnetic field

1
1
H = H + (q, )eiqrit et i
(7.44)
2
0
in the x-y-plane. The Zeeman term then simplifies to
HZ =

gB +

H (q, )Sbq
+ h.c.,
~

(7.45)

x iS
by . In the following the Hermitian conjugate (h.c.) part of Eq.(7.45) will
where Sbq
= Sbq
q
be ignored. We use
~X

ckq ,
(7.46)
=
b
ck b
Sbq

in the c-operator representation. In the framework of linear response theory, this coupling will
+
(q, t) = (B /~)hSbq+ (t)i eit+t . Using the same equations of
induce a magnetization Mind
motion as in Section 3.2,
i~

b+
+
S = [Sbk,q
, H],
t k,q

(7.47)

+
ck+q , we can determine this induced magnetization, first without the interaction
=b
ck b
with Sbk,q
term (U=0). We obtain for the given Fourier component,

i~

b+
+
ck+q )H + (q, )eit+t .
ck b
ck+q b
(t) g~B (b
ck b
S (t) = (k+q k )Sbk,q
t k,q k,q

(7.48)

Using the monochromatic time dependence of the field and the response (eit+t ) and applying
the thermal average we obtain,
+
(k+q k ~ i~)hSk,q
(t)i = g~B (nk+q nk )H + (q, )eit+t ,

(7.49)

which then leads to the induced spin density - magnetization,

~
1X +
+
+
Mind
(q, )ieit+t = hSind
(q, )ieit+t =
hSk,q (t)i
B

(7.50)

~
(q, )H + (q, )eit+t ,
=
B 0
with
0 (q, ) =

nk+q nk
g2B X
.

k+q k ~ i~
k

145

(7.51)

Note that the form of the bare susceptibility 0 (q, ) is similar to the Lindhard function (3.63),
actually identical, if there is no spin polarization. The result (7.50) describes the induced spin
density within linear response approximation.
In a next step, we want to include the effects of the interaction. Analogously to the induce
charge modulation found in Section 3.2, the induced spin density generates an effective field on
the spin of the electrons (mean field). The induced spin polarization may be represented as
an effective magnetic field through the exchange interaction. To implement this feature let us
rewrite the contact interaction term in Eq.(7.40) in the form
Hint =

U X
b
c
c b
c 0
b
c 0
0 b
0 0 k+q k k q0 k
k,k ,q

U X
=
b
c b
c
c 0 b
c 0 0 + const.
0 b
0 0 k k+q k k q

(7.52)

k,k ,q

U X b+ b
Sq0 Sq0 .
= 2
~
0
q

+
The induced spin polarization hSind
(q, )i acts through the exchange interaction as an effective
+
(q, )iq,q0 in Eq.(7.53) ,
(local) field, as can be seen by replacing Sbq+0 hSind

g
U X b+ b
U
+

Sq0 Sq0 2 hS + (q, )iSbq

= B Hind
(q, )Sbq
2
~
~
~
0

(7.53)

+
(q, ) finally reads
where the effective magnetic field Hind
+
(q, ) =
Hind

U
hS + (q, )i.,
gB ~

(7.54)

with the same monochromatic time dependence as above. This induced field acts on the spins
as well, such that the total response of the spin density on the external field becomes

M + (q, ) = B hS + (q, )i
~
+
= 0 (q, )[H + (q, ) + Hind
(q, )]
U
= 0 (q, )H + (q, ) + 0 (q, )
hS + (q, )i
(7.55)
gB ~
U
= 0 (q, )H + (q, ) + 0 (q, ) 2 M + (q, ).
gB
In the last step we introduce self-consistency taking the induced magnetization as the real
magnetization. With the definition
M + (q, ) = (q, )H + (q, )

(7.56)

of the susceptibility we find

(q, ) =

0 (q, )
.
1 2U2 0 (q, )

(7.57)

which corresponds to the random phase approximation (RPA) (see Section 3.2). This form of
the susceptibility is found to be valid for all field directions, as long as spin-orbit coupling is
neglected. Within the random phase approximation, the generalization of the Stoner criterion
for the appearance of an instability of the system at finite temperature reads
1=

U
(q, ).
22B 0
146

(7.58)

For the limiting case (q, ) (0, 0) corresponding to a uniform, static external field, we obtain
for the bare susceptibility
0 (q, 0) =

22B X nk+q nk

k+q k
k

22 X f (k )
q0
B
= 0 (T ),

k

(7.59)

which corresponds to the Pauli susceptibility (g = 2). Then, (T ) from equation (7.57) is again
cast into the form (7.34) and describes the instability of the metal with respect to ferromagnetic
spin polarization, when the denominator vanishes. Similar to the charge density wave, the
isotropic deformation for q = 0 is not the leading instability, when 0 (Q, 0) > 0 (0, 0) for a
finite Q. Then, another form of magnetic order is more favored.

7.2.2

Instability with finite wave vector Q

In order to show that, indeed, the Stoner instability does not always prevail among all possible
magnetic instabilities, we first go through a simple argument based on the local susceptibility.
For that, we define the local magnetic moment along the z-axis, M (r) = B hb
(r) b (r)i, and
consider the nonlocal relation
Z
M (r) = d3 r0
0 (r r 0 )Hz (r 0 ),
(7.60)
within the linear response approximation. In Fourier space, the same relation reads
Mq = 0 (q)Hq ,

(7.61)

with
Z
0 (q) =

d3 r
0 (r)eiqr .

(7.62)

Figure 7.5: R0 , the ratio between the local and the extended susceptibility, plotted for a boxshaped band with width 2D. Depending where the Fermi energy lies = F /D, the susceptibility
is dominated by the contribution 0 (q = 0) or by the susceptibility at finite q.
Now, compare 0 (q = 0) with 0 (q) defined as
Z
Z
1X
1X
3
iqr
0 (q) =
(q) =
d r
0 (r)e
= d3 r
0 (r)(r) =
0 (r = 0) .
q 0
q
147

(7.63)

This q-averaged susceptibility corresponds to the local susceptibillity. For a paramagnetic metal
at T = 0 we may write
Z
Z
22B X nk+q nk
2B
f () f (0 )
0 (q) = 2
dN () d0 N (0 )
=
,
(7.64)

k k+q
2
0 
k,q

and must be compared to 0 (q = 0) = 2B N (F ) (f () = (F )). The local susceptibility
depends on the density of states and the Fermi energy of the system. A very good qualitative
understanding can be obtained by a very simple form

, D  D,
(7.65)
N () = D

0,
|| > D,
for the density of states which does not correspond to particular band structure but mimics
a band of width 2D. With this rough approximation, the integral in equation (7.64) is easily
evaluated. The ratio between (q) and 0 (q = 0) is then found to be
0 (q)
R0 =
= ln
0 (q = 0)

4
1 2


+ ln

1
1+


,

(7.66)

with = F /D where D < F < +D. For both small and large band fillings (F close to the
band edges), the tendency towards ferromagnetism dominates (cf. Figure 7.5), whereas when
F lies in the center of the band, the susceptibility 0 (q) is not maximal at q = 0 anymore, and
magnetic ordering with a well-defined finite q = Q becomes more probable.

7.2.3

Influence of the band structure

Whether magnetic order arises at finite q or not depends strongly on the details of the band
structure. The argument given above, comparing the local (r = 0) to the uniform (q = 0)
susceptibility is nothing more than a vague indicator for a possible instability at nonzero q. A
crucial ingredient for the appearance of magnetic order at a given q = Q is the so-called nesting
of the Fermi surface. Within extended areas of the Fermi surface the energy dispersion satisfies
the nesting condition,
k+Q = k

(7.67)

where k = k F and Q is some fixed vector. The nesting conditions connects for given k an
electron- and hole-like band states (at T = 0 filled and empty states, respectively). If the Fermi
surface of a material features such a nesting trait, the susceptibility will be dominated by the
contribution from this vector Q. In order to see this, let us investigate the static susceptibility
0 (q) for q = Q under the assumption, that equation (7.67) holds for all k (see tight-binding
example below). Thus,
Z
22B X nk+Q nk
d3 k f(k ) f(k )
2
0 (Q; T ) =
= B
,

k k+Q
(2)3
k

(7.68)

where f() = f ( + F ) = f () = [exp(/kB T ) + 1]1 and f is the Fermi-Dirac distribution.

Under the further assumption that k is weakly angle dependent, we find
2

0 (Q; T ) = B

2B
d3 k tanh(k /2kB T )
=
(2)3
k
2

148

Z
dN ( + F )

tanh(/2kB T )
.

(7.69)

In order to approximate this integral properly, we notice that the integral has a logarithmic
divergence a infinite energies . The band width gives a natural cutoff. Let us, therefore, take
the density of states in Eq.(7.65) with F = 0,
ZD

tanh(/2kB T )

0

 


 
4e
1.14D
D
+ ln
2B N (F ) ln
,
= 2B N (F ) ln
2kB T

2kB T
2

0 (Q; T ) B N (F )

(7.70)
(7.71)

where we assumed D  kB T , and where 0.57721 is the Euler constant. The bare susceptibility 0 diverges logarithmically at zero temperatures. Inserting the result (7.71) into the
generalized Stoner relation (7.58), results in


U N (F )
1.140
0=1
,
(7.72)
ln
2
2kB Tc
with the critical temperature
kB Tc = 1.140 e2/U N (F ) .

(7.73)

A finite critical temperature persists for arbitrarily small positive values of U N (F ). The nesting
condition for a given Q leads to a maximum of 0 (q, 0; T ) at q = Q and triggers the relevant
instability in the system. The latter finally stabilizes in a magnetic ordered phase with wave
vector Q, the so-called spin density wave. The spin density distribution takes, for example, the
form
S(r) = zS cos(Q r),

(7.74)

without a uniform component. In comparison, the charge density wave was a modulation of the
charge density with a much smaller amplitude than the height of the uniform density, i.e.,
(r) = 0 + cos(Q r),

(7.75)

with  0 . The spin density state frequently appear in low-dimensional systems like organic
conductors, or in transition metals such as chromium (Cr) for example. In all cases, nesting
plays an important role (cf. Figure 7.6).
In quasi-one-dimensional electron systems, a main direction of motion dominates over two other
directions with weak dispersion. In this case, the nesting condition is very probable to be
fulfilled, as it is schematically shown in the center panel of Figure 7.6. Chromium is a threedimensional metal, where nesting occurs between a electron-like Fermi surface around the -point
and a hole-like Fermi surface at the Brillouin zone boundary (H-point). These Fermi surfaces
originate in different bands (right panel in Figure 7.6). Chromium has a cubic body centered
crystal structure, where the H-point at (/a, 0, 0) leads to the nesting vector Qx k (1, 0, 0) and
equivalent vectors in y- and z-direction, which are incommensurable with the lattice.
The textbook example of nesting is found in a tight-binding model of a simple cubic lattice with
nearest-neighbor hopping at half filling. The band structure is given by
k = 2t[cos(kx a) + cos(ky a) + cos(kz a)],

(7.76)

where a is the lattice constant and t the hopping term. Because of half filling, the chemical
potential = F lies at = 0 such that k = k . Obviously, k+Q = k holds for all k, for the
nesting vector Q = (/a)(1, 1, 1). This full nesting trait is a signature of the total particle-hole
symmetry, i.e. in the ground state there are as many occupied as empty states. Analogously to
the Peierls instability, the spin density wave induces the opening of a gap at the Fermi surface.
This is another example of a Fermi surface instability. In this situation, the gap is confined to
the areas of the Fermi surface obeying the nesting condition. Contrarily to the ferromagnetic
order, the material can become insulating when forming the spin density wave state.
149

Chrom

quasieindimensional

eindimensional

H
Q

Q
Q

BZ
BZ

BZ
lochartige
Fermiflche

elektronartige
Fermiflche

Figure 7.6: Sketch of Fermi surfaces favorable for nesting. In purely one-dimensional systems
(left panel) there is a well-defined nesting vector pointing from one end of the Fermi surface to
the other one. In quasi-one-dimensional systems nesting is almost perfect (central panel). In
special cases (e.g. Cr) the Fermi surface(s) of three-dimensional systems show nesting properties
(right panel) promoting an instability of the susceptibility at finite q.

7.3

Stoner excitations

In this last section, we discuss the elementary excitations of the ferromagnetic ground state
with n > n , including both particle-hole excitations and collective modes. We focus on spin
excitations, for which we make the Ansatz
X
ck |g i.
(7.77)
|q i =
fk b
ck+q b
k

In this excitation, an electron is extracted from the ground state |g i and is replaced by an
electron with opposite spin. We consider excitations with a fixed momentum transfer q. We have
to ensures that an electron with (k, ) is available to be removed, and that the state (k + q, )
is unoccupied. The excitation energy can be obtained by solving the following Schrodinger
equation
H|q i = (Eg + ~q )|q i ,

(7.78)

with Eg as the ground state energy (H|g i = Eg |g i). Actually it is more convenient to rearrange
this equation into the following form eliminating Eg ,
~q |q i =

X
k

fk [H, b
ck+q b
ck ]|g i

(7.79)

with the Hamiltonian

H=

k b
cks b
cks +

k,s

U X
b
c
c b
c 0
b
c 0 .
0 b
0 0 k+q k k q0 k

(7.80)

k,k ,q

Evaluating the commutator in Eq.(7.79) we obtain

~q |q i = (k+q k )|q i
+

o
U X Xn
fk
b
ck0 +q0 b
ck0 b
ck+qq0 b
ck b
ck+q b
ckq0 b
ck0 q0 b
ck0 |g i .

0 0
k

k ,q

(7.81)

150

The four-b
c-operator terms are not so easy to handle. Therefore we decouple them through the
following approximation,
b
ck0 +q0 b
ck0 b
ck+qq0 b
ck b
ck+q b
ckq0 b
ck0 q0 b
ck0
q0 ,0 (nk0 nk0 ) b
ck+q, b
ck + (nk+q k0 ,k+q nk k,k0 +q0 ) b
ckq0 +q b
ckq0
(7.82)
where nks is the Fermi-Dirac distribution for electrons of spin s with the energy ks . We insert

this approximation now into the Schrdingier

equation (7.79) and multiply from the left with

to obtain the projected equation,

ck+q
hg |b
ck

b
o UX
n
+ k U (n n ) =
fk ~q k+q
fk (nk+q nk ) = Rq
(7.83)

where we used that the density of electrons of spin s is given by

1X
ns =
n 0 .
0 ks

(7.84)

and we defined Rq . We solve now Eq.(7.83) for the wave function fk and obtain
fk =

Rq

(7.85)

~q k+q + k

with ks = k + U ns. We use now fk to reexpress Rq which yields the implicit equation for the
excitation energies ~q ,
nk+q nk
1X
Rq = U Rq
(7.86)

~q k+q + k
k

where we may divide by Rq 6= 0. This equation contains a continuum of states corresponding

to the electron-hole excitations given by
~q = k+q k = k+q k + U (n n ) = k+q k + .

(7.87)

This part of the spectrum is depicted by the shaded region in Fig.7.7. Note that the spin
splitting of the spectrum opens a window in the low-energy low-momentum transfer sector of
the excitations.
There is also a collective mode. It is easy to see from Eq.(7.86) that in the q 0 limit we find
1=

U (n n )

~0 U (n n )

~0 = 0

(7.88)

indicating that there are excitations in the mentioned window. We calculate now the dispersion
for small q ( kF ). Using the fact that ~q  we can expand Eq.(7.86) leading to
#"
"
#1
U X nk+q nk
U X nk+q nk
~q 1 +

.
(7.89)

k+q k

(k+q k )2
k

For small q we expand up order

U X nk+q nk

k+q k +
k

q2

(actually q-linear contributions cancel),

(

 )
k+q k
k+q k 2
U X nk+q nk

1
+



U X nk + nk ~2 q 2 nk nk (2~2 k q)2
1 +

2
2m
2
4m2
k

~2 q 2
1 +
2m2



4F
U n0
3
151

(7.90)

and analogously

U X nk+q nk
1

(k+q k + )2

(7.91)

such that we obtain

~q

~2 q 2 2F
~2 q 2
(U N (F ) 2)
2m 3
2m

U
1
Uc

1/2
.

(7.92)

Were we use that (U Uc )1/2 for U Uc . Hence, if U > Uc = 2/N (F ) we find ~q > 0.
This collective excitation features a q 2 -dependent dispersion, vanishing in the limit q 0. This
is a consequence of the ferromagnetic state breaking a continuous symmetry. The continuous
rotation symmetry is broken by the choice of a given direction of magnetization. A uniform q = 0
rotation of the magnetization does not cost any energy 0 = 0. This result was already found
in equation (7.88) and is predicted by the so-called Goldstone theorem.3 Such an infinitesimal
rotation is induced by a global spin rotation,
X

b
ck b
ck = Sbtot
(7.93)
k

which acting on the ground state |g i leads to a degenerate state and corresponds to the state
(7.77) with a wave function fk = const.. Since the elementary excitations have an energy gap of
the order of at small q, the collective excitations, which are termed magnons, are well-defined
quasiparticles describing propagating spin waves. When these modes enter the electron-hole
continuum, they are damped in the same way as plasmons decaying into electrons and holes
(see Figure 7.7). Being a bound state composed of an electron and a hole, magnons are bosonic
quasiparticles.

~!q

electron(hole*
con+nuum*
magnon&

kF "

kF #

kF " + kF #

Figure 7.7: Schematic diagram of the elementary spin excitationspectrum (light gray) and collective modes (magnons, solid line) of the Stoner ferromagnet.

The Goldstone theorem states that, in a system with a short-ranged interaction, a phase which is reached
by the breaking of a continuous symmetry features a collective excitation with arbitrarily small energy, so-called
Goldstone modes. These modes have bosonic character. In the case of the Stoner ferromagnet, these modes are
the magnons or spin waves.

152

Chapter 8

Magnetism of localized moments

Up to now, we have mostly assumed that the interaction between electrons leads to secondary
effects. This was, essentially, the message of the Fermi liquid theory, the standard model of
condensed matter physics. There, the interactions of course renormalize the properties of a
metal, but their description is still possible by using a language of nearly independent fermionic
quasiparticles with a few modifications. Even in connection with the magnetism of itinerant
electrons, where interactions proved to be crucial, the description in terms of extended Bloch
states. Many properties were determined by the band structure of the electrons in the lattice,
i.e., the electrons were preferably described in k-space.
However, in this chapter, we will consider situations, were it is less clear wether we should
describe the electrons in momentum or in real space. The problem becomes obvious with the
following Gedanken experiment: We look at a regular lattice of H-atoms. The lattice constant
should be large enough such that the atoms can be considered to be independent for now. In the
ground state, each H-atom contains exactly one electron in the 1s-state, which is the only atomic
orbital we consider at the moment. The transfer of one electron to another atom would cost the
relatively high energy of E(H + )+E(H )2E(H) 15eV, since it corresponds to an ionization.
Therefore, the electrons remain localized on the individual H-atoms and the description of the
electron states is obviously best done in real space. The reduction of the lattice constant will
gradually increase the overlap of the electron wave functions of neighboring atoms. In analogy
to the H2 molecule, the electrons can now extend on neighboring atoms, but the cost in energy
remains that of an ionization. Thus, transfer processes are only possible virtually, there are
not yet itinerant electrons in the sense of a metal.

weak overlapp

strong overlapp

Figure 8.1: Possible states of the electrons in a lattice with weak or strong overlap of the electron
wave functions, respectively.
On the other hand, we know the example of the alkali metals, which release their outermost nselectron into an extended Bloch state and build a metallic (half-filled) band. This would actually

153

work well for the H-atoms for sufficiently small lattice constant too.1 Obviously, a transition
between the two limiting behaviors should exist. This metal-insulator transition, which occurs,
if the gain of kinetic energy surpasses the energy costs for the charge transfer. The insulating
side is known as a Mott insulator.
While the obviously metallic state is reliably described by the band picture and can be sufficiently
well approximated by the previously discussed methods, this point of view becomes obsolete
when approaching the metal-insulator transition. According to band theory, a half-filled band
must produce a metal, which definitely turns wrong when entering the insulating side of the
transition. Unfortunately, no well controlled approximation for the description of this metalinsulator transition exists, since there are no small parameters for a perturbation theory.
Another important aspect is the fact, that in a standard Mott insulator each atom features
an electron in the outermost occupied orbital and, hence, a degree of freedom in the form of
a localized spin s = 1/2, in the simplest case. While charge degrees of freedom (motion of
electrons) are frozen at small temperatures, the same does not apply to these spin degrees of
freedom. Many interesting magnetic phenomena are produced by the coupling of these spins.
Other, more general forms of Mott insulators exist as well, which include more complex forms
of localized degrees of freedom, e.g., partially occupied degenerate orbital states.

8.1

Mott transition

First, we investigate the metal-insulator transition. Its description is difficult, since it does
not constitute a transition between an ordered and a disordered state in the usual sense. We
will, however, use some simple considerations which will allow us to gain some insight into the
behavior of such systems.

8.1.1

Hubbard model

We introduce a model, which is based on the tight-binding approximation we have introduced in

chapter 1. It is inevitable to go back to a description based on a lattice and give up continuity.
The model describes the motion of electrons, if their wave functions on neighboring lattice sites
only weakly overlap. Furthermore, the Coulomb repulsion, leading to an increase in energy, if a
site is doubly occupied, is taken into account. We include this with the lattice analogue of the
contact interaction. The model, called Hubbard model, has the form
X
X
H = t
(b
cis b
cjs + h.c.) + U
n
bi n
bi ,
(8.1)
i

hi,ji,s

where we consider hopping between nearest neighbors only, via the matrix element t. Note,
()
that b
cis are real-space field operators on the lattice (site index i) and n
bis = b
cis b
cis is the density
operator. We focus on half filling, n = 1, one electron per site on average. There are two obvious
limiting cases:
Insulating atomic limit: We put t = 0. The ground state has exactly one electron on
each lattice site. This state is, however, highly degenerate. In fact, the degeneracy is 2N
(number of sites N ), since each electron has spin 1/2, i.e.,
Y
|A0 {si }i =
b
ci,si |0i,
(8.2)
i

where the spin configuration {si } can be chosen arbitrarily. We will deal with the lifting
of this degeneracy later. The first excited states feature one lattice site without electron
1

In nature, this can only be induced by enormous pressures metallic hydrogen probably exists in the centers
of the large gas planets Jupiter and Saturn due to the gravitational pressure.

154

and one doubly occupied site. This state has energy U and its degeneracy is even higher,
i.e., 2N 2 N (N 1). Even higher excited states correspond to more empty and doubly
occupied sites. The system is an insulator and the density of states is shown in Figure 8.2.
Metallic band limit: We set U = 0. The electrons are independent and move freely
via hopping processes. The band energy is found through a Fourier transform of the
Hamiltonian. With
1 X
b
cis =
b
cks eikri ,
(8.3)
N k
we can rewrite
t

(b
cis b
cjs + h.c.) =

hi,ji,s

k b
cks b
cks ,

(8.4)

k,s

where
k = t

X
a



eika = 2t cos kx a + cos ky a + cos kz a ,

(8.5)

and the sum runs over all vectors a connecting nearest neighbors. The density of states is
also shown in Figure 8.2. Obviously, this system is metallic, with a unique ground state
Y
(8.6)
ck |0i.
|B0 i =
(k )b
ck b
k

Note, that F = 0 at half filling, whereas the bandwidth 2D = 12t.

E

U
N(E)

N(E)

metallic limit

atomic limit

Figure 8.2: Density of states of the Hubbard model in the atomic limit (left) and in the free
limit (right).

8.1.2

Insulating state

We consider the two lowest energy sectors for the case t  U . The ground state sector has
already been defined: one electrons sits on each lattice site. The lowest excited states create the
sector with one empty and one doubly occupied site (cf. Figure 8.3). With the finite hopping
matrix element, the empty (holon) and the doubly occupied (doublon) site become mobile. A
155

fraction of the degeneracy (2N 2 N (N 1)) is herewith lifted and the energy obtains a momentum
dependence,
Ek,k0 = U + k + k0 > U 12t.

(8.7)

Even though ignoring the spin configurations here is a daring approximation, we obtain a qualitatively good picture of the situation.2 One notices that, with increasing |t|, the two energy
sectors approach each other, until they finally overlap. In the left panel Figure8.2 the holondoublon excitation spectrum is depicted by two bands, the lower and upper Hubbard bands,
where the holon is a hole in the lower and the doublon a particle in the upper Hubbard band.
The excitation gap is the gap between the two bands and we may interpret this system as an
insulator, called a Mott insulator. (Note, however, that this band structure depends strongly on
the correlation effects (e.g. spin correlation) and is not rigid as the band structure of a semiconductor.) The band overlap (closing of the gap) indicates a transition, after which a perturbative
treatment is definitely inapplicable. This is, in fact, the metal-insulator transition.

sector

sector

Figure 8.3: Illustration of the two energy sectors, and .

8.1.3

The metallic state

On the metallic side, the initial state is better defined since the ground state is a filled Fermi sea
without degeneracy. The treatment of the Coulomb repulsion U turns out to become difficult,
once we approach the Mott transition, where the electrons suffer a strong impediment in their
mobility. In this region, there is no straight-forward way of a perturbative treatment. The socalled Gutzwiller approximation, however, provides a qualitative and very instructive insights
into the properties of the strongly correlated electrons.
For this approximation we introduce the following important densities:
1: electron density
s : density of the singly occupied lattice sites with spin
s : density of the singly occupied lattice sites with spin
d: density of the doubly occupied sites
h: density of the empty sites
It is easily seen, that h = d and s = s = s/2, as long as no uniform magnetization is present.
Note, that d determines the energy contribution of the interaction term to U d, which we regard
as the index of fixed interaction energy sectors. Furthermore,
1 = s + 2d

(8.8)

holds. The view point of the Gutzwiller approximation is based on the renormalization of the
probability of the hopping process due to the correlation of the electrons,exceeding restrictions
2

Note that the motion of an empty site (holon) or doubly occupied site (doublon) is not independent of the
spin configuration which is altered through moving these objects. As a consequence, the holon/doublon motion
is not entirely free leading to a reduction of the band width. Therefore the band width seen in Figure8.2 (left
panel) is smaller than 2D, in general. The motion of a single hole was in detail discussed by Brinkman and Rice
(Phys. Rev. B 2, 1324 (1970).

156

due to the Pauli principle. With this, the importance of the spatial configuration of the electrons
is enhanced. In the Gutzwiller approximation, the latter is taken into account statistically by
simple probabilities for the occupation of lattice sites.
We fix the density of the doubly occupied sites d and investigate the hopping processes which
keep d constant. First, we consider an electron hopping from a singly occupied to an empty
site (i j). Hopping probability depends on the availability of the initial configuration. We
compare the probability to find this initial state for the correlated (P ) and the uncorrelated (P0 )
case and write
P ( 0) + P ( 0) = gt [P0 ( 0) + P0 ( 0)].

(8.9)

The factor gt will eventually appear as the renormalization of the hopping probability and, thus,
leads to an effective kinetic energy of the system due to correlations. We determine both sides
statistically. In the correlated case, the joint probability for i to be singly occupied and j to be
empty is obviously
P ( 0) + P ( 0) = sh = sd = d(1 2d).

(8.10)

where we used equation (8.8). In the uncorrelated case (where d is not fixed), we have
P0 ( 0) = ni (1 ni )(1 nj )(1 nj ) =

1
.
16

(8.11)

The case for follows accordingly. In order to collect the total result for hopping processes which
keep d constant, we have to do the same calculation for the hopping process (, ) (, ),
which leads to the same result. Processes of the kind (, 0) (, ) leave the sector of fixed
d and are ignored.3 With this, we obtain in all cases the same renormalization factor for the
kinetic energy,
gt = 8d(1 2d),

(8.12)

i.e., t gt t. We consider the correlations by treating the electrons as independent but with a
renormalized matrix element gt t. The energy in the sector d becomes
E(d) = gt kin + U d = 8d(1 2d)kin + U d,

kin

Z0

1
=
N

d N ().

(8.13)

For fixed U and t, we can minimize this with respect to d (note that this in not a variational
calculation in a strict sense, the resulting energy is not an upper bound to the ground state
energy), and find
1
d=
4

U
1
Uc


and

gt = 1

U
Uc

2
,

(8.14)

with the critical value

Uc = 8|kin | 25t 4D.

(8.15)

For u Uc , double occupancy and, thus, hopping is completely suppressed, i.e., electrons
become localized. This observation by Brinkman and Rice [Phys. Rev. B 2, 4302 (1970)]
provides a qualitative description of the metal-insulator transition to a Mott insulator, but
3

This formulation is based on plausible arguments. A more rigorous derivation can be found in the
literature, e.g., in D. Vollhardt, Rev. Mod. Phys. 56, 99 (1984); T. Ogawa et al., Prog. Theor.
Phys.
53, 614 (1975); S. Huber, Gutzwiller-Approximation to the Hubbard-Model (Proseminar SS02,
http://www.itp.phys.ethz.ch/proseminar/condmat02).

157

takes into account only local correlations, while correlations between different lattice sites are not
considered. Moreover, correlations between the spin degrees of freedom are entirely neglected.
The charge excitations contain contributions between different energy scales: (1) a metallic part,
described via the renormalized effective Hamiltonian
X
Hren =
gt k b
cks b
cks + U d,
(8.16)
k,s

and (2) a part with higher energy, corresponding to charge excitations on the energy scale U ,
i.e., to excitations raising the number of doubly occupied sites by one (or more).
We can estimate the contribution to the metallic conduction. Since in the tight-binding description the current operator contains the hopping matrix element and is thus subject to the same
renormalization as the kinetic energy, we obtain
1 () =

p2
4

() + 1high

energy

(),

(8.17)

where we have used equation (6.12) for a perfect conductor (no residual resistivity in a perfect
lattice). There is a high-energy part which we do not specify here. The plasma frequency is
renormalized, p2 = gt p2 , such that the f -sum rule in equation (6.13) yields
Z
I=

d1 () =

p2
8

gt + Ihigh

energy

p2
8

(8.18)

For U Uc , the coherent metallic part becomes weaker and weaker,

 2 ! 2
p
p2
U
gt = 1
.
8
Uc
8

(8.19)

According to the f -sum rule, the lost weight must gradually be transferred to the high-energy
contribution.

8.1.4

Fermi liquid properties of the metallic state

The just discussed approximation allows us to discuss a few Fermi liquid properties of the metallic
state close to metal-insulator transition in a simplified way. Let us investigate the momentum
distribution. According to the above definition,
1 X
kin =
k ,
(8.20)
N
kFS

where the sum runs over all k in the Fermi sea (FS). One can show within the above approximation, that the distribution is a constant within (nin ) and outside (nout ) the Fermi surface for
finite U , such that, for k in the first Brillouin zone,
1
1 X
1 X
1
=
nin +
nout = (nin + nout )
(8.21)
2
N
N
2
kFS
/

kFS

and
gt kin =

1 X
1 X
nin k +
nout k .
N
N
kFS
/

kFS

Taking into account particle-hole symmetry, i.e.,

X
X
X
k =
k +
k = 0,
k

(8.22)

kFS
/

kFS

158

(8.23)

we are able to determine nin and nout ,

nin + nout = 1
nin nout = gt

nin = (1 + gt )/2
nout = (1 gt )/2

(8.24)

With this, the jump in the distribution at the Fermi energy is equal to gt , which, as previously,
corresponds to the quasiparticle weight (cf. Figure 8.4). For U Uc it vanihes, i.e., the
quasiparticles cease to exist for U = Uc .

nk
gt
kF

Figure 8.4: The distribution function in the Gutzwiller approximation, displaying the jump at
the Fermi energy.
Without going into the details of the calculation, we provide a few Fermi liquid parameters. It
is easy to see that the effective mass
m
= gt ,
(8.25)
m
and thus


3U 2
F1s = 3 gt1 1 = 2
,
(8.26)
Uc U 2

where t = 1/2m and the density of states N (F ) = N (F )gt1 . Furthermore,
U N (F ) 2Uc + U
U,
4
(U + Uc )2 c
U N (F ) 2UC U
F0s =
U,
4
(U Uc )2 c

F0a =

2B N (F )
,
1 + F0a
N (F )
= 2
.
n (1 + F0s )
=

(8.27)
(8.28)

It follows, that the compressibility vanishes for U Uc as expected, since it becomes more and
more difficult to compress the electrons or to add more electrons, respectively. The insulator is, of
curse, incompressible. The spin susceptibility diverges because of the diverging density of states
N (F ) . This indicates, that local spins form, which exist as completely independent degrees of
freedom at U = Uc . Only the antiferromagnetic correlation between the spins would lead to a
renormalization, which turns finite. This correlation is, as mentioned above, neglected in the
Gutzwiller approximation. The effective mass diverges and shows that the quasiparticles are
more and more localized close to the transition, since the occupation of a lattice site is getting
more rigidly fixed to 1.4 As a last remark, it turns out that the Gutzwiller approximation is
well suited to describe the strongly correlated Fermi liquid 3 He (cf. [D. Vollhardt, Rev. Mod.
Phys. 56, 99 (1984)]).
4
This can be observed within the Gutzwiller approximation in the form of local fluctuations of the particle
number. For this, we introduce the density matrix of the electron states on an arbitrary lattice site,
s
= h|0ih0| + d|ih| + (|ih| + |ih|) ,
(8.29)
2

159

8.2

The Mott insulator as a quantum spin system

One of the most important characteristics of the Mott insulator is the presence of spin degrees of
freedom after the freezing of the charge. This is one of the most profound features distinguishing
a Mott insulator from a band insulator. In our simple discussion, we have seen that the atomic
limit of the Mott insulator provides us with a highly degenerate ground state, where a spin-1/2
degree of freedom is present on each lattice site. We lift this degeneracy by taking into account
the kinetic energy term Hkin (t  U ). In this way new physics appears on a low-energy scale,
which can be described by an effective spin Hamiltonian. Prominent examples for such spin
systems are transition-metal oxides like the cuprates La2 CuO4 , SrCu2 O3 or vanadates CaV4 O9 ,
NaV2 O5 .

8.2.1

The effective Hamiltonian

In order to employ our perturbative considerations, it is sufficient to observe the spins of two
neighboring lattice sites and to consider perturbation theory for discrete degenerate states. Here,
this is preferably done in real space. There are 4 degenerate configurations, {| , i, | , i, | ,
i, | , i}. The application of Hkin yields
Hkin | , i =

Hkin | , i = 0,

(8.31)

Hkin | , i = Hkin | , i = t| , 0i t|0, i,

(8.32)

where, in the last two cases, the resulting states have an energy higher by U and lie outside the
ground state sector. Thus, it becomes clear that we have to proceed to second order perturbation,
where the states of higher energy will appear only virtually (cf. Figure 8.5).

E=U

or
final
spin configurations

initial
spin configuration

virtual states

Figure 8.5: Illustration of the origin of the superexchange.

We obtain the matrix elements
Ms1 ,s2 ;s0 ,s0 =
1

X
n

hs1 , s2 |Hkin |ni

1
hn|Hkin |s01 , s002 i,
hn|HCoul |ni

(8.33)

where |ni = | , 0i or |0, i, such that the denominator is always U . We end up with
2t2
.
U

(8.34)

hn2 i hni2 = hn2 i 1 = tr(

n2 ) 1 = 4d + s 1 = 2d.

(8.30)

M; = M; = M; = M; =
from which we deduce the variance of the occupation number,

The deviation from single occupation vanishes with d, i.e., with the approach of the metal-insulator transition.
Note that the dissipation-fluctuation theorem connects hn2 i hni2 to the compressibility.

160

Note that the signs originates from the anti-commutation properties of the Fermion operators.
In the subspace {| , i, | , i} we find the eigenstates of the respective secular equations,
1
(| , i + | , i) ,
2
1
(| , i | , i) ,
2

E = 0,
E=

(8.35)
4t2
.
U

(8.36)

Since the states | , i and | , i have energy E = 0, the sector with total spin S = 1 is
degenerate (spin triplet). The spin sector S = 0 with the energy 4t2 /U is the ground state
(spin singlet).
An effective Hamiltonian with the same energy spectrum for the spin configurations can be
b and S
b on the two lattice sites
written with the help of the spin operators S
1
2


2
4t2
b S
b ~ ,
Heff = J S
J=
> 0.
(8.37)
1
2
4
U ~2
This mechanism of spin-spin coupling is called superexchange and introduced by P.W. Anderson
[Phys. Rev. 79, 350 (1950)].
Since this relation is valid between all neighboring lattice sites, we can write the total Hamiltonian as
X
b S
b + const.
HH = J
S
(8.38)
i
j
hi,ji

This model, reduced to spins only, is called Heisenberg model. The Hamiltonian is invariant
under a global SU (2) spin rotation,
X
b
b=
b .
Us () = eiS ,
S
S
(8.39)
j
j

Thus, the total spin is a good quantum number, as we have seen in the two-spin case. The
coupling constant is positive and favors an antiparallel alignment of neighboring spins. The
ground state is therefore not ferromagnetic.

8.2.2

Mean field approximation of the anti-ferromagnet

There are a few exact results for the Heisenberg model, but not even the ground state energy
can be calculated exactly (except in the case of the one-dimensional spin chain which can be
solved by means of a Bethe Ansatz). The difficulty lies predominantly in the treatment of
quantum fluctuations, i.e., the zero-point motion of coupled spins. It is easiest seen already
with two spins, where the ground state is a singlet and maximally entangled. The ground state
of all antiferromagnetic systems is a spin singlet (Stot = 0). In the so-called thermodynamic
limit (N ) there is long-ranged anti-ferromagnetic order in the ground state for dimensions
D 2. Contrarily, the fully polarized ferromagnetic state (ground state for a model with J < 0)
is known exactly, and as a state with maximal spin quantum number S 2 it features no quantum
fluctuations.
In order to describe the antiferromagnetic state anyway, we apply the mean field approximation
again. We can characterize the equilibrium state of the classical Heisenberg model (spins as
simple vectors without quantum properties) by splitting the lattice into two sublattices A and
B, where each A-site has only B-sites as neighbors, and vice-versa.5 On the A-(B-)sublattice,
the spins point up (down). This is unique up to a global spin rotations. Note, that this spin
5
Lattices which allow for such a separation are called bipartite. There are lattices, where this is not possible,
e.g., triangular or cubic face centered lattices. There, frustration phenomena appear, a further complication of
anti-ferromagnetically coupled systems.

161

configuration doubles the unit cell.

We introduce the respective mean field,

iA
m + (Sbiz m)
Sbiz =
.

z
b
m + (Si + m) i B

(8.40)

This leads to the mean field Hamiltonian

Hmf = HA + HB = Jzm

Sbiz + Jzm

iA

X
iB

m
Sbiz + Jz
N + ,
2

(8.41)

with the coordination number z, the number of nearest neighbors (z = 6 in a simple cubic
lattice). It is simple to calculate the partition sum of this Hamiltonian,

 h

iN
2
Z = tr eHmf = eJmz~/2 + eJmz~/2 eJzm /2 .
(8.42)
The free energy per spin is consequently given by
F (m, T ) =

m2
1
kB T ln Z = Jz
kB T ln (2 cosh(Jzm~/2)) .
N
2

(8.43)

At fixed temperature, we minimize the free energy with respect to m to determine the thermal
equilibrium state,6 i.e., set F/m = 0 and find


~
Jzm~
m = tanh
.
(8.44)
2
2kB T
This is the self-consistency equation of the mean field theory. It provides a critical temperature
TN (Neel temperature), below which the mean moment m is finite. For T TN , m approaches
0 continuously. Thus, TN can be found from a linearized self-consistency equation,

Jzm~2 
m=
,
(8.45)
4kB T T =TN
and thus
TN =

Jz~2
.
4kB

(8.46)

This means, that TN scales with the coupling constant and with z. The larger J and the more
neighbors are present, the more stable is the ordered state.7 For T close to TN , we can expand
the free energy in m,
"
#

 3
TN
TN
Jz
2
2
4
F (m, T ) = F0 +
1
m + 2
m .
(8.47)
2
T
3~
T
This is a Landau theory for a phase transition of second order, where a symmetry is spontaneously broken. The breaking of the symmetry (from the high-temperature phase with high
symmetry to the low-temperature phase with low symmetry) is described by the order parameter
m. The minimization of F with respect to m yields (cf. Figure 8.6)

T > TN ,

0,
m(T ) =
(8.48)
q

~ 3(T /T 1), T T .
N
N
2
6
Actually, a magnetic field pointing into the opposite direction on each site would be another equilibrium
variable (next to the temperature). We set it to zero.
7
At infinite z, the mean field approximation becomes exact.

162

m
T>T

T< T

TN

Figure 8.6: The free energy and magnetization of the anti-ferromagnet above and below TN .

8.3

Collective modes spin wave excitations

Besides its favorable properties, the mean field approximation also has a number of insufficiencies. Quantum and some part of thermal fluctuations are neglected, and the insight into the
low-energy excitations remains vague. As a matter of fact, as in the case of the ferromagnet,
collective excitations exist here. In order to investigate these, we write the Heisenberg model in
its spin components, i.e.,

X
1  b+ b b b+ 
z bz
b
HH = J
S S + Si Sj
.
(8.49)
Si Sj +
2 i j
hi,ji

In the ordered state, the moments shall be aligned along the z-axis. P
c = N f Sb on the ground
To observe the dynamics of a flipped spin, we apply the operator W
l=0 l l
state |0 i, and determine the spectrum, by solving the resulting eigenvalue equation
c | i = [H , W
c ]| i = ~ W
c | i,
(HH E0 )W
0
H
0
0

(8.50)

with the ground state energy E0 . Using the spin-commutation relations

h
i
Sbj+ , Sbj = 2Sbjz ij ,
i
h
Sbjz , Sbj = Sbj ij ,

(8.51)
(8.52)

then yields the equation

X
l

where

X0

Sbjz Sbl + J

X0
j

Sbj Sblz ~ Sbl |0 i = 0,

(8.53)

P0

runs over all neighbors of l. We decouple this complicated problem by replacing

the operators Sbz by their mean fields. Therefore, we have to distinguish between A and B
sublattices, such that we end up with two equations,
!
X
X

flA Jmz Sbl + Jm

Sbl+a
~ Sbl |0 i = 0,
l A,
(8.54)
j

lA

!
X
l0 B

flB0

Jmz Sbl0 Jm

X
a

Sbl0 +a ~ Sbl0

|0 i = 0,

l0 B.

(8.55)

Choosing
r
flA

=
r

flB0 =

2 iqrl
e
N

lA,

2 iqrl0
e
N

l0 B ,

163

(8.56)

we introduce the operators

Sbl =
Sbl0 =

2 X iqrl
b
a e
,
N q q

(8.57)

2 X b iqrl0
b e
,
N q q

(8.58)

2 X b iqrl
,
Sl e
N

(8.59)

2 X b iqrl0
Sl 0 e
,
N 0

(8.60)

with l A and l0 B, and, vice versa,

b
aq =

lA

bb =
q

l B

and insert them into the equation and obtain,

!
(Jmz ~)

X
lA

(Jmz ~)

l0 B

Sbl eiqrl

+ Jm

Sbl0 eiqrl0 Jm

iqa

Sbl0 eiqrl0

l0 B

eiqa

|0 i = 0,

(8.61)

|0 i = 0.

(8.62)

!
Sbl eiqrl

lA

From this follows that


(Jmz ~)b
aq + Jmqbbq |0 i = 0,


aq |0 i = 0,
(Jmz ~)bbq Jmq b


(8.63)
(8.64)

P
with q = a eiqa = 2(cos qx a + cos qy a + cos qz a). This eigenvalue equation is easily solved
leading to the description of spin waves in the antiferromagnet. The energy spectrum is given
by
q
~q = Jm z 2 q2 .
(8.65)
Note, that only the positive energies make sense. It is interesting to investigate the limit of
small q,
z 2 q2 z 2 q 2 + O(q 4 ),

(8.66)

~q = Jmz|q| + .

(8.67)

where

This means that, in contrast to the ferromagnet, the spin waves of the antiferromagnet have a
linear low-energy spectrum (cf. Figure 8.7). The same applies here if we expand the spectrum
around Q = (1, 1, 1)/a (folding of the Brillouin zone due to the doubling of the unit cell).
After a suitable normalization, the operators b
aq and bbq are of bosonic nature; this comes about
since, due to the mean field approximation, the Sb are bosonic as well,
l

[Sbl+ , Sbj ] = 2Sblz lj 2mlj ,

(8.68)

where the sign depends on the sublattice. The zero-point fluctuations of these bosons yield quantum fluctuations, which reduce the moment m from its mean field value. In a one-dimensional
164

boundary of reduced
Brillouin zone

hq

2a

Figure 8.7: Spectrum of the spin waves in the antiferromagnet.

spin chain these fluctuations are strong enough to suppress antiferromagnetically order even for
the ground state. The fact that the spectrum starts at zero has to do with the infinite degeneracy of the ground state. The ordered moments can be turned into any direction globally. This
property is known under the name Goldstone theorem, which tells that each ordered state that
breaks a continuous symmetry has collective excitations with arbitrary small (positive) energies.
The linear spectrum is normal for collective excitations of this kind; the quadratic spectrum of
the ferromagnet has to do with the fact that the state breaks time-inversion symmetry.
These spin excitations show the difference between a band and a Mott insulator very clearly.
While in the band insulator both charge and spin excitations have an energy gap and are inert,
the Mott insulator has only gapped charge excitation. However, the spin degrees of freedom for
a low-energy sector which can even form gapless excitations as shown just above.

165