Solid State Theory

Spring Semester 2013

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/fs13/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 B anden: 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-lling and materials properties . . . . . . . . . . . . . . . 1.6.1 Electron count and band lling . . . . . . . . . . . . . . 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 . . . . . . . . . . . . . . . . . . . . . . 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 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 8 8 8 10 11 12 15 15 16 17 19 21 21 23 24 24 27 27 28 28 35 36 36 38 39 40 42 44 44 45 46 46 47 48 51 51 52 54

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 . . . . . . . . . . . . . . . . 2

3.2

3.3

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

57 57 59 60 62 66 66 68 69 73 76 76 76 78 79 80 82 83 89 93 93 96 98 99 101 102 103 105 105 108 109 111 111 113 113 115 118 119 119 121 122 125 126 128 128 130 132 132 135 136

4 Itinerant electrons in a magnetic eld 4.1 The de Haas-van Alphen eect . . . . . . . . . . 4.1.1 Landau levels . . . . . . . . . . . . . . . . 4.1.2 Oscillatory behavior of the magnetization 4.1.3 Onsager equation . . . . . . . . . . . . . . 4.2 Quantum Hall Eect . . . . . . . . . . . . . . . . 4.2.1 Hall eect of the two-dimensional electron 4.2.2 Integer Quantum Hall Eect . . . . . . . 4.2.3 Fractional Quantum Hall Eect . . . . . . 5 Landaus Theory of Fermi Liquids 5.1 Lifetime of quasiparticles . . . . . . . . . . 5.2 Phenomenological Theory of Fermi Liquids 5.2.1 Specic 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? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

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 eect . . . . . . . . . . . . . . . 6.4 Electron-phonon interaction . . . . . . . . . . . 6.5 Electron-electron scattering . . . . . . . . . . . 6.6 Matthiessens rule and the Ioe-Regel limit . . 6.7 General transport coecients . . . . . . . . . . 6.7.1 Generalized Boltzmann equation . . . . 6.7.2 Thermoelectric eect . . . . . . . . . . . 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 . . . . . . . . . . 3

7 Magnetism in metals 7.1 Stoner instability . . . . . . . . . . . . . . . . . . . . . . . 7.1.1 Stoner model within the mean eld 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 nite wave vector Q . . . . . . . . 7.2.3 Inuence of the band structure . . . . . . . . . . . 7.3 Stoner excitations . . . . . . . . . . . . . . . . . . . . . . 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 eective Hamiltonian . . . . . . . . . . . . . . 8.2.2 Mean eld approximation of the anti-ferromagnet . 8.3 Collective modes spin wave excitations . . . . . . . . . .

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

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

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

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

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

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

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

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

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

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

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

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

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

137 138 138 139 142 143 143 146 147 149 152 153 153 154 155 157 159 159 160 162

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 subeld 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, H = He + Hn + Hne , with He =
i

(1)

p2 1 i + 2m 2
2 Pj

i= i

e2 , |r i r i | Zj Zj e2 , |R j R j | (2)

Hn =
j

2 Mj

1 2

j =j

Hne =

i,j

Zj e2 , |r i R j |

where He (Hn ) describes the dynamics of the electrons (nuclei) and their mutual interaction and Hne includes the interaction between ions and electrons. The parameters appearing are 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 ne 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 aected by quantum eects, as EPlanck = c2 c 1019 GeV, G lPlanck = G 1.6 1035 m, c3 (3)

where G = 6.673 1011 m3 kg1 s2 is the gravitational constant. This is the realm of the GUT (grand unied theory) and string theory. The goal is not to provide a better description of electrons or atomic cores, but to nd the most fundamental theory of physics. solid state physics 10 meV metals semiconductors magnets superconductors ferroelectrics ...... eective models known and established 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 dierent 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 (eective, 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 eect. 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 rst principles. Consequently, the formulation of eective (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 10 eV electrons, cores atom high-energy physics astrophysics and cosmology 1 MeV phenomenological particle physics standard model GUT string theory M-theory

most fundamental theory

Fermi liquids, the phonons connected to lattice vibrations, and magnons in ferromagnets. The idea is to treat the ground state as an eective 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 dierent 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 eect. 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 nally 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 Schr odinger 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 inuence 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 classication 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 dierent space groups1 (cf. Table 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 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 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. a b = c G.
1

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 coecients, Rn = n1 a1 + n2 a2 + n3 a3 . (1.1)

a2 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 gure. 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 = g r + a = {g |a}r , (1.2)

where g represents a rotation, reection 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 rotations, reections, inversions screw axes, glide planes {E |a}, {g |0}, {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 reection at a plane followed by a translation along the same plane. The symmetry operations {g |a}, together with the associative multiplication {g |a}{g |a } = {gg |g a + 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 |a }{g |a}1 = {E |g a } (1.4) (1.5)

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

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 groups)

point groups
Sch onies symbols

space group numbers

international tables

triclinic (2,2) monoclinic (3,13) orthorhombic (3,59) tetragonal (7,68) trigonal (5,25) hexagonal (7,27) cubic (5, 36)

C1 , C 1 C2 , Cs , C2h D2 , C2v , D2h C4 , S4 , C4h , D4 , C4v , D2d , D4h C3 , S6 , D3 , C3v , D3d C6 , C3h , C6h , D6 , C6v , D3h , D6h T, Th , O, Td , Oh

1-2 3-15 16-74 75-142 143-167 168-194 195-230

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

1.1.2

Reciprocal lattice

We dene now the reciprocal lattice which is of importance for the electron band structure and x-ray diraction on a periodic lattice. The reciprocal lattice is also perfectly periodic with a translation symmetry with a basic set {bi } dening arbitrary reciprocal lattice vectors as Gm = m1 b1 + m2 b2 + m3 b3 , where mi are integers and ai bj = 2ij , such that bi = 2 aj ak ai (aj ak ) i, j = 1, 2, 3 , bj bk . bi (bj bk ) (1.7) (1.8) (1.6)

and ai = 2

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 fG eiGr (1.10) f (r ) =
G

with the obvious property: f (r + R) = f (r ). The coecients are given by fG = 1 UC d3 rf (r )eiGr

UC

(1.11)

where the integral runs over the unit cell of the periodic lattice with the volume UC . Finally, we dene the (rst) 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 Ta on the Hilbert space commutes with the Hamiltonian H = He + Hie (purely electronic Hamiltonian He , interaction between electrons and ions Hie ), [Ta , H] = 0. (1.12)

This translation operator is dened through Ta |r = |r + a . Neglecting the interactions among electrons, which would be contained in He , we are left with a single particle problem H H0 = p2 + V (r ), 2m (1.13)

where r and p are position and momentum operators, and V (r ) =

j

Vion (r Rj ),

(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 Ta . For a Hamiltonian H0 commuting with the translation operator Ta , the eigenstates of H0 are simultaneously eigenstates of Ta . Blochs theorem states that the eigenvalues of Ta lie on the unit circle of the complex plane, which ensures that these states are extended. This means Ta (r ) = (r a) = a (r ) ,
l Tla (r ) = (r la) = Ta (r ) = l a (r )

(1.15)

with l an integer. The wave function delocalized in a periodic lattice satises,

2 2 | (r )|2 = | (r la)|2 = |l a | | (r )| ,

(1.16) (1.17)

requiring This condition is satised 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 ) . 11 (1.18) |a | = 1 a = eia .

with Ta un,k (r ) = un,k (r a) = un,k (r ), Ta n,k (r ) = n,k (r a) = eika n,k (r ), H0 n,k (r ) =

n,k n,k (r ).

(1.19) (1.20) (1.21)

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 Ta , 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 rst Brillouin zone. Blochs theorem simplies the initial problem to the so-called Bloch equation for the periodic function uk , (p + k)2 + V (r ) uk (r ) = 2m
k uk (r ),

(1.22)

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

which can be used for more complex forms of the Hamiltonian as well. There are various numerical methods which allow to compute rather eciently 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, V (r ) =
G

VG eiGr , 1 UC
UC

(1.24) (1.25)

VG =

d3 r V (r )eiGr .

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

cG eiGr ,

(1.26)

where the coecients cG = cG (k) are functions of k, in general. Inserting this Ansatz and the expansion (1.24) into the Bloch equation, (1.22), we obtain a linear eigenvalue problem for the band energies k ,
2

2m

(k G)2

cG +
G

VG G cG = 0.

(1.27)

This represents an eigenvalue problem in innite dimensions with eigenvectors cG (k) and eigenvalues k as band energies. These k include corrections to the bare parabolic dispersion, (0) 2 k2 /2m, due to the potential V (r ). Obviously, the dispersion (0) is naturally parabolic k = k 12

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 rst 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)). We illustrate here the nearly free electron method using the case of a one-dimensional lattice. 2 G2 /2m (taking G Assuming that the periodic modulation of the potential is weak, say |VG | as a characteristic wavevector in the periodic system), the problem (1.27) can be simplied. Let us start with the lowest energy values around the center of the rst Brillouin zone, i.e. k 0 (|k | /a). For the lowest energy eigenvalue we solve Eq.(1.27) by 1 for G = 0 cG (1.28) 2mVG 1 for G = 0 2 {(k G)2 k 2 } leading to the energy eigenvalue
k 2 k2

2m

[(k G=0 2m E0 =

G)2 k 2 ] |G |2
2

|VG |2

2 k2

2m

+ E0

(1.29)

with
G=0

2m

G2

(1.30)

and

1 1 = 14 m m

G=0

|G |2

(1.31)

with G = VG /{ 2 G2 /2m} (|G | 1). We observe that k is parabolic with a slightly modied (eective) 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 eective 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 . (1.32) 2 cG )2 k V (k + G
2G 2m

The eigenvalues are obtained through the secular equation, 2 )2 k (k G V2 G det 2m 2 2 V2G 2m (k + G) leading to
k

=0,
k

(1.33)

1 )2 + (k G )2 (k + G 2 2m
2

2m

)2 (k G )2 (k + G

2 + 4|V2G |

2m

2 |V | + G 2G

2m

k2 (1.34) 13

with the eective mass

where m 1. We observe a energy band gap separating two bands | + > 0 and m < 0 as |G 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 ()

1 1 1 1 2|G | m m

(1.35)

cG =

V2 G |V2G |

c+G = c+G

()

()

(1.36)

where we have chosen V2G to be real and positive. Thus, uk=0 (x) = sin Gx for
k=0,+

(1.37)
k=0,

cos Gx for

one being even and the other odd under parity operation x x. A similar analysis can be done at the boundary of the rst Brillouin zone where two energy parabolas cross. For example at k = /a we nd 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 rst Brillouin zone in one-dimension) with positive and negative eective 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 dierent reciprocal lattice points contributes to a band gap. By construction we can extend the band structure beyond the rst Brillouin zone and nd a periodic energy spectrum with
k +G

k k

(1.38) =
k

where G is a reciprocal lattice vector. Moreover, we nd in Fig.1.2 that as well as time reversal symmetry, like for free electrons.

due to parity

band gap

1st Brillouin zone 2a a 0

a 2 a

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

14

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) = for an atom located at position R, so that Ha (R) = 2 p + Va (r R) 2m (1.40)
n n (r

R) ,

(1.39)

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= 2 p + 2m Va (r Rj ) = Ha (Rj ) + VRj (r ) (1.41)

Rj

where we single out one atomic potential (the choice of Rj is arbitrary) and introduce the correction VRj (r ) = Va (r Rj ) . (1.42)
R j =R j

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.

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 n eikRj n (1.43) (r R j ) , k (r ) = N R
j

15

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

(k ) =

d3 rn k (r ) n k (r ) =

1 N

d3 reik(Rj
R j ,R j

R j ) n (r

Rj )n (r Rj )

=
Rj

d3 reikRj (r ) n (r Rj )n e i k R j n n (R j )
Rj =0

= n n +

(1.44) where due to translational invariance in the lattice we may set Rj = 0 eliminating the sum over Rj and dropping the factor 1/N . To estimate the energy we calculate, H
n n

(k) =

1 N

d3 reik(Rj
R j ,R j

Rj ) n (r

Rj ){Ha (Rj ) + VRj (r )}n (r Rj )

Rj ) n (r

= En 1

n n

(k ) +

1 N

d3 reik(Rj
R j ,R j

Rj )VRj (r )n (r R j )

= En 1

n n

(k) + En n +
Rj =0

eikRj n n (R j ) (1.45)

where En n = and n n (R j ) = d3 r n (r Rj )VRj

=0 (r )n

d3 r n (r )VRj

=0 (r )n

(r )

(1.46)

(r ) .

(1.47)

From this we can now calculate the band energies through the secular equation, det [ H
n n

(k)

n n

(k)] = 0.

(1.48)

The merit of the approach is that the tightly bound atomic orbitals have only weak overlap such that both n n (Rj ) and n n (Rj ) fall o very quickly with growing Rj . Mostly it is sucient to take Rj connecting nearest-neighbor and sometimes next-nearest-neighbor lattice sites. This is for example ne 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 rst 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 16

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 (1.49) ss (Rj ) = t Rj connects next nearest neighbors which leads immediately to
n.n. k n.n.n.

= Es + Es t

e
Rj

ikRj

eikRj
Rj

= Es + Es 2t{cos(kx a) + cos(ky a) + cos(kz a)} 4t [cos(kx a) cos(ky a) + cos(ky a) cos(kz a) + cos(kz a) cos(kx a)}

(1.50)

Note that VRj (r ) 0 in most cases due to the attractive ionic potentials. Therefore t, t > 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 eective mass
2 k

= Es + Es + 6t + 12t +

2m

k2 +

(1.51)

with

1 2 = 2 (t + 4 t ) . (1.52) m Note that t and t 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) , (1.53) 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 nd, Ex = Ey = Ez = Ep and En n = Ep n n . (1.54)

The overlaps Eq.(1.47) for nearest neighbors, Rj = (a, 0, 0) x ( bonding) t xx (Rj ) = t Rj = (0, a, 0), (0, 0, a) x ( bonding)

(1.55)

and analogous for yy and zz , while n =n . For next-nearest neighbors by symmetry n = 0, if n we obtain, Rj = (a, a, 0), (a, 0, a) t xx (Rj ) = (1.56) t Rj = (0, a, a) 17

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 = 0.1t, t = 0.05t and t = 0.15t. choose the parameters: t = 0.2t for the s-orbitals; t = 0.2t, t 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.

and analogous for yy and zz . Next-nearest neighbor coupling also allows for inter-orbital matrix elements, e.g. sign(Rnx Rny ) xy (Rj ) = yx (Rj ) = t (1.57) for Rj = (a, a, 0) and analogous for yz (Rj ) and zx (Rj ). The dierent conguration of nearest- and next-nearest-neighbor coupling is shown in Fig.1.5. Now we may setup the coupling matrix, sin(kx a) sin(ky a) 4t sin(kx a) sin(kz a) Ex (k) 4t 4t sin(kx a) sin(ky a) sin(ky a) sin(kz a) (1.58) Ey (k) 4t Hn n = 4t sin(kx a) sin(kz a) 4t sin(ky a) sin(kz a) Ez (k) with Ex (k) = Ep + Ep + 2t cos(kx a) 2t (cos(ky a) + cos(kz a)) cos(kx a)(cos(ky a) + cos(kz a)) 4t cos(ky a) cos(kz a) +4t 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.(1.48) and shown in Fig.1.4. Also in this case we may consider a k p-approximation around a symmetry point in the Brillouin zone. For the -point we nd the expansion around k = 0: 2 + B (k 2 + k 2 ) Akx Ckx ky Ckx kz y z 2 + B (k 2 + k 2 ) Ckx ky Aky Cky kz Hn (1.60) n = E + z x 2 2 2 Ckx kz Cky kz Akz + B (kx + ky ) 18 (1.59)

nearest neighbors

next- nearest neighbors

" - bonding

! - bonding

no coupling

Figure 1.5: The congurations for nearest- and next-nearest-neighbor coupling between porbitals on dierent 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.

4t , A = a2 (t + 4t ), B = a2 (t 2t + 2t ) and C = 4t . with E = Ep + Ep + 2t 4t + 4t These band energies have to be determined through the secular equations and lead to three bands with anisotropic eective masses.

1.4.4

Wannier functions

An alternative approach to the tight-binding approximation is through Wannier functions. These are dened as the Fourier transformation of the Bloch wave functions, 1 k (r ) = N eikR w(r R) 1 w(r R) = N eikR k (r )
k

(1.61)

where the Wannier function w(r R)3 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
Ambiguity of the Wannier functions: The Wannier function is not uniquely dened, because there is a gauge freedom for the Bloch function which can be multiplied by a phase factor k (r ) ei(k) k (r ) (1.62)
3

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

19

there is only one Wannier function per site. Wannier functions obey the orthogonality relation d3 r w (r R )w(r R) = 1 = N e
k ,k ikRik R

1 N

eikRik R
k ,k

d3 rk (r )k (r )

(1.63) kk = RR .

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
k

d3 r k (r )Hk (r ) =

1 N

eik(R R)
R ,R

d3 r w (r R )Hw(r R) (1.64)

=
R

eikR

d3 r w (r R)Hw(r ) ,

where we took translational invariance of the lattice into account. With the denitions
0

d3 r w (r )Hw(r ), d3 r w (r R)Hw(r ) for R = 0

(1.65) (1.66)

t(R) = the band energy can be written as

+
R

t(R)eikR .

(1.67)

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 dene 1 nk (r ) = eikR cnn (1.68) ( k ) wn (r R ) N R ,n where for all k,
cnn (k)cn n (k) = nn . n

(1.69)

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

= =

d3 rn k (r )Hnk (r ) i k R c (k)e n n (k)cnn R n ,n d3 rwn (r ) (r R)Hwn

(1.70)

=
R n ,n

i k R tn c (k)e n (R) . n n (k)cnn

20

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 dene the following Fermionic operators, c j,s creates an electron of spin s on lattice site Rj , c j,s annihilates an electron of spin s on lattice site Rj , in the corresponding Wannier states. We introduce the following Hamiltonian, H= j,s 0c j,s c
j,s

(1.71)

+
i,j

tij c j,s i,s c

(1.72)

with tij = tji real. These coecients tij are called hopping matrix elements, since c j,s i,s c 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 c j,s = N
ikRj a ks e k

and

1 c j,s = N

a k s e i k R j
k

(1.73)

where a aks ) creates (annihilates) an electron in the Bloch state with pseudo-momentum k ks ( and spin s. Inserting Eq.(1.73) into the Hamiltonian (1.72) leads to 1 1 i(kk )Ri i k R j i k R i H= + ks (1.74) e t e a a = 0 ij k a ks a N k s ks N
k,k ,s i i,j k,s

where a ks = n ks constitutes the number operator for electrons. The band energy is the same ks a as obtained above from the tight-binding approach. The Hamiltonian (1.72) 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 ) = 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 S{g|a} of an element {g |a} of the space group on a Bloch wave function k (r ) 4
4

In Dirac notation we write for the Bloch state with pseudo-momentum k as k (r ) = r |k . (1.75)

The action of the operator S{g|a} on the state |r is given by S{g|a} |r = |g r + a such that r |S{g|a} |k = k (g 1 r g 1 a). The same holds for pure translations. (1.77) and r |S{g|a} = g 1 r g 1 a|, (1.76)

21

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

(1.79)

Because {g |a} belongs to the space group of the crystal, we have [S{g|a} , H0 ] = 0. Applying a pure translation Ta = S{E |a } to this new wave function and using Eq.(1.5) Ta S{g|a} k (r ) = S{g|a} Tg1 a k (r ) = S{g|a} eik(g = S{g|a} ei(gk)a k (r ) = ei(gk)a S{g|a} k (r ), (1.80)
1 a

k (r )

the latter is found to be an eigenfunction of Ta with eigenvalue ei(gk)a . Remember, that, according to the Bloch theorem, we chose a basis {k } diagonalizing both Ta and H0 . Thus, apart from a phase factor, the action of a symmetry transformation {g |a} on the wave function5 corresponds to a rotation from k to g 1 k. S{g|a} k (r ) = {g|a} gk (r ), with |{g|a} |2 = 1, or S{g|a} |k = {g|a} |g k . in Dirac notation. Then it is easy to see that
gk 1 = g k|H0 |g k = k|S{ g |a} H0 S{g |a} |k = k|H0 |k = k.

(1.82)

(1.83)

(1.84)

Consequently, there is a star-like structure of equivalent points g k 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 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,g k ,

nk

n,k ,

nk

n,k+G .

(1.85)

Note that this denition has also implications on the sequential application of transformation operators such as S{g1 |a1 } S{g2 |a2 } k (r ) = r |S{g1 |a1 } S{g2 |a2 } |k = {g1 |a1 }1 r |S{g2 |a2 } |k (1.78) = {g2 |a2 }1 {g1 |a1 }1 r |k = k {g2 |a2 }1 {g1 |a1 }1 r . This is important in the context of Eq.(1.80). 5 Symmetry behavior of the wave function. S{g|a} k (r ) 1 = S{g|a} eikr
1 1 1 cG (k) eiGr = eik(g rg a)

cG (k) eiG(g
G

r g 1 a)

1 = ei(gk)a ei(gk)r 1 = ei(gk)a ei(gk)r

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

cg1 G (k) eiGr

G

(1.81)

where we use the fact that cG = cG (k) is a function of k with the property cg1 G (k) = cG (g k) i.e. S{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)} + 4t cos(kx a) cos(ky a). The dots correspond to the star of k with degenerate energy values, demonstrating nk = n,gk .

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, k } satisfying the eigenvalue equation, H| ; n, k = nk | ; n, k (1.86) For given k we consider the little group operations. Operating an element g of the little group on a state | ; n, k we obtain again an eigenstate of the Hamiltonian H, as S{g,0} commutes with H, S{g,0} |, nk = | , nk , nk|S{g,0} |, nk . (1.87)
=M
, (g )

We transform only within the subspace of degenerate states |, nk and the matrix M , (g ) is a representation of the group element g on the vector space of eigenstates {|, nk }. 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 [110] it is C2v containing only four elements leaving k invariant: C2v D4h Oh .

1.6

Band-lling and materials properties

Due to the fermionic character of electrons each of the band states |n, k, s 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

23

of materials. Here we would like to look at the most simple classication of materials based on independent electrons.

1.6.1

Electron count and band lling

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
N

H = t

j =1 s=,

{c j,s + c j +1,s } j +1,s c j,s c

(1.88)

where we impose the equivalence j + N = j (periodic boundary conditions). This Hamiltonian can be diagonalized by the Fourier transform 1 c j,s = N leading to H= Now we request eiRj k = ei(Rj +L)k Lk = N ak = 2n k= 2 2 n n= L a N (1.91) k,s ka k,s a
k,s

a k,s eiRj k
k

(1.89)

with

= 2t cos ka .

(1.90)

with the pseudo-momentum k within the rst Brillouin zone ( a < k < a ) and n being an integer. On the real-space lattice an electron can take 2N dierent states. Thus, for k we nd that n should take the values, n + N/2 = 1, 2, . . . , N 1, N . Note that k = /a and k = +/a dier 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 ll 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-lled band and (2) 2N electrons exhausting all possible states representing a completely lled band. In the case of half-lling we dene 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 dierence between (1) and (2) is that the former allows for many dierent 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 = 2 (1 + N/4)/N a (n = 1 + N/4), we nd the energy dierence

E =

2t

2 1 N N

(1.92)

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

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.

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 dierent k -levels whereby it has to be noticed that +/a and /a are equivalent. The condition of half-lling and complete lling are shown, where for half-lling a ground state conguration is shown (note there are 4 degenerate states). For the completely lled 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.

The bands can be either completely lled 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 identied with a Fermi energy but lies within the energy gap separating highest lled and the lowest empty band (see Fig.1.8). There is a nite 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 lled 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 lling bands. Note that we will encounter another form of an insulator, the Mott insulator, whose insulating behavior is not governed by a band structure eect, but by a correlation eect through strong Coulomb interaction. If the number of electrons per unit cell is odd, the uppermost non-empty band is half lled (see Figure 1.8). Then the system is a metal, in which electrons can move and 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 rst column of the periodic table (Li, Na, K, Rb, Cs), as all of them have the conguration [noble gas] (ns)1 , i.e., one mobile electron per ion. In general, band structures are more complex. Dierent bands need not to be separated by energy gaps, but can overlap instead. In particular, this happens, if dierent orbitals 25

insulator semiconductor E E

metal E

metal semimetal

EF EF lled k lled k

EF

lled k

Figure 1.8: Material classes according to band lling: left panel: insulator or semiconductor (partially lled 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).

are involved in the structure of the bands. In these systems, bands can have any fractional lling (not just lled or half-lled). 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 dierent atoms, A and B, where A has a small ionization energy while B has a large electron anity. 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 conguration. 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 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 26

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. 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 k (r , t) =
k

gk (k )eik ri

(1.93)

where gk (k ) 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 dened. Furthermore, the applied electric and magnetic elds have to be small enough not to induce transitions between dierent 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 elds without proof: The band index of an electron is conserved, i.e., there are no transitions between the bands. The equations of motion read6 nk , k = eE (r , t) e v n (k) H (r , t). k c = v n (k) = r (1.96) (1.97)

All electronic states have a wave vector that lies in the rst 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

(1.98)

Each state of given k and spin can be occupied only once (Pauli principle).
A plausibility argument concerning the conservation of energy leading to the equation (1.97) is given here. The time derivative of the energy (kinetic and potential) E= has to vanish, i.e., 0= n (k ) dE = v n (k ) = k e r dt k e . k (1.95)
n (k(t)) 6

e(r (t))

(1.94)

From this, equation (1.97) follows directly for the electric eld E = and the Lorentz force is allowed because the force is always perpendicular to the velocity v n .

27

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 (1.97) is not the force that acts on the electron, as the forces exerted by the periodic lattice potential is not included. The latter eect is contained implicitly through the form of the band energy (k), which governs the rst 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 eld. For illustration, consider a one-dimensional system where the band energy k = 2 cos ka leads to the solution of the semi-classical equations (1.96,1.97) = eE k eEt k= x = 2a sin eEat , (1.99) (1.100) (1.101)

in the presence of a homogenous electric eld E . It follows immediately, that the position x of the electron oscillates like x(t) = 2 cos eE eEat . (1.102)

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 eld. This eect can only be observed under very special conditions where the probe is absolutely clean. The eect is easily destroyed by damping or scattering.

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 v 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

28

follows from j = 2e
BZ

d3 k v (k)n(k) = 2e (2 )3
BZ

d3 k (k ) n(k ) , 3 (2 ) k

(1.103)

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 nite current density j , the momentum distribution n(k) has to deviate from the equilibrium Fermi-Dirac distribution in equation (1.98). It is straight forward to show that the current density vanishes for an empty band. The same holds true for a completely lled band (n(k) = 1) where equation (1.96) implies j = 2e
BZ

d3 k 1 (k) =0 (2 )3 k

(1.104)

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 lled bands can carry currents. An interesting aspect of band theory is the picture of holes. We compute the current density for a partially lled band in the framework of the semi-classical approximation, j = e
BZ

d3 k n(k)v n (k) 4 3 d3 k v (k ) 4 3
BZ BZ

(1.105) d3 k [1 n(k)]v (k) 4 3

= e = +e
BZ

(1.106)

d3 k 4 3

[1 n(k)]v (k).

(1.107)

This suggests that the current density comes either from electrons in lled 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 lled, and with electrons if the lling of an energy band is small.

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 dierent reciprocal lattice vectors are sucient 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.

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 lled 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 29

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).

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. We introduce the core electron states as |j , with H|j = Ej |j 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 |n,k = |nk |j j |n,k , (1.108)

with |n,k an orthonormal set of states. Then, n,k |j = 0 holds for all j . If we choose plane waves for the |nk , the resulting |n,k are so-called orthogonalized plane waves (OPW). The Bloch functions are superpositions of these OPW, |n,k = bk+G |n,k+G , (1.109)

where the coecients bk+G converge rapidly, such that, hopefully, only a small number of OPWs is needed for a good description. First, we again consider an arbitrary |nk and insert it into the eigenvalue equation H|nk = Enk |nk , H|nk or H|nk + (Enk Ej ) |j j |n,k = Enk |nk . (1.111)
j

H|j j |n,k = Enk |nk

|j j |n,k

(1.110)

We introduce the integral operator in real space V = j (Enk Ej )|j j |, describing a nonlocal and energy-dependent potential. With this operator we can rewrite the eigenvalue equation in the form (H + V )|n,k = (H0 + V + V )|n,k = (H + Vps )|n,k = Enk |nk . (1.112)

This is an eigenvalue equation for the so-called pseudo-wave function (or pseudo-state) |nk , instead of the Bloch state |nk , where the modied potential Vps = V + V is called pseudopotential. The attractive core potential V = V (r ) is always negative. On the other hand, 30

Enk > Ej , such that V is positive. It follows that Vps is weaker than both V and V . An arbitrary number of core states j aj |j may be added to |nk without violating the orthogonality condition (1.108). 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.

Rc

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

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 exemplied 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 vps (r) = V0
e2 Zion r

r < Rc , r > Rc ,

(1.113)

where Zion is the charge of the ionic core and Rc its eective 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 exible approaches allow for the incorporation of more experimental input into the pseudo-potential.

31

The full pseudo-potential of the lattice can be constructed from the contribution of the individual atoms, Vps (r ) =
n

vps (|r Rj |),

(1.114)

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, Vps,G = 1 d3 r Vps (r )eiGr = N d3 r vps (r )eiGr . (1.115)

For the AHA form (1.113), this is given by Vps,G = 4Zion e2 G2 cos(GRc ) + V0 2 2 (Rc G 2) cos(GRc ) + 2 2Rc G sin(GRc ) Zion e2 G . (1.116)

For small reciprocal lattice vectors, the zeroes of the trigonometric functions on the right-hand side of (1.116) 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 Vps (r ) =
n, vps (|r (Rj + R )|),

(1.117)

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,

Vps,G = =

eiGR

d3 r vps (|r |)eiGr

(1.118) (1.119)

eiGR F,G .

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

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 mun-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 cell7 (Figure 1.13). The advantage of this decomposition is that the problem can be solved using a divide-andconquer strategy. Inside the mun-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
7 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.

32

Figure 1.13: Mun-tin potential.

solutions at the boundaries. The spherically symmetric problem for |r | < rs is solved with the standard Ansatz (r ) = ul (r) Ylm (, ), r (1.120)

where (r, , ) are the spherical coordinates or r and the radial part ul (r) of the wave function obeys the dierential equation d2 + 2m dr2
2 2 l(l

+ 1) + V (r) E ul (r, E ) = 0. 2mr2

(1.121)

We dene an augmented plane wave (APW) A(k, r , E ), which is a pure plane wave with wave vector k outside the Mun-tin sphere. For this, we employ the representation of plane waves by spherical harmonics, e
i k r

= 4

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

(1.122)

l=0 m=l

where jl (x) is the l-th spherical Bessel function. We parametrize r u (r, E ) 4 il jl (krs ) s l Y (k)Ylm ( r ), r < rs , rul (rs , E ) lm UC l,m A( k , r , E ) = 4 il j (kr)Ylm (k)Ylm ( r ), r > rs , UC l,m l

(1.123)

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 (1.18) and (1.26)), k (r ) =
G

cG (k)A(k + G, r , E ),

(1.124)

where the G are reciprocal lattice vectors. The unknown coecients can be determined variationally by solving the system of equations
G

Ak (E )|H E |Ak+G (E ) cG (k) = 0, 33

(1.125)

where Ak (E )|H E |Ak (E ) = with

2 Vk,k = 4rs 2k 2k

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

(1.126)

k E 2m

j1 (|k k |rs ) |k k | +
l=0

k )j (kr )j (k r ) ul (rs , E ) . (1.127) (2l + 1)Pl (k l s l s 2m ul (rs , E )

Here, Pl (z ) is the l-th Legendre polynomial and u = du/dr. The solution of (1.125) yields the energy bands. The most dicult parts are the approximation of the crystal potential by the mun-tin potential and the computation of the matrix elements in (1.125). 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 as = ne2 , m (2.1)

where n is the density of mobile electrons, is the average time between two scattering events of the electrons, and m and e are the electronic mass and charge, respectively. In metals, n is independent of temperature, whereas the scattering time decreases with increasing temperature. Thus, determines majorly the temperature dependence of the conductivity in metals. On the other hand, insulators and semiconductors have no mobile charges at T = 0. At nite temperature, charges are induced by thermal excitations which have to overcome the band gap1 Eg between the valence and the conduction band, yielding n n0 T T0
3/2

eEg /2kB T ,

(2.3)

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 eective masses and scattering times are not the same for the valence and conduction bands.

35

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 conguration (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 1 1 1 located at (0, 0, 0) and ( 4 , 4 , 4 ) (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 = |ns + |npx + |npy + |npz , |2 = |ns + |npx |npy |npz ,

|3 = |ns |npx + |npy |npz ,

|4 = |ns |npx |npy + |npz .

(2.4)

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

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.

The band structure of these materials can be found around the -point by applying the freeelectron approximation discussed in the previous chapter. 1 is the corresponding representation to the parabolic band centered around the center of the Brillouin zone (0, 0, 0). Note that the reciprocal lattice of a face-centered cubic (fcc) lattice is body-centered cubic (bcc). The Brillouin zone of a fcc crystal is embedded in a bcc lattice as illustrated in Figure 2.3). The next multiplet with an energy of = 6 2 2 /ma2 derives from the parabolic bands emanating from the neighboring Brillouin zones, with G = G(1, 1, 1) and G = 2/a. The eight states 36

kz

X kx

L K

ky

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

are split into 1 2 15 25 . The magnitude of the resulting energies are obtained from band structure calculations, lifting the energy degeneracy 25 < 15 < 2 < 1 in the presence of a periodic potential V (r ). The essential behavior of the low-energy band structure of C and Si is shown in Figure 2.4.

Si

" L !

# X L

" !

# X

Figure 2.4: Band structure of C and Si. With two atoms per unit cell and four valence electrons, there are totally eight electrons per unit cell. It follows that the bands belonging to 1 (non-degenerate) and 25 (threefold degenerate) are completely lled. The maximum of the valence band is located at the -point and belongs to the representation 25 . Because of the existence of an energy gap between valence and conduction band, the system is not conducting at T = 0. The gap is indirect, meaning that the minimum of the conduction band and the maximum of the valence band lie at dierent points in the Brillouin zone, i.e., the gap is minimal between the -point of the valence band and some nite momentum k0 along the [100]-direction of the conduction band.2 A typical example of a
2 Energy gaps in semiconductors and insulators are said 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 Figure 2.5).

37

semiconductor with a direct gap is GaAs, composed of one element of the third and fth group, respectively. E E

direct energy gap

indirect energy gap

Figure 2.5: Illustration of direct and indirect band gaps. Next we list some facts about these materials of the group IV: Carbon has an energy gap of around 5.5eV in the diamond structure. Thus, in this conguration 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 has a direct energy gap of 1.43eV .

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 H= where
V,k V,k cV,k,s cV,k,s k,s

+
k,s

C,k cC,k,s cC,k,s ,

(2.5)

and

The operator (annihilates) an electron with (pseudo-)momentum k and spin s in the band n, n {V, C }. In the ground state |0 , |0 =
k,s

C,k are the band cnks (cnks ) creates

energies of the valence band and conduction band, respectively.

c V,k,s |0 ,

(2.6)

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

(2.7)

38

The total momentum and spin of the ground state vanish. We next consider single electron excitations from that ground state.3

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, s = c C,k+q ,s cV,k,s |0 , (2.12)

where we remove an electron with pseudo-momentum k and spin s 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 s to s and of shifting the wave vector of conduction electrons by q is included. Furthermore |k + q , s; k, s is assumed to be normalized. The electron-hole pair may either be in a spin-singlet (pure charge excitation s = s ) or a spin-triplet state (spin excitation s = s ). Apart from spin, the state is characterized by the wave vectors k and q . The excitation energy is given by E=
C,k+q

V,k .

(2.13)

The spectrum of the electron-hole excitations with xed q is determined by the spectral function I (q , E ) =
k,s,s 2 | k + q , s; k, s |c C,k+q ,s cV,ks |0 | (E ( C,k+q

V,k )).

(2.14)

Excitations exist for all pairs E and q for which I (q , E ) does not vanish, thus, only above a q dependent threshold, which is minimal for q = k0 , where k0 = 0 (k0 = 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 2.6). 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.
3 Where is the chemical potential? We assume a simple band structure for a direct-gap semiconductor based on the k p approximation: 2 2 2 2 k k and (2.8) C,k = Eg + V,k = 2mV 2mC Now we calculate the electron distribution. The overall electron density is n:

n=
kBZ s

1=2
BZ

d3 k 1. (2 )3

(2.9)

The density of electrons in the conduction and valence band for small nite temperatue are given by nC =2
BZ

d3 k nF ( (2 )3

C,k )

=
BZ

d3 k 4 3 e (

1
C,k )

+1

BZ

d3 k ( e 4 3

C,k )

e (Eg ) (2mC kB T )3/2 , 4 3 =

e (2mV kB T )3/2 . 4 3 +1 BZ BZ BZ (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 At T = 0 the chemical potential is exactly in the center of the band gap and then moves with increasing T , if the eective masses are dierent. 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. n nV = d3 k (1 nF ( 4 3
V,k ))

d3 k 4 3 e (

V,k )

d3 k ( e 4 3

V,k )

39

E continuum

E q k0 Figure 2.6: Electron-hole excitation spectrum for k0 = 0. Excitations exist in the shaded region, where I (q, E ) = 0.

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, V =
s,s d3 r d3 r s (r )s (r )

e2 (r )s (r ), |r r | s

(2.15)

where the eld operators are dened by 1 s (r ) = n=V,C un,k (r )eikr cn,ks ,
k

(2.16)

where un,k (r ) are the Bloch functions of the band n = C, V . Now, we consider a general particle-hole state, |q = A(k)c C,k+q ,s cV,k,s |0 = A(k)|k + q , s; k, s , (2.17)

and demand that it satises the stationary Schr odinger equation (H + V )|q = (E0 + Eq )|q . This two-body problem can be expressed as k + q , s; k, s |H + V |k + q , s; k , s A(k ) = (E0 + Eq )A(k). (2.18)

The matrix elements are given by k + q , s; k, s |H|k + q , s; k , s = k,k {E0 + and k + q , s; k, s |V |k + q , s; k , s = 2ss 2 1 2
C,k+q

V,k }

(2.19)

iq (r r ) d3 r d3 r u C,k+q (r )uV,k (r )uC,k +q (r )uV,k (r )e d3 r d3 r u C,k+q (r )uV,k (r )uC,k +q (r )uV,k

e2 |r r | e2 (r )ei(k k)(rr ) , (2.20) |r r |

The rst 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 u n,k (r )un,k (r ) 1 d3 ru n,k (r )un,k (r ) = 40 1 u |u 1, n,k n,k (2.21)

which is reasonable for k k (= k + q ). In the same manner, we see that u C k+q (r )uV,k (r ) 1 1 uC,k+q |uV,k u |u = 0. C,k V,k (2.22)

Note that the semiconductor is a dielectric medium with a dielectric constant (D = E ). Classical electrodynamics states that E = 4 , (2.23)

Thus, we can write the stationary equation (2.18) as

C,k+q

i.e., the Coulomb potential is partially screened due to dielectric polarization. Including this eect in the Schr odinger equation phenomenologically, the matrix element (2.20) takes on the form 4e2 k + q , s; k, s |V |k + q , s; k , s = . (2.24) |k k |2 1 4e2 A(k ) = 0. |k k |2

V,k

E A(k)

(2.25)

We include the band structure using the k p - approximation which, for a direct energy gap, leads to 2 k2 2 k2 = + E and = . (2.26) C,k g V,k 2mC 2mV For simplicity we assume now that mV = mC = m , We dene a so-called envelope function F (r ) by 1 A(k)eikr . (2.27) F (r ) = k This function satises the dierential equation 2ex
2 2

e2 F (r ) = |r |

E Eg

2q2

2Mex

F (r ),

(2.28)

1 = 2/m and total mass M = 2m . where ex is the reduced mass, i.e., ex ex The stationary equation (2.28) is equivalent to the Schr odinger equation of a hydrogen atom. The energy levels then are given by

Eq = Eg

2q2 ex e4 + , 22 2 n2 2Mex

(2.29)

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-dened 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 (2.30)

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 41

E electron-hole continuum excitons

q Figure 2.7: Qualitative form of the exciton spectrum below the electron-hole continuum.

the excitons are inside the gap, as shown schematically in Figure 2.7. 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 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 s . If s = s the exciton is a spin singlet, while for s = s 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 q= e 2 2 2 = 2 = c hc hc a a 2 , a (2.31)

where c denotes the speed of light, a the lattice constant, and 1/137 the ne 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 ( ) (2.32) ( 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 uV,k |r |uC,k are nite 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 42

transfer, as Q with Q = cs |Q| and the sound velocity cs phonon assisting in the transition reduces the transition rate to

c. The requirement of a (2.33)

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 2.8). E Q E Q Phonon k Photon Eg Photon k Eg

abs ( ) c+ ( + Q Eg )2 + c ( Q Eg )2 ,

Phonon

E = + Q

E = Q

Figure 2.8: Phonon-assisted photon absorption in a semiconductor with indirect gap: phonon absorption (left panel) and phonon emission (right panel). 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 2.9, we show the situation for a direct-gap semiconductor. excitons n=1 n=2 n=3 electron-hole continuum

Eg

Figure 2.9: 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 . 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 43
/kB T

(2.34)

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 congurations are given by Al : P: [(1s)2 (2s)2 (2p)6 ] (3s)2 (3p), [(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 the conduction band by a single isotropic band with eective mass mC ,
k

2 k2

2mC

+ Eg .

(2.35)

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 2mC
2 2

e2 | r |

F (r ) = EF (r ),

(2.36)

which is nothing else than the static Schr odinger 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 mC = 2 Ry, 2 2 2 2 n m n2 (2.37)

for n N and the eective radius (corresponding to the renormalized Bohr radius in the material) of the lowest bound state reads r1 =
2

m e2

m a , mC B

(2.38)

where aB = 0.53 A is the Bohr radius for the hydrogen atom. For Si we nd mC 0.2m and 12, such that E1 20meV and r1 30 A. (2.40) (2.39)

Thus, the resulting states are weakly bound, with energies inside the band gap. We conclude that the net eect of the P-impurities is to introduce additional electrons into the crystal, whose 44

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 2.10). conduction band valence band no doping conduction band impurity levels valence band n-doped impurity levels valence band p-doped conduction band

Figure 2.10: Position of the chemical potential in semiconductors. 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 dierently 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.

2.3.2

Carrier concentration

Let us briey 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 , (2.41)

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, ne nh = n2 0 T T0
2 3

eEg /kB T = n2 (T ),

(2.42)

where T0 , n0 and Eg are parameters specic to the semiconductor. In the case of undoped silicon at T = 300K, ne nh 1020 cm3 . Thus, for the undoped semiconductor we nd 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 ne nD and nh = n2 (T ) 103 cm3 . ne 45 (2.44) 1017 cm3 (2.43)

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 rectiers.4 When contacting the two types of doped 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 2.11). 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 2.11: Occupation of the impurity levels of a pn-junction. In the absence of a voltage U over the junction, the net current ow 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 diusion current Jdi . From the point of view of the electrons, the dipole eld 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 diusion current Jdi from the n-side to the p-side. The diusion current is directed against the potential gradient, so that the diusing electrons have to overcome a potential step. The equilibrium condition for U = 0 is given by 0 = Jtot (U = 0) = Jdi + Jdrift C1 (T )eEg /kB T C2 (T )eEg /kB T = 0, (2.45)

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 signicantly. For the diusion current however, the factor
4

dt. Gleichrichter

46

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 modied according to Eb Eg eU , where eU = n p is the dierence 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 . (2.46)

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 2.12) shows a clearly asymmetric behavior, which can be used to rectify ac-currents. Rectiers (or diodes) are an important component of many integrated circuits.
J
eU eU

eU

eU

reverse bias

forward bias U

p-doped

n-doped

p-doped

n-doped

Figure 2.12: 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-dened 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 dierent semiconductors with dierent 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 eciency 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 dierence in refractive indices inside nSC 3 and outside nair 1 the device leads to large reective losses. Thus, the ecacy of diode light sources, dened as the number of photons emitted per created particle-hole pair, is small, but still larger than the eciency 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 47

leads to an excess of hole carriers (minority charge carriers). Some of these holes will diuse towards the pn-interface and will be drawn to the p-side by the dipole eld. In this way, they induce an additional current JL modifying the current-voltage characteristics to Jtot = Jpn JL = Js (eeU/kB T 1) JL . (2.47)

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 UL = kB T ln e JL +1 . Js (2.48)

The maximum eciency is reached by applying an external voltage Uc < UL such that the product Jc Uc is maximized, where Jc = Jtot (U = Uc ) (cf. Figure 2.13). contacts non-reecting layer

J U UL

n p JL

maximal power rectangle

Figure 2.13: Solar cell design and shifted current-voltage characteristics. The eciency 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 dierent architectures. Here we shortly introduce the MOSFET (Metal-OxideSemiconductor-Field-Eect-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 2.14): 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. p-type Si source n-type Si
SiO2 , insulator

z x, y

metal gate

drain

n-type Si

Figure 2.14: 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 ow between the electrode and the semiconductor when a gate voltage is applied. The switchable currents in the MOSFET ow between 48

p-type Si conduction band SiO2 d valence band depleted layer UG small z=0 z z=0 SiO2 inversion layer d z

p-type Si conduction band

valence band depleted layer UG large

Figure 2.15: Depletion layer at the SiO2 -Si interface for 0 < e UG < Eg (left panel) and the inversion layer Eg < e UG (right panel).

the source and the drain which are heavily n-doped semiconductor regions. Depending on the applied gate voltage UG three dierent regimes can be realized: 1. UG = 0 Virtually no current ows, 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 lled 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 potential (z ), where z is the distance from the boundary between SiO2 and Si. This potential (z ) satises the simple one-dimensional Poisson equation d2 4(z ) (z ) = , 2 dz where the charge density originates in the occupied acceptor levels, (z ) = enA , z < d, 0, z > d, (2.50) (2.49)

and nA is the density of acceptors. The boundary conditions are given by (z = 0) = UG The solution for 0 z d then reads (z ) = 2enA (z d)2 , with d2 = UG . 2enA (2.52) and (z = d) = 0. (2.51)

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

49

3. 1 <

e UG Eg

When the applied gate voltage is suciently large, a so-called inversion layer is created (cf. Figure 2.15). 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 eect (QHE), which is characterized by highly unusual charge transport properties in the presence of a large magnetic eld, 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 lled 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 lled. The fundamental dierence 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 conguration consists of closed shell cores and one single valence electron in an ns-orbital. Neglecting the core electrons (completely lled bands), we consider the valence electrons only and apply the approximation of nearly free electrons. The lowest band around the -point is then half-lled. First, we will also neglect the inuence 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 simplication 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 ) such that the reciprocal space is discretized as k=
1

(3.2)

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

(3.3)

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

51

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 lling 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 |0 =
s

|k|kF

c k,s |0

(3.4)

where the operators c k,s (ck,s ) create (annihilate) an electron with wave vector k and spin s. 2 /2m is determined The Fermi wave vector kF with the corresponding Fermi energy F = 2 kF by equating the lled electronic states with the electron density n. We have n= which results in kF = (3 2 n)1/3 . Note kF is the radius of the Fermi sphere in k -space around k = 0.2 (3.6) 1 1=2
|k|kF ,s 3 d3 k 4 kF 1 = 2 , (2 )3 3 (2 )3

(3.5)

3.1.1

Theory of metals - Sommerfeld and Pauli

In a rst 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) = and the density of states N (E ) =
k,s

1 e(
k )/kB T

+1

(3.7)

(E 2m
2

k)

= 2

2 k2 d3 k E (2 )3 2m 1/2

= 3 4

ddk k

m k 2k

2mE

2 2

3/2

E 1/2 =

(3.8) for E > 0 with N = n as the total number of electrons. We rst address the temperature dependence of the chemical potential up second power in T for xed electron number N , by using the equation N=
k,s

3N 2 F

E
F

f ( k) =

+ 0

dE f (E )N (E ) =
0

dEN (E ) +

2 (k T )2 N () + , 6 B
F /kB . 3

(3.9)

where we used the Sommerfeld expansion assuming T

0
2 3 F

TF = (
F)

We now use
F )N

dEN (E )

dEN (E ) + (

F )N

= N + (

F)

(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
+

dE g (E )

f (E ) E

=
+

dE

g () + (E )g () + f (E ) E

( E ) 2 g ( ) + 2

f (E ) E (3.10)

g ( ) = g ( ) + 2

dE (E )

2 + = g ( ) + g ()(kB T )2 + 6

52

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

F) + F.

2 (k T )2 N ( ) 6 B

(T ) =

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

F )/N ( F ) 0
F

= 1/2

Now we also determine the internal energy

U (T ) =

dE EN (E )f (E ) dE EN (E ) +
F

dE EN (E ) +
0

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

F )N (

F) +

2 (k T )2 N ( 6 B

2 (k T )2 N ( 6 B

F)

=0

= U0 +

2 (k T )2 N ( 6 B

F)

, (3.16) 1 U 2 2 = k TN( T 3 B

where we used Eq.3.15. The specic heat is then given by C=

F)

= T

(3.17)

and shows a T -linear behavior where is the Sommerfeld coecient, proportional to the density of states at the Fermi energy. Next we consider the eect of a magnetic eld 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, M = B (N+ N ) = B 2
0

B 2

dEN (E )f (E B H ) 2B H 2 B HN (
F)

0 0

dEN (E )f (E + B H ) f (E ) E = 2 B HN (
F)

dEN (E )

f (E ) E

dE

(3.18)

By taking the derivative with respect to H we nd for the susceptibility, p = M = 2 BN ( H

F)

(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
+

dE

g (E ) E
+

f ( E ) = g ( ) +

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

(3.11)

With the denition

E

g (E ) =

dE (E )

dE (E )f (E ) =

dE (E ) +

2 (kB T )2 () + . 6

(3.12)

Note that

dx
4

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

(3.13)

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

M 2 BH
0

dEN (E )

f (E ) E

= 2 BH

N ( ) +

2 ( k B T ) 2 N ( ) 6

(3.20)

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 from Eq.(3.4) for given kF . The Hamiltonian splits into four terms H = Hkin + Hee + Hei + Hii with Hkin = Hee =
k cks cks k,s

(3.23)

(3.24) e2 (r )s (r ) |r r | s (3.25) (3.26) (3.27)

1 2

d3 r d3 r s (r )s (r ) s,s

Hei = Hii = 1 2

d3 r d3 r
s

d3 r d3 r

n2 e 2 , |r r |

ne2 (r )s (r ) |r r | s

where we have used in second quantization language the electron eld operators 1 s (r ) = 1 s (r ) =
i k r c k,s e k

(3.28) (3.29)

ck,s eikr
k

The variational energy which we want to minimize with respect to n can be computed from Eg = 0 |H|0 and consists of four dierent contributions: First we have the kinetic energy Ekin = 0 |Hkin |0 = = 2 d3 k (2 )3
k k k,s

0 |c ks cks |0 = nk s 3 5

(3.30)

nk s = N

(3.31)

where we used N = n the number of valence electrons and 1 |k| kF nk s = . 0 |k| > kF
With Eq.(3.15) we write N ( ) N which leads to M 2 B HN (
F) F

(3.32)

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

N(

F)

N ( 2 (kB T )2 6 N(

2 F) F)

(3.21)

1+

N ( F) N( F )

N ( F) N( F )

= p (T )H ,

(3.22)

and denes 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, Eee = = 1 2 1 2 d3 r d3 r d3 r d3 r e2 |r r |
0 | s (r )s (r )s (r )s (r )|0

(3.33) (3.34)

s,s

e2 n2 G(r r ) = EHartree + EFock . |r r |

For this contribution we used the fact, that the two-particle correlation function from equation (3.33) may be expressed5 as
2 0 | s (r )s (r )s (r )s (r )|0 = n G(r r )

(3.42)

s,s

where G(r ) = 9n2 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, Eei = =
5

d3 r d3 r

e2 d3 r d3 r n2 . |r r |
1 2 ei(kk
k,k ,q ,q )r i(q q )r

e2 n |r r |

0 | s (r )s (r )|0

(3.44) (3.45)

We shortly sketch the derivation of the pair correlation function. Using equations (3.28) and (3.29) we nd
0 | s (r )s (r )s (r )s (r )|0 =

0 |c k s cq s cq

c k s | 0 .

(3.35)

We distinguish two cases: First, consider s = s ,

0 |c ks cq s cq s

ck s |0 = kk qq nks nqs 1 2 nks nq,s =

k,q

(3.36) n2 . 4

leading to
0 | s (r )s (r )s (r )s (r )|0 =

(3.37)

Secondly, assume s = s such that

0 |c ks cq s cq s ck s |0 = (kk qq kq qk )nks nq s ,

(3.38)

which in turn leads to

0 | s (r )s (r )s (r )s (r )|0 =

1 2

1 ei(qk)(rr
k ,q

nks nq,s .

(3.39)

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

2

2 d3 k ikr e (2 )3 (3.40)

G(r ) = 2

eikr nks
k

= 2
|k|kF

kF

2 dk k sin kr = 2 1 sin kF r kF r cos kF r 2 2 r3

1 = 2 2 2 r
0 3 and n = kF /3 2 (k = |k| and r = |r |).

(3.41)

55

Were the expectation value 0 | s (r )s (r )|0 corresponds to the uniform density n, as is easily calculated from the denitions (3.28) and (3.29). Finally we have the repulsive ion-ion interaction Eii = 0 |Hii |0 = 1 2 d3 r d3 r n2 e 2 . |r r | (3.46)

n 2 G(rr)

n2

n2/2

k F (rr)

Figure 3.1: Pair correlation function (schematic graph).

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 eect. The remaining terms are the kinetic energy and the Fock term. The latter is negative and reads EFock 9n2 = 4 e2 d r |r |
3

sin kF |r | kF |r | cos kF |r | (kF |r |)3 2.21 0.916 2 rs rs

= N

3e2 k . 4 F

(3.47)

Eventually, the total energy per electron is given by Eg N =

2 3 2 kF 3e2 k = 5 2m 4 F

Ry

(3.48)

where 1Ry = e2 /2aB and the dimensionless quantity rs is dened via n= and rs = d = aB 9 4
1/3

3 4d3

(3.49)

me2 . 2k F

(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. 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. 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 conguration and, thus, relatively inert. In the variational approach outlined above correlation eects among the electrons due to the Coulomb repulsion 56

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 + . . . 2 rs rs Ry (3.51)

which can be obtained from a more sophisticated quantum eld 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 (r) er/l r (3.52)

i.e. the Coulomb potential is modied into the short-ranged Yukawa potential with screening length l. In contrast to metals, the nite 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 (r) 1 . r (3.53)

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

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 H = Hkin + HV =
k cks cks k,s

+
s

d3 r Va (r , t) s (r )s (r )

(3.54)

where the second term is considered as a small perturbation. In a rst 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)

57

Using equations (3.28) and (3.29) we obtain for the density operator in momentum space, q =
s i q r d3 r = s (r )s (r )e

c ks ck+q s =
k,s

k,q,s ,
k,s

(3.57)

where we dene k,q,s = c ks ck+q s . The perturbation term HV now reads HV =

it t d3 r Va (r , t) e = s (r )s (r ) = q Va (q , )e s

it t e . k,q ,s Va (q , )e k,s

(3.58)
with k,q ,s = ck+q s cks . The density operator q (t) in Heisenberg representation is the relevant quantity needed to describe the electron density in the metal. We introduce the equation of motion for k,q,s (t):

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

k+q

(3.59) (3.60)

it t k,q,s + c e . ks cks ck+q s ck+q s Va (q , )e

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

k,q,s + n0k,s n0k+q,s Va (q , )

(3.61)

where n0k,s = c ks cks and, therefore, nind (q , ) = 1 k,q,s =

k,s

n0k+q,s n0k,s
k,s k +q

Va (q , ).

(3.62)

With this, we dene the dynamical linear response function as 0 ( q , ) = 1 n0k+q,s n0k,s
k,s k +q

(3.63)

such that nind (q , ) = 0 (q , )Va (q , ), where 0 (q , ) is known to be the Lindhard function. So far we treated the linear response of the system to an external perturbation without considering feedback eects due to the interaction among electrons. In fact, the density uctuation 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) or in Fourier space Vn (q , ) = 4e2 n(q , ). q2 (3.65) (3.64)

If we allow feedback eects in our system with external perturbation Va (q , ), the eective potential V felt by the electrons is determined self-consistently via V (q , ) = Va (q , ) + Vn (q , ) = Va (q , ) + 58 4e2 q2 n(q , ), (3.66) (3.67)

where n(q , ) = 0 (q , )V (q , ). The relation between V and Va may then be written as V (q , ) = with ( q , ) = 1 4e2 (q , ), q2 0 (3.70) V a (q , ) (q , ) (3.69) (3.68)

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 dene the response function (q , ) within random phase approximation (q , ) = 0 ( q , ) = (q , ) 0 (q , ) . 4e2 1 2 0 ( q , ) q
6

(3.71) to be (3.73)

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

3.2.2

Electron-hole excitation

For simple particle-hole excitations in metals, neglecting Coulomb interaction between the electrons, it is sucient 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
0+

lim

1 =P z i

1 z

+ i (z )

(3.74)

where the Cauchy principal value P of the rst term has to be taken, we separate the Lindhard function (3.63) into 01 (q , ) = 1 P n0,k+q n0,k
k+q

k,s

k+q

(3.75) ) (3.76)

1 02 (q , ) =
6

k,s

(n0,k+q n0,k ) (

The equation (3.73) can be written in the form of a geometric series, (q , ) = 0 (q , ) 1 + 4e2 0 (q , ) + q2 4e2 0 (q , ) q2
2

+ .

(3.72)

From the point of view of perturbation theory, this series corresponds to summing a limited subset of perturbative terms to innite order. This approximation is called Random Phase Approximation (RPA) and is based on the assumption the phase relation between dierent 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

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 .

k
FermiSee FermiSee

k+q

Figure 3.2: Schematic view of an particle-hole pair creation (electron-hole excitation). The relevant excitations originating from the Lindhard function are particle-hole excitations. Starting from the ground state of a completely lled 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 3.2). The energy dierence is then given by Ek,q =
k +q

> 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 3.3. Note the absence of an energy gap for excitations.

Plasmaresonanz

2k F

Figure 3.3: Excitation spectrum in the -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.

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 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 nite frequency for small momenta q . We derive it here using the response function
7

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

60

(q , ). Assuming |q |

kF we expand 0 (q , ) in q , starting with

k+q

+ q k

n0,k+q = n0,k +

n0 q k

+
k

(3.78) + (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 0 (q , ) 2 = 2 (2 )2 d3 k q v F ( k ) (2 )3 q v F i
+1

d cos
1

2 kF vF

qvF cos + + i

qvF cos + i

qvF cos + i

(3.80) (3.81) (3.82)

3 q2 2 q2 kF 3 vF 1 + 3 2 m( + i )2 5 ( + i )2 2 q2 n0 q 2 3 vF = 1 + . m( + i )2 5 ( + i )2

According to equation (3.70), we nd lim (q , ) = 1

2 p

|q |0

(3.83)

for the dielectric function in the long wavelength limit (|q | 0), with
2 p =

4e2 n0 . m

(3.84)

We now use the result in Eq.(3.82) to approximate (q , ), (q , ) = where we introduced R(q, )2 = 1+

2 q2 3vF 5 2

n0 q 2 R(q, )2 2 m( + i )2 4e2 n2 0 R(q, ) n0 q 2 R(q, ) 2mp 1 1 + i p R(q, ) + i + p R(q, )

(3.85) (3.86)

(3.87)

Applying the relation (3.74) in Eq.(3.82) we obtain Im (q , ) n0 q 2 R(q, p ) p ( p R(q, p )) ( + p R(q, p )) (3.88)

which yields a sharp excitation mode, (q ) = p R(q, p ) = p 1+

2 q2 3vF + 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-dened energy-momentum relation and may consequently be viewed as a quasiparticle (plasmon) which has bosonic character. When the plasmon dispersion 61

metal Li Na K Mg Al

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

(theo) [eV] p 8.5 6.2 4.6 -

Table 3.1: Experimental values of the plasma frequency for dierent 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.

merges with the electron-hole continuum it is damped (Landau damping) because of the allowed decay into electron-hole excitations. This results in a nite life-time of the plasmons within the electron-hole continuum corresponding to a nite width of the resonance of the collective excitation. 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 eld E = 4 P which acts as a restoring force. The equation of motion for an individual electron describes harmonic oscillations d2 m 2 r = eE = 4e2 n0 r . (3.90) dt with the same oscillation frequency as in Eq.(3.84), the plasma frequency, 4e2 n0 . (3.91) m Classically, the plasma resonance can therefore be thought as an oscillation of the whole electron gas cloud on top of a positively charged background.
2 = p

+
Figure 3.4: 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 eld ( 0). Using the expansion (3.79) we obtain 0 (q , 0) = 1 (
k,s k

)= F

2 3n 1 kF = 0 2 vF 2 F

(3.92)

62

and thus (q , 0) = 1 +
2 kT F q2

(3.93)

2 = 6e2 n / . The eect of the renormalized with the so-called Thomas-Fermi wave vector kT F 0 F 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 ) = or in real space V (r ) =

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

(3.94)

e2 kT F r e . r

(3.95)

The potential is screened by a rearrangement of the electrons and this turns the long-ranged 1 , Coulomb potential into a Yukawa potential with exponential decay. The new length scale is kT 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 elds 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. Friedel oscillations The static dielectric function can be evaluated exactly for a system of free electrons, resulting for 3 dimensions in (q , 0) = 1 +
8

4e2 mkF q 2

2 q2 2kF + q 1 4kF + ln 2 8kF q 2kF q

(3.102)

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 ( with ne (
F F

+ e(r)) ne (

(3.96)

)=

3 kF 1 = 3 2 3 2

(2m

)3/2

(3.97)

2 where F = 2 kF /2m. This approach is justied, if the spacial change of the potential (r ) is slow compared to 1 kF , so that locally we may describe the electron gas as a lled Fermi sphere of corresponding electron density. The Poisson equation may now be formulated as

2 (r ) = 4 [ind (r ) + ex (r )] 4 e2 (r )
2 = kT F (r ) 4ex (r )

ne ( )

ex (r )
= F

(3.98) (3.99)

with the Thomas-Fermi momentum kT F dened as,

2 2 kT F = 4e

ne ( )

=
= F

6e2 ne
F

(3.100)

and ne = ne (

). 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.

63

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 n(r ) = with g (q ) = q (q ) 1 . 2 2 (q ) (3.105) d3 q (2 )3 1 1 1 na (q )eiqr = ( q ) r

g (q )na (q ) sin qr dq
0

(3.104)

Note that g (q ) vanishes for both q 0 and q . Using partial integration twice, we nd n 0 n(r ) = a r3 where g (q ) A ln|q 2kF | and g (q ) dominate around q 2kF . Hence, for kF r Ana0 n(r) r3
2kF +

g (q ) sin qrdq
0

(3.106)

(3.107)

A q 2kF 1,

(3.108)

2kF

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

(3.109)

Ana0

cos 2kF r . r3

(3.110)

with a cuto . 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 . Q = en = e d3 r n(r ) = lim 1 1 na (q ) = ena0 (q ) (3.111)

q 0

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 nd a perfect compensation which corresponds to perfect screening.
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) (3.103)

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

64

ni

standard ThomasFermi

Lindhard form
(Friedel oscillation)

r
Figure 3.5: 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 s2 4 1/2 1 (3.112) 0 (q, = 0) = 1 1 2 (s 2) , 2D 2 s k s 4 s+2 1 2 ln , 3D F2 1 2 4 s s2 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.

(q ,0) (0,0)
1

1D

2D 3D

0 0

2k F

Figure 3.6: Lindhard functions for dierent dimensions. The lower the dimension the stronger the singularity at q = 2kF .

65

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 sucient to obtain some essential features of the interaction between lattice vibrations and electrons. In particular, renormalized screening eects 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 innitesimal volume element d3 r around a point r to a dierent point r (r ). We can introduce here the so-called displacement eld u(r ) = r (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 Eel = 2 d3 r ( u(r , t))2 (3.113)

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

(3.115)

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, 2 c2 s t (3.116)

which is a wave equation with sound velocity c2 s = /0 . The resulting displacement eld can be expanded into normal modes, 1 u(r , t) = where every qk (t) satises the equation d2 2 q + k qk = 0, dt2 k
10

ek qk (t)eikr + qk (t) eikr

k

(3.117)

(3.118)

Note that the most general form of the elastic energy of an isotropic medium takes the form Eel = d3 r
, =x,y,z

( u )( u ) + ( u )( u ) , 2

(3.114)

where = /r . The Lam e coecients 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

with the frequency k = cs |k| = cs k and the polarization vector ek has unit length. Note that within our simplication for the elastic energy (3.114), all modes correspond to longitudinal waves, i.e. u(r , t) = 0 and ek k. The total energy expressed in terms of the normal modes reads E=
k 2 0 k [qk (t)qk (t) + qk (t)qk (t)] .

(3.119)

Next, we switch from a Lagrangian to a Hamiltonian description by dening the new variables Qk = Pk =
0 (qk + qk )

(3.120) (3.121)

d Q = ik 0 (qk qk ) dt k

in terms of which the energy is given by E= 1 2

2 2 2 Pk + k Qk . k

(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 dening the canonical conjugate operators Pk Pk and Qk Qk which obey, by denition, the commutation relation, [Qk , Pk ] = i k,k . (3.123)

As it is usually done for quantum harmonic oscillators, we dene the raising and lowering operators bk = b k = satisfying the commutation relations ] = k ,k , [bk , b k [bk , bk ] = 0,
[b k , bk ] = 0.

1 k Qk + iPk 2 k 1 k Qk iPk , 2 k

(3.124) (3.125)

(3.126) (3.127) (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 H= k b k bk +
k

1 2

(3.129)

In analogy to the treatment of the electrons in second quantization we say that the operators b k (bk ) create (annihilate) a phonon, a quasiparticle with well-dened energy-momentum relation, k = cs |k|. Using Eqs.(3.118, 3.121, and 3.125) the displacement eld operator u(r ) can now be dened as 1 u(r ) = ek
k ikr bk eikr + b . ke

20 k

(3.130)

67

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 dene 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 eect 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-dened collective plasma excitation at the ionic plasma frequency 2 p = 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 eects of the electrons that react nearly instantaneously to the slow ionic motion, due to their much smaller mass. The nite 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
2 k =

2 p (k, 0)

k 2 2 p
2 k 2 + kT F

(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 nally recovered, and the renormalized velocity of sound cs reads c2 s 2 p k2
TF

2 Zmp

Mi

k2

TF

1 m 2 = Z v . 3 Mi F

(3.133)

For the comparison of the energy scales we nd, D c s = TF vF where we used kB TF =

11 F

1 m Z 3 Mi

1,

(3.134)

and kD kF .

See course of Statistical Physics HS09.

68

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

2 p (k, 0)

(3.135)

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 nd k 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 eect 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 eects 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 eld 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 , where contributions of the isolated electronic and ionic systems are included in Hisol =
2 k2 k,s 2

(3.137)

2m

c k s ck s +

dx

du (x) dx

(3.138)

whereas the interactions between the system comes in via the coupling Hint = n0 dx dx V (x x ) d u(x) s (x )s (x ) 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 uctuations of the positively charged background12 mediated by the screened Coulomb interaction V (x x ). 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 N=
s

L 2

+kF

dk 1 = 2
kF

L 2k 2 F

(3.140)

leading to kF = n . 2 0 (3.141)

12 Note that only phonon modes with a nite value of u couple in lowest order to the electrons. This is only possible of longitudinal modes. Transverse modes are dened by the condition u = 0 and do not couple to electrons in lowest order.

69

Perturbative approach - instability Now we consider the Kohn anomaly of this system. For a small background modulation u(x) = 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 =
2 k2 k,s

2m

c k s ck s +

0 2

2 q uq uq

(3.142) (3.143)

Hint = i

k,q,s

u c q V q q k+q,s ck,s Vq uq ck,s ck+q,s ,

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

1 u(x) = L 1 V (x) = L

uq eiqx ,
q

(3.144) (3.145)

eiqx , V q
q

= 4e2 /q 2 (q, 0) const. - we consider an eectively short-ranged potential. We with V q compute the second order correction to the ground state energy using Rayleigh-Schr odinger perturbation theory (note that the linear energy shift vanishes) E
(2)

=
k,q,s

|2 u u q |V q q q
2

2 2 | 0 |c k,s ck+q,s |n | + | 0 |ck+q,s ck,s |n | n

E0 En

(3.146)

=
q

|2 q 2 u u |V q q q

nk + q nk
k k +q

(3.147) (3.148)

=
q

|2 q 2 (q, 0)u u |V q 0 q q

where the virtual states |n are electron-hole excitations of the lled Fermi sea. This term gives a correction to the elastic term in (3.142). In other words, the elastic modulus and, thus, the 2 phonon frequency q 2 = /0 q 2 = c2 s q is renormalized according to
ren q 2 2 q +

|2 q 2 |V q 0

2 0 (q, 0) = q

|2 q |V q 20

ln

q + 2 kF q 2kF

(3.149)

From the behavior for q 0 we infer that the velocity of sound is renormalized. However, a much more drastic modication occurs at q = 2kF . Here the phonon spectrum is softened, i.e. the frequency vanishes and even becomes negative. The latter eect 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.
13

Note that indeed the expression

2 q =

2 p (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 = e|| |coh Q
2

/2 n=0

n (b Q) n |0 n!

(3.151)

70

q q
(ren)

2k F

Figure 3.7: Kohn anomaly for the one-dimensional system with electron-phonon coupling. The renormalization of the phonon frequency is divergent at q = 2kF .

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)

where Q = 2kF and u0 remains to be determined variationally. We investigate the eect 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 det
2 k2 2m

2 (k Q)2

2m

=0

(3.155)

where follows from the Fourier transform of the potential V (x), = iQu0 nV Q with = V Q dx eiQx V (x). (3.157) (3.156)

The equation (3.155) leads to the energy eigenstates

Ek = 2

4m

(k Q)2 + k 2

{(k Q)2 k 2 }2 + 16m2 ||2 /

(3.158)

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

(3.152)

and, moreover, we nd
coh coh = Q |u(x)|Q

1 eiQx + eiQx = u0 cos(Qx) L 20 Q

(3.153)

with u0 = /0 LQ , assuming being real.

71

The total energy of the electronic and ionic system is then given by Etot (u0 ) = 2
0k<Q

Ek +

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
2

(3.160)
Q

2 32Q2 m2 n2 V Q
4 +kF

2m

u0
0

dk 2

1 {(k 1 Q)2 k 2 }2 +
4

2 u2 / 4 16m2 Q2 n2 V Q 0

2 Q u0 2

(3.161)

2 4Qmn2 V Q = u0 2 = u0

dq
kF

2 u2 / q 2 + 4m2 n2 V Q 0
2k
F

2 Q u0 2

(3.162)

2 8Qmn2 V Q arsinh 2

u 2mnV Q 0

2 Q u0 . 2 1.

(3.163)

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

2k
F

mnV Q

exp

2k

2 8mn2 V

2 F 1/N (0)g e kF nV Q

(3.164)

2 /2m is the Fermi energy and N (0) = 2m/ 2 k is the density of states at the where F = 2 kF F 2 / that describes the phononFermi energy. We introduced the coupling constant g = 4n2 V Q induced eective electron-electron interaction. The coupling is the stronger the more polarizable (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, + = 2|| = 8 Ek E = Ek F F

exp

1 N (0)g

(3.165)

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 nite energy gap for all electron-hole excitations. 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 k (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 k (x) = e|k (x)|2 = 2( k Ek )|| e 1 sin Qx (Ek k )2 + ||2 (3.167)

72

+Q k

Figure 3.8: Change of the electron spectrum. The modulation of the ionic background yields gaps at the Fermi points and the system becomes an insulator.

and its modulation from the homogeneous distribution en is given by (x) =

k

e k (x) (en) = 2

kF

dk 2

4 k2 k 2
F

m|| sin Qx + m2 ||2

(3.168)

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 eect 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.

2 F en|| = ln sin(2kF x). 16 F ||

(3.169)

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 dene 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 eects 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 . 73 (3.173)

Adding (3.171) to (3.172) and subtracting (3.170), we obtain (el + ion )Vren = Va + Vel + Vion which simplies with (3.173) to = el + ion 1 (3.175) (3.174)

In order to nd 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 e (3.176)

i.e. the potential Va /el that results from bare screening of the polarizable electrons is additionally screened by an eective ionic dielectric function ion e which includes electron-phonon interactions. Using equation (3.175) and the denition of ion e via (3.176) we obtain ion e = 1 + or using the denition (3.70) ion 0,e = ion 0 . el (3.178) 1 ion ( 1), el (3.177)

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

, 2 k2 F . el = 1 + T2 k

=1

2 p

(3.179) (3.180)

For the electrons we used the result from the quasi-static limit in (3.93). The full dielectric function now reads
2 2 kT p F =1+ 2 2 = k

1+

2 kT F k2

2 k 2

(3.181)

k,

k+q, +

q,

k,

kq,

Figure 3.9: Diagram for the electron-electron interaction involving also electron-phonon coupling.

74

The time-independent Coulomb interaction Va = 4e2 q2 (3.182)

between the electrons is replaced in a metal by an eective interaction Vren (q , ) = = 4e2 q 2 ( q , ) 4e2 2 + q2 kT F 2 2 2 q . (3.183) (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 eect 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 inuence 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 eld

Electrons couple through their orbital motion and their spin to external magnetic elds. 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 eld is the Quantum Hall eect of a two-dimensional electronic system. In both case the Landau levels with play an important role and will be introduced here in a rst step.

4.1

The de Haas-van Alphen eect

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 eld.

4.1.1

Landau levels

Consider a free electron gas subject to a uniform magnetic eld B = (0, 0, B ). The one-particle Hamiltonian for an electron is given by H= 1 e i A 2m c
2

gB

Sz B.

(4.1)

We x 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) simplies to H= 1 2m
2

2 e Bx + i x2 y c

g 2 B Sz B. z 2

(4.2)

In this gauge, the vector potential acts like a conning 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 Hn [(x ky 2 )/ ]e(xky 76
2 )2 /2 2

(4.4)

where Hn (x) are the Hermite polynomials and represents the magnetic length dened via 2 = c/|eB |. The eigenenergies of the Hamiltonian (4.2) read En,kz ,s =
2 k2 z

2m

+ c n +

1 2

gB

Bs

(4.5)

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 dierences 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 conned 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)

xes 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 ux quantum (e.g. Aharonov-Bohm interference eect). Thus the degeneracy Ndeg corresponds to the number of ux quanta included in the total magnetic ux 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 1 2 2
kz

(E En,kz ,s )
2 k2 dkz g 1 z E c n + + B Bs 2 2m 2

(4.10) (4.11) (4.12)

(2m)3/2 c 8 2 2

1 E c (n + 1/2) + gB Bs/

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 eld, N0 (E, B = 0) = 1 E
2 k2

k,s

2m

(2m)3/2 E. 2 2 3

(4.13)

The density of states for nite applied eld is shown in Fig. 4.1 for one spin-component.
Like the vector potential, the wave function is a gauge dependent quantity. To see this, observe that under a gauge transformation A(r , t) A (r , t) = A(r , t) + (r , t) the wave function undergoes a position dependent phase shift (r , t) (r , t) = (r , t)ei
c(r ,t)/e 1

(4.6)

(4.7)

77

N0

B=0

B=0

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

4.1.2

Oscillatory behavior of the magnetization

In the presence of a magnetic eld, 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 eld. In order to understand the resulting eect on the magnetization, we consider the free energy F = N T S = N kB T ln 1 + e(En,kz ,s )/kB T
kz ,ky ,n,s

(4.14)

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 4 B k T 1 B F . M = N P B 1 + L + B (4.15) P B B B B sinh 2 kB T
=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 suciently low temperatures, kB T < B B c , the magnetization as a function of the applied eld 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 inuence the magnetic moment upon successively passing through the Fermi energy as the magnetic eld is varied. The period in 1/B of the oscillations of the term = 1 is easily found to be F B or 1 B = 2e 1 c A(kF ) (4.17) 1 B = 2 (4.16)

2 of the Fermi where we used that B = e/2mc and dened the cross sectional area A(kF ) = kF sphere perpendicular to the magnetic eld.
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 eld. The semiclassical equations of motion for the center of mass of the wave packet (1.96, 1.97) simplify in the absence of an electric eld to = vk = k r k e k = vk B . c (4.18) (4.19)

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

Pn

with n being an integer and a system specic shift, irrelevant for the nal result and h = 2 . The momentum within the semiclassical approach is expressed as e e p = k A = (r B + A) , c c where we used Eq.(4.19) integrated over time. Thus, Eq.(4.20) can be now evaluated as p dr = e c {r B + A} dr = f racecn + e B c r dr = h(n + ) , (4.22) (4.21)

Pn

Pn

Pn

=2eB Sn /c=2en /c

where Sn is the area encircled by the path Pn and n = B Sn the magnetic ux threading. With this we nd 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 |dr | = 0 |dk| 2B Sn = 0 2B
2

An

(4.24)

With the ux quantization in Eq.(4.23) we obtain for a given magnetic eld the area An , An = (n + )4 2 while for a given area A we obtain, A = (n + )4 2 Bn 4 2 = (n + ) B . 0 0 n (4.26) B 0 (4.25)

Thus changing the eld 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 B = 1 Bn+1 79 1 4 2 = . Bn A0 (4.27)

band botton

extremal areas

band top

extremal area

Fermi surface

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 k -direction (k B ). Left panel: 3-dimensional view of system with a simple Fermi surface. Right panel: cross section including k . 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 k B. This corresponds the extremal area A(k ). Note, also a band top yields the same type of singularity and corresponds to a minimum of A(k ). Therefore, in experiment the magnetization maxima oscillate with a period in the inverse magnetic eld 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 eld 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 eect. For both methods it is crucial that the Landau levels are suciently clearly recognizable. Apart from low temperatures this necessitates suciently clean samples. In this context, suciently 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 c . (4.28)

4.2

Quantum Hall Eect

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

the original Hall eect. To this end we consider the classical equation of motion of an electron, subject to an electric and a magnetic eld m v dv = e E + B , dt c (4.29)

where m is the eective 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 xed current j = (0, jy , 0) = (0, n0 ev, 0) and magnetic eld B = Bz , the steady state condition (4.30) simplies to Ex + vBz = 0. c (4.31)

The solution Ex = vBz /c yields the Hall voltage that compensates the Lorentz force. The Hall conductivity H is dened as the ratio between the longitudinal current jy and the transverse electric eld Ex , leading to H = jy Ex = n0 ec e2 = , Bz h (4.32)

where = n0 hc/Be. 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 x

Bz

Figure 4.3: Schematic view of a Hall bar. The current runs a long the y -direction and the magnetic eld 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 Eects When measuring the Hall eect in a special two-dimensional electron system, Klaus von Klitzing and his collaborators made 1980 an astonishing discovery.3 The system was the inversion layer of GaAs-MOSFET device with a suciently 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 conned in the third dimension like in a
3

See [von Klitzing, Dorda, and Pepper, Phys. Rev. Lett. 45, 494 (1980)] for the original paper.

81

potential well (cf. Section 2.4.3). In high magnetic elds between 1 30T and at suciently 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 (4.33)

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

yy
0

Figure 4.4: Integer Quantum Hall eect: As a function of the lling factor plateaus in xy appear at multiples of e2 /h. The longitudinal conductance yy is only nite for llings where xy changes between plateaus. In 1982, Tsui, St ormer, and Gossard4 discovered an additional quantization of H , corresponding to certain rational multiples of e2 /h. Correspondingly, one now distinguishes between the integer quantum Hall eect (IQHE) and the fractional quantum Hall eect (FQHE). These discoveries marked the beginning of a whole new eld in solid state physics that continues to produce interesting results.

4.2.1

Hall eect of the two-dimensional electron gas

Here we rst discuss the Hall eect 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 conning the electronic system to two dimensions, the Hamiltonian reduces to H= 1 2m
2

2 e + i + Bx 2 x y c

(4.34)

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 |/m c. Here, we will concentrate on the lowest Landau level
4

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

82

(n = 0) with the wave function 0,ky = 1 2

2

e(xx0 )

2 /2 2

eiky y .

(4.35)

where the magnetic length = c/|eB | gives the extension of the wave function in the presence of the magnetic eld. 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 eld Ex along the x-direction. The Hamilton operator (4.34) is then modied 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 ) x0 (ky ) = ky
2

eEx . 2 m c

(4.36)

Moreover the degeneracy of the Landau level is lifted since the energy becomes ky -dependent and (after completing the square) takes the form En=0 (ky ) = c m + eEx x0 (ky ) + 2 2 cEx B
2

(4.37)

The energy (4.37) corresponds to the wave function 0ky from (4.35) where x0 is replaced by x0 . The velocity of the electrons is then given by vy (ky ) = eE 1 dEn=0 (ky ) = x dky
2

cEx , B

(4.38)

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 (4.39)

where = n0 2 2 is the lling of the Landau level.5 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 Eect

The plateaus observed by von Klitzing in the Hall conductivity H of the two-dimensional electron gas as a function of the magnetic eld 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, (4.40)

and only becomes nite 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 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 nd out, even small amounts of disorder lead to the localization of electronic states in this two-dimensional
Note that 1 = B/n0 0 where 0 = hc/e represents the ux quantum, i.e. 1 B is the number of ux quanta 0 per electron.
5

83

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

2m

1 r r r r

e 1 i Br r 2 c

(r, ) + U (x, y ) (r, ) = E (r, ) .

(4.41)

Without the external potential U (x, y ) we nd the ground state solutions n=0,m (r, ) = 1 2
2 2m m !

eim er

2 /4 2

(4.42)

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

= 2mB

c hc =m = m0 , eB e

(4.43)

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

|| 2

Figure 4.5: Wavefunction of a Landau level state in the symmetric gauge. Now we consider the eect 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 (4.44)

the exact expression of all eigenstates of equation (4.41) in the lowest Landau level is obtained using the Ansatz (r, ) = 0,m 1 2
2 2 (

r + 1)

eim er
2

2 /4 2

(4.45)

= 2 m C / After introducing the dimensionless parameters C1 1 quantities and from equation (4.45) can be expressed via 2 = m2 + C1 , 2

= 8 4 m C / 2 , the and C2 2

(4.46) (4.47)

. 1 + C2

Indeed, the Ansatz (4.45) describes eigenstates of the disordered problem (4.41). The degeneracy of the ground state energy (the lowest Landau level) is now lifted, E0,m = c 2
2 2

( + 1) m + C3 . 84

(4.48)

The wave functions are concentrated around the radii rm = , C C1 1 and m 1 the energy is approximatively given by 2 E0,m

2 .

For weak potentials

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

(4.49)

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 eld 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 eld is varied the lling = 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 lled or empty depending on the strength of the magnetic eld, 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 ll now water into such a landscape. The trajectories of the particles is restricted to the shore line. For small lling, we nd 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 lling, 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 lled, rst all occupied state are localized (low lling). At some special intermediate lling level, the extended states are lled and contribute to the current transport. At higher chemical potential (lling) the states would be localized again. In the following argument, going back to Robert B. Laughlin, the presence of lled 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 eld B along the z -axis. In addition we can introduce an arbitrary ux through the hole without inuencing the uniform eld in the disc. This ux 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 ux is increased adiabatically by , the vector potential is changed according to A A + A = A + , which in our case means (A) = 85 . 2r (4.50)

closed

extended

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

At the same time, the wave function acquires a phase factor eie/
c

= ei(/0 ) .

(4.51)

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 (4.42), 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 ux 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.

B
y

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 eld B and the ux 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 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 eld 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

(4.52)

86

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

(4.53)

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 ux is increased by . Following the Aharonov-Bohm argument that the energy of the system is invariant under a ux 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 (4.54)

We conclude from this argument, that each lled Landau level containing percolating states will contribute e2 /h to the total Hall conductivity. Hence, for n N0 lled 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 eld is given by N2DEG (E ) = 2
kx ,ky

2 (k 2 x

2) + ky

2m

Lx Ly m 2

(4.55)

with twice degenerate energy states for the spins, whereas for the Landau levels in a clean sample, we have NL (E ) = Lx Ly 2
2 n,s

(E En,s ).

(4.56)

Here the prefactor is given by the large degeneracy (4.9) of each Landau level.

N(E)

B=0

N(E)

B=0

N(E)

B=0
lokalisiert ausgedehnt

E
translationsinvariant

E
Unordnung

Figure 4.8: Density of states for the two-dimensional electron gas in three dierent cases. On the left panel, the system without applied magnetic eld. On the middle panel, an external magnetic eld is applied to a clean system showing sharp strongly degenerate Landau levels. The right panel visualizes the eect of disorder in the two-dimensional system with magnetic eld; 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. According to our previous discussion, the main eect of a potential is to lift the degeneracy of the 87

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 lled (see Fig. 4.8). For partially lled 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 ux is changed by 0 (in Laughlins argument) are occupied. Thus, the charge transferred does not amount to a complete e. The appearance of partially lled extended states marks the transition from one plateau to the next and is accompanied by a nite 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 Eect is only observable for suciently low temperatures (T < 4K). Edge states and B uttikers argument A conning 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.(4.37), we nd that the energy is not symmetric in ky , the wave vector along the edge, i.e. E (ky ) = 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 veried by inspection of (4.37) based on Fig.4.9. The total current owing along the edge for a given Landau level is I=
ky

e v , Ly y

(4.57)

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 (4.38). In summary, we have I= e 2
occupied

dky

dEn (ky ) dky

e h
occupied

dx0

e dE (0) = ( En ) dx0 h

(4.58)

where x0 = ky 2 is the transversal position of the wave function and is the chemical potential. Suciently far away from the boundary En is independent of x0 and approaches the value (0) = (1/2 + n) of a translationally invariant electron gas. The potential dierence between En c the two opposing edges leads to a net current along the edge direction of the Hall bar, A B = eVH = eEx Lx = and with that H = IH e2 = , Ex Lx h (4.60) h h (IA + IB ) = IH , e e (4.59)

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 88

E E
n=2 n=1 n=0

IA
B A

x E B A

IB

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 dierence is shown, where the latter yields a nite 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 xes the Hall conductance value n to 2.

shape of the conning edge potential. 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 the resistivity is a tensor: j = E E = j (4.61) (4.62)

where the conductivity and the resistivity are both a symmetric 2 2 tensor with xx = yy and xx = yy . Therefore we nd yy = xy = yy 2 yy + 2 xy xy , . (4.63) (4.64)

2 2 yy + xy

In the following argument, we explain why the longitudinal resistivity yy in two dimensions has to vanish in the presence of a nite 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 nite 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 Eect

Only two years after the discovery of the Integer Quantum Hall Eect, St ormer, 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 plateaus is observed at a lling 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 . (4.65)

The emergence of these new plateaus is a clear evidence of the so-called Fractional Quantum Hall Eect (FQHE). 89

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

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 eects. Laughlin specially investigated the case p,m = 1/3 and made the Ansatz 1/m (z1 , . . . , zN ) (zi zj )m exp |zi |2 4 2 (4.66)

i<j

for the N -body wave function, where z = x iy 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.

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. 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 (4.66) can be written as 1/m =
i<j

(zi zj )m1 S

(4.67)

90

where S is the Slater determinant6 describing the completely lled lowest Landau level. We see that the prefactor of S in equation (4.67) acts as a so-called Jastrow factor that introduces correlation eects 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 exp(i(m 1) ) = exp i e m1 0 , c 2 (4.72)

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.
composite Fermion
composite Fermion s

Bex

im externen Feld

2 0
Beff

Beff = B ex 2 0 n0
FeldKompensation IQHE

Figure 4.12: Sketch of the composite Fermion concept. Electrons with attached magnetic ux lines, here for the state of = 1/3. According to the Footnote 5 (see equation (4.39)), the case p,m = 1/3 implies that there are three ux 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
6 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 (4.42)) they are given by m |z | m (z ) = z e
2

/4 2

(4.68)

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 ) 1 det = N! 1 1 . . . 1 z1 z2 . . . zN
2 z1 2 z2 . . . 2 zN

(4.69)

N 1 z1 N 1 z2 . . . N 1 zN

exp |zi |2 4 2 . (4.70)

The remaining determinant is a so-called Vandermonde determinant, which can be reexpressed in the form of a product, such that S =
i<j

(zi zj ) exp
i

|zi |2 4 2

(4.71)

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 dened total angular momentum Lz = N .

91

an electron that has two (in fact m 1) negative ux quanta attached to it. These objects may be considered as independent fermions since the attached ux quanta compensate the Jastrow factor in equation 4.67 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 (4.72). Due to the presence of the ux 20 per electron, the composite fermions are subject to an eective eld composed of the external eld and the attached ux quanta: Be = B 1 = B 3 20 (zi )
i

(4.73) (4.74)

2 20 (zi ) B 3

For an external eld B = 3n0 0 , the expression in the brackets of equation (4.74) vanishes and the composite fermions feel an eective eld Be = 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 lled Landau levels with composite fermions consisting of an electron with an attached ux of 2k 0 , the eective eld reads Be = n0 From this we infer, that 1 1 + 2k = n p,m or equivalently p,m = p n = . m 2kn + 1 (4.77) (4.76) 0 2kn0 0 = n0 0 . p,m n (4.75)

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 eective 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.7

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

while detailed literature on the fractional quantum Hall eect 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 eect 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-dened energy-momentum relation, and that their groundstate is a lled 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 justied. This quasiparticle picture will lead us to Landaus phenomenological theory of Fermi liquids.

5.1

Lifetime of quasiparticles

We rst consider the lifetime of a state consisting of a lled 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 Hee =
V (q )c kq ,s ck +q ,s ck ,s ck,s ,

(5.1)

k,k ,q s,s

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 q 2 (q , 0) = q2 4e2 2 + kT F (5.2)

from equation (3.94) will be used. As we are only interested in very small energy transfers , the static approximation is admissible. F
k kq k+q

Figure 5.1: The decay of an electron state above the Fermi energy happens through scattering by creating particle-hole excitations. In a perturbative treatment, the lowest order eect of the interaction is the creation of a particlehole excitation in addition to the single electron above the Fermi energy. As the additional 93

electron changes its momentum from k to k q , a hole appears at k and a second electron with wavevector k + q is created outside the Fermi sea. The transition is allowed whenever both energy and momentum are conserved, meaning k = (k q ) k + (k + q ), and
k

(5.3)

k q

k +q .

(5.4)

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

k ,q s

k +q )).

(5.5)

Note that the term n0,k (1 n0,kq )(1 n0,k +q ) takes care of the Pauli principle, by ensuring that the nal state after the scattering process exists, i.e. the hole state k lies inside and the two particle states k q and k + q lie outside the Fermi sea. First the integral over k is performed under the condition that the energy k +q k of the excitation is small. With that, the integral reduces to S (q,k , q ) = 1 n0,k (1 n0,k +q ) (
k q

k +q ))

(5.6) (5.7) (5.8)

1 = (2 )3 =

d3 k n0,k (1 n0,k +q ) (

k +q

q,k )

N ( F ) q,k 4 qvF

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
Small are justied, 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 =
2 2 2 2 with k1 = kF k2,0 and k2 = kF (k k1 kF 1

dk k
k2 0

dk

q + 2m

2 2

qk m

(5.9) (5.10)

m 2 2 k1 k2 , 4 2 2 q
,0

+ q )2 , where k

,0

= (2m q 2 )/2 q is enforced by the delta function.

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,k (1 n0,k +q ) > 0 and can be obtained by simple geometric considerations. equation (5.8) follows immediately.

94

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 coordinates, the integral reads 1 2 N ( F ) = k 4vF = N( F) (2 )2 2 vF |V (q )|2 q,k q q , k q
2

(5.11) (5.12)

q ,s

d3 q |V (q )|2 dq |V (q )| q

N( F) = (2 )4mvF = N( F) (2 )4mvF

2 2 1

d sin (2k cos q ) 1 (2k cos q )2 4k

2

(5.13)

dq |V (q )|2 q 2

.
1

(5.14)

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

1 2 k 2 kF 4k

(5.15) (5.16) (5.17)

N( F) m 1 ( (2 )4vF kF 4 1 8
3

)2

dq |V (q )|2 dq |V (q )|2 .

N( F) ( 2 vF

)2

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 fullled. Essentially, the result states that 1 ( k
k

)2

(5.18)

for k slightly above the Fermi surface. This implies that the state |ks 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). The resonance becomes arbitrarily sharp as the Fermi surface is approached
kkF
2

lim

/k = 0, k F

(5.21)

The spectral function is dened as A(E, k)

n

| n |cks |0 |2 (E En )

(5.19)

where |0 is the exact (renormalized) ground state and |n 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 +
2

(5.20)

k2

where Zk is the quasiparticle weight, smaller than 1.

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.

so that the quasiparticle concept is asymptotically valid. The equation (5.21) can also be seen as a verication 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-dened 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, dened as n (k), is subject to the condition N=
k,

n (k).

(5.22)

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)

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 that n0ks = ck ck of the real electrons in the ground state are not identical (Figure 5.3). Interestingly, n0ks is still discontinuous at the Fermi surface, but the height of the jump is, in general, smaller than unity. The modication of the electron distribution function from a step
3

This is the content of the Luttinger theorem [J.M. Luttinger, Phys. Rev. 119, 1153 (1960)].

96

n0ks

n (k)

kF

kF

Figure 5.3: Schematic picture of the distribution function: Left panel: modied distribution function of the original electron states; right panel: distribution function of quasiparticle states making a simple step function.

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 dierence. 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, (k)n (k)

1 2

f (k, k )n (k)n (k ) + O(n3 )

k,k ,

(5.25)

where E0 denotes the energy of the ground state. Moreover, the phenomenological parameters (k) and f (k, k ) have to be determined by experiments or by means of a microscopic theory. The variational derivative (k) = E = n (k)
(k)

f (k, k )n (k )
k ,

(5.26)

yields an eective energy-momentum relation (k), whose second term depends on the distribution of all quasiparticles. A quasiparticle moves in the mean-eld of all other quasiparticles, so that changes n (k) in the distribution aect (k). The second variational derivative describes the coupling between the quasiparticles 2E 1 = f (k, k ). n (k)n (k ) (5.27)

We introduce a parametrization for these couplings f (k, k ) 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|, |k | kF . Therefore the dependence of f (k, k ) on k, k can be reduced to the relative angle k, k k ) + f a (k, k ) f (k, k ) = f s (k, 97 (5.28)

 = k/|k|. The symmetric (s) and antisymmetric (a) part of f (k, k ) can be expanded where k in Legendre-polynomials Pl (z ) 4 , leading to k )= f s,a (k,
l=0

fls,a Pl (cos k, k ).

(5.31)

The density of states at the Fermi surface is dened as N( F) = 2 ( (k) )= F

2 kF m kF = 2 vF 2 2

(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 eective mass, although we will be only interested at the spectrum in the immediate vicinity of the Fermi energy. With this denition, we also introduce the so-called Landau parameters Fls = N ( F )fls , Fla = N( )fla , F (5.34) (5.35)

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 specic heat, compressibility, spin susceptibility among others.

5.2.1

Specic heat - Density of states

1 e[(k)]/kB T +1

Since the quasiparticles are fermions, they obey Fermi-Dirac statistics n(0) (T, k) = (5.36)

with the chemical potential . For low temperatures, we nd

(0) n (k) = n(0) (T, k) n (0, k)

(5.37)

which, using Eq.(5.26) leads to (k) = (k), because here 1 f (k, k )n (k ) = 0 .

k ,

(5.38)

In n (k ) 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 ) =
4

1 e[ (k)
F ]/kB T

+1
2ll 2l + 1

(5.39)

Legendre-Polynomials: The Legendre polynomials satisfy the orthogonality relation d k Pl (cos k )Pl (cos k ) = 4
+1

dzPl (z )Pl (z ) =
1

(5.29)

for all l = 0, 1, 2, .... The rst few polynomials are P 0 (z ) = 1 ,

5

P 1 (z ) = z ,

P2 (z ) =

1 (3z 2 1) , 2

P 3 (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 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 specic 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 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 specic heat C (T ) = T S T k T = B k T = B (5.41) e(k)/kB T (k) ln ( e(k)/kB T + 1 )2 kB T 2 n (k) 1 n (k) (5.42) (5.43)

k,

k,

(k) (k) 1 2 k T2 k T 4 cosh ( (k)/2kB T ) B B

F

is obtained, where we introduced (k) = (k) N( F) C (T ) T 4kB T 3 k2 N ( F ) B 4 = 2 k2 N (

B

. In the limit T 0 we nd 2 cosh2 (/2kB T ) dy y2 cosh2 (y/2) (5.44)

+
F

(5.45)

(5.46)

which is the well-known linear behavior C (T ) = T for the specic heat at low temperatures, 2 N ( )/3. Since N ( ) = m k / 2 2 , the eective mass m of the quasiparticles with = 2 kB F F F can directly determined by measuring the specic heat of the system.

5.2.2

Compressibility - Landau parameter Fs 0

A Fermi gas has a nite compressibility because each fermion occupies a nite amount of space due to the Pauli principle. The compressibility is dened as6 = 1 p (5.48)
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 xed. We consider the response of the Fermi liquid upon application of uniform pressure p. The shift of the quasiparticle energies is given by (k) =
6

(k ) (k ) k (0) p = p = v k k p = k (0) p p k p 3

(5.49)

An alternative denition considers the change of particle number upon change of the chemical potential, = 1 n2 n (5.47)
T,

with n = N/.

99

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 1 1 f, (k, k )n (k )
k ,

= k (0) p +

f, (k, k )
k ,

n (k ) (k ) (k )
F )k

(5.50)

= k (0) p

k ,

f, (k, k ) ( (k )

p
F /3

Changes are concentrated on the Fermi surface such that we can replace k = 2 = (0) N ( Therefore we nd =
F)

so that (5.51)

dk s s f (k, k ) = (0) F0 . 4 (0) s . 1 + F0

(5.52)

Now we determine (0) from the volume dependence of the energy E (0) =
k,

3 (k) = N 5

2 k2 3 3 2N F = N = 5 2m 10 m

3 2

2/3

(5.53)

Then we determine the pressure p= and E (0) =

N

1 2N 5 m

3 2

2/3

(5.54)

1 p 1 2N 2 = = (3 2 n)2/3 = n (0) 3m 3

(5.55)

kF
k F k F

kF
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.

100

5.2.3

Spin susceptibility - Landau parameter Fa 0

In a magnetic eld 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 eld. The shift of the renormalized quasiparticle energy due to the applied eld is (k) = (H, k) (H = 0, k) 1 f, (k, k )n (k ) = gB H + 2
k ,

(5.57)

= g B H . 2 Note that by symmetry, n (k) = n (k). Due to interactions, the renormalized gyromagnetic factor g diers from the value of g = 2 for free electrons. We focus on the second term in Eq.(5.57), which can be reexpressed as 1 f (k, k )n (k ) =
k ,

1 1

f (k, k )
k ,

n (k ) (k ) (k )
F

k ,

f (k, k ) ( (k )

) g B H

(5.58)

Combining this result with the Eqs. (5.57) and (5.58), we derive g =gg N ( F ) or equivalently g = g a. 1 + F0 (5.60) dk a a , f (k, k ) = g g F0 4 (5.59)

The magnetization of the system can be computed from the distribution function, M = gB
k,

n (k) = gB 2 ( (k) 2
F

k,

n (k) (k ) 2 (k) 2 (5.61)

= gB
k,

) g B H

from which the susceptibility is immediately found to be = 2 N ( F ) M = B a . H 1 + F0 (5.62)

The changes in the distribution function induced by the magnetic eld feed back into the susa > 0) or enhanced (F a < 0). For the ceptibility, so that the latter may be either weakened (F0 0 a magnetic susceptibility, the Landau parameter F0 and the eective mass m (through N ( F )) lead to a renormalization compared to the free electron susceptibility.

101

5.2.4

Galilei invariance - eective mass and Fs 1

We initially introduced by hand the eective mass of quasiparticles in (k). In this section we show, that overall consistency of the phenomenological theory requires a relation between s ). The reason is, that the eective mass the eective mass and one Landau parameter (F1 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 kF n=+1 q

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 = with v (k) = = 1 1 k (k) k
(k)

v (k)n (k) =
k,

v (k)n (k)
k,

(5.64)

k ,

k f (k, k )n (k ) .

(5.65)

Thus we obtain for the current density, jq = 1 k 1 n (k) + 2 m 1 [n(0) (k) + n (k)] k f (k, k )n (k ) 1
k, k , 2 [k n(0) (k)]f (k, k )n (k ) + O (q )

k,

k, k ,

1 = = 1

k,

k 1 n (k) 2 m k 1 n (k) + 2 m

(5.66)

k,

k, k ,

f (k, k ) ( (k )

k n (k) + O(q 2 ) = j 1 + j 2 . m

where, for the second line, we performed an integration by parts and neglect terms quadratic in n and, in the third line, used f (k, k ) = f (k , k) and k n(0) (k ) = n(0) (k) (k) k
(k )

= ( (k) 102

)k

(k )

= ( (k)

2k

(5.67)

The rst 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 (backow) of the other particles due to interactions. From a dierent 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 jq = N q 1 = m k 1 n (k) = m k n (k). m (5.68)

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 = + m m = k/k then leads to which with k F 1 1 = + N( F) m m k dk 1 1 s k f (k, k ) = + 4 m m m dk k ) N ( F )f s (k, 4
l=0

k ,

f (k, k ) ( (k )

k m

(5.69)

cos k
P1 (cos k )

(5.70)

Fls Pl (cos k )

or by using the orthogonality of the Legendre polynomials, m 1 s . = 1 + F1 m 3 (5.71)

where 1/3 = 1/(2l + 1) for l = 1 originates from the orthogonality relation of Legendres in order for polynomials, as shown above. Therefore, the relation (5.71) has to couple m to F1 s > 0 so that Landaus theory of Fermi liquids to be self-consistent. Generally, we nd that F1 quasiparticles in a Fermi liquid are eectively heavier than bare electrons.

5.2.5

Stability of the Fermi liquid

m = , 0 m m 1 = s, 0 m 1 + F0 m 1 = a 0 m 1 + F0

Upon inspection of the renormalization of the quantities treated previously (5.72) (5.73) (5.74)

with

m 1 s = 1 + F1 , m 3 and the the response functions of the non-interacting system are given by 0 =
2 mk kB mkF 3m F , 0 = and 0 = 2 B 2 2 2 2 2 3 n kF

(5.75)

(5.76)

s 1 (F a 1), one notes that the compressibility (susceptibility ) diverges for F0 0 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

103

deformation of the quasiparticle distribution function may vary over the Fermi surface, so that arbitrary deviations of the Fermi liquid ground state may be classied by the deformation ) = n (k
+l

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 anisotropy Ylm (k , k ) is given by 7 l,m = l,m 1+
Fls,a 2l+1 (0)

(5.85)

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, dened as R 0 1 = = a R0 0 1 + F0
7

(5.87)

General response and distribution deformations: We consider a force eld F with conjugate polarisation P which yields a modication of the quasiparticle dispersion,
(k )

= (k)F

and

(k) = (k)F

(5.78)

m where we assume that (k) = Yl,m (k ) without spin dependence. Then we can write , k ) = (1) Yl,m (k , k

(k) =

(k )

k )n (k ) = f (k,
k ,

(k)

k ,

k ) n (k ) . f (k, (k )

(5.79)

In the last step we take for n (k ) the self-consistent value taking the feedback of the quasiparticle coupling into account. We now use the relation

k )= f s,a (k,
l=0

k ) = 4 fls,a Pl (k
l=0

fls,a 2l + 1

+l Ylm (k , k )Ylm (k , k ) m=l

(5.80)

and insert Eq.(5.78) to nd (k)F = (k)F N ( which leads straightforwardly to = Fls 2l + 1 = 1+

s Fl 2l+1

F)

fls Ylm (k , k ) 2l + 1

dk Ylm ( k , k ) (k )F

(5.81)

(5.82)

Now the polarisation is calculated which we may dene as P = 1 (k)n (k) =

k,

(k)
k,

n (k) = N ( (k )

F )F

(5.83)

such that the linear response is given by =

8

2 N ( F ) P = . s Fl F 1 + 2l+1

(5.84)

I.J. Pomeranchuk, JETP 8, 361 (1958)

104

2 . Note that the Wilson ratio does not depend on the eective where R0 = 0 /0 = 62 / 2 kB B 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 solidies (pc 2.5MPa = 25bar).

pressure p=0 p < pc

m /m 3.0 6.2

s F0 10.1 94

a F0 -0.52 -0.74

s F1 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 solidication. The trends show obviously, that the higher the applied pressure is, the denser the liquid becomes and the stronger the quasiparticles interact. Approaching the solidication 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 eective 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 eective mass is induced by the hybridization of itinerant conduction electrons with strongly interacting (localized) electron states in partially lled 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 eld theory and would go beyond the scope of these lectures. However, plain Rayleigh-Schr odinger 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 ), described by the Hamiltonian H= =
k,s k cks cks k,s k cks cks

+ + U

d3 r d3 r (r ) (r ) U (r r ) (r ) (r )
c k+q ck q ck ck .

(5.88) (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 nd well-dened quasiparticles, the interaction between the Fermions has to be short ranged. This specially holds for the contact interaction.

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 = c ks cks = nks + nks , we can expand the energy resulting form equation (5.89) following the Rayleigh-Schr odinger perturbation method, E = E (0) + E (1) + E (2) + (5.90)

105

with E (0) =
k,s k nk s ,

(5.91) (5.92) . (5.93)

E (1)

U =

k ,k

nk nk , nk nk (1 nk+q )(1 nk q ) +

E (2) =

U2 2
k ,k ,q

k +q

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 dierent contributions. We rst consider the term quadratic in nk and combine it with the rst order term E (1) , which has the same structure,
2 (1) = E (1) + U E 2

k ,k , q k

nk nk
k

k+q

k q

k ,k

nk nk .

(5.94)

through, In the last step, we dened the renormalized interaction U

2 =U+U U

1
q k

k+q

.
k q

(5.95)

In principle, restricted to compared to

depends on the wave vectors k and k . However, when the wave vectors are U the Fermi surface (|k| = |k | = kF ), and if the range of the interaction is small 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 cubic in nk and reads 2 (2) = U E 2 nk nk (nk + q + nk q )
k ,k ,q k

k+q

(5.100)

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+

nk

2 U 2

nk (nk+q + nk q ) nk+q nk q
k ,q k

k+q

(5.101)

k q

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 cuto 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, Qc

1
q k

k+q

=
k q

1 (2 )3
o

dq q 2
+1

dq

m (k k) q q 2
Qc

(5.96)

m = (2 )2 = where we use K = |k k| 2kF weak. m (2 )2 2mQc (2 )2

dq q
1

d cos m = K cos q (2 )2
0

dq q ln

qK q+K

(5.97)

Qc + 1

K 2 Q2 Qc K c ln 2K Qc + K K4 Q4 c ,

(5.98) (5.99)

K2 +O Q2 c

is indeed Qc . From this we conclude that the momentum dependence of U

106

and an analogous expression is found for (k). The coupling parameters f (k, k ) may be determined using the denition (5.26). Starting with f (kF , kF ) with wave-vectors on the Fermi surface (|kF | = |kF | = kF ), the terms contributing to the coupling can be written as 2 U 2 nk + q + nk q nk
k

k ,q

k +q

k +q k F k q

kF

2 U nk F 2 U
F

nk q nk
k k

(0)

(0)

k q

q =k F k F

(5.102) (5.103)
(0)

=
(0)

nk
kF

0 ( k F k F ) ,

where we consider nk = nk + nk . Note that the part in this term which depends on nk 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 dened in (3.63). With the help of equation (5.26), it follows immediately, that f (kF , kF ) = f (kF , kF ) = 2 U (k k F ). 2 0 F (5.104)

The other couplings are obtained in a similar way, resulting in f (kF , kF ) = f (kF , kF ) = U where the function 0 (q ) is dened as 1 0 (q ) = nk + q + nk
F

2 U 2 0 (kF + kF ) 0 (kF kF ) , 2

(5.105)

(0)

(0)

k +q

(5.106)
k

If the couplings are be parametrized by the angle between kF and kF , they can be expressed as f () = U 2 1+ 1+ N( ) U F 2 N( ) U F 2 2+ 1 cos 1 + sin(/2) ln 2 sin(/2) 1 sin(/2) sin(/2) 1 + sin(/2) ln 2 1 sin(/2) . (5.107) (5.108)

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), 1 u 1+u 1 + (2 + ln(2)) 6 2 a F0 = u 1+u 1 (1 ln(2)) 3 2 s F1 = u 2 (7 ln(2) 1) 15
s F0 =

u + 1.449 u 2 ,

(5.109) (5.110) (5.111)

= u 0.895 u 2 , 0.514 u 2 ,

N ( ) has been introduced for better readability. Since the Landau parameter F s where u =U F 1 is responsible for the modication of the eective 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 aect the renormalization of the eective 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 107

and the compressibility depends on the sign of the interaction. If the interaction is repulsive s > 0), implying that it is harder to compress the ( u > 0), the compressibility decreases (F0 a < 0) in this case, so that it is easier to polarize Fermi liquid. The susceptibility is enhanced (F0 the spins of the electrons. Conversely, for attractive interactions ( u < 0), the compressibility is s , whereas the susceptibility is suppressed with enhanced due to a negative Landau parameter F0 a ), with F a > 0. The attractive case is more subtle because the Fermi liquid a factor 1/(1 + F0 0 becomes unstable at low temperatures, turning into a superuid 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 eect of interactions on the ground state properties, using again Rayleigh-Schr odinger perturbation theory. The calculation of the corrections to the ground state |0 , the lled Fermi sea can be expressed as | = |(0) + |(1) + where |(0) = |0 |(1) = U
k,k ,q s,s

(5.112)

(5.113)
c k+q ,s ck q ,s ck ,s ck,s k

k+q

k q

|0 .

(5.114)

The state |0 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) aects the distribution function, will be discussed next. The momentum distribution nks = c ks cks is obtain as the expectation value, nk s =
(0)

|c (0) (2) ks cks | = nks + nks + |

(5.115)

where nks is the unperturbed distribution (kF |k|), and U2 2 U2 2 (1 nk1 )(1 nk2 )nk3 ( k + k3 k1 k2 )2 k+k3 ,k1 +k2 nk1 nk2 (1 nk3 ) ( k1 + k2 k k3 )2 k+k3 ,k1 +k2 |k| < kF . |k| > kF (5.116)

nks =

(2)

k 1 ,k 2 ,k 3

k 1 ,k 2 ,k 3

This yields the modication 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, nk F nk F + = 1 where nk F =
|k|kF 0

UN( F) 2

ln(2),

(5.117)

lim

nk + nk

(0)

(2)

(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.

108

nk

nk

3D
kF k

1D
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 justied 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 briey below. The Landau parameters have been expressed above in terms of the response functions 0 (q = kF kF ) and 0 (q = kF kF ). For the one-dimensional system, as given in Eqs.(5.104 - 5.106), the relevant contributions come from two congurations, since there are two Fermi points only (instead of a two-dimensional Fermi surface), (kF , kF ) q = kF kF = 0, 2kF . (5.119)

We nd 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. 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 nite) quasiparticle weight. The
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 2 kF + q 2 kF q ln |q | < 2 kF 2 2 2 kF + q + 2 kF q 4 kF q2 0 (q ) = (5.120) 2 m q + 2 k q 2 k F F arctan + arctan |q | > 2 kF 2 2 q 2 kF q + 2 kF q 2 4kF which diverges as
q 0 10

lim 0 (q ) =

m
2k F

ln(

q ), 2kF

q 2kF

lim

0 (q ) =

m
2k F

and

q 2kF +

lim

0 (q ) =

m 1 . (5.121) 2 kF q 2 kF

109

analogous calculation as in Sect.5.3.2 leads here to k+ 1 U2 ln 2 2 8 2 vF k kF (2) nk s k 1 U2 2 2 2 ln 8 kF k vF

k > kF . k < kF (5.124)

Here, k are cuto 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

Spin

q=e

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-lled 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 dierent kinds of domain walls, and are free to move at dierent 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 inuence the transport and yield a nite residual resistivity, as found in real materials. At nite temperature, electron-electron and electron-phonon scattering lead to a temperature-dependent resistivity. Furthermore, an external magnetic eld may inuence the resistivity, a phenomenon called magnetoresistance, and also leads to the previously studied Hall eect. In this chapter, the eects of a magnetic eld 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 eect) 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 eld E (q , ). For a homogeneous isotropic metal, we dene 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 or, in Fourier transformed, (q , ) q j (q , ) = 0. (6.3) (6.2)

It is interesting to see that a relation between the conductivity (q , ) and the dynamical dielectric susceptibility 0 (q , ) dened in equation (3.68) of chapter 3 arises from the equations (6.1) and (6.3). For this, we can calculate 0 (q , ) = (q , ) q j (q , ) = eV (q , ) eV (q , ) (q , ) q E (q , ) (q , ) [iq 2 V (q , )] = = . e V (q , ) e2 V (q , )

(6.4)

In an anisotropic material, the conductivity would be a full 3 3 tensor.

111

In the rst line, we used the denition (3.68) of 0 (q , ) and the continuity equation (6.3). To the second line, we made use of the denition (6.1) of (q , ) and then replaced E (q , ) by iq V (q , ), which is nothing else than the Fourier transform of the equation eE (r , t) = r V (r , t). From this calculations we conclude that 0 (q , ) = and thus (q , ) = 1 4i 4e2 0 (q , ) = 1 + (q , ). 2 q (6.7) iq 2 (q , ), e2 (6.6) (6.5)

In the limit of large wavelengths q kF , we know from previous discussions2 that (0, ) = 2 / 2 . Then the conductivity simplies to 1 p ( ) =
2 ip

(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. Dening 1 (2 ) as the real (imaginary) part of , this relation states that + 1 1 1 ( ) = P d ( ) (6.9) 2

and 2 ( ) = 1 P + d

1 ( ) . 1

(6.10)

A simple calculation with 2 from equation (6.8), yields 1 ( ) = ( ), 4 2 p 2 ( ) = . 4

2 p

(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,
0

1 d 1 ( ) = 2

d 1 ( ) =

2 p

e2 n . 2m

(6.13)

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

6.2

Transport equations and relaxation time

We introduce here Boltzmanns transport theory as as rather simple and ecient 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 nite 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 inuence, 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 innitesimal 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 . The F total number of particles N is given by N =2 d3 k 3 d rf (k, r , t). (2 )3 (6.15)

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 D f (k, r , t) = Dt r + k +r k t f (k, r , t) = f t ,
coll

(6.17)

where the substantial derivative in phase space D/Dt is dened 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 eld, we nd k , m = eE , k = r (6.18) (6.19)

, which is our central interest, originates from the electric eld. The collision i.e., the force k integral may be expressed via the probability W (k, k ) to scatter a quasiparticle with wave vector k to k . For simple scattering on static potentials, the collision integral is given by f t
3

coll

d3 k W (k, k )f (k, r , t)(1 f (k , r , t)) (2 )3 W (k , k)f (k , r , t)(1 f (k, r , t)) .

(6.20) (6.21)

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 rst term, describing the scattering4 from k to k , requires a quasiparticle at k, hence the factor f (k, r , t), and the absence of a particle at k , therefore the factor 1 f (k , 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, k ) = W (k , k). Assuming this, we can combine both terms and end up with f t =
coll

d3 k W (k, k ) f (k , r , t) f (k, r , t) . (2 )3

(6.22)

The Boltzmann equation is a complicated integro-dierential 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 k . Then, we make the Ansatz f t = f (k, r , t) f0 (k, r , t) . ( k) (6.23)

coll

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 eld 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 nd, after linearizing in f , if (k, ) with f (k, t) =
+

eE ( )

k f0 (k) =

f (k, ) ( k)

(6.24)

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 ) ( k )[E ( ) v k ]v k f0 ( k ) , 1 i ( k ) k (6.26)

This result leads straightforwardly to the quasiparticle current j ( ), j ( ) = 2e d3 k e2 v f ( k , ) = k (2 )3 4 3 d3 k (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 eect) and would not be conserved independently.

114

with v k = k k , which in turn can be simplied to j ( ) =

( )E ( ),

(6.28)

where the conductivity tensor reads = e2 4 3 d f0 ( ) ( ) 1 i ( ) dk k 2 vk v k |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 dierent 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 ( ) i e2 m2 vF 4 3 3
2 dk vF z =i 2 p 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 = e2 n m d
2 p f0 e 2 n ( ) = = . m 4

(6.31)

Since the function f0 / 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 nite frequency,
2 p ( ) = = 4 1 i 4 2 p

2 i + 1 + 2 2 1 + 2 2

= 1 + i2 .

(6.32)

Note that the real part satises the f -sum rule,

d 1 ( ) =
0 0

2 p = 4 1 + 2 2 8

2 p

(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
5

2 p

(i + )

=1

2 p 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 ( ) =
2 p 4 1 + 2 2

2 p ( ) . 4

(6.34)

115

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 reectivity of light hitting a metal surface perpendicularly. Then the reectivity 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/ = Relaxation-free regime ( 1 p )
2 2 1 ( ) p ,

(6.37) (6.38)

dening the penetration length

c .

In this limit, the real (1 ) and imaginary (2 ) part of the dielectric function (6.35) read (6.39)
2 p

. (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 n( ) ( ) 2 ( ) 2
2 p

2 ( )

1,

(6.41)

As a result, the reectivity R is practically 100%. The absorption index k ( ) determines the penetration depth through ( ) = c c ( ) p 2 . (6.42)

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 )
2 p 2 p

Here, we can expand the dielectric function (6.35) in ( )1 , yielding ( ) = 1 2 +i 3 . (6.43)

2 / 2 is large and negative and dominates in magnitude over . For the The real part 1 p 2 optical properties, we obtain p (6.44) ( ) p . (6.45) n( ) 2 2

We nd ( ) n( ) 1, which implies a large reectivity 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 reectivity 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
2 p

(6.46)

such that the reectivity 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 ( ) =
2 p

(6.47)

This inuences the reecting properties of metals; particularly, the value of p is reduced to p = p / , where is the frequency-independent part of the dielectric function. With this, the reectivity for frequencies above p approaches R = ( 1)2 /( + 1)2 , and 0 < R < 1 (see Figure 6.2 and 6.3). Color of metals The practically full reectance 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 reectance on the whole visible frequency range, while for colored metals there is a certain threshold above which the reectance drops and frequencies towards blue are not or much weaker reected. 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 reectance drop is caused by the transition from the 117

completely occupied d-band to the partially lled s-band, 3d 4s in case of Cu. For copper, this drop appears below 2.5eV so that predominantly red light is reected (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 reectance. On the other hand, aluminum shows a reectance rather close to the expected behavior (see. Figure 6.3). Also here, there is a small reduction of the reection due to interband absorption. However, this eect is weak and the strong drop occurs at the plasma frequency of p = 15.8eV . Like silver also polished aluminum is white shiny.

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, k ) and the relaxation time . This relation, f (k) f0 (k) = ( k) d3 k W (k, k ){f (k) f (k )}, (2 )3 (6.48)

will be studied for an isotropic system, with elastic scattering, and a small external eld E . The solution of equation (6.24) is of the form f (k) = f0 (k) + A(k )k E , such that f (k) f (k ) = A(k )(k k ) E . 118 (6.50) (6.49)

llclpoJd eqt ot

Figure 6.2: Reectance spectra for silver and copper. In both cases the drop of reectance is due to optical transitional between the completely lled d-band and the partially lled s-band. Note the logarithmic scale for the reectivity. (Source: An introduction to the optical spectroscopy of inorganic solids, J. Garc a Sol e, L.E. Baus a and D. Jaque, Wiley (2005))

slels 3o {lrsuaP soLule le JnJJO uJec eql ^\oleq tndpelleJ-os eql

l uleldxe urlceds rqlIA\ pelelJossu el qJIq^\ 'se8pe

lo,{ltutctn eql ut l dlp eql 'snqJ ul uI luelJgJeoJ

'l ]e ueql Je^\ol ql se 'JeAe,^AoH e /Y\o[eqspueq

ul uI suolllsueJl ,rt8 'lE '{8leue g'lepotu JoIrIuJ Ae E'l lnoqu l !

ueaqe^eq pa>lretu
J d' ulnutulnle JO

e fg n J e J !J ur drP r,r.t1ceger

ql JoJsJunocJu

Puu (nJ) ,Le Z 't) uorlenbg ,(q t oz{leue oI 9Z 9202910190

'Q961'ddrpq4pue qJIaJuoJqE qlr,,r,r ruor; uoISSrr-urad pecnpo:de:) esecqlee ur peluJrpur sr ,{cuanbe:.; eusuld eql ol Surpuodse"rroc ,{3reue uoto{d aq1 ':eddocpue Je^lrs;o erlceds,(1r,trlcegar aq1 91'p arn8rg

polelnrler '(Ae 6 - dcD4)3y pue (na 8'0t :'totl) nJ ot Surpuodserrot serSreua 'y uoloqd etuseld eqJ 're^lrs pue eJt3eds{1r,tr1ceger

0z

(1e) {6reu3

9t

A ' 8'0t - " rr 4

---l,, I
nc

l a $= "4! t

0y

0t

OL n o = o 001 o

001 OL

t' 0

100

L'0 L0'0 + L
\v

=.

AND INSULATORS SEMICONDUCTORS

::-: . :.lle tillS - .. : . . h or t er . , :: - i r t ha n
. r rf tl g - :. L lp Io . t. 1nc es _ _.

r27

1.0 0.8

a 0 .6
g
o E
o

0.4 0.2

1 -l -),

0.0

10

15

20

25

hot(eV)
*. -- i I

(full line)compared predicted withthose spectrum of Aluminum Figure4.5 Thereflectivity

with f : anda damped oscillator modelwith ltato: 15.8 eV (dotted fromtheideal metal Figure 6.3: Reectance spectrum of aluminum. Theline) slight reduction of reectivity below p is withpermission fromEhrenreich line)(experimental data reproduced 1.25x 16ta r-t (dashed -:.1+ due to interband transitions. The thin solid line is the theoretical behavior for = 0 and the et al.,1962). dashed line for nite . (Source: An introduction to the optical spectroscopy of inorganic solids, J. Garc a Sol e, L.E. Baus a and D. Jaque, Wiley (2005)) with of aluminumis compared reflectivityspectrum In Figure4.5,theexperimental thosepredictedby the ideal metal andthe dampedmetal models.Al hasa free electron per atom)andso,according .*-3 (threevalence electrons densityof ly' : 18.I x 1922 l5.SeV.Thus,thereflectivityspectrum toEquation(4.20),itsplasmaenergyisltc,;o: Without loss of generality, we dene z k, and introduce the parametrization of the angles , for the ideal metal can be now calculated.Comparedto the experimentalspectrum, polar angle of Ethe and ( ) polar (azimuth) k ), leading to into account when taking slightly of improved ideal metal model spectrumis only angle from DC conductivity the dampingterrn,with f : I.25 x 10la s-1, & valuededuced kThe Emain = kE cos , between the two calculated spectraare that (6.51) differences measurements. the ultraviolet below op and less one a reflectivity slightly than dampingproduces k k = kk cos , (6.52) out. transmission edgeis slightly smoothed k E = k E (cos cos the + sin metal sin model cos ).the damped (6.53) nor neither ideal be mentioned that Finally, it should metal model are able to explain why the actualreflectivity of aluminum is lower than For elastic scattering, k = one k , (R wery obtain 1) at frequencies lower thanrt r. Also, thesesimple models the calculated do not reproducefeaturessuchas the reflectivity dip observedaround 1.5eV. In order (k) for f (these k)= A(k ) kEthen cos (1 cos understanding ) sin sin cos . (6.54) of real metals, have and to a better to f account aspects, at the end of the band structuremust be taken into account.This will be discussed Inserting this into the right-hand side in Section 4.8. of equation (6.48), the -dependent part of the integration this chapter,

vanishes for an isotropic system, and we are left with f (k) f0 (k) 4.5 SEMICONDUCTORS = dk AND [f (k)INSULATORS f (k )]W (k, k ) ( k) (6.55) (6.56) (6.57)

and insulatorshave bound valenceelectrons.This Unlike metals,semiconductors next of this and givesrise to interband transitions. aspect = A(k )kE cos The objective dk (1 cos the )W (ksection , k ) is

= [f (k) f0 (k)]

dk (1 cos )W (k, k ).

The factor [f (k) f0 (k)] can be dropped on both sides of the equation, resulting in 1 = ( k) d3 k W (k, k )(1 cos ), (2 )3 (6.58)

where one should remember that, for elastic scattering, the quasiparticle energy k = k is conserved in the collision process. The scattering probability W (k, k ) accounts for this restriction. In the next few sections we discuss dierent 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, k ) for a potential V can be determined applying Fermis golden rule,7 W (k, k ) = 2 nimp | k |V |k |2 (
k

).

(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 eects between dierent impurities. According to equation (6.58), the relaxation time of a quasiparticle with momentum k is given by 1 2 = nimp ( k) d3 k k ) ( | k |V |k |2 (1 k k (2 )3 d k ) dk , v ) (k, k )(1 k = nimp (k k d 4
k

(6.60) (6.61)

= k/|k|. Here, we used the connection8 with the dierential scattering cross section d/d and k between Fermis golden rule and the Born approximation. Note the dierence in the expressions for the relaxation time in equation (6.58) and for the lifetime , 1 = d3 k W (k, k ), (2 )3 (6.65)

given by Fermis golden rule. The factor (1 cos ) in equation (6.58) gives more weight to backscattering ( ) compared to forward scattering ( 0), since the former has more inuence in impeding transport. This explains why is also termed transport lifetime. Assuming defects in the form of point charges Ze, whose screened potential is k |V |k = 4Ze2 2 . |k k |2 + kTF (6.66)

In the limit of very strong screening, kTF kF , the dierential cross section becomes independent of the deviation (k k ), the transport and the usual lifetime become equal, = , 9 and 1 N ( F )nimp 4Ze2 2 kTF
2

(6.67)

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 rst Born approximation in scattering theory. Note, that this approximation is insucient to describe resonant scattering. 8 The scattering of particles with momentum k into the solid angle dk around k yields

W (k, k )dk =

2 2

| k |V |k |2 (
k dk

) 2 dk N ( )| k |V |k |2 .

(6.62)

dk
k dk

d3 k k |V |k | 2 ( (2 )3

)=

(6.63)

The scattering per incoming particle current jin d (k, k ) = W (k, k )dk determines the dierential cross section v k d (k, k ) = 2 N ( ) | k |V |k |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. (6.64)

120

or, for the specic resistivity = 1/ , = m e2 n (

F

(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 dened 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 eect

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 o magnetic impurities with a spin degree of freedom, yielding a dramatic energy dependence of the scattering rate. This problem was rst 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 VK =
i

VK

=J
i

S i s(r ) (r Ri )

(6.70) (6.71) (6.72)

=J
i

1 + 1 + z z s (r ) + Si Si s (r ) + Si s (r ) (r Ri ) 2 2
+ i(kk )Ri z . (ck ck c Si k ck ) + Si ck ck + Si ck ck e

J = 2

k,k ,i

Here, it becomes important that spin ip 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, W (k, k ) J 2 S (S + 1) 1 + 2JN ( F ) ln where D is the bandwidth and we have assumed that JN ( F ) to be J 2 S (S + 1) 1 N ( ) 1 + 2JN ( F ) ln ( k) | D | k | , (6.73)

1. The relaxation time is found D k

(6.74)

Note that W (k, k ) 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 eect involving the electrons very near the Fermi surface. The fact that the local spins S i can ip, makes the scattering center dynamical, because the scatterer is constantly changing. The scattering process of an electron is inuenced by previous scattering events, leading to the singularity at F . This cannot be described within the rst Born approximation, but requires at least the second approximation or the full solution.10 As mentioned before, the resonant
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.
10

121

behavior induces a strong temperature dependence of the conductivity. Indeed, (T ) =

3 e2 kF 6 2 m

e2 n 8mkB T

1 4kB T cosh (
2

)/2kB T )

( )

(6.75) (6.76)

J 2 S (S + 1)[1 2JN ( F ) ln(D/)] . cosh2 (/2kB T )

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 nite 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 nite temperature. The thermally induced distortions of the lattice, phonons, act as uctuating scattering centers. In the language of electron-phonon interaction, electrons are scattered via absorption and emission of phonons, which induce local uctuations in volume (cf. Chapter 3). The corresponding coupling term was given in equation (3.143) and simplies with the denition (3.130) to Hint = 2i V q
k,q ,s |q |(bq b q )ck+q ,s cks .

20 q

(6.79)

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 k due to the emission of a phonon with momentum q , we have k = k + q + G,
k

(6.80) (6.81)

= q +

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 . In order to calculate F
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,

k + q ; Nq |(bq b q )ck+q ,s cks |k; Nq = k + q |ck+q ,s cks |k

Nq N

,Nq 1

q,q q,q ,

Nq + 1 N need to be calculated. From Fermis golden rule we obtain f t = 2

q

,Nq +1

(6.82)

coll

|g (q )|2 f (k)(1 f (k + q ))(Nq + 1) f (k + q )(1 f (k))Nq ( f (k + q )(1 f (k))(Nq + 1) f (k)(1 f (k + q ))Nq (

k+q k +q

+ q )

q ) ,

(6.83)

|q | 2 / . Each of these four terms describes one of the single phonon where g (q ) = V q 0 q scattering processes depicted in Figure 6.4. k+q q k k+q k q k+q k q k k+q q

Figure 6.4: The four single-phonon electron-phonon scattering processes. The collision integral leads to a complicated integro-dierential 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, k =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 t =
coll

2
q

|g (q )|2 2N (q )[f (k + q ) f (k)] (

k +q

k ),

(6.84)

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 ).
In analogy to the discussion on electromagnetic radiation, the phenomenon of spontaneous phonon emission due to zero-point uctuations appears. It is formally visible in the additional +1 in the factors (Nq ) + 1.
12

123

In analogy to previous approaches, we obtain for with relaxation-time Ansatz 1 2 = ( k) N( F) d3 q N (q )(1 cos ) ( (2 )3 q
k+q

k ),

(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 kF k q

Figure 6.5: The geometry of electron-phonon scattering. In order to perform the integration in equation (6.86), we have to re-express ( writing
2

k+q

k)

by

k +q

k)

2m

(q 2 2kF q cos )

m
2k

q F

q cos , 2kF

(6.88)

where is dened in Figure 6.5. From there, we also see that 2 + = , and thus, nd 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 = ( F , T) N( F) = m 2 k F
2kF /2

dq qq N (q )
0 2kF 0

d sin cos2 ( )

q cos 2kF

(6.90)

mcs 3 4N ( F ) 2 kF

q 4 dq e
cs q/kB T

1 y 4 dy , ey 1

(6.91)

2 mcs kF = 4N ( F ) 2

T D

2D /T

(6.92)

124

where we have approximated the Debye temperature by kB D 2 cs kF . We notice the two distinct characteristic temperature regimes, kB D T 5 , T D , 6 (5) 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 nally 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 D , T , T (T ) (6.96) 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 deects 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, the lifetime of a quasiparticle strongly increases close to the Fermi surface. The basic reason was the constraint of the scattering phase space due to the Pauli principle. The lifetime, which we can identify with the relaxation time here, has the form 1 ( ( )
F

)2 .

(6.98)

Note that this phase space reduction does not restrict the momentum transfer, so that in principle 0 < q < 2kF is valid for all energies . This allows determining the conductivity from equation (6.29), and we nd (T ) 1 T2 (6.99)

The resistivity T 2 increases quadratically in T . This is a key property of a Fermi liquid and is often considered an identifying criterion.

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 nite 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 satised 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 = k + G, (6.100)

where G is a reciprocal vector of the lattice, allows for scattering to other Brillouin zones (G = 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)).

)&6'+2&7$ "324-&'+5$

./01(211$$ "324-&'+5$

~ k0

~ k 00
~ k0 ~ q

~ k+~ q

~ k 00 ~ q

!"#$%&'(()*'+$,)+-$

~ k

Figure 6.6: Considering energy and momentum conservation in two-electron collision there are ordinary scattering processes (k, k ) (k + q , k q ) which does not lead to a momentum relaxation, and there are Umklapp scattering processes (k, k ) (k + q , k q ) leading to a neighbouring Brillouin zone, here visualized by an extended Brillouin zone scheme.

6.6

Matthiessens rule and the Ioe-Regel limit

Matthiessens rule states, that the scattering rates of dierent scattering processes can simply be added, leading to W (k, k ) = W1 (k, k ) + W2 (k, k ), or, expressed in the relaxation time approximation, 1 1 1 = + , 1 2 and = m m = 2 ne2 ne 1 1 + 1 2 126 = 1 + 2 . (6.103) (6.102) (6.101)

This rule is not a theorem and corresponds eectively to a serial coupling of resistors in a classical circuit. It is applicable, if the dierent 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 inuences of impurities, e.g., through interference eects due to the coherent scattering of an electron at dierent impurities, would invalidate this simplication. An example where Matthiessens rule is violated is a one-dimensional system, where a single scatterer i induces a nite resistance Ri . Two serial scatterers then lead to a total resistance R = R1 + R2 + 2e2 R R R1 + R2 . h 1 2 (6.104)

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 , (6.105)

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 modied 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 ) (6.106) (6.107)

= 0 + T 5 + (1 + 2JN ( F ) ln(D/kB T )) + T 2 ,

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 indenitely with temperature. In most cases, however, the resistivity will saturate at a nite 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 k . If the de Broglie wavelength becomes comparable to the 1 would become a boundary for . In mean free path, the framework becomes unfounded and kF 1 most systems a and kF are comparable lengths. Empirically, the resistivity is described via the formula 1 1 1 = + , (T ) BT (T ) max (6.108)

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 eects will arise. The saturated resistivity max can be estimated from the Jellium model, max = 3 2 m h 3 m = = 2 2 2 2 3 e n ( F ) e kF ( F ) e 2kF h 3 2 , e 2kF 127 (6.109) (6.110)

where we used 1 kF . For a typical value kF 108 cm1 of the Fermi wave vector, we nd max 1mcm, which is called the Ioe-Regel13 limit. Establishing a quantitative estimate of max for a given material is often dicult. There are even materials whose resistivity surpasses the Ioe-Regel limit.

6.7

General transport coecients

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 )), where f0 (k, T (r ), (r )) = 1 e(
k (r ))/kB T (r )

(6.111)

+1

(6.112)

In addition, we require the charge density to remain constant in space, i.e., d3 k f (k; r ) = n0 (2 )3 (6.113)

for all r . In this section, we introduce the electrochemical potential (r ) = e(r ) + (r ) generating the general force eld E = (e + ), where (r ) denotes the electrostatic potential which produces the electric eld E = . With this, the Boltzmann equation for the stationary situation reads f t f = v k r f + k k = f v k k
k

(6.114) (6.115)

coll

r T E . T

In the relaxation time approximation for the collision integral, we obtain the solution f (k) = f0 ( k )v k E k
k

r T T

(6.116)

which allows us to calculate the charge and heat currents, J e = 2e Jq = 2

13

d3 k v f (k), (2 )3 k d3 k ( )v k f (k), (2 )3 k

(6.117) (6.118)

The saturated resistivity max 1mcm= 1000cm should be compared to the room-temperature resistivity of good conductors, metal [cm] which are all well below max . Cu 1.7 Au 2.2 Ag 1.6 Pt 10.5 Al 2.7 Sn 11 Na 4.6 Fe 9.8 Ni 7 Pb 21

128

a)

b)

ky

ky

fk kx kx

fk

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 eld along the negative x-direction, on the right panel (b) for a temperature gradient in x-direction.

respectively. Inserting the solution (6.116) into the two denitions above yields e (1) K (T ) , T (1) E + 1 K (2) (T ) , J q = K T (0) E J e = eK (n) (n N ) are dened as where should be understood as r and the tensors K 0 K =
(n)

(6.119) (6.120)

1 4 3

f0 ( )( )n

dk

vF vF |v F |

2 . kF

(6.121)

For an isotropic system these transport coecients are no longer tensors but represented by scalars, n f0 K (n) = 0 d ( )( )n (6.122) m In the case T TF we can calculate the coecients,14 K (0) ( F ) = n0 ( F ) ( F ) m 2 (0) K (1) ( F ) = (kB T )2 K ( ) 3 K (2) ( F ) = 2 (k T )2 K (0) ( F ). 3 B (6.125) ,
=
F

(6.126) (6.127)

We measure the electrical resistivity assuming thermal equilibrium, T = 0 for all r . With this, we nd the expression = e2 K (0) .
If a function g ( ) depends only weakly on in the vicinity of to derive a general approximation for following integrals and d g ( )(
F F

(6.128)
F

14

, we can use the Sommerfeld (Taylor) expansion

d g( )

f0 = g(

)+

2 g( ) 2 ( kB T ) 2 6 2

+ ...
= F

(6.123)

g ( ) f0 2 = ( kB T ) 2 3

+ ...,
= F

(6.124)

in the limit T 0. We used that

in that asymptotic case.

129

In order to determine the thermal conductivity relating the heat current J q to T when no external electric eld E is applied, we set J e = 0 as for an open circuit. Then, the equations (6.119) and (6.120) reveal the appearance of an general force eld E= Thus, the heat current is given by Jq = 1 T K (2) K (1)2 K (0) T = T. (6.130) K (1) T. T K (0) (6.129)

In simple metals, the second term in (6.130) is often negligible as compared to the rst one and we obtain in this case 2 2 2 1 (2) 2 kB (0) = kB T = K = TK , (6.131) T 3 3 e2 which is the well-known Wiedemann-Franz law. Note, that we can write the thermal conductivity in the form C = 2 , (6.132) e N( F)
2 T /3 denotes the electronic specic heat. where C = 2 N ( F )kB

6.7.2

Thermoelectric eect

Equation (6.129) shows, that a temperature gradient induces an electric eld. For a simple, isotropic system, this relation reduces to E = QT = with the Seebeck coecient Q=
2T 2 kB ( ) 3 e ( )

K (1) T, T K (0)

(6.133)

.
=
F

(6.134)

This is the so-called Mott formula which looses its validity at high-temperatures or very anisotropic scattering. Using ( ) = n( )e2 ( )/m, we investigate ( ), ( )= ( ) n( ) ( ) N( ) ( ) + ( ) = ( ) + ( ), ( ) n( ) ( ) n( ) (6.135)

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 eect). In the opposite situation, namely, when the rst term is irrelevant, the Seebeck coecient 2 T N( ) 2 kB S F Q= = (6.136) 3 e n( F ) ne 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 (6.136) Q=
2T 2 kB 2 kB T = 2 e F 2 eTF

(6.137)

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 dierent. Dierences occur through phonon eects, such as the so-called phonon drag which we have neglected here. In the following, we consider two dierent types of thermoelectric eects. 130

100

Cs

Li

50

Q [nV/K]

Na

50 0 1 2

K
T [K]

Figure 6.8: Seebeck coecient for the Alkali metals Li, Na, K, and Cs at low temperatures. The dashed line represents the estimate for Na and K following Eq.(6.137). (adapted from D.K.C. MacDonald, Thermoelectricity: an introduction to the principles, Dover (2006).)

Seebeck eect The rst is the Seebeck eect, where a thermoelectric voltage appears in a bi-metallic system (cf. Figure 6.9). With equation (6.133), a temperature gradient across metal B induces an electromotive force15 UEMF = dl E
T1 T2 T0

(6.138) dl T + QA
T1 T2

= QA
T0

dl T + QB

dl T

(6.139) (6.140)

= (QB QA )(T2 T1 ).

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 conguration was chosen to reveal voltage dierences across the contacts which are absent in a single metal.
The term electromotive force, rst introduced by Alessandro Volta, is misleading in the sense, that it measures a voltage instead of a force.
15

131

a) T2 E metal B T1 metal A T0 Vtherm T0 metal A Je

b)

Jq metal A T0 metal B T0 Jq T0 Je T0 metal A

Figure 6.9: Schematics of thermoelectric eects. On the left panel (a) a representation of Seebeck eect is given, where the symbol E is used instead of E . On the right panel (b) the Peltier eect is represented. In our analysis, both systems were eectively one-dimensional.

Peltier eect The second phenomenon, termed Peltier eect, 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 one reservoir (top) to another (bottom). This follows from the equations (6.119) and (6.120) by assuming T = 0, where 2 (0) Je = e K E (6.141) Jq = K (1) E implies Jq = K (1) J = Je = QT Je . K (0) e (6.142)

The coupling = T Q between Jq and Je is called Peltier coecient. 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 . (6.143)

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 eects which both represent bulk eects of the two metals A and B. By no means, it should be mistaken as an eect originating from the inter-metal contacts.

6.8

Anderson localization in one-dimensional systems

Transport in one spatial dimension is very special, since there are only two dierent 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 eects of multiple scattering at dierent 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 reection at an arbitrary potential with nite support16 in one dimension can be described by a transfer matrix T .
16

The support (dt. Tr ager) of a function is the set of all points, where the function takes non zero values.

132

V a1+ I1 I2 a1 x Figure 6.10: Transfer matrices are sucient to describe potential scattering in one dimension. T a2 a2+

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 1 (x) = a1+ eikx + a1 eikx , 2 (x) = a2+ eikx + a2 eikx , (6.144) (6.145)

where 1 (2 ) is dened in the area I1 (I2 ). The vectors ai = (ai+ , ai ) i {1, 2} are connected via the linear relation, a = a2 = T 1 T11 T12 T21 T22 a1 , (6.146)

. The conservation of current (J = J ) requires that T is with the 2 2 transfer matrix T 1 2 unimodular, i.e., det T = 1. Here, d (x) d (x) (x) (x) , (6.147) dx dx such that, for a plane wave (x) = (1/ L)eikx in a system of length L, the current results in J= i 2m J = v/L (6.148)

with the velocity v dened as v = k/m. Time reversal symmetry implies that, simultaneously with (x), the complex conjugate (x) is a solution of the stationary Schr odinger equation. From this, we nd T11 = T22 and T12 = T21 , such that = T T11 T12 T12 T11 . (6.149)

It is easily shown that a shift of the scattering potential by a distance x0 changes the coecient by a phase factor ei2kx0 . Meanwhile, the coecient T remains unchanged. T12 of T 11 With the Ansatz for a right moving incoming wave ( eikx ), producing a reected ( 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 can be determined explicitly in this situation, resulting in The coecients of T = T
1 t r t r t 1 t

(6.150) (6.151)

(6.152)

133

Here, the conservation of currents imposes the condition 1 = |r|2 + |t|2 . Furthermore, we can nd 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 reected 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 (6.153) (6.154) (6.155)

with the velocity v = k/m, the electron charge e, and the particle densities n0 , nr , and nt corresponding to the incoming, reected 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 1 + |r |2 , L (6.156) (6.157)

From this consideration, a density dierence 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 dened by the ratio between the voltage drop over the resistor V and the transmitted current Jt , i.e. R= V Jt (6.158)

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 dierence E = eV between the two sides of the resistor, such that the expression n = dn E dE dn = (eV ) dE (6.159) (6.160)

produces the wished relation. Here, interval [E, E + dE ] and we nd 1 dn = dE L E

dn dE dE

is the number of states per unit length in the energy

2 k2 dk E 2 2m

2 k2

k,s

2m

=2

1 . v (E )

(6.161)

The resistance R is nally obtained from the equations (6.158), (6.159), and (6.161), leading to R= h |r|2 , e2 |t|2 (6.162)

The Klitzing constant RK = h/e2 25.8k is a resistance quantum named after the discoverer of the Quantum Hall Eect. The result (6.162) 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.

134

T1

T2

and T Figure 6.11: Two spatially separated scattering potentials with transmission matrices T 1 2 respectively.

6.8.2

Scattering at two impurities

and T each We consider now two spatially separated scattering potentials, represented by T 1 2 determined by r1 , t1 and r2 , t2 respectively. The particles are multiply scattered at these potentials in a unknown manner, but the global = T T result can again be expressed via a simple transfer matrix T 1 2 , given by the matrix multiplication of each transfer matrix. All previously found properties remain valid for the new , given by matrix T r r1 r2 r1 2 1 1 r + t t t t t t 1 t2 1 t2 1 t2 1 t2 = T = . (6.163) r1 r2 1 r r1 r2 1 t t t t t t t t + t t
1 2 1 2 1 2 1 2

For the ratio between reection and transmission probability we nd |r|2 1 = 2 1 2 |t| |t| (6.164) (6.165) 1. (6.166)

t 2 r1 r2 1 2 = 1+ 1 |t1 |2 |t2 |2 t 2 t r t r1 r2 r1 1 2 2 2 2 2 1 + | r | | r | + + = 1 2 |t1 |2 |t2 |2 t t 2 2

Assuming a (random) distance d = x2 x1 between the two potential barriers, we may average over this distance. Note, that for x1 = 0, we nd r2 e2ikd , while r1 , t1 , and t2 are independent vanish after averaging over on d. Consequently, all terms containing an odd power of r2 or r2 d. The remainders of equation (6.166) can be collected to |r|2 |t|2 =
avg

|t1 |2 |t2 |2 |r1 |2 |r2 |2 + |t1 |2 |t2 |2

1 + |r1 |2 |r2 |2 1 +2 |r1 |2 |r2 |2 . |t1 |2 |t2 |2

(6.167) (6.168)

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 (6.169)

with RK = h/e2 This result is a consequence of the unavoidable multiple scattering in one dimensions. This eect is particularly prominent if Ri h/e2 for i {1, 2}, where resistances are then multiplied instead of summed.

135

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 innitesimal is found from equation (6.169), resulting in dR = d + 2 which yields
R( )

R( ) d , RK

(6.170)

d =
0 0

dR , 1 + 2R/RK

(6.171)

and thus, = h ln 1 + 2R( )/RK . 2e2 (6.172)

Finally, solving this equation for R( ), we nd R( ) = RK 2 e 2

/RK

1 .

(6.173)

Obviously, R grows almost exponentially fast for increasing . This means, that for large , the system is an insulator for arbitrarily small but nite > 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 innitely many bound states with dierent energy exist. The mean localization length of individual states, related to the mean extension of a wave function, is found from equation (6.173) to be = /RK . The transmission amplitude is reduced17 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 states18 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 eects of multiple scattering are far less drastic and the Ohmic law is applicable. Localization eects in two dimensions is a very subtle topic and part of todays research in solid state physics.

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. 18 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.

17

136

Chapter 7

Magnetism in metals
Magnetic ordering in metals can be viewed as an instability of the Fermi liquid state. We enter this new behavior of metals through a detailed description of the Stoner ferromagnetism. The discussion of antiferromagnetism and spin density wave phases will be only brief here. 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 eect 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. On the other hand, 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 insucient to trigger magnetism. It is, however, possible to nd 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 specic 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 .

137

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 rst 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 a, m 1 + F0 (7.1)

a 1 and leads to a correlation driven instability of the Fermi which obviously diverges for F0 liquid, as we will discuss here.

7.1.1

Stoner model within the mean eld approximation

Consider a model of conduction electrons with a repulsive contact interaction, H=

k cks cks k,s

+U

d3 r d3 r (r ) (r r ) (r ),

(7.2)

where we use the electron density s (r ) = s (r )s (r ) and the eld operator s (s ) follows from the denition (3.28) [(3.29)]. 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 two-particle correlation between electrons of identical spin (cf. Figure 3.1). The derivation of the full solution of this model goes beyond the scope of this lecture. However, a mean eld approximation will already provide very useful insights 1 . We rewrite,

s (r ) = ns + [s (r ) ns ], where ns = s (r ) ,

(7.3)

(7.4)

and represents the expectation value of the argument. We stipulate that the deviation from the mean value ns shall be small. In other words [s (r ) ns ]2 n2 s. (7.5)

Inserting equation (7.3) into the Hamiltonian (7.2), we obtain Hmf = =

k,s k cks cks k,s

+U

d3 r [ (r )n + (r )n n n ]

(7.6) (7.7)

+ U ns ) cks cks U n n ,

the so-called mean eld 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 are ignored here. The advantage of this approximation is, that the many-body problem is now reduced to an eective one-particle problem, where only
Note that the following mean eld calculation is equivalent to a variational approach using simple many-body wavefunction (Slater determinant) with dierent concentrations of up and down spins.
1

138

the mean electron interaction is taken into account. This is equivalent to a generalized HartreeFock approximation and enables us to calculate certain expectation values, such as the density of one spin species n = = = 1 c k ck = 1 ( 1
k

f(
k

+ U n )

(7.8) (7.9) (7.10)

d d

U n )f ( )

1 N ( U n )f ( ). 2

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 eld 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 dened via n + sm 1 (n + n ) + s(n n ) = 0 , 2 2 where n0 is the total particle density. This leads to the two coupled equations ns = n0 = 1 2 1 m= 2 d d N ( U n ) + N ( U n ) f ( ), N ( U n ) N ( U n ) f ( ), (7.11)

(7.12) (7.13)

or equivalently n0 = m= 1 2 1 2 d N
s

U n0 Um s 2 2 U n0 Um s 2 2

f ( ), f ( ),

(7.14) (7.15)

s
s

d N

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 . In this limit, the equations (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 the equations (7.14) and (7.15) has been absorbed into F . The Fermi-Dirac distribution takes the form f( ) = 1 e [ (m,T )]
2

+1

(7.17)

where = (kB T )1 . After expanding equation (7.14) for small m, one obtains n0
F

d f( ) N( ) +

1 2

Um 2

N ( ) 2 1 (kB T )2 N ( F ) + 6 2 Um 2
2

(7.18)

d N ( ) + N ( F ) +
0

N ( F ),

(7.19)

139

where we introduced the abbreviations N ( ) = dN ( )/d and N ( ) = d2 N ( )/d 2 . Since the rst term on the right side of equation (7.19) is identical to n0 , (m, T ) is immediately found to be given by (m, T ) N ( F ) 2 1 (kB T )2 + N( F) 6 2 Um 2
2

(7.20)

Analogously, the expansion of equation (7.15) in m and T , results in m Um 1 d f( ) N ( ) + N ( ) 2 3! 2 1 (kB T )2 N ( F ) + 6 3! Um 2 Um 2

3

(7.21)
2

N( F) +

N ( F ) + N ( F )

Um 2

(7.22)

and nally, inserting the result for from (7.20) into (7.22), we nd m = N( F) 1 where
2 ( F) = 1 2

2 2 (k T )2 1 ( F) 6 B

Um 2

2 N ( F )2 ( F)

Um 2

(7.23)

N ( F) N( F)

N ( F) , N( F)

and

2 ( F) = 2

1 2

N ( F) N( F)

N ( F) . 3!N ( F )

(7.24)

The structure of equation (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 conguration is shifted by U m/2, resulting in a nite total magnetization. In our situation, this condition corresponds to 1 2 2 1 = U N ( F ) 1 (kB TC )2 1 ( F) , 2 6 yielding kB TC = 6 1 ( F ) 1 2 UN( F) 1 Uc U (7.27) (7.26)

140

for U > Uc = 2/N ( F ). This is an instability condition for the nonmagnetic 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 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 nite 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 uctuations destroy the ordered state at some nite temperature via entropy increase, entropy is irrelevant at T = 0. The order is now suppressed by quantum uctuations (Heisenbergs uncertainty principle).
T 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 an external 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 insucient 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)
50

T(K)

UGe2 FM

1000

Fe fcc

Fe
Fe hpc

Fe bcc FM

0.5

1.5

p (GPa)

10

20

p (GPa)

Figure 7.3: Phase diagrams of UGe2 and Fe.

141

uous, rst 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. Let us eventually 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 instability 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. Copper is known to be a better conductor than Ni for the same reason.

7.1.3

Spin susceptibility for T > TC

Next, we study the response of a metallic system to an external perturbation. For this purpose, we apply an innitesimal magnetic eld H along the z -axis, which induces a spin polarization due to the Zeeman coupling. HZ = gB d3 r H 1 (r ) (r ) 2 . (7.29)

From the self-consistency equations (7.14) and (7.15) we obtain m= 1 2 d f( )

s

sN

B sH s

Um 2

(7.30) (7.31)

d f ( )N ( )

Um + B H 2 Um + B H 2

= N( F) 1

2 (k T )2 1 ( F )2 6 B

(7.32)

2 The Stoner instability is a simplication of the quantum phase transition. In most cases, a discontinuous phase transition originates in the band structure or in uctuation eects, which were ignored here, For more details consult [D. Belitz and T.R. Kirkpatrick, Phys. Rev. Lett. 89, 247202 (2002)].

142

to lowest order in m and H . Solving this equation for m yields M = B m = 0 (T ) H, 1 U 0 (T )/22 B (7.33)

and consequently, the magnetic susceptibility reads = 0 ( T ) M = , H 1 U 0 (T )/22 B 2 (k T )2 1 ( F )2 . 6 B (7.34)

where the bare susceptibility 0 is given by 0 (T ) = 2 N( F) 1 B (7.35)

We see, that the denominator of the susceptibility (T ) vanishes exactly when the Stoner instability criterion is fullled. The diverging susceptibility (T ) 0 (TC )
2 TC T2

(7.36)

indicates the instability. Note that for T TC from above, the susceptibility diverges like (T ) |TC T |1 corresponding to the mean eld behavior, since the mean eld 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. Examples are spin density wave (SDW) states, antiferromagnets and spin spiral states. In this section, we analyze general instability conditions for metallic systems.

7.2.1

General dynamic spin susceptibility

We consider a magnetic eld, oscillating in time and with spacial modulation H (r , t) = H 0 eiqrit et , (7.37)

and calculate the resulting magnetization, for the corresponding Fourier component. For that, we proceed analogously as in Chapter 3 and dene the spin density operator S (r ) in real space, (r ) (r ) + (r ) (r ) S (r ) = (7.38) i (r ) (r ) + i (r ) (r ) s (r ) ss s (r ) = 2 2 s,s (r ) (r ) (r ) (r ) with momentum space representation Sq = d3 rS (r )eiqr = 2 c k,s ss ck+q ,s =
k,s,s

S k ,q ,
k

(7.39)

where S k,q = ( /2)c k,s ss ck+q ,s . The Hamiltonian of the electronic system with contact interaction is given by H = H0 + HZ + Hint , 143 (7.40)

where H0 =
k cks cks , k,s

(7.41) (7.42) (7.43)

HZ = Hint = U

gB

d3 rH (r , t) S (r ),

d3 r (r ) (r ).

The operator HZ describes the Zeeman coupling between the electrons of the metal and the perturbing eld. We investigate a magnetic eld 1 1 H = H + (q , )eiqrit et i (7.44) 2 0 in the xy -plane. The Zeeman term then simplies to HZ = gB
H + (q , )S q + h.c.,

(7.45)

where the Hermitian conjugate (h.c.) part will be ignored in the following. We use
S q =

c k ckq .

(7.46)

+ In the framework of linear response theory, this coupling will induce a magnetization Mind = + (B / ) Sind (q , ) . Using the same equations of motion as in Section 3.2,

+ + S = [Sk ,q , H], t k,q

(7.47)

+ with Sk ,q = ck ck+q , we can determine this induced magnetization, rst without the interaction term (U=0). We obtain for the given Fourier component,

+ S (t) = ( t k,q k,q

k +q

+ k )Sk,q (t)

+ it+t . g B (c k ck ck+q ck+q )H (q , )e

(7.48)

Performing the Fourier transform from frequency space to real time, and applying the thermal average we obtain, (
+ + i ) Sk ,q = g B (nk+q nk )H (q , ),

k +q

(7.49)

which then leads to the induced spin density ( magnetization),

+ Sind (q , ) =

+ Sk ,q = k

0 (q , )H + (q , ),

(7.50)

with 0 (q , ) = g2 B nk+q nk
k k +q

(7.51)

Note that the form of the bare susceptibility 0 (q , ) is similar to the Lindhard function (3.63). The result (7.50) describes the induced spin density within linear response approximation. In the next step, we want included the eects of the interaction. Analogously to the charge density wave state found in Section 3.2, the induced spin density generates in turn an eective 144

eld. The induced spin polarization appears in the exchange interaction as an eective magnetic eld. We rewrite the contact interaction term in equation (7.40) in the form Hint = U
c k+q ck ck q ck c k ck+q ck ck q + const. + Sq Sq .

k , k ,q

= =

(7.52)

k ,k ,q

U 2

+ The induced spin polarization Sind (q , ) acts through the exchange interaction as an eective + + (local) eld, as can be seen by replacing Sq Sind (q , ) q,q in Eq.(7.53) ,

U 2

+ Sq Sq

U
2

S + (q , ) S q =

gB

+ Hind (q , )S q

(7.53)

+ where the eective magnetic eld Hind (q , ) nally reads + Hind (q , ) =

U gB

S + (q , ) .

(7.54)

This induced eld acts on the spins as well, such that the total response of the spin density on the external eld becomes M + (q , ) = B S + (q , )
+ = 0 (q , )[H + (q , ) + Hind (q , )] U S + (q , ) = 0 (q , )H + (q , ) + 0 (q , ) gB U = 0 (q , )H + (q , ) + 0 (q , ) 2 M + (q , ). gB

(7.55)

With the denition M + (q , ) = (q , )H + (q , ) (7.56)

of the susceptibility in the random phase approximation (RPA) (see Section 3.2), we nd (q , ) = 0 (q , ) . U 1 2 2 0 ( q , )
B

(7.57)

This form of the susceptibility is found to be valid for all eld directions, as long as spin-orbit coupling is neglected and the spin is isotropic (Heisenberg model). Within the random phase approximation, the generalization of the Stoner criterion for the appearance of an instability of the system at nite temperature reads 1= U 0 (q , ). 2 2 B (7.58)

For the limiting case (q , ) (0, 0) corresponding to a uniform, static external eld, we obtain for the bare susceptibility 0 (q , 0) =
q 0

2 2 B 2 2 B

nk+q nk
k k+q

f ( k ) = 0 (T ), k

(7.59)

145

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 nite Q. Then, another form of magnetic order will occur.

7.2.2

Instability with nite wave vector Q

In order to show that, indeed, the Stoner instability does not always prevail among all possible magnetic instabilities, we rst go through a simple argument based on the local susceptibility. For that, we dene the local magnetic moment along the z -axis, M (r ) = B (r ) (r ) , and consider the nonlocal relation M (r ) = d3 r 0 (r r )Hz (r ), (7.60)

within the linear response approximation. In Fourier space, the same relation reads Mq = 0 (q )Hq , with 0 ( q ) = d3 r 0 (r )eiqr . (7.62) (7.61)

Figure 7.5: R0 , the ratio between the local susceptibility and the static susceptibility, plotted for a box-shaped 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 nite q . Now, compare 0 (q = 0) with 0 (q ) = 0 (r = 0), i.e., the uniform and the local susceptibility at T = 0. The local susceptibility appears to be the average of 0 (q ) over all q , 0 (q ) = 22 B 2 nk + q nk
k ,q k

k +q

2 B 2

d N( )

d N( )

f( ) f( ) ,

(7.63)

and must be compared to 0 (q = 0) = 2 N ( F ) (f ( ) = ( F )). The local susceptibility B 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 1 , D D, D (7.64) N( ) = 0, | | > D, 146

for the density of states. N ( ) forms a band with the shape of a box with band width 2D. With this rough approximation, the integral in equation (7.63) is easily evaluated. The ratio between (q ) and 0 (q = 0) is then found to be R0 = 0 (q ) = ln 0 (q = 0) 4 1 2 + ln 1 1+ , (7.65)

with = F /D. For both small and large band llings ( F close to the band edges), the tendency towards ferromagnetism dominates (cf. Figure 7.5), whereas when F tends towards the middle of the band, the susceptibility 0 (q ) will cease to be maximal at q = 0, and magnetic ordering with a well-dened nite q = Q becomes more probable.

7.2.3

Inuence of the band structure

Whether magnetic order arises at nite 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, i.e. within extended areas in close proximity to the Fermi surface the energy dispersion satises the nesting condition, k+Q = k (7.66)

where k = k F and Q is some xed vector. 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 (7.66) holds for all k. Thus, 0 ( Q ; T ) = 22 B nk + Q nk
k

k k + Q

= 2 B

( ) f ( ) d3 k f k k , (2 )3 k

(7.67)

( ) = f ( + ) = f ( ) and f is the Fermi-Dirac distribution. Under the further where f F assumption that k is weakly angle dependent, we nd 0 (Q; T ) = 2 B 2 d3 k tanh(k /2kB T ) B = (2 )3 k 2 dN ( +
F

tanh(/2kB T ) .

(7.68)

In order to approximate this integral properly, we notice that the integral only converges if a cuto 0 is introduced which we take at half the bandwidth D, i.e. 0 D. Furthermore, the leading contribution comes from the immediate vicinity of the Fermi energy ( 0), so that, within the integration range, a constant density of states N ( F ) can be assumed. Finally, the susceptibility reads
0

0 (Q; T ) B N ( F )

d
0

tanh(/2kB T )
0

(7.69) 2 N ( F ) ln B 1.14 0 2kB T , (7.70)

= 2 N ( F ) ln B

2kB T

+ ln

4e

where we assumed 0 kB T , and where 0.57721 is the Euler constant. The bare susceptibility 0 diverges logarithmically at low temperatures. Inserting the result (7.70) into the generalized Stoner relation (7.58), results in 0=1 UN( F) ln 2 147 1.14 0 2kB Tc , (7.71)

with the critical temperature kB Tc = 1.14 0 e2/U N (

F)

(7.72)

A nite 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 nally 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 ) = z S cos(Q r ), (7.73)

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 ), (7.74)

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).
eindimensional quasieindimensional Chrom

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-dened 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 nite q . 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 fullled, 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 dierent 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 (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 lling. The band structure is given by
k

= 2t[cos(kx a) + cos(ky a) + cos(kz a)],

(7.75)

where a is the lattice constant and t the hopping term. Because of half lling, the chemical potential lies at = 0. Obviously, k+Q = k holds for all k, for the nesting vector 148

Q = (/a)(1, 1, 1). This full nesting trait is a signature of the total particle-hole symmetry. 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 conned 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.

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 |q = fk c k+q ck |g . (7.76)

In this excitation, an electron is extracted from the ground state |g and is replaced by an electron with opposite spin. We consider excitations with a xed 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 Schr odinger equation H|q = (Eg + q )|q , (7.77)

with Eg as the ground state energy (H|g = Eg |g ). Actually it is more convenient to rearrange this equation into the following form eliminating Eg , q |q = with the Hamiltonian H=
k cks cks k,s

fk [H, c k+q ck ]|g

(7.78)

k , k ,q

c k+q ck ck q ck .

(7.79)

Evaluating the commutator in Eq.(7.78) we obtain q |q =(

k+q

k )|q

fk
k k ,q

c c c c c k+q ckq ck q ck |g . k +q k k +q q k

(7.80) The four-c-operator terms are not so easy to handle. Therefore we decouple them through the following approximation,
c c c c c k+q ckq ck q ck k +q k k +q q k

q ,0 (nk nk ) c k+q , ck + (nk+q k ,k+q nk k,k +q ) ckq +q ckq (7.81) 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 (7.78) and multiply from the left with g |ck c to obtain the projected equation, +q k

fk

+q k

U (n n ) = 149

fk (nk+q nk ) = Rq

(7.82)

where we used that the density of electrons of spin s is given by ns = 1 nk s .

k

(7.83)

and we dened Rq . We solve now Eq.(7.82) for the wave function fk and obtain fk = Rq q
k+q

(7.84)
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 1 Rq = U Rq (7.85) q k+q + k
k

where we may divide by Rq = 0. This equation contains a continuum of states corresponding to the electron-hole excitations given by q =
k +q

k +q

+ U (n n ) =

k +q

+.

(7.86)

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.(7.85) that in the q 0 limit we nd 1= 0 U (n n ) U (n n ) 0 = 0 (7.87)

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.(7.85) leading to U q 1 + nk+q nk
k k+q

nk + q nk
k +q

1 2

k )

(7.88)

For small q we expand up order q 2 (actually q -linear contributions cancel), U nk+q nk

k k +q

nk + q nk
k

k +q

2q2

k +q

1 + 1 + and analogously U such that we obtain q

2q2

nk + nk
k 2q2

2m

nk nk (2 2 k q )2 2 4m2

(7.89)

2m2

U n0

3 1

k +q

nk + q nk

2 k + )

(7.90)

2m 3

(U N (

F ) 2)

2q2

2m

U 1 Uc

1/2

.
F)

(7.91) we nd q > 0.

Were we use that (U Uc )1/2 for U Uc . Hence, if U > Uc = 2/N ( 150

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 (7.87) and is predicted by the so-called Goldstone theorem.3 Such an innitesimal rotation is induced by a global spin rotation,
c k ck = Stot

(7.92)

which acting on the ground state |g leads to a degenerate state and corresponds to the state (7.76) 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-dened 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

!"#$%$&

!"!#$%&'()&"!* #&'+',,-*

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.

151

Chapter 8

Magnetism of localized moments

Up to now, we have mostly assumed that the interaction between electrons leads to secondary eects. 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 modications. 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-lled) band. This would actually

152

work well for the H-atoms for suciently 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 suciently 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-lled band must produce a metal, which denitely 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 dicult, 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 H = t (c is cjs + h.c.) + U
i,j ,s i

n i n i ,

(8.1)

where we consider hopping between nearest neighbors only, via the matrix element t. Note, () that cis are real-space eld operators on the lattice (site index i) and nis = c is cis is the density operator. We focus on half lling, 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., |A0 {si } = c i,si |0 , (8.2)

where the spin conguration {si } can be chosen arbitrarily. We will deal with the lifting of this degeneracy later. The rst excited states feature one lattice site without electron
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.
1

153

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 cis = N we can rewrite t where
k

cks eikri ,
k

(8.3)

(c is cjs + h.c.) =
i,j ,s k,s

k cks cks ,

(8.4)

= t

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 |B 0 = Note, that
F

( k )c k ck |0 .

(8.6)

= 0 at half lling, whereas the bandwidth 2D = 12t.

E E

U N(E) N(E)

atomic limit

metallic 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 dened: 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 nite hopping matrix element, the empty (holon) and the doubly occupied (doublon) site become mobile. A 154

fraction of the degeneracy (2N 2 N (N 1)) is herewith lifted and the energy obtains a momentum dependence, Ek,k = U +
k

> U 12t.

(8.7)

Even though ignoring the spin congurations 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 nally 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 eects (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 denitely 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 dened since the ground state is a lled Fermi sea without degeneracy. The treatment of the Coulomb repulsion U turns out to become dicult, once we approach the Mott transition, where the electrons suer 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 xed 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
Note that the motion of an empty site (holon) or doubly occupied site (doublon) is not independent of the spin conguration 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).
2

155

due to the Pauli principle. With this, the importance of the spatial conguration 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 x 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 conguration. We compare the probability to nd 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 eective 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). where we used equation (8.8). In the uncorrelated case (where d is not xed), we have P0 ( 0) = ni (1 ni )(1 nj )(1 nj ) = 1 . 16 (8.11) (8.10)

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 xed 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 + U d = 8d(1 2d) + U d, 1 = N
0

kin

kin

kin

d N( ) .
D

(8.13)

For xed 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 nd d= with the critical value Uc = 8|
kin |

1 4

U Uc

and

gt = 1

U Uc

(8.14)

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
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).
3

156

takes into account only local correlations, while correlations between dierent lattice sites are not considered. Moreover, correlations between the spin degrees of freedom are entirely neglected. The charge excitations contain contributions between dierent energy scales: (1) a metallic part, described via the renormalized eective Hamiltonian Hren = gt k c ks cks + U d,
k,s

(8.16)

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 ( ) =
2 p

high ( ) + 1

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 2 = g 2 , such that the f -sum rule in equation (6.13) yields renormalized, p t p

I=
0

d1 ( ) =

2 p

gt + Ihigh

energy

2 p

(8.18)

For U Uc , the coherent metallic part becomes weaker and weaker,

2 p

gt =

U Uc

2 p

(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 simplied way. Let us investigate the momentum distribution. According to the above denition,
kin

1 N

k, kFS

(8.20)

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 nite U , such that, for k in the rst Brillouin zone, 1 1 = 2 N and gt
kin

nin +
kFS

1 N

k / FS

1 nout = (nin + nout ) 2

(8.21)

1 N

nin
kFS

1 N

nout k .
k / FS

(8.22)

Taking into account particle-hole symmetry, i.e.,

k k

=
kFS

+
k / FS

= 0,

(8.23)

157

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 k

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 eective mass m = gt , (8.25) m and thus 3U 2 1 s 1 = 2 = 3 gt F1 , (8.26) Uc U 2
a = F0

1 where t = 1/2m and the density of states N ( F ) = N ( F )gt . Furthermore,

It follows, that the compressibility vanishes for U Uc as expected, since it becomes more and more dicult 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 nite. This correlation is, as mentioned above, neglected in the Gutzwiller approximation. The eective 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 xed 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 uctuations of the particle number. For this, we introduce the density matrix of the electron states on an arbitrary lattice site, s = h|0 0| + d| | + (| | + | |) , (8.29) 2

U N ( F ) 2Uc + U U, 4 (U + Uc )2 c U N ( F ) 2UC U s F0 = U, 4 (U Uc )2 c

2 N ( F ) B a , 1 + F0 N ( F ) = 2 s) . n (1 + F0 =

(8.27) (8.28)

158

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 eective 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 eective Hamiltonian

In order to employ our perturbative considerations, it is sucient 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 congurations, {| , , | , , | , , | , }. The application of Hkin yields Hkin | , = Hkin | , = Hkin | , = t| , 0 t|0, , Hkin | , = 0, (8.31) (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).

t
initial spin configuration

E=U

t
or
final spin configurations

virtual states

Figure 8.5: Illustration of the origin of the superexchange. We obtain the matrix elements Ms1 ,s2 ;s
1 ,s2

s1 , s2 |Hkin |n

1 n|Hkin |s1 , s2 , n|HCoul |n

(8.33)

where |n = | , 0 or |0, , such that the denominator is always U . We end up with M; = M; = M; = M; =

from which we deduce the variance of the occupation number, n2 n
2

2t2 . U

(8.34)

= n2 1 = tr( n2 ) 1 = 4d + s 1 = 2d.

(8.30)

The deviation from single occupation vanishes with d, i.e., with the approach of the metal-insulator transition. Note that the dissipation-uctuation theorem connects n2 n 2 to the compressibility.

159

Note that the signs originates from the anti-commutation properties of the Fermion operators. In the subspace {| , , | , } we nd the eigenstates of the respective secular equations, 1 (| , + | , ) , 2 1 (| , | , ) , 2 E = 0, E= 4t2 . U (8.35) (8.36)

Since the states | , and | , 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 eective Hamiltonian with the same energy spectrum for the spin congurations can be written with the help of the spin operators S 1 and S 2 on the two lattice sites
2

He = J S 1 S 2

J=

4t2 > 0. U 2

(8.37)

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 HH = J S i S j + const. (8.38)

i,j

This model, reduced to spins only, is called Heisenberg model. The Hamiltonian is invariant under a global SU (2) spin rotation, Us ( ) = eiS , S=
j

Sj .

(8.39)

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 eld 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 diculty lies predominantly in the treatment of quantum uctuations, 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 uctuations. In order to describe the antiferromagnetic state anyway, we apply the mean eld 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.

160

conguration doubles the unit cell. We introduce the respective mean eld, z m) iA m + ( Si z Si = . z m + (Si + m) i B This leads to the mean eld Hamiltonian Hmf = HA + HB = Jzm
z Si + Jzm iA iB z Si + Jz

(8.40)

m2 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, Z = tr e Hmf = eJmz
/2

+ eJmz

/2

eJzm

2 /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 xed temperature, we minimize the free energy with respect to m to determine the thermal equilibrium state,6 i.e., set F/m = 0 and nd m= 2 tanh Jzm 2kB T . (8.44)

This is the self-consistency equation of the mean eld theory. It provides a critical temperature TN (N eel temperature), below which the mean moment m is nite. For T TN , m approaches 0 continuously. Thus, TN can be found from a linearized self-consistency equation, m= and thus TN = Jz 2 . 4kB (8.46) Jzm 2 4kB T ,
T =TN

(8.45)

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, F (m, T ) = F0 + Jz 2 1 TN T m2 + 2 3 2 TN T
3

m4 .

(8.47)

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) 3(TN /T 1), T TN . 2
6 Actually, a magnetic eld pointing into the opposite direction on each site would be another equilibrium variable (next to the temperature). We set it to zero. 7 At innite z , the mean eld approximation becomes exact.

161

F T>T
N

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 eld approximation also has a number of insuciencies. Quantum and some part of thermal uctuations 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., HH = J
z z Sj + Si i,j

1 + S + S + Si Sj 2 i j

(8.49)

In the ordered state, the moments shall be aligned along the z -axis. To observe the dynamics of a ipped spin, we apply the operator W = N l=0 fl Sl on the ground state |0 , and determine the spectrum, by solving the resulting eigenvalue equation (HH E0 )W |0 = [HH , W ]|0 = W |0 , with the ground state energy E0 . Using the spin-commutation relations
+ z ij , Sj , Sj = 2Sj z , Sj = Sj ij , Sj

(8.50)

(8.51) (8.52)

then yields the equation

l

z Sj Sl + J j j z Sl Sl |0 = 0, Sj

(8.53)

where

runs over all neighbors of l. We decouple this complicated problem by replacing

the operators S z by their mean elds. Therefore, we have to distinguish between A and B sublattices, such that we end up with two equations, flA
lA

Jmz Sl + Jm
a

Sl +a Sl

|0 = 0, |0 = 0,

l A, l B.

(8.54) (8.55)

flB
l B

Jmz Sl Jm

Sl+a Sl 2 i q r l e N 2 iqrl e N 162

Choosing flA = flB = lA, l B,

(8.56)

we introduce the operators Sl = Sl = with l A and l B , and, vice versa, a q = b q = 2 N 2 N Sl eiqrl ,

lA

2 N 2 N

i q r l a , qe q i q r l b , qe q

(8.57) (8.58)

(8.59) (8.60)

Sl eiqrl ,
l B

and insert them into the equation and obtain, (Jmz ) (Jmz ) From this follows that
(Jmz )a q + Jmq bq |0 = 0, (Jmz )b q Jmq aq |0 = 0,

Sl eiqrl + Jm
l A a

e i q a
l B

Sl eiqrl Sl eiqrl
l A

|0 = 0, |0 = 0.

(8.61) (8.62)

l B

Sl eiqrl Jm

eiqa
a

(8.63) (8.64)

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 = Jm
2. z 2 q

(8.65)

Note, that only the positive energies make sense. It is interesting to investigate the limit of small q ,
2 z 2 q z 2 q 2 + O (q 4 ),

(8.66)

where q = Jmz |q | + . (8.67)

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 aq and bq are of bosonic nature; this comes about since, due to the mean eld approximation, the Sl are bosonic as well,
[Sl+ , Sj ] = 2Slz lj 2mlj ,

(8.68)

where the sign depends on the sublattice. The zero-point uctuations of these bosons yield quantum uctuations, which reduce the moment m from its mean eld value. In a one-dimensional 163

boundary of reduced Brillouin zone

h q

q
2a

Figure 8.7: Spectrum of the spin waves in the antiferromagnet.

spin chain these uctuations 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 innite 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 dierence 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.

164