Sunteți pe pagina 1din 320

Advanced Quantum Mechanics

Geert Brocks
Faculty of Applied Physics, University of Twente

August 2002
ii
Contents

Preface xiii

I Single Particles 1

1 Quantum Mechanics 3
1.1 Wave Mechanics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2 Quantum Mechanics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.3 Representations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.3.1 General Formalism . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.3.2 The Position Representation; Wave Mechanics Revisited . . . . . . . 9
1.4 Many Particles and Product States . . . . . . . . . . . . . . . . . . . . . . . 13

2 Time Dependent Perturbation Theory 17


2.1 Time Evolution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.1.1 The Huygens Principle . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.2 Time Dependent Perturbations . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.3 Fermi’s Golden Rule . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.4 Radiative Transitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.4.1 Atom in a Radiation Field . . . . . . . . . . . . . . . . . . . . . . . . 30
2.4.2 Einstein Coefficients and Rate Equations . . . . . . . . . . . . . . . 32
2.4.3 Population and Lifetime . . . . . . . . . . . . . . . . . . . . . . . . . 35
2.5 Epilogue . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
2.6 Appendix I. The Heisenberg Picture . . . . . . . . . . . . . . . . . . . . . . 37
2.7 Appendix II. Some Integral Tricks . . . . . . . . . . . . . . . . . . . . . . . 40

3 The Quantum Pinball Game 43


3.1 A Typical Experiment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
3.2 Time Evolution; Summing the Perturbation Series . . . . . . . . . . . . . . 45
3.2.1 Adapt Integration Bounds; Green Functions . . . . . . . . . . . . . . 47
3.2.2 Fourier Transform to the Frequency Domain . . . . . . . . . . . . . 49
3.2.3 Sum the Perturbation Series; Dyson Equation . . . . . . . . . . . . . 50
3.2.4 Green Functions; Closed Expressions . . . . . . . . . . . . . . . . . . 51
3.2.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
3.3 Connection to Mattuck’s Ch. 3 . . . . . . . . . . . . . . . . . . . . . . . . . 54
3.4 Appendix. Green Functions; the Lippmann-Schwinger Equation . . . . . . . 55
3.4.1 The Huygens Principle Revisited . . . . . . . . . . . . . . . . . . . . 59

iii
iv CONTENTS

4 Scattering 61
4.1 Scattering by a Dilute Concentration of Centers . . . . . . . . . . . . . . . . 62
4.1.1 The Scattering Cross Section . . . . . . . . . . . . . . . . . . . . . . 65
4.1.2 Forward Scattering; the Optical Theorem . . . . . . . . . . . . . . . 67
4.2 Scattering by a Single Center . . . . . . . . . . . . . . . . . . . . . . . . . . 69
4.3 Re-summation of the Series; the Self-Energy . . . . . . . . . . . . . . . . . 72
4.4 The Physical Meaning of Self-Energy . . . . . . . . . . . . . . . . . . . . . . 74
4.5 The Scattering Cross Section . . . . . . . . . . . . . . . . . . . . . . . . . . 77
4.5.1 The Lippmann-Schwinger Equation . . . . . . . . . . . . . . . . . . 77
4.5.2 The Scattering Amplitudes and the Differential Cross Section . . . . 82
4.5.3 The Born Series and the Born approximation . . . . . . . . . . 83
4.6 Epilogue . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
4.7 Appendix I. The Refractive Index . . . . . . . . . . . . . . . . . . . . . . . . 85
4.8 Appendix II. Applied Complex Function Theory . . . . . . . . . . . . . . . 87
4.8.1 Complex Integrals; the Residue Theorem . . . . . . . . . . . . . . . 87
4.8.2 Contour Integration . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
4.8.3 The Principal Value . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
4.8.4 The Self-Energy Integral . . . . . . . . . . . . . . . . . . . . . . . . . 94

II Many Particles 97

5 Quantum Field Oscillators 99


5.1 The Quantum Oscillator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
5.1.1 Summary Harmonic Oscillator . . . . . . . . . . . . . . . . . . . . . 102
5.1.2 Second Quantization . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
5.2 The One-dimensional Quantum Chain; Phonons . . . . . . . . . . . . . . . 104
5.3 My First Quantum Field . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
5.3.1 Classical Chain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
5.3.2 Continuum Limit: an Elastic Medium . . . . . . . . . . . . . . . . . 110
5.3.3 Quantizing the Elastic Medium; Phonons . . . . . . . . . . . . . . . 113
5.4 The Three-dimensional Quantum Chain . . . . . . . . . . . . . . . . . . . . 115
5.4.1 Discrete Lattice . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
5.4.2 Elastic Medium . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
5.4.3 Are Phonons Real Particles ? . . . . . . . . . . . . . . . . . . . . . 117
5.5 The Electro-Magnetic Field in Vacuum . . . . . . . . . . . . . . . . . . . . . 118
5.5.1 Classical Electro-Dynamics . . . . . . . . . . . . . . . . . . . . . . . 119
5.5.2 Quantum Electro-Dynamics (QED) . . . . . . . . . . . . . . . . . . 121
5.5.3 Are Photons Real Particles ? . . . . . . . . . . . . . . . . . . . . . . 124

6 Bosons and Fermions 127


6.1 N particles; the Stone Age . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
6.1.1 The Slater Determinant . . . . . . . . . . . . . . . . . . . . . . . . . 131
6.1.2 Three Particle Example Work-out . . . . . . . . . . . . . . . . . . . 132
6.1.3 One- and Two-particle Operators . . . . . . . . . . . . . . . . . . . . 133
6.2 N particles; the Modern Era . . . . . . . . . . . . . . . . . . . . . . . . . . 134
6.2.1 Second Quantization for Bosons . . . . . . . . . . . . . . . . . . . . 134
6.2.2 Second Quantization for Fermions . . . . . . . . . . . . . . . . . . . 137
CONTENTS v

6.2.3 The Road Travelled . . . . . . . . . . . . . . . . . . . . . . . . . . . 141


6.3 The Particle-Hole Formalism . . . . . . . . . . . . . . . . . . . . . . . . . . 143
6.3.1 The Homogeneous Electron Gas . . . . . . . . . . . . . . . . . . . . 143
6.3.2 Particles and Holes . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
6.3.3 The Quantum Field Theory Connection . . . . . . . . . . . . . . . . 149
6.4 Second Quantization and the Electron Field . . . . . . . . . . . . . . . . . . 150
6.5 Appendix I. Identical Particle Algebra . . . . . . . . . . . . . . . . . . . . . 154
6.5.1 Normalization Factors and Orthogonality . . . . . . . . . . . . . . . 154
6.5.2 Second Quantization for Operators . . . . . . . . . . . . . . . . . . . 156
6.6 Appendix II. Identical Particles . . . . . . . . . . . . . . . . . . . . . . . . 162
6.6.1 Indistinguishable Particles . . . . . . . . . . . . . . . . . . . . . . . . 162
6.6.2 Why Symmetrize ? . . . . . . . . . . . . . . . . . . . . . . . . 163
6.6.3 Symmetrize The Universe ? . . . . . . . . . . . . . . . . . . . . . . . 165

7 Optics 169
7.1 Atoms and Radiation; the Full Monty . . . . . . . . . . . . . . . . . . . . . 169
7.1.1 Absorption; Fermi’s Golden Rule . . . . . . . . . . . . . . . . . . . . 173
7.1.2 Spontaneous Emission . . . . . . . . . . . . . . . . . . . . . . . . . . 175
7.2 Electrons, Holes and Photons . . . . . . . . . . . . . . . . . . . . . . . . . . 176
7.2.1 Electrons and Radiation . . . . . . . . . . . . . . . . . . . . . . . . . 177
7.2.2 Free Electrons and Holes . . . . . . . . . . . . . . . . . . . . . . . . 179
7.2.3 Light Absorption by Electrons and Holes . . . . . . . . . . . . . . . 182
7.2.4 Light Scattering by Free Electrons . . . . . . . . . . . . . . . . . . . 187
7.3 Higher Order Processes; the Quantum Pinball Game . . . . . . . . . . 188
7.4 Appendix I. Interaction of an Electron with an EM field . . . . . . . . . . . 190
7.4.1 Dipole Approximation . . . . . . . . . . . . . . . . . . . . . . . . . 192
7.5 Appendix II. Relativistic Electrons and Holes . . . . . . . . . . . . . . . . . 193

III Interacting Particles 199

8 Propagators and Diagrams 201


8.1 The Single Particle Propagator . . . . . . . . . . . . . . . . . . . . . . . . . 202
8.1.1 A Gedanken Experiment . . . . . . . . . . . . . . . . . . . . . . . . . 203
8.1.2 Particle and Hole Propagators . . . . . . . . . . . . . . . . . . . . . 204
8.2 A Single Particle or Hole . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205
8.2.1 Particle Scattering . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205
8.2.2 The Second Quantization Connection . . . . . . . . . . . . . . . . . 207
8.2.3 Hole Scattering . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209
8.3 Many Particles and Holes . . . . . . . . . . . . . . . . . . . . . . . . . . . . 211
8.3.1 Atom Embedded in an Electron Gas . . . . . . . . . . . . . . . . . . 214
8.3.2 Goldstone Diagrams; Exchange . . . . . . . . . . . . . . . . . . . . . 215
8.3.3 Diagram Expansion . . . . . . . . . . . . . . . . . . . . . . . . . . . 218
8.3.4 Diagram Summation . . . . . . . . . . . . . . . . . . . . . . . . . . . 219
8.3.5 Exponential Decay . . . . . . . . . . . . . . . . . . . . . . . . . . . 222
8.4 Interacting Particles and Holes . . . . . . . . . . . . . . . . . . . . . . . . . 223
8.4.1 Two-Particle Diagrams . . . . . . . . . . . . . . . . . . . . . . . . . 225
8.4.2 The Homogeneous Electron Gas Revisited . . . . . . . . . . . . . . . 227
vi CONTENTS

8.4.3 The Full Diagram Dictionary . . . . . . . . . . . . . . . . . . . . 228


8.4.4 Radiation Diagrams . . . . . . . . . . . . . . . . . . . . . . . . . . . 229
8.5 The Spectral Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 230
8.5.1 Physical Content . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 232
8.6 (Inverse) Photoemission and Quasi-particles . . . . . . . . . . . . . . . . . 233
8.6.1 Photoemission . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 233
8.6.2 Inverse Photoemission . . . . . . . . . . . . . . . . . . . . . . . . . . 235
8.7 Appendix I. The Adiabatic Connection . . . . . . . . . . . . . . . . . . . . . 237
8.7.1 The Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 237
8.7.2 The Solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 238
8.8 Appendix II. The Linked Cluster Expansion . . . . . . . . . . . . . . . . . . 241
8.8.1 Denominator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 242
8.8.2 Numerator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 245
8.8.3 Linked Cluster Theorem . . . . . . . . . . . . . . . . . . . . . . . . . 250

9 The electron-electron interaction 253


9.1 Many interacting electrons . . . . . . . . . . . . . . . . . . . . . . . . . . . . 253
9.2 The Hartree approximation . . . . . . . . . . . . . . . . . . . . . . . . . . . 255
9.2.1 The Hartree (Coulomb) interaction . . . . . . . . . . . . . . . . . . . 256
9.2.2 The Hartree Self-Consistent Field equations . . . . . . . . . . . . . . 260
9.2.3 Pro’s and con’s of the Hartree approximation . . . . . . . . . . . . . 265
9.3 The Hartree-Fock approximation . . . . . . . . . . . . . . . . . . . . . . . . 266
9.3.1 The exchange interaction . . . . . . . . . . . . . . . . . . . . . . . . 267
9.3.2 The Hartree-Fock Self-Consistent Field equations . . . . . . . . . . . 271
9.3.3 The homogeneous electron gas revisited . . . . . . . . . . . . . . . . 274
9.3.4 Pro’s and con’s of the Hartree-Fock approximation . . . . . . . . . . 280
9.3.5 Screening . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 282
9.4 The Random Phase Approximation (RPA) . . . . . . . . . . . . . . . . . . 283
9.4.1 The RPA diagram . . . . . . . . . . . . . . . . . . . . . . . . . . . 284
9.4.2 The RPA screened interaction . . . . . . . . . . . . . . . . . . . . . . 287
9.4.3 The GW approximation . . . . . . . . . . . . . . . . . . . . . . . . . 290
9.4.4 The homogeneous electron gas re-revisited . . . . . . . . . . . . . . . 293
List of Figures

2.1 Progagation of a wave using the Huygens principle. . . . . . . . . . . . . . . 20


2.2 Feynman diagram. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.3 Dictionary of Feynman diagrams. . . . . . . . . . . . . . . . . . . . . . . . . 23
2.4 The Born approximation in Feynman diagrams. . . . . . . . . . . . . . . . . 24
2.5 Typical line-shape function F (ω). . . . . . . . . . . . . . . . . . . . . . . . 29
2.6 Optical processes involving the levels i and f ; Einstein A, B coefficients. . . 32
2.7 Radiative processes. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
2.8 The function 4T (ω) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41

3.1 Typical quantum experiment. . . . . . . . . . . . . . . . . . . . . . . . . . . 44


3.2 Feynman diagram of the absorption of a photon. . . . . . . . . . . . . . . . 45
3.3 A ‘physical’ theta function. . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
3.4 Propagation of a wave using the Huygens principle. . . . . . . . . . . . . . . 59

4.1 Scattering of a single particle by a fixed target. . . . . . . . . . . . . . . . . 62


4.2 Incoherent scattering of wave packets in a dilute sample. . . . . . . . . . . . 64
4.3 Scattering geometry. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
4.4 Perturbation series for single particle scattering. . . . . . . . . . . . . . . . 71
4.5 Perturbation series rewritten in terms of self energy. . . . . . . . . . . . . . 72
4.6 the self energy for single particle scattering . . . . . . . . . . . . . . . . . . 72
4.7 Angle resolved scattering detection. . . . . . . . . . . . . . . . . . . . . . . . 78
4.8 Adding layers to calculate the index of refraction. . . . . . . . . . . . . . . . 86
4.9 Contour integration = integration along a path in the complex plane. . . . 87
4.10 A closed contour C in the complex plane. . . . . . . . . . . . . . . . . . . . 88
4.11 Cauchy’s integral formula. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
4.12 Closing the contour in the lower half plane. . . . . . . . . . . . . . . . . . . 90
4.13 Closing the contour in the upper half plane. . . . . . . . . . . . . . . . . . . 91
4.14 Lorenzian line shape function. . . . . . . . . . . . . . . . . . . . . . . . . . . 93

5.1 A linear chain of masses and springs. . . . .p. . . . . . . . . . . . . . . . . 104


κ
5.2 Dispersion relation of a linear chain ω (k) = 2 m | sin 12 ka|. . . . . . . . . . 107
5.3 An elastic medium with displacements u at points xm (top figure). Artist’s
impression of continuum (bottom figure). . . . . . . . . . . . . . . . . . . . 109
5.4 Scattering of a neutron and emission (left) or absorption (right) of a phonon.118
5.5 Compton scattering of X-rays. . . . . . . . . . . . . . . . . . . . . . . . . . . 125

6.1 An N -boson state. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134


6.2 An N -fermion state. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138

vii
viii LIST OF FIGURES

6.3 Absorption of a photon. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142


6.4 Shells of constant ²k . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
6.5 Particle and hole energies relative to the Fermi level. . . . . . . . . . . . . . 149
6.6 Two particles in two separated wave packets. . . . . . . . . . . . . . . . . . 163
6.7 Two particles in overlapping wave packets. . . . . . . . . . . . . . . . . . . . 165

7.1 Absorption of a photon by an electron. . . . . . . . . . . . . . . . . . . . . . 179


7.2 Absorption of a photon by an electron. . . . . . . . . . . . . . . . . . . . . . 180
7.3 Absorption of a photon, creating an electron-hole pair. . . . . . . . . . . . . 182
7.4 Creation of a particle-hole pair by a photon. . . . . . . . . . . . . . . . . . . 184
7.5 Creation of a particle-hole pair by a photon. . . . . . . . . . . . . . . . . . . 185
7.6 Absorption of a photon by a bound hole. . . . . . . . . . . . . . . . . . . . . 186
7.7 Annihilation of a particle-hole pair creates a photon. . . . . . . . . . . . . . 186
7.8 A contribution to electron-photon (or Compton) scattering. . . . . . . . . . 187
7.9 An 8th order electron-electron scattering diagram. . . . . . . . . . . . . . . 189
7.10 Creation of a particle-hole pair; (a) non-relativistic, (b) relativistic. . . . . . 194
7.11 Relativistic particle-hole spectrum. . . . . . . . . . . . . . . . . . . . . . . . 195
7.12 Energy-momentum dispersion relations of an electron (top thick curve), a
hole (bottom thick curve), and a photon (thin straight line). . . . . . . . . . 196
7.13 Cascade induced by a high energy electron. . . . . . . . . . . . . . . . . . . 197

8.1 A visual interpretation of the propagator i~G+ (l, k,t1 − t0 ). . . . . . . . . . 204


8.2 Second order potential scattering of a particle. . . . . . . . . . . . . . . . . 208
8.3 Second order potential scattering of a hole. . . . . . . . . . . . . . . . . . . 210
8.4 Mattuck’s table 4.2, p.75 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 211
8.5 Mattuck’s table 4.3, p.86. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 229
8.6 Photoemission: incoming photon of energy ~ω and outgoing electron of
energy ²q . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 233
8.7 Inverse photoemission: incoming electron of energy ² and outgoing photon
of energy ~ωq . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 236

9.1 The Hartree diagram . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 257


9.2 Summing over Hartree diagrams . . . . . . . . . . . . . . . . . . . . . . . . 258
9.3 The self-consistent Hartree approximation . . . . . . . . . . . . . . . . . . . 264
9.4 The dressed (self-consistent) Hartree potential . . . . . . . . . . . . . . . . . 265
9.5 The exchange potential . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 268
9.6 The exchange potential . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 269
9.7 The exchange diagram . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 269
9.8 A fourth order Hartree-Fock diagram . . . . . . . . . . . . . . . . . . . . . . 270
9.9 Fourth order Hartree-Fock diagram . . . . . . . . . . . . . . . . . . . . . . . 270
9.10 The Hartee-Fock approximation . . . . . . . . . . . . . . . . . . . . . . . . . 273
9.11 The self-consistent (dressed) Hartree-Fock potential. . . . . . . . . . . . . . 274
9.12 Exchange energy as function of kkF . . . . . . . . . . . . . . . . . . . . . . . 277
9.13 Hartree-Fock energies ²k and kinetic energies ²0,k for kF = 1a−10 . . . . . . . 278
1 j1 (r)
9.14 ρX (r) = 2π2 r as function of r . . . . . . . . . . . . . . . . . . . . . . . . 279

9.15 ²0 (k) = dk in the Hartree-Fock approximation. . . . . . . . . . . . . . . . . . 282
9.16 An eighth order RPA + Hartree-Fock diagram. . . . . . . . . . . . . . . . . 285
LIST OF FIGURES ix

9.17 The RPA approximation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 286


9.18 The RPA or GW approximation . . . . . . . . . . . . . . . . . . . . . . . . 290
9.19 The self-energy according to Thomas-Fermi screening . . . . . . . . . . . . 300
9.20 The RPA quasi-particle energies using Thomas-Fermi screening . . . . . . . 301
9.21 k-derivative of the HF (upper) and RPA/Thomas-Fermi (lower) particle
energies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 302
9.22 Quasi-particle energies as calculated within RPA. . . . . . . . . . . . . . . . 304
9.23 The factor |Im ΣX,RP A (k, k,ω k )| which determines the inverse quasi-particle
lifetime, in units of ²F , in the RPA approximation (lower curve). . . . . . . 305
9.24 The weight factor zk at the Fermi level, |k| = kF as a function of rs . . . . 306
x LIST OF FIGURES
List of Tables

1.1 discrete basis representation . . . . . . . . . . . . . . . . . . . . . . . . . . . 9


1.2 continuous basis representation . . . . . . . . . . . . . . . . . . . . . . . . . 9

5.1 Comparison between a discrete classical chain and an elastic medium . . . . 113

6.1 Equation roadmap from 1st to 2nd quantization . . . . . . . . . . . . . . . . 142


6.2 Particles and holes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148

xi
xii LIST OF TABLES
Preface

“The labour we delight in physics pain”, Shakespeare, Macbeth.1

This manuscript contains the lecture notes of the course “voortgezette quantum me-
chanica” (advanced quantum mechanics). The course is intended for physics students earn-
ing their master degree (4th/5th year) who are interested in modern theoretical physics.
It is meant to form a bridge between elementary courses in quantum mechanics and the
more advanced topical quantum mechanics books. Rather than reviewing all the topics of
an introductory course once again on a deeper level, I have chosen to make the quantum
mechanics of many-particle systems one of the main themes in this course. The emphasis is
on general methods and interpretations—often borrowed from quantum field theory—, such
as second quantization and quasi-particles. Our main tool is time-dependent perturba-
tion theory, which can be represented graphically by Feynman65 diagrams in a physically
intuitive way.2 Model systems such as the electron gas and the elastic medium are used
to introduce the general structure of the physics, examples of relevant applications are
mainly taken from condensed matter physics and optics. The organization of these notes
is as follows. Part I introduces some of the basic tools of quantum mechanics, in par-
ticular time-dependent perturbation theory. Part II describes some of the basic concepts
of many-fermion and -boson physics, (anti)particles, quasi-particles, second quantization
and quantum fields. Part III contains an introduction to more advanced topics, such as
propagators and diagrammatic expansions.

I have tried to make these notes self-contained with as few phrases such as “A little
algebra yields ...” or “It can be shown that ... ” as possible. Such sentences always
annoyed me when I was a student; either show how it works, or leave the subject alone.
Preparing these notes as a lecturer however I discovered that there is probably a good rea-
son for such phrases. Without them, the text easily assumes the shape of an “omgevallen
boekenkast”. I have tried to bring some order in this pile by presenting the more basic
material in the first sections of each chapter, collecting the more special (but often very
interesting) topics in the final sections, and shifting some detailed algebra and background
material to appendices. Topics such as scattering theory or quantum electro-dynamics are
only discussed superficially, since a thorough discussion would require too much space.
Other topics, such as relativistic quantum mechanics or symmetry (group theory), are
only touched upon. They are more suitable for a specialized course. I stayed away from
1
“It is a good thing for an uneducated man to read books of quotations”, Winston Churchill, My Early
Life.
2
The contributors to quantum theory and many-particle physics (or quantum field theory) make up a
list of “who-is-who” in physics. I will mark Nobel prize winners by the year in which they recieved their
award.

xiii
xiv PREFACE

the finite temperature formalism, since that is too difficult for me; in all the material
presented here T = 0 is assumed.
I would like to thank Els Braker-Peerik for word-processing the first version of these
notes from my hand-written papers. Also I wish to acknowledge Jeroen Hegeman for point-
ing out an error in the original notes, which resulted from my urge to take a shortcut. More-
over, I did my best to remove inconsistencies in notation, units and phase/normalization
factors which plagued the original notes. At least the errors are more consistent now; if
you spot any, please let me know.

A good working knowledge of elementary quantum mechanics is assumed, as well


as a knowledge of the basic mathematical tools of analysis, linear algebra and Fourier
transforms. Knowledge of complex function theory is useful, but not absolutely essential.
The book you used in your introductory course is still helpful: D. J. Griffiths, Introduction
to Quantum Mechanics, (Prentice Hall, New Jersey, 1995) . Although the present notes
are reasonably self-contained, it is useful to have the following book; R. Mattuck, An
Introduction to Feynman Diagrams in Many Particle Physics, (Dover, New York, 1992).
I will make comparisons to the relevant passages of this book in these lecture notes. In
(numerous) footnotes references are given to additional literature, if you are interested.
In the lecture notes I mainly discuss the general structure of the theory. Lecture notes
are designed to make the lecturer superfluous. This is not so for the assignments which
are handed out each week. These are an integral part of this course. These exercises are
vital in order to get familiarized with the theory. Moreover, they contain applications to
physical problems.
Part I

Single Particles

1
Chapter 1

Quantum Mechanics

“Though this be madness, yet there is method in it”, Shakespeare, Hamlet.

This chapter starts by summarizing the basic postulates of (Schrödinger33 ’s) wave me-
chanics, which is the usual subject of introductory courses on quantum mechanics. These
postulates are actually applicable in a wider sense, which has lead Dirac33 to formulate
the structure of quantum mechanics in a more general way. Wave mechanics is only a spe-
cific representation of quantum mechanics. The term “representation” has a well-defined
mathematical meaning here, which is explained next in this chapter. Finally, it is shown
how to systematically construct quantum states for many particles and many degrees of
freedom. This last section might only be glanced through at first reading; we will come
back to it at a later stage.

1.1 Wave Mechanics


Experiments and theoretical analysis have resulted in a number of postulates (or laws)
upon which quantum mechanics is founded. They describe (1) how microscopic particles
must be represented, (2) how to obtain quantities that can be observed, (3) how time
evolution must be described and (4) what the logical structure of (a series of) measure-
ments is. The form of quantum mechanics you are probably most familiar with, is wave
mechanics (or Schrödinger’s quantum mechanics as it is called in Appendix A of Mattuck).
For a single particle the postulates of wave mechanics are summarized below. Mind you,
the summary is very compact, and you should consult your introductory books, such as
Griffiths, for more detail. If you find quantum mechanics strange, let me quote Richard
Feynman65 : “...the way we have to describe Nature is generally incomprehensible to us.”.
In other words we don’t know why the physical postulates or laws are as they are, but
they are highly successful in explaining the phenomena, which is ultimately what counts
in physics. The same probably holds for classical mechanics or electrodynamics as well;
we just seem to be more familiar with Newton’s or Maxwell’s laws.

POSTULATES (single particle):

1. A particle is represented by a complex wave function ψ(r, t) where r = x, y, z is a


point in space and t is the time. Noteworthy properties are:

3
4 CHAPTER 1. QUANTUM MECHANICS

(a) If ψ(r, t) is an acceptable wave function and φ(r, t) is an acceptable wave func-
tion, then aψ(r, t) + bφ(r, t) must also be an acceptable wave function, where
a, b are complex numbers. This is the superposition principle. Wave functions
thus form a linear vector space. In physical terms, the superposition principle
describes the phenomenon of interference of waves.
(b) The wave function itself is not directly observable; however the probability P (r)
of finding the particle inside a small volume dV around the point r is given by
the intensity of the wave
1
P (r) = |ψ(r, t)|2 dV (1.1)

R R
where the constant Nψ = |ψ(r, t)|2 dV is chosen such that P (r)dV = 1;
(i.e., the particle has to be somewhere).
Only wave functions for which Nψ can be calculated p as a finite number are
acceptable. Defining the norm of a wave function as Nψ , acceptable wave
functions thus have a finite norm.
(c) If ψ(r, t) is an R acceptable wave function and φ(r, t) is an acceptable wave

function, then ψ (r, t)φ(r, t)dV = c, where c is a complex number, defines the
inner product of the two wave functions. Wave functions thus form an inner
product space.1
|c|2
If the particle starts in ψ at t = 0, then Nφ Nψ is the probability of finding the
particle in φ at t.

2. Every observable property A of the particle corresponds to a mathematical object


b , which operates in the wave function space defined
called operator , notation A
above. Noteworthy is:

(a) The average of property A at t over a series of measurements in which the


particle is represented by the wave function ψ(r, t), is given by
R ∗
b
ψ (r, t)Aψ(r, t)dV
hA(t)i = R ∗ (1.2)
ψ (r, t)ψ(r, t)dV
It is called the expectation value of A.
Familiar examples of observables in wave mechanics are:
b is a simple
• The position, e.g. along the x axis. The position operator x
multiplication with the number x. The quantity hx(t)i gives the average
position along the x axis as a function of t.
• The momentum px , again e.g. in the x direction. The momentum operator
pbx is a differential operator
~ ∂
pbx = (1.3)
i ∂x
the average momentum in the x direction is given by hpx (t)i.
1
Norms and inner products are strongly related; you can’t have one without the other. The properties
of inner product (vector) spaces should be familiar to you from your linear analysis courses, consult
your dictaten. This particular one, the wave function space, is called L2 . It has an inifinite number of
dimensions; mathematically, it is an example of a so-called Hilbert space.
1.1. WAVE MECHANICS 5

(b) All operators are linear operators; i.e. A b {aψ(r, t) + bφ(r, t)} = aAψ(r,
b t) +
b
bAφ(r, t).
(c) One can define powers of operators A bA,
b2 = A b multiply different operators A bB,
b
sum operators A b + B,
b and even define functions of operators f (A).
b The latter
are defined by their power series, e.g.
b2 b3
b + A + A + ...
b ≡I+A
exp(A)
2! 3!
An important operator is the Hamiltonian given by

pb2x + pb2y + pb2z


b=
H x, yb, zb)
+ V (b (1.4)
2m
Bear in mind that in multiplying different operators their order is important.
This is usually emphasized by defining a quantity called the commutator of Ab
b
and B,
h i
bB
A, b =A bBb−B bAb (1.5)

which need not be zero, for instance [b x, pbx ] = i~.


Operator algebra (additions, multiplications, power series, commutators) is of
great practical use. One uses it, for instance, to define basic statistical quantities
such as the statistical spread ∆A
­ ®
(∆A)2 = A2 − hAi2 (1.6)

3. The evolution of a wave function ψ(r, t) in time is given by the Schrödinger equation

∂ b
i~ ψ(r, t) = Hψ(r, t) (1.7)
∂t

where the Hamiltonian (or Hamilton operator) H b is given by eq. 1.4. Since the
Schrödinger equation is a linear differential equation, which is first order in time,
we only need to specify the initial condition at t = 0, i.e. ψ(r, 0), to completely fix
the evolution of the wave function ψ(r, t).

4. The result of a single measurement of the observable property A always gives one
of the eigenvalues an of the operator A, b whatever the wave function ψ(r, t) the
particle is in. Moreover, whatever the wave function of the particle is before the
measurement, after the measurement it is the eigenfunction φn (r, t) that belongs
to the measured value an ; Aφb n (r, t) = an φn (r, t). This phenomenon is called the
collapse of the wave function after measurement.
This postulate is probably the most confusing part of quantum mechanics. Note that
the average result over a series of measurements, each one on a particle which starts
out in ψ(r, t), is given by the expectation value hA(t)i, cf. eq. 1.2. Postulate no. 4
tells you what happens in a single measurement. Each measurement must give you
a real number (since only real numbers can be measured), so all the eigenvalues an
of an observable A b must be real. Sometimes the eigenvalue can be any number; for
6 CHAPTER 1. QUANTUM MECHANICS

instance the position or the momentum of a free particle can take any value x or px .
In other cases the eigenvalue must be one of a set of discrete numbers. For instance,
if you measure the energy of a hydrogen atom, the result of that measurement is one
of the values En = − 13.6 n2
eV; n = 1, 2, ...2 . Here is the magic part: after you have
measured a certain property A of a particle and have obtained a value an , then if you
follow that same particle and repeat the measurement of A you will always obtain
the same value an !3 You can do this again and again; after a first measurement
select the particles for which the result was an . Following measurements on this
selected set of particles always gives the same result an . So these particles are in
a state for which the expectation value hA(t)i = an . I leave it up to you to prove
that the wave function associated with these particles can only be φn (r, t), i.e. the
eigenstate which belongs to an ; Aφ b n (r, t) = an φn (r, t)4 . In other words, whatever
the state ψ(r, t) was in which the particle set out, once we have performed the first
measurement and selected an an , we are absolutely sure that after this process the
particle is in state φn (r, t). In more fancy terminology this is phrased as “the process
of measurement collapses the wave function ψ(r, t) onto the eigenstate φn (r, t)”.

The measuring process thus plays a very active role in quantum mechanics, since it
changes the wave function the particle is in. This is in contrast with its passive role
in classical mechanics, where one assumes that one can always set up an experiment
in which the measuring process does not disturb what is measured. For instance an
object can be probed by bouncing a test particle of it. In classical mechanics one
can always, at least in thought, make the test object very light such that it disturbs
the object only in an infinitesimal way. Bouncing a test particle of a quantum
mechanical hydrogen atom in its ground state leaves it either untouched in its ground
state, or in one of its excited states and the latter possibility is a large and far from
infinitesimal disturbance. Until the present day this active role of measurement in
quantum mechanics is controversial. It never seizes to fuel a heated debate and it
has a number of famous adversaries, among them Einstein21 and Schrödinger. The
latter formulated a famous paradox known today as “Schrödinger’s cat”.5 To my
knowledge these debates, and, more importantly, the experimental data, have not
resulted in a widely accepted alternative interpretation of the measuring process,
different from the one I just gave you.

2
The zero of energy here is where the electron and the proton of the hydrogen atom are infinitely far
apart.
3
Note this is not the same situation as described in postulate no. 2. The latter states that when you
start anew with a new particle in state ψ(r, t), then the measurement can give another result an0 , which
may or may not be equal to the first result an . The average over a large series of such measurements must
give hA(t)i according to eq. 1.2.
4
If you find this hard to grasp, think it through on a specific example for A, e.g. position, momentum,
or energy. Take the latter. What the postulate says is: once you have measured the energy of a particle,
you have measured it; after the measurement, it does not change anymore (provided you do not cheat, and
disturb the particle by sending it into an external field, for instance). But a state with such a well-defined
energy must be an eigenstate of the energy operator (i.e. the Hamiltonian).
5
For those of you interested in this sort of stuff, discussions can be found in most modern introductory
quantum mechanics books, such as D. J. Griffith, Introduction to Quantum Mechanics (Prentice Hall,
Upper Saddle River, 1995); B. H. Brandsen and C. J. Joachain, Quantum Mechanics (Prentice Hall,
Harlow, 2000). Heated discussions flame up once and a while in Physics Today and the NNV blaadje; the
magazines of the American and Dutch physical societies, respectively.
1.2. QUANTUM MECHANICS 7

This set of postulates constitute what is called the “Kopenhagen” interpretation of


quantum mechanics, as formulated by Bohr22 together with a large group of people visiting
him at his institute in the Danish capital.

1.2 Quantum Mechanics


It turns out that the postulates of wave mechanics are more general than wave mechanics
itself, if we rephrase the theory a bit. Start by defining a shorthand notation |ψ(t)i =
ψ(r, t). This now is called a state or a ket, instead of a wave Rfunction. In a similar
shorthand notation, the inner product is written as hψ(t)|φ(t)i = ψ ∗ (r, t)φ(r, t)dV . The
object hψ| is called a bra, hence the name bra-ket notation.6 States form an inner product
space (Hilbert space). Observable quantities can be connected with linear operators, their
expectation values are written as
D E
b
ψ(t)|A|ψ(t)
hA(t)i = (1.8)
hψ(t)|ψ(t)i
Operator algebra’s, commutators, the Schrödinger equation, are defined as before and so
the postulates can be rewritten in this new notation.
Not only is the bra-ket notation a shorthand notation for wave mechanics, but it makes
the formalism more general. This most general formulation of quantum mechanics was set
up by Paul Dirac, hence the name Dirac notation for the bra-ket formalism. Analogous to
the old Heineken beer commercial: bra-ket’s apply to parts of quantum mechanics which
ordinary wave mechanics does not reach.

As an example of the latter statement, consider the electron spin. From observations
on the magnetism of electrons it is clear that electrons have a property called spin.7 The
same observations tell us that the spin of an electron is completely independent of its
position and motion in space; it is an independent degree of freedom. It also means that
there is no wave function α(r) that can be associated with the spin of an electron. Careful
analysis of the magnetic data leads to the following. As far as states are concerned,
we only know that two states can be associated with spin, |αi and |βi (or any linear
combination of these, hence we need a two-dimensional vector space to describe these
spin states). As far as operators are concerned, we can define sbx , sby , sbz as the (familiar)
spin operators, of which we can deduce from experiments that algebraically £ 2 ¤they behave
like angular momentum operators; i.e. [b sz , etc., and sb , sbx,y,z = 0 where
sx, sby ] = i~b
sb2 = sb2x +b s2z . Furthermore we can deduce the properties sb2 |αi = 34 ~2 |αi; sbz |αi = 12 ~|αi
s2y +b
and sb2 |βi = 34 ~2 |βi; sbz |βi = − 12 ~|βi. Oddly enough this rather formal knowledge of
spin states and operators suffices to find out everything one would like to know about
observations related to the spin of an electron. And except for this rather formal procedure
there is no other way of doing it !
Bra-ket or Dirac notation and operator algebra can also be extended to many-particle
systems, electrodynamics, solid state physics and all other branches of modern physics. It
is certainly most handy in theoretical manipulations, so we will use it in the rest of the
course.
6
Mathematically speaking, the bra’s also form a vector space. However for our purposes the bra’s only
function is to form an inner product with a ket.
7
A highlight of Dutch physics. The names of Zeeman, Goudsmit and Uhlenbeck are associated with it.
8 CHAPTER 1. QUANTUM MECHANICS

1.3 Representations
Before you read the following sections, you might want to refresh the mathematics related
to quantum mechanics. Ch. 3 of Griffiths gives a nice summary.

1.3.1 General Formalism


Even for a single (spinless) particle quantum mechanics is more general than the wave me-
chanics we have discussed in Section 1.1. Wave mechanics is just one of the representations
of quantum mechanics. It is worth while to consider a more general point of view.

We start from the time-dependent Schrödinger equation, eq. 1.7 in Dirac notation

d b (t)i
i~ |ψ (t)i = H|ψ (1.9)
dt

where |ψ (t)i is the time-dependent state and H b is the Hamiltonian. A representation is


defined starting from a basis set. An orthonormal basis set is a set of fixed (i.e. time-
independent) states |φi i; i = 1, 2, ..., having the properties

hφi |φj i = δ ij orthonormal (1.10)


X
|φi ihφi | = I complete ( ≡ resolution of indentity) (1.11)
i

Proof: if the
P states |φi i; i = 1, 2, ... form a basis set, then every possible |ψi can be written
as |ψi = i ci |φi i, with ci complex P numbers. Since P the states |φi iPare orthonormal, these
numbers are given by hφk |ψi = i ci hφk |φi i = i ci δ ki = ck , so k |φk ihφk |ψi = |ψi =
I|ψi. This proves the property 1.11. It is trivial to prove P the reverse: if eqs. 1.10, 1.11
hold then every possible |ψi can be written as |ψi = i ci |φi i.

A basis set is used to define a representation as follows. Rewrite the Schrödinger


equation, eq. 1.9, by inserting resolutions of identity, eq. 1.11:

d X X
b
X
i~ |φi ihφi |ψ (t)i = |φi ihφi |H |φj ihφj |ψ (t)i
dt
i i j

b j i ≡ Hij ; i, j =
Now use the short hand notation hφi |ψ(t)i ≡ ψ i (t) ; i = 1, 2, ... and hφi |H|φ
1, 2, ... Then the Schrödinger equation can be rearranged to
 
X d X
|φi i i~ ψ i (t) − Hij ψ j (t) = 0
dt
i j

Since all basis states |φi i are independent, the [....] need to be zero for all i. Using a basis
set, a state |ψ(t)i is thus represented by a vector with components ψ i (t) ; i = 1, 2, ... ;
an operator A b by a matrix with elements Aij ; i, j = 1, 2, ... and the Schrödinger equation
∂ P
becomes a matrix-vector equation with components i~ ∂t ψ i (t) − j Hij ψ j (t) = 0; i =
1, 2, ... Depending upon the physical problem at hand the number of components (and
1.3. REPRESENTATIONS 9

Dirac discrete basis representation


state |ψ(t)i ψ i (t); i = 1, 2... vector
operator Ab Aij ; i, j = 1, 2... matrix
d b (t)i d P
Schr. eq. i~ dt |ψ (t)i = H|ψ i~ dt ψ i (t) − j Hij ψ j (t); i = 1, 2...

Table 1.1: discrete basis representation

Dirac continuous basis representation


state |ψ(t)i ψ(x, t) function
operator A b A(x, x )0 matrix
R 0
d
Schr. eq. i~ dt b (t)i
|ψ (t)i = H|ψ ∂
i~ ∂t ψ(x, t) = dx H (x, x0 ) ψ (x0 , t)

Table 1.2: continuous basis representation

basis states) can be finite or infinite. A summary of the discrete basis representation is
shown in Table 1.1.
.
Apart from basis sets |φi i which are labeled by a discrete index i = 1, 2, ..., one can
also define basis sets which are labeled by a continuous variable x; the notation for the
basis states is |xi. The “orthonormality” and “completeness” relations of eqs. 1.10 and
1.11 are then generalized in an obvious way
¡ ¢
hx|x0 i = δ x − x0 orthonormal (1.12)
Z
dx |xihx| = I complete (resolution of identity) (1.13)
R
In terms of a continuous basis |xi every state |ψi can then be written as |ψi = dx c (x) |xi,
with c (x) a complex function given by c (x) = hx|ψi. The proof is analogous to the discrete
case. Using a continuous basis set, a state |ψ(t)i is thus represented by a function (a
“continuous” vector with components:) ψ (x, t) ≡ hx|ψ (t)i. An operator A b is represented
by a continuous matrix hx|A|x b i ≡ A (x, x ) and the Schrödinger equation becomes a
0 0

R
differential equation i~ ∂t ψ(x, t) = dx0 H (x, x0 ) ψ (x0 , t). Again the proof is completely
analogous to the discrete case. A summary of the continuous basis representation is given
in Table 1.2.
.
b i.e. A|a
The set of eigenstates |ai i; i = 1, 2... of any observable A, b i i = ai |ai i, forms
a basis set. In other words it has the properties given by eqs. 1.10 and 1.11 (for a proof
of this statement, see the exercises). This also holds for any continuous set of eigenstates.
In other words, the eigenstates of any observable can be used to form a representation.
This flexibility comes in very handy.

1.3.2 The Position Representation; Wave Mechanics Revisited


As an example we will consider the case of one particle in one dimension x in more
b,
detail. As a continuous basis set we take the eigenstates |xi of the position operator x
b|xi = x|xi.8 Obviously x can be any real number, and thus the set of eigenstates
i.e. x
8
The notation can be confusing, but it is consistent; x b denotes the position operator, and |xi an
eigenstate of this operator. This implies that when the particle is in state |xi, the probability of observing
10 CHAPTER 1. QUANTUM MECHANICS

form a continuous basis set. The representation of the state |ψ(t)i on this basis set,
i.e. ψ (x, t) = hx|ψ (t)i corresponds to what we ordinarily call the wave function. All of
wave mechanics can be derived using this so-called position representation. For instance,
the norm Nϕ of a state |ψi in wave function notation can be obtained by inserting the
resolution of identity, eq. 1.13
Z Z
Nψ = hψ|ψi = dx hψ|xi hx|ψi = dx ψ ∗ (x) ψ (x) (1.14)

The trick of inserting resolutions of identity also works for expectation values, for instance
for the position operator hb xi = hψ|bx|ψi (assume Nψ = 1 for simplicity), using eqs. 1.12
and 1.13
ZZ ZZ
0
­ 0
®­ 0 ® ¡ ¢ ­ ®
hψ|b
x|ψi = dxdx ψ|x x |b x|x hx|ψi = dxdx0 ψ ∗ x0 x x0 |x ψ (x) (1.15)
ZZ Z
¡ 0¢ ¡ ¢
= dxdx ψ x xδ x − x ψ (x) = dx x|ψ (x) |2
0 ∗ 0

It now remains to be proven that in the position representation the Schrödinger equation
gets its familiar wave mechanical form. The potential part is easy; the operator V (b x)
becomes a matrix hx0 |V (b b
x) |xi which is diagonal on the eigenstates of x
­ ® ­ ® ­ ® ¡ ¢ ¡ ¢ ¡ ¢
x) |x0 = x|V (x) |x0 = x|x0 V x0 = δ x − x0 V x0
x|V (b (1.16)
The kinetic energy part is more complicated. We first consider the momentum operator pb,
and start from the familiar commutation relation: [b x, pb] = i~ . This means hx| [b x, pb] |x0 i =
0 0 0
x, pb] |x i = hx|b
i~ hx|x i = i~δ (x − x ). But also hx| [b xpb − pbx 0
b|x i = hx|b 0
xpb|x i − hx|b
px b|x0 i =
0 0
p|x i − x hx|b
x hx|b 0
p|x i. Combining these two, it follows that
­ ® δ (x − x0 ) ¡ ¢
p|x0 = i~
x|b = −i~ δ̇ x − x 0
(1.17)
x − x0

Intermezzo on δ-functions
The last step follows from the following relation which holds for δ-functions
df
−xδ̇ (x) = δ (x) ( notation: f˙ ≡ ) (1.18)
dx
Proof: a simple integration by parts does the trick, using the familiar property of the
δ-function
Z ∞
dx δ (x) f (x) = f (0)
−∞
Z ∞ Z ∞ n o
− dx xδ̇ (x) f (x) = − [xδ (x) f (x)]∞−∞ + dx δ (x) f (x) + xf˙ (x)
−∞ −∞
= 0 + f (0) + 0f˙ (0) = f (0)
In a similar way one can prove by integration by parts
Z ∞
dxδ̇ (x) f (x + a) = − f˙ (a)
−∞
note also δ̇ (−x) = −δ̇ (x) (1.19)
the particle at that particular position x is one, and the probability of finding the particle at any other
position x0 6= x is zero.
1.3. REPRESENTATIONS 11

Resume Main Text


The momentum operator operating on a state |φi ≡ pb|ψi, now becomes in x representation
Z Z
­ ® ~ ¡ ¢ ¡ ¢
p|ψi = dx00 hx|b
hx|φi = hx|b p|x00 i x00 |ψ = dx00 δ̇ x − x00 ψ x00
i

using eq. 1.17 and Table 1.2. Changing the integration variable to x0 = x00 − x yields
Z Z
0 ~
¡ 0¢ ¡ 0
¢ ~ ¡ ¢ ¡ ¢
hx|b
p|ψi = dx δ̇ −x ψ x + x = − dx0 δ̇ x0 ψ x + x0
i i
Z
£ ¡ 0¢ ¡ ¢¤∞ ~ ¡ ¢ ¡ ¢
= − δ x ψ x + x0 −∞ + dx0 δ x0 ψ̇ x + x0
i

using eq. 1.19. The first term on the bottom line obviously gives zero, and the integral of
the second term can be done to give

~ d
hx|b
p|ψi = ψ (x) (1.20)
i dx
R
remembering the notation of eq. 1.18. From hx|b p|ψi = dx0 hx|b
p|x0 i hx0 |ψi it then also
follows that the matrix representation of the momentum operator is diagonal

~ d ¡ ¢
0 ~ d
p|x0 i = hx|x0 i
hx|b = δ x − x (1.21)
i dx0 i dx0

and using eqs. 1.16 and 1.20 it the Hamilton matrix must also be diagonal9
· ¸
pb2 ¡ ¢ ~2 ∂ 2 ¡ 0¢
b 0
hx|H|x i = hx| + V (b 0 0
x)|x i = δ x − x − +V x (1.22)
2m 2m ∂x02

Finally, the Schrödinger equation of Table 1.2 in the x (position) representation becomes
· ¸
∂ ~2 ∂ 2
i~ ψ (x, t) = − + V (x) ψ (x, t) (1.23)
∂t 2m ∂x2

Note that we have recovered all of wave mechanics essentially by using only (a) the
formal properties of a continuous basis set, eqs. 1.12 and 1.13, and the formal commutation
relation between position and momentum operators, [b x, pb] = i~.

Dirac states that a full description of quantum mechanics is given by the following
restatement of the postulates.

POSTULATES
1. A system is represented by a state (ket). All possible states |ψi form an linear inner
product space (which mathematicians call a Hilbert space), i.e. one can construct
linear combinations a|ψi + b|φi, inner products hφ|ψi = c, norms etcetera (a, b, c
are complex numbers).
9 ∂ d
We now use ∂x
instead of dx
because the wave function also has a time dependence.
12 CHAPTER 1. QUANTUM MECHANICS

2. Observables are hrepresented bB


by operators A, b . Their algebra in the form of commu-
i
tation relations A,b B
b =C b is vital because it structures the possible observations
(see the exercises).
d
3. The time propagation of states is described by the Schrödinger equation i~ dt |ψ (t)i =
b (t)i.
H|ψ

4. The measurement postulate states that each measurement of A b results in one of its
eigenvalues an and projects the wave function on the corresponding eigenstate |φn i.

The postulates are stated in a mathematical way. The physical interpretation (of
expectation values, etcetera) is the same as in Section 1.1. All sorts of representations
can be constructed from this formal quantum mechanics. We have seen the position
representation, which gives the ordinary wave mechanics of Section 1.1. Other examples
are

• the momentum representation, which uses the eigenstates of the momentum operator
as a basis set pb|pi = p|pi (see the exercises). This representation is useful in cases
when we are dealing with waves and scattering of waves, e.g. in free space or in the
solid state.

• One can also use some discrete basis set representation |φi i, which is tailored to
a specific problem. For instance, in calculations on molecules often a basis set of
atomic orbitals is used. In the solid state a similar basis set leads to the so-called
tight-binding representation.

All it needs to construct such a representation is defining of a specific (discrete or


continuous) basis set.

Why is the Dirac formalism useful?


There are two main reasons, practical, as well as basic.

1. Specific representations are often clumsy to work with. For instance, for 3 particles
in 3 dimensions wave functions look like ψ (x1 , y1 , z1 , x2 , y2 , z2 , x3 , y3 , z3 , t) and we
do not like to manipulate such a lengthy notation. In Dirac notation we simply use
|ψ (t)i instead. The same holds for operators; in a representation we often have to
manipulate differential operators. In Dirac notation we use commutator algebra as
much as possible, which only involves additions and multiplications.

2. Sometimes the formal Dirac notation is all we have. For instance, in Section 1.2 we
discussed the electron spin, which does not involve wave functions, but only states
|αi, |βi. Full knowledge of the electron spin is obtained from the spin operators
sbx,y,z and sb2 = sb2x + sb2y + sb2z ; their commutations relations [b
sx, sby ] = i~b
sz , etcetera;
and the way the latter operate on the spin states sb2 |αi = 34 ~2 |αi sbz |αi = 12 ~|αi;
sb2 |βi = 34 ~2 |βi sbz |βi = − 12 ~|βi.
1.4. MANY PARTICLES AND PRODUCT STATES 13

Lots of microscopic physical objects do not have wave functions, e.g. spins, photons
and many other quantum particles. But they can be described in quantum mechanics
using the Dirac formalism. For a calculation in which one wants to produce a number
that can be compared to a particular experiment, one can always choose a representation
that is best suited for the purpose at hand.

1.4 Many Particles and Product States


In Section 1.1 we considered wave functions, states, and observables to describe the quan-
tum properties of one particle. As far as we know, the postulates of quantum mechanics
are however valid for any number of particles in any number of dimensions. In this section
we consider how to construct states and observables for a many particle system in many
dimensions in a systematic way. On first reading I would suggest you just scan this section
to see whether you can get its general meaning. The details will be used in Chapter 6.
Let’s start with two particles.

TWO PARTICLES
A general two-particle wave function has the obvious form ψ(r1 , r2 , t). It can always be
expanded in products of one-particle wave functions.
Proof: Let φn (r); n = 1, 2... be a complete set of basis functions. Then we can write
X
ψ(r1 , r2 , t) = cn (r1 , t)φn (r2 )
n

Think of it: if we fix r1 , t then ϕ(r2 ) = ψ(r1 , r2 , t) can be Rexpanded in the basis functions
φn (r2 ) by assumption, with expansion coefficients cn = φ∗n (r)ϕ(r)d3 r; obviously cn =
cn (r1 , t). Since the latter are again functions, it is possible to expand them; cn (r1 , t) =
P
m cnm (t)φm (r1 ). Thus we get
XX
ψ(r1 , r2 , t) = cnm (t)φm (r1 )φn (r2 ) (1.24)
n m

The principle can be used to systematically construct many particle spaces. In the more
general Dirac notation, let the states |m1 i; m1 = 1, 2, ... form a basis set for one particle,
and |m2 i; m2 = 1, 2, ... a basis set for a second particle. Then the product states

|m1 (1)i|m2 (2)i; m1 = 1, 2, ..., m2 = 1, 2, ...

form a basis set for the two particle space. The notation is as follows: mi labels the
individual states; and (j) indicates the j’th particle. In quantum mechanics often the
notation |m1 (1)i|m2 (2)i ≡ |m1 (1)m2 (2)i is used as a short-hand. A general two-particle
state |ψ(t)i can then be written as a linear combination of such states
XX
|ψ(1, 2, t)i = cm1 m2 (t)|m1 (1)m2 (2)i (1.25)
m1 m2

If the two particles are distinct, for instance an electron and a proton, this would be the full
story. If the particles are identical, for instance two electrons, it turns out that the actual
two-particle state is more restricted. This is the result of the so-called symmetry postulate,
14 CHAPTER 1. QUANTUM MECHANICS

which prescribes the following. Two bosons must be in a symmetric state with respect to
the interchange of the two particles, i.e. |ψ(1, 2, t)i = |ψ(2, 1, t)i. According to eq. 1.25
the expansion coefficients must then be related as cm1 m2 (t) = cm2 m1 (t). Two fermions
on the other hand must be in an anti-symmetric state with respect to the interchange of
the two particles, |ψ(1, 2, t)i = −|ψ(2, 1, t)i, which leads to cm1 m2 (t) = −cm2 m1 (t). The
symmetry postulate is fundamental and it has far-reaching consequences, which will be
considered in more detail in Chapter 6. It can be considered as the 5’th postulate of
quantum mechanics.

A note for mathematicians. The formal mathematical notation for the product states
is |m1 (1)i|m2 (2)i ≡ |m1 (1)i ⊗ |m2 (2)i. It is called the direct product of |m1 (1)i and
|m2 (2)i or also the tensor product. If N1 is the n1 -dimensional vector space spanned by
the basis |m1 i, and N2 is the n2 -dimensional vector space spanned by the basis |m2 i, then
N1 ⊗ N2 is the n1 × n2 -dimensional direct or tensor product space; it is spanned by the
basis |m1 i ⊗ |m2 i.

Product states more or less behave as we might expect. An inner product, for instance,
goes like

hm1 (1)m2 (2)|n1 (1)n2 (2)i = hm1 (1)|n1 (1)i hm2 (2)|n2 (2)i (1.26)

i.e. we split the product state, and recombine each term individually. The same thing
written in terms of wave functions is self-evident:
Z
φ∗m1 (r1 )φ∗m2 (r2 )φn1 (r1 )φn2 (r2 )d3 r1 d3 r2 =
Z Z
φm1 (r1 )φn1 (r1 )d r1 · φ∗m2 (r2 )φn2 (r2 )d3 r2
∗ 3

The next thing is to define operators in this product space. For each operator, we must
define on which part it works. For instance, if particle no. 1 has a charge and particle no.
2 has not, an electric field would of course only operate on particle no. 1. We then have
b
an operator of the form A(1), where
³ ´
b
A(1)|m 1 (1)m2 (2)i = b
A(1)|m 1 (1)i |m2 (2)i (1.27)

The matrix elements of such an operator are given by


D ¯ ¯ E D ¯ ¯ E
¯b ¯ ¯b ¯
m1 (1)m2 (2) ¯A(1)¯ 1
n (1)n2 (2) = m 1 (1) ¯ A(1)¯ 1
n (1) hm2 (2)|n2 (2)i
D ¯ ¯ E
¯b ¯
= m1 (1) ¯A(1) ¯ n1 (1) δ m2 n2 (1.28)

assuming we have an orthonormal basis set. Operators that work only on particle no.1
and operators that work only on particle no.2 are completely independent of each other;
they must commutate.
b
Proof: Let A(1)|m b
1 (1)i = |a1 i and A(2)|m2 (2)i = |a2 i. Then

b A(2)|m
A(1) b b
1 (1)m2 (2)i = A(1)|m1 (1)a2 i = |a1 a2 i
b
= A(2)|a b b
1 m2 (2)i = A(2)A(1)|m1 (1)m2 (2)i
1.4. MANY PARTICLES AND PRODUCT STATES 15
h i
b
Since this holds for all basis states, A(1), b
A(2) =0
b
A(j) are examples of so-called one-particle operators (since they operate on only one
particle). Operators which are sums of one-particle operators are also called one-particle
operators. One often encounters operators of a type
N
X
b = A(1)
A b + A(2)
b b=
or A b
A(j) in the N -particle case (1.29)
j=1

For instance, if both particles have a charge, an electric field operates on particle no.1 as
q1 Vb (r1 ), where Vb (r1 ) is the electrostatic potential at the position r1 of particle no.1. In a
similar way it operates on particle no.2 as q2 Vb (r2 ). The full operation of the electrostatic
potential on the two particle system is then given by q1 Vb (r1 ) + q2 Vb (r2 ). Such an operator
works on the states as
³ ´
b 1 (1)m2 (2)i = A(1)
A|m b + A(1)b |m1 (1)m2 (2)i (1.30)
³ ´ ³ ´
b
= A(1)|m b
1 (1)i |m2 (2)i + |m1 (1)i A(2)|m2 (2)i

A matrix element is given by


D ¯ ¯ E
¯ b¯
m1 (1)m2 (2) ¯A ¯ n1 (1)n2 (2) =
D ¯ ¯ E D ¯ ¯ E
¯b ¯ ¯b ¯
m1 (1) ¯A(1)¯ 1
n (1) hm 2 (2)|n 2 (2)i + hm 1 (1)|n1 (1)i m2 (2) ¯ A(2)¯ 2
n (2)
D ¯ ¯ E D ¯ ¯ E
¯b ¯ ¯b ¯
= m1 (1) ¯A(1) ¯ n1 (1) δ m2 n2 + m2 (2) ¯A(2) ¯ n2 (2) δ m1 n1 (1.31)

Besides one-particle operators we can also have operators working on both particles which
cannot be written as a sum. For instance, our two charged particles will have a Coulomb in-
teraction given by v(r1 , r2 ) = |rq11−r
q2
2|
. Obviously, for the corresponding operator vb(r1 , r2 ) =
b
A(1, 2), it is not possible to split the operation as in eqs. 1.30 and 1.31, and we must write
in full
D ¯ ¯ E
¯b ¯
m1 (1)m2 (2) ¯A(1, 2)¯ n1 (1)n2 (2) (1.32)

MULTIPLE DIMENSIONS
• So far we considered product states formed from states belonging to two different
particles. However products can also be formed from states that describe different
independent degrees of freedom of one particle. For instance, let |φi be a state
describing one particle, and hr|φi =φ(r) its wave function representation. Let |si
describe the spin state of the particle. The probability of finding the particle at a
position r is completely independent of the probability of finding the particle in a
certain spin state s (in absence of an external field). The total probability is then
the product of those two probabilities, and the complete state must be the product
state |ψi = |φi⊗|si =|φi|si =|φsi. There are operators that work on only one of
the two components, such as an electrostatic potential Vb (r) which operates on φ(r)
σ, where σ
only, or B·b b is the spin operator, which describes a magnetic field working
on the spin. In addition one can have operators that work on both components.
16 CHAPTER 1. QUANTUM MECHANICS

For instance in atomic physics one has the spin-orbit coupling operator L·b b σ , which
b
couples the angular momentum L = b r× p b of an orbiting electron to its spin. The
most generalPstate when such a coupling is active then is a linear combination of the
form |ψi = n,s cns |φn si, where |φn i is some complete basis in real space and |si is
some basis in spin space.

• Pursuing the idea of the previous point, different dimensions are also different degrees
of freedom. Let |xi be the complete set of eigenstates of the (one-dimensional) posi-
tion operator x b, as used in the previous section. Let |yi,|zi be a similar set of eigen-
states of the position operators yb, zb in the other two directions. Then the product
|ri =|xyzi
 =|xi⊗|yi⊗|zi is the eigenstate of the three-dimensional position operator
b
x
b
r=  yb , i.e. b
r|ri = r|ri. Note this holds because it holds for each of the compo-
zb
nents, e.g. yb|ri =by(|xi⊗|yi⊗|zi) =|xi⊗b y|yi⊗|zi =|xi⊗y|yi⊗|zi =y(|xi⊗|yi⊗|zi) =y|ri.
Product states are thus a way of constructing higher dimensional states from single
dimensional states.10

10
The technique is related ro the technique of separation of variables when solving a partial differential
equation in higher dimensions. For instance, the eigenstates of a particle in a three-dimensional square
box can be written as Ψklm (r) = φk (x)φl (y)φm (z), where φk (x) are the eigenstates of a particle in a
one-dimensional box in the x-direction, and similar for y and z. The most general state can of course again
be written as a linear combination of such states, with coefficients cklm .
Chapter 2

Time Dependent Perturbation


Theory

“Rest, rest, perturbed spirit”, Shakespeare, Hamlet.

As you already know from your introductory courses on quantum mechanics, the num-
ber of systems for which we are able to find exact analytical solutions is very limited: the
harmonic oscillator, the hydrogen atom, just to mention a few (actually the list is not
that much longer).1 The situation is hopeless, but not desperate. We can often think
of approximations that are physically reasonable. Perturbation theory is one of the few
systematic techniques available to us to construct such a reasonable approximation. Es-
pecially time dependent perturbation theory is very versatile. Since we need it later on
in its full glory, I will introduce time dependent perturbation theory in a very general
way in this chapter. Via the so-called adiabatic theorem it is even possible to derive the
results of the perhaps more familiar time independent perturbation theory from it (see
the exercises). A general idea used throughout this lecture course, is that we try to do as
many manipulations as we can on operators instead of on states, because operator algebra
is easier. After a general exposé on time evolution, time dependent perturbation theory
is explained. A simple first order approximation then leads to Fermi’s golden rule, which
obtains its golden status because it plays a vital role in all kinds of spectroscopy (optical or
otherwise). The treatment of external perturbations as classical fields yields certain prob-
lems which can only be solved by turning to quantum field theory. However, a simple and
pragmatic solution to these problems is supplied by Einstein’s phenomenological theory of
radiative transitions. The theoretical problems are summarized in the final section. The
first appendix contains a discussion on the Heisenberg32 picture, which is another way of
looking at the time evolution in quantum mechanics. The second appendix contains some
integral tricks.

2.1 Time Evolution


b (t, t0 )
We start by defining the formal time evolution operator U
b (t, t0 ) |ψ (t0 )i
|ψ (t)i = U (2.1)
1
This situation is not unique for quantum mechanics. The list in classical mechanics is just as long (or
short, depending upon your expectations).

17
18 CHAPTER 2. TIME DEPENDENT PERTURBATION THEORY

It is also called the time development operator or time propagation operator. Of course
like all operators in quantum mechanics it has to be a linear operator. The time evolution
operator operates on a state |ψi at time t0 and evolves it to the state |ψi at time t. Using
this definition in the Schrödinger equation 1.9, one gets
· ¸
d b bUb (t, t0 ) |ψ (t0 )i = 0
i~ U (t, t0 ) − H
dt
Since this must hold or any state |ψ (t0 )i it follows that
d b bUb (t, t0 ) = 0
i~ U (t, t0 ) − H (2.2)
dt
This is an operator equation for the time evolution operator U b (t, t0 ) . Solving this equation
is completely equivalent to solving the Schrödinger equation. Since the initial condition is
Ub (t0 , t0 ) = I , by virtue of eq. 2.1, the formal solution of this equation is

Z t ¡ ¢
b (t, t0 ) = I − i
U b 0 )U
H(t b t0 , t0 dt0 (2.3)
~ t0

b (t0 , t0 ) on
as one can easily check by substituting this in eq. 2.2. One can substitute for U
the right-hand side, the whole expression of the right-hand side and get

Z t µ ¶2 Z t Z 0
¡ ¢ ¡ 0¢ t ¡ ¢ ¡ 00 ¢ 00 0
b (t, t0 ) = I − i
U b t0 dt0 +
H
i b
H t b t00 U
H b t , t0 dt dt (2.4)
~ t0 ~ t0 t0

b (t). The case for


Using this substitution repeatedly one constructs an infinite series in H
b
which the Hamiltonian H is time independent is one that we will encounter frequently.
The time integral then gives
Z t
b 0 = (t − t0 ) H
Hdt b (2.5)
t0

All the integrals in the series of eq. 2.4 can be now done easily, which leads to

µ ¶2
b (t, t0 ) = I − i (t − t0 ) H
U b+1 i b 2 + ...........
(t − t0 )2 H
~ 2 ~
· ¸
i b
= exp − (t − t0 ) H (2.6)
~

PROPERTIES
From its definition, eq. 2.1, we have the initial condition
b (t0 , t0 ) = I
U (2.7)

i.e. nothing happens if we do not move in time, t0 → t0 . Also from its definition we have
the time product

U b (t2 , t1 ) U
b (t2 , t0 ) = U b (t1 , t0 ) (2.8)
2.1. TIME EVOLUTION 19

i.e. evolving in time from t0 to t1 , followed by evolving from t1 to t2 is equivalent to


evolving from t0 to t2 . From eqs. 2.7 and 2.8 we can also define what happens if we
move backwards in time
b (t0 , t) U
U b (t, t0 ) = I

This means that the inverse time evolution operator can be expressed as
b −1 (t, t0 ) = U
U b (t0 , t) (2.9)

which also makes sense; the inverse of moving from t0 to t is moving from t to t0 . Using
eq. 2.6 the adjoint operator can be expressed as
· ¸ · ¸
b † i b † i b b (t0 , t)
U (t, t0 ) = exp + (t − t0 ) H = exp − (t0 − t) H = U
~ ~

b† = H
since H b is a self-adjoint (or Hermitian) operator (every observable must be a
Hermitian operator, see the exercises). Combining this result with eq. 2.9 we get
b −1 (t, t0 ) = U
U b † (t, t0 ) (2.10)

In other words the time evolution operator U b is a unitary operator.2 This means it
“conserves” the norm (see the exercises) of a state
D E D E
hψ(t)|ψ(t)i = U b (t, t0 ) ψ(t0 ) = ψ(t0 )|U
b (t, t0 ) ψ(t0 )|U b † (t, t0 ) U
b (t, t0 ) |ψ(t0 )

= hψ(t0 )|ψ(t0 )i

Again this makes sense; if a particle is in state |ψi at time t0 , i.e. the probability of finding
it in this state is 1 or hψ(t0 )|ψ(t0 )i = 1; then by propagating it to time t, the probability
of finding it in this state is still 1, i.e. hψ(t)|ψ(t)i = 1. In other words a particle cannot
spontaneously appear or disappear.

2.1.1 The Huygens Principle


This patriotic subsection is just for fun. The formal time evolution of eq. 2.1 has an ancient
interpretation in wave mechanics, which was given by Christiaan Huygens (pronounced in
English as “hojgens”) already in the 17th century. Huygens was actually thinking about
optics, but the main idea and the mathematics are similar, or as Feynman said: “the same
equations have the same solutions”. Write eq. 2.1 in the position representation
Z
b
hx|ψ (t)i = hx|U (t, t0 ) |ψ (t0 )i = dx0 hx|U b (t, t0 ) |x0 ihx0 |ψ (t0 )i ⇔
Z
¡ ¢
ψ (x, t) = dx0 U (x, t; x0 , t0 )ψ x0 , t0 (2.11)

b (t, t0 ) |x0 i. The interpretation is as follows; suppose at time


defining U (x, t; x0 , t0 ) ≡ hx|U
t0 we know the wave form ψ (x0 , t0 ) over the complete space x0 . Then the wave at a later
2
Also in case the Hamiltonian is time dependent, the time evolution operator is unitary. The proof can
be based upon the general series expression of eq. 2.4.
20 CHAPTER 2. TIME DEPENDENT PERTURBATION THEORY

time t can be constructed for any point x by assuming that ψ (x0 , t0 ) acts as a source.
For each point x0 at time t0 the source sends out a secondary wave which propagates to
x at time t. This secondary wave is described by U (x, t; x0 , t0 ). The wave ψ (x, t) can
be constructed by summing (or integrating) all the secondary waves U (x, t; x0 , t0 ) over all
the points x0 , each of these secondary waves properly weighted with ψ (x0 , t0 ), which is
the strength of the source at x0 . This idea works in any number of dimensions; in three
dimensions one obtains the familiar Huygens construction for the propagation of waves as
shown in Fig. 2.1.

x'
U (x, t ; x ', t0 )

ψ (x ', t0 ) ψ (x, t )

Figure 2.1: Progagation of a wave using the Huygens principle.

Admittedly, this picture is a bit misleading, since a secondary wave U (x, t; x0 , t0 ) is


emitted from each point in space x0 , and the total wave ψ (x, t) is obtained by interference
of all these secondary waves, each weighted with the strength of their source ψ (x0 , t0 ), i.e.
Z
¡ ¢
ψ (x, t) = d3 x0 U (x, t; x0 , t0 )ψ x0 , t0 (2.12)

One can prove that in the limit where the potential varies slowly over a scale which is much
larger than the typical wave length associated with such a wave, the Huygens construction
leads to classical mechanics! It was actually Hamilton who, when he reformulated classical
mechanics in the 19th century, proved this statement. Huygens’s idea was again picked up
by Feynman in the 20th century, when he reformulated the whole of quantum mechanics
on the basis of path integrals, starting from eq. 2.12. I bet you that Huygens never
dreamed that he would be one of the founding fathers of modern physics, centuries after
he wrote down his idea.
The idea is the same in optics, where an equation like eq. 2.12 can be used to describe
the propagation of waves. In the limit where the refractive index varies slowly over a scale
that is much larger than the wave length, wave optics becomes classical ray optics. The
latter is of course the field for which Huygens originally formulated his principle.3
3
A very nice book describing the connection between classical mechanics, optics, wave mechanics, path
integrals, and many other things, is D. A. Park, Classical Dynamics and its Quantum Analogues, Lecture
2.2. TIME DEPENDENT PERTURBATIONS 21

2.2 Time Dependent Perturbations


Suppose we have a Hamiltonian
b =H
H b 0 + Vb (t) (2.13)

where H b 0 is the Hamiltonian of an unperturbed system (a molecule or a crystal, for


instance) and Vb (t) is a small time-dependent perturbation (an external electromagnetic
field, for instance). Furthermore suppose that we know all there is to know about the
unperturbed system. In other words, we have solved the equation

∂ b b0 (t, t0 )
b0U
i~ U0 (t, t0 ) = H (2.14)
∂t
and we know the time evolution of the unperturbed system as given by U b0 . We would of
b , including the perturbation,
course like to solve the full time evolution U

∂ b h i
i~ U (t, t0 ) = Hb 0 + Vb (t) Ub (t, t0 ) (2.15)
∂t
Since we know U b0 already, we split off this unperturbed part and define an operator
b
UI (t, t0 ) by writing

U b0 (t, t0 )U
b (t, t0 ) = U bI (t, t0 ) ⇔ U
bI (t, t0 ) = U
b † (t, t0 ) U
b (t, t0 ) (2.16)
0

Use eq. 2.16 in eq. 2.15 (skipping the argument (t, t0 ) in the notation for the moment)
∂ hb b i b0Ub0 U
bI + Vb (t) U bI
b0 U
i~ U0 UI = H applying the chain rule gives
∂t · ¸ · ¸ · ¸
∂ b b ∂ b ∂ b b
⇐⇒ i~ U0 UI + U0 i~ UI = i~ U0 UI + Vb (t) U
b bI
b0 U
∂t ∂t ∂t
where we have used eq. 2.14. Deleting the first terms at each side yields
· ¸
b0 i~ ∂ U
U bI = Vb (t) U bI
b0 U
∂t
b † and defining
multiplying from the left with U 0

VbI (t) = U
b † (t, t0 )Vb (t)U
0
b0 (t, t0 ) (2.17)

finally gives
∂ b
i~ UI (t, t0 ) = VbI (t)U
bI (t, t0 ) (2.18)
∂t
We now have a Schrödinger like equation similar to eqs. 2.14, 2.15, but with a time
evolution which is determined by the perturbation VbI (t) only. We have achieved this by
notes in physics, Vol. 110 (Springer, Berlin, 1979).
I would also suggest a wonderfull little booklet, written for the layman by one of physics’ heroes: R. P.
Feynman, QED, the strange theory of light and matter. A must-have for every physicist!!
22 CHAPTER 2. TIME DEPENDENT PERTURBATION THEORY

splitting off the time evolution Ub0 , which we already knew, and using eqs. 2.16 and 2.17.
Quantities defined like in eq. 2.17, which we give the subscript I, constitute the so-called
interaction picture. It resembles the Heisenberg picture discussed in Appendix I, cf. eq.
2.63. The main idea of the interaction picture is to focus on the external perturbation
(which is the “interaction”). We already know the formal solution of eq. 2.18 from the
previous section, namely eq. 2.3, or as a series, eq. 2.4.4 We can write

X
bI (t, t0 ) = I +
U b (n) (t, t0 )
U (2.19)
I
n=1

b (n) is a term of the series expansion


where UI
µ ¶n Z t Z τn Z τ n−1 Z τ3 Z τ2
b (n) 1
UI (t, t0 ) = dτ n dτ n−1 dτ n−2 ...... dτ 2 dτ 1
i~ t0 t0 t0 t0 t0
VbI (τ n ) VbI (τ n−1 ) VbI (τ n−2 ) ......VbI (τ 2 ) VbI (τ 1 ) (2.20)

The main idea should be clear by now. U b (n) ∝ Vb n .


bI is sort of a power series in VbI , with U
I I
One assumes that the perturbation VbI is small, hoping that the series converges quickly
and only the first few terms will be sufficient for a decent approximation. We can now
step back from the “interaction picture” to our original “Schrödinger picture”. Using eqs.
2.16, 2.17, 2.19 and 2.20 gives us our final result for the time evolution operator

X
U b0 (t, t0 ) +
b (t, t0 ) = U b (n) (t, t0 )
U (2.21)
n=1

where the first term on the right hand side describes the time evolution of the unperturbed
system and the second term is a series of perturbation corrections. U b (n) is the term of
n’th order in the perturbation V b
µ ¶n Z t Z τn Z τ3 Z τ2
b (t, t0 ) =
(n) 1
U dτ n dτ n−1 ............ dτ 2 dτ 1 (2.22)
i~ t0 t0 t0 t0
b0 (t, τ n ) Vb (τ n ) U
U b0 (τ n , τ n−1 ) .......
b0 (τ 3 , τ 2 ) Vb (τ 2 ) U
U b0 (τ 2 , τ 1 ) Vb (τ 1 ) U
b0 (τ 1 , t0 )

Perturbation theory is one of the most prominent tools in quantum mechanics, and expres-
sions like eqs. 2.21, 2.22 are used all over the place. Richard Feynman invented a pictorial
representation for time integrals like eq. 2.22, based upon an intuitive “physical” inter-
pretation. Starting from right to left in the integrand, the system moves “unperturbed”
from the initial time t0 to a time τ 1 described by U b0 (τ 1 , t0 ). At time τ 1 an interaction
with the perturbing potential takes place, described by Vb (τ 1 ). Then the system again
moves unperturbed from time τ 1 to τ 2 before an interaction Vb (τ 2 ) takes place, etcetera.
In a so-called Feynman diagram this is pictured as in Fig. 2.2.
The diagram is defined to be completely equivalent to eq. 2.22. The direction of the
arrows gives the direction of time propagation. Each arrow describes an unperturbed
propagation between two times, and each dot (or vertex, as it is called mathematically)
represents an interaction with the perturbation at a specific time. The dictionary that
translates diagrams into mathematical equations is given in Fig. 2.3.
4
Eqs. 2.5 and 2.6 are not valid here since VbI (t) is time dependent.
2.3. FERMI’S GOLDEN RULE 23

a simple Feynman diagram


τ1 τ2 τn
t0 t
V (τ 1 ) V (τ 2 ) V (τ n )

Figure 2.2: Feynman diagram.

t0 τ1 = Uˆ (t 0 ,τ 1 )
τ1
= i=1 Vˆ (τ 1 )

Figure 2.3: Dictionary of Feynman diagrams.

1
Note that the constant i~ is absorbed in the interaction. Since the interactions can
take place at any time, integrations over the intermediate time labels τ 1 , τ 2 , ..., τ n are
always assumed, also in the diagrams. The diagrams (and thus the integrals) are time
ordered, i.e. t0 ≤ τ 1 ≤ τ 2 .... ≤ τ n ≤ t. The integrals are always written from right to
left, i.e. starting at the right with t0 , and increasing the time when going to the left. It
has to be like this since, by convention, operators work on states from right to left. Like
many other authors I write diagrams from left to right, since this is the natural order in
which we are used to read. Some authors, like Mattuck, write diagrams from bottom to
top. Several different conventions are used but confusion should not arise, since the arrows
always indicate the direction of time flow and the points at which to start and stop.
As a diagram, Fig. 2.2 looks fairly trivial. Later on in many particle physics we will see
more much complicated diagrams. In addition we will see that manipulating diagrams is
sometimes easier than manipulating mathematical equations. Since we have a dictionary
like Fig. 2.3 to translate diagrams into equations and vice versa, we can always obtain
one from the other.

2.3 Fermi’s Golden Rule

“The rule is jam tomorrow and jam yesterday - but never jam today”, Lewis Caroll, Through the
Looking-Glass.

The terms Ub (n) (t, t0 ) in the perturbation series of eqs. 2.21 and 2.22 are, loosely
speaking, proportional to Vb n . Now suppose the perturbation Vb is very, very small. It
is then sufficient to consider the first order term only and neglect all higher order terms.
24 CHAPTER 2. TIME DEPENDENT PERTURBATION THEORY

This approximation is called the Born 54 approximation.5

Z t
U b0 (t, t0 ) + U
b (t, t0) ≈ U b0 (t, t0 ) + 1
b (1) (t, t0 ) = U b0 (t, τ ) Vb (τ ) U
dτ U b0 (τ , t0 ) (2.23)
i~ t0

The corresponding Feynman diagrams are given in Fig. 2.4, where the double arrow
indicates the full, perturbed propagation, and the single arrow the unperturbed one.

τ
t0 t ≈ t0 t + t0 t
V (τ )

Figure 2.4: The Born approximation in Feynman diagrams.

An everyday example is that of a molecule or crystal subjected to an external electro-


magnetic (EM) field. In most cases the external field is much weaker than the internal
electrostatic (Coulomb) fields which hold the molecule or crystal together, so the external
field indeed represents a small perturbation and one expects the Born approximation to
be accurate. Only in very intense EM fields the approximation becomes inaccurate. With
modern lasers such intense fields are possible nowadays. We will come back to the latter
situation later on, but for the moment we will stick to the “normal” case of weak fields.
Suppose at time t0 the system is in an eigenstate of H b 0 (for instance the molecule in its
ground state) and then we switch on the perturbation, i.e. the external field. We are
interested in the probability that at time t the system is in a different eigenstate of H b0,
which is the probability that the molecule has made a transition. We use the notation i
for the “initial” state with energy ²i , and f for the “final” state with energy ²f .

b 0 |ii = ²i |ii ,
H b 0 |f i = ²f |f i
H

The transition probability Wi→f from the initial to the final state is found from elementary
quantum mechanics
¯D ¯ ¯ E¯2
¯ ¯b ¯ ¯
Wi→f = ¯ f ¯U (t, t0 )¯ i ¯ (2.24)

You have to read this equation as follows. The system starts in state |ii at t0 and it
then evolves under the time evolution operator U b that corresponds to the full Hamiltonian
b b b (t, t0 ) |ii
H0 + V . At time t we determine the overlap |hf |i0 i|2 of the evolved state |i0 (t)i = U
with state |f i, which gives usDthe 6
¯ probability
¯ E that our system is in state |f i. We thus
¯b ¯
focus on the matrix element f ¯U (t, t0 )¯ i . Using eq. 2.23 and inserting the resolution
P
of identity I = n |nihn| between each pair of operators, where |ni; n = 1, 2... are the

5
Max Born was one of the founding fathers of quantum mechanics. He derived this approximation in
the context of scattering theory for the case of a weak scattering potential. The equation is generally
accurate for any weak perturbing potential or field.
6
We assume the states |i, f i to be normalized.
2.3. FERMI’S GOLDEN RULE 25

(normalized) eigenstates of H b 0 , i.e. Hb 0 |ni = ²n | ni, one obtains


D ¯ ¯ E D ¯ ¯ E
¯b ¯ ¯b ¯
f ¯U (t, t0 )¯ i = f ¯U 0 (t, t0 )¯ i
X 1 Z t D ¯¯ ¯ ED ¯ ¯ ED ¯ ¯ E
+ dτ f ¯U b0 (t, τ )¯¯ n n ¯¯Vb (τ )¯¯ m m ¯¯U
b0 (τ , t0 )¯¯ i (2.25)
n,m
i~ t0

b0 and the fact that the states |ni are eigenstates


We now make use of our knowledge of U
b0
of H 7

D ¯ ¯ E D ¯ i ¯ E
¯b ¯ ¯ − ~ (t−t0 )Hb 0 ¯
n ¯U 0 (t, t0 ¯
) m = n ¯ e ¯m
i i
= e− ~ (t−t0 )²m hn|mi = e− ~ (t−t0 )²m δ nm

Using this in eq. 2.25 gives8


D ¯ ¯ E Z
¯b ¯ 1 t i i
f ¯U (t, t0 )¯ i = dτ e− ~ (t−τ )²f Vf i (τ ) e− ~ (τ −t0 )²i (2.26)
i~ t0
D ¯ ¯ E
¯ ¯
where Vf i (τ ) = f ¯Vb (τ )¯ i . We are interested in the special case of an harmonic per-
turbation, which we can write as9

Vb (t) = Vb0† eiωt + Vb0 e−iωt ω>0 (2.27)

Using eq. 2.27 in eq. 2.26 then yields


D ¯ ¯ E ½ Z t Z t ¾
¯b ¯ 1 − i (t²f −t0 ²i ) ∗ i(ωf i +ω)τ i(ω f i −ω)τ
f ¯U (t, t0 )¯ i = e ~ V0,if dτ e + V0,f i dτ e
i~ t0 t0
(2.28)
D ¯ ¯ E
¯ ¯
where ω f i = ~1 (²f − ²i ) and V0,f i = f ¯Vb0 ¯ i . Without loss of generality we may put
t0 = 0. The integrals can be done, and the result for the transition probability of eq. 2.24
from state i to state f becomes

Wi→f = |V0,if |2 · t 4t (ωf i + ω)
~2

+ 2 |V0,f i |2 · t 4t (ω f i − ω)
~
µ ¶2

+{ V0,if V0,f i It (ωf i − ω, ωf i + ω)
~
+ complex conjugate} (2.29)

The first term in this equation results from the first term between brackets in eq. 2.28,
the second from the second, and the third (and fourth) terms are cross terms between the
7 b 0 is time independent we use eq. 2.6.
Since H
8
Note that the zero’th order terms gives zero, since |ii and |f i are different states.
9
Here is a short explanation for this form. Since Vb has to be an observable, it must be a Hermitian
operator, i.e. Vb † = Vb . You can checkDthat
¯ ¯this
E is the £case here.¤ You can also check that the expectation
¯b¯
value with respect to a state |υi is υ ¯V ¯ υ = Re V0,υυ eiωt , i.e. the usual expression for harmonic
driving potentials.
26 CHAPTER 2. TIME DEPENDENT PERTURBATION THEORY

first and second terms in eq. 2.28. Note that the phase factor in front of the brackets
in eq. 2.28 does not play a role. The functions 4t and It are studied in more detail in
Appendix II. It is relatively straightforward (but a bit tricky, that’s why I moved it to an
appendix) to show that in the “long” time limit one obtains


tÀ =⇒ 4t (x) → δ(x), It (x) → 0 (2.30)
ω
In fact the long time limit is not just an asymtotic limit here, but it represents the real
physics of this problem. One needs a fair amount of cycles to be able to define an harmonic
potential characterized by a single frequency ω as in eq. 2.27. Elementary Fourier trans-
form analysis tells us that a signal of a finite duration t always has a spread in frequency
domain of δω = 2π t . So in order that δω ¿ ω, we have to be in the time limit given by eq.
2.30.
Because of the δ-function, the first term in eq. 2.29 can only give a contribution if
²f < ²i . Likewise, the second term only gives a contribution if ²f > ²i . The third and
fourth terms give zero contribution in the limit of eq. 2.30. If we now adapt our convention
such that i always labels the level which is lowest in energy, and f always labels the upper
level, we can write the transition probability as


Wi→f = |V0,f i |2 · t · δ (ω f i − ω) (2.31)
~2
Of more interest than the transition probability is the so-called transition rate, which is
the transition probability per unit time

dWi→f 2π ¯¯D ¯¯ ¯¯ E¯¯2


wi→f = = 2 ¯ f ¯Vb0 ¯ i ¯ δ (ωf i − ω) (2.32)
dt ~

This final result looks quite elegant and simple. It is the world famous “Fermi’s
Golden rule”.10 It is the main tool in interpreting transitions in spectroscopic experiments.
Fermi’s golden rule gives a simple, yet mostly adequate description of transition rates
in such experiments. Because of its prominent role in spectroscopy, I will discuss its
ingredients in more detail.

1. Conservation of energy. Because of the δ-function, the transition rate wi→f is non-
zero only for perturbations with a frequency which conserve the energy

~ω = ²f − ²i (2.33)

This enables one to perform spectroscopic experiments. For instance, starting with
a system in its ground state i, one can study its excited states f by scanning the
frequency ω of the external field over a selected range and picking out the states f
one at a time.
10
Enrico Fermi38 has his name attached to a wide variety of phenomena in atomic and solid state physics.
He also build the world’s first nuclear reactor at the university of Chicago (with some help from the college’s
American football team), where for the first time plutonium was produced in a sizable quantity (to the
dismay of the people of Nagasaki).
2.3. FERMI’S GOLDEN RULE 27

2. Oscillator strengths; selection rules. The “intensity” of the transition is determined by


the matrix elements of the perturbation (squared)
¯D ¯ ¯ E¯2
¯ ¯b ¯ ¯
¯ f ¯V0 ¯ i ¯ (2.34)

These intensities are also called the oscillator strengths of the transitions (this stan-
dard, but unfortunate name originates from a not very helpful classical model). Eqs.
2.33 and 2.34 are sufficient to determine a spectrum under normal conditions. How-
ever, not all transitions which are possible in principle are observed in practice. This
depends on the kind of external perturbation that is applied. To see what I mean,
consider a simple one-dimensional example, where D ¯we ¯can
E think of the matrix element
¯b ¯ R∞
as an integral in the position representation f ¯V0 ¯ i = −∞ dx f ∗ (x)V0 (x)i(x).
Suppose the system has a symmetry, and all its states i and f are either even or
odd functions; for example i(−x) = i(x) is even and f (−x) = −f (x) D ¯is odd.
¯ E If
¯b ¯
now, for instance, the perturbation V0 is even, V0 (−x) = V0 (x), then f ¯V0 ¯ i = 0
, because the integrand f ∗ (x)V0 (x)i(x) is an even × even × odd = odd function.
The corresponding transition rate is zero, even if we apply a perturbation having
the correct frequency, eq. 2.33 . Transitions which have D ¯ a ¯zero
E rate are called
¯b ¯
forbidden. Given an initial state |ii the matrix elements f ¯V0 ¯ i determine which
transitions to which |f i’s are allowed , i.e. have non-zero rates. The collection of
all possible pairs i, f together with the assignment forbidden, or allowed are called
the selection rules for the transitions belonging to that particular perturbation Vb0 .
Usually one can determine selection rules on forehand, using arguments based on
symmetry of the type presented above.11

3. Micro-reversibility. The transition rate from state f to state i is the same as the tran-
sition rate from state i to state f

wf →i = wi→f (2.35)

This follows trivially from eqs. 2.29-2.32. It is called the principle of micro-reversibility.12

Since Fermi’s golden rule is based upon first order perturbation theory (the Born
approximation), it is of course only valid in case the higher order terms can be neglected.
This however is frequently the case in practical situations, which gives the rule a very
prominent place in quantum physics.

LOOSE ENDS
Some of you may still be a bit puzzled by Fermi’s golden rule, eq. 2.32, especially by the
δ-function part. What does it mean to have a transition rate that is zero for all frequencies
except one, at which it is supposed to be infinity ? Below I will give a couple of reasons why
11
A general derivation of selection rules can be based upon group theory, which comprises a systematic
treatment of symmetry. Symmetry is a prominent subject in modern quantum physics. A little bit of it is
explored in the exercises; a full treatment would be the subject of a separate lecture course.
12
Why microscopic laws are reversible, whereas the macroscopic world clearly is not, is considered in
statistical mechanics.
28 CHAPTER 2. TIME DEPENDENT PERTURBATION THEORY

in real physical measurements one never really obtains δ-functions, but (at best) sharply
peaked function with a finite height and a finite width.

1. For finite times t, the function 4t (ω f i − ω) in eq. 2.29 will of course not be exactly a δ-
function. A picture of this function is shown in Fig. 2.8 (see Appendix II). Standard
Fourier analysis shows that this function has a finite width δω ' 2π t ¿ ω around
its center at ωf i − ω = 0. In optical absorption spectroscopy typically t À 10−6 s,
so δω ¿ 106 Hz. (mind you, these are order-of-magnitude numbers to give you just
an impression). A typical optical frequency is ω v 1015 Hz, which is also a measure
for the frequency scan in a spectroscopic experiment. This means that δω ω is indeed
very small. In fact, it is usually too small to be of any practical interest and thus we
can mathematically approximate 4t by a δ-function and It by zero as in eq. 2.30.

2. The second reason is more or less the same as the first, but it focuses more on the source
of the perturbing potential. We started with a pure harmonic driving potential in eq.
2.27. Such a perfect monochromatic potential is ideal to work with mathematically,
but physically it is a bit artificial, since in practice it can never be achieved. As an
example, electromagnetic (EM) radiation is one of the most commonly used sources
for harmonic driving potentials. EM waves can be made monochromatic to a good
degree (by using lasers, for instance), but never perfectly; there will always be a small,
but finite, frequency spread. So insteadi of eq. 2.27 we should have used a potential of
R h † 0 iω0 t
the form b
V0 (ω )e + Vb0 (ω 0 )e−iω t dω 0 . For a nearly monochromatic potential
0

Vb0 (ω 0 ) is sharply peaked around a central frequency ω0 and it goes rapidly to zero
for an ω 0 away from ω0 . The whole calculation of eqs. 2.28-2.32 can also be done for
this “nearly monochromatic potential”, but the expressions become clumsy, so I will
only give you the flavor. Eqs. 2.29-2.32 Dwill¯ contain ¯ EaDlarge
¯ number
¯ E of cross terms
¯b 0 ¯ ¯ b 00 ¯
between different frequencies of a type f ¯V0 (ω )¯ i f ¯V0 (ω )¯ i . If we assume
that the phases of all these factors for different frequencies are unrelated (and this
is indeed the case for the usual sources), then one can show that the sum of all such
cross terms is zero. The resulting expression then becomes a rather simple extension
of eq. 2.32
Z
2π ¯¯D ¯¯ b 0 ¯¯ E¯¯2 ¡ ¢
wi→f = 2 ¯ f ¯V0 (ω )¯ i ¯ δ ω f i − ω 0 dω 0 (2.36)
~

We define an oscillator strength gf i by a simple extension of eq. 2.34 as


Z ¯D ¯ ¯ E¯
¯ ¯ b 0 ¯ ¯2
gf i = ¯ f ¯V0 (ω )¯ i ¯ dω 0 (2.37)

Now we formally define a line-shape function F (ω0 − ω 0 ) by

1 ¯¯D ¯¯ b 0 ¯¯ E¯¯2
F (ω 0 − ω0 ) = ¯ f ¯V0 (ω )¯ i ¯ (2.38)
gf i

Since we assumed that Vb0 (ω0 ) is sharply peaked around ω 0 = ω 0 , F (ω 0 − ω 0 ) is


sharply peaked around 0; let us call the typical width of this peak 4ω. Note that
2.3. FERMI’S GOLDEN RULE 29

F(ω)

∆ω

Figure 2.5: Typical line-shape function F (ω).

R
F (ω 0 − ω 0 ) is normalized, i.e. F (ω 0 − ω 0 ) dω0 = 1. A typical line-shape function is
shown in Fig. 2.5.

Using these definitions the resulting expression for the rate becomes


wi→f = gf i F (ω f i − ω0 ) (2.39)
~2

which is a bit more complicated than our original expression of eq. 2.32, but at
least everything if finite now, i.e. no δ-function infinities. In the monochromatic limit
we take lim4ω→0 F (ω f i − ω 0 ) = δ(ω f i − ω 0 ), while keeping the oscillator strength
gf i fixed. In this sense Fermi’s golden rule of eq. 2.32 is the mathematical limit of
the more physical expression of eq. 2.39.

An optical light source such as a mercury lamp has a typical frequency ω0 v 1015 Hz
and a typical width 4ω v 1010 Hz. The line-shape function thus indeed defines a
very narrow and sharp peak. Laser light can have a typical width 4ω which is even
a few orders of magnitude lower, thus defining an even narrower line shape. As we
will see in the next section, in many calculations we will integrate the transition
rate over a frequency range which certainly covers the entire line width. So from a
pragmatic point of view it does not matter then whether we use the expressions of
eq. 2.32 or eq. 2.39.

3. The third reason is a very fundamental one. Even if the light source would be perfectly
monochromatic, then still the line shape of a transition would not be a δ-function.
The reason is that excited states of atoms and molecules are not infinitely sharp,
but they have a finite width called the natural line width. In order to understand
this one has to look at the quantum character of EM radiation, which we will do at
a later point. The natural line width 4ω is typically around 1010 Hz, which means
it can in principle be scanned using a laser source with a very narrow width.
30 CHAPTER 2. TIME DEPENDENT PERTURBATION THEORY

2.4 Radiative Transitions


2.4.1 Atom in a Radiation Field
Detailed applications of quantum mechanics are not part of this course; they are supposed
to be part of (advanced) courses on molecular or condensed matter physics, for example.
Yet Fermi’s golden rule plays such an important role in the interpretation of all kinds
of spectroscopy that I will consider its application in somewhat more detail by a simple,
yet physical, example of an atom or molecule in an external electromagnetic field. The
complete Hamiltonian which describes the interaction between an electromagnetic field
and matter has a somewhat complicated form. A full discussion of it is not possible
within the constraints of this course, so I will present a simplified description, which is
not wrong, but limited.
Consider an electromagnetic (EM) wave, the electric part of which is given by
 
E0
E (r, t) =  0  cos (kz − ωt) (2.40)
0
The electric field E oscillates along the x-direction, and z is the direction of wave propa-
gation. The magnetic part of the wave is then of course given by
 
0
B(r, t) =  B0  cos (kz − ωt)
0

The size of an atom or molecule is typically in the range 1-10Å, whereas the wave length
λ = 2π 3
k of visible light is typically in the range 4-8 × 10 Å. To a very good approximation
we may therefore neglect the spatial dependence of the EM wave over the molecule and
treat the fields as homogeneous, i.e. spatially independent. For a molecule in the origin
we can use z = 0 in eq. 2.40. All electric charges in the molecule interact with the electric
field. Assuming that the molecule as a whole is neutral, its energy V in a uniform electric
field E is given by simple electrostatics as
N
X n
X
V = −µ · E where µ = ZI RI − e ri (2.41)
I=1 i=1

is the total dipole moment of the molecule; ZI , RI are the charges and positions of the
nuclei, −e and ri are the charges and positions of the electrons and N, n are the number of
nuclei and electrons, respectively. We now assume that the correct quantum mechanical
operator is given by

Vb = −b
µ·E (2.42)

b is the dipole moment operator, which is the same expression as eq. 2.41, but
where µ
with positions replaced by position operators. Using eqs. 2.40 and 2.41 and the spatial
homogeneity of the field we get

Vb (t) = −b
µx E0 cos ωt (2.43)

A couple of questions might puzzle the critical observer


2.4. RADIATIVE TRANSITIONS 31

• Is there no interaction with the magnetic field? Yes there is, but a more elaborate
treatment shows that this interaction is much weaker, so we neglect it here. It gives
rise to much weaker magnetic dipole transitions.

• What happens if the molecule is larger and/or the wave length is shorter such that
we may no longer assume that the electric field is homogeneous? We have to do a
multipole expansion and consider quadrupoles and higher multipoles.

• What if we have a very large system for which we expect even a multipole expansion
to be invalid, such as, for instance, a crystal. Or what happens if we use very short
wave length radiation, such as X rays. In those cases we have to take a different
approach right from the start, which (hopefully) you will find in your advanced
topical courses.

Let’s now consider the “simplest” system: a hydrogen atom.13 The dipole moment of
a hydrogen atom is simply µ = −er, where r points from the proton to the electron, so
eq. 2.43 becomes
E0
Vb (t) = eb x (eiωt + e−iωt )
xE0 cos ωt = eb (2.44)
2
Fermi’s Golden rule, eq. 2.32, gives for the transition rate between two states of the
hydrogen atom

πe2 2
wi→f = x| ii|2 δ (ω f i − ω)
E |hf |b (2.45)
2~2 0

The cycle averaged intensity I of the EM wave is proportional to I v E02 . In other words
the transition rate is proportional to the intensity of the EM wave. This is called the linear
term. If, instead of using only first order perturbation theory, we use additional higher
order n = 2, 3... perturbation terms in eq. 2.23, the latter lead to terms in the transition
rate that are proportional to I n ; n = 2, 3... These higher order terms are collectively called
the non-linear terms. Usually, in ordinary spectroscopy the intensity is sufficiently small
such that the linear term is by far dominant. In very intense laser beams the non-linear
terms can become non-negligible. We are then in the realm of non-linear spectroscopy,
where Fermi’s gold no longer rules alone, but in principle you know how to produce new
rules; simply add higher order perturbation terms. It is not too difficult to prove that in
an “ordinary” laser beam, where E0 ∼ 103 V/m, even the Fermi Golden rule rate wi→f ,
eq. 2.45, is small and higher order terms are certainly negligible (see the exercises).

Some final remarks concerning eq. 2.45


¡ ¢
x| ii|2 is of dimension “surface” m2 and in fact of order a20 (where a0 =
• |hf |b
0.529177249 × 10−10 m is the Bohr radius of the hydrogen atom). It roughly corre-
sponds to the “size” of a cross-section through the hydrogen atom and is therefore
called the absorption cross-section . This holds for a molecule in general, the absorp-
tion cross section is proportional to the size of the molecule.
13
Conceptually, a general molecule is not more complicated in principle, but the notation gets a bit
messy. In practical calculations, a general molecule is just more work (actually a lot more).
32 CHAPTER 2. TIME DEPENDENT PERTURBATION THEORY

• We are often not interested in the fine grained frequency details of a transition, but
in the transition as a whole. A convenient average transition rate is defined as
Z ωf i + 12 4ω
1
wi→f = wi→f dω (2.46)
4ω ωf i − 12 4ω

where we choose 4ω to be a frequency interval that encompasses the whole line


shape, see the previous section. So once again the δ-function only plays a mathe-
matical role.

2.4.2 Einstein Coefficients and Rate Equations


A straight-forward use of the transition rate as given by eq. 2.45 leads to a fundamental
problem. The reason is that our transition probability is proportional to the time t, cf.
eq. 2.31

Wi→f = wi→f t (2.47)


1
This can obviously not be true for all times since for t > wi→f the transition probability
Wi→f > 1, which is obvious nonsense. To make matters even worse, the non-linear
terms discussed above lead to terms in the transition probability that are proportional to
tn ; n = 2, 3... The conclusion is that t cannot become too large in order to be able to use the
transition rate wi→f of eq. 2.45. The way out of this dilemma is not to focus on one type
of transition like we did up till now, but to consider all possible types of optical transitions
between the two levels i and f at the same time. It is possible to derive all these processes
starting from quantum mechanical first principles. To do this however, one has to treat
the EM field as a quantum field and not as a classical field as we did up till now. A full
discussion of the quantum interaction between the EM field and the molecule is lengthy
and requires a separate course.14 Instead I will describe a semi-phenomenological model
which was devised by Einstein. It starts from the three basic types of optical processes
which involve the two levels i and f , as shown in Fig. 2.6.

εf

Bif D(ω ) A fi B fi D(ω )

εi
absorption spontaneous stimulated
emission emission

Figure 2.6: Optical processes involving the levels i and f ; Einstein A, B coefficients.

The Einstein model contains the following processes:


14
A relatively accessible introduction can be found in: R. Loudon, The Quantum Theory of Light (Claren-
don, Oxford, 1981).
2.4. RADIATIVE TRANSITIONS 33

1. Absorption at a rate Bif D (ω), in which photons of energy ~ω are absorbed from the
external radiation field by the atom and the latter is excited from state i to state
f . Einstein postulated the absorption rate to be proportional to D (ω), which is the
energy density of an electric radiation field of frequency ω per unit of frequency.15
In general it can be expressed as D (ω) = DT (ω) + DE (ω), where DT is the energy
density of the thermal radiation (i.e. the background, so-called “black body” radia-
tion which is always present at a finite temperature T ), and DE is the energy density
due to the radiation field of the external EM source. The proportionality constant
for the absorption rate Bif is called the Einstein B-coefficient for absorption. It is
supposed to be a coarse grained quantity that is representative for the transition
between the states i and f as a whole and does not depend on the frequency. So in-
stead of using a fine grained energy density D (ω), one should use the coarse grained
D(ω f i ) defined as
Z ω f i + 12 4ω
1
D(ωf i ) = D (ω) dω (2.48)
4ω ωf i − 12 4ω

where 4ω is a frequency interval that covers the whole line shape of the transition.16
From standard EM theory we have for a monochromatic electric wave of a single
frequency ωf i
Z ω f i + 12 4ω
1
DE (ωf i ) 4 ω = DE (ω) dω = ε0 E02 (2.49)
ω f i − 12 4ω 2

where ε0 is the electric permittivity of free space. At the right hand side of this
expression one finds the familiar energy density (energy per volume) of a monochro-
matic wave. With usual light sources (including lasers) in the visible part of the
spectrum, at not too high a temperature DE À DT , so we may neglect DT when-
ever an external field is present. In a quantum mechanical interpretation D (ω) is
proportional to the number of photons present with an energy ~ω, as we will see in
Chapter 5. From the foregoing discussion it will be clear that Bif is related to our
previously calculated quantum transition rate wi→f , eq. 2.45. A consistent way of
defining Bif is by integrating the rates over the frequency interval 4ω that covers
the whole transition, as in eq. 2.46
1
wif 4 ω ≡ Bif DE (ωf i ) 4 ω = Bif · ε0 E02 (2.50)
2
Using eqs. 2.46 and 2.45 means that the Einstein B-coefficient for absorption is
given by the expression
πe2
Bif = 2
x| ii|2
|hf |b (2.51)
ε0 ~
In the following I will use D (ω) as shorthand notation for the coarse grained D(ω f i )
of eq. 2.48. Hopefully this will not lead to a misunderstanding.
Absorption is not the only possible optical process; there are two more.
15
So D (ω) has the dimension of Js/m3 , i.e. energy per volume per frequency
16
∆ω is a frequency interval around ω as in eq. 2.46. It might seem rather arbitrary, but we need not
worry, since in the final results it drops out.
34 CHAPTER 2. TIME DEPENDENT PERTURBATION THEORY

2. Stimulated emission at a rate Bf i D (ω), in which photons of energy ~ω are put back
into exactly the same mode as that of the external radiation field, i.e. at the same
frequency and direction as the external field. The atom is de-excited from state f
to state i.

3. Spontaneous emission at a rate Af i , in which also photons of energy ~ω are send out,
but in any mode. Spontaneously emitted photons can have any direction. In fact,
the probability that a spontaneously emitted photon has the same direction as the
incoming field is negligibly small.
The difference between stimulated and spontaneous emission is shown schematically
in Fig. 2.7.

non-absorbed radiation +
stimulated emission

incoming spontaneous
wave emission

Figure 2.7: Radiative processes.

In summary, Bif , Af i and Bf i are called the Einstein coefficients for the absorption,
spontaneous and stimulated emission and Figs. 2.6 and 2.7 summarize the Einstein model
for radiative processes.17 Remarkably, the A, B coefficients are not independent but they
are related by
µ 3¶

Bif = Bf i ; Bf i = Af i (2.52)
π 2 c3
Proof: As remarked before, these relations can in principle be proven from basic
quantum mechanics, but the proof is lengthy. Einstein gave a very elegant alternative
proof, which is essentially based upon statistical physics. Call the occupation probability
(or population) of the lower level Ni , and that of the upper level Nf . The system is
supposed to be in one of these two states, so Ni + Nf = 1. If the processes given in Fig.
2.6 are the only ones possible, we can write down a phenomenological rate equation for
these populations
dNi dNf
=− = −Ni Bif D (ω) + Nf Bf i D (ω) + Nf Af i (2.53)
dt dt
The rate Nf Af i at which the upper level f is depopulated by spontaneous emission is
given by the transition rate Af i times its population Nf , etcetera. It is identical to the
17
In various branches of physics there are Einstein models around, since Einstein was a very active man
who worked on very diverse problems in his younger years. Einstein’s A and B coefficients are still used
today.
2.4. RADIATIVE TRANSITIONS 35

rate at which the lower level i is populated, since the system has to be in one of these two
states. In thermal equilibrium, in absence of any external radiation, the level populations
must be constant (which is what equilibrium means) so

dNi dNf
=− =0 (2.54)
dt dt
From eq. 2.53 we then obtain the energy density of the thermal radiation, which is present
in thermal equilibrium

Af i
DT (ω) =
(Ni /Nf ) Bif − Bf i

In addition we know from statistical physics that in thermal equilibrium the level popula-
tions are related by a Boltzmann factor, i.e.

Nf ~ω
= e− kT
Ni
using eq. 2.33. This yields

Af i
DT (ω) = ~ω (2.55)
e kT Bif − Bf i

On the other hand, the energy density of thermal (black body) radiation is also given by
Planck’s law18
~ω 3 1
DT (ω) = 2 3 ~ω (2.56)
π c e kT − 1

For the two expressions of eqs. 2.55 and 2.56 to be identical, the A, B coefficients must
be related as in eq. 2.52.

2.4.3 Population and Lifetime


Back to our main route; we now apply the external radiation field and neglect the thermal
radiation field. The rate equations of eq. 2.53 can be solved in a straightforward way
Ni + Nf = 1. Starting from the initial conditions Ni (0) = 1 and Nf (0) = 0 the solution is
µ ¶−1
Af i
Nf (t) = +2 {1 − exp [− (Af i + 2Bif DE ) t]} (2.57)
Bif DE
Ni (t) = 1 − Nf (t)

using Ni + Nf = 1. Since rates are always positive, the population Nf (t) increases mono-
tonically in time, but only to a maximum
µ ¶−1
Af i
Nf (∞) = +2 (2.58)
Bif DE
18
See e.g. S Gasiorowicz, Quantum Physics, (Wiley, New York, 1974), Ch.1, or B. H. Brandsen & C. J.
Joachain, Quantum Mechanics, (Prentice Hall, New Jersey, 2000), Ch. 1.
36 CHAPTER 2. TIME DEPENDENT PERTURBATION THEORY

One observes that always Nf (t) < 12 , which means that Ni (t) > 12 and thus Nf (t) <
Ni (t). For light sources of ordinary intensity (including most lasers) the spontaneous
emission dominates, i.e. Af i À Bif DE . This yields Nf ¿ 12 and Nf ¿ Ni , i.e. the
upper level is much less populated than the lower level. The net rate of loss of energy
from the incoming beam of radiation in the propagation direction is equal to the rate
Ni Bif DE at which photons are taken out of the beam by absorption minus the rate
Nf Bif DE at which photons are put back into the beam by stimulated emission, since the
probability that spontaneous emission puts a photon back in the propagation direction is
negligible. In other words the net rate of loss of energy from the incoming beam is given
by (Ni −Nf )Bif DE ~ω. One loosely speaks of “absorption” of light to describe this result,
although strictly speaking this is not correct, since all three optical processes are involved.
The fact that always Nf < Ni indeed shows that the beam is attenuated. If we could find
a way such that Nf > Ni , which is called population inversion, then the beam intensity
would be amplified. We would have energy gain instead of energy loss, or in other words
we would have produced a laser. The foregoing analysis shows that population inversion
is impossible in a simple two-level system. Other tricks have to be used to get a laser; for
instance, a three-level system.

Note that by considering rate equations involving all three optical processes we obtain
a finite (and usually small) population of the upper level, which solves the problem we had
with eq. 2.47. “To shed more light on it”, suppose at a time t1 we turn off the external EM
field when the occupation of the upper level is Nf,1 ≡ Nf (t1 ). In absence of an external
field and, as usual, neglecting the thermal radiation field, spontaneous emission is the only
optical process left. The upper level depopulates by spontaneous emission

dNf
= −Nf Af i ⇒ Nf (t) = Nf,1 e−Af i t (2.59)
dt
The characteristic time τ R = A1f i is called the radiative lifetime (also called the fluorescent
lifetime) of the level f . The quantum mechanical expression for the hydrogen atom is given
by

e2 ω 3
τ −1
R = Af i = x| ii|2
|hf |b (2.60)
ε0 π~c3

using eqs. 2.51 and 2.52. A typical value is τ R ∼ 10−9 s. The typical time t to be inserted
then in eq. 2.47 to calculate the transition probability is of order τ R , since after that time
the state f , populated by absorption, is depopulated again by emission. It is as if we are
trying to fill a bucket f which has a large hole in it. The time τ R is long enough compared
to the “optical” time τ v ω −1 v 10−15 s to make the “near δ-function” approximation
discussed below eq. 2.35 valid. However, it is short enough for the transition probability
of eq. 2.47 Wi→f ¿ 1, as one can easily prove by a direct calculation. This shows that
our analysis of the problem is internally consistent.

2.5 Epilogue
The discussion regarding radiative processes was not completely satisfactory from a fun-
damental point of view (at least in my opinion).
2.6. APPENDIX I. THE HEISENBERG PICTURE 37

1. We had to involve some delicate assumptions in Section 2.4 regarding the observation
time scale t. It should be large enough to get a sharply peaked δ-like transition
line shape, but small enough such that the transition probability Wi→f would not
grow to an unphysically large value. By using the phenomenological Einstein model
for absorption and emission of radiation we could argue that the lifetime of the
excited state τ R sets the physical time scale t. The numerical values of the Einstein
coefficients in “normal” (i.e. not too strong) radiation fields are consistent with our
assumptions regarding this time scale, which is somewhat reassuring.

2. Our quantum mechanical treatment of the external field as a perturbation can de-
scribe the absorption Wi→f and the stimulated emission Wf →i rates using Fermi’s
golden rule. However it cannot describe spontaneous emission (we needed the phe-
nomenological Einstein model to do that). Once the external field is switched off,
the perturbation is gone and the system can stay in an excited state forever, since
b 0 . This is clear nonsense, since
it is an eigenstate of the unperturbed Hamiltonian H
we know experimentally that excited states have a finite lifetime τ R .

3. Another unsatisfactory point is that, whereas we treated our hydrogen atom quan-
tum mechanically, we assumed the external radiation field to be given by the classical
expression of eq. 2.40. We mentioned photons, but where are they in the theory ?

All these problems have the same origin ! If we want to understand the full behavior
of a quantum system, we have to treat everything quantum mechanically, including the
external fields. The full quantum mechanical Hamiltonian contains an atom, molecule
or crystal part as before, but also a term describing the (quantum) radiation field, as
well as a term that describes the coupling between the atom/molecule and the radiation
field. It is this last term that can be considered as a perturbation term. As we will see in
Chapter 5, the states of the radiation field can be interpreted in terms of quantum particles
called photons. The “ground state” of the radiation field then is a state describing the
situation in which there are no photons present (this is the lowest energy state). The first
excited state contains one photon, etcetera. The coupling term in the total Hamiltonian
can describe a transition between (1) a molecule in an excited state (f ) + the radiation
field in its ground state (no photons) and (2) the molecule in its ground state (i) + the
radiation field in its first excited state (one photon). This is what we call spontaneous
emission! The details of this problem require a lengthy discussion, so we will not pursue
it here.

2.6 Appendix I. The Heisenberg Picture


Just before Schrödinger formulated his wave mechanics, another form of quantum me-
chanics was formulated by Heisenberg. The latter is nowadays called the Heisenberg pic-
ture, whereas the form of quantum mechanics introduced in the first chapter is called the
Schrödinger picture. Both these pictures appeared to be able to solve the same problems,
such as the spectrum of the hydrogen atom, so for a while in the 1920’s people were puzzled
by the fact that two apparently quite different quantum theories existed that both gave
the same results for measurable quantities. The main difference between the two pictures
is the way in which time evolution is treated. Later on it was realized how these two pic-
tures are connected and in the modern formulation of quantum mechanics, using the time
38 CHAPTER 2. TIME DEPENDENT PERTURBATION THEORY

evolution operator, we will see that the relation between the two pictures is almost trivial.
I will briefly discuss the Heisenberg picture here. Since it focuses on operators, rather
than on states, it still has a lot of relevance to modern quantum mechanics, especially to
quantum field theory.

I will restrict the discussion to operators that do not explicitly depend on time. In the
Schrödinger picture such operators A bS are fixed, i.e. time independent. I use the subscript
“S” to indicate the Schrödinger picture. The states |ψ S (t)i are of course time dependent.
Their time evolution is given by the Schrödinger equation, and from Section 2.1 we know
that formally the solution can be given as
b (t, t0 )|ψ S (t0 )i
|ψ S (t)i = U (2.61)
where t0 is some fixed starting point in time. The states in the Heisenberg picture are
simply defined as
b † (t, t0 )|ψ S (t)i
|ψ H i ≡ |ψ S (t0 )i = U (2.62)
using eq. 2.10 and using the subscript “H” to indicate the Heisenberg picture. Note that
the states in the Heisenberg picture are fixed, i.e. time independent. The Heisenberg
operators AbH (t) are considered to be time dependent. Their time dependence can be
deduced from the fact that expectation values, which are the measurable quantities, must
be identical in both the Schrödinger and Heisenberg pictures.
bS |ψ S (t)i = hψ S (t0 )|U
hψ S (t)|A b † (t, t0 )A b (t, t0 )|ψ S (t0 )i
bS U
= hψ H |AbH (t)|ψ H i

using eq. 2.61. From the definition eq. 2.62 it is then self-evident that one must define
bH (t) = U
A b † (t, t0 )A b (t, t0 )
bS U (2.63)
Note that at the starting point Heisenberg and Schrödinger operators are identical, i.e.
bH (t0 ) = A
A bS (2.64)
The algebra of Heisenberg operators is the same as that of Schrödinger operators, provided
we look at “equal times”. One can easily prove from eq. 2.63 that
h i h i
A bS = C
bS , B bS ⇔ A bH (t), B
bH (t) = C bH (t) (2.65)
h i
Note that the commutator A bH (t), B
bH (t0 ) ; t 6= t0 is not so easy to derive, but luckily we
won’t be needing it.

All that is needed now is to derive an equation that describes the time evolution of
these Heisenberg operators.
bH (t)
dA d ³ b† b (t, t0 )
bS U
´
i~ = i~ U (t, t0 )A
dt dt à ! à !
dUb (t, t0 ) dUb † (t, t0 )
b † (t, t0 )A
= U bS i~ + i~ bS U
A b (t, t0 )
dt dt
= U bS U
bS H
b † (t, t0 )A b † (t, t0 )H
b (t, t0 ) − U bS U
bS A b (t, t0 )
2.6. APPENDIX I. THE HEISENBERG PICTURE 39

where the Schrödinger equation, eq. 2.2, and its complex conjugate is used to derive the
last line. One can use eqs. 2.10 and 2.63 to transform the last line into

bH (t)
dA
i~ b † (t, t0 )A
= U bS U(t, b † (t, t0 )H
b t0 )U bS U b † (t, t0 )H
b (t, t0 ) − U bS U b † (t, t0 )A
b (t, t0 )U bS U
b (t, t0 )
dt
= A b H (t)A
b H (t) − H
bH (t)H bH (t)

The last line can be condensed to give the final result.

bH (t) h
dA i
i~ b b
= AH (t), HH (t) (2.66)
dt
This is called the Heisenberg equation. It describes time evolution in the Heisenberg pic-
ture, and it obviously plays the same role as the Schrödinger equation in the Schrödinger
picture.

In summary, the Heisenberg picture is characterized by fixed states and time dependent
operators, whereas in the Schrödinger picture it is the other way around, i.e. fixed operators
and time dependent states. Time evolution in the Heisenberg picture is given by the
Heisenberg equation, eq. 2.66. Obviously both pictures give the same expectation values
and thus the same measurable quantities. In fact the relation between the two pictures is
given by a unitary transformation defined by the time evolution operator, cf. eqs. 2.62
and 2.63.

In view of the simple relation between the two pictures, one might wonder what the
fuss in the 1920’s was about. Well, Heisenberg used a completely different route to arrive
at his equation, eq. 2.66, which at first sight indeed looks very different from the wave
equation derived by Schrödinger. Heisenberg started from the classical Hamilton equa-
tions of motion. We shall work our way backwards. Consider a simple one-dimensional
Hamiltonian H b = pb2 + V (b
x). We drop the subscript “H” and assume that we are in the
2m
Heisenberg picture. Using eq. 2.66 and the commutation relation [b x(t), pb(t)] = i~, it is
then straightforward to show that the time propagation of the position and momentum
operators is given by

db
x(t) pb db
p(t) dV (b
x)
= and =− (2.67)
dt m dt dbx
These equations look the same as the Hamilton equations of motion, except that here
the quantities are operators x b, pb instead of simple numbers x, p as they are in classical
mechanics. As usual the step from quantum to classical mechanics is easy —one drops
the operator aspects— but the other way around is not easy, if you have to invent the
concept of operators first. Heisenberg did just that; at some point he realized that in
quantum mechanics x b, pb are non-commutating quantities, unlike in classical mechanics.
By a clever reasoning he succeeded in deriving the commutation relation [b x(t), pb(t)] = i~
and from there he derived his time evolution equation 2.66, which, by the way, also lead
him to his famous uncertainty principle. The Heisenberg route is still in practice today,
especially in quantum field theory. Consider for instance the electro-magnetic (EM) field.
From Maxwell’s equations it is possible to derive a classical Hamiltonian. It proves to
be difficult to write down quantum states (for a good reason, there aren’t any quantum
40 CHAPTER 2. TIME DEPENDENT PERTURBATION THEORY

mechanical wave functions associated with the EM field). This hinders the Schrödinger
route to quantization, since it is difficult to guess a correct Schrödinger equation for the EM
field without having the states. But it is possible to identify “position” and “momentum”
operators for the EM field and by using the familiar commutation relation between them,
the Heisenberg route to quantization is open. More about this in Chapter 5, Section 5.5.

2.7 Appendix II. Some Integral Tricks


This appendix is included for reasons of completeness. It summarizes some tricks associ-
ated with Fourier transforms, δ-functions, etcetera. In principle, you can find them in any
book on Fourier transforms. The idea is to show where the functions 4t (ω) and It (ω),
which are used in eqs. 2.29 and 2.30 in Section 2.3 come from and how they are handled.

¯Z ¯2 Z
¯ T ¯ T
¯ e iωτ ¯
dτ ¯ = T · eiωτ dτ (2.68)
¯
0 0

for T = 2π
ω · n, where n is a natural number
Proof
Z T Z T Z T Z T
2 iωτ −iωτ 0 0 0
| | = e dτ · e dτ = eiω(τ −τ ) dτ dτ 0
0 0 0 0

change to a new variable τ 00 = τ − τ 0


Z T Z T −τ 0
00
= dτ 0 eiωτ dτ 00
0 0
Z "−τ
Z 0 Z #
T T −τ 0
iωτ 00 iωτ 00
= dτ 0 e dτ 00 + e dτ 00
0 −τ 0 0

change to a new variable τ 000 = T + τ 00


Z T " Z Z #
T T −τ 0
0 −iωT iωτ 000 iωτ 00
= dτ e e dτ 000 + e dτ 00
0 T −τ 0 0
Z T Z T
0
= dτ eiωτ dτ QED
0 0

For n even we have


Z T Z T
2 0 T
eiωτ dτ = eiωτ dτ 0 using τ 0 = τ −
0 − T2 2

We now define the function 4T (ω) by


Z T
1 2
4T (ω) = eiωτ dτ (2.69)
2π − T2
2.7. APPENDIX II. SOME INTEGRAL TRICKS 41

which according to eq. 2.68 gives


¯Z ¯2
¯ T ¯
¯ e iωτ
dτ ¯¯ = 2πT 4T (ω) (2.70)
¯
0

Note that for T → ∞ eq. 2.69 corresponds to the regular definition of a δ-function

lim 4T (ω) = δ (ω) (2.71)


T →∞

T
∆ T (ω) 2π


T

Figure 2.8: The function 4T (ω) .

In Section 2.3 also the following integral occurs, which by eq. 2.69 can be written as
Z T Z T
¡ ¢ 2 2 0 ¡ ¢
IT ω, ω0 = eiωτ
dτ · eiω τ dτ = (2π)2 4T (ω) 4T ω 0 (2.72)
− T2 − T2

Obviously, from eq. 2.71 it follows that for ω 6= ω 0


¡ ¢
lim IT ω, ω 0 = 0 (2.73)
T →∞

So far, we have only considered special times T , but a general time t can always be
written as t = T + 4t, where 0 ≤ 4t < 2π ω · 2. In this case we write

¯Z ¯2
¯ T +4t ¯
¯ iωτ
e dτ ¯¯ = 2πT 4T (ω)
¯
0
Z T +4t Z T 0
Z T Z T +4t
iωτ −iωτ 0
+ e dτ · e dτ + ....
T 0 0 T
Z T +4t Z T +4t
+ ....
T T

The integrals can be done straight-forwardly and give something like

= 2πT 4T (ω) + 2π 4 t f (ω 4 t) 4T (ω) + O(4t2 )


42 CHAPTER 2. TIME DEPENDENT PERTURBATION THEORY

where f (ω 4 t) oscillates between ±1 as a function of 4t. In the limit of large T (or


t), i.e. T À 2π
ω , the first term of this expression is dominant. In other words, a good
approximation is
¯Z t ¯2
¯ ¯
¯ e dτ ¯
iωτ
≈ 2πt 4t (ω)
¯ ¯
0
→ 2πtδ (ω) (2.74)

and at the same time


¡ ¢
It ω, ω0 ≈ 0 (2.75)

The function 4T (ω) of eq. 2.69 can be examined for fixed T and varying ω. It corresponds
the familiar function found in any book on Fourier transforms
¡ ¢
sin ω T2
4T (ω) = (2.76)
πω
The function is shown in Fig. 2.8. It is strongly peaked around ω = 0 and it has zero’s at
ω = ± 2π
T etcetera. The peak’s width at half height is approximately given by


4ω ≈ (2.77)
T
which gives a measure of the resolution in the ω domain.
Chapter 3

The Quantum Pinball Game

“Degene in balbezit doet iets met de bal”, J. Cruijff over storingstheorie. (“The one in possession
of the ball does something with the ball”, the Dutch philosopher J. Cruijff on perturbation theory.)

Many experiments in which one aims at obtaining quantum information about a sys-
tem consists of scattering a beam of microscopic particles of a target and collecting the
fragments afterwards. This not only holds for high energy physics with its large accelera-
tors and storage rings, but also the spectroscopic experiments of “low energy” molecular
and condensed matter physics. Mattuck calls this kind of experiment the quantum pinball
game. In the first section I will explain in more detail what is meant by that. Several
chapters of these lecture notes are concerned with describing quantum pinball games. We
will use time dependent perturbation theory for that purpose. In view of the complicated
pinball machines presented to us by nature, the first order perturbation term (Fermi’s
golden rule) will often be insufficient to describe what is going on. We need techniques
that allow us to handle the full perturbation series. An introduction to such techniques is
given in the second section, which comprises the main part of this chapter. Mattuck gives
a discussion in the first chapters of his book, so I will discuss the connection especially
with his chapter 3. The techniques involve Green functions, a summary of which is given
in the appendix.

3.1 A Typical Experiment


A general template for an experiment is given in Fig. 3.1. A well-prepared beam of micro-
scopic particles is send at a target, and the outgoing particles are collected and analyzed in
order to obtain information about the target. An example of such an experiment from con-
densed matter physics is electron scattering (or EELS; electron energy loss spectroscopy).
The incoming quantum particles are then electrons, usually prepared such that they have
p2
a well-defined momentum p = ~k and energy E = 2m . The outgoing particles are also
p02
electrons of which we measure their momenta p = ~k and energies E 0 = 2m
0 0 . The target
usually is a crystal, which is a quantum many-electron system. Our job is to find out
what happens inside the target after it has been hit by an incoming particle. This is an
example of a quantum pinball game.
We can play such a game with any kind of particle. For instance, the optical experiment
of the previous chapter can be described in terms of: photons in (momentum p = ~k and

43
44 CHAPTER 3. THE QUANTUM PINBALL GAME

incoming outgoing
particles many particle particles
target

Figure 3.1: Typical quantum experiment.

energy E = ~ω = pc), and photons out. In the two-level system discussed in the previous
chapter, the outgoing photons have the same energy E = ~ω, but the ones resulting from
spontaneous emission can have their momentum in a different direction p0 = ~k0 , p0 = p. In
a multi-level system the outgoing photons can also have a different energy. Other example
are: photons in, electrons out, which is called photoemission, and electrons in, photons
out, called inverse photoemission. In molecular and condensed matter physics we operate
mostly with electrons and photons, but other particles like positrons or neutrons are also
used. In high energy physics a similar set of particles is used, but at a higher energy (what
a surprise). The goal of any such experiment is to study the properties of the target by
analyzing the probability of energy and momentum transferred to it; ∆E = E − E 0 and
4p = p − p0 . The collection probabilities of ∆E, 4p transfer as a function of energy and
momentum of incoming/outgoing particles is called a spectrum. It gives information on
the excited states of the target, since hitting a target with an incoming particle usually
excites it. A spectrum obviously also depends upon the kind of particles used; one has
EELS, optical, photoemission spectra, etcetera, each of which can give information on
different excited states of the target.

Most remarkable is that such excitations can be described in terms of the creation of
one or more quantum particles. Depending the species of incoming particles, their energy,
and of course the target, one can create particles such as phonons, plasmons, excitons,
polarons, magnons, but also electrons and holes. The ending -on denotes a quantum
particle. The word hole existed before this nomenclature set foot; a more consistent name
would actually be holon. In high energy physics there are many more examples of such
quantum particles, e.g. positrons, which are holons in free space.1 A whole cornucopia
of particles or -ons exist. For example, in the electron scattering experiment discussed
above, a (quasi-)electron can be created in the target with momentum 4p = ~K, where
K is the Bloch52 wave vector (or crystal momentum) and energy 4E = ²(K), where ²(K)
is a band energy.

With some imagination the schematic experiment of Fig. 3.1 can directly be trans-
formed into Feynman diagrams like that of Fig. 2.2. The dots Vb (τ i ) then describe what
happens inside the target box, such as, for instance, collisions between an incoming electron
and the ions inside a crystal. This imagination is useful as we will apply time dependent
1
Since the author does not know much about high energy physics, his examples of quantum particles
mainly come from condensed matter physics.
3.2. TIME EVOLUTION; SUMMING THE PERTURBATION SERIES 45

perturbation theory in order to describe what happens in the target. If more than one
species of quantum particle is involved, the perturbation expansion and the diagrams be-
come more complicated. For instance, absorption of a single photon, which we previously
derived from the right most diagram in Fig. 2.4, must in a fully quantum mechanical
description be represented by a diagram as shown in Fig. 3.2. The picture represents a

t0 t
τ
εi εf

Figure 3.2: Feynman diagram of the absorption of a photon.

“collision” of a target of energy ²i at time τ with a photon of energy ~ω, in which the
photon gets absorbed and the target emerges with energy ²f . The conservation of energy,
as expressed by the δ (ω f i − ω) part of Fermi’s golden rule, is given by ~ω = ²f − ²i and
the arrows in the diagram thus also indicate the flow of energy. The diagram is completely
equivalent to the mathematical expression for the matrix element hf (t)|U b (1) (t, t0 ) |i(t0 )i
as discussed in the previous chapter. Considering the typical experiment of Fig. 3.1 again,
we can either focus on what happens to the target system, which is what we did in the
previous chapter, or we can focus on the incoming particles. Which point of view we take
does not matter, because we need to consider all events such as shown in Fig. 3.2 anyway.
Whether we follow the solid lines, which describe how the target system evolves, or the
wiggly line, which describes what happens to the incoming photon, we always have to
include the event in which the two collide. So we need both lines.2 The focus on incoming
particles is called the quantum pinball game by Mattuck in his introductory chapters.

In modern quantum physics all events are re-interpreted in terms of interactions be-
tween quantum particles. In the rest of this course we will investigate the properties of
some of these quantum particles and their interactions in more detail.

With real life targets an incoming quantum particle can be scattered by up to O(1023 )
scattering centers (electrons, nuclei) in the target. Fermi’s golden rule will often be insuffi-
cient to describe such a complicated process. We need techniques that can handle the full
perturbation series and not only the first order term. The next section gives a systematic
mathematical approach.

3.2 Time Evolution; Summing the Perturbation Series


By applying a series of well-defined, but rather formal mathematical steps, we shall see
that it is possible to handle the time dependent perturbation series more easily. As a
2
A 2001 political analoque : the system approach is equivalent to socialism (PvdA), the incoming
particle approach is equivalent to individualism (VVD). Let the two interact in one cabinet and you can’t
tell one from the other.
46 CHAPTER 3. THE QUANTUM PINBALL GAME

warning in advance it is fair to say that the formalism not always leads to a simplification
of practical calculations. However, the method will guide our path in the labyrinth of
many particle physics in the following chapters of these lecture notes. I will use a slightly
more general operator formalism than Mattuck, which has the advantage that you can
apply it to any kind of basis set (representation). The spirit of this section however is the
same as Mattuck’s chapter 3. First I will demonstrate the necessary mathematical steps
and then show the connection to Mattuck’s chapter 3.
The central idea to make the time dependent perturbation series easier to handle is to
perform the following mathematical steps.
1. Adapt the integration bounds in the time integrals of eq. 2.22 using Green functions.
2. Fourier transform the time integrals to the frequency domain.
3. Sum the perturbation series in the frequency domain.
4. Find a closed expression for the Green function associated with the time evolution
operator.
More on Green functions can be found in the appendix.
We recapitulate the expressions found for the perturbation expansion of the time evo-
lution operator, cf. 2.21 and 2.22


X
U b0 (t − t0 ) +
b (t − t0 ) = U b (n) (t − t0 )
U (3.1)
n=1
where
µ ¶n Z t Z τn Z τ2
b (n) (t − t0 ) = 1
U dτ n dτ n−1 ......... dτ 1
i~ t0 t0 t0
b0 (t − τ n ) Vb U
U b0 (τ n − τ n−1 ) Vb ........Vb U
b0 (τ 1 − t0 ) (3.2)
We are now restricting ourselves to the case of an time independent perturbation Vb and
since
b0 (t − t0 ) = e− ~i (t−t0 )Hb 0
U (3.3)
where H b 0 is also time independent, all the terms in the time evolution operator only
depend upon the time difference t − t0 and not on the individual times t and t0 . Without
loss of generality, we can choose t0 = 0. We wish to study the situation in which Vb is not
necessarily very small, so we expect to be needing a lot of terms in the expansion of eq. 3.1
(maybe even an infinite number). Calculating integrals such as eq. 3.2, which contain a lot
of oscillating terms such as eq. 3.3 is a nuisance. Those of you who are experienced Fourier
transformers (and you engineers per definition are) know what to do; Fourier transform
to the frequency domain. Eq. 3.2 is in fact a multiple convolution integral; the part in τ 1
for example is
Z τ2
b0 (τ 2 − τ 1 ) Vb U
dτ 1 U b0 (τ 1 ) (3.4)
0
By Fourier transforming a convolution integral in the time domain we will get a product
function in the frequency domain, which simplifies matters considerably. There is however
one catch; Fourier integrals usually run from −∞ to +∞.
3.2. TIME EVOLUTION; SUMMING THE PERTURBATION SERIES 47

3.2.1 Adapt Integration Bounds; Green Functions


b+ (t)
Luckily it is not a catch-22. Here is the way out; define Green function operators G
b +
and G0 (t)

b+ (t) ≡ Θ (t) U
i~G b (t) (3.5)
b+ (t) ≡ Θ (t) U
i~G b0 (t)
0

where Θ (t) is the so-called “theta function” or “step function” which mathematicians also
call the “Heaviside function”. It is a very peculiar function given by

Θ (t) = 0 t ≤ 0
= 1 t>0 (3.6)

It takes care of what physicists call causality. If we decide to start our system at a time
t = 0, then obviously we cannot have evolution before t = 0. Or put in a different way, if
the cause starts at t = 0, then the effect must be at times t > 0 (this is the definition of
causality). Causality is expressed by the integration limits of eq. 3.2, where it prescribes
the time order 0 ≤ τ 1 ≤ τ 2 ≤ ... ≤ τ n ≤ t. Using the theta function, causality can be
incorporated in a Green function operator. With the help of Green function operators we
can rewrite eq. 3.4, using eq. 3.5, as
Z ∞
b+ (τ 2 − τ 1 ) Vb G
dτ 1 G b+ (τ 1 ) (3.7)
0 0
−∞

b and
and actually we can rewrite the whole series of time integrals of 3.2 in terms of G’s
Gb0 ’s with the integrals running from −∞ to +∞ . Note that the theta functions now take
care of the correct time order, cf. eq. 3.6. The factor “i~” in the Green function operator
is just a convention which is used by most authors. The + on G b+ is a “plus” and not a
“dagger”; it expresses the fact that this Green function operator differs from zero only for
positive times.

mathematical intermezzo on Green functions


“Verde que the quiero verde. Verde viento. Verde ramas”, Federico Garcia Lorca (Green how I
love you green. Green wind. Green branches).

The operators G b+ and Gb+ are examples of so-called Green function operators.3 Re-
0
member that the operator equation for the time evolution operator, which is equivalent
to the Schrödinger equation, is

∂ b bUb (t)
i~ U (t) = H (3.8)
∂t
3
I have been told that the correct terminology is Green function and not Green’s function like you find
in some textbooks. It is like Fourier transform and Bessel function (and not Fourier’s transform or Bessel’s
function). Mind you I was educated in Nijmegen, so this could be a catholic convention. In any case,
Green functions are named after the English mathematician George Green (1st half 19th century).
48 CHAPTER 3. THE QUANTUM PINBALL GAME

b (0) = I. From this we can derive an equation for G


with the initial condition U b+ (t) (see
also Mattuck § 3.4)
∂ b+ i~ ∂ b (t) + Θ (t) ∂ U
b (t) = δ (t) U b (t)
i~ G (t) = Θ (t) U
∂t i~ ∂t ∂t
b (0) + 1 Θ (t) H
= δ (t) U bU b (t) = δ (t) + HbGb+ (t)
i~
or
µ ¶
∂ b b+ (t) = δ (t)
i~ − H G (3.9)
∂t

In deriving this I have used ∂t Θ (t) = δ (t) (see if you can derive this property yourself)
and δ (t) f (t) = δ (t) f (0) (think of an integral with this function).

b be an
A general definition of a Green function operator can be given as follows. Let L
b
operator; then the Green function operator G(t) belonging to this operator is defined by
b G(t)
L b = δ (t) (3.10)

George Green in the 19th century did not know about quantum mechanics of course. He
was interested in partial differential equations. We can try and make a connection with his
spirit. For example, take one particle in one dimension and write eq. 3.9 in the position
representation as
¿ ¯µ ¶ ¯ À
¯ ∂ ¯ ­ ®
x ¯¯ i~ − H b G b+ (t)¯ x0 = x|x0 δ (t) ⇔
∂t ¯
Z ¿ ¯ ¯ ÀD ¯ ¯ E
¯ ∂ ¯ 00 ¯ b + ¯ 0 ­ ®
dx00 x ¯¯i~ − H b ¯ x00 x ¯ G (t)¯x = x|x0 δ (t)
∂t ¯

use the following properties


(i) 0 0
­hx|x ® − x ) 00
¯ i ∂=¯ δ 00(x
(ii) ¯ ¯ ∂
Dx ¯i~ ¯∂t xE = hx|x n i i~ ∂t ; now use (i) o
¯ b ¯ 00 ~2 ∂ 2
(iii) x ¯H ¯ x = hx|x i − 2m ∂x002 + V (x00 ) ; as proven in Section 1.3.2, eq. 1.22
00
D ¯ ¯ E
¯ b+ ¯ 0
(iv) x ¯G (t)¯ x ≡ G+ (x, x0 , t)

Using all of this to simplify the equation gives


· ¸
∂ ~2 ∂ 2 +
¡ 0
¢ ¡ 0
¢
i~ + − V (x) G x, x , t = δ x − x δ (t) (3.11)
∂t 2m ∂x2

The function G+ (x, x0 , t) is an example of what a standard mathematician in the field of


differential equations would call a Green function. In the formal language of differential
equations, we can state the following. Let L (− →
x ) be a differential operator in a set of


variables x = (x1 , x2 , x3 , ......) . Then the Green function is defined as the solution of the
equation4
¡→ − ¢ ¡→ − ¢
L (− →x )G −x,→ x0 =δ − x −→ x0 (3.12)
4
δ (−

x −−

x 0 ) is a short hand notation for δ (x1 − x01 ) × δ (x2 − x02 ) × δ (x3 − x03 ) × ....
3.2. TIME EVOLUTION; SUMMING THE PERTURBATION SERIES 49

You probably have encountered and/or will encounter Green functions anywhere where
partial differential equations occur and in physics these are not uncommon as you know; the
Poisson equation in electrostatics, the wave equation in electrodynamics, the Schrödinger
equation in wave mechanics, just to mention a few. Some additional information on Green
functions can be found in the appendix. A couple of final remarks on Green functions.

• The Green function does not exist. It is similar to the phrase “Fourier transform”;
one should always mention the function of which you want to have the Fourier
transform. Similarly, if you hear the phrase Green function, one should always
mention the operator for which you want the Green function. The operator L b (or


L ( x ) in a representation) is defined first and then the Green function is determined
by eq. 3.10 (or eq. 3.12)

• In addition to an operator one should have initial or boundary conditions in order


to fix a Green function operator. For instance, the Green function operator G b+ (t)
defined by eqs. 3.5 and 3.9 has Gb (t) = 0 for t ≤ 0. This is an initial condition in
+

partial differential equation language. In principle, all sorts of initial and/or bound-
ary conditions can be worked into a Green function and each condition leads to a
slightly different Green function. Labels like the “+” on G b+ (t) refer to the condi-
tions used to construct it. By these labels the Green functions acquire (politically
incorrect) names like “advanced” or “retarded” Green functions.

3.2.2 Fourier Transform to the Frequency Domain


In terms of the Green function operators defined by eq. 3.5, we can write eqs. 3.1 and 3.2
(setting t0 = 0 without loss of generality)

X
b+ (t) = G
G b+ (t) + b(n) (t)
G (3.13)
0
n=1

where
Z ∞ Z ∞ Z ∞
b(n) (t) =
G dτ n ......
dτ 1
−∞ −∞ −∞
b+ (t − τ n ) Vb G
G b+ (τ n − τ n−1 ) Vb ...........Vb G
b+ (τ 1 ) (3.14)
0 0 0

We now perform our Fourier transforms. From the convolution theorem we know that
convolutions in the time domain become simple products in the frequency domain.5 For
instance, Fourier transforming eq. 3.4 gives
Z ∞
b
F (t) = b+ (τ 2 − τ 1 ) Vb G
dτ 1 G b+ (τ 1 ) F T Fb (ω) = G
b+ (ω) Vb G+ (ω) (3.15)
0 0 −→ 0 0
−∞

Using the same trick repeatedly in eq. 3.14 gives

b(n) (ω) = G+ (ω) Vb G+ (ω) Vb .....Vb G+ (ω)


G (3.16)
0 0 0

5
If you have forgotten the convolution theorem, you might want to read Mattuck’s chapter 2 to find a
proof.
50 CHAPTER 3. THE QUANTUM PINBALL GAME

and we obtain the equivalent of eq. 3.13 in the frequency domain

G b+ (ω) + G
b+ (ω) = G b+ (ω) Vb G
b+ (ω) + G
b+ (ω) Vb G
b+ (ω) Vb G
b+ (ω) + .... (3.17)
0 0 0 0 0 0

This is just our perturbation series for the time evolution operator of eq. 3.1 transformed
to the frequency domain (using the definition of eq. 3.5)

3.2.3 Sum the Perturbation Series; Dyson Equation


We are now in the position of summing the perturbation series (at least formally). Eq.
3.17 can be written as
" ∞ ³
#
X ´n
b+ (ω) = G
G b+ (ω) I + Vb G
b+ (ω) (3.18)
0 0
n=1
P
The term [...] looks like a simple geometric series like ∞ n
n=0 r = (1 − r)
−1 which allows

us to write
³ ´−1
b+ (ω) = G
G b+ (ω) I − Vb Gb+ (ω) (3.19)
0 0

b and Vb ’s are operators, so we must keep then in the right order. The −1
All the G’s
superscript means “inversion” now; in the following we will show that this is correct.
Meanwhile, an equivalent expression is
³ ´−1
G b+ (ω) Vb
b+ (ω) = I − G b+ (ω)
G (3.20)
0 0

Eq. 3.19 can also be derived formally in a better way by noting that eq. 3.18 is equivalent
to
b+ (ω) = G
G b+ (ω) + G
b+ (ω) Vb G
b+ (ω) (3.21)
0 0

Recursion of the right hand side then gives the series of eq. 3.18. Rewriting this as
b+ (ω) − G
G b+ (ω) Vb G
b+ (ω) = G
b+ (ω)
0 0
³ ´
⇔ I−G b (ω) Vb G
+ b (ω) = G
+ b (ω)
+
(3.22)
0 0

then leads to eq. 3.20.

Eq. 3.21 is an important equation called the Dyson equation.6 The Dyson equation is
the equivalent of the whole perturbation series of eqs. 3.1 and 3.2 in the frequency domain.
Going back to eq. 3.17 and expanding it as a series is called the Dyson expansion. Eq.
3.19 or 3.20 give a formal solution of the Dyson equation which includes all orders n of
the perturbation Vb . We can, if we wish to, now Fourier transform the final result G b+ (ω)
b + b
back to the time domain and obtain G (t), eq. 3.14, or U (t), eq. 3.5. Since knowing the
time evolution operator Ub (t) is equivalent to solving the Schrödinger equation, the Dyson
6
after the english mathematician and theoretical physicist Freeman Dyson (mid 20th century), who
worked with Feynman and Schwinger. In the 2nd world war he also worked as a theoretician for the
“bomber command” of the RAF.
3.2. TIME EVOLUTION; SUMMING THE PERTURBATION SERIES 51

equation is thus equivalent to the Schrödinger equation. Moreover, we have just discussed
a procedure for solving the Schrödinger equation; well, at least formally. However, whether
this formal procedure is useful in practical calculations depends strongly upon the physical
problem at hand. For single particles or so-called independent particle system it can be
applied successfully. In a system that consists of many interacting particles, we have severe
problems, as we will see, but still the general technique is very useful.

3.2.4 Green Functions; Closed Expressions


b+ . According to eqs. 3.3 and 3.5 we
Up till now we have not yet used our knowledge of G0
have
b0 (t) = Θ (t) e− ~ t.H0 i b
i~G (3.23)

Its Fourier transform is


Z ∞
b0 (ω) = − i
G
i b
dt eiωt Θ (t) e− ~ t.H0 (3.24)
~ −∞

As explained by Mattuck on p. 40 the integrand is not well-behaved, since it keeps on


oscillating for t → ∞, and does not go to zero. Thus formally its Fourier transform is not
very well defined (which is a physicists understatement; a mathematician would say: does
not exist). We apply a typical physicists type-of-reasoning to get rid of this nuisance. In
eq. 3.23 we are not really interested in “infinite” times. A very large time, compared to
the time in which we do our measurements, is infinite enough for all practical purposes.
Therefore we modify the theta function of eq. 3.6

Θδ (t) = 0 t ≤ 0
= e−δ.t t>0 (3.25)

where δ > 0 is such a small number that for 0 < t < tphysical (a practical measuring time)
e−δt is indistinguishable from unity for any practical measuring accuracy, i.e. e−δt = 1
for 0 < t < tphysical . However since δ is not zero, e−δt ultimately goes to zero for t large
enough. The idea is shown in Fig. 3.3. Mathematicians don’t like such tricks, but

Θ δ (t )
1

t 0 t physical ∞

Figure 3.3: A ‘physical’ theta function.

as physicists we are interested in practical end-results, not in infinite hair-splitting. Just


like in the case of Fermi’s golden rule, where we did not need a true δ-function, but only
a very sharply peaked function, here we don’t need a true theta function, but only one
52 CHAPTER 3. THE QUANTUM PINBALL GAME

which behaves like a theta function over a practical time scale. In practical calculations,
we can always choose the parameter δ as small as we like (but not zero). In any case, the
integral of eq. 3.24 becomes well-behaved
Z ³ ´−1
i ∞ i b
b +
G0 (ω) = − dt e ~ (~ω−H0 +iδ)t = ~ω − H
b 0 + iδ (3.26)
~ 0
b =H
Since also the full Hamiltonian H b 0 + Vb is time independent, we can write G
b+ (t) =
i b
b+ (ω) we find
e− ~ tH and similar to G0
³ ´−1
b+ (ω) = ~ω − H
G b + iδ (3.27)
³ ´−1
b+ (E) =
Writing E = ~ω and using units in which ~ = 1, we have G b + iδ
E−H .
Thus we can write
³ ´ ³ ´
b b + b +
lim E − H G (E) = lim G (E) E − H = I b (3.28)
δ→0 δ→0
³ ´
So loosely speaking Gb+ (E) is the “inverse” of the operator E − H
b . By mathematicians
b+ (E) is also called the resolvent.7
G

We now use a basis set |ni; n = 1, 2..., which consists of eigenstates of the Hamiltonian
b
H|ni = ²n |ni. Applying resolutions of identity to eq. 3.27 it follows
X X
b+ (E) =
G |nihn|Gb+ (E) |mihm| = δ n,m |ni (E − ²m + iδ)−1 hm| (3.29)
n,m n,m

which gives
X |nihn|
b+ (E) =
G (3.30)
n
E − ²n + iδ

This way of writing down a Green function operator on the basis of eigenstates of the
Hamiltonian is very popular. Using the notation of the mathematical intermezzo and
defining φn (x) = hx|ni for wave functions as usual, we have in “wave function” notation
¡ ¢ D ¯¯ + ¯ E X hx|ni hn|x0 i X φ (x)φ∗ (x0 )
¯
+ 0 b
G x, x , E = x ¯G (E)¯ x0 = = n n
(3.31)
n
E − ²n + iδ n
E − ²n + iδ

which might be familiar to you from previous encounters with Green functions. Note that
b+ (E) would go to infinity, if it were not for the convenient iδ term.
at the points E = ²n , G
b
The inverse of E − H does not really exist here, that is why we need the iδ to avoid these
points.8 Fourier transforming this expression to the time domain (using complex contour
integration, see the appendix of the next chapter) gives the expression
¡ ¢ X
G+ x, x0 , t = φn (x)φ∗n (x0 ) e−i²n t Θ(t) (3.32)
n
7
Whether they really think it resolves all our problems I don’t know.
8
In the terminology of complex function theory we say that G b + (E) has a “pole”at E = ²n for δ → 0,
or a “pole” at E = ²n − iδ in the complex plane, which becomes real for δ → 0. See the appendix of the
next chapter.
3.2. TIME EVOLUTION; SUMMING THE PERTURBATION SERIES 53

From the expression of the Green function operator of eq. 3.27 we can also recover the
Dyson equation of eq. 3.21.
³ ´− 1 ³ ´−1 ³ −1 ´−1
Gb+ (E) = E − H b + iδ = E−H b 0 + iδ − Vb = G b+ (E) − Vb (3.33)
0

or, in other words


³ −1 ´
b+ (E) − Vb G
G b+ (E) = I (3.34)
0

or
³ −1 ´
G b+ (E) − Vb G
b+ (E) G b+ (E) = Gb+ (E)
0 0 0
³ ´
⇐⇒ I − Gb+ (E) Vb Gb+ (E) = Gb+ (E) (3.35)
0 0

which can be rewritten as


b+ (E) = G
G b+ (E) + G
b+ (E) Vb G
b+ (E) (3.36)
0 0

which is the Dyson equation again.

3.2.5 Summary
In case you don’t see the wood for the trees anymore, here is a short summary of the
previous sections. We have a Hamiltonian
b =H
H b 0 + Vb (3.37)

where Hb 0 describes an “unperturbed” system and Vb is a “perturbation”. We can write


the time-evolution operator of the perturbed system, described by the full Hamiltonian as
X∞ µ ¶n Z t Z τn Z τ2
b b
− ~i tH b0
− ~i tH 1
U (t) = e =e + dτ n dτ n−1 ....... dτ 1
n=1
i~ 0 0 0
i b i b i b0
e− ~ (t−τ n )H0 Vb e− ~ (τ n −τ n−1 )H0 Vb .......e− ~ τ 1H
(3.38)

This expression can be simplified in a series of steps.

1. Define a Green function operator by

b+ (t) = Θδ (t) U
i~G b (t) (3.39)
³ ´
We can prove i~ ∂t∂
−Hb Gb+ (t − t0 ) = δ (t − t0 ), which makes it a “Green function
operator” in the mathematical sense.

2. Using the definition 3.39 in 3.38 and Fourier transforming to the frequency domain
we get the series
∞ ³
X ´n
b+ (ω) = G
G b+ (ω) + G
b+ (ω) Vb G b+ (ω) + G
b+ (ω) = G b+ (ω) Vb G
b+ (ω) (3.40)
0 0 0 0 0
n=1
54 CHAPTER 3. THE QUANTUM PINBALL GAME

which is also called the Dyson equation.

3 Eq.3.40 can be “solved” formally as

³ ´−1
G b+ (ω) Vb
b+ (ω) = I − G b+ (ω)
G (3.41)
0 0

4 which can also be shown to be identical to the resolvent operator

³ ´−1
G b + iδ
b+ (ω) = ~ω − H (3.42)

Usually the expression of step 4 is only useful in formal manipulations, but not in practical
calculations. Note for instance that the expression of eq. 3.30 assumes that we know all
the eigenstates of the full Hamiltonian; if we did, we wouldn’t have any problems to start
with. Whether the expression of step 3 is useful in practical calculations depends upon
how difficult it is to calculate the inverse, i.e. (....)−1 ; for single or independent particles
it can be done. If it proves to be too difficult, we have to solve Dyson’s equation in step
2 via another route. More of this in Chapter 4.

3.3 Connection to Mattuck’s Ch. 3


Mattuck starts his pinball game simple with just a single electron. The scattering poten-
tials in his pinball machine are the ion cores or nuclei in a crystal. The properties of his
single electron pinball are summarized in his table 3.1. This table is equivalent to our
time dependent perturbation series of eqs. 2.21 and 2.22 and Figs. 2.2 and 2.3. Mattuck
gives sort of a description of the foregoing formalism in his chapter 3, which I advice you
to read after this point. I will explicitly describe the connection with the present chapter.
Mattuck’s eq. 3.1 on p.38 gives the general many particle definition of a (one-particle)
propagator; the rest of chapter 3 just deals with the specific case of a single particle.
Suppose |r1 i is an eigenstate of the position generator

b
r|r1 i = r1 |r1 i (3.43)

At t = t1 we start with a particle in this particular state |r1 (t1 )i. We let the particle evolve
(propagate) under the full Hamiltonian to t2 and ask for the probability amplitude to find
it in |r2 (t2 )i, or in other words to find it at position r2 . The probability amplitude is given
by
D ¯ ¯ E
¯b ¯
r2 (t2 ) ¯U (t2 − t1 )¯ r1 (t1 ) t2 > t1 (3.44)

This is identical to Mattuck’s definition of his “propagator” of eq. 3.1, iG+ (r2 , r1 , t2 − t1 )
Actually, we could have used our definition of G b+ (t2 − t1 ) in eq. 3.44, which for t2 > t1
b
is identical to U (t2 − t1 ). In other words Mattuck’s eq. 3.1 is just the position represen-
tation of our Green function operator. Mattuck’s eq. 3.2 on p. 39 is simply a matrix
representation of the same Green function operator using another basis set. The example
3.4. APPENDIX. GREEN FUNCTIONS; THE LIPPMANN-SCHWINGER EQUATION55

Mattuck worked in detail has H b =H b 0 = pb 2 is the Hamiltonian of a “free


b 0 + Vb , where H
2m
particle”. The eigenstates of the latter are also the eigenstates of the momentum operator
b |pi = p|pi
p

b 0 |pi = ²p |pi p2
H ²p = (3.45)
2m
In wave function notation we can write
i
hr|pi ≡ φp (r) ∝ e ~ p·r (3.46)

as we have seen in one of the exercises. Using this set of eigenstates we get, in the
“momentum representation”
D ¯ ¯ E ¡ ¢ ¡ ¢
0 ¯ b+ ¯ i
p ¯G0 (t2 − t1 )¯ p = δ p0 − p Θ (t2 − t1 ) e− ~ ²p (t2 −t1 ) ≡ δ p0 − p G+
0 (p, t2 − t1 )
(3.47)

which corresponds to Mattuck’s eq. 3.9 on p. 40. In the example of § 3.2, p. 43 Mattuck
uses a “perturbation” of the form V (bp) where
­ 0 ® ¡ ¢
p)| p = δ p0 − p V (p)
p |V (b (3.48)

in the momentum representation. The time evolution operator of eq. 3.38 can be R trans-
formed into the momentum representation by inserting a resolution of identity I = dp |pihp|
between each pair of operators. The result becomes simple, since in all matrices only the
diagonal terms survive, by virtue of eqs. 3.47 and 3.48. The final results are Mattuck’s
eqs. 3.30 and 3.31. Mattuck’s §3.3 and 3.4 describe the same mathematical steps we
have performed for our Green function, but now for this special case in the momentum
representation.

3.4 Appendix. Green Functions; the Lippmann-Schwinger


Equation
What use is a Green function ? This question has haunted many generations of
physics students. A popular story on the origin of “Green functions” states that, whenever
a a theorist mentions the word Green function, all experimentalists get dizzy and green in
the face, and have to leave the room.9 The followers of George Green, who gave his name
to these functions, were interested in solving differential equations of the form

L (−

x ) φ (−

x ) = ρ (−

x) (3.49)

where L (− →x ) is some differential operator, ρ (−



x ) is a fixed function called the source, and


φ ( x ) is the solution of the equation we are after. For instance from electrostatics we
know the Poisson equation

∇2 φ (r) = 4πρ (r) (3.50)


9
I don’t think this story is entirely true. Some of them become red in the face and very agitated,...
before they leave the room.
56 CHAPTER 3. THE QUANTUM PINBALL GAME

The source ρ (r) is the charge density, which we assume to be given, and φ (r) is the
electrostatic potential, which we want to determine. Suppose we know the solution to the
equation
¡ ¢ ¡ ¢
∇2 G r, r0 = δ r − r0 (3.51)

which defines a Green function à la George Green. The solution to the Poisson equation
can then be written as
Z
¡ ¢ ¡ ¢
φ (r) = φ0 (r) + 4π G r, r0 ρ r0 d3 r0 (3.52)

where φ0 (r) is a general solution of the homogeneous equation ∇2 φ0 (r) = 0 (the Laplace
equation). Substituting eq. 3.52 in eq. 3.50 one easily proves that this is the solution. We
have “solved” the Poisson equation in two steps. The differential operator part is taken
care of by defining a Green function as in eq. 3.51 and then the source term ρ (r) part
is incorporated by applying eq. 3.52. The differential equation part, eq. 3.51, is solved
first, subject to the relevant boundary conditions (in electrostatics boundary conditions
are usually determined by the presence of conducting or dielectric objects). Here is the
crux: once we have obtained the Green function,10 we can calculate the potential for any
charge density ρ (r) simply by applying eq. 3.52.

The Poisson equation is inhomogeneous, i.e. it can be expressed like eq. 3.49 with
a non-zero source term. The Schrödinger equation is homogeneous, i.e. it has the form
L (−

x ) φ (−

x ) = 0. Let us take the simple one-dimensional example from wave mechanics
again. The two variables are −→
x = (x, t) and the Schrödinger equation is
½ ¾
∂ ~2 ∂ 2
i~ + − V (x) φ (x, t) = 0 (3.53)
∂t 2m ∂x2
How can we apply Green magic to a homogeneous equation like the Schrödinger equation
in the same way as above ? Well, by using a trick. We rewrite the Schrödinger equation
as
½ ¾
∂ ~2 ∂ 2
i~ + φ (x, t) = V (x)φ (x, t) (3.54)
∂t 2m ∂x2

and we treat the right hand side as a source term (even though formally this is not true,
since it contains the solution we are after, rather than being a fixed function like in the
Poisson equation). The rest is on automatic pilot. First we obtain a Green function by
solving
½ ¾
∂ ~2 ∂ 2 ¡ ¢ ¡ ¢ ¡ ¢
i~ + 2
G+ 0 0 0
0 x, t; x , t = δ x − x δ t − t
0
(3.55)
∂t 2m ∂x
and then, in analogy to the Poisson case, the solution to eq. 3.54 can be written as
ZZ
¡ ¢ ¡ 0 0¢ 0 0
φ (x, t) = φ0 (x, t) + G+ 0 0 0
0 x, t; x , t V (x )φ x , t dx dt (3.56)

10
How this is done can be found in Electricity & Magnetism books, e.g. J. D. Jackson, Classical
Electrodynamics, (Wiley, New York, 1975).
3.4. APPENDIX. GREEN FUNCTIONS; THE LIPPMANN-SCHWINGER EQUATION57

where φ0 (x, t) is a general solution of the homogeneous equation


½ ¾
∂ ~2 ∂ 2
i~ + φ0 (x, t) = 0
∂t 2m ∂x2
Actually this part is easy. The solution of the homogeneous equation is a simple plane
wave.
r
1 i(kx−ωt) (~k)2
φ0,k (x, t) = e with ~ω =
2π 2m
The Green function G+ 0 0
0 (x, t, x , t ) can then be calculated from eq. 3.32 noting that the
continuous index k labels the eigenstates.11
Z
+ 0 0
¡ ¢ i (~ k)
2
0
G0 (x, t, x , t ) = dk φ0,k (x, 0) φ0,k x0 , 0 e− ~ 2m (t−t ) Θ(t − t0 )
µ ¶1 Ã !
−mi 2 m |x − x0 |2
= exp − Θ(t − t0 ) (3.57)
2π~(t − t0 ) 2i~(t − t0 )

The Green function incorporates causality, i.e. t > t0 . By direct substitution in eq. 3.55
one can see that this is a solution. There are a couple of snakes in the grass. First and
foremost, eq. 3.56 is not a solution to the problem, since it contains the unknown function
φ (x, t) on the right-hand side as well as on the left-hand side. Instead, it is an integral
equation from which φ (x, t) has to be solved. This is different from an inhomogeneous
equation like the Poisson equation where eq. 3.52 directly gives you the solution. In the
way we have derived eq. 3.56 you can easily see that it must be completely equivalent to
the Schrödinger equation.

The integral equation of eq. 3.56 is called the Lippmann 08 -Schwinger 65 equation. Is
it easier to solve than the Schrödinger (differential) equation? That depends upon the
physical problem at hand. The Lippmann-Schwinger equation is used a lot in “scatter-
ing theory” in which the scattering of quantum particles (by fixed targets and/or other
particles) is studied. There is also a Lippmann-Schwinger equation in electromagnetism
where propagation of electromagnetic waves is studied. It is also used in studying the
propagation of acoustical and seismological waves, so it is a versatile technique.

What does this have to do with the Green functions we encountered in the main text?
First of all, there is a good reason for the notation G+ 0 for the Green function in eq. 3.56.
One just has to follow the reasoning which lead to eq. 3.11 backwards to see that it can
be written as
¡ ¢ D ¯¯ + ¡ ¢¯ E D ¯ ¯ E
G+ x, t; x0 0
, t = x ¯ b t − t0 ¯¯ x0 = 1 x ¯¯e− ~i (t−t0 )Hb 0 ¯¯ x0 Θ(t − t0 )
G (3.58)
0 0
i}
We now recognize eq. 3.56 as the position representation of the general form of the
Lippmann-Schwinger equation
Z
¡ ¢
|φ(t)i = |φ0 (t)i + G b+ t − t0 Vb |φ(t0 )i dt0 (3.59)
0

11
More on analytical expressions of Green functions for a number of (simple) cases can be found in
mathematical physics books, e.g. P. R. Wallace, Mathematical Analysis of Physical Problems, (Dover, New
York, 1984).
58 CHAPTER 3. THE QUANTUM PINBALL GAME
D ¯ ¯ E
¯ ¯
(remember hx|φ(t)i = φ (x, t) and x ¯Vb ¯ x0 = V (x)δ(x − x0 ) ). Now let us write |φ(t)i =
b (t − t0 ) |φ(t0 )i = 1 G
U b+ (t − t0 ) |φ(t0 )i for the formal time evolution of the state with
i}
respect to an arbitrary starting point t0 as we did before. Eq. 3.59 becomes, dropping the
1
i} at the right and left hand sides
½Z ¾
¡ ¢ ¡ ¢ 0
b + b +
G (t − t0 ) |φ(t0 )i = G0 (t − t0 ) |φ0 (t0 )i + b + 0 b b+ 0
G0 t − t V G t − t0 dt |φ(t0 )i
(3.60)

Now there is one last trick. Consider the typical scattering experiment of Fig. 3.1. At the
starting time t0 , just when the incoming particles start to leave their source, they are far
from the target. At that point in time they do not yet experience the perturbing potential
V which originates from the target. In other words, they still behave like free particles,
which means at time t0 we have |φ(t0 )i = |φ0 (t0 )i. But then eq. 3.60, which must be
obeyed for any starting condition |φ0 (t0 )i, becomes an equation for the operator
Z
¡ ¢ ¡ ¢
b + b +
G (t − t0 ) = G0 (t − t0 ) + G b+ t − t0 Vb G
b+ t0 − t0 dt0 (3.61)
0

If one Fourier transforms this equation to the frequency domain, one immediately rec-
ognizes the Dyson equation, eq. 3.21. This means that one can tackle the Lippmann-
Schwinger equation of eq. 3.56 using the Green function techniques (Dyson equation,
summation of perturbation series, etcetera) discussed in this chapter.

In the Born approximation, eq. 2.23, one only includes the perturbation in first order.
This corresponds to using G b+ ’s only on the right hand side of eq. 3.61. Working our way
0
backwards to eqs. 3.59 and 3.56 one easily sees that the Born approximation corresponds
to using φ0 (x, t)’s only on the right hand side of the Lippmann-Schwinger equation.
ZZ
¡ ¢ ¡ 0 0¢ 0 0
φ (x, t) = φ0 (x, t) + G+ 0 0 0
0 x, t; x , t V (x )φ0 x , t dx dt

In the Born approximation the equation becomes of the same type as eq. 3.52, i.e. the
unknown φ (x, t) only occurs on the left hand side, and can thus be calculated straightfor-
wardly. Because of its simplicity, the Born approximation is used quite a lot in scattering
calculations, or more generally, in studying the scattering and propagation of all sorts of
waves. It is even used in situations where the perturbation is not really small enough in
order to justify using it only in first order. But, as often in practical physics, one stretches
the limit.

In summary: what use are Green functions ?

• They provide a general approach for solving an inhomogeneous partial differential


equation, see eqs. 3.49—3.52.

• For a homogeneous partial differential equation they provide an alternative: an


integral (Lippmann-Schwinger) equation. This can be approached using the Green
function techniques (Dyson equation, summation of perturbation series) discussed
in this chapter.
3.4. APPENDIX. GREEN FUNCTIONS; THE LIPPMANN-SCHWINGER EQUATION59

3.4.1 The Huygens Principle Revisited


Once more, a patriotic subsection just for fun. In view of eqs. 2.11 and 3.5, wave propa-
gation in one dimension can be written as
Z
ψ(x, t) = i~ dx0 G+ (x, t, x0 , t0 )ψ(x0 , t0 ) (3.62)

We interpreted this equation before in terms of the Huygens principle for wave propagation,
see eq. 2.11. The Green function has to obtained by solving eq. 3.11. For simple cases
this can be done. For instance, if the potential is constant, V (x) = V0 , the Green function
becomes that of eq. 3.57 multiplied by a phase factor
i 0
G+ (x, t, x0 , t0 ) = G+ 0 0 − ~ V0 (t−t )
0 (x, t, x , t ) e (3.63)

∂ 22
In three dimensions we have to make the substitutions x → r, x0 → r0 , and ∂x 2 → ∇
in eqs. 3.11 and 3.62. The Green function is similar to the one above if we make the
1 3
substitution x → r, x0 → r0 in eq. 3.57 and substitute (...) 2 by (...) 2 . It is seen that the
functional form looks like
¯ ¯
G+ (r, t, r0 , t0 ) = FV0 (¯r − r0 ¯ , t − t0 ) (3.64)

This form represents a spherical wave, i.e. a wave with spherical wave fronts, which
originates from a center r0 at t0 . Usually a spherical wave is used for the secondary waves
in the Huygens construction for wave propagation, cf. Fig. 2.1. If the potential is not
constant, we can still use the Huygens construction, but in an infinitesimal step-by-step
way. The idea is shown in Fig. 3.4.

G (r, t , r ', t ')

ψ (r ', t ')
ψ (r, t )

V (r ) ≈ V0 V (r ) ≈ V1

Figure 3.4: Propagation of a wave using the Huygens principle.

We consider only points r which are very close to the “source” points r0 , i.e. |r − r0 | <
∆r small, such that we may approximate the potential by a constant V0 in this region.
Then we may use the spherical waves of eq. 3.64 and the three-dimensional variant of eq.
3.62 to propagate the wave function in this region as shown in Fig. 2.1. Now we calculate
at the outer edge of this region the exact potential, which usually gives a slightly different
60 CHAPTER 3. THE QUANTUM PINBALL GAME

value V1 . We approximate in an adjacent region the potential by this new constant V1 .


We still can use spherical waves to propagate the wave in this new region, but we have
to recalculate it using eq. 3.63 with V1 instead of V0 . We can now propagate our wave
through the next small step ∆r, ∆t in space and time. Using this procedure iteratively
we can propagate our wave throughout the whole space and time. Fig. 3.4 is again a
little bit misleading is the sense that a spherical wave should emerge from each point r0
in space, and the wave function ψ(r, t) should be constructed from the interference of all
such waves.

The beauty of it all is that Huygens got his idea in 1678, before anything was known
about Green functions, or indeed anything was known about wave equations. Huygens
only used his intuition. Personally I feel that “Green functions” should be called “Huygens
functions”. Almost three centuries later, in 1948, Richard Feynman was again inspired
by Huygens’s idea, which he implemented in his famous path integral method for solving
wave equations.12

12
R. P. Feynman, Space-Time Approach to Non-Relativistic Quantum Mechanics, Rev. Mod. Phys. 20,
367 (1948).
Chapter 4

Scattering

Scattering theory is an very broad subject. It is used to describe scattering phenomena in


a wide range of fields in physics, from the collisions of elementary particles to the motion
of seismic waves through the earth’s interior. Naturally our discussion will mainly be
within the context of quantum mechanics, and more specifically within the context of
the quantum pinball game or the typical experiment of Fig. 3.1. Even here we have to
restrict ourselves severely. We will only consider the scattering of a single particle by a
collection of fixed scattering centers which we are far apart, a so-called “dilute sample”.
The classical example is scattering of α-particles through a thin gold foil, which was used
by Rutherford08 to demonstrate the existence of a massive, but small nucleus inside the
atoms. Other examples are the motion of a quantum particle (e.g. an electron) through
an atomic gas; or the motion of an electron through an ideal metal containing impurity
ions in a dilute concentration. These are examples of a quantum pinball game as explained
in the previous chapter. The outgoing particle is the same as the incoming particle and
the “pinball machine” contains no other particles, only fixed scattering centers. Even this
problem we will at not consider in its full detail. In principle the outgoing particle can
be in a different quantum state than the incoming particle. We will only consider elastic
scattering, i.e. the energy of the incoming and outgoing particle is the same. Furthermore
we assume that the scattering potential cannot change the spin state of the incoming
particle, so we do not have to consider spin explicitly. There are many cases for which these
assumptions are very reasonable. Consider scattering of electrons by atoms, for instance.
Atoms are much heavier than electrons, so in an elastic collision the energy transfer from
electrons to atoms is indeed very small. In an inelastic scattering the electron excites the
atom; such processes do happen, but we won’t consider them here.1 Furthermore, for
not too heavy atoms (all but the bottom rows of the periodic table) the spin state of an
electron is indeed unchanged in a collision. In summary, we will be looking at the case
in which only the direction of the momentum p0 of the outgoing particle can differ from
p2
that of the incoming particle p, whereas the size (which is given by the energy ² = 2m )
0
remains the same, i.e. |p| = |p |.
First I will give a general introduction to the scattering of waves, which, although we
will use electron waves as an example, is not very specific to quantum mechanics persé.
This is mainly to show that the most important measurable quantities can all be derived
from the so-called scattering amplitudes. The size and phase of these amplitudes are char-
1
The general formalism discussed here can be extended in a straight-forward way to include inelastic
processes. The notation becomes a bit messy however, which is why I skipped it.

61
62 CHAPTER 4. SCATTERING

acteristic for the specific incoming particle / target combination; the physics of scattering
is universal for all types of waves. The rest of this chapter discusses the quantum me-
chanical calculation of the scattering amplitudes. At first we will look at the transmitted
component, where the outgoing particle has the same momentum as the incoming parti-
cle, i.e. p = p0 . The main idea of that part is to introduce important concepts, such as
self-energy and quasi-particle, which will also be useful to us later on in the many-particle
world. In the final sections I will make contact with the more traditional quantum scatter-
ing theory, and focus upon the total and differential cross sections. Appendix I contains
a short discussion on the index of refraction, merely to show that for a dilute sample this
also can be obtained straight-forwardly from the scattering amplitudes. In this chapter
some use is made of the results of complex function theory, a short summary of which is
given in Appendix II.

4.1 Scattering by a Dilute Concentration of Centers


We consider a single particle being scattered by a collection of scattering centers that are
far apart. To be specific, think of an electron scattered by an atomic gas.2 We assume
that our single particle comes in as a free particle. Its quantum state is a plane wave
2 |k |2
ei(k1 ·r−ωt) with a well-defined momentum p1 =~k1 and a well-defined energy ~ω = ~ 2m 1
.
Each center (i.e. an atom) can scatter this wave. Each scattering event will lead to
scattered waves as in Fig. 4.1, which are sent in all directions from the ion or atom. In
a quantum mechanical interpretation, the scattered wave determines the probability that
the particle is scattered and the transmitted wave gives the probability that the particle
is not scattered.

scattered particle k2

k1 k1

incoming particle target transmitted particle

Figure 4.1: Scattering of a single particle by a fixed target.

Like waves on a water surface after you have thrown a stone on it, the amplitude of
the scattered waves becomes smaller the further you are away from the center (the stone).
Intuitively one expects the wave that is represented by Fig. 4.1 to have the form
" #
i(k1 ·r−ωt) ei(kr−ωt)
φ(r,t) = A0 e + fk2 k1 (4.1)
rÀλ r
2
Many example from optics, condensed matter physics or high energy physics would do, but this one
seems to be the easiest conceptually.
4.1. SCATTERING BY A DILUTE CONCENTRATION OF CENTERS 63

at a large distance from the target (distance measured in units of wave length). The first
term on the right hand side describes an incoming plane wave. The “intensity” of the
incoming wave is3
¯ ¯2
¯ ¯
I in = ¯A0 ei(k1 ·r−ωt) ¯ = |A0 |2 (4.2)

The second term on the right hand side of eq. 4.1 describes the outgoing scattered wave. In
an isotropic medium far from the target, scattered waves always have spherical wave fronts,
irrespective of the “shape” of the target. Note that we may write k = |k1 | = |k2 |; in elastic
scattering the wave number does not change. The amplitude A of the outgoing spherical
wave goes as A(r) ∝ 1/r, where r is the distance to the center. The total intensity (or in
quantum mechanical terms the total probability) has to be distributed over the surface of
a sphere of radius r, and since the total must be constant, the local intensity (probability,
which is A2 ) goes as 1/r2 . The amplitude of the outgoing spherical wave in general
depends upon the direction k2 with respect to that of the incoming wave k1 . This angular
dependence is represented by the scattering factors fk2 k1 in eq. 4.1. The fk2 k1 are called
the scattering amplitudes. In general these will be complex numbers, so they have a phase
as well as an amplitude. They contains all the target specific information.

Once we know how a single microscopic target scatters the incoming particles, we can
try and figure out how a real macroscopic sample used in an experiment behaves. A
macroscopic sample consists of many microscopic targets (atoms) and spherical scattered
waves emerge from all these targets. How the sample as a whole scatters depends very
much upon the conditions of the experiment. Suppose we can divide our system into cubic
boxes of size L (volume Ω = L3 ) such that each box contains one target and the probability
of finding two targets in one box is negligible. If nI is the density of the gas, then simply
Ω = 1/nI . If we are not near a thermodynamic critical point, density fluctuations are
small, so the probability of finding more than (or less than) one target per Ω is indeed
very small. The basic idea is shown in Fig. 4.2.
In particle scattering the conditions are usually quite different from what you might
know from diffraction. We have stated that our source produces incoming particles of
well-defined energy and momentum such that its quantum state can be represented by a
plane wave ei(k1 ·r−ωt) . This is of course an idealization; a real source usually produces an
incoherent beam of particles. In wave terms, an incoherent beam consists of a series of
pulses, so-called wave packets. Each wave packet has a size ` that is sufficiently long, so
that the packet is nearly monochromatic, i.e. nearly a wave with a well-defined frequency
(energy) and wave length (momentum). ` is called the coherence length, divided by the
wave’s speed v = p/m = ~k1 /m it gives the coherence time τ c = `/v. Obviously one must
have ` À λ for a nearly monochromatic wave. There is however no phase relation between
3
£ ¤
A “proper intensity” should £have the¤ ¯ dimension
¯ W/m2 , whereas the “intensity” given here has the
dimension of a “particle density” 1/m 3 ¯ 2¯ 3
£¯ ¯¤ £ ( φ(r,t)
¤ d r gives the probability of finding¯ a ¯particle in a volume
3 ¯
d r, so the dimension of φ(r,t) 2¯
= 1/m¯ ).¯ Multiplying the particle density ¯φ2 ¯ by the speed v of
3

the particles gives the “flux” of particles v ¯φ2 ¯, i.e. the number of particles crossing a unit ¯surface ¯ per
unit time. Multiplying this £with the ¤energy ~ω per particle gives the “proper intensity” ~ωv ¯φ2 ¯, which
has the required dimension J/(sm2 ) . Since v and ω are constants here, and we normalize with respect
to the incoming beam anyway, the concepts “particle density”, “flux”, “proper intensity” can all be used
to represent “intensity”. Here we use “particle density” because it is the easiest. In quantum scattering
theory one frequently uses “flux”; in scattering of electromagnetic or acoustic waves one normally uses the
“proper intensity”.
64 CHAPTER 4. SCATTERING

scattered particles
incoming particles
k2

ψ1
ψ2
λ

k1

L
A

Figure 4.2: Incoherent scattering of wave packets in a dilute sample.

subsequent wave packets in the incoherent beam.4 In our sample the targets are separated
by an average distance L. If L & `, which is typical for a dilute sample or gas, the scattered
waves emitted from different targets at ant point in time have no phase relation.5 If we
put our detector at a distance D À L far from the sample, which is the usual setup, we
measure a spherical wave like that of eq. 4.1 coming from each target. Since the relative
phases of these waves are unrelated, we may simply sum their intensities to get the total
intensity. For example, the total intensity produced by a wave ψ 1 (r, t) + ψ 2 (r, t) is
I(r, t) ∝ |ψ 1 (r, t) + ψ 2 (r, t)|2 = |ψ 1 (r, t)|2 + |ψ 2 (r, t)|2 + 2 Re(ψ ∗1 (r, t)ψ 2 (r, t)) (4.3)
If we average this expression over a time larger than the coherence time, then the last
term averages to zero, because the relative phase of ψ 1 (r, t) and ψ 2 (r, t) is random for
incoherent waves. So
Z t+∆t
I(r, t) = I(r, τ )dτ ∝ |ψ 1 (r, t)|2 + |ψ 2 (r, t)|2 = I1 + I2 (4.4)
t−∆t

for ∆t À τ c . In summary, for incoherent scattering we can obtain the scattering intensity
of the sample for each direction k2 by summing over the scattering intensities of all the
targets in the direction k2 .
4
Think of the light produced by e.g. a mercury discharge lamp. Each wave packet is produced by
spontaneous emission related to a specific transition of a mercury atom. The optical emission frequency
is around 1015 Hz, so the cycle time T ∼ 10−15 s. The duration of the wave packet is typically in the
order of τ c ∼ 10−9 s. We have `/λ = τ c /T ∼ 106 , which gives a nearly monochromatic wave packet with a
well-defined frequency (energy) and wave length (momentum). Emissions of different mercury atoms are
not coupled; they emit at independent times. The result is a wave train that consists of wave packets 10−9
s long, the relative phase of which is unrelated; i.e. an incoherent light wave. An electron gun, used as
a source for electrons, operates by (thermal) emission of electrons from a metal filament. It produces an
incoherent beam of electrons using similar arguments.
5
From P V = nRT one obtains that the average distance L between atoms in an ideal gas at T = 300
2 k2
K and P = 1 atm is 34 Å. From ² = ~2m one obtains that the wave length λ of 1 keV electrons is 0.4
Å. So the distance between the scattering targets L ∼ 102 λ. One could lower the pressure to create even
better “incoherence” conditions, P = 10−3 atm gives L ∼ 103 λ.
4.1. SCATTERING BY A DILUTE CONCENTRATION OF CENTERS 65

Now consider the situation where the average distance between the targets L ¿ `, the
coherence length of the incoming beam. We call this coherent scattering. The relative
phase of ψ 1 (r, t) and ψ 2 (r, t) in eq. 4.3 is then fixed. The last term on the right hand
side averaged over time does not give zero and results in an interference pattern (such
as in Young’s double slit). Whether this interference pattern can be observed depends
upon the conditions. In a random sample, i.e. no correlation between the positions of
the microscopic targets, it is usually not possible to observe an interference pattern. The
intensity fluctuations resulting from interference are very closely spaced and the surface
or aperture of an ordinary detector samples over a large number of such fluctations. This
spatial averaging has the same effect as the temporal averaging in eq. 4.4, i.e. it washes
out the interference term. So even for coherent scattering the total intensity can usually
be obtained simply by summing over the scattering intensities of the individual targets.
Only under very special circumstances interference can be observed, such as when we have
only a very small number of targets, or the targets are placed on a regular lattice. The
latter gives diffraction if λ is comparable to the lattice spacing. So diffraction requires
coherent scattering.
So far we have only discussed single scattering events. If we have multiple scattering,
i.e. a wave scattered from one target is scattered again by the next target, things become
much more complicated as each target receives such scattered waves from its surrounding.
For a “small” sample the sum of the intensities of these scattered (secondary) waves is
small as compared to the intensity of the (primary) incoming beam. We may neglect
such waves being scattered a second time, since this contributes very little to the total
scattered intensity. Dilute samples such as gases actually need not be that small. Even
for macroscopic samples the neglect of multiple scattering usually works very well.6

4.1.1 The Scattering Cross Section


According to eqs. 4.1—4.4, the total scattered intensity coming from the sample in the
direction k2 , relative to that of the incoming beam k1 , at a distance D from the sample
is given by
¯ ¯
I k2 k1 ¯A0 D−1 ei(kD−ωt) fk k ¯ |fk2 k1 |2
2 1
=N =N
I in |A0 |2 D2

where N is the number of targets in the sample; since they are all the same, their scattering
amplitudes fk2 k1 are all identical. Imagine that we surround the sample by a sphere of
radius D to collect all the scattered waves. The geometry is shown in Fig. 4.3.
Each surface element D2 df = D2 sin θdθdϕ receives a relative “power”

I k2 k1 2
D df = N |fk2 k1 |2 df
I in
Per solid angle df and per target a scattering “power” can thus be defined as
dσ k2 k1
= |fk2 k1 |2 (4.5)
df
6
For dense samples in which the individual microscopic targets scatter the incoming particles very
strongly, multiple scattering becomes important. An example where this is the case is electron scattering
(or diffraction) by crystals.
66 CHAPTER 4. SCATTERING

D k2

D2d =
ϕ
D 2 sin θ dθ dϕ
θ z

target k1

Figure 4.3: Scattering geometry.

In quantum scattering theory “intensity” is usually expressed as the number of particles


crossing a unit surface per unit time, it is also called “flux”; “power” is then the number
of particles per unit time.


k2 k1
The quantity df is called the differential (scattering) cross section. The differential
cross section of a single scattering center is defined as: “the number of particles that is
scattered per unit of time into a solid angle df, divided by the flux of incoming particles”.
Obviously it is a function of the relative orientation of k2 and k1 , or the angles θ and ϕ,
see Fig. 4.3. Summing over all possible k2 corresponds to integrating over all angles, i.e.
over the complete sphere. One obtains the total (scattering) cross section σ k1 . The total
cross section, or simply: the cross section of a single scattering center is defined as: “the
total number of particles that is scattered per unit of time, divided by the flux of incoming
particles”. It is given by
X Z
2 dσ k2 k1
σ k1 = |fk2 k1 | df = df (4.6)
df
k2 6=k1

One has to exclude the k1 direction, because that contains the incoming beam, but in an
integral this carries no weight of course. σ k1 has the dimension of surface, i.e. [σ k1 ] =
s−1 /(s−1 m−2 ) = m2 . Conceptually it is the “size” or the “area” which the target presents
to the incoming wave. Obviously it is a function of the state k1 of the incoming wave. If
the target is spherically symmetric, it is a function of |k1 | only, or put in another way, of
2
the wave length λ = |k2π1 | or frequency ω = ~|k1| 7
2m of the incoming wave. Note that there is
not much particular quantum character in these definitions. In fact, similar definitions can
7
It also depends very much upon the kind of incoming wave. In case this is an electron wave with energy
1 keV (wave length 0.4 Å), the cross section of a Carbon atom is of order 1 Å2 . For X-rays of the same
wave length, its cross section is only 10−8 Å2 . Electrons thus see condensed matter, where the interatomic
distances are of order 1 Å, as one “solid target”. They can hardly penetrate and scatter mainly from the
surface region. For X-rays even condensed matter is largely empty space and they can penetrate easily.
4.1. SCATTERING BY A DILUTE CONCENTRATION OF CENTERS 67

be given for scattering of electromagnetic waves by obstacles (radar waves, for instance),
or acoustic waves (including seismic waves).

4.1.2 Forward Scattering; the Optical Theorem


Since all the scatterers are independent, consider just one of them, i.e. one box in Fig.
4.2. We denote the direction k1 of the incoming wave as the z-axis, cf. Fig. 4.3. Suppose
we look at the wave in the forward direction, i.e. in the k1 - direction, or at small angles
from that direction. The wave of eq. 4.1 is given by
" #
ei(kr−ωt)
φ(r,t) = A0 ei(kz−ωt) + fk1 k1
rÀλ r
" #
eik(r−z)
= A0 1 + fk1 k1 ei(kz−ωt)
r
" √ 2 2 2 #
eik( x +y +z −z)
= A0 1 + fk1 k1 ei(kz−ωt) (4.7)
z

Small angles implies (x2 + y2 ) ¿ z 2 ; thus we may approximate


p p
x2 + y2 + z 2 = z 1 + z −2 (x2 + y 2 )
· ¸
1 −2 2 1
≈ z 1 + z (x + y ) = z + z −1 (x2 + y 2 )
2
2 2
Inserting this in eq. 4.7 gives for the wave in the forward direction
" k 2 2
#
ei 2z (x +y )
φ(r,t) = A0 1 + fk1 k1 ei(kz−ωt) (4.8)
rÀλ z

and the intensity is given by


½ h k 2 2 i¾
2
|φ(r,t)|2 = |A0 |2 1 + Re ei 2z (x +y ) fk1 k1 (4.9)
rÀλ z

We are going to integrate the intensity of eq. 4.9 over a small surface of dimension a2
which represents a detector of this dimension placed in the forward direction. The detector
signal, which is the number of particles detected per unit time, per unit of incoming flux,
therefore is
Z Z · Z Z ¸
1 2 2 2 k
i 2z (x2 +y2 )
s= dxdy |φ(r,t)| = a + Re fk1 k1 dxdy e (4.10)
I in a a z a a

Since I in = |A0 |2 according to eq. 4.2. We choose the surface a2 such, that the phase
factor in the integrand of the right hand side at the edge of the surface
k 2 ka2
(x + y 2 ) ∼ À 2π (4.11)
2z 2z
whereas, because of our original small angle assumption

(x2 + y 2 ) ∼ a2 ¿ z 2
68 CHAPTER 4. SCATTERING

These two requirements lead to


r
4πz √
= 2λz ¿ a ¿ z (4.12)
k

Because λ ∼ 10−10 m and z ∼ 1 m, a ∼ 10−2 m will do just fine. Because of eq. 4.11,
the integrals of the right hand side of eq. 4.10 can be extended from −∞ to ∞ without
appreciably changing their value, since the integrand is a rapidly varying function in the
outer region. We get
Z ∞ Z ∞ k 2 +y 2 ) 2πiz
dxdy ei 2z (x =
−∞ −∞ k

So eq. 4.10 becomes


s = a2 + Im fk1 k1
k

The term a2 represents the particles that arrive from the incoming beam into the detector.
The second term describes the attenuation of this beam due to the scattering process. Since
there is conservation of the number of particles, this attenuation must be equal to the total
number of particles which are scattered in all other but the forward direction. But the
latter is just the definition of the total cross section, according to eq. 4.6. In other words,
the total cross section is related to the imaginary part of the forward scattering amplitude
by


σ k1 = Im fk1 k1 (4.13)
k

This important relation is called the optical theorem.

Again there is nothing very quantum mechanical about this derivation. It holds for
scattering of all sorts of waves. The relation of eq. 4.13 was first derived in the 19th
century for (classical) scattering of electromagnetic radiation, hence its name.8 Eqs. 4.6
and 4.13 present a sum rule for the scattering amplitudes fk2 k1 .
Z

Im fk1 k1 = |fk2 k1 |2 df (4.14)
k

Quantum mechanically, this expresses the conservation of the number of particles in the
scattering process (or the conservation of energy for that matter, since each particle rep-
resents an energy ~ω). What doesn’t end up in the forward beam must be scattered in the
other directions. There is a lot more interesting stuff that can be said about scattering of
waves. An example is found in Appendix I.
8
It is attributed to Lord Rayleigh, although the books don’t seem to be certain on this. The optical
theorem seems to have been rediscovered a couple of times in various subfields of physics. In quantum
scattering theory it is also called the Bohr 75 -Peierls-Placzek relation.
4.2. SCATTERING BY A SINGLE CENTER 69

4.2 Scattering by a Single Center


Now we know that the relevant physical quantities can be derived from the scattering
amplitudes of a single target fk2 k1 , we set about calculating these quantities quantum
mechanically. Scattering of particles by a spherical potential is usually treated in intro-
ductory textbooks on quantum mechanics using the so-called partial wave analysis. Since
this is rather specific for spherical symmetry and relies strongly on the properties of special
functions, I will skip it here. More generally, scattering of single particles is usually treated
via the Lippmann-Schwinger equation, eq. 3.59. I will come back to this in the last part
of this chapter. Here I will start from another approach, the so-called “propagator”, in
which we will use the Green function techniques of the previous chapter. We will focus on
the time evolution of a single incoming particle and we will obtain from it the total cross
section of eq. 4.13.
The particle entering into the box is a free particle. Its quantum states are plane
waves characterized by a well-defined momentum; we will label these states by k = p~ .
These states form a continuous basis set. Continuous basis sets require the use of delicate
mathematics, whereas discrete basis sets are much easier to work with. In order to avoid
the mathematical acrobatics needed for continuous basis sets, we use a little trick borrowed
from solid state physics. If we apply periodic boundary conditions on the cubic box of
size L, then only a discrete set of ki , where ki = (i1 , i2 , i3 ) × 2π
L ; i1,2,3 = 0, 1, ... is allowed.
Periodic boundary conditions (called Born-von Karman conditions in solid state physics)
may seem a bit artificial, but they are easy to work with and provided the box is large
enough, they have a negligible effect on the end results. The ki form a cubic lattice in
reciprocal space, where the size of the lattice cell is

2π 3 (2π)3
∆3 k = ( ) = (4.15)
L Ω

and Ω = L3 is the volume of the box. We can now work with a basis set of normalized
plane waves.

1
φki (r) ≡ hr|ki i = √ eiki ·r (4.16)

One can easily appreciate that this basis set is orthonormal and complete

hki |kj i = δ ki kj
X
I= |ki ihki | (4.17)
ki

For instance in wave function form one has


Z D E Z
3 1
hki |kj i = ki |rihr|kj d r = ei(kj −ki )·r d3 r =δ ki kj
Ω Ω Ω
X X Z D E
0 0 0
hr| |ki ihki |ψi = hr|ki i ki |r ihr |ψ d3 r
ki ki Ω
µZ ¶
1 X iki ·r −iki ·r 0 3 0
= e e ψ(r ) d r = ψ(r) = hr|ψi
Ω Ω
ki
70 CHAPTER 4. SCATTERING

The first line is proven by straight-forward integration. In the last line one recognizes
the expansion of a function as a Fourier series. The idea now is that we do all our
intermediate calculations in the discrete basis set |ki i. Ω is large enough such that ∆3 k
is small, whichP means Rthat at any time we wish, we may replace a sum over ki by an
integral, i.e. ∆3 k → d3 k.

The single scattering center is placed in the origin of the box and it is represented
by a potential V (r). It will act as a “perturbation” on the free particle with incoming
momentum p1 = ~k1 . There is a probability that the particle is scattered in some other
direction k2 and a probability that the particle goes through unscattered in k1 . Later, far
from the scattering center, the scattered (or unscattered) particle won’t feel the influence of
the potential anymore and it is again a free particle in state |k2 i (or |k1 i). We assume the
scattering to be elastic. In other words the scattering center is so massive as compared to
the incoming particle that it can be considered as fixed. It does not take any momentum
from the particle and, since it has no internal degrees of freedom, it does not absorb
any energy. In other words, the scattering event can only change the direction of the
momentum of the incoming particle, i.e. |k1 | = |k2 |. Since a single proton is already
∼ 2 × 103 times heavier that an electron, this is a good approximation for the scattering
of electrons by atoms.
Analogous to eq. 2.24, the probability of scattering of the incoming particle is
b (t2 − t1 ) |k1 i|2
Wk1 →k2 = |hk2 |U (4.18)

and the probability that the particle goes through unscattered is


b (t2 − t1 ) |k1 i|2
Wk1 →k1 = |hk1 |U (4.19)

where t1 is the time at which we send in the particle and t2 is the time at which we measure
the scattered (or unscattered) particle. Obviously t2 > t1 . In view of the previous chapter
we know that it is easier to switch to the frequency domain.
Z
b i~ ∞ b+ (ω) |k1 i
hk2 |U (t2 − t1 ) |k1 i = dω e−iω(t2 −t1 ) hk2 |G (4.20)
2π −∞

b (t2 − t1 ) |k1 i is also called the propagator . It determines the transition


The quantity hk2 |U
amplitudes, and thus how the particle “propagates” through the system.

We will focus on the “unscattered” particle and on the relevant matrix element
b+ (ω) |k1 i ≡ G+ (k1 , k1 , ω)
hk1 |G

This determines the probability that the particle goes through unscattered. One might
think of using eq. 3.27 or 3.30, since these give exact expressions for the Green function
operator. However, since we do not know the eigenstates of the full Hamiltonian, which
includes the scattering potential V (r), these expressions are not of much practical use
in this case. The same thing is more or less true for eq. 3.19. The reason is that the
“inversion” implied by (....)−1 is damn hard to do analytically for most potentials.9 The
9
This should not surprise us too much. Even the classical motion of a particle in a central potential
can only be obtained analytically for a limited number of potentials V (r). Newton was lucky with his 1r .
4.2. SCATTERING BY A SINGLE CENTER 71

expression can however be used in numerical calculations, where it leads to quite an elegant
procedure. Since that is a topic which belongs more in a course in computational physics,
we will follow a different route here. The starting point is the series expression
P of eq. 3.17,
i.e. the Dyson expansion, where we insert resolutions of identity I = ki |ki ihki | between
each pair of operators.
X
b+ (ω) |k1 i = hk1 |G
hk1 |G b+ (ω) |k1 i + b+ (ω) |ki ihki |Vb |kj ihkj |G
hk1 |G b+ (ω) |k1 i
0 0 0
ki ,kj
X
+ b+ (ω) |ki ihki |Vb |kj ihkj |G
hk1 |G b+ (ω) |kl ihkl |Vb |km ihkm |G
b+ (ω) |k1 i + ....
0 0 0
ki ,kj ,kl ,km
(4.21)

Since the states |ki are eigenstates of H b 0 , we can simplify this expression using eq. 3.26
³ ´−1 1
hkj |G b 0 + iδ
b+ (ω) |kl i = hkj | ~ω − H |kl i = hkj |kl i = G+
0 0 (kl , ω) δ kj kl
~ω − ²kl + iδ
(4.22)

Let us also simplify our notation defining kl ≡ k; ki ≡ i, etcetera. A consistent use of eq.
4.22 now transforms eq. 4.21 into

G+ (k, k, ω) = G+ + +
0 (k, ω) + G0 (k, ω) Vk,k G0 (k, ω)
X
+ G+ + +
0 (k, ω) Vk,l G0 (l, ω) Vl,k G0 (k, ω) + .... (4.23)
l

where Vk,l ≡ hk|Vb |li.

We can represent this equation by Feynman diagrams similar to Figs. 2.2 and 2.4.
The result is shown in Fig. 4.4.

k k = k k + k k +
Vkk
l
k k + ........
Vkl Vlk

Figure 4.4: Perturbation series for single particle scattering.

The lines in the diagram are now labeled by the state in which the particle travels. The
particle has to start in state k and it has to stop in state k in the particular situation which
we are interested in. In the first diagram on the right hand side the particle goes through
without ever encountering the potential (“free” propagation). In the second diagram the
particle is scattered once from its initial state k to its final state k. In the third diagram the
particle is scattered to an intermediate state l in which it propagates until it is scattered
back to state k. You can easily construct all the higher order diagrams yourself. Since the
particle starts and stops in the same state k we call it an unscattered particle, despite all
the intermediate interactions with the potential ! Perhaps it is better to use the phrase
forward scattered or transmitted particle. It all means the same thing.
72 CHAPTER 4. SCATTERING

4.3 Re-summation of the Series; the Self-Energy


I will now explain a very useful procedure which will (ultimately) lead to a simple physical
picture of the “unscattered” probability. The same procedure will also be used in more
complicated situations in many-particle cases. The central idea is to define a quantity
Σ (k, ω) such that the diagram expansion of Fig. 4.4 can be represented as in Fig. 4.5

k k = k k + k Σ k +

k
k Σ Σ k + ........

Figure 4.5: Perturbation series rewritten in terms of self energy.

In mathematical equation form this becomes

G+ (k, k, ω) = G+ + +
0 (k, ω) + G0 (k, ω) Σ (k, ω) G0 (k, ω)
+ G+ + +
0 (k, ω) Σ (k, ω) G0 (k, ω) Σ (k, ω) G0 (k, ω) + .... (4.24)

Comparing with eq. 4.23 and pondering a little shows that the expression for Σ (k, ω)
must be given by
X X
Σ (k, ω) = Vk,k + Vk,l G+
0 (l, ω) Vl,k + Vk,l G+ +
0 (l, ω) Vl,m G0 (m, ω) Vm,k + ....
l6=k l6=k,m6=k
(4.25)

In diagram form this becomes Fig. 4.6

l≠k l≠k m≠k


Σ = + + + ........
Vkk Vkl Vlk Vkl Vlm Vmk

Figure 4.6: the self energy for single particle scattering

In other words Σ (k, ω) consists of all diagrams (or terms in the mathematical series)
where none of the intermediate lines has the label k of the incoming particle. Σ (k, ω) is
called the self-energy of the particle in state k. This meaning of this term will be clarified
below. It is a useful exercise to try and see that all the diagrams in the expansion of Fig.
4.4 can be constructed by selecting diagrams of Fig. 4.6 and connecting them by lines
labeled with k. Also none of the diagrams in Fig. 4.4 are obtained more than once in this
way. So the expansion of Fig. 4.5 is indeed equivalent to that of Fig. 4.4. Alternatively,
you may prove algebraically that eqs. 4.24 and 4.25 are equivalent to eq. 4.23.

It is now extremely simple to sum the series of eq. 4.24


( ∞
)
X £ ¤n G+
+ + + 0 (k, ω)
G (k, k, ω) = G0 (k, ω) 1 + Σ (k, ω) G0 (k, ω) = (4.26)
n=1
1 − Σ (k, ω) G+0 (k, ω)
4.3. RE-SUMMATION OF THE SERIES; THE SELF-ENERGY 73

All the terms in this expression are numbers (not operators), so we do not have to use the
(....)−1 notation. This expression can be simplified even further by multiplying numerator
and denominator with 1/G+ 0 (k, ω)

1 1
G+ (k, k, ω) = = (4.27)
1/G+
0 (k, ω) − Σ (k, ω) ~ω − ²k − Σ (k, ω) + iδ

using eq. 4.22. Now the matrix element of the “full” Green function G+ (k, k, ω) has the
same form as that of the “free particle” Green function G+ 0 (k, ω), but the “free particle”
energy ²k has been replaced by ²k + Σ (k, ω). The term Σ (k, ω) acts as a sort of energy
correction on the particle in state k, which is caused by intermediate excursions of the
particle to other states, cf. Fig. 4.6. It is called the self-energy of the particle in state
k.10 However, it is a somewhat strange sort of energy, as we will examine now. For most
potentials it far from straight-forward to calculate the self energy Σ (k, ω), as you might
imagine in view of eq. 4.25 or Fig. 4.6; one has to use numerical techniques. In order to
avoid such complicated calculations, for our present purpose it is sufficient to take only
the first two terms on the right hand side of eq. 4.25. We assume for the moment that
we may neglect the higher order terms in V (the typical argument used in perturbation
theory). Now is the time to convert sums into integrals. For the second term on the right
hand side of eq. 4.25 we have
X Z
1 X 3 Ω
... = 3 ...∆ l ←→ ...d3 l (4.28)
∆ l (2π)3
l6=k l6=k

using eq. 4.15(the l 6= k point in carries no weight in the integral, so we can neglect this
constraint in the integral). The matrix element Vk,l deserves some extra attention. Using
eq. 4.16 one derives
Z
1 1
Vk,l = hk|Vb |li = e−ik·r V (r)eil·r d3 r = V (k − l) where
Ω Ω Ω
Z
V (k) = e−ik·r V (r) d3 r is the Fourier transform of V (r) (4.29)

Whether we take the integral over the full space or over the volume Ω does not matter.
Since we assumed the potential V (r) to be well localized within Ω, its Fourier transform
is always well defined. In this approximation the self energy becomes
Z
1 1
Σ (k, ω) ≈ V (0) + d3 l V (k − l)G+
0 (l, ω) V (l − k)
Ω Ω(2π)3
Z 2
1 1 3 |V (k − l)|
= V (0) + d l (4.30)
Ω Ω(2π)3 ~ω − ²l + iδ

We have used Vk,k = Ω1 V (k − k) = Ω1 V (0). What remains is taking the limδ→0 as discussed
in Section 3.2.4. In order to do this in a neat and controlled way, we write the denominator
10
The term probably comes from quantum electrodynamics where the event in which an electron changes
its state from k to l (at a vertex Vkl in Fig. 4.6) is accompanied by emission of a photon. Changing its
state back from l to k ( at a vertex Vlk in Fig. 4.6) then involves reabsorbing this photon. This constant
emission and recapturing of photons influences the energy of the electron. Because it is considered to be
the electron’s own doing, it is called the “self-energy” of the electron. Another term sometimes used for
the same object is the “optical potential”.
74 CHAPTER 4. SCATTERING

of the integrand as

1 (~ω − ²l ) iδ
= 2 − (4.31)
~ω − ²l + iδ 2
(~ω − ²l ) + δ (~ω − ²l )2 + δ 2

The first term on the right hand side of this expression contributes to the real part of the
self energy Σ (k, ω) and the second term contributes to its imaginary part. The real part
gives the so-called “principal value” or Pv of the integral.11
½ Z 2¾
1 1 3 |V (k − l)|
Re Σ (k, ω) = V (0) + Pv d l (4.32)
Ω (2π)3 ~ω − ²l

More about Pv ’s can be found in the appendix on complex analysis. Also there you can
find there the proof of
1 η
lim = δ(x) (4.33)
η→0+ π x + η2
2

i.e. a Lorentzian line-shape function in the limit of vanishing width becomes a δ-function.
Using this property for the second term on the right hand side of eq. 4.31 then gives for
the imaginary part of the self-energy
Z
1
Im Σ (k, ω) = −π d3 l |V (k − l)|2 δ(~ω − ²l ) (4.34)
Ω(2π)3

The integrals in eqs. 4.32 and 4.34 can usually be done, and thus one finds a closed
expression for the self energy Σ (k, ω). We see now why it is a peculiar object. It has
the dimension of energy and it depends upon the state k and upon the frequency ω. The
strange thing is that it has an imaginary part as well as a real part. This might seem
weird for an “energy”, but it has clear physical consequences, which are discussed in the
next subsection.

4.4 The Physical Meaning of Self-Energy


Let us go back to our starting point. Starting with the particle in state k at time t1 , the
probability amplitude that the particle is still in state k at time t2 is given by
Z ∞
b (t2 − t1 ) |ki = i~ e−iω(t2 −t1 )
hk|U dω (4.35)
2π −∞ ~ω − ²k − Σ (k, ω)

according to eqs. 4.20 and 4.27. The integral can be done by complex contour integration
(a short summary of complex function analysis is given in the appendix).12 What is needed
are the “poles” of the integrand, i.e. the roots of the denominator, or the solutions ω of
the equation

~ω − ²k − Σ (k, ω) = 0 (4.36)
11
The principal value of the integral is calculated by using the first term on the right handside of eq.
4.31 in the integral. Then one does the integration, and finally one takes limδ→0
12
We have taken limδ→0 already. This is o.k. here; since Im Σ (k, ω) < 0 the integral can be done without
needing the δ-trick.
4.4. THE PHYSICAL MEANING OF SELF-ENERGY 75

Read the appendix if this does not make any sense to you. Because of the ω dependence
of Σ (k, ω) this is a nontrivial equation to solve, but we could solve it iteratively, using
knowledge of our approximated Σ (k, ω), cf. eqs. 4.32 and 4.34. Because we have used
only the first few terms of a perturbation series to derive those equations, we have already
assumed that the perturbation V is small. But this then must mean that Σ (k, ω) cannot be
too large. We can start solving eq. 4.36 by setting Σ (k, ω) = 0 in zeroth order, ¡ which ¢gives
(0) (1)
~ω = ²k . In the next iteration we then solve the equation ~ω − ²k − Σ k, ω (0) = 0,
which gives ~ω(1) = ²k +Σ (k, ²k /~). Note that ω (1) is a complex number, since ¡ Σ (k, ²k¢/~)
(n)
is complex. In principle we can pursue these iterations solving ~ω − ²k − Σ k, ω (n−1) =
0; n = 2, 3... which converges to some point ω(∞) . In numerical analysis one calls this
mathematical algorithm “fixed point iteration”, since one converges onto a “fixed” point.
Having found the pole, the next step is to find the “residue”. The details of the derivation
are a bit messy, but they are explained at great length in the appendix, Subsection 4.8.4.
We assume that |Im Σ (k, ω)| is small, so we can use eq. 4.97. The result is
Z ∞
e−iω(t2 −t1 ) 2π i i~
dω = zk e− ~ [²k +∆k − 2 Γk ](t2 −t1 ) for t2 > t1
−∞ ~ω − ²k − Σ (k, ω) i~
= 0 for t2 < t1 (4.37)

where the second line follows from the fact that Im Σ (k, ω) < 0 (eq. 4.34 and the ap-
pendix). The factors in this equation are given by (see eq. 4.97)
µ ¶−1
1 ∂ Re Σ (k, ω)
zk = 1− |ω=²k /~ =1
~ ∂ω
∆k = Re Σ (k, ²k /~)
2zk
Γk = − Im Σ (k, ²k /~) (4.38)
~

Using these results in eq. 4.35 we finally get the answer we were looking for

b (t2 − t1 ) |ki = zk e− ~i [²k +∆k ](t2 −t1 ) e− 12 Γk (t2 −t1 )


hk|U for t2 > t1 (4.39)

Comparing this with the time evolution of a free particle without scattering potential
present
i
b0 (t2 − t1 ) |ki = e− ~ ²k (t2 −t1 )
hk|U

we see that three remarkable things happened by the introduction of the scattering po-
tential V

1. The energy level of the particle has shifted from ²k to ²k + ∆k . Note that the
energy shift ∆k within our approximation is exactly what you would have found
by applying up to 2nd order time-independent perturbation theory on the state |ki.
This is most clearly seen be rewriting this equation in the discrete sum representation
again (using eqs. 4.28 and 4.29).
X |Vkl |2
∆k = Re Σ (k, ²k /~) = Vkk + (4.40)
²k − ²l
l6=k
76 CHAPTER 4. SCATTERING

2. The probability amplitude is decaying exponentially by a characteristic lifetime given


by τ k = Γ1k . The probability that a particle stays in state k is given by
b (t2 − t1 ) |ki|2 ∝ e−Γk (t2 −t1 )
Wk→k = |hk|U

Thus the characteristic decay time of the exponential decay is given by τ k = Γ−1 k .
If the decay time is long, the particle still resembles very much a free particle and
the standard term for such a particle is quasi-particle. The decay rate Γk contains
a factor Ω−1 , cf. eqs. 4.32 and 4.34. Referring to our introduction in Section 4.1,
we observe that
Ω−1 = nI
the concentration of gas atoms. It makes clean physical sense that the decay rate
is proportional to the concentration of scattering centers the incoming particle en-
counters.
3. The probability amplitude is multiplied by a factor zk . In this particular case we can
easily show that zk = 1. This is because hk|U b (t1 − t1 ) |ki = hk|I||ki = 1, according
to eq. 4.17 and eq. 2.7 in Chapter 2. So here we have
∂ Re Σ (k, ²k /~)
=0 (4.41)
∂ω
This is true for a single particle, but is no longer true in the many particle case.
Anticipating later chapters, one can prove that in the general many particle case one
still finds the quasi-particle form of eq. 4.39, but now 0 < zk ≤ 1. So zk acts as a
weight factor for the quasi-particle probability amplitude. In the single particle case
this weight factor is always one.

Connection with Fermi’s Golden Rule


Now have a look back at Fermi’s golden rule, eq. 2.32. The perturbation V is frequency
independent in the scattering considered here, so we can set ω = 0 in eq. 2.32; moreover
ω lk = ²k −²
~
l
and we can use the relation δ(ax) = a1 δ(x). So from Fermi’s golden rule we
deduce that the transition rate from state k to a different state l is
2π 2π
wk→l = |Vk,l |2 δ(²k − ²l ) = |V (k − l)|2 δ(²k − ²l ) (4.42)
~ ~Ω2
using eq. 4.29. From eqs. 4.34 and 4.38 we now have established
Z X
Ω 3
Γk = d l wk→l = wk→l (4.43)
(2π)3
l6=k

when returning to the discrete sum representation again (eq. 4.28 ) This expresses the fact
that the decay rate of the probability of staying in state k is equal the sum of transition
rates to all other states l. Thus the sum of the probabilities of the particle being in state
k or in one of the other states l is one. So eq. 4.43 is the quantum mechanical expression
of “conservation of particles”. Conservation of particles must of course always be valid.
So even if one has to include higher order perturbation terms to calculate the transition
rates wk→l (and not just the first order term that lead to Fermi’s golden rule), then still
eq. 4.43 is valid.
4.5. THE SCATTERING CROSS SECTION 77

4.5 The Scattering Cross Section


Setting up a scattering experiment, one usually does work with only a single particle.
Instead, one tries to prepare a stable beam of incoming particles, all with the same mo-
2
mentum p = ~k and energy ²k = (~k) 2m , i.e. a “monochromatic” beam. Experimentally,
it is the particle flux Φin of the beam one aims to stabilize at a constant value. The
flux is defined as Φin = N v, where N is the density of particles in the beam and v= ~k m
is their speed. The total number of scattered particles per second is equal to the sum
of the transition rates to all possible states, which according to eq. 4.43 is equal to Γk .
According to its definition in Section 4.1.1, the total cross section is therefore given by

σ k = Γk /Φin (4.44)

Often one omits the subscript k on σ k since the dependence is clear from the context.
In the calculations of the previous three sections we have assumed a density of incoming
particles of N = Ω−1 , since we have chosen to normalize our incoming plane wave such
that we have one particle per box of volume Ω, see eq. 4.16. Therefore in our case the
flux is Φin = Ωv . Using eq. 4.38, and realizing that Ω−1 = nI , since we also have only one
scattering center per box. The scattering cross section σ = Γk /Φin then becomes

2 Im Σ (k, ²k /~)
σ=− (4.45)
nI ~v

Comparing this to eq. 4.13 in Section 4.1.1 we must have13


m
fkk = − Σ (k, ²k /~) (4.46)
2π~2 nI

Eq. 4.45 is thus merely a restatement of the optical theorem.14 ; fkk is called the forward
scattering amplitude and we have found a way to calculate it (and the cross section). In
the simple perturbation approximation which gave rise to eqs.4.30 and 4.34 we get the
expression
Z
m 1
Im fkk = − d3 l |V (k − l)|2 δ(²k − ²l ) (4.47)
2π~2 (2π)3

which is simply restating eq. 4.43 again. Comparing eqs. 4.47 and 4.14, one gets the
feeling that the right hand side of eq. 4.47 must incorporate an approximation to the
scattering amplitudes flk . We will have a look at these in the next sections.

4.5.1 The Lippmann-Schwinger Equation


No doubt the reader will have noticed a difference in point of view between sections 4.2
and 4.1. In section 4.2 we introduced a single particle at a specific time t1 and followed
its time evolution up to time t2 . In section 4.1 we considered a beam of particles in a
steady state situation, i.e. the number of incoming and scattered particles per unit time
is constant. The scattered wave form of eq. 4.1 describes such a steady state situation.
Obviously the times of creation and detection of individual particles do not play a role
13
Formally, we have only proved it for the imaginary part, but it holds for the real part as well.
14
Which is probably why the self-energy is also called the optical potential.
78 CHAPTER 4. SCATTERING

detector

scattered particle k’

d

k dθ
θ

target

incoming particle

Figure 4.7: Angle resolved scattering detection.

in steady state. We have seen that quantities relevant to the latter situation, such as
the scattering cross section σ, can be obtained from parameters obtained from the time
evolution of a single particle. In this section we will show that the steady state situation
can also be decribed more directly via the Lippmann-Schwinger equation (which we will
derive). In particular we will find expressions for the scattering amplitudes flk .
Let us consider the experimental situation once more. Experimentally, it is possible
to measure the angular distribution of scattered particles. This can be done by using
a particle detector with a relatively small aperture, and moving this detector around in
space, or by using a whole array of such detectors, distributed such as to cover all possible
angles. Each detector at a certain position aims at capturing all scattered particles within
a solid angle df = sin θdθdϕ. The setup is shown in Fig. 4.7. One is measuring the

differential cross section df as defined in Section 4.1.1.

Integrating over all possible solid angles (which means integrating over all possible
angles covering the sphere in Fig. 4.7) must give the total cross section of the previous
section.
Z Z
dσ dσ
σ= df = sin θdθdϕ (4.48)
df df
The differential cross section provides detailed information on the properties of the target.
From Fig. 4.7 it will be clear that df dσ
(k, k0 ) depends upon k and k0 ,the states on the
incoming and the detected particles. As before, the target does not contain any internal
degrees of freedom, so it cannot absorb any energy, and one can only have elastic scattering;
i.e. ²k = ²k0 or k = k 0 . Since k and k0 fix the geometry of the measurement, i.e. the angles
dσ dσ
θ and ϕ, one has df = df (k, θ, ϕ) for elastic scattering.
4.5. THE SCATTERING CROSS SECTION 79

The rest of this section aims at (1) finding an equation which describes the steady
state situation more directly and then (1) obtaining a quantum mechanical expression for
the differential scattering cross section. Let us start as in Section 4.2 by looking at the
time evolution of a single particle in an incoming state |ki, or its Fourier transform, cf.
eq. 4.20
FT
U b+ (ω) |ki ; t2 > t1
b (t2 − t1 )|ki À G (4.49)

We then can use the Dyson equation of eq. 3.21 and write the right hand side of eq. 4.49
as

b+ (ω) |ki = G
G b+ (ω) |ki + G
b+ (ω) Vb G
b+ (ω) |ki (4.50)
0 0

We recognize that by Fourier transforming Gb+ (ω) |ki we get the complete time evolution
of the incoming particle. Obviously from Gb+ (ω) |ki one then gets the time evolution of
0
the unperturbed system, in absence of the scattering potential Vb . We know that

b+ (ω) |ki = |ki


G0 (4.51)
~ω − ²k + iη

The factor (~ω −²k +iη)−1 is handled the following way. We Fourier transform to the time
domain according to eq. 4.20. The Fourier transform can be calculated using complex
contour integration (see the Appendix, Section 4.8.2). One gets
Z
i~ ∞ b+ (ω) |ki = |kie− ~i (²k −iη)t ; t > 0
|ψ k,0 (t)i = dω e−iωt G 0 (4.52)
2π −∞
= 0; t≤0

see eq. 4.85 in Appendix II. Remember that η is a very small positive number, which leads
to a factor exp(−ηt/~). This damps the state, but as long as ~/η À any measuring time,
this factor has no physically observable effect, and is only a mathematically convenient
trick. Eq. 4.52 describes the time evolution of an unperturbed particle which is created
at t = 0 (and does not exist before that time) in state |ki. Sudden creation of a particle
obviously does not describe a steady state. But now have a look at the following state
i
|φk,0 (t)i = |kie− ~ (²k −iη)t ; t > 0 (4.53)
− ~i (²k +iη)t
= |kie ; t≤0

The bottom line involves a factor exp(+ηt/~), which goes to zero for t → −∞. This
expression describes a state which is very slowly switched on from t = −∞ to t = 0 and
dies out (very slowly) again to t = ∞. It is called an adiabatically switched-on (and -off)
state, in contrast to Eq. 4.52, which is called a suddenly switched-on state. Adiabatically
switching-on allows for a gentle approach to a steady state situation. Letting limη→0 gives
i
lim |φk,0 (t)i = |kie− ~ ²k t (4.54)
η→0

which is the familiar expression known from elementary quantum mechanics. Eq. 4.53 is
very similar to this, but it describes a state with well-defined limits at t = ∓∞. We expect
80 CHAPTER 4. SCATTERING

this to be a state of a well-defined energy ²k from the similarity with Eq. 4.54. This can
be verified explicitly by Fourier transformation
Z
1 ∞
|φk,0 (ω)i = dt eiωt |φk,0 (t)i
i~ −∞
|ki |ki
= −
~ω − ²k + iη ~ω − ²k − iη

= |kiδ(ω − ²k /~) (4.55)
i~
where we have used Eq. 4.93 in Appendix II. Now we are almost there. An expression for
the suddenly switched-on state can be found by rewriting the Dyson equation, Eq. 4.50,
as
³ ´−1
|ψ k (ω)i = G b+ (ω) |ki = I − G b+ (ω) Vb b+ (ω) |ki
G (4.56)
0 0
³ ´−1
= I−G b+ (ω) Vb |ψ k,0 (ω)i
0

Since the Dyson equation is only the Schrödinger equation in disguise, the same must hold
for an adiabatically switched-on state
³ ´−1
|φk (ω)i = I−G b+ (ω) Vb |φk,0 (ω)i
0
2π ³ b+ (²k /~) Vb
´−1
= I−G 0 |kiδ(ω − ²k /~) (4.57)
i~

≡ |φ iδ(ω − ²k /~) (4.58)
i~ k
where we recognize that G b+ can only give a contribution if ω = ²k /~, since f (x)δ(x−a) =
0
f (a)δ(x − a). The bottom line is a definition of the state |φk i. To get the time dependence
one can Fourier transform according to the prescription
Z
i~ ∞
|φk (t)i = dω e−iωt |φk (ω)i
2π −∞
Z ∞
i
= dω e−iωt δ(ω − ²k /~)|φk i = |φk ie− ~ ²k t (4.59)
−∞
It descibes a state with at an energy ²k , in which the effect of the scattering potential
is fully taken into account. Comparing Eqs. 4.57 and 4.58 and deleting the δ(ω − ²k /~)
factors we get
³ ´−1
|φk i = I − Gb+ (²k /~) Vb |ki
0

which is equivalent to
b+ (²k /~) Vb |φk i
|φk i = |ki + G (4.60)
0

This equation is called the Lippmann-Schwinger equation.. As we can see, the ω-dependence
is not really necessary. We can work with |φk i and |ki, which are related by eq. 4.60; their
i
time dependence is obtained by multiplying with e− ~ ²k t , cf. eqs. 4.54 and 4.59. From
the Lippmann-Schwinger equation on can calculate the scattering state |φk i from the un-
perturbed state |ki, where both have the fixed energy ²k . Remember that quantum me-
chanically fixed energy means fized frequency and it is clear that the Lippmann-Schwinger
equation describes the situation of section 4.1.
4.5. THE SCATTERING CROSS SECTION 81

Connection to the time-independent Schrödinger equation


The Lippmann-Schwinger equation, Eq. 4.60, describes the scattering of particles at a
2 k2
fixed energy ²k = ~2m , which is the energy of the incoming particle. It is easy to show
that |φk i is the eigenstate of the full Hamiltonian with eigenvalue ²k . From eq. 4.60 one
gets

b 0 )|φk i = (²k − H
(²k − H b 0 )|ki + (²k − Hb 0 )G
b+ (²k /~) Vb |φk i
0
b
= (²k − ²k )|ki + IV |φk i ⇔
b 0 + Vb )|φk i = ²k |φk i
(H (4.61)

where we have used eq. 3.28 of Section 3.2.4. The bottom line is the time-independent
Schrödinger equation, familar from elementary quantum mechanics. Following the steps
from eq. 4.61 to eq. 4.60 backwards is another way to derive the Lippmann-Schwinger
equation. It shows that the latter is completely equivalent to the time-independent
Schrödinger equation.15 In conclusion, the main difference between sections 4.2 and 4.1
is the use of a time-dependent Schrödinger picture in the former, and a time-independent
one in the latter.

Wave functions
We are interested in the spatial behavior of the Lippmann-Schwinger equation, i.e. in the
position representation or the wave function φk (r) = hr|φk i and we will show that it will
give the wave form of eq. 4.1 in Section 4.1. Inserting a couple of resolutions of identity
in eq. 4.60 one obtains

b+ (²k /~) Vb |φk i


φk (r) = hr|φk i = hr|ki + hr|G
Z Z 0

= hr|ki + 3 0 3 00
d r d r hr|G b+ (²k /~)|r0 ihr0 |Vb |r00 ihr00 |φk i
0
Z
1 b+ (²k /~) |r0 iV (r0 )φk (r0 )
= √ eik·r + d3 r0 hr|G 0 (4.62)

where we have used eqs. 4.16; the last line follows from hr0 |Vb |r00 i = V (r00 )hr0 |r00 i =
V (r00 )δ(r0 − r00 ). The interesting bit is
X
b+ (²k /~) |r0 i =
hr|G b+ (²k /~) |k0 ihk0 |r0 i
hr|k00 ihk00 |G
0 0
k00 ,k0
X hr|k00 ihk0 |r0 i
= δ 0
² − ²k0 + iδ k,k
00 0 k
k ,k
Z 0 0
1 3 0 eik ·(r−r )
= d k (4.63)
(2π)3 ²k − ²k0 + iδ
15
The nomenclature is slightly confusing. Both the forms of eq. 4.60 and 3.59 in the appendix of the
previous chapter, are called the Lippman-Schwinger equation. Eq. 3.59 works in the time domain and is
equivalent to the time-dependent Schrödinger equation. Eq. 4.60 works at a fixed energy and is equivalent
to the time-independent Schrödinger equation. If you want a delicate mathematical proof that both forms
give equivalent results, see R. G. Newton, Scattering Theory of Waves and Particles, (Springer, New York,
1982); Ch. 6,7.
82 CHAPTER 4. SCATTERING

where we have used eqs. 4.16, 4.22 and 4.28. Note that we could have immediately
2 k02
obtained this result from eqs. 3.30 and 3.31. Knowing that ²k0 = ~2m , the integral of eq.
4.63 can be done.16 The end result is quite simple and elegant
0
b+ (²k /~) |r0 i = − m eik|r−r |
hr|G0 (4.64)
2π~2 |r − r0 |
So eq. 4.62 becomes
Z 0
1 m eik|r−r |
φk (r) = √ eik·r − d3 r0 V (r0 )φk (r0 ) (4.65)
Ω 2π~2 |r − r0 |
This is the real space representation of the Lippmann-Schwinger equation.

4.5.2 The Scattering Amplitudes and the Differential Cross Section


The first term on the right hand side of eq. 4.65 is the plane wave which describes the
incoming particle; the second term on the right hand side describes the scattered wave.
The potential V (r0 ) which describes Rthe target has a finite range and it decays fast outside
the target. So whereas the integral d3 r0 in principle goes over Ω (the volume of the box
we used to normalize our states), in practice V (r0 ) ensures that one only has to integrate
over a small volume which contains the target. The position r at which the scattered
particles are detected is far from the target. The target is a microscopic particle, so
r0 . 10−14 − 10−9 m, whereas the detector is part of a macroscopic laboratory experiment,
so typically r ∼ 10−2 − 101 m. In other words, we are interested in the limit r À r0 in eq.
4.65. In this limit we can make the approximation
¯ ¯
¯r − r0 ¯ ≈ r − er · r0

where er is a unit vector in the direction of the detector r and therefore


0
eik|r−r | eikr −iker ·r0
≈ e
|r − r0 | r
Defining k0 = ker , a wave vector in the direction of the detector with the same size as
that of the incoming wave, we get
Z
1 ik·r m eikr 0 0
φk (r) −→ √ e − 2
d3 r0 e−ik ·r V (r0 )φk (r0 )
r large Ω 2π~ r
√ ikr Z
1 m Ωe
= √ eik·r − 2 r
d3 r0 hk0 |r0 iV (r0 )hr0 |φk i
Ω 2π~

1 ik·r m Ω eikr 0 b
= √ e − hk |V |φk i (4.66)
Ω 2π~2 r
Defining a quantity fk0 k by
mΩ 0 b
fk0 k = hk |V |φk i (4.67)
2π~2
0
16
If you want to try and do it yourself. Write d3 k0 = k 2 dk0 d(cos θ)dϕ in spherical coordinates and
ik·(r−r0 ) 0
e = eik|r−r | cos θ , where θ is defined as the angle between k0 and (r − r0 ). First do the integral over
dϕ and then the integral over d(cos θ). The remaining integral over dk0 can be done by contour integration,
as discussed in the Appendix.
4.5. THE SCATTERING CROSS SECTION 83

we obtain
· ¸
1 eikr
φk (r) −→ √ eik·r − fk0 k (4.68)
r large Ω r

i
Adding the time factor e− ~ ²k t as in eq. 4.59, we see that this corresponds to the wave form
of eq. 4.1 in Section 4.1 (which actually proves that this wave form that was postulated
on “intuitive” grounds, is correct). Eq. 4.67.now gives a way to calculate the scattering
amplitudes fk0 k .

Just one more step brings us to the differential cross section. Imagine that the large
sphere in Fig. 4.7 has a radius r, then the flux Φ(θ, ϕ) of scattered particles through the
black surface element of that sphere is

Φ(θ, ϕ) = probability density at surface × area surface element × speed particle


|fk0 k |2
= × r2 df × v (4.69)
Ωr2

One has to divide this by the incoming flux Φin = Ωv (see the previous section) and the
solid angle df to get the differential cross section according to its definition

dσ Φ(θ, ϕ)
= = |fk0 k |2 (4.70)
df Φin df

Comparing eq. 4.66 to eq. 4.68 one finds the final elegant result

dσ m2 Ω2
= |fk0 k |2 = 2 4 |hk0 |Vb |φk i|2 (4.71)
df 4π ~

4.5.3 The Born Series and the Born approximation


The occurrence of |φk i in this expression supposes that one knows the solution of the
Lippmann-Schwinger equation, eq. 4.60. If one does not, one can get successive better
approximations by applying a series expansion of eq. 4.60 in the same spirit as the Dyson
expansion, cf. eq. 3.17

b+ (²k /~) Vb |ki + G


|φk i = |ki + G b+ (²k /~) Vb G
b+ (²k /~) Vb |ki + ..... (4.72)
0 0 0

Although it is actually the same thing as the Dyson expansion, in scattering theory this
is called the Born series. If one neglects all terms beyond the first one on the right hand
side of eq. 4.72, one defines the so-called “Born approximation” for eq. 4.71

dσ 1 m2 Ω2 m2
= 2 4 |hk0 |Vb |ki|2 = 2 4 |V (k0 − k)|2 (4.73)
df 4π ~ 4π ~

according to eq. 4.29. In the context of scattering theory “the Born approximation” has
this very specific meaning. We can prove that it is equivalent to what can be obtained by
84 CHAPTER 4. SCATTERING

Fermi’s golden rule. The total cross section in this approximation is given by
Z Z
dσ 1 1 dσ 1 2
σ1 = sin θdθdϕ = 2 k sin θdθdϕ
df k df
Z
m2
= |V (k − k0 )|2 δ(k − k 0 )k02 sin θdθdϕ
4π 2 ~4 k 2
Z
m2 ~2 k
= |V (k − k0 )|2 δ(²k − ²k0 )d3 k 0
4π 2 ~4 k 2 m
Z
1
= |V (k − k0 )|2 δ(²k − ²k0 )d3 k 0 (4.74)
4π 2 ~v
1
using δ(b(x2 − a2 )) = 2ab δ(x − a) and v = ~k
m . This corresponds to eqs. 4.45 and 4.47. We
have now found three separate ways to find the total cross section in this approximation.
1. Using eq. 4.44, which needs the decay rate Γk . This is related to the self-energy Σ,
via eq. 4.38. The self-energy is then approximated via eq. 4.30.
2. Again using eq. 4.44, but now obtaining the decay rate Γk using the “conservation
of particles” and Fermi’s golden rule, cf. eqs. 4.42 and 4.43.
3. Using the differential cross section in the “Born approximation”, eq. 4.73, and
integrating over all angles.
The equivalence of 2. and 3. shows that the physical content of Fermi’s golden rule
and the “Born approximation” is essentially the same. This should not be surprising,
since if we follow the derivation of the differential cross section backwards, from eq. 4.74
to eq. 4.60, one observes that the “Born approximation” of eq. 4.73 can be obtained by
replacing |φk i by |ki at the right hand side of eq. 4.60. So we make the approximation
b+ (²k /~) Vb |ki
|φk i ≈ |ki + G (4.75)
0
Going back from eq. 4.60 to eq. 4.50, one observes that this corresponds to the approxi-
mation for the Green function operator
Gb+ (ω) ≈ Gb+ (ω) + G
b+ (ω) Vb G
b+ (ω) (4.76)
0 0 0
But this is essentially nothing else than the Fourier transform of the expression we got in
first order perturbation theory, cf. eq. 2.23 in Section 2.3, from which we derived Fermi’s
golden rule. “First order perturbation theory” was called the “Born approximation” in
that section; now we see that it is consistent with the use of this phrase in scattering
theory.

The Born approximation is used a lot in scattering theory, because it is so simple. To


calculate the differential cross section one only needs the Fourier transform of the scattering
potential, cf. eqs. 4.73 and 4.29. Unfortunately it does not always give accurate results;
if the scattering potential Vb is large, neglect of the higher order perturbation terms is not
allowed. For moderately strong potentials, one can add subsequent higher order terms
as in eq. 4.72. Nowadays, for single particle scattering it is also possible to solve the
Lippmann-Schwinger equation (eq. 4.60) numerically on a fast computer, usually using
the real space representation of eq. 4.65. From the numerical function |φk i one then
calculates the differential cross section of eq. 4.71, and by numerical integration over all
angles the total cross section. The procedure can be quite involved, since one is usually
interested in knowing the latter for a whole range of k-values.
4.6. EPILOGUE 85

4.6 Epilogue
A final word about this chapter. It might seem that we made it lot of fuss about things
that could have been derived simply from Fermi’s golden rule, eq. 2.32. We can obtain
the transition rates wk→l , eq. 4.42 from Fermi’s golden rule. From the “conservation of
particles” we can establish that this leads to an exponentially decaying probability of the
particle staying in state k, with a rate Γk given by eq. 4.43. This short-cut would have
saved us an enormous amount of equation deriving.17 However I did not take the detour
for you just to admire the scenery; it has given us some important road signs for the future
(quasi-particles and self energies). The quasi-particle result, eq. 4.39, is much more general
than what could be obtained under Fermi’s rule. We can make better approximations to
the self energy Σ (k, ω) than just the first two terms of the perturbation expansion as in
eq. 4.30. The “pole” equation 4.36 can also be solved to any accuracy to obtain a root
ω k . It may require extra work and even some numerical computing to obtain accurate
results, but the general equation, eq. 4.39, still stands.
Also it might seem strange that we speak of a quasi-particle when obviously we have
been dealing with a well-defined single particle all the time. This term however remains
useful in a more complicated environment. Imagine that we have a large number scattering
centers that are not very far apart, like in a dense gas, or a dirty metal full of impurities.
The particle (e.g. an electron) is then a pinball in a giant (1017 —1023 atoms) pinball ma-
chine. Multiple scattering from many centers leads to a complicated interference pattern
of scattered waves and to very complicated particle states. Yet one can prove that the
general equation 4.39 is still valid. Even if one takes into account all the electrons in the
crystal when adding an extra one - we then have a giant number (1023 ) of pinballs in a
giant pinball machine - even then eq. 4.39 is still relevant for describing the time evolution
or propagation of the extra electron. Of course one looses track of the individual particles
amidst all this scattering. As we will see later on, quantum mechanics tells us that elec-
trons are indistinguishable, so it is even fundamentally impossible to track a single electron
in the pinball machine. Yet, as long as we have a time dependence which is given by eq.
4.39, we quasi still have a single particle, or in other words a quasi-particle. Whether the
quasi-particle concept is useful in practical situations depends on the numerical value of
the lifetime τ k = Γ1k . If it is shorter than any time which is experimentally accessible,
then obviously the concept is without much practical use. Surprisingly, in “typical” ex-
periments on “typical” materials, lifetimes of “typical” particles, like electrons and holes,
are quite large. This makes the quasi-particle concept extremely useful in practice. It
helps you to avoid feeling like the pinball kicked around by the system, and puts you in
the deriver’s seat.

4.7 Appendix I. The Refractive Index


Averaging the wave of eq. 4.8 over the small surface a2 under the same conditions as in
Section 4.1.2 gives
Z Z · ¸
1 A0 2 2πi
φ(z,t) = 2 dxdy φ(r,t) = 2 a + fk1 k1 ei(kz−ωt) (4.77)
zÀλ a a a a k
17
Get a feeling for why Fermi’s rule is called “Golden” ?
86 CHAPTER 4. SCATTERING

The term between [...] represents the change of the forward propagating wave, i.e. the
transmitted wave, due to the scattering of one box in Fig. 4.2. If the sample becomes
thicker one might wonder what the collective effect of the targets in all boxes is on the
transmission of the wave through the sample. All boxes in a layer perpendicular to k1 ,
i.e. in an xy-layer, produce the same change on φ(z,t). The number Nxy of such boxes
which fall into the surface a2 is given by a2 /L2 (where L2 is the surface of one box).
Incorporating the contributions of a complete xy-layer gives
· ¸ · ¸
2πi 2πi
φ(z,t) = A0 1 + 2 Nxy fk1 k1 ei(kz−ωt) = A0 1 + fk k ei(kz−ωt)
zÀλ ka kL2 1 1

In the propagation direction z we get an additional factor [...] for each layer of thickness
L we add, because we enter a new layer of targets. The basic idea is shown in Fig. 4.8.

one box

xy-layer

Figure 4.8: Adding layers to calculate the index of refraction.

So at a position z we get
· ¸z
2πi L
φ(z,t) = A0 1 + fk k ei(kz−ωt)
kL2 1 1

· ¸z
2πiz L L i(kz−ωt)
= A0 1 + fk k e
kL3 1 1 z
· ¸z
2πinI z L L i(kz−ωt)
= A0 1 + fk1 k1 e
k z

since L−3 = nI is the density of the targets (gas atoms, i.e. one per box). If M = Lz is
x M
very large, then we can use l’Hospital’s rule, i.e. limM→∞ (1 + M ) = ex to write
2πinI z
fk1 k1
φ(z,t) = A0 e k ei(kz−ωt) = A0 ei(nk1 kz−ωt) with
2πnI
nk1 = 1 + 2 fk1 k1 (4.78)
k

The wave ei(nk1 kz−ωt) is recognized as a plane wave, where nk1 is the index of refraction.
It is a function of the state k1 of the incoming beam. In a dilute system, it can be directly
coupled to the forward scattering amplitude fk1 k1 of a single target. Note that the index
4.8. APPENDIX II. APPLIED COMPLEX FUNCTION THEORY 87

of refraction is in general a complex number, because fk1 k1 is complex; fk1 k1 = Re fk1 k1 +


i Im fk1 k1 . The imaginary part i Im fk1 k1 gives rise to “absorption” in the medium, i.e.

¯ ¯ 4πnI
¯φ(z,t)¯2 = |A0 |2 e−γ k1 z with γ k1 = Im fk1 k1 = nI σ k1 (4.79)
k

according to eq. 4.13. The intensity of the wave is decaying exponentially, and γ k1
is usually called the extinction coefficient or the absorption coefficient. Note however
that absorption in our sample is not caused by “absorption” by the individual targets
(gas atoms). The number of particles (or energy) is fully conserved; they are simply
scattered by the targets out of the propagating wave, the intensity of which then decays
exponentially. It makes good sense that the decay constant γ k1 is proportional to the
density of scattering targets nI and their individual cross section σ k1 . The real part of
nk1 in eq. 4.78 describes the change in wave length λ = 2π(k Re nk1 )−1 in the medium as
compared to 2πk −1 in free space (assuming a fixed frequency ω).

4.8 Appendix II. Applied Complex Function Theory


“Nothing is real, nothing to get hung about”, Lennon & McCartney, Strawberry Fields Forever.

The following is just a summary of some of the elements of complex function theory;
it gives theorems without proofs. You should have a look into a mathematical physics
book for the “real stuff”; e.g. G. Arfken, Mathematical methods for physicists, (Academic
Press, New York, 1985). Or better still, do a mathematics course on complex function
theory.

4.8.1 Complex Integrals; the Residue Theorem


Most of the elementary real functions you know can be extended into the complex plane,
e.g. √ez = ea+bi = ea (cos b+i sin b) is a well-defined function (z a complex, a, b real numbers,
i = −1). A function f (z) in the complex plane is called analytic at a point z = z0 , when
it is differentiable in and around z0 . Integrals in the complex plane are done by contour
integration, which is a path integration in the complex plane, see Fig. 4.9.

Im z '
z4 ' z2 = zn '
C z3 '

z2 ' ς2'
z1 '
ς1'

z1 = z 0 '
Re z '

Figure 4.9: Contour integration = integration along a path in the complex plane.
88 CHAPTER 4. SCATTERING

Z z2 n
X
f (z 0 )dz 0 = lim
n→∞
f (ζ i )∆zi0 (4.80)
C, z1 max |∆zi0 |→0 i=1

where ∆zi0 = zi+1


0 − zi0 and zi0 ≤ ζ i ≤ zi+1
0 and C denotes the “contour”, which is the path
taken. We now cite a couple of theorems

1. Cauchy’s integral theorem: if f (z)


H is analytic across a domain which includes the closed
contour C then the integral C f (z) dz = 0, see Fig. 4.10.

Im z '
C

Re z '

Figure 4.10: A closed contour C in the complex plane.

2. This can be used to prove the very useful Cauchy’s integral formula: Let f (z) be
analytic across a domain which includes the closed contour C. If C is traversed in a
counter-clockwise direction then
I
f (z)
dz = 2πi f (z0 ) if the point z0 is enclosed by C; Fig. 4.11(a)
C z − z0
= 0 if z0 is outside C; Fig. 4.11(b) (4.81)
1
Because of the z−z0 factor the integrand is obviously not analytic in the point z = z0

C C z0 C’
z0 z0

(a) (b) (c)

Figure 4.11: Cauchy’s integral formula.

(it is not differentiable in this point; the derivative depends upon the direction from
which one approaches the point). This point z = z0 is called a pole of the function
4.8. APPENDIX II. APPLIED COMPLEX FUNCTION THEORY 89

f (z)
g(z) = z−z 0
. If the contour is traversed in a clockwise direction, as C 0 , see Fig.
4.11(c), then
I I
...dz = − ...dz
C0 C

Cauchy’s integral formula leads to the most important theorem of complex function
theory.

3. It is called the residue theorem: Suppose a function f (z) has poles at the points
z = z0 , z1 , ..... zn . Define the so-called residues by

ai = [(z − zi ) f (z)] at z=zi ; i = 0, 1, ..... n (4.82)

then the residue theorem states that the integral of f (z) over a closed contour C is
given by
I enclosed
X
f (z) dz = 2πi ai (4.83)
C i

i.e. the sum of all residues at the poles zi which are enclosed by the contour C.
Note that the residue theorem can be seen as a generalization of Cauchy’s integral
formula.

4.8.2 Contour Integration


Oddly enough in practice, complex contour integration is usually performed not to calcu-
late integrals over a “complex” path, but to calculate integrals along the “real” axis. Let
us say we are interested in the integral
Z ∞
e−iωt
dω ; t>0
−∞ ω − ² + iδ

which obviously is related to the inverse Fourier transform (from the frequency to the time
domain) of the Green function of eq. 3.26. In order to solve this integral we make use of a
standard trick of complex contour integration. The integral obviously is along the real ω
axis, but we make it part of a complex contour integral, by making ω a complex variable.
The integrand then has a pole at ω = ² − iδ (δ, ² are real and δ > 0). We choose to do
the integration along the closed contour C given in Fig. 4.12, where in the end we take
limR→∞ .
It is called “closing the contour in the lower half plane”. Because the pole is enclosed
e−iωt
by C and, except at the pole, the function f (ω) = ω−²+iδ is analytical, we have according
to the residue theorem
I
f (ω) dω = 2πie−i(²−iδ)t (4.84)
C
H R −R
RObviously we can split the integral over C into two parts C f (ω) dω = R f (ω) dω +
S f (ω) dω, where the first part describes the path along the real axis and the
R second part
describes the path along the semi-circle S. We are now going to talk the S integral to
90 CHAPTER 4. SCATTERING

Re ω

C
−R R
Im ω
ε − iδ

Figure 4.12: Closing the contour in the lower half plane.

zero. A path integral can be parametrized as usual. For ω on a semi-circle of radius R, we


can write ω = R eiθ ; in order to describe S we let θ run from π to 2π; note dω = R eiθ idθ.
We have
Z Z
e−ωt
f (ω) dω = dω or
S S ω − ² + iδ
Z 2π ³ ´ Z 2π tR sin θ −itR cos θ
iθ iθ e e
f R e R e idθ = iθ
R eiθ idθ
π π R e − ² + iδ

Here comes the trick: since sin θ < 0 for π < θ < 2π and we had t > 0, the factor etR sin θ
behaves like e−αR with α > 0. The rest of the integrand is limited, its absolute value is
less than some fixed number, independent of R. Then in the limit lim , the e−αR factor
R→∞
in the integrand makes the integral zero
Z 2π
lim ... dθ = 0
R→∞ π

so in this limit only the part of the path along the real axis remains
I Z −R Z ∞
lim f (ω) dω = lim f (ω) dω = − f (ω) dω
R→∞ C R→∞ R −∞

using the result we already have found for the contour integral, eq. 4.84, we finally find
for our integral
Z ∞
e−iωt
dω = −2πie−i(²−iδ)t ; t > 0 (4.85)
−∞ (ω − ² + iδ)
We obtain this integral essentially without doing any work; the residue theorem is a
brilliantly elegant piece of mathematical equipment!

If t < 0 the same trick does not work because then the factor etR sin θ behaves like eαR
with α > 0, which obviously explodes if R → ∞. However in this case we can use the
contour given in Fig. 4.13. It is called “closing the contour in the upper half plane”.
The semicircle S can again be parametrized as ω = R eiθ ; but now we let θ run from
0 to π. Since sin θ > 0 for 0 < θ < π and we had t < 0, the factor etR sin θ again behaves
4.8. APPENDIX II. APPLIED COMPLEX FUNCTION THEORY 91

Re ω
S

Im ω
−R R
ε − iδ
C

Figure 4.13: Closing the contour in the upper half plane.

R H
like e−αR with α > 0 so lim f (ω) dω = 0. However f (ω) dω = 0 according to the
R→∞ S C
residue theorem, since the pole ² − iδ now lies outside the closed contour C and thus
Z ∞
e−iωt
dω = 0 for t < 0 (4.86)
−∞ (ω − ² + iδ)

NOTE that the results of eqs. 4.85 and 4.86 can be used to prove that the inverse
b+ (ω), cf. eq. 3.26, i.e.
Fourier transform of G0
Z ∞
b + 1 b+ (ω)
G0 (t) = dω G0
2π −∞

is given by eq. 3.23. Note especially how the +iδ term in the denominator of the integrand
in eq. 4.86 transforms into the Θδ (t) function of eq. 3.25. Thus the little δ takes care of
the physical “causality” which was discussed around eq. 3.6. Had we taken a −iδ term
instead (δ > 0), then we would have a pole at ω = ² + iδ in the upper half plane. The
contour integral over the lower half plane, which is appropriate for t > 0, then leads to
zero, since the contour does not enclose the pole. However, closing the contour in the
upper half plane, which is appropriate for t < 0, encloses the pole. The result of this
contour integration is
Z ∞
e−iωt
dω = 0 for t > 0
−∞ (ω − ² − iδ)
Z ∞
e−iωt
dω = 2πie−i(²+iδ)t for t < 0 (4.87)
−∞ (ω − ² − iδ)

The Green function operator corresponding with eq. 3.26 using a −iδ instead of the +iδ
b− (ω). From the results above we conclude that its inverse Fourier transform G
is G b− (t) is
0 0
different from zero only for t < 0. There seems to be little physics in such an non-causal
function, but we will have a use for this Green function later on in the program.

We have to say something about limits now, which is a confusing subject. Up until
the last paragraph we have used a finite number δ > 0. But, as stated in the main text at
some point we usually want to take limδ→0 . Let us look again at the main example of this
92 CHAPTER 4. SCATTERING

section, the results of which are given in eqs. 4.85, 4.86 and 4.87. It is clear that it is not
a good idea to let δ → 0 in the integrand right away, since the result of the integration
obviously depends upon whether we start with +iδ or −iδ. Setting δ = 0 in the integrand
gives a pole for ω = ² on the real axis, right on the path of integration. Having such a
singularity on your integration path is bad news. In order to get any sensible result at all
you have to define a limiting procedure. What we did here, moving the pole off the real
axis by defining a finite δ, then do the integral and finally take limδ→0 , is one well defined
limiting procedure. The prize we pay for trying to do this horrible integral is that the end
result depends upon the particular limiting procedure that is used. For instance, whether
we start with +iδ or −iδ makes a difference. Mathematically, any procedure is fine, as
long as we define exactly what we are doing. Physically, the boundary conditions defined
by the physical system usually impose the limiting procedure to be used. As we discussed
before, the principle of causality defines a Gb+ (t), which can only be achieved by starting
0
the limiting procedure with +iδ.

EXAMPLE FOR EXERCISES


In the exercises we will encounter the following integral, which can be easily done using
the residue theorem, eqs. 4.82 and 4.83, and the trick of closing the integral in the lower
and upper half plane, respectively
Z ∞ " #
e−iωt e−i(²1 −iδ)t e−i(²2 −iδ)t 2π
dω = + for t > 0
−∞ (ω − ²1 + iδ) (ω − ²2 + iδ) ²1 − ²2 + iδ ²2 − ²1 + iδ i
= 0 for t < 0

We can rewrite the first expression a bit defining ² = 12 (²1 + ²2 ) and ∆ = 1


2 (²1 − ²2 ); we
have ²1 = ² + ∆ ²2 = ² − ∆, so
Z ∞ Ã !
2π e−i(²+∆)t e−i(²−∆)t
...dω = +
−∞ i −2∆ 2∆
2π −i²t
= e sin ∆t (4.88)

where we have taken limδ→0 to obtain this expression.

4.8.3 The Principal Value


Another way of looking at the integral is rewriting it as
Z ∞ Z ∞
e−iωt e−iωt (ω − ² − iδ)
dω = dω
−∞ (ω − ² + iδ) −∞ (ω − ² + iδ) (ω − ² − iδ)
Z ∞ Z ∞
e−iωt (ω − ²) e−iωt iδ
= dω 2 − dω (4.89)
−∞ (ω − ²) + δ 2 −∞ (ω − ²)2 + δ 2

The function (δ, ² constants; ω variable)

1 δ
L(ω − ²) =
π (ω − ²)2 + δ 2
4.8. APPENDIX II. APPLIED COMPLEX FUNCTION THEORY 93

1
πδ
L(ω−ε)

ε ω

Figure 4.14: Lorenzian line shape function.

is called a Lorentzian line shape function (or for short: Lorentzian). It looks like in Fig.
1
4.14, centered around ω = ² with a peak R ∞ height of πδ and a width at half height of 2δ
It is not difficult to prove that (a) −∞ dω L(ω − ²) = 1 . A little bit more difficult it is
R∞
to prove that, given a well-behaved function f (ω), limδ→0 −∞ dω L(ω−²)f (ω) = f (²). But
these are just what is needed to define a δ-function, which means limδ→0 L(ω−²) = δ(ω−²).
This takes care of the second integral in the bottom line of eq. 4.89. The first integral
is an example of what is called a Cauchy principal value, notation Pv. It is actually a
definition of a (yet another) limiting procedure defined as follows. Suppose the function
f (x) has a pole on the real axis at the point x = x0 ; then the principal value of its integral
over the real axis is defined as
Z ∞ ½Z x0 −a Z ∞ ¾
Pv dx f (x) = lim dx f (x) + dx f (x)
−∞ a→0 −∞ x0 +a

One approaches the singularity “sideways”. In our case we are interested in the following
principal value
Z ∞ ½Z ²−δ Z ∞ ¾
e−iωt e−iωt e−iωt
Pv dω = lim dω + dω
−∞ (ω − ²) δ→0 −∞ (ω − ²) ²−δ (ω − ²)
Z ∞ −iωt
e (ω − ²)
= lim dω
δ→0 −∞ (ω − ²)2 + δ 2

The proof for the last line in this equation is given by Mattuck, p. 61. In summary,
taking the limδ→0 , the integral of eq. 4.89 is given by
Z ∞ Z ∞ Z ∞
e−iωt e−iωt
lim dω = Pv dω − πi dω δ(ω − ²) e−iωt (4.90)
δ→0 −∞ (ω − ² + iδ) −∞ (ω − ²) −∞
Z ∞
e−iωt
= Pv dω − πie−i²t
−∞ (ω − ²)

This type of expression is used quite a lot; we have used it in eqs. 4.30—4.34. Using the
principal value approach for the integral of eq. 4.90 does not make too much sense, since
we have already obtained the result by other means (I have used it mainly for illustrative
94 CHAPTER 4. SCATTERING

purposes). Comparison with eqs. 4.85 and 4.86 however does give the principal value of
this particular integral
Z ∞
e−iωt
Pv dω = −πie−i²t for t > 0 (4.91)
−∞ (ω − ²)
= +πie−i²t for t < 0

We have used the function e−iωt in the numerator of the integrand as an example, but we
can replace it in eq. 4.90 by any well-behaved (real or complex) function f (ω) without a
problem.
Z ∞ Z ∞ Z ∞
f (ω) f (ω)
I = lim dω = Pv dω − πi dω δ(ω − ²) f (ω) (4.92)
δ→0 −∞ (ω − ² + iδ) −∞ (ω − ²) −∞
R∞
If the function f (ω) happens to be a real function, then Pv −∞ corresponds with the real
part of the integral Re I, and the δ-function part corresponds with Im I. Eq. 4.92 is often
loosely written down as

1 Pv
lim = − πiδ(ω − ²) (4.93)
δ→0 ω − ² + iδ ω−²

It is called the well-known theorem from complex function theory.18 What is meant by
this odd short-hand notation is actually the full integral of eq. 4.92.

The integrals we have seen so far have been relatively simple. Often we encounter
integrals in which ²(k) with k a wave vector and we have to integrate over all wave vectors
in three dimensions, cf. eqs. 4.28— 4.34. Contour integration is not easy in that case and
it becomes advantageous to use the separation in a Pv and a δ-function part. The latter
is usually quite doable (although not as easy as in the one dimensional case); the Pv part
requires somewhat more work.

4.8.4 The Self-Energy Integral


In the text we have encountered an integral of the form
Z ∞ Z ∞
e−iωt
dω f (ω) = dω
−∞ −∞ ω − ² − Σ (ω)

where Σ (ω) = Re Σ (ω)+i Im Σ (ω) is a complex function, cf. eq. 4.37. Suppose Im Σ (ω) <
0 and Σ (ω) is an analytical function. Furthermore suppose Σ (ω) is such that ω − ² −
Σ (ω) = 0 has only one (complex) solution. Call that solution ω1 . It is clear that the
integrand f (ω) has a pole at ω = ω1 . As before, for t > 0 the integral can be done using
the trick of closing the integral in the lower half plane and applying the residue theorem,
eqs. 4.82 and 4.83. We need to calculate the residue

a1 = [(ω − ω1 ) f (ω)] at ω=ω1


18
Calling something “well-known” is a well-known trick to intimidate the audience in order to avoid
questions that would embarasse the speaker.
4.8. APPENDIX II. APPLIED COMPLEX FUNCTION THEORY 95

Expand the analytical function Σ (ω) around ω = ω 1


© ª
Σ (ω) = Σ (ω1 ) + Σ0 (ω 1 ) (ω − ω 1 ) + O (ω − ω1 )2

Then, using ² − Σ (ω1 ) = ω 1

(ω − ω1 ) e−iωt
lim (ω − ω 1 ) f (ω) = lim
ω→ω1 ω→ω 1 ω − ² − Σ (ω 1 ) − Σ0 (ω 1 ) (ω − ω 1 ) − −O {(ω − ω 1 )2 }

e−iωt
= lim
ω→ω 1 (1 − Σ0 (ω 1 )) − O {(ω − ω 1 )1 }

e−iω1 t
=
1 − Σ0 (ω 1 )

which finally gives for our integral


Z ∞
e−iωt e−iω1 t
dω = −2πi ; t>0
−∞ ω − ² − Σ (ω) 1 − Σ0 (ω 1 )
e−i(Re ω1 −i|Im ω1 |)t
= −2πi (4.94)
1 − Σ0 (ω 1 )

One observes that the factor (1 − Σ0 (ω1 ))−1 corresponds to the factor zk in eq. 4.37. Note
that ω 1 is a complex number and since Im Σ (ω) < 0 we also have Im ω 1 < 0; thus this
factor leads to exponential decay. Of course for t < 0 one can prove by closing the contour
in the upper half plane that the integral is zero.

In case |Im Σ (ω)| is very small, as one would expect for a quasi-particle, we can make
a further approximation. Although as such this is not a part of complex analysis, this
approximation is often done for quasi-particles and we will discuss it here. One starts
from the equation for the pole

ω − ² − Σ (ω) = 0 ⇔
ω − ² − ΣR (ω) − iΣI (ω) = 0

where ΣR (ω) ≡ Re Σ (ω) and ΣI (ω) ≡ Im Σ (ω). Denote ω = ω 0 the (real) solution to
ω − ² − ΣR (ω) = 0. Because |ΣI (ω)| is small, the solution ω = ω 1 of the pole equation
will be close to ω 0 . Make the following approximations

ΣI (ω) ≈ ΣI (ω 0 ) (4.95)
ΣR (ω) ≈ ΣR (ω 0 ) + Σ0R (ω 0 ) (ω − ω0 )

the pole equation becomes

ω − ² − ΣR (ω 0 ) − Σ0R (ω 0 ) (ω − ω 0 ) − iΣI (ω0 ) = 0 ⇔


ω − ω 0 − Σ0R (ω 0 ) (ω − ω 0 ) − iΣI (ω 0 ) = 0 ⇔
iΣI (ω 0 )
ω − ω0 =
1 − Σ0R (ω 0 )
96 CHAPTER 4. SCATTERING

or
i
ω 1 = ω 0 − Γ0 where
2
2 |ΣI (ω 0 )|
ω 0 = ² + ΣR (ω 0 ) and Γ0 = (4.96)
1 − Σ0R (ω0 )

Consistent with the approximation of eq. 4.95 is using Σ0 (ω1 ) ≈ Σ0R (ω 0 ) in eq. 4.94
Z i

e−iωt e−i(ω0 − 2 Γ0 )t
dω = −2πi ; t>0 (4.97)
−∞ ω − ² − Σ (ω) 1 − Σ0R (ω 0 )

This equation is an approximation of eq. 4.94 when |ΣI (ω)| (or Γ0 ) is small.
Part II

Many Particles

97
Chapter 5

Quantum Field Oscillators

“Make everything as simple as possible, but not simpler”, A. Einstein.

In the previous chapter, we assumed that we could easily obtain all the eigenstates
of an unperturbed Hamiltonian H b 0 . However, you already know from previous encoun-
ters with quantum mechanics that in practice this is rare. Especially when dealing with
a many particle system (and here few is already many) the best we can hope for is to
find decent approximations. Choosing the right representation (i.e. basis set) helps a lot.
This is discussed in Mattuck’s Appendix A; you might want to check out this appendix.
This chapter and the next one contain a few many particle systems that can be solved
“exactly”. As we will see, all of these examples can be rewritten in terms of “indepen-
dent”, i.e. non-interacting, particles. Sometimes this becomes clear only after choosing
the right representation. It will be illustrated in this chapter on a couple of many boson
problems which are related to the ubiquitous harmonic oscillator. We will introduce the
representation which is most widely used in many particle physics; it is called the occu-
pation number representation. The techniques developed for handling this representation
are called second quantization. This chapter is organized as follows. After learning the
art of “solving the harmonic oscillator problem by algebra only”, this art is then applied
to the problem of lattice dynamics. The continuum limit of the lattice problem serves to
illustrate how to quantize a classical system which is described by a field, and not by a
collection of mechanical particles. In the last section I briefly discuss the quantization of
the classical electro-magnetic field using this method. In general, quantizing a classical
field using the second quantization technique (or occupation number representation) is the
subject of quantum field theory.

5.1 The Quantum Oscillator


The harmonic oscillator is a true work horse both in quantum as well as in classical me-
chanics. In this section I will reproduce the all-time-number-one of the quantum mechanics
charts; “how to solve the harmonic oscillator problem with just a bit of operator algebra”.
In one dimension the Hamiltonian of the harmonic oscillator is given by

2
b = pb + 1 mω2 x
H b2 (5.1)
2m 2

99
100 CHAPTER 5. QUANTUM FIELD OSCILLATORS

In order to get rid of the clumsy constants we rescale the operators


³ mω ´ 1
b − 12 b 2
P = (m~ω) pb and Q = b
x (5.2)
~
The Hamiltonian then gets the simple form
³ ´
Hb = 1 Pb2 + Q
b2 ~ω (5.3)
2
x, pb] = i~
From the fundamental commutation relation [b one obtains
h i
b Pb = i
Q, (5.4)

b by a simple
Eqs. 5.3 and 5.4 are sufficient to find the eigenvalues and eigenstates of H
bit of algebra. The first step is to define an operator
1 ³b ´
b
a= √ Q + iPb (5.5)
2
b and Pb are Hermitian operators,
from which it follows, since Q
1 ³b ´
a† = √ Q
b − iPb
2
Note that b a† , so this operator is not an observable. From the commutation relation,
a 6= b
eq. 5.4, it is easy to prove
h i 1 nh b bi h i h i h io
b a† =
a, b Q, Q + i Pb, Q b − i Q,b Pb + Pb, Pb
2
1
= {0 + i · −i − i · i + 0} = 1
2
Furthermore
1 n b2 o ³
bPb + Pb2 = 1 Pb2 + Q
´ h
b2 + i Q,
i
a† b
b a= Q − iPbQ
b + iQ b Pb
2 2 2
The Hamiltonian eqs. 5.3 is then rewritten using eq. 5.4 as
µ ¶
b † 1
H = b
ab a+ ~ω (5.6)
2
h i
b a† = 1
a, b (5.7)

These are the two equations we need. For instance, it is easy to show that
h i n³ ´ ³ ´o
b b
H, a = a†b
b a ba−b a ba† b
a ~ω
n³ ´ ³ ´ o
= ba† − 1 b
ab a− b aba† b
a ~ω = −b a~ω

1
(the constant 2 in the Hamiltonian commutates with everything). From this it follows
that
³ ´
ba = b
Hb a H b − ~ω (5.8)
5.1. THE QUANTUM OSCILLATOR 101

b i.e. H|νi
Here jumps the rabbit out of the high hat. Let |νi be an eigenstate of H, b = ²ν |νi,
then b b
a|νi is also an eigenstate of H with eigenvalue ²ν − ~ω, since
³ ´
b a|νi = b
Hb a H b − ~ω |νi = b
a (²ν − ~ω) |νi = (²ν − ~ω) ba|νi (5.9)

By repeatedly applying ba we can now step down the eigenvalue ladder, i.e. b an |νi is an
b with eigenvalue ²ν − n~ω; simply repeat the rule given by eq. 5.8. This
eigenstate of H
falling down the ladder has to stop for some value of n, when we arrive at the bottom of
the ladder, at the ground state. Moreover, the eigenvalue ²ν − n~ω has to stay positive
and ≥ 12 ~ω

an |νi = |ν 0 i,
Proof: Call b ²ν − n~ω = ²ν 0 , b 0 i = ²ν 0 |ν 0 i. Then b
H|ν a† b
a|ν 0 i =
³ ´ ¡ ²ν 0 ¢ 0
Hb 1 0 1 0 0 a† b
~ω − 2 |ν i = ~ω − 2 |ν i ≡ bν |ν i, according to eq. 5.6. Also hν |b
0 a|ν 0 i = bν 0 hν 0 |ν 0 i.
On the other hand hν 0 |b a† b
a|ν 0 i = hb
aν 0 |b
aν 0 i = kb a|ν 0 ik2 > 0, because the norm of the
vector b a|ν 0 i always has to be positive (one of the axiom’s of an inner product space).
Since hν 0 |ν 0 i > 0 by the same rule, we therefore conclude that bν 0 > 0, or in other words
²ν 0 > 12 ~ω. The only exception is if |ν 0 i = |i, the “null” vector of the vector space, for which
h|i = 0. We can repeat ³ the ´trick with |ν 00 i = b
an−1 |νi and, knowing that b a|ν 00 i = |ν 0 i = |i,
Hb ¡ ²ν 00 ¢
we have b a† ba|ν 00 i = ~ω − 12 |ν 00 i = ~ω − 12 |ν 00 i = |i or ²ν 00 = 12 ~ω.

In conclusion, the only way to end falling down the eigenvalue ladder by repeated
application of the step-down operator b
a is to have some power m of b
a giving the “null”
vector |i, i.e.

am |νi = |i
b (5.10)

(Note this then holds for any power n ≥ m by elementary linear algebra, since any linear
b = |i). The last power (m − 1) which doesn’t
operation maps the null vector onto itself, A|i
give the null vector must be the ground state |0i = b am−1 |νi. According to the proof given
above its eigenvalue must be 12 ~ω. From this ground state |0i we can step up the eigenvalue
ladder again, using the step-up operator b a† . Analogous to eq. 5.8, we can prove
³ ´
b a† = b
Hb a† H b + ~ω

so
µ ¶
b a† |0i = 1
Hb a† |0i
+ 1 ~ωb
2
and
³ ´n µ ¶ ³ ´n
b † 1
H a |0i = n + a† |0i ;
~ω b n = 1, 2, .... (5.11)
2

We have found a way to construct all the eigenstates of the harmonic oscillator (ponder a
bit about why we find all the eigenstates this way). The only remaining thing we have to
take care of is normalization, because we like our eigenstates normalized.
¡ ¢ This we do in the
following way. Suppose |ni, the eigenstate with eigenvalue ²n = n + 12 ~ω is normalized,
then ba|ni = cn |n− 1i steps down the ladder with cn a constant to be determined. Its norm
102 CHAPTER 5. QUANTUM FIELD OSCILLATORS

­ † ® Hb
is, using eq. 5.7, hb
an|b ab
ani = n|b a|n = hn| ~ω − 12 |ni = n hn|ni = n, since by assumption
|ni was normalized. However, we also have hb ani = |cn |2 hn − 1|n − 1i = n. So if we
an|b

choose cn = n, the vector |n − 1i is also normalized.1 In other words

b
a|ni = n|n − 1i (5.12)

In a similar way we can prove



a† |ni = n + 1|n + 1i (5.13)

Using eq. 5.13 we can climb the eigenstate ladder step by step
1 † 1 † 1 ³ † ´2
|1i = √ ba |0i ; |2i = √ b a |1i = √ b
a |0i ;
1 2 2.1
1 † 1 ³ † ´2 1 ³ † ´3
|3i = √ ba |2i = √ b
a |1i = √ b
a |0i
3 3.2 3.2.1
In general, all the eigenstates of the Harmonic oscillator can be expressed as
¡ † ¢n
a
|ni = √ |0i (5.14)
n!

If we start with a normalized ground state |0i then all the states are normalized.

5.1.1 Summary Harmonic Oscillator


The Hamiltonian is given by

2
b = pb + 1 mω 2 x 1 ³ b2 b2 ´
H b2 = P + Q ~ω
2m 2 2
where
³ ´1
b = mω 2 x
1
Pb = (m~ω)− 2 pb Q b
~
define
1 ³b ´ 1 ³b ´
b
a= √ Q + iPb a† = √ Q
b − iPb
2 2
then we can rewrite the Hamiltonian as
µ ¶
b= 1 †
H +b
ab a ~ω
2

h i
b a† = 1
a, b

1 √
We might add a phase factor, i.e. cn = eiαn n ; αn arbitrary. The resulting vector would still be
normalized. We take the liberty to set αn = 0; ∀n
5.1. THE QUANTUM OSCILLATOR 103

b can be labeled
Using only operator algebra we have found that all the eigenstates of H
by a natural number n, for which
b
H|ni = ²n |ni ; n = 0, 1, 2, .....

the eigenvalues are given by


µ ¶
1
²n = n+ ~ω
2

The operators b a† can be used to step-down and step-up the eigenvalue ladder,
a and b
respectively
√ √
b
a|ni = n|n − 1i a† |ni = n + 1|n + 1i
b

In fact, starting from the ground state |0i we can step-up the whole ladder
¡ † ¢n
b
a
|ni = √ |0i
n!

5.1.2 Second Quantization


¡ ¢
It is tempting to interpret the eigenstates |ni and their energies ²n = n + 12 ~ω in the
following way:

• The state |ni contains n identical quanta, each with energy ~ω.

• The operator b a† , which increases n by 1, creates an additional quantum and is thus


called creation operator . Similarly b
a is the annihilation operator since it annihilates
a quantum.
b =b
• The operator N a† b
a is called the number operator since
1
b |ni = b
N a† b a† |n − 1i = n|ni
a|ni = n 2 b

it counts the number of quanta of the state |ni

• The state |0i has no quanta, it is called the vacuum state (or vacuum for short).
Don’t confuse it with the null vector; the state |0i is the ground state of the system,
and thus a non-trivial vector.

• All states can be created by letting the creation operator repeatedly operate on the
vacuum to create quanta of energy ~ω.

The language of b a† , |ni etcetera, is called the language of second quantization. The
a, b
idea is that writing down the Hamiltonian, eq. 5.1, as a quantum operator is “first
quantization”. The language of first quantization can be translated into wave mechanics
using the position representation. The interpretation of the eigenstates of the Hamiltonian
as being composed of n individual but identical quanta is then “second quantization”.
Since |ni contains n quanta, a representation using |ni; n = 0, 1, 2, .... as a basis set is
called the number representation (or also the occupation number representation). It is
104 CHAPTER 5. QUANTUM FIELD OSCILLATORS

tempting to call the quantum of energy ~ω a “particle”. Obviously these “particles” are
indistinguishable . Apparently, an arbitrary number of them can have the same energy
~ω, which means that these particles are bosons. You might think it a bit far-fetched to
call something a “particle” only because it is associated with an energy quantum of ~ω.
That is true; for a simple harmonic oscillator the second quantization point of view is only
language. It does not give anything that cannot be obtained by first quantization or wave
mechanics. However, in the following we encounter examples in which the quantum of
energy ~ω also attains a momentum ~k. Typical experiments as discussed in Chapter 3
can be interpreted in a straight-forward way in terms of “colliding” quanta and applying
the rules of conservation of energy and momentum for these “collisions”. For objects
behaving this way, “particle” is a natural name.2 In the following I will use the phrases
“second quantization” and “(occupation) number representation” interchangeably for the
language, as well as for the basis set.

5.2 The One-dimensional Quantum Chain; Phonons


The first step up in complexity is a familiar example from solid state physics: longitudinal
vibrations in a linear chain. The system is an infinite chain of identical masses m connected
by identical springs with spring constants κ. In equilibrium we have perfect translational
symmetry; the distance between two consecutive masses is a. We denote the displacement
from equilibrium of the n’th mass by un . The linear chain is a simple model for describing
lattice vibrations in a solid, for instance.3 It is shown in Fig. 5.1. The black/gray balls
indicate displaced/equilibrium positions, respectively.

un u n+1 u n+ 2 u n +3 u n+4
k

m
a

Figure 5.1: A linear chain of masses and springs.

The quantum Hamiltonian of this chain is given by

XN
b= pb2n 1
H + κ (b bn )2
un+1 − u with un , pbm ] = i~δ nm
[b (5.15)
n=1
2m 2

An infinite chain presents some subtle mathematical hurdles, so we simplify the problem
by taking N masses (where N is large, but finite). To get rid of boundary effects, it is
custom to connect the first mass 1 to the last mass N by a spring. Or in more fancy
2
In classical mechanics our objects are particles. Wave mechanics (first quantization) makes them
into waves. It seems that second quantization turns them into particles again. However, these quantum
particles are not classical objects like point masses! They have energy, momentum and, depending upon
the particle, other properties like mass and spin, but in contrast to classical particles they can be created,
annihilated or transformed. Maybe calling them quanta would have been better after all, but alas....
3
The extension to three dimensions is discussed later. The mathematics becomes more complicated,
but the essential physics remains the same as in the one-dimensional model.
5.2. THE ONE-DIMENSIONAL QUANTUM CHAIN; PHONONS 105

terms, one applies periodic boundary conditions, i.e. u bN+1 = u b1 , etcetera. This trick is
similar to the one we used in Section 4.2; in the end we can always take limN →∞ . We
rewrite the Hamiltonian a bit using (b
un+1 − u bn )2 = P
b2n + u
u b2n+1P− 2b bn+1 . Since we have
un u
applied periodic boundary conditions it follows that n u 2
bn = n u 2
bn+1 (the sums are over
all masses). Defining
r

ω0 = (5.16)
m
we get the form

XN 2 µ ¶
b= b
pn 1 2 2 1 1
H + mω0 u bn − u bn+1 − u
bn u bn
bn+1 u (5.17)
n=1
2m 2 2 2

bn+1 and u
Since u bn commutate: [b bj ] = 0 and [b
ui , u pi , pbj ] = 0 ∀i,j , we may freely interchange
their order. I prefer to use this symmetric form. You have solved this problem a couple of
times before in classical mechanics and solid state physics, where it leads to normal modes
and phonons. As a quantum problem, it is our first many body problem. In principle, one
could try and solve this problem head-on in the position representation. We then need a
wave function Φ that keeps track of the positions un of all the N masses involved. As
usual we would set pbn → ~i ∂u∂n ; ubn → un and the Schrödinger equation becomes
" N µ ¶#
X ~2 ∂ 2 1 1 1
− 2
+ mω 20 u2n − un un+1 − un+1 un Φ (u1 , u2 , u3, ......, uN−1 , uN , t)
n=1
2m ∂u n 2 2 2

= i~ Φ (u1 , u2 , u3, ......, uN−1 , uN , t) (5.18)
∂t
Needless to say that blindly trying to solve this N -dimensional differential equation is a
hopeless exercise, even on a computer (unless N is small).

We can however reformulate the problem in terms of a new set of particles, called
phonons. As by magic, our many-body problem will decouple into a set of one-particle
problems.4 Each phonon will involve all the old particles of Fig. 5.1. In some books
the phonon is called a collective excitation; others call it a quasi-particle, analogous to
the term used in Chapter 4. Here is how the mathematics goes; we define new operators
pbk , qbk
N−1
1 X −ik` na
pbk` = √ e pbn
N n=0
N−1
1 X −ik` na
qbk` = √ e bn
u (5.19)
N n=0


where k` = ` Na ; ` = 0, 1....N − 1 and a is equilibrium distance between masses, see Fig.
5.1. You will recognize this as a discrete Fourier transform. Using the familiar relations
4
Of course, magic has very little to do with it. We use a transformation which decouples a many coupled
harmonic oscillator problem into uncoupled normal mode oscillations, as in classical mechanics.
106 CHAPTER 5. QUANTUM FIELD OSCILLATORS

for discrete Fourier transforms


N
X −1
0
ei(k−k )na = N δ k,k0
n=0
N−1
X 0
eik` (n−n )a = N δ n,n0 (5.20)
`=0

we can invert the relations of eq. 5.19

N−1
1 X ik` na
bn =
u √ e qbk`
N `=0
N−1
1 X ik` na
pbn = √ e pbk` (5.21)
N `=0

The Hamiltonian of eq. 5.17 can be rewritten as (dropping the ` subscripts, and using a
notation where k denotes one of the allowed k` values)

X pb† pbk µ ¶
b= k 1 2 † 1 ika 1 −ika
H + mω 0 qbk qbk 1 − e − e (5.22)
2m 2 2 2
k

Since 12 eika + 12 e−ika = cos ka and defining


r
κ 11
ω (k) = ω 0 (1 − cos ka) = 2 | sin ka|
2 (5.23)
m 2

we can write
X pb† pbk 1
b=
H k
+ mω (k)2 qbk† qbk (5.24)
2m 2
k

Comparing this to eq. 5.1 we see that it is in fact a sum of independent harmonic oscilla-
tors, each labeled by k, with a characteristic frequency given by ω (k). Since qbk`+N = qbk`
and pbk`+N = pbk` , cf. eq. 5.19 we have a periodicity also in k space. We have some

freedom in choosing the interval over which we let the independent k` = ` Na run. In solid
state physics it is custom to use ` = −N
2 + 1, ....., N
2 , such that − π
2 < k ≤ π
2 this defines
;
the so-called Brillouin zone.5 The relation between the frequency ω of an oscillator and
k (which turns out to be a wave number) is called a dispersion relation. The one given
by eq. 5.23 is shown in Fig. 5.2 (assuming that N is very large, so we may plot it as a
continuous function).

We have solved the harmonic oscillator in the previous section; nothing much changes
since all the oscillators are independent. We can work through it again, and add an index
5
Had we chosen as our interval k`0 ; ` = 0, 1....N − 1 then 0 ≤ k0 < 2π. Because of the periodicity in k
space the index k`0 would run over the same values as the index k` , and the physical results would be the
same. We prefer the more symmetric interval − π2 < k ≤ π2 .
5.2. THE ONE-DIMENSIONAL QUANTUM CHAIN; PHONONS 107

ω(k ) κ
2
m

π π
− k
2 2


Figure 5.2: Dispersion relation of a linear chain ω (k) = 2 m | sin 12 ka|.

k to any Pb, Q,
b ω, b a† , n encountered. So in second quantization language the Hamiltonian
a, b
becomes
X µ1 ¶
b
H= +b †
ak b
ak ~ω (k) (5.25)
2
k

with commutation relations


h i h i
b a†k0 = δ kk0 ; [b
ak , b ak , b a†k , b
ak0 ] = b a†k0 = 0 (5.26)

Since the oscillators which belong to different k’s are independent, their operators com-
mutate. Each k-oscillator has its own eigenstates (nk = 0, 1, ...). The eigenstates of the
full Hamiltonian consists of products; see Section 1.4
Y
|nk1 i|nk2 i.....|nkN i = |nk1 nk2 .....nkN i ≡ |nk i ≡ | {nk }i (5.27)
k

The eigen-energies are given by


Hb | {nk }i = E{n } | {nk }i
k
Xµ 1

where E{nk } = nk + ~ω (k) (5.28)
2
k

The state | {0}i with no quanta in any of the k-oscillators is of course called the vacuum state.
All other states can be formed by creating a number of quanta, applying creation operators
for each k
³ ´nk
Y b a†k
| {nk }i = √ | {0}i (5.29)
k
nk !
The quanta are called phonons. In second quantization interpretation, phonons are con-
sidered to be particles. In this case each phonon can be labeled by a wave number
k ; − π2 < k ≤ π2 . A k-phonon has an energy ²(k) = ~ω (k). Obviously k is a quantum
b such that
number; one can define an operator K,
à !
X
b {nk }i =
~K| ~knk | {nk }i (5.30)
k
108 CHAPTER 5. QUANTUM FIELD OSCILLATORS

One can also define operators b b2 , ....b


k1, k kn operating on a single k-phonon state, i.e.
b
~k1 |nk1 i = ~k1 nk1 |nk1 i etcetera. Obviously one has
X
b =b
K k1 + b k2 ..... + b
kN = b
k (5.31)
k

Analyzing processes in which phonons interact with other particles like neutrons, electrons,
or photons, we can assign to each phonon a momentum p = ~k. (We will discuss this in
more detail later on). The operator ~b k is then the momentum operator for k-phonons;
~Kb is the total momentum operator. In contrast to the simple harmonic oscillator in
Section 5.1, the quanta (phonons) of the chain have a well-defined momentum as well as
an energy; reason enough to call them particles. P{nk } = h{nk } |~K| b {nk }i = P ~knk
k
is obviously the total momentum of the state | {nk }i. One can easily check that the
states | {nk } (t)i = exp(− ~i E{nk } )| {nk }i with the total energy E{nk } are solutions of the
Schrödinger equation
∂ b {nk } (t)i
i~ | {nk } (t)i = H|
∂t
So by choosing a convenient representation, the occupation number representation, we have
been able to solve the Schrödinger equation in a fairly straight-forward way and avoid the
problems, had we tried to solve the same equation in the position representation, see eq.
5.18.6

5.3 My First Quantum Field


“Het gaat altijd om tijd en ruimte. Dus moet je zorgen dat het middenveld goed in evenwicht is”,
J. Cruijff over veldentheorie. (“The key issues are always time and space. Thus you have to make
sure that the midfield is always well-balanced”, the Dutch philosopher J. Cruijff on field theory.)

In the foregoing section we extended the techniques developed for the single quantum
harmonic oscillator in Section 5.1 to the case of many coupled harmonic oscillators. The
most important trick was the transformation of eq. 5.19; from there on the system de-
coupled into simple harmonic oscillators again. Imagine that the masses are not point
masses like in Fig. 5.1, but they are spread out in a uniform way and constitute an elastic
medium. Each mass element of the medium can have a displacement from its equilibrium
position. The displacement u does not any longer depend upon a discrete index i, which is
associated with discrete (point) masses, but it depends upon a continuous variable, namely
the position x along the chain, i.e. u(x). A continuous physical function like u(x) is called
a field. The position x labels the degrees of freedom of the system. A field thus has an
infinite, continuous set of degrees of freedom, in contrast to the discrete chain, where i
might run to infinity, but at least you can count the degrees of freedom one by one. An
artists impression of the displacement field is given in Fig. 5.3.
We want to give a quantum mechanical description of a field. Since the “infinities” of
a field are a bit tricky to handle mathematically, I will develop the theory of the classical
field of the elastic medium first, before I come to the quantum field description.
6
If you really want to, it is possible to construct wave functions from the states | {nk }i. It would involve
Hermite polynomials, etcetera, familiar to you from elementary quantum mechanics. However, this would
not add anything new to the physics. On the contrary, it makes things less transparent.
5.3. MY FIRST QUANTUM FIELD 109

u ( xn ) u ( xn +1 ) u ( xn + 2 ) u ( xn +3 ) u ( xn + 4 )

Figure 5.3: An elastic medium with displacements u at points xm (top figure). Artist’s
impression of continuum (bottom figure).

5.3.1 Classical Chain


The classical Hamiltonian which corresponds to eq. 5.15 is given by
X p2 1 X1 1
Hclass = n
+ κ (un+1 − un )2 = mu̇2n + κ (un+1 − un )2 (5.32)
n
2m 2 n
2 2

The classical equations of motion which correspond to this Hamiltonian are given by7

d2 un 1
2
= − ω20 (2un − un+1 − un−1 ) ; n = 1, 2, ..., N (5.33)
dt 2
We have already found the way to solve this set of N coupled differential equations. One

defines normal coordinates qk` according to eq. 5.19, with k` = ` Na and one chooses
−N N π π
` = 2 + 1, ...... 2 , such that − 2 < k ≤ 2 runs over the Brillouin zone. The classical
Hamiltonian which corresponds to eq. 5.24 is then given by
X p∗ pk 1
Hclass = k
+ mω (k)2 qk∗ qk (5.34)
2m 2
k

The normal coordinates behave as independent harmonic oscillators. The dispersion rela-
tion ω (k) is given by eq. 5.23. Using eq. 5.21 it is easy to show that the classical equations
of motion, eq. 5.33, become
−N N
q̈k` = −ω (k) qk` ; ` = + 1, ...... (5.35)
2 2
with the general solution

qk` (t) = ak` ,0 e−iω(k)t + ak` ,1 eiω(k)t (5.36)

The (complex) constants ak` ,0;1 give the phase and amplitude, which can be determined
from the initial conditions. The full solution is obtained from eq. 5.21
N
1 X
2

un (t) = √ eik` na qk` (t) (5.37)


N
`= −N
2
+1
P
7 ∂V
The right handside is obtained from − ∂u n
, where V = 14 mω20 n (un+1 − un )2 . Remember we use
periodic boundary conditions, such that u0 = uN ; u1 = uN +1 , etcetera
110 CHAPTER 5. QUANTUM FIELD OSCILLATORS

Obviously these displacements have to be real numbers, i.e. u∗n = un , from which one
derives the relation ak` ,1 = a∗−k` ,0 , so eq. 5.36 becomes

qk` (t) = ak` ,0 e−iω(k)t + a∗−k` ,0 eiω(k)t (5.38)

Each k` labels a so-called normal mode of the chain (or simply mode for short).8 The
displacements ui of all the masses along the chain in case only a single (normal) mode is
excited can be found by setting qk` (t) = 0 for all k` except one. One then obtains wave-like
displacement pattern eik` ia qk` (t) and the most general solution is a linear combination of
such waves. One reserves the name normal mode for the spatially dependent part eik` na ;
the (harmonic) time-dependent qk` (t) part incorporates the phase and amplitude of the
wave.

5.3.2 Continuum Limit: an Elastic Medium


We now take the continuum limit of this system. Imagine that the masses are not point
masses like in Fig. 5.1, but that they are spread out, like in Fig. 5.3. In equilibrium we
have a constant mass density µ = m/a. We denote the position of a mass element along the
continuous chain by x. Each mass element can have a displacement. The displacement
u does not any longer depend upon a discrete index n, which is associated with point
masses, but it becomes a variable u(x), which depends continuously upon the position
along the chain x. In the discrete case u is only defined for the discrete set of points
xn = na ; n = 1, 2..., N ; i.e. u(xn ) = un . A continuous physical variable like u(x) is called
a field, and since we are doing classical mechanics for the moment, it is a classical field.
The equation of motion which governs this classical field is easily obtained by taking the
continuum limit of eq. 5.33. One uses on the right handside

−[2un − un+1 − un−1 ] = [un+1 − un ] − [un − un−1 ] =


( u(x +a)−u(x ) u(x )−u(x −a) )
n n n n

[u(xn+1 ) − u(xn )] − [u(xn ) − u(xn−1 )] = a2 a a
a

The expression between {} is recognized as the discrete approximation to the second


2 u(x)
derivative ∂ ∂x2 and the latter is obtained by taking the continuum limit lima→0 of this
expression. The prefactor on p the right handside of eq. 5.33 now is (aω 0 )2 . For discrete
point masses we defined ω 0 = 2κ/m by eq. 5.16. The spring constant κ of a spring scales
with its length a as κ = τ /a with τ a constant determined by the materials properties of
the spring, which is called the p
“elasticity” or “tension”of the spring. With a mass density
µ = m/a we can write ω0 = a1 2τ /µ. Now define the constant v0 = aω 0 or
p
v0 = 2τ /µ (5.39)

in terms of the materials properties of the continuum. Note that v0 has the dimension of
£ ¤1
velocity (m/s) = kgm/s2 · m/kg 2 . The continuum limit of eq. 5.33 thus becomes

∂ 2 u(x, t) 2
2 ∂ u(x, t)
= v0 (5.40)
∂t2 ∂x2
8 √1
We absorbed the prefactor N
into the phase/amplitude factors ak` ,0 .
5.3. MY FIRST QUANTUM FIELD 111

The time and position dependence of the displacement field of a continuous chain thus
obeys a wave equation. The continuous chain is of course nothing else than a homogeneous
elastic medium, and eq. 5.37 describes the elastic waves (or sound waves). The constant
v0 is the speed of sound in the medium. The general solution of the wave equation can be
found by taking the continuum limit of eq. 5.37. We use a trick similar to eq. 4.28; the
spacing between our discrete k values was ∆k = 2π L , where L = N a is the total length of
the medium. We take L large such that ∆k becomes small enough to take the integral
X Z
1 X L
... = ...∆k −→ ...dk
∆k 2π
k k

Incorporating the prefactor into a normalization constant N to be determined later on,


we get as the general solution of the wave equation
Z ∞
u(x, t) = N eikx q(k, t)dk (5.41)
−∞

where, according to eq. 5.38

q(k, t) = a(k)eiω(k)t + a(−k)∗ e−iω(k)t (5.42)

We recognize as a continuous superposition of waves. As usual, the phases and amplitudes


a(k) (now a function of the wave number k), are determined by initial conditions. Again
the name mode or normal mode is reserved for the spatial dependent part eikx . The
function q(k, t) replaces the collection of amplitudes and phases of the discrete case; one
can also say that since we have an infinite system, there are infinite degrees of freedom
which can be labeled by k. Trying out eq. 5.41 in eq. 5.37 gives
Z ∞ Z ∞
(±iω (k))2 eikx q(k, t)dk = v02 (ik)2 eikx q(k, t)dk
−∞ −∞

In order that this is true for any a(k) we must have the following dispersion relation

ω (k) = v0 k (5.43)

The constant v0 thus determines the speed at which the sound waves travel, as we have
foreseen. The total energy H for the discrete chain is given by
X1 1
H= mu̇2n + κ (un+1 − un )2
n
2 2

We use the by now familiar tricks


Pto go to the
P continuum limit. The spacing between the
lattice sites n was ∆x = a. Use n ... = a1 n ...∆x and un+1 − un = a( u(xn +a)−u(x
a
n)
) →
∂u(x)
a ∂x . Furthermore, use Pthe definitions of mass Rdensity µ = m/a and elasticity τ = κa
and transform the sum n ...∆x into an integral ...dx
Z ∞ " µ ¶ µ ¶ #
1 ∂u(x, t) 2 1 ∂u(x, t) 2
H= dx µ + τ (5.44)
−∞ 2 ∂t 2 ∂x
112 CHAPTER 5. QUANTUM FIELD OSCILLATORS

This is the classical energy of an elastic medium of length L. The expression can be
transformed to an integral over k by using of eq. 5.41. For instance, defining q̇(k, t) =
∂q(k,t)
∂t
Z ∞ Z ∞ Z ∞ Z ∞
∂u(x, t) ∂u(x, t) 0
dx = N2 eikx q̇(k, t)dk ·
dx eik x q̇(k0 , t)dk0
−∞ ∂t ∂t
Z−∞∞
−∞
Z ∞−∞
0
= N2 dk 0 dk q̇(k 0 , t)q̇(k, t) dx ei(k+k )x
Z−∞∞
−∞

= N2 dk 0 dk q̇(k 0 , t)q̇(k, t) 2πδ(k + k0 )


−∞
Z ∞ Z ∞
2 2
= 2πN dk q̇(−k, t)q̇(k, t) = 2πN dk q̇∗ (k, t)q̇(k, t)
−∞ −∞

1
R∞ ikx . The last line fol-
where we used the expression for a δ-function, δ(k) = 2π −∞ dx e

lows from q(−k, t) = q (k, t), which can be derived from eq. 5.41 by noting that the
displacement field must be real, i.e. u∗ (x, t) = u(x, t). In a similar way one can show
Z ∞ Z ∞
∂u(x, t) ∂u(x, t)
dx = 2πN 2 dk k2 q ∗ (k, t)q(k, t)
−∞ ∂x ∂x −∞

Moreover, since the time dependence is simply given by eq. 5.42, we have q ∗ (k, t)q(k, t) =
q ∗ (k, 0)q(k, 0). In the following we write q(k) ≡ q(k, 0) simplify the notation. We can now
rewrite the energy as
Z · ¸
L ∞ 1 ∗ 1 2 ∗
H= dk µ q̇ (k)q̇(k) + τ k q (k)q(k)
2π −∞ 2 2

where we have chosen the convenient normalization constant



L
N = (5.45)

This expression is purely in the amplitude/phase factors q(k). Using the dispersion rela-
tion, eq. 5.43, and eq. 5.39 and defining a “momentum” by

p(k) = µq̇(k) (5.46)

we obtain the final result


Z ∞ · ¸
L p∗ (k)p(k) 1
H= dk + µω (k)2 q ∗ (k)q(k) (5.47)
2π −∞ 2µ 2

This looks fantastically like the continuum analogue of eq. 5.34, which also explains why
I have chosen the symbol “H” to represent the energy. It is nothing else than the classical
Hamiltonian of an elastic medium.

SUMMARY
Table 5.1 gives a short summary of the similarities and differences between a classical
discrete harmonic chain and an elastic continuum.
5.3. MY FIRST QUANTUM FIELD 113

Property Discrete Chain Elastic medium


degrees of freedom countable set; uj ; j = 1, 2... continuous field u(x); −∞ < x < ∞
equations of motion Newton’s equations, eq. 5.33 Wave equation, eq. 5.40
normal modes eikja ; j = 1, 2...; − π2 < k ≤ π2 eikx ; −∞ < x, k < ∞
Hamiltonian eqs. 5.32,p5.34 eqs. 5.44,
p5.47
dispersion relation ω (k) = 2k/m | sin 12 ka| ω (k) = 2τ /µ k

Table 5.1: Comparison between a discrete classical chain and an elastic medium

5.3.3 Quantizing the Elastic Medium; Phonons


The way we have analyzed the classical field of the elastic medium now makes it easy
to present the corresponding quantum field. The classical Hamiltonian H of eq. 5.47 is
simply transformed into a quantum mechanical Hamiltonian H b by defining the operators
qb(k) and pb(k) and assuming that these have familiar commutation relations
Z ∞ · † ¸
b = L pb (k)b
p(k) 1
H dk + µω (k)2 qb† (k)b q(k) (5.48)
2π −∞ 2µ 2
h i ¡ ¢ £ ¤ £ ¤
qb† (k), pb(k 0 ) = i~δ k−k 0 ; qb(k), qb(k0 ) = pb(k), pb(k0 ) = 0

As before, the problem can be solved using the second quantization formalism and
defining creation/annihilation operators ba† (k) and b a(k) in full analogy to eqs. 5.25—5.29.
The Hamiltonian becomes
Z ∞ · ¸
Hb = L dk b a† (k)b
a(k) +
1
~ω (k) with
2π −∞ 2
h i £ ¤ h † i
b a† (k0 ) = δ(k − k0 ) ; b
a(k), b a(k), ba(k 0 ) = b a† (k 0 ) = 0
a (k), b (5.49)

Note that one can also substitute the classical amplitudes/phases a(k) in eq. 5.42 by
annihilation operators ba(k) (and their complex conjugates a∗ (k) by creation operators
a† (k)) and obtain exactly the same results!!
b
µ ¶1 h i
~ 2
qb (k, t) = a (k) e−iωt + b
b a† (−k) eiωt (5.50)
2µω
You have to take into account the correct prefactors of eq. 5.2 to be consistent with
eq. 5.5. In case you are wondering where the time factors go, we are actually following
the Heisenberg route to quantization in which operators acquire a time dependence, see
Section 2.6. One can switch to the ordinary Schrödinger picture, by simply setting t =
t0 = 0, cf. eq. 2.64. The eigenstates of this Hamiltonian can be written as | {n(k)}i.
Note that each wave number k defines an independent harmonic oscillator with states
|n(k)i; n(k) = 0, 1, .... The eigenstates of the full Hamiltonian consist of products of such
states for all k, see eq. 5.27. Since k is a continuous variable, the notation is a bit difficult
but the idea should be clear.
b | {n(k)}i = E{n(k)} | {n(k)}i
H
Z · ¸
L ∞ 1
where E{n(k)} = dk n(k) + ~ω (k) (5.51)
2π −∞ 2
114 CHAPTER 5. QUANTUM FIELD OSCILLATORS

Note that E has the dimension of energy here, which is consistent. The quanta which are
a† (k) are of course called phonons. The term is even better at its place here
created by b
than in Section 5.2, since the Greek ϕoνoς = sound, and a phonon thus is a ‘quantum
of sound’. They can be considered as particles since they have a well-defined momentum
p = ~k and a well-defined energy ²(k) = ~ω (k). All the quantum states of the elastic
medium can be constructed in a systematic way by applying creation operators b a† (k) on
the vacuum state | {0}i, the ground state of the system without any phonons present. In
other words, all states of the quantum field can be constructed by creating phonons in the
vacuum.

CLOSING REMARKS
Using the second quantization formalism, the quantum field problem of a continuous
elastic medium is no more difficult to solve than the discrete quantum chain of Section
5.2. However, there is a qualitative difference. For the discrete chain it is in principle
still possible to try and solve the problem in the position representation (i.e. in terms
of wave functions, cf. eqs. 5.15—5.18); it is a stupid approach, but, be my guest. The
second quantization formalism is merely the fastest way to solve the problem. For the
discrete chain, the relation between the classical Hamiltonian, eq. 5.34 and its quantum
mechanical counter part, eq. 5.15, is straight-forward. Not so for the continuous elastic
medium ! Its classical Hamiltonian is given by eq. 5.44. If I only had this form, it is
not immediately clear how to transform this expression into a quantum mechanical form.
We could derive its quantum Hamiltonian by first quantizing the discrete chain and then
taking the continuum limit. In other words, the paths described by the single arrows in
the table below represent well-defined procedures. With that help we can also define a
procedure for path represented by the double arrow in this table.

classical discrete chain, Section 5.3.1 → elastic medium, Section 5.3.2


↓ continuum ⇓
quantum discrete chain, Section 5.2 → elastic medium, Section 5.3.3

I simply rewrite eq. 5.44 in terms of normal coordinates and the associated ampli-
tude/phase factors and their conjugate momenta, as in eq. 5.47. The latter form can
directly be translated into a quantum mechanical expression, eq. 5.48, assuming that
these amplitude/phase factors and momenta can be interpreted as quantum mechanical
operators with the usual commutation relations. In view of the table above we know that
this is a correct procedure. We then automatically obtain an expression within the second
quantization formalism.

You might wonder why I make so much fuss about a continuous medium. After all, on a
microscopic scale a crystal is not a continuous medium, but is closer to the quantum chain
of Fig. 5.1. A continuous medium can only be an approximation which is appropriate
for waves with long wave lengths (small wave numbers k). That is certainly true, but the
elastic medium is the simplest example of a field. The way we approached its quantum
problem guides us when we have to deal with other fields for which there is no physical
equivalent of a discrete chain, or, in other words, we only have the right column of the table
5.4. THE THREE-DIMENSIONAL QUANTUM CHAIN 115

above. A non-trivial example of the latter will be given in the final section of this chapter,
namely the quantization of the electro-magnetic field, or quantum electrodynamics.9 For
such fields the second quantization formalism is not merely the fastest way to approach the
corresponding quantum field problem, it is the only way ! For such quantum fields it is no
longer possible to define a “position” representation; the occupation number representation
is all we have got.

5.4 The Three-dimensional Quantum Chain


5.4.1 Discrete Lattice
The previous two Sections 5.2 and 5.3 have been concerned with one-dimensional problems.
It is tedious, but not difficult in principle to extend the discussion to the three-dimensional
case. The 3-dim chain is a good model to describe phonons in a simple crystal with one
atom per unit cell.10 Instead of the expression given in eq. 5.19, one obtains similar
three-dimensional expressions
1 X −ik·R
pbk,s = √ e bR
εk,s · p
N R
1 X −ik·R
qbk,s = √ e bR
εk,s · u (5.52)
N R

It is assumed that the masses form a three-dimensional lattice, where R =la+mb+nc;


l, m, n = 0, 1... are the lattice vectors describing the positions of the atoms in equilibrium.
The atoms interact with their neighbors via harmonic potentials (the “springs”; if the
displacements from equilibrium are small enough we may always approximate the full
potential by an harmonic potential). uR = (ux , uy , uz )R is the displacement vector of
the atom at the lattice position R; pR = (px , py , pz )R is its corresponding momentum. k
is a vector in reciprocal space belonging to the first Brillouin zone. The vectors εk,s =
(εx , εy , εz )k,s ; s = 1, 2, 3; |εk,s |2 = 1 are called the polarization vectors. In a cubic
crystal, there is one longitudinal mode εk,1 kk and two transversal modes εk,2 , εk,3 ⊥k (
and εk,2 ⊥εk,3 ). In an anisotropic crystal, the relation between the direction of k of the
wave and the polarization vectors can be quite complicated.
Eq. 5.24 is replaced by

X pb†k,s pbk,s 1
b=
H †
+ mω s (k)2 qbk,s qbk,s (5.53)
2m 2
k,s

9
A general method to quantize classical fields exists, not based upon normal modes. The first step is
to formulate a “classical mechanics” for fields within the Lagrangian formalism. A Hamiltonian is then
defined in the usual way via canonical momenta and coordinates. This is then quantized by turning the
latter into operators, assuming the usual commutation relations. This, in a nutshell is “quantum field
theory”. Useful references are: H. Goldstein, Classical Mechanics, (Addison-Wesley, Reading, 1980); for
classical fields, and W. Greiner, J. Reinhardt, Field Quantization (Springer, Berlin, 1996) for quantum
fields.
10
The case of multiple atoms per unit cell is not much more difficult to treat, but there the notation
becomes really messy. If you want to know more about it, you should consult solid state physics books. A
good reference is: N. W. Ashcroft and N. D. Mermin, Solid State Physics, (Holt-Saunders, Philadelphia,
1976).
116 CHAPTER 5. QUANTUM FIELD OSCILLATORS

where the sum over k covers the first Brillouin zone and s = 1, 2, 3 sums over the polariza-
tions. The dispersion relations ω s (k) can be calculated straight-forwardly, once the lattice
(fcc, bcc, hcp, etcetera all give different dispersion relations) and the interaction potential
between the atoms (the “spring” constants) are known.11 From here on we can switch to
second quantization language and simply copy all the steps after eq. 5.24 to obtain the
phonons of the three-dimensional lattice. The Hamiltonian is rewritten as
X µ1 ¶
b
H = +b †
ak,sb
ak,s ~ωs (k) with
2
k,s
h i £ ¤ h † i
a†k0 ,s0 = δ k,k0 δ s,s0 ; b
ak,s , b
b ak0 ,s0 = b
ak,s , b a†k0 ,s0 = 0
ak,s , b (5.54)

The eigenstates |{nk,s }i can be constructed systematically by operating with the creation
operators ba†k,s on the vacuum state |{0}i. The operator b
a†k,s creates a phonon with polar-
ization s, momentum p =~k, and energy ²s (k) = ~ωs (k).

5.4.2 Elastic Medium


As in Section 5.3 we can take the continuum limit to describe (quantized) sound waves
in an elastic medium. A homogeneous and isotropic medium can be characterized by a
speed of sound for longitudinal and transversal waves vl , vt , respectively. The classical
wave equations in three dimensions, equivalent to eq. 5.40 are
∂ 2 us (r, t)
= vs2 ∇2 us (r, t) (5.55)
∂t2
where
P the subscript s indicates longitudinal or transversal. The total displacement field
s us (r, t) ≡ u(r, t) = (ux (r, t), uy (r, t), uz (r, t)) is a three dimensional vector, where each
component depends upon the position in space r =(x, y, z) and time t. The transition to
quantum mechanics follows exactly the procedure outlined in Section 5.3 in one dimension,
where we assign different polarization directions to the one longitudinal and two transversal
modes, as for the isotropic (cubic) crystal. So the equivalent of eq. 5.41 in three dimensions
becomes
√ Z
Ω X ∞ ik·r
u(r, t) = e εk,s qs (k, t)d3 k (5.56)
(2π)3 s −∞

with ω s (k) = vs k, as before. The normalization constant is the three-dimensional analogue


of eq. 5.45, where Ω = L3 is some convenient normalization volume. The spatial dependent
functions εk,s eik·r labeled by s = 1, 2, 3 and k are again called the modes of the system.
Per k there is one longitudinal mode εk,1 kk and two transversal modes εk,2 , εk,3 ⊥k ( and
εk,2 ⊥εk,3 ) with speeds vl , vt , respectively.12 The quantum mechanical operators qbs (k) and
pbs (k) equivalent to eq. 5.52 can be written as
Z ∞
1
pbs (k) = √ b (r) d3 r
e−ik·r εk,s · p
Ω −∞
Z ∞
1
qbs (k) = √ b (r) d3 r
e−ik·r εk,s · u (5.57)
Ω −∞
11
See Ashcroft & Mermin.
12
This is only true in an isotropic medium. In an anisotropic medium polarizations become much more
complicated. Also the speed of sound is different for different directions in the medium.
5.4. THE THREE-DIMENSIONAL QUANTUM CHAIN 117

and the Hamiltonian equivalent to eqs. 5.48 and 5.53


" #
Ω XZ ∞ b
p†
s (k)b
ps (k) 1
Hb= d3 k + µω (k)2 qbs† (k)b
qs (k) (5.58)
(2π)3 s −∞ 2µ 2

Again, referring to the closing remarks of the previous section, a continuous medium is
only an approximation for a real crystal in the long wave length limit (i.e. for small k),
but the procedure will guide us in quantizing the electro-magnetic field in vacuum which
we will discuss in the next section (the vacuum is a continuous “medium”).

5.4.3 Are Phonons Real Particles ?


For those of you who think that the “particles” obtained by second quantization are merely
a mathematical construction, let me try and convince you a bit that these are real particles.
Go back to the typical experiment of Fig. 3.1 and let us shoot “real” particles at a crystal.
If we wish to observe phonons, an incoming beam of neutrons is convenient. Since neutrons
are neutral, they are not hindered by the strong Coulomb fields inside a crystal, and they
can easily penetrate. Because they have a mass which is within a few orders of magnitude
of the mass of an atomic nucleus, they are ideal for transferring momentum and energy
in a collision with the latter. Experimentally we can send in a beam of neutrons having
Pi2
a near monochromatically well-defined momentum Pi and energy ²i = 2M . Of all the
f P2
neutrons coming out again we can analyze their momenta Pf and energies ²f = 2M . We
can calculate the transition probabilities Pi , ²i → Pf , ²f using the methods explored in the
previous chapters. Fermi’s golden rule, which turns out to be quite accurate for this case,
leads to following selection rules. If the scattering is inelastic, i.e. ²f 6= ²i , the transition
probabilities between the states characterized by Pi , ²i and Pf , ²f are zero, unless13

²f − ²i = ±~ω s (k) (5.59)


Pf − Pi = ~k

The first selection rule can be interpreted as the conservation of energy and the second rule
as the conservation of momentum, if we re-interpret the inelastic scattering of neutrons by
crystals in simple physical terms. Consider the “−” in eq. 5.59 first. In comes a neutron
of momentum Pi and energy ²i ; out goes a neutron of momentum Pf and energy ²f and
a particle with momentum ~k and energy ~ω s (k), such that the elementary conservation
laws of energy and momentum are obeyed. The event is shown schematically on the left
side of Fig. 5.4.
In the last particle we recognize our phonon of course. The incoming neutron creates
a phonon which takes a part of its energy and momentum. The “+” in eq. 5.59 then
corresponds with an event in which a neutron collides with a phonon, absorbs it and takes
up its energy and momentum. It is shown on the right side of eq. 5.59. Quantum particles
are not mere mathematical constructs ! You don’t have to call the quantum (phonon) of
the harmonic oscillator chain a “particle” if you don’t want to, but if you do, it enables
you to interpret the results of inelastic scattering experiments in terms of simple, almost
13
See, for instance, Ashcroft & Mermin. I have cheated a little bit; a momentum ~K, where K is a
reciprocal lattice vector might be added. It leads to selection rules which are a bit more complicated. Such
details are best left to solid state physicists.
118 CHAPTER 5. QUANTUM FIELD OSCILLATORS

p f ,ε f p f ,ε f

pi , εi pi , εi

=k , =ω s (k ) =k , =ω s (k )

Figure 5.4: Scattering of a neutron and emission (left) or absorption (right) of a phonon.

classical, collisions between particles.14 Yet, admittedly, the phonon is a strange particle.
As remarked below eq. 5.18, a phonon involves a single mode, which is an oscillation in
the complete quantum chain (i.e. it involves all the masses in the quantum chain). The
phonon particle therefore is a completely delocalized phenomenon. As we will see, this is
also true for other such boson particles, such as photons or plasmons. Some people refuse
to call these “particles”; they use the phrase “collective excitations” instead. It is fine by
me; the physics stays the same.

5.5 The Electro-Magnetic Field in Vacuum

“We all agree that your theory is crazy, but is it crazy enough?”, N. Bohr.

A complete theory of “quantum electro-dynamics” incorporates the interaction of


the electro-magnetic (EM) field with charges and spins (of electrons, protons, etcetera)
and is quite involved.15 ,16 The theory for the EM field in vacuum, without any charges
or currents present, is much easier and I will reproduce a version of it here. In fact, the
procedure followed in the “quantization of the elastic medium” in Section 5.3 can be copied
for the quantum theory of the vacuum EM field. We begin with a short review of classical
electro-magnetism.

14
Although meant to illustrate the real scattering event, Fig. 5.4 almost looks like a set of Feynman
diagrams. This is no accident; these figures can indeed be used to calculate transtitions probabilities within
Fermi’s golden rule, see also Figs. 2.4 and 3.2.
15
The books mentioned in the footnote at the end of section 5.3 are a useful reference. If you are
interested in the details of quantum electro-dynamics, you can have a look at the two comprehensive books:
C. Cohen-Tannoudji97 , J. Dupont-Roc, G. Greenberg, Photons and Atoms and Atom-Photon Interactions
(Wiley, New York, 1998), known as the green book and the blue book, respectively.
16
A wonderfull little booklet, written for the layman (no mathematics !), by one of the founding fathers
of quantum electro-dynamics is: R. P. Feynman, QED, the strange theory of light and matter. A must-have
for every physicist!!
5.5. THE ELECTRO-MAGNETIC FIELD IN VACUUM 119

5.5.1 Classical Electro-Dynamics


The classical electro-magnetic field is given by Maxwell’s equations, which in vacuum (no
charges or currents present) read17

∂B
∇×E = − ∇·E=0
∂t
∂E
∇ × B = ε0 µ0 ∇·B=0 (5.60)
∂t

Calculating ∇×∇ × E and ∇ × ∇ × B , using ∇ × ∇ × a = ∇(∇ · a) − ∇2 a and ∇ · a =


0 for a = E, B, and using ε0 µ0 = c12 , we get

1 ∂2E 1 ∂B2
∇2 E = and ∇2 B = (5.61)
c2 ∂t2 c2 ∂t2
Both the electric field E (r, t) and the magnetic field B(r, t) are solutions of a wave equa-
tion. Note the resemblance with eq. 5.55. The general solutions of these wave equations
eq. 5.61 can then be written as in analogy with eq. 5.56
√ Z

E (r, t) = d3 k E (k, t) eik·r
(2π)3
√ Z

B (r,t) = d3 k B (k, t) eik·r (5.62)
(2π)3

where again Ω is some convenient normalization volume. Assuming that these fields can
be represented by real numbers one has the general form

E (k, t) = E0 (k) e−iωt + E∗0 (−k) eiωt


B (k, t) = B0 (k) e−iωt + B∗0 (−k) eiωt (5.63)

You can check it explicitly that eqs. 5.62 and 5.63 constitute a solution of eq. 5.61 if the
following holds

ω = ck (5.64)

(k = |k|), which is the familiar dispersion relation for EM waves in vacuum. Again note the
resemblance with eq. 5.56. Apparently, the general solution of the EM field in vacuum
can be written as a linear combination of modes eik·r . The coefficients E0 (k) , B0 (k)
can in principle be completely arbitrary but Maxwell’s equations, eq. 5.60, place some
restrictions. Using eq. 5.62 in eq. 5.60, the div’s lead to

k · E0 (k) = 0 k · B0 (k) = 0

whereas the rot’s give

k × E0 (k) = ωB0 (k) k × B0 (k) = −ωε0 µ0 E0 (k)


17
In SI (rationalized mks) units. For a discussion on the various systems of units, see the appendix in W.
K. H. Panofsky and M. Phillips, Classical Electricity and Magnetism, (Addison-Wesley, Reading, 1978).
120 CHAPTER 5. QUANTUM FIELD OSCILLATORS

Or in other words

E0 (k) ⊥ B0 (k) ⊥ k and |E0 (k)| = c |B0 (k)| (5.65)

The wave vector k denotes the direction of propagation of the plane wave eik·r . Maxwell’s
equations show that in vacuum EM waves are transversal, i.e. both electric and magnetic
fields are perpendicular to the direction of propagation, and the amplitudes and phases of
the latter are coupled. We can define two polarization vectors εk,1 , εk,2 ⊥k and εk,1 ⊥εk,2 ,
and write
√ Z
Ω X
E (r, t) = d3 k Es (k, t) εk,s eik·r (5.66)
(2π)3 s=1,2

where Es (k, t) denotes the amplitudes and phases, which classically can have any contin-
uous value. The functions εk,s eik·r are again called the modes of the system; they can
be labeled with the wave vector k and polarization s. The general solutions of the wave
equation can be written as a linear combination of the modes. The magnetic field can be
obtained using the relations given in eq. 5.65.

If we are not in free space, but in a cavity or wave guide, for instance, the modes can
have a much more complicated form than just simple plane waves. This depends upon
the boundary conditions on the walls of the wave guide or cavity.18 Whatever the shape
of the wave guide or cavity, one can always obtain a set of modes fk1 k2 k3 s (r) which are
labeled by three numbers, k1 , k2, k3 and a polarization s = 1, 2. One can always write the
electric field as a linear combination of these modes.
√ Z

E (r, t) = dk1 dk2 dk3 Es (k1 k2 k3 ; t) fk1 k2 k3 s (r) (5.67)
(2π)3

In a closed cavity, the index ki is not continuous, but discrete (remember a particle in
aR box); inPthat case we simply replace the integral by a sum over the discrete index
dki → ki as usual. The magnetic field can be obtained from the electric field by
relations similar (but not identical !!) to eq. 5.65. The time dependence is set by the wave
equation, eq. 5.61.

It is well known from classical EM theory that instead of working with E and B fields
plus the restrictions of eq. 5.65 which are dictated by Maxwell’s equations, it is convenient
to work with potentials. As usual we choose a vector potential A (r, t) with ∇ · A = 0

(the so-called Coulomb gauge), such that

B=∇×A

From Maxwell’s equations one then shows that


∂A
E=−
∂t
18
If a wall is an ideal conductor, the electric field should be perpendicular to the wall, etcetera; you
know this stuff from your electricity & magnetism courses.
5.5. THE ELECTRO-MAGNETIC FIELD IN VACUUM 121

Note that the scalar potential φ (r,t) = 0 in the absence of charges. By choosing this
vector potential we automatically obey the two equations ∇ · B = 0 and ∇ · E = 0. The
two other Maxwell’s equations can then be combined to the wave equation

1 ∂2A
∇2 A = (5.68)
c2 ∂t2
with the obvious general solution
√ Z
Ω X
A (r, t) = d3 k As (k, t) εk,s eik·r (5.69)
(2π)3 s=1,2

Because ∇ · A = 0, it follows k · ²ks = 0; s = 1, 2; we have the familiar two transversal


waves. Because A (r, t) has to be real and a solution of eq. 5.68, the numbers As (k, t)
have the general form

As (k, t) = as (k) e−iωt + a∗s (−k) eiωt (5.70)

with ω = ck, as before. Again note the resemblance with eqs. 5.55 and 5.56. One can
easily check that the relations of eq. 5.65 are automatically fulfilled. The vector potential
this is the easiest way to set up EM theory; everything else can be derived from it.

5.5.2 Quantum Electro-Dynamics (QED)


Classically, eqs. 5.69, 5.70 solve the EM problem in vacuum. From the foregoing I guess
that by now you got the idea of how this is quantized. The vacuum is not a medium
in the sense of Sections 5.3 and 5.4; we have no “quantum chain” to start from.19 This
means that we cannot start from a Schrödinger equation of a quantum chain. Quantum
is more complex and more detailed than classical. We can go from quantum to classical
mechanics by making well-defined approximations. Going the other way, from classical to
quantum mechanics, one has to guess what these approximations could look like. There is
no systematic way for doing this. Schrödinger and Heisenberg were independently inspired
by classical mechanics, made an educated guess for quantum mechanics, and arrived at the
Schrödinger and Heisenberg equations, respectively. For a classical field which is not based
upon classical mechanical particles, like the EM field, we must use a different guess. One
option is to first devise a version of “classical mechanics” which can describe fields; see the
footnote at the end of Section 5.3. That route is very general, but also very formal. We
take a short-cut here. In Sections 5.3 and 5.4 we quantized the elastic medium, which can
be based upon mechanical particles (the “quantum chain”), in a way which was consistent
with the Schrödinger equation for the particles in the quantum chain. Inspired by this,
we close our eyes, and follow the same recipe for the EM field. So comparing eq. 5.69 to
eq. 5.56 we know what gets quantized: the numbers As (k, t) become operators qbs (k, t).
In quantum mechanics the vector potential becomes an operator.
√ Z
b Ω X
A (r, t) = d3 k qbs (k, t) εk,s eik·r (5.71)
(2π)3 s=1,2

19
In the 19th century EM waves were thought to propagate in a sort of elastic medium called the “aether”.
Already Maxwell hadn’t much use for it; his equations work fine without it. After Einstein’s theory of
relativity, it became completely obsolete.
122 CHAPTER 5. QUANTUM FIELD OSCILLATORS

where
µ ¶1 h i
~ 2
qbs (k, t) = as (k) e−iωt + b
b a†s (−k) eiωt (5.72)
2ε0 ω

Compare 5.70 to eqs. 5.38 and 5.50. Again the time dependence emerges because we follow
the “Heisenberg route” to quantization, i.e. we start from the classical time dependent
“observable” and directly quantize it into an operator, as explained in Section 2.6. The
connection to the usual “Schrödinger picture”, where the operators are time-independent
(and instead the states are time-dependent), can be made by simply setting t = t0 = 0
in the equations above, cf. eq. 2.64. The basic idea of quantizing the EM field is that,
like in the elastic medium case, we want the intensity of the EM field to be proportional
to the number of photons (so much we know from experiment), so we must quantize the
amplitudes in order to achieve this.

IMPORTANT MESSAGE
The classical modes εk,s eik·r are not changed, so these are not quantum wave functions.
Don’t make the mistake of calling εk,s eik·r the “wave function of the photon in vacuum”.
It is not; it simply is the classical mode. It is the amplitude which gets quantized, not
the mode. Compare to the discrete quantum chain. The modes are defined by eq. 5.37;
the wave function for the particles in the chain is given by eq. 5.18; these are completely
different things !! In a wave guide or cavity, which has classical modes fk1 k2 k3 s (r), the
quantum equivalent of eq. 5.66 for the vector potential becomes
√ Z
b (r, t) = Ω
A dk1 dk2 dk qbs (k1 k2 k3 , t) fk1 k2 k3 s (r)
(2π)3

which stresses again that the amplitudes, not the modes, are quantized. The position r is
b operator, like R was a discrete (lattice) position index
simply a continuous index for the A
in the quantum chain of eq. 5.52. Again, do not get confused with wave functions; the
r-dependence has nothing to do with that.

The classical E and B fields can be derived from eq. 5.69


√ Z
i Ω X
B (r, t) = ∇ × A = d3 k eik·r (k × εk,s ) qs (k,t)
(2π)3
√ Z
∂A Ω X
E (r, t) = − =− d3 k eik·r εk,s q̇s (k,t) (5.73)
∂t (2π)3 s

The transition to quantum mechanics can be completed by defining a conjugate momentum


analogous to eq. 5.46

ps (k,t) = ²0 q̇s (k,t) (5.74)

and making the p0 s and q0 s quantum operators


b (r, t) → qbs (k,t)
B
b (r,t) → pbs (k,t)
E (5.75)
5.5. THE ELECTRO-MAGNETIC FIELD IN VACUUM 123

Note that the E and B fields become operators E b and B. b Such operators which depend
upon a continuous index (r) are called quantum fields. We assume that the q’s and p’s
have familiar commutation relations
h i ¡ ¢
qbs† (k), pbs0 (k0 ) = i~δ k − k0 δ ss0 and
£ ¤ £ ¤
qbs (k), qbs0 (k0 ) = pbs (k), pbs0 (k0 ) = 0 (5.76)

where we have set t = 0 in order to obtain operators in the “Schrödinger picture”. Having
defined the relevant operators, how do we find the quantum “states” of the EM field? As
in Section 5.46 we do this by starting from the classical energy. The energy of a classical
EM field in a volume Ω is given by
Z · ¸
3 1 2 1 2
H= d r ε0 |E (r, t)| + |B (r, t)| (5.77)
2 µ0

We quantize by identifying this classical energy with the quantum mechanical Hamiltonian
b Using eqs. 5.73—5.77, the relations ε0 µ0 = 12 and ω = kc we find
H. c
" #
XZ pb†s (k)b
b= Ω
H d3 k
ps (k) 1 2 †
+ ε0 ω qbs (k)b
qs (k) (5.78)
(2π)3 s 2ε0 2

This looks like the familiar sum of harmonic oscillators again, see e.g. 5.48 and by now
we know how to solve this. Write
Z ½ ¾
b Ω X 3 1 †
H= d k +b as (k)bas (k) ~ω
(2π)3 s 2
h i ¡ ¢
b a†s0 (k0 ) = δ k − k0 δ ss0 and
as (k), b
h i £ ¤
b a†s0 (k0 ) = b
a†s (k), b as0 (k0 ) = 0
as (k), b (5.79)

in terms of creation and annihilation operators, where

³ε ω ´1 µ ¶1
0 2 1 2
b
as (k) = qbs (k) + i pbs (k)
2~ 2~ε0 ω

Note that these relations are consistent with eq. 5.72. The eigenstates of this Hamiltonian
are given by
Y
| {ns (k)}i = |ns (k)i = |n1 (k1 )i|n2 (k1 )i|n1 (k2 )i|n2 (k2 )i....|n1 (ki )i|n2 (ki )i...
ks

Each state |ns (ki )i contains a number of ns (ki ) quanta in mode ki and polarization s.
Each quantum has an energy ~ω = ~cki and a momentum ~ki . All the quanta belonging
to one mode ki and polarization s are indistinguishable and we can put as many into
each mode as we want. They can be considered as particles and are called photons, after
the Greek ϕoτ oς which means “light”. A “photon” thus is a “light particle”. Obviously,
photons are bosons.
124 CHAPTER 5. QUANTUM FIELD OSCILLATORS

CLOSING REMARKS
• We have derived the quantum form of the vacuum EM field by analogy with the
elastic medium for phonons. You cannot prove a priori that this is the correct way to
quantize the classical (Maxwell) equations. But then again you also cannot “prove”
the Schrödinger wave equation starting from classical mechanics for particles. You
have to postulate that this is the correct form and find out whether this predicts the
right results for experiments (it does).

• The quantum chain and the vacuum EM field were examples in which the particles
(phonons or photons) do not interact with each other, i.e. they are independent par-
ticles. If we introduce interaction terms, things get more complicated. For instance,
anharmonic interactions in the quantum chain of a type x bm x
bn x bl , or non-linear terms
in the EM field, which are caused by interactions with matter. Such interaction
terms can play the role of a perturbation Vb , for which we developed time-dependent
perturbation theory or the Green function techniques in the previous chapters.

• Another type of interaction is letting the bosons (photons/phonons) interact with


or decay into fermions. For instance, when a photon is absorbed in a crystal, the
photon (boson) is annihilated while at the same time an electron and a hole (two
fermions) are created. The photon energy is then usually on the scale of 1 eV . In
high energy physics a similar process can happen on a slightly higher energy scale.
A γ-photon with an energy ∼ 106 eV is annihilated and an electron and a positron
are created. By analogy, the latter is a hole in the vacuum. Again such interactions
can play the role of a perturbation Vb .

5.5.3 Are Photons Real Particles ?


This same questions we had with phonons arise again and I can use the same arguments
to try and convince you that photons are particles. Here I have a little bit more to start
from. We limit our discussion to a single mode A (r) = qs (k) εk,s eik·r . First we prove
that the photon momentum is given by p =~k. A classical EM field possesses momentum
as well as energy. Using Maxwell’s equations one can prove that the momentum of an EM
wave is given by (using Poynting’s theorem)
Z
p = ε0 d3 r E × B

It is perpendicular both to E, the electric, and B the magnetic field and thus, in view of eq.
5.65, it is in the direction of propagation of the wave k. Since E and B are perpendicular
as well, its size is
ε0
|p| = ε0 |E||B| = |E|2 = ε0 c|B|2 =
½ c ¾
1 1 1
= ε0 |E| + |B|2 = H
2
2c µ0 c

using eq. 5.77. In other words, the classical momentum is given by


Hk H
p= = k
c k ω
5.5. THE ELECTRO-MAGNETIC FIELD IN VACUUM 125

As we know, the quantum energy of a photon is given by H = ~ω; but then its momentum
must be given by p =~k.

We can also use a pure particle argument to show the same. From special relativity
we know that the energy E and the momentum p = |p| of any particle are given by

m0 c2 m0 v
E = mc2 = q and p = mv = q (5.80)
2 2
1 − vc2 1 − vc2

where m0 and v are the rest mass and the velocity of the particle, respectively. From these
relations one derives
q
pc2
E = p2 c2 + m20 c4 and v = p (5.81)
p2 c2 + m20 c4
For the photon v = c per definition. This can only be true if m0 = 0, the photon has zero
rest mass. But then E = pc, and since also E = ~ω = ~ck it then follows that p = ~|k|.

To prove the particle properties of photons we can do a scattering experiment anal-


ogous to the one shown in Fig. 5.4. The famous photon scattering experiment is called
Compton 27 scattering in which one scatters high energy X-rays of a crystal. The incoming
X-ray photons with momentum ~k and energy ~ω(k) = ~ck can collide with the electrons
in the crystal, which have an energy ²i and momentum pi . Out go X-ray photons with
momentum ~k0 and energy ~ω(k0 ) = ~ck 0 and recoiled electrons with momentum pf and
p2f
energies ²f = 2me , as shown in Fig. 5.5, both of which can be observed experimentally.

pi , εi p f ,ε f

=k , =ω(k ) =k ' , =ω(k ' )

Figure 5.5: Compton scattering of X-rays.

Of all the X-ray photons and electrons coming out we can analyze their momenta and
energies. We are helped by the fact that the binding energy of the valence electrons of
the atoms in the crystal is very small compared to the high energy of the X-rays and
may be neglected. The valence electrons thus have (approximately) an energy ²i = 0 and
momentum pi = 0, i.e. they can be considered at rest. The remaining quantities turn out
to be related to the momentum and the energy of the incoming photon as
¡ ¢
~ω (k) = ²f + ~ω k0 (5.82)
~k = pf + ~k0

But these are nothing else than the conservation of energy and the conservation of momen-
tum, if we interpret Compton scattering simply as a “billiard ball”-like collision between
126 CHAPTER 5. QUANTUM FIELD OSCILLATORS

a photon and an electron particle. The conservation rules of eq. 5.82 can be derived as
selection rules, using the by now familiar perturbation theory. Compton’s experiment was
very important historically in convincing people of the particle properties of EM radia-
tion, since it is extremely difficult, if not impossible, to explain the results by any other
means.20

As we have seen in the case of phonons, photons are kind of strange particles. Each
photon is created into a fixed mode with spatial dependence eik·r and polarization εk,s
which is just the classical mode (that is, in vacuum; in a wave guide/cavity, the mode is a
more complicated fk1 k2 k3 s (r)). Do not confuse this with a wave function in the quantum
mechanical sense; a photon has no wave function. The photon “exists” at all positions
in the mode and is completely delocalized. Think of the analogy to the quantum chain,
where a single phonon is a quantum of a vibration which extends over the whole chain.
Again, some people prefer to use the term “collective excitation” (in this case: of the
vacuum). Photons are very fragile particles, they interact with any particle which has
a charge, such as electrons and protons. They also interact with any quasi-particle or
collective excitation which involves a charge distribution, such as phonons and plasmons.
As a result of such interactions, photons are very easily created or annihilated and their
number is almost never constant or even very well-defined.

20
If you want to know more about Compton scattering, have a look at S. Gasiorowicz, Quantum Physics
(Wiley, New York, 1974), Ch.1; or B. H. Brandsen and C. J. Joachain, Quantum Mechanics (Prentice Hall,
Harlow, 2000), Ch.1.
Chapter 6

Bosons and Fermions

“The creatures outside looked from pig to man, and from man to pig, and from pig to man again;
but already it was impossible to say which was which”, George Orwell, Animal Farm.

In the previous chapter we discussed many-boson systems that could entirely be de-
scribed in terms of harmonic oscillators. We obtained an elegant simple picture by using
a special set of states and operators to set up the so-called occupation number represen-
tation. In this chapter we extend this technique to make it more generally applicable
to bosons, as well as fermions. In this case we start from a “traditional” wave function
representation and work our way to the number representation. The latter is in fact more
general and also applies to systems like the EM field, where the former does not apply, as
we have seen in the previous chapter. At first the material will be a bit tedious, since many
particle states involve a lot of book-keeping, I can’t help that. However, the final results
will be again be very elegant and physical applications will follow. There is a good dis-
cussion of many-fermion states in Mattuck’s book. Chapter 4, p. 64-72 (§ 4.1 - 4.3) gives
a general introduction. Chapter 7, p. 123-141 gives a full picture. In this chapter I will
start from a single particle and then build in the wave-function-like way a many-particle
state (first quantization). Then I will introduce the occupation number representation, or
second quantization technique, for bosons and fermions, respectively. Second quantization
(for fermions here) allows for a reinterpretation of the results in terms of particles and
anti-particles. I will also try and give some explanation for the seemingly strange name
“second quantization” and give you some physical insight in particles and holes. Appendix
I contains all the detailed algebra of occupation number representation and proofs of the
statements made in the text. For those of you who are interested in some background,
Appendix II discusses why identical particles create a problem in quantum mechanics and
what the way out is (the symmetry postulate). You might want to start by re-reading
Chapter 1, Section 1.4.

6.1 N particles; the Stone Age


Let φm (r1 ) be a wave function for a single particle, and the set of functions m = 1, 2, ....
form a basis set. If we have a system of N of such particles (N ≥ 2), a basis set is formed
by the product functions

φm1 (r1 ) · φm2 (r2 ) · ........ · φmN (rN ) (6.1)

127
128 CHAPTER 6. BOSONS AND FERMIONS

where all the indices mi ; i = 1, ......N independently can assume all allowed values, mi =
1, 2, .... (see Section 1.4). In the following we will use Dirac notation and write |m (1)i
instead of φm (r), meaning that particle no. 1 is in a state |mi. In this notation, the
product state of eq. 6.1 becomes

|m1 (1)i|m2 (2)i.......|mN (N )i ≡ |m1 (1) m2 (2) .......mN (N )i (6.2)

which means that particle 1 is in state |m1 i, particle 2 is in state |m2 i, ..., particle N is
in state |mN i. The most general N -particle state can written as a linear combination of
such states.
XX X
|Ψ(1, 2, ..., N )i = ... Cm1 m2 ...mN |m1 (1) m2 (2) .......mN (N )i
m1 m2 mN
with the coefficients Cm1 m2 ...mN = hm1 (1) m2 (2) .......mN (N ) |Ψ(1, 2, ..., N )i

Let us consider the single basis state of eq. 6.2. If the particles are identical, they should
per definition be indistinguishable (Appendix II). However, writing down in which state we
put each particle implies that we can distinguish the particles (particle 1 is in state |m1 i,
particle 2 is in state |m2 i, etcetera), since these states can be different. We could use the
state numbers m1 , m2 , ... to label the particles. We could call the particle marked m1 the
“blue” particle, the one called m2 the “red” particle, etcetera. In principle, m can label
the eigenstates of an observable, so the “color” of the particle could be measured. Clearly
we have found a way to distinguish the particles, see also the discussion in Appendix II.
Thus the state of eq. 6.2 is not entirely physical. We should in fact not be able to make a
distinction with, for instance, a state in which particle 1 is in state |m2 i and particle 2 is
in state |m1 i. To be specific, let Pb12 be an operator which interchanges the two particles
1 and 2.

Pb12 |m1 (1) m2 (2) ....mN (N )i = |m1 (2) m2 (1) .......mN (N )i

The only way (within a linear vector space) to ensure that we are not able to tell which
of the particles 1 or 2 occupies |m1 i or |m2 i, is to make a linear combination between this
state and the one from eq. 6.2

|m1 (1) m2 (2) .......mN (N )i + eiϕ |m1 (2) m2 (1) .......mN (N )i =


(1 + eiϕ Pb12 )|m1 (1) m2 (2) .......mN (N )i

Both terms in this expression have an equal weight, but the coefficient’s phase ϕ we don’t
know a priori.1 Realizing that a similar thing must hold for any permutation Pb in which
the N particles are distributed/permutated over the states |m1 i|m2 i.....|mN i, it follows
that states that are allowed for indistinguishable particles, must be linear combinations in
which all these permutated states appear with equal weights
X
eiϕp Pb|m1 (1) m2 (2) .........mN (N)i
Pb
1
The most general linear combination has a form like c1 φ(1, 2) + c2 φ(2, 1), using an obvious short-hand
notation. Since both terms must have equal weight, |c1 | = |c2 |. Writing c1 = |c1 |eiϕ1 and c2 = |c2 |eiϕ2 ,
one can rewrite this linear combination in the form c1 [φ(1, 2) + ei(ϕ2 −ϕ1 ) φ(2, 1)]. Setting c1 = 1 and
ϕ = ϕ2 − ϕ1 results in the form given.
6.1. N PARTICLES; THE STONE AGE 129

where Pb is a permutation operating on the particles and the sum is over all N ! possible
permutations, i.e. over all possible orderings of the N particles. A priori, it may seem that
we cannot say anything about the phase factors eiϕp . However, a fundamental postulate
of quantum mechanics, called the symmetry postulate, states that only two forms of N-
particle states are actually allowed .2

The first special form is called the totally symmetric state, where all the phase factors
eiϕp = 1
X
|m1 m2 .......mN i(s) = Ns Pb|m1 (1) m2 (2) .........mN (N )i (6.3)
Pb

where Ns is a normalization constant we will determine later on. It is straightforward to


prove that if Pbij is a permutation which interchanges particles i and j then

Pbij |m1 m2 .....mN i(s) = |m1 m2 ......mN i(s) (6.4)

See Appendix I for this proof. The idea is that, using the right hand side of eq. 6.3, if
Pb runs over all possible N ! permutations, the product Pbij Pb also runs over all possible
N ! permutations. Since every permutation can be decomposed into a series of successive
interchanges of two particles, if eq. 6.4 holds for any interchange Pbij , it also holds for any
arbitrary permutation Pb. The state |m1 m2 .......mN i(s) does not change if we permutate
the particles. We say that this state is symmetric with respect to all possible interchanges,
or permutations of particles. In other words, we can no longer identity which particular
particle is in which state, which is what we wanted.

The second special form is called the totally anti-symmetric state and is written as
X
|m1 m2 ....mN i(a) = Na (−1)p Pb|m1 (1) m2 (2) ....mN (N )i (6.5)
Pb

The phase factors are eiϕp = ±1 according to the following rule: the number (−1)p = 1,
or p = 0, if the permutation Pb can be written as an even number of interchanges and
(−1)p = −1, or p = 1, if the permutation Pb is equivalent to an odd number of interchanges.
We have

Pbij |m1 m2 .....mN i(a) = −|m1 m2 ......mN i(a) (6.6)

See Appendix I. The idea is that each permutation Pb in eq. 6.5 is even or odd, i.e.
(−1)p = ±1; the single interchange Pbij is odd per definition, i.e. (−1)pij = −1, so the
product Pbij Pb is odd/even, i.e. (−1)p+pij = ∓1). We say that this state is anti-symmetric
with respect to all possible interchanges of particles. As always in quantum mechanics the
state |m1 m2 .....mN i(a) = |.....i(a) is not directly observable. What can be measured are
b
expectation values (a) h....|A|....i b is some (observable) operator. So minus signs
(a) , where A

(−) like in eq. 6.6 seem to have no influence on expectation values.

2
This postulate should be added to the ones given in the first chapter. It is a true postulate; one can
make it plausible (see appendix II), but it cannot be proved from other physical postulates.
130 CHAPTER 6. BOSONS AND FERMIONS

One should however not draw the conclusion that the minus sign is unimportant. It has
very important consequences, because it leads to the Pauli 45 exclusion principle. This can
be derived as follows; suppose two of the numbers ml in eq. 6.6 are identical, for instance
mi = mj . Then we have

Pbij |...mi ...mj ...i(a) = |...mj ...mi ...i(a) = |...mi ...mj ...i(a) (a)

since obviously a state cannot change if we interchange two identical numbers. On the
other hand, because of eq. 6.6 we have

Pbij |...mi ...mj ...i(a) = −|...mi ...mj ...i(a) (b)

By an elementary theorem of linear vector algebra, both (a) and (b) can only be true
simultaneously if |...mi ...mj ...i(a) =|i the “null” vector of vector space. It follows that for
any state |...mi ...mj ...i(a) which is not the (trivial) null vector, all the numbers ml ; l =
1, .., N must be different from one another. Or, in other words, in the anti-symmetric state
all the particles must be in different one-particle states. If one particle occupies a state
|m1 i, then the other particles cannot occupy this state anymore, i.e. they are excluded
from this state. This is the Pauli exclusion principle or in German: “das Pauli Verbot”
(somehow the rule sounds stricter in German).

The symmetry postulate states that all many-particle systems which consist of identical
particles either have symmetric or anti-symmetric states. Moreover, it states that there
are two kinds of particles

• If the particles are bosons, all their states must be symmetric.

• If the particles are fermions, all their states must be anti-symmetric.

Bosons cannot become fermions and vice versa, and all elementary particles must
be either bosons or fermions.3 The most general N -particle state for bosons (fermions)
respectively can written as a linear combination of the symmetric (anti-symmetric) basis
states of eqs. 6.3 and 6.5
XX X
|Ψs/a (1, 2, ..., N )i = ... Cm1 m2 ...mN |m1 (1) m2 (2) .......mN (N)i(s)/(a)
m1 m2 mN

with the coefficients Cm1 m2 ...mN =(s)/(a) hm1 (1) m2 (2) .......mN (N ) |Ψs/a (1, 2, ..., N )i

Using eqs. 6.4 and 6.6 it is easy to prove that general boson (fermion) states are symmetric
(anti-symmetric), i.e.

Pbij |Ψs (1, 2, ..., N )i = |Ψs (1, 2, ..., N )i


Pbij |Ψa (1, 2, ..., N )i = −|Ψa (1, 2, ..., N)i

In addition, using relativistic quantum mechanics, Pauli argued the following.


3
In order to explain the fractional quantum hall effect a new kind of particle has recently been introduced
by Laughlin 98 , where an interchange of two particles does not introduce a ± sign, but a more general phase
factor, i.e. Pbij |M1 M2 .....MN i(ϕ) = eiϕ |M1 M2 ......MN i(ϕ) . These particles are called anyons. However,
these are not elementary, but composite particles.
6.1. N PARTICLES; THE STONE AGE 131

• If the particles have integer spin, i.e. S = 0, 1, ... they are bosons.

• If the particles have half-integer spins, i.e. S = 12 , 32 , ... they are fermions.

Examples of bosons are phonons, photons, plasmons; examples of fermions are elec-
trons, protons, neutrons. Composite particles are also bosons or fermions; for instance
the 4 He nucleus (2 protons and 2 neutrons) has spin zero and is a boson, whereas the
3 He nucleus (2 protons and 1 neutron) has spin 1 , and is a fermion. As you know from
2
your statistical mechanics course, whether particles are bosons or fermions has profound
consequences for the thermal physical properties of a many-particle system. Whether a
composite particle must be considered as a single particle, or as a collection of elemen-
tary particles depends upon the experiment you are doing. For instance, if you study the
properties of liquid 4 He, you do not observe the internal structure of the 4 He nuclei, so
the He atoms behave like bosons (which leads to macroscopic quantum phenomena at low
temperature like Bose-Einstein condensation or superfluidity ). However, doing a nuclear
experiment, for instance by scattering high energy particles off a He nucleus, you probe
its internal structure and you do see that it is actually composed of protons and neutrons,
which behave like fermions.

To have the states |....i(s) and |....i(a) normalized i.e. (s) h....|....i(s) = 1 and (a) h....|....i(a) =
1, one has to calculate the factors Ns and Na in eqs. 6.3 and 6.5
1 1
Ns = (n1 !n2 !...nN !N !)− 2 Na = (N !)− 2 (6.7)

where for bosons (s) we assumed that n1 of the particles are in state |m1 i; n2 are in state
|m2 i etcetera. For a more detailed discussion of (anti-)symmetric states and normalization
factors see Appendix I.

6.1.1 The Slater Determinant


Given the definition of the determinant, det (M ) or |M |, of a matrix M ; det (M ) =
P
(−1)p M1P1 · M2P2 · ......MNPN (see your linear algebra courses), we can see that it is
P
possible to write eq. 6.5 as a determinant. We get, using eq. 6.7
¯ ¯
¯ |m1 (1)i |m1 (2)i ... |m1 (N )i ¯
¯ ¯
(a) 1 ¯¯ |m2 (1)i |m2 (2)i ... |m2 (N )i ¯
¯
|m1 m2 ...mN i = √ ¯ ¯ (6.8)
N! ¯ ... ... ... ... ¯
¯ |mN (1)i |mN (2)i ... |mN (N )i ¯

This is called a Slater determinant. Slater determinants are a systematic way of enumerat-
ing all the allowed states for fermions. Note that a Slater determinant automatically takes
care of the Pauli exclusion principle; if two single particle states are equal, i.e. mi = mj ,
then two rows of the determinant will be identical, and, as we know, the determinant of a
matrix with two identical rows is zero.
Needless to say, for bosons there is no exclusion principle. We can put all bosons in
the same single particle state, or in different ones, they don’t care; everything is allowed.
Both bosons and fermions represent a perfect implementation of the communist ideal: all
particles are equal, and none are more equal than others. Fermions make up a funny team:
all particles have to be in a different state, but it is impossible to say which particle is
132 CHAPTER 6. BOSONS AND FERMIONS

in which state. It is like a football team with positions as keeper, defender, mid-fielder,
forward etcetera, where each player has to play at each position in turn, but never with
two players at the same position. (It remotely resembles the system the Dutch football
team prefers. There it usually goes wrong because players forget to occupy at least one
of the defending positions. In quantum mechanics it works). Bosons make up an even
funnier team: all particles not only are equal, but they can also be all in the same state.
(Like the Italian football team, which prefers to play with eleven defenders).

6.1.2 Three Particle Example Work-out


Suppose we have three particles in states m1 6= m2 6= m3 . Forming product states the
possible permutations of the particles over the states are

|m1 (1)m2 (2)m3 (3)i ≡ |123i


|m1 (2)m2 (3)m3 (1)i ≡ |231i
|m1 (3)m2 (1)m3 (2)i ≡ |312i
|m1 (2)m2 (1)m3 (3)i ≡ |213i
|m1 (1)m2 (3)m3 (2)i ≡ |132i
|m1 (3)m2 (2)m3 (1)i ≡ |321i

There are 3! = 6 possible permutations. The notations |123i etcetera are just a short-hand.
The permutations can be connected by defining operators Pbij , that interchange particles
i and j.

Pb13 Pb12 |123i = Pb13 |213i = |231i

i.e. first we interchange particles 1 and 2, and then we interchange particles 1 and 3. Note
that in the same way

Pb12 Pb13 |123i = |213i

which means that Pb12 Pb13 6= Pb13 Pb12 . In general, the permutation operators do not commu-
tate. The symmetric state, which is appropriate for bosons can be written as

1
|m1 m2 m3 i(s) = √ [|123i + |231i + |312i +
3!
|213i + |132i + |321i]

For the anti-symmetric state, which is appropriate for fermions, we have to determine the
sign (−1)p for each permutation, where p is the number of interchanges needed to produce
the permutation. In the two examples given above we have two interchanges, i.e. p = 2
and thus the sign is (−1)2 = 1. Another example is |213i = Pb12 |123i, which obviously
carries the sign (−1)1 = −1. It is then easy to show that the anti-symmetric state is

1
|m1 m2 m3 i(a) = √ [|123i + |231i + |312i −
3!
|213i − |132i − |321i]
6.1. N PARTICLES; THE STONE AGE 133

Another way to write this state is the Slater determinant form


¯ ¯
¯ |m1 (1)i |m1 (2)i |m1 (3)i ¯
(a) 1 ¯¯ ¯
¯
|m1 m2 m3 i = √ ¯ |m2 (1)i |m2 (2)i |m2 (3)i ¯
3! ¯ |m (1)i |m (2)i |m (3)i ¯
2 2 2

Using the familiar expansion rules for determinants with signs


¯ ¯
¯ + − + ¯
¯ ¯
¯ − + − ¯
¯ ¯
¯ + − + ¯

gives
¯ ¯
1 ¯ |m2 (2)i |m2 (3)i ¯
|m1 m2 m3 i (a)
= √ {|m1 (1)i ¯ ¯ ¯
3! |m3 (2)i |m3 (3)i ¯
¯ ¯ ¯ ¯
¯ |m1 (2)i |m1 (3)i ¯ ¯ |m1 (2)i |m1 (3)i ¯
¯
−|m2 (1)i ¯ ¯ + |m (1)i ¯ ¯}
|m3 (2)i |m3 (3)i ¯ 3 ¯ |m2 (2)i |m2 (3)i ¯

Expanding this further gives the same state |m1 m2 m3 i(a) as before. The determinant form
thus takes care of all the signs.

6.1.3 One- and Two-particle Operators


Knowing all the allowed basis states for fermions and bosons, we can now try to calculate
matrix elements of operators, which, according to Chapter 1, is what we need to set
up a representation. In principle one can have operators which work on all particles
simultaneously. In practice such operators can usually be decomposed into simpler ones.
1. The “simplest” operator is an operator which only operates on one particle b h (i) .
Because the N particles are identical, we then must have such an operator working
on each particle. Moreover, for N identical particles only operators are allowed
which are invariant, i.e. they do not change, if we permute the particles. A bit of
contemplation results in the form
N
X
b
h= b
h (i) (6.9)
i=1

If all the b
h (i)’s are identical (except from the particle they operate on), it is clear
that since we sum over all the particles, b h does not change if we interchange two
particles. The operator b h is loosely called a one-particle operator (despite the fact
that it operates on all the particles). Familiar examples of one-particle operators are
the total kinetic energy of N electrons
N
X
b b 2i
p
h= (6.10)
2me
i=1

or the Coulomb potential that a nucleus of charge Ze at position R exerts on the N


electrons
N
X
b −Ze2
h=
|R−b
ri |
i=1
134 CHAPTER 6. BOSONS AND FERMIONS

2. The next “simplest” operator works on two particles Vb (i, j). Again, because the
particles are identical, in N -particle space such an operator must have the form
N N
b 1X X b
V = V (i, j) (6.11)
2
i=1 j6=i=1

Since we sum over all the different pairs of particles, and all the terms Vb (i, j) have
an identical form, Vb does not change if we interchange two particles, which is what
we want. A Vb (i, i) part is usually excluded since, if it exists, it can be considered as
a one-particle operator. The factor 12 is simply included for convenience; it has no
deep meaning. The operator Vb is called a two-particle operator . A familiar example
is the Coulomb repulsion between the N electrons
1X e2
Vb = (6.12)
2 ri − b
|b rj |
i6=j

3. In principle, one can have operators working on 3 particles, 4 particles etcetera. In


every day practice, such more-than-two-particle operators are rare. Therefore we
stick to one- and two-particle operators.

6.2 N particles; the Modern Era


6.2.1 Second Quantization for Bosons
The states of eq. 6.3 and 6.5 (or eq. 6.8) form a perfect basis set for constructing a
representation for an N -boson or N -fermion system, but they can be very clumsy to work
with. Suppose we let each mi ; i = 1, 2, ... run from 1 to M , then we have M N product
states of the type of eq. 6.2. We have to combine N ! of such products to form a single
state of type eq. 6.3 or 6.5 (in the case of fermions, most of them are not allowed because
of the Pauli exclusion principle). To complete a representation (see Chapter 1, Table 1.1),
we have to calculate matrix elements of one- and two-particle operators over these states.
It is all very systematic and produces the correct results, but it is also very cumbersome
and not very elegant. Is there no easier way to set up a representation ? Of course there
is, it is second quantization or the occupation number representation.

In the previous chapter we have already used it for the many-harmonic oscillator case,
and we now are going to generalize it for any system of bosons. Any basis state for bosons
can be written as

| m1m1 ...m1m2 m2 ...m2 ......mN 〉 ( s ) ≡| ...〉 ( s )


n1 n2 nM

Figure 6.1: An N -boson state.

where we have listed the particles such that the first n1 of them are the same state
|m1 i, the next n2 are in state |m2 i, etcetera. Note that any N -boson state can be written
6.2. N PARTICLES; THE MODERN ERA 135

this way, the only restriction is that n1 + n2 + ... + nM = N , the total number of particles.
We assumed that there are M different possible states |mi i. This number could be infinite;
it does not matter, the formalism is valid anyway.

As an example take a 6-particle boson state, i.e. N = 6, where each particle can be in
one of the seven states |mi; m = 0, ..., 6, i.e. M = 7. Writing down a state in the notation
of the previous sections, for each particle we list in which state it is; for example
|exi(s) = |m1 m2 m3 m4 m5 m6 i(s) = |1 2 1 0 1 4i(s) (6.13)
The first particle is in state |1i, the second in state |2i, the third in state |1i again, etcetera,
and we symmetrize the total state such, that we don’t know which particle is in which
state. We now rewrite this state such that first we list all the particles that are in state
|0i, then all the particles that are in state |1i, etcetera.
|exi(s) = |0 1 1 1 2 4i(s) (6.14)
Note that this is perfectly OK, since all we do in the rewriting process is to change the
order of the particles. Since we are dealing with a totally symmetric state |exi(s) anyway,
this is allowed; the state is the same for any permutation of the particles, as we have seen
in the previous section.

Now comes the magic hat trick : the state of Fig. 6.1 can also be written in a new
notation as
|...i(s) = |n1 n2 ......nM i (6.15)
We don’t list the states anymore, but simply enumerate the number n1 of particles which
are in state |m1 i, the number n2 of particles which are in state |m2 i etcetera. This can
never lead to any confusion if we agree to always list the states in an order such that
m1 < m2 < ... < mN as in the example above. In the new notation the example state of
eq. 6.14 becomes
|exi(s) = |1 3 1 0 1 0 0i (6.16)
We have one particle in state |0i, three particles in state |1i, one particle in state |2i, zero
particles in state |3i, one particle in state |4i, and zero particles in state |5i and |6i. Stop
here and think about it; it is important to understand what the change of notation going
from eq. 6.14 to eq. 6.16 is.

As you will have guessed, the new notation is nothing else than the occupation num-
ber representation. In a sense it is just a change of notation and just another way of
enumerating the possible basis states. However, by explicitly focussing on the occupation
numbers nj rather than on the single particle states |mi i the representation becomes easier
to handle. For instance, we can define creation and annihilation operators, analogous to
the previous chapter
1
a†i |n1 n2 ...ni ...nM i = (ni + 1) 2 |n1 n2 ...ni + 1...nM i
b
1
b
ai |n1 n2 ...ni ...nM i = (ni ) 2 |n1 n2 ...ni − 1...nM i with
h i h i
ai , b
b a†j = δ ij and [b ai , b a†i , b
aj ] = b a†j = 0 (6.17)
136 CHAPTER 6. BOSONS AND FERMIONS

and construct new states with these operators. In fact as we saw in the previous chapter,
we can construct all possible states by repeatedly operating with creation operators on
a state without any particles (in any of the states). The latter state is written as |0i; is
called the vacuum state or vacuum, for short.
³ ´n1 ³ ´n2 ³ ´nM
a†1
b a†2
b a†M
... b
|n1 n2 ...nM i = 1 |0i (6.18)
(n1 !n2 !...nM !) 2

The real power of this representation does not lie in a method of enumerating all possible
states more easily.

The real power is revealed by writing all operators in terms of creation and annihilation
operators, like we did for the harmonic oscillator. In deriving these operator some algebraic
manipulation is required, which you can find in Appendix PNI. The results of that algebra
are given here. The one-particle operator of eq. 6.9, h = i=1 b
b h (i), can be written as
M
X
b
h= a†k b
hkl b al where
k,l=1
D E
hkl = k(1)|b
h(1)|l(1) (6.19)

is the single-particle matrix element of one of the N terms bh (1) between the single-particle
states |ki and |li. It does not matter which of the N single-particle terms b h (i) we take,
since they are all identical. Therefore the argument (1) is a dummy argument, just to
indicate that the matrix element is a single particle integral. To be specific, take the
b2
p
example of eq. 6.10 in a position representation, where we know b h (1) = 2m1e → − 2m~2
e
∇21
and |li → φl (r1 ). The matrix element then becomes
Z · ¸
~2
hkl = d3 r1 φ∗k (r1 ) − ∇21 φl (r1 )
2me

Note again that the argument r1 is dummy, we might have used r2 or any symbol for
that matter. The really important part of eq. 6.19 is that we have written the one-
particle operator as a sum of M 2 products of a creation and an annihilation operator.
The summation now is over all M 2 possible combinations of basis states, instead of over
the N particles as in eq. 6.10. The proof of the correctness of eq. 6.19 is given in Appendix
I.
PN
The two-particle operator of eq. 6.11, Vb = 1
2 j6=i=1 V
b (i, j), can be written as

M
X
1
Vb = a†l b
Vklmnb a†k b
amb
an where
2
k,l,m,n=1
D ¯ ¯ E
¯ ¯
Vklmn = k(1)l(2) ¯Vb (1, 2)¯ m(1)n(2) (6.20)

is the two-particle matrix element of one of the 12 N (N −1) identical terms Vb (1, 2) between
the states |ki, |li, |mi, and |ni. Again to be specific, take the example of eq. 6.12, where
6.2. N PARTICLES; THE MODERN ERA 137

e2
Vb (1, 2) → |r1 −r2 | and
Z Z
e2
Vklmn = d3 r1 d3 r2 φ∗k (r1 )φ∗l (r2 ) φ (r1 )φn (r2 ) (6.21)
|r1 − r2 | m
Note again that the arguments r1 , r2 are dummy, we can use any symbols, as long as we
use two different ones. The important part again is that we have written the two-particle
operator as a sum of M 4 products of two creation and two annihilation operators. The
summation now is over all M 4 possible combinations of basis states, instead of over the
1
2 N (N − 1) pairs of particles as in eq. 6.12. Note the interchange of the indices k and l
on the creation operators as compared to the matrix element in eq. 6.20 !! Now stop and
note it again; you are now warned twice, disregard it and it will cause you trouble !!

The states and operators of eqs. 6.15—6.20 constitute the occupation number repre-
sentation or second quantization for bosons. Why is this representation much easier than
using the representation defined by Fig. 6.1 and eqs. 6.9 and 6.11 ?

Remember, if we want to set up a representation in order to solve the Schrödinger


equation, or calculate expectation values of observables (to predict the outcome of exper-
iments), we have to calculate matrix elements. In the “old” notation we have to sandwich
operators like eqs. 6.9 and 6.11 between states like Fig. 6.1. The algebra is very cumber-
some; see Appendix I. Setting up the operators of eqs. 6.19 and 6.20, one has to calculate
single- and two particle matrix elements. But as soon as we have these, we are in the
derivers seat again and it is plain cruising from there on. We just apply the algebraic rules
given for creation/annihilation operators, cf. eq. 6.17. Similar to the harmonic oscillator,
these rules are sufficient to calculate all desired N -particle matrix elements in an easy way.
The operators

n a†i b
bi = b ai (6.22)

have a very special meaning. From eq. 6.17 one proves

bi |n1 n2 ...ni ...nM i = ni |n1 n2 ...ni ...nM i


n (6.23)

These operators counts the number of particles which occupy state |mi i and are thus called
occupation number operators or number operators for short. The states |n1 n2 ...ni ...nM i,
which are just another way of writing down the states |m1 m2 .......mN i(s) , are eigenstates
of these number operators. That is why the representation using these states as a basis
set, is called the (occupation) number representation. The operator
M
X
b=
N bi
n (6.24)
i=1

counts the total number of particles present, and (surprise, surprise) it is called the total
number operator.

6.2.2 Second Quantization for Fermions


Enlightened by the previous section we will now develop the same tools for many-fermion
systems. As before, we start by enumerating the fermion states, as in Fig. 6.2.
138 CHAPTER 6. BOSONS AND FERMIONS

| m1 .m2 m3 ......mN 〉 ( a ) ≡| ...〉 ( a )


n1 n2 n3 nM

Figure 6.2: An N -fermion state.

Since we are dealing with fermions, all the |mi i’s must be different on accord of Pauli’s
exclusion principle. All the occupation numbers nj must either be 0 or 1. Furthermore,
we have to pay special attention to the order in which the states |mi i appear, since by
interchanging the order, “−” signs appear

|m1 ...mi ...mj ...mM i(a) = −|m1 ...mj ...mi ...mM i(a)

The proof of this is most easily seen from the Slater determinant of eq. 6.8; interchanging
two rows in a determinant changes its sign. This unfortunately means that we have to
keep a good track of all the “−” signs. The convention of writing down an anti-symmetric
state as in Fig. 6.2 is to use the order

m1 < m2 < .... < mN (6.25)

This defines the order in a unique way, and all the states can be enumerated.

Let us look at an example again; a 5-particle fermion state, i.e. N = 5, where each
particle can be in one of the seven states |mi; m = 0, ..., 6, i.e. M = 7.

|exi(a) = |m1 m2 m3 m4 m5 i(a) = |0 1 2 4 6i(a)

Here is the magic hat trick again; the state of Fig. 6.2 is written in a new notation as

|...i(a) = |n1 n2 ......nM i (6.26)

n1 (0 or 1) particle(s) in state |m1 i, n2 particle(s) in state |m2 i etcetera. For the example
state this becomes

|exi(a) = |1 1 1 0 1 0 1i (6.27)

which means 1 particle in states |0i, |1i and |2i, 0 particles in states |3i and |5i, and 1
particle in states |4i and |6i. As you can see, for fermion states the occupation number
representation results in a sort of digital notation.

c†i and b
Again we can define creation and annihilation operators, which we now call b ci ,
respectively.

c†i |n1 n2 ...ni ...nM i = (−1)Σi (1 − ni )|n1 n2 ...ni + 1...nM i


b
ci |n1 n2 ...ni ...nM i = (−1)Σi (ni )|n1 n2 ...ni − 1...nM i
b (6.28)

The prefactors on the right hand side need some explanation. First the “ni ” factors, which
take care of Pauli’s exclusion principle.
6.2. N PARTICLES; THE MODERN ERA 139

• If ni = 1, one cannot create another particle in the same state because of Pauli’s
exclusion principle and if ni = 0 one can. The factor (1 − ni ) takes care of this.
• If ni = 0, one cannot annihilate a particle from this state per definition, but if ni = 1
one can. The factor (ni ) takes care of this.

The factors (−1)Σi come from a decent book keeping of signs. See also Mattuck, §7.3.
One way to fix the sign-and-ordering problem is to let a creation operator bc†k always create
its particle in a state at the left most position. In old-fashioned notation this reads

c†k |m1 m2 ....mN i(a) = |km1 m2 ....mN i(a)


b (6.29)

where we have added one particle in state |ki to the N which were already present in
states |m1 i, |m2 i, ..., |mN i. Note k must be different from any of the mi , otherwise we
get zero. What happens is again best seen from the Slater determinant of eq. 6.8. The
creation operation of eq. 6.29 adds an extra row and column. If k is equal to one of the
mi , two rows in the determinant are equal, and the determinant is zero.
¯ ¯
¯ ¯ ¯ |k (1)i |k (2)i ... ... |k (N + 1)i ¯
¯ |m1 (1)i |m1 (2)i ... |m1 (N )i ¯ ¯ ¯
¯ ¯ ¯ |m1 (1)i |m1 (2)i ... ... |m1 (N + 1)i ¯
¯ |m (1)i |m (2)i ... |m (N )i ¯ ¯ ¯
c†k ¯¯
b 2 2 2 ¯ = ¯ |m2 (1)i |m2 (2)i ... ... |m2 (N + 1)i ¯
¯ ¯ ¯
¯ ... ... ... ... ¯ ¯ ¯
¯ |mN (1)i |mN (2)i ... |mN (N)i ¯ ¯ ... ... ... ... ... ¯
¯ |mN (1)i |mN (2)i ... ... |mN (N + 1)i ¯

(we ignore the normalization constant for the moment). Since k can be any number, in
general the state of eq. 6.29 does not comply with the ordering defined by eq. 6.25. We
have to “transport” k to its proper position; i.e. if mi−1 < k < mi then the correct order
is

|m1 m2 ...mi−1 kmi ...mN i(a)

We can achieve this step by step from eq. 6.29 by interchanging k with its neighbor. Each
time we do a “−” sign appears.

|km1 m2 ....mN i(a) = −|m1 km2 ....mN i(a) =


(−1)2 |m1 m2 k....mN i(a) = (−1)Σi |m1 m2 ...mi−1 kmi ...mN i(a)

where Σi is the total number of states that are actually occupied and come before k in
the ordering of states. In other words

Σi = n1 + n2 + ... + ni−1 (6.30)

A similar sign convention is handy for the annihilation operator b


ck . Here we do the steps
in reverse; first transport k to the left most position

|m1 m2 ...mi−1 kmi ...mN i(a) = (−1)Σi |km1 m2 ...mi−1 mi ...mN i(a)

and then annihilate the particle in this state

ck |km1 m2 ....mN i(a) = |m1 m2 ....mN i(a)


b (6.31)

In summary, the sign (−1)Σi is produced as follows


140 CHAPTER 6. BOSONS AND FERMIONS

1. We define the creation, annihilation operators bc†k , b


ck such that they always operate
at the front of the anti-symmetric state |m1 m2 ....mN i(a) .

2. Then we reorder the states such that the rule of eq. 6.25 is obeyed.

The general result of all this book keeping is given by eq. 6.28. Apart from the
prefactors, these fermion creation/annihilation operators look just like the boson ones of
eq. 6.17. This is however slightly misleading. There is one very (and I mean very, very)
important difference. The fermion operators obey anti-commutation relations:
h i
ci , b
b c†j c†j + b
ci b
≡ b c†j b
ci = δ ij
+
h i
c†i , b
b c†j ci , b
= [b cj ]+ = 0 (6.32)
+

Note the “+” sign; an extremely important difference with bosons ! The relations are
easily proved from the definitions, eq. 6.28 (see also Mattuck, p. 130). I will give one
example, the rest you can easily prove yourself.

ci b
b c†i |n1 n2 ...ni ...nM i = (−1)Σi (1 − ni )b
ci |n1 n2 ...ni + 1...nM i
= (−1)Σi (1 − ni )(−1)Σi (ni + 1)|n1 n2 ...ni ...nM i
= (1 − n2i )|n1 n2 ...ni ...nM i

c†i b
b c†i |n1 n2 ...ni − 1...nM i
ci |n1 n2 ...ni ...nM i = (−1)Σi ni b
= (−1)Σi ni (−1)Σi (1 − ni + 1)|n1 n2 ...ni − 1...nM i
= (2ni − n2i )|n1 n2 ...ni ...nM i (6.33)

from which it follows

c†i + b
ci b
(b c†i b
ci )|n1 n2 ...ni ...nM i = (1 + 2ni − 2n2i )|n1 n2 ...ni ...nM i

Since ni can only be 0 or 1 (remember: fermions), 1 + 2ni − 2n2i = 1 and since this result
is true for any basis state, we have b c†i + b
ci b c†i b
ci = 1.

The difference in sign between the relations obeyed by fermion and by boson opera-
tors, eq. 6.32 vs. eq. 6.17 reflects the sign difference we introduced in the |....i(a) and
|....i(s) states, eq. 6.5 vs. eq. 6.3. This difference is very fundamental since it reflects the
anti-symmetry or symmetry of the many-particle states, and thus the difference between
fermions and bosons, the two fundamentally different kinds of elementary particles we
find in nature. You might already get some idea on the usefulness of second quantization.
Instead of the “sign book keeping” you have to do when manipulating the complex expres-
sions of the |....i(a) states, one can use the “sign rules” of the operator anti-commutation
relations of eq. 6.32. As always, operator algebra is simpler than manipulating states.
In order to make explicit use of the power of second quantization, one has to express
operators accordingly. Remarkably, the second quantization expressions for the one- and
two-particle operators for fermions have exactly the same form as for bosons, cf. eq. 6.19
6.2. N PARTICLES; THE MODERN ERA 141

and eq. 6.20.


N
X
b
h= b
h (i) ⇔
i=1
XM
b
h= c†k b
hkl b cl where (6.34)
k,l=1
D E
hkl = k(1)|b
h(1)|l(1)

N
b 1 X b
V = V (i, j) ⇔
2
j6=i=1
M
b 1 X
V = c†l b
Vklmn b c†k b
cm b
cn where (6.35)
2
k,l,m,n=1
D ¯ ¯ E
¯ ¯
Vklmn = k(1)l(2) ¯Vb (1, 2)¯ m(1)n(2)

Note again the k, l interchange between matrix element and operators in the two-particle
operator. Again, the proofs can be found in Appendix I. So the only difference between
bosons and fermions is commutation relations, eq. 6.17, vs. anti-commutation relations,
eq. 6.32 ! The operators

n c†i b
bi = b ci (6.36)

are again called the number operators. From eq. 6.33 it is seen that they give 0 if ni = 0
and 1 if ni = 1, as required. Again, as for bosons, the operator
M
X
b=
N bi
n (6.37)
i=1

is the total number operator, which counts the total number of particles present. The
states |n1 n2 ...ni ...nM i form the basis set for the (occupation) number representation.

6.2.3 The Road Travelled


“One small step for a man, a giant leap for mankind”, N. A. Armstrong.

Table 6.1 gives a road map for the change of representation we have performed in the
previous two sections.
Following Mattuck, I have rechristened the “old” notation as “first quantization” or
“stone-age notation”, indicating that it belongs to the dark ages. The new notation is
called “second quantization” or “number representation”, which is the language of modern
enlightened physicists. It is also the language of quantum field theory. The new language
has appeal for a number of reasons.

1. Calculations become easier. We just let the creation and annihilation operators do
their work. Like in the case of the harmonic oscillator we can get matrix elements
and expectation values with a minimum of work.
142 CHAPTER 6. BOSONS AND FERMIONS

1st quantization or 2nd quantization


stone-age notation number representation
states bosons 6.3 6.15
fermions 6.5,6.8 6.26
operators bosons 6.9,6.11 6.17,6.19,6.20
fermions 6.9,6.11 6.32,6.32,6.34,6.35

Table 6.1: Equation roadmap from 1st to 2nd quantization

2. An intuitive physical picture emerges through the notion of creating and annihilat-
ing particles. For instance, a Hamiltonian can contain a term

Vb ∝ b c†f b
ak b ci (6.38)

This annihilates a fermion in state |ii and creates a fermion in state |f i; or in other
words it transfers the fermion from state |ii to state |f i. At the same time a boson
in state |ki is annihilated. If the fermion is an electron in an atomic state i or f and
the boson is a photon with momentum ~k, then Vb describes an absorption process.

3. It is perfectly suitable for a diagrammatic representation by Feynman diagrams. The


Vb of eq. 6.38 describes what happens at the vertex (the point where the lines come
together) of the following diagram. More of such diagrams will appear in the future.

i f

=k

Figure 6.3: Absorption of a photon.

Note the resemblance with real physical processes like the ones shown in Fig. 5.4.
This resemblance is actually what led Feynman to invent and use his diagrams not
only for visualisation, but also for actual calculations.

4. Whereas in stone-age notation we are always obliged to work with a fixed total
number of particles N , in the number representation there is nothing in the final
equations which depends upon the number of particles; cf. Table 6.1. The expres-
sions have the same form whatever the number of particles. We can make particles
appear and disappear (provided we obey the basic conservation rules of energy and
momentum, for instance). This ability is certainly very useful when considering bo-
son particles such as photons or phonons, since these are easily created or annihilated
in many physical processes. For fermions we are less used to this idea. At first sight
it seems only useful in high energy physics, where e.g. an electron and a positron can
be created from a γ-ray photon in much the same process as is shown in Fig. 6.3;
the energy of the photon is on the scale of 106 eV. However, a direct parallel of such
a process exists also in condensed matter physics. In a semiconductor absorption of
6.3. THE PARTICLE-HOLE FORMALISM 143

a photon across the band gap creates an electron (in the conduction band) and a
hole (in the valence band); here the energy of the photon is on the scale of the band
gap, i.e. 1 eV.4 ,5

6.3 The Particle-Hole Formalism


“Entia non sunt multiplicanda praeter necessitatem”, William Occam (No more things should be
assumed to exist than necessary).

Each fermion occupies a different state according to Pauli. Filling a system with
fermions (e.g. electrons in a molecule or crystal), the system being in thermal equilibrium
at not too high a temperature, is like filling a bucket with water. The lowest level is
filled first, then the next lowest level etcetera. In a large system one fills energy levels
which are distributed over a range over tens or even hundredths of eV’s. In many (optical,
electrical transport) experiments we only probe the energy levels in a range of a few eV
around the highest occupied one. So it is a nuisance to have to drag along all these very
low lying energy levels which never change their occupation. To avoid this nuisance, the
second quantization formalism for fermions can be adapted somewhat. This adaptation is
called the particle-hole formalism. As usual, it is not only a formalism, but it also paints
a new physical picture. How the particle-hole formalism works is best explained using a
pet model of solid state physics, the so-called homogeneous electron gas. As much as the
harmonic oscillator or quantum chain is the prime example of a many-boson system, the
homogeneous electron gas is the prime example of a many-fermion system.

6.3.1 The Homogeneous Electron Gas


The homogeneous electron gas consists of an infinite number of electrons in infinite space,
such that their average density is a fixed ρ. To make the whole system neutral, the electrons
move in a fixed, positively charged background which has the same charge density ρ. The
positive background is necessary, otherwise the Coulomb repulsion between the infinite
number of electrons would lead to an infinite positive energy. The homogeneous electron
gas is also called “jellium”; it is the metallic equivalent of a homogeneous medium. It is
also a simple model for a metal, in which the positive charge of the nuclei or ion cores of the
metal is smeared out homogeneously over the background. The model is far less academic
than might seem at first sight; it turns out to be quite a reasonable first approximation
for many metals.
In this section we will consider the so-called non-interacting electron gas. That is, we
neglect all the electron-electron Coulomb repulsions. Since this repulsion is very large,
this may seem not at all appropriate. In fact, it is not at all appropriate, but we will
4
Hendrik Casimir was a famous dutch physicist who could (should?) have, but unfortunately did not
get the nobel prize. He mentions that Dirac’s electron-positron concept might have been inspired by the
electron-hole concept in semiconductors. H. B. G. Casimir, Het toeval van de werkelijkheid, (Meulenhoff,
Amsterdam, 1983) bevat de memoires van Casimir. Het is een leuk boek, vol anecdotes over mensen en
gebeurtenissen uit de begindecennia van de quantum mechanica. Het bevat o.a. zijn fameuze lezing over
“broken English”, de internationale taal van de wetenschap waarin ook deze lecture notes zijn geschreven.
5
The difference in energy scale of at least six orders of magnitude between condensed matter and high
energy physics means that we have to use relativistic quantum mechanics for the latter, whereas we can
stick to non-relativistic quantum mechanics (i.e. the Schrödinger equation) for the former.
144 CHAPTER 6. BOSONS AND FERMIONS

show in the last chapter how to introduce the Coulomb repulsion in a way such that the
physics of the interacting electron gas resembles that of the non-interacting electron gas
(by transforming particles into quasi-particles). We will leave that subject until later and
stick to the non-interacting gas for the moment. The Hamiltonian is
X p
b 2i
b
h= + Vb (6.39)
2me
i

The first term is the sum of kinetic energies of all the electrons. The second term is the
potential due to the uniform positive background. Since it is just a constant, we can
b2
p
neglect it. The eigenstates of the single particle Hamiltonian 2m e
are written as usual as
|ki; they are also the eigenstates of the momentum operator.

b |ki = ~k|ki
p
b2
p ~2 |k|2
|ki = ²k |ki with ²k = (6.40)
2me 2me
In the position representation, these momentum eigenstates are of course plane waves, as
in Chapter 4, see Section 4.2
1
hr|ki = √ eik·r ≡ φk (r) (6.41)

We are using a periodic box of volume Ω again to normalize these states and produce a
discrete spectrum in k. Electrons are fermions and we use these single particle states to
form anti-symmetric states for N particles; or, in stone-age notation6

|k1 k2 ....kN i(a) (6.42)

Filling the system with electrons, we start from the lowest energy level ²k = 0 for k =
(0, 0, 0), see eq. 6.40. All states |ki for which k has the same length |k| have the same
energy ²k . So filling the levels according to increasing energy ²k fills up spherical shells,
when we make a plot in k-space. This is shown in Fig. 6.4.The plot is in two dimensions
k = (kx , ky ); in three dimensions k = (kx , ky , kz ) the shells will of course be spheres.
In the ground state of our homogeneous electron gas all states k with ²k ≤ ²F will be
occupied and all states with ²k > ²F will be unoccupied.7 ²F is called the Fermi energy.
We can write it as
~2 kF2
²F = (6.43)
2me
6
Electrons also have a spin. In this course we will not be looking at any processes in which an interaction
with the electronic spin takes place. Under these circumstances, spin is then a “silent” quantum number,
and electrons behave just like spinless particles. Only when counting electrons we have to take into
consideration that each state |ki can in fact hold two electrons, one with spin up, and one with spin down.
Keeping this in the back of our mind, we will not use an explicit notation for the spin quantum number.
If one wishes to explicitly specify the spin state, one can always substitute |ki by |k; σ z i = |ki|σ z i with
σ z = ± 12 .
7
That is at zero temperature. At a finite temperature, there will be some thermal excitations, giving
rise to the Fermi-Dirac distribution you know from your statistical physics course. Since usually ²F & 10
eV and kT ≈ 0.026 eV at room temperature, the effect of thermal excitations on the occupation numbers
of the levels is small, except for very high temperatures.
6.3. THE PARTICLE-HOLE FORMALISM 145

k y contours of constant ε k

ε k1 ε k 2

kx

Figure 6.4: Shells of constant ²k .

kF is called the Fermi wave number ; all states with |k| ≤ kF are occupied. Plotted as in
Fig. 6.4 these k’s fill a sphere, called the Fermi sphere. Its surface is given by |k| = kF
and it is called the Fermi surface.8 We can calculate kF by counting the electrons in the
states. The total number of electrons is
kF
X
N= 2
k=(0,0,0)

since we can have 2 electrons per k state; one with spin-up and the other with spin-down.
In the continuum limit we have, see Section 4.2, eq. 4.28
X Z

...←→ ...d3 k (6.44)
(2π)3
k

which means
Z Z kF
N 1 3 2 kF3
≡ρ= 2d k= 4πk 2 dk =
Ω (2π)3 (2π)3 0 3π 2

Since we assumed the density ρ of the electrons to be given, we then easily find the
expressions
2
2 1 ~2 (3π 2 ρ) 3
kF = (3π ρ) 3 and ²F = (6.45)
2me

6.3.2 Particles and Holes


For realistic metallic densities ²F is in the range 10-20 eV. In many experiments we are
interested in excitations from the ground state which are at least an order of magnitude
8
If we introduce real atomic nuclei or ion cores into our system, the electron gas becomes inhomogeneous.
The fermi sphere gets distorted and can obtain a very complicated shape. However, the concepts of a Fermi
energy and a Fermi surface (though not spherical anymore) are still valid. They are very universal concepts
in solid state physics.
146 CHAPTER 6. BOSONS AND FERMIONS

smaller in energy, Eexc ¿ 1 eV. The majority of states with energies below ²F − Eexc are
untouched by the experiment and it is a nuisance to drag them along. For that reason we
introduce the particle-hole formalism (see also Mattuck §4.2, 4.3, and 7.5). We define two
sets of new fermion creation/annihilation operators for electrons and holes, respectively,
which are related to our previous fermion operators of eqs. 6.28 and 6.32 as

a†k = b
b c†k ; b
ak = b
ck if |k| > kF (electrons)
bb† = b
ck ; bbk = b
c†k if |k| ≤ kF (holes) (6.46)
k

For energies above ²F the operators b a†k , b


ak correspond to creating an electron in a state |ki,
with |k| > kF just as before (such states are unoccupied in the ground state). However,
for energies below ²F the annihilation of an electron by b ck is now associated with the
creation of a hole by bb†k (which makes sense, we are creating a hole in a state which is
occupied in the ground state). The annihilation of a hole by bbk is then obviously associated
with creating an electron by b c†k . The bak ’s and bbk ’s must obey the same anti-commutation
relations as the b
ck ’s, see eq. 6.32.
h i h i
b a†k0 = δ k,k0 ; bbk , bb†k0 = δ k,k0 etcetera
ak , b (6.47)
+ +

Furthermore the b ak ’s and bbk ’s always anti-commute, since they are related to b
ck ’s with
different k’s, eq. 6.46.
h i
ak , bbk0 = 0 etcetera
b (6.48)
+

c†k b
The Hamiltonian in b ck -number representation is given by
X
b
h= c†k0 b
hk0 k b ck
k0 ,k

since the Hamiltonian of eq. 6.39 only contains a one-particle operator. We have

b2
p
hk0 k = hk0 (1)|b
h(1)|k(1)i = hk0 (1)| 1 |k(1)i
2me
0
= ²k hk |ki = ²k δ k0 ,k

see eq. 6.40.9 . The Hamiltonian thus becomes


X
b
h= c†k b
²k b ck (6.49)
k

ak ’s and bbk ’s we get


In b
X X
b
h = ²kbbkbb†k + a†kb
²kb ak
|k|≤kF |k|>kF
X X X
= ²k − ²kbb†kbbk + a†kb
²kb ak
|k|≤kF |k|≤kF |k|>kF

9
Note the index (1) is used as a dummy index. We usually discard it when there is no confusion.
6.3. THE PARTICLE-HOLE FORMALISM 147

where we have obtained the last line by using the anti-commutation rule bbkbb†k + bb†kbbk = 1.
We can give this expression a new interpretation. First note that
X
E0 = ²k (6.50)
|k|≤kF

is the sum of energies of all states occupied in the ground state, i.e. the ground state energy.

Here comes the main idea: we consider the ground state as the new reference point
from which we create and annihilate particles. So we consider the ground state, being the
lowest energy state of a system of fermions, as our new vacuum state ! The Hamiltonian
is written as
X X
b
h = E0 − ²kbb†kbbk + a†kb
²kb ak (6.51)
|k|≤kF |k|>kF

Since ba†kb
ak = nbk are number operators, according to eq. 6.36, the last term of this
expression counts the number of electrons present with energies above the Fermi level.
Each electron can be assigned an energy ²k and a momentum ~k. The second term also
contains number operators bb†kbbk . It counts the number of a new kind of particle, which
is present below the Fermi level, and has an energy −²k . Originally it corresponded to
the annihilation of an electron with energy ²k ; so taking away an electron with energy
²k corresponds to creating a “new particle” with energy −²k . From the standpoint of
conservation of energy this makes perfect sense. Similarly one can argue that taking away
an electron with momentum ~k and charge −e corresponds to creating a new particle
with momentum −~k and charge +e (consider the laws of conservation of momentum and
charge). The Hamiltonian part
X
b
hh = − ²kbb†kbbk
|k|≤kF

has the same form as eq. 6.49. Since the latter has been derived from eq. 6.39, b
hh has the
same eigenstates, cf. eqs. 6.40 and 6.42, the only difference being the “−” sign in front of
the eigenvalues −²k . In particular, the “new particle” has the same wave functions of eq.
6.41. As you have guessed by now, the “new particle” created by bb†k is called a hole.

SUMMARY
In the particle-hole formalism, both particles and hole are created from the vacuum. The
vaccuum has the following properties

• The vacuum state is the ground state of the electron gas.


• The energy of the vacuum is the ground state energy E0 , given by eq. 6.50.

Particles (electrons) are created above the Fermi level; holes are created below the
Fermi level. Table 6.2 gives a comparison between the properties of particles and holes.
The particle-hole formalism describes both electrons and holes as additions to the
ground state, or in more fancy words as excitations of the vacuum. The electron is called
a particle, and its counter part the hole is called an anti-particle. The particle/anti-particle
idea is quite universal and commonplace.
148 CHAPTER 6. BOSONS AND FERMIONS

particle hole
creation a†k
b bb†
k
energy ²k −²k
momentum ~k −~k
charge −e +e
wave function φk (r) φk (r)
particle anti-particle

Table 6.2: Particles and holes

LOOSE ENDS
1. You might worry about the sign of the energy −²k of the hole. Actually, the signs on
the energy levels have little meaning. It is only their relative position with respect
to each other (and the Fermi level) that is important. We can easily change the sign
by choosing a different zero-point for the potential. As we stated in the beginning
of this section, the Hamiltonian contains a constant potential Vb , cf. eq. 6.40. It can
be written as
X
Vb = c†k b
V0 b ck
k
X X X
= V0 − V0bb†kbbk + a†kb
V0b ak
|k|≤kF |k|≤kF |k|>kF

analogous to eqs. 6.49 and 6.50. If one sums this term to the Hamiltonian of eq.
6.51 one obtains a new Hamiltonian

h0 = b
b h + Vb
X X
= E00 − ²0kbb†kbbk + a†kb
²0kb ak where
|k|≤kF |k|>kF
X
²0k = ²k + V0 and E00 = ²0k (6.52)
|k|≤kF

The new Hamiltonian is the same as the old one, but with all the energy levels
shifted by the constant V0 . This has little meaning however, because as always in
physics one can choose the zero-point of a potential where ever one likes. Absolute
energies have little meaning, it is only relative energies that count. The old levels ²k
were chosen such that they were all positive, cf. eq. 6.51 and Fig. 7.3; this makes
the energies −²k of the holes all negative. If V0 is a large negative number, which is
likely since it represents the potential of a background positive charge which keeps
the electron gas together, then ²0k is negative for all ²k < |V0 |, according to eq. 6.52.
For corresponding holes, −²0k is then positive. It does not matter, the physics stays
the same; only the relation between particle and hole energies via the “−” sign as
in Table 6.2 is important.

2. Some authors prefer to choose the arbitrary zero point of the potential such that
V0 = −²F . Then all the energy levels are given with respect to the Fermi level, i.e.
²0k = ²k − ²F . The energy for all particle states, where |q| >kF , and thus ²q > ²F , is
then positive. Also the energy of all hole states, where |k| ≤kF , and thus ²k ≤²F , is
6.3. THE PARTICLE-HOLE FORMALISM 149

εq ' = εq − ε F
εq
εF

−ε k ' = ε F − ε k

εk

Figure 6.5: Particle and hole energies relative to the Fermi level.

positive, since −²0k = ²F − ²k > 0. This is depicted in Fig. 6.5. Although this gives
a nice “symmetric” energy level spectrum, we will stick to the old Hamiltonian of
eq. 6.51, with positive particle and negative hole energies.

6.3.3 The Quantum Field Theory Connection


When we quantized the EM field in Section 5.5, we ended up with a Hamiltonian (eq.
5.79) which in discretized form, using eq. 6.44, can be written in simplified form as
X 1
b =
H ~ωk ( + ba†kb
ak )
2
k
X
= E0 + a†kb
~ω k b ak
k
P
This looks very much like eq. 6.51, with E0 = 12 k ~ωk being the energy of the ground
state, the vacuum, and ²k = ~ω k = ~c|k|. In case of the EM field b a†k creates a boson
particle called a photon. As discussed in the previous chapter, this Hamiltonian describes
the quantum EM field, also called the photon field. The particles, photons, are excitations
of the vacuum state of this field.

The Hamiltonian of eq. 6.51 has a similar form. By analogy, it is said to describe the
quantum field for electrons, or the electron field. Two kinds of particles are associated with
it; electrons and holes (actually holons would be a better word), being created by b a†k and bb†k
respectively. Between these two particles exists a particle/anti-particle relationship. This
description is called quantum field theory (which in this context is a synonym for second
quantization or occupation number representation). In quantum field theory everything
(all matter, light, you name it) is described by a specific (quantum) field. There are
fields for photons (the EM field), electrons, phonons, protons, etcetera; all the objects
you know from solid state or elementary particle physics. Each field has a Hamiltonian
and an equation of motion (such as the wave equation for the EM field).10 All particles
10
The equations of motion can be derived from the (classical) Hamiltonian. See, e.g., H. Goldstein et
al., Classical Mechanics, (Addison Wesley, San Francisco, 2002).
150 CHAPTER 6. BOSONS AND FERMIONS

mentioned are excitations of (the vacuum state of) their quantum field. Bosons fields like
the EM field, have only one kind of particle. Fermion fields, like the electron field, have
two kinds of particle; a particle and an anti-particle. This point of view carries over from
condensed matter physics to high energy physics, which each have their own elaborate sets
of particles and anti-particles.11

So far, we have been considering particles that do not interact with each other. This
would make the world very easy to understand, but also very boring. However,

• Fields of a different kind can interact with each other. For instance, electrons interact
with photons; Figs. 3.2, 5.5, 6.3. It is custom in such diagrams to represent fermions
by solid lines and bosons by “wavy” lines or dashed lines.

• Particles of one field can interact with each other. For instance, eqs. 6.20 and 6.21
describes the Coulomb repulsion between electrons.

6.4 Second Quantization and the Electron Field


In this section I want to spend some words on the origin of this seemingly strange phrase
“second quantization”. When we quantized the electro-magnetic (EM) field in free space
we had a classical wave equation to start from, which can be derived from Maxwell’s
equation, cf. Section 5.5. For the vector potential A (r, t) it reads

1 ∂ 2 A (r, t)
∇2 A (r, t) − =0 (6.53)
c2 ∂t2
and we derived its general solution (cf. eqs. 5.68 and 5.69)

√ Z
Ω X
A (r, t) = d3 k Ak,s (t) εk,s eik·r (6.54)
(2π)3
s=1,2

The functions eik·r are the modes in free space; εk,s are the polarization vectors; and the
coefficients Ak,s are the amplitudes. The time dependence of the latter has the general
form

Ak,s (t) = ak,s e−iωk t + a∗−k,s eiωk t (6.55)

and in order that eqs. 6.54 and 6.55 give the solution to eq. 6.53, the dispersion relation
is

ω k = c|k| (6.56)

The move from classical electrodynamics to quantum electrodynamics was made by making
the amplitudes quantum operators

Ak,s → qbk,s (6.57)


11
The physics can be slightly different, however. In high energy physics, one can also have boson fields
that have particles and anti-particles
6.4. SECOND QUANTIZATION AND THE ELECTRON FIELD 151

and leaving all the other classical objects and relations untouched. From there on, all we
had to do to derive the states of the quantum EM field (photons), is to use eq. 6.57 to
transform the classical energy into a quantum mechanical Hamiltonian.

Now consider the following equation

~2 2 ∂ψ (r, t)
− ∇ ψ (r, t) − i~ =0 (6.58)
2m ∂t
We immediately recognize this as the Schrödinger wave equation of a single electron in
free space. Its general solution can be described in the same way as in eq. 6.54, i.e. as a
Fourier integral.
√ Z

ψ (r, t) = d3 k ck (t) eik·r (6.59)
(2π)3

where we recognize eik·r as the eigenfunctions for the free electron, which we can interprete
as the “modes”. The coefficients ck (t) then give the amplitudes and phases, as before.
There are however a couple of differences with the EM field.

1. Since ψ (r, t) is a scalar function, and not a vector function like A (r, t) we need no
polarization vectors.

2. The time dependence of its amplitudes are given by

ck (t) = ck e−iωk t (6.60)

In order that eqs. 6.59 and 6.60 present a solution to the Schrödinger equation of
eq. 6.58, the dispersion relation must be

~2 |k|2
~ω k = (6.61)
2m

p2
The latter is of course the familiar relation between energy and momentum E = 2m if
one accepts Planck18 ’s and De Broglie29 ’s relations E = ~ω, and p = ~k. The fact that the
relation between frequency and wave number has to be quadratic instead of simply linear
as in eq. 6.53 urged Schrödinger to derive a “wave” equation which has a first order time
derivative, cf. eq. 6.58, instead of the second order time derivative of a conventional wave
equation like eq. 6.53. Of course he also had to put an “i” in front of the time derivative,
since otherwise it would not have been a wave equation, but a “diffusion” equation which
only has non-wave like e±κ·r solutions.

Now suppose poor me is just a simple chemist who does not understand quantum
mechanics very deeply.12 I have missed all the introductory courses in quantum mechanics
and I just heard about quantizing the EM field following the prescription going from eqs.
6.53 to 6.57. Being faced with eq. 6.58, I am thinking that this is just a “classical”
wave equation like eq. 6.53. The function ψ (r, t) then plays the role of a classical wave,
or classical field as in the previous chapter. It is a scalar, and not a vector like the
12
All three statements are true.
152 CHAPTER 6. BOSONS AND FERMIONS

EM potential, but that makes it only simpler. It describes electrons, so I am told by


Schrödinger, so I call it the “classical” electron field. You know that this is wrong, since
you are experts in quantum mechanics, but stay with me for the sake of the argument.
I think I know how to quantize this “classical” electron field. I simply follow the same
prescription of quantum electrodynamics and turn the amplitudes into operators

ck → b
ck (6.62)

One of my clever fellow students told me that the energy an electron wave can be calculated
as
Z · ¸
3 ∗ ~2 2
E = d r ψ (r, t) − ∇ ψ (r, t) (6.63)
2m
Using eq. 6.59 and doing some δ-function manipulation as in the previous chapter this
can be rewritten as
Z

E= d3 k ~ω k c∗k ck (6.64)
(2π)3
with ~ω k given by eq. 6.61. This looks fine by me; essentially it is just Planck’s relation
applied to each mode, and all modes are summed over using their “intensity” |ck |2 as
weight factors. So now I say, let’s turn this “classical” energy into a quantum mechanical
Hamiltonian by using the prescription of eq. 6.62.
Z
b Ω
H= d3 k ~ωk bc†k b
ck (6.65)
(2π)3
But this looks just like the Hamiltonian in number representation of the homogeneous
electron gas, cf. eq. 6.49 ! We have to set ~ω k = ²k , but that is o.k. if we compare eqs.
6.40 and 6.61. Furthermore, we have switched from a sum over k to an integral, i.e. from
a discrete to a continuous spectrum. But we have done this before; it poses no problems.
Switching back to a discrete spectrum in a box using eq. 6.44, we get eq. 6.49
X
Hb= c†k b
~ω k b ck
k

Since the b ck ’s are operators, we have to figure out what their algebra is, i.e. their
(anti)commutation relations. But that is what we have been doing in this chapter all
along, so we already know the answer: since electrons are fermions, the b ck ’s are re-
quired to obey the anti-commutation rules of eq. 6.32. I have found the solution for
the quantum electron field (see also Mattuck, the end of chapter 7). For the EM field this
procedure gave the many photon states. Each photon can be created by an operator b a†k
working on the (photon) vacuum and in this sense it is called an excitation of the EM field.
c†k working on the (electron) vacuum; it
Similarly, each electron is created by an operator b
can be called an excitation of the electron field.

Now of course, after I did all this work, you come and tell me that the Schrödinger wave
equation, eq. 6.58, is in fact not a “classical” wave equation at all, but a quantum wave
equation. No problem, I say, let’s call the Schrödinger wave equation first quantization and
the step of eq. 6.62, or equivalently the step from eq. 6.64 to 6.65, second quantization.
6.4. SECOND QUANTIZATION AND THE ELECTRON FIELD 153

The result of quantizing the amplitudes ck of the modes of the Schrödinger wave equation
is a many-body Hamiltonian in second quantized form. Since that is exactly what we have
been deriving in this chapter all along, the procedure must be right. Such are the wonders
of quantum field theory.

Note that we just as well could have made the substitution of eq. 6.62 in eq. 6.59.
b (r, t)
From the latter (and its complex conjugate) the so-called electron field operators ψ

b (r, t) are defined as
and ψ
√ Z
b (r, t) = Ω
ψ d3 k b
ck (t) eik·r ; b ck (t) = bck e−iωk t
(2π)3
√ Z
b † (r, t) = Ω
ψ c†k (t) e−ik·r ; b
d3 k b c†k (t) = b
c†k eiωk t (6.66)
(2π)3
They play a similar role for the quantum electron field as the electric and magnetic field
operators in the EM theory, cf eq. 5.75 in Section 5.5, Chapter 5. The Hamiltonian then
becomes
Z · ¸
b 3 b† ~2 2 b
H = d r ψ (r, t) − ∇ ψ (r, t) (6.67)
2m

by (second) quantization of eq. 6.63. One can easily check that this is equivalent to eq.
6.65. The annihiliation operator b ck “lowers the amplitude of the wave” by annihilating a
particle in the mode characterized by wave vector k, cf. eq. 6.59. The creation operator
c†k obviously creates a particle in this mode. According to eq. 6.41 one can also write
b

c†k |0i → |ki


b (6.68)

where the left hand side is in second quantization, and the right hand side is in stone-age
notation. Then also, using eqs. 6.66
Z Z

b
ψ (r) |0i = 3
d ke −ik·r †
ck |0i = d3 k e−ik·r |ki
b
Z
= d3 k hk|ri |ki = |ri (6.69)

b † (r) creates a par-


(apart from normalization factors). Thus the electron field operator ψ
b (r) then annihilates a particle at a
ticle at a position r. The electron field operator ψ
position r.

Obviously I did not think of this myself; the quantum field procedure is a classic in
quantum mechanics by now. You can start from a wave-like equation like eqs. 6.53 or
6.58 and find the solution in terms of modes, eqs. 6.54 or 6.59. Then you quantize the
amplitudes, eqs. 6.57 or 6.62 and use this to construct a quantum Hamiltonian from the
“classical” energy, eqs. 5.76 or 6.63. In case the wave equation is already quantum me-
chanical, such as the Schrödinger equation, you call it “first quantization” and quantizing
the amplitudes “second quantization”. The eigenstates of the resulting Hamiltonian, eqs.
5.77 or 6.65, are many-particle states. The nature of the particles, whether they are bosons
or fermions, is only determined by the algebra of their creation/annihilation operators,
154 CHAPTER 6. BOSONS AND FERMIONS

i.e. whether these commutate or anti-commutate. It works for any kind of wave equation
and any kind of particle in any kind of physics (condensed matter, high energy, etcetera).
I think that, in view of the cumbersome algebra required for the stone-age quantum way
(see also Appendix I), we can all agree that this simple route followed by quantum field
theory is much faster. Moreover it applies to situations such as the EM field, in which the
stone-age quantum route to many particles by first quantization does not work. For the
EM field, “first quantization” gives the modes of the classical wave equation. It is only
“second quantization” that gives you photons and quantum mechanics.

6.5 Appendix I. Identical Particle Algebra


In this Appendix I have gathered the bulk of the algebra that is needed to prove some
of the statements made it in the text. In the first subsection it is proved that different
(anti)symmetric states are orthogonal, and the normalization factors are calculated. In
the second subsection the second quantization form of the operators given in the text is
proved to be correct. The algebra is lengthy, but straight-forward, as the textbook cliché
states.

6.5.1 Normalization Factors and Orthogonality


The factors Ns and Na are defined such that the states |....i(s) and |....i(a) are normalized;
i.e. (s) h....|....i(s) = 1 and (a) h....|....i(a) = 1, cf. eqs. 6.3, 6.5 and 6.7. We do the anti-
symmetric state first. Calculating (a) h....|....i P
(a) by the right hand side of eq. 6.5, leads to

a double summation over permutations Pb ,Pb 0 which has (N !)2 terms. These terms can
be divided into two classes

1. Pb =
6 Pb0 ; the “ off-diagonal ” terms. The corresponding matrix element in (a) h....|....i(a)
is
hPbm1 (1) m2 (2) ....mN (N) |Pb0 m1 (1) m2 (2) ....mN (N)i

Because the two permutations Pb, Pb0 are different from one another, there is at least
one particle i where on the left (the bra) side we have hmj (i)| and on the right (the
ket) side we have |mk (i)i, where mj 6= mk . (Note that because of Pauli’s exclusion
principle all the numbers mi ; i = 1, ..., N must be different from one another). If
we work out this matrix element as in Section 1.4, eq. 1.26, we see that this leads
to a factor hmj (i)|mk (i)i = 0, since the single particle basis set we started from is
orthogonal. In conclusion, all “off-diagonal” terms give a zero contribution.
2. Pb = Pb0 ; the “ diagonal ” terms. These are all of the same type

hm1 m2 ....mN |m1 m2 ....mN i = hm1 |m1 ihm2 |m2 i...hmN |mN i = 1
because our single particle basis set is orthonormal. Since there are N ! possible
permutations, there are N ! of these “diagonal” terms.

Combining all the terms gives


X
(a)
h....|....i(a) = (Na )2 = (Na )2 N ! = 1
Pb ,Pb 0
6.5. APPENDIX I. IDENTICAL PARTICLE ALGEBRA 155

and thus the normalization factor is


1
Na = √ (6.70)
N!
Along the same lines one can prove
N
Y
(a)
hm01 m02 ....m0N |m1 m2 ....mN i(a) = δ m0k ,mk (6.71)
k=1

i.e. the various anti-symmetric states are orthogonal. It goes as


(a)
hm01 m02 ....m0N |m1 m2 ....mN i(a) =
X 0
(Na )2 (−1)p+p hPb0 m01 (1) m02 (2) ....m0N (N ) |Pbm1 (1) m2 (2) ....mN (N )i =
Pb ,Pb 0
X
(Na )2 (−1)2p hPbm01 (1) m02 (2) ....m0N (N ) |Pbm1 (1) m2 (2) ....mN (N)i =
Pb
N
Y
2
(Na ) N ! δ m0k ,mk
k=1

The step from the second to the third line can be made because we have defined the anti-
symmetric state such that m1 < m2 < ... < mN and m01 < m02 < ... < m0N . This means
that the “off-diagonal” terms certainly give zero contribution (check this yourself). The
“diagonal” terms all give the same contribution, and there are N ! of them.

For the symmetric state |....i(s) things are a bit trickier, since some of the numbers
mi ; i = 1, ... can be the same (there is no Pauli principle for bosons). Suppose we have a
state
|m1 (1) m1 (2) ...m1 (n1 )m2 (n1 + 1)...m2 (n1 + n2 )...mM (N )i
i.e. n1 particles
P in state |m1 i, n2 particles in state |m2 i etcetera. The sum over all N !
permutations Pb in eq. 6.3 then contains only
N!
n1 !n2 !...nM !
distinct terms (assuming we have only M different states mi ), since interchanging particles
which are in the same state does not give a new term. (s) (s)
P When we calculate h....|....i ,
there are again two types of terms in the double sum Pb ,Pb 0 .
1. The “ off-diagonal ” terms between states which are distinct in the sense described
above. This all give zero for the same reason as in the case of the anti-symmetric
state. They always contain a factor of type hmj (i)|mk (i)i where mj 6= mk .
2. The “ diagonal ” terms between states which are not distinct in the sense described
above. There are a whole bunch of them which result from permutating particles
which are in the same state. Permutating the n1 particles in state |m1 i gives n1 !
terms, all identical. In the double sum these give (n1 !)2 matrix elements. All these
matrix elements are 1, as before, because we have an orthonormal basis. We can do
the same with permuting the n2 particles in state |m2 i, etcetera. These terms thus
give a contribution (n1 !)2 (n2 !)2 ...(nM !)2
156 CHAPTER 6. BOSONS AND FERMIONS

N!
Since we have n1 !n2 !...nM ! of these “diagonal” terms which give a contribution (n1 !)2 (n2 !)2 ...(nM !)2 ,
we get

X N!
(s)
h....|....i(s) = (Ns )2 = (Ns )2 (n1 !)2 (n2 !)2 ...(nM !)2 = 1
n1 !n2 !...nM !
Pb ,Pb 0

and thus the normalization factor is

1
Ns = √ (6.72)
N !n1 !n2 !...nM !

The orthogonality of the symmetric states is proved along the same lines

N
Y
(s)
hm01 m02 ....m0N |m1 m2 ....mN i(s) = δ m0k ,mk (6.73)
k=1

6.5.2 Second Quantization for Operators

Here we prove that the second quantization form of the operators given in the text, i.e.
eqs. 6.19, 6.20, 6.34, and 6.35, is correct. The proof proceeds by showing that the matrix
elements of operators in stone-age form, eq. 6.9—6.12 are identical to the matrix elements
of operators in second quantized form. Since this holds for any possible matrix element,
and the states form a basis set, this proves that the two operator form are identical.
Calculating matrix elements in stone-age form is the largest task. It can be shortened a
little bit by using a trick in which one initially considers special states only, and only in
the final stage one generalizes the result.

Stone-age Matrix Elements

First we will consider how one- and two-particle operators work on simple product states
in stone-age notation. A typical matrix element of a one-particle operator is

N
X
hm01 (1) m02 (2) ....m0N (N ) | b
h (i) |m1 (1) m2 (2) ....mN (N )i
i=1
N
X Y
= hm0i (i) |b
h (i) |mi (i)i hm0k (k) |mk (k)i
i=1 k6=i
N
X Y
= hm0i (i) |b
h (i) |mi (i)i δ m0k ,mk (6.74)
i=1 k6=i
6.5. APPENDIX I. IDENTICAL PARTICLE ALGEBRA 157

by a straight-forward generalization of eq. 1.28 in Section 1.4. In a similar way one can
show that a matrix element of a two-particle operator becomes
N
1 X b
hm01 (1) m02 (2) ....m0N (N ) | V (i, j) |m1 (1) m2 (2) ....mN (N )i
2
j6=i=1
N
1 X Y
= hm0i (i) m0j (j) |Vb (i, j) |mi (i) mj (j)i hm0k (k) |mk (k)
2
j6=i=1 k6=i,k6=j
N
1 X Y
= hm0i (i) m0j (j) |Vb (i, j) |mi (i) mj (j)i δ m0k ,mk (6.75)
2
j6=i=1 k6=i,k6=j

We will now limit our discussion to special single-particle states |mi i that are eigenstates
b
of h (i) and Vb (i, j) . Later on we will generalize the discussion again to a more general
basis set. So
b
h (i) |mi (i)i = hmi |mi (i)i (6.76)

with hmi as the eigenvalue. We keep the dummy label (i) here because we will need it
later on. The eigenstate of the two-particle operator Vb (i, j) is of course a two-particle
state

Vb (i, j) |mi (i) mj (j)i = Vmi mj |mi (i) mj (j)i (6.77)

where Vmi mj is the eigenvalue. If you wonder whether this procedure is correct, for the
proof it is not really necessary that the states are simultaneous eigenstates of b h (i) and
b b
V (i, j). We could consider one basis set to complete the proof for h (i) and another one
for Vb (i, j). For instance, if b 1
h (i) = 2m pi |2 , the kinetic energy of a particle, one would use
|b
a basis set of momentum eigenstates, i.e. |mi (i)i = |pi i. If Vb (i, j) = e2 |b rj |−1 , the
ri − b
Coulomb repulsion between two particles, one would use a basis set of position eigenstates,
i.e. |mi (i) mj (j)i = |ri i|rj i, which has the obvious eigenvalues Vmi mj = e2 |ri − rj |−1 . In
order not to complicate our notation too much, we stick to eqs. 6.76 and 6.77.
Using this special set of states the calculation of matrix elements become easier

hm0i (i) |b
h (i) |mi (i)i = hmi hm0i (i) |mi (i)i = hmi δ m0i ,mi
hm0i (i) m0j (j) |Vb (i, j) |mi (i) mj (j)i = Vmi mj hm0i (i) m0j (j) |mi (i) mj (j)i
= Vmi mj δ m0i ,mi δ m0j ,mj

which simplifies eqs. 6.74 and 6.75 to


N
X
hm01 (1) m02 (2) ....m0N (N ) | b
h (i) |m1 (1) m2 (2) ....mN (N )i
i=1
N
X N
Y
= hmi δ m0k ,mk
i=1 k=1
XN
= hmi hm01 (1) m02 (2) ....m0N (N ) |m1 (1) m2 (2) ....mN (N )i (6.78)
i=1
158 CHAPTER 6. BOSONS AND FERMIONS

and

N
1 X b
hm01 (1) m02 (2) ....m0N (N ) | V (i, j) |m1 (1) m2 (2) ....mN (N )i
2
j6=i=1
N
X N
Y
1
= Vmi mj δ m0k ,mk
2
j6=i=1 k=1
N
X
1
= Vmi mj hm01 (1) m02 (2) ....m0N (N ) |m1 (1) m2 (2) ....mN (N )i (6.79)
2
j6=i=1

Now we attack the more complicated task of calculating matrix elements between
(anti)symmetric states, since these are the correct basis states for identical particles.
P b We
show the algebra for fermions only; for bosons it is similar. So writing bh= N i=1 h (i) as
before

hm01 m02 ....m0N |b


(a)
h|m1 m2 ....mN i(a) =
X 0
Na2 (−1)p+p hPb0 m01 (1) m02 (2) ....m0N (N ) |b
h|Pbm1 (1) m2 (2) ....mN (N )i (6.80)
Pb ,Pb 0

One word about how the permutation operators work. The operators Pb, Pb0 permutate the
particles 1, ..., N and one should write

|Pbm1 (1) m2 (2) ....mN (N )i = |m1 (P 1) m2 (P 2) ....mN (P N )i

Since a permutation does not alter the states but only redistributes the particles, we can
also write

|Pbm1 (1) m2 (2) ....mN (N )i = |mP −1 1 (1) mP −1 2 (2) ....mP −1 N (N )i

i.e. pretend we keep the particles fixed but permutate the states. In order that the two
expressions are the same, we have to permutate the states using the inverse permutation
P −1 . In the following this is only a minor point since we sum over all permutation anyway,
and if a permutation P is even (odd) than its inverse is also even (odd).
One observes that all the terms in the double sum of eq. 6.80 obtain a form as in eq.
6.5. APPENDIX I. IDENTICAL PARTICLE ALGEBRA 159

6.78. We can write


(a)
hm01 m02 ....m0N |b
h|m1 m2 ....mN i(a)
X N
X
2 p+p0
= (Na ) (−1) hmP −1 i
Pb ,Pb 0 i=1

hm0P 0−1 1 (1) m0P 0−1 2 (2) ....m0P 0−1 N (N) |mP −1 1 (1) mP −1 2 (2) ....mP −1 N (N )i
N
X X 0
= hmi (Na )2 (−1)p+p
i=1 Pb ,Pb 0
hmP 0−1 1 (1) mP 0−1 2 (2) ....m0P 0−1 N (N) |mP −1 1 (1) mP −1 2 (2) ....mP −1 N
0 0
(N )i
XN
= hmi (a) hm01 m02 ....m0N |m1 m2 ....mN i(a)
i=1
XN YN
= hmi δ m0k ,mk (6.81)
i=1 k=1
P PN
The step from the second to the third line can be made since N i=1 hmi = i=1 hmP −1 i ,
i.e. the mP −1 i labels are just permutated mi labels, and as long as we sum over all of
them, nothing changes. The rest essentially follows from eq. 6.71.

Number Representation for Special One-particle Operators


Now we switch to number representation and we write
|m1 m2 ....mN i(a) = |n1 n2 ....nM i
But we have to be careful ! In stone-age notation only those states |mi i are listed that
are occupied by particles (there are N of such states). In number representation the
occupations of all possible states |mi i is mentioned (there are M of them, and usually
M À N ). The ones |mj i that are not occupied simply get an occupation number nj = 0.
P
The sum N i=1 hmi in eq. 6.81 is again only over the states that are actually occupied. So
if we want to write this matrix element using number representation, we must find a way
to single out only the occupied states. The trick is following
(a)
hm0 m0 ....m0 |b
h|m1 m2 ....mN i(a)
1 2 N
M
X
= hn01 n02 ....n0M | bi |n1 n2 ....nM i
hmi n (6.82)
i=1
M
X
= hmi ni hn01 n02 ....n0M |n1 n2 ....nM i
i=1
N
X
(a)
= hmi hm01 m02 ....m0N |m1 m2 ....mN i(a)
i=1
The trick is to sum over all possible states, but multiply each term with the number
operator. Since
bi |n1 n2 ...ni ...nM i = ni |n1 n2 ...ni ...nM i
n
160 CHAPTER 6. BOSONS AND FERMIONS

and for the unoccupied states ni = 0, only the occupied states are filtered out as re-
quired. Neat trick, eh ? Now, since eqs. 6.81 and 6.82 give the same result between
any pair of states, and since these states span the entire N -fermion space, we must have
the operator identity
M
X M
X
b
h= bi =
hmi n c†i b
hmi b ci (6.83)
i=1 i=1

Number Representation for Special Two-particle Operators


The two-particle operator can be handled in the same way. As a first step one can prove,
completely analogous to eq. 6.81.
(a)
hm01 m02 ....m0N |Vb |m1 m2 ....mN i(a)
N
1 X (a)
= Vmi mj hm01 m02 ....m0N |m1 m2 ....mN i(a) (6.84)
2
j6=i=1

If we want to turn this into number representation we again must find a trick to single
out the occupied states only. Following eq. 6.82 we write the ansatz
(a)
hm01 m02 ....m0N |Vb |m1 m2 ....mN i(a)
M
1 X bij |n1 n2 ....nM i
= hn01 n02 ....n0M | Vmi mj Q (6.85)
2
i,j=1
N
1 X (a)
= Vmi mj hm01 m02 ....m0N |m1 m2 ....mN i(a)
2
j6=i=1

A bit of contemplation reveals that the third line results from the second if we define
b ij = n
Q bi n
bj − δ ij n
bi (6.86)

The δ ij term serves to filter out the i = j term in the double sum. Using the fermion
anti-commutation relations we can rewrite this operator a bit
bij = b
Q c†i b c†j b
ci b c†i b
cj − δ ij b ci
³ ´
c†i δ ij − b
= b c†j b
ci b c†i b
cj − δ ij b ci
c†i b
= −b c†j b cj = b
ci b c†i b
c†j b
cj b
ci (6.87)

Again, eq. 6.85 is valid for every possible combination of basis states, so we must have
the operator identity
M
1 X
Vb = c†i b
Vmi mj b c†j b
cj b
ci (6.88)
2
i,j=1

Eqs. 6.83 and 6.88 are almost what we want.


6.5. APPENDIX I. IDENTICAL PARTICLE ALGEBRA 161

Number Representation for General One- and Two-particle Operators


We have given the proof for a special set of states, namely eigenstates of b
h and Vb , cf. eqs.
6.76 and 6.77. What remains now is to generalize it to any sort of basis set |kj i. The
latter can always be expanded in the “special” basis set |mi i and vice versa.
M
X
|kj i = |mi ihmi |kj i
i=1
XM
|mi i = |kj ihkj |mi i (6.89)
j=1

using the resolution of identity as in the first chapter. We now switch to a notation
c†mi ≡ b
b c†i for the creating a particle in the “special” state |mi i, in order to distinguish it

from bckj , which creates a particle in the “general” state |kj i. The two can be related as

c†m |0i
|mi i = b
Xi X †
= |kj ihkj |mi i = b
ckj |0ihkj |mi i
j j

This can only be true if


M
X
c†mi =
b c†kj hkj |mi i
b (6.90)
j=1

From this one easily proves


M
X
b
cmi = b
ckj hmi |kj i
j=1
M
X
c†kj
b = c†mi hmi |kj i
b
i=1
M
X
b
ckj = b
cmi hkj |mi i (6.91)
i=1

Writing hmi = hmi |b


h (1) |mi i using eq. 6.7613 , we can now transform eq. 6.83 to a general
basis set
M
X M
X M
X
b
h= c†mi b
hmi b cmi = b
hmi |h (1) |mi i c†kj hmi |kl ib
hkj |mi ib ckl
i=1 i=1 j,l=1

We reorder the terms in this triple sum a bit and note that
M
X
hkj |mi ihmi |b
h (1) |mi ihmi |kl i
i=1
XM
= hkj |mi ihmi |b
h (1) |mp ihmp |kl i = hkj |b
h (1) |kl i
i,p=1
13
Again the particle label (1) is dummy; it does not matter which particle we take, they are all the same.
162 CHAPTER 6. BOSONS AND FERMIONS

The first step follows from the fact that b


h (1) is diagonal
P on the “special”
P basis |mi i; the
second step simply removes the resolutions of identity i |mi ihmi | and p |mp ihmp |. We
arrive at the number representation form of the one-particle operator on a general basis
set
M
X
b
h = c†kj b
hkj kl b ckl with
j,l=1

hkj kl = hkj (1)|b


h (1) |kl (1)i (6.92)

In a similar way we can transform eq. 6.88 to its most general form
M
X
1
Vb = c†kl b
Vkk kl km kn b c†kk b
ckm b
ckn with
2
k,l,m,n=1

Vkk kl km kn = hkk (1)kl (2)|Vb (1, 2) |km (1)kn (2)i (6.93)

Note again the interchange of the k, l indices of the matrix element, as compared to the
creation operators.

So finally we have proved eqs. 6.34 and 6.35. It is worth the trouble however, since
with these two expressions we can avoid the cumbersome calculation of matrix elements
in stone-age notation like we had to do in this appendix. Instead, in future we will focus
upon operator algebra, i.e. (anti-)commutation rules, which are much easier to handle.

6.6 Appendix II. Identical Particles


This appendix was essentially taken from A. Messiah, Quantum Mechanics (North-Holland,
1961, Ch. XIV). We will try and answer the following questions

1. What does it mean, identical quantum particles are indistinguishable ?

2. Why do the states of identical particles need to be symmetric or anti-symmetric ?

3. Do we have to form (anti)symmetric states using all identical particles, e.g. electrons,
in the universe ?

6.6.1 Indistinguishable Particles


We will use good old fashioned wave mechanics to answer these questions. Suppose we do
a scattering experiment, starting with two particles a large distance apart. Initially, one
particle is in a wave packet φ(r) that is fairly localized around r = r1 ; the other one is in
another wave packet ψ(r) localized around r = r2 . Both wave packets are well separated,
each of them localized in a different region of space. An artists impression of this situation
is given in Fig. 6.6.
The initial wave function describing both particles is given by

Ψ12 (r, r0 ) = φ(r)ψ(r0 ) (6.94)


6.6. APPENDIX II. IDENTICAL PARTICLES 163

φ(r ) ψ (r )
1st particle 2nd particle

r1 r r2

Figure 6.6: Two particles in two separated wave packets.

Indistinguishable particles means that there is absolutely nothing we can do to distinguish


the state Ψ12 from the state which is given by

Ψ21 (r, r0 ) = ψ(r)φ(r0 ) (6.95)

where particle 1 is in wave packet ψ(r) and particle 2 in wave packet φ(r). The states
Ψ12 and Ψ21 carry the same energy, momentum, etcetera. In practice it turns out that
absolutely any experiment we can think of gives the same result for Ψ12 and Ψ21 . Let
us consider the consequences. We prepare our scattering experiment such that φ(r) and
ψ(r) are different, i.e. linear independent functions. Since the wave packets for the two
particles are well separated in space, the functions Ψ12 (r, r0 ) and Ψ21 (r, r0 ) are also linear
independent. Now any linear combination of Ψ12 and Ψ21

Ψ(r, r0 ) = αΨ12 (r, r0 ) + βΨ21 (r, r0 ) (6.96)

with α, β constants, is also a function which gives the same result for all thinkable ex-
periments. For instance, if Ψ12 is an eigenstate of the Hamiltonian with energy E, then
also Ψ21 is, and thus also Ψ. This general conclusion cannot be right, as we will show
now. Nature must have found a way to fix the constants α, β to limit the Ψ’s that are
acceptable. (α = β gives a symmetric state, and α = −β gives an anti-symmetric state,
for instance).

6.6.2 Why Symmetrize ?


Let us start by changing the “basis” states from Ψ12 , Ψ21 to their symmetric and anti-
symmetric linear combinations
1 £ ¤
Ψs (r, r0 ) = √ Ψ12 (r, r0 ) + Ψ21 (r, r0 )
2
1 £ ¤
Ψa (r, r0 ) = √ Ψ12 (r, r0 ) − Ψ21 (r, r0 ) (6.97)
2
Since from eqs. 6.94 and 6.95 one has Ψ12 (r0 , r) = Ψ21 (r, r0 ), it follows

Ψs (r0 , r) = Ψs (r, r0 ) and Ψa (r0 , r) = −Ψa (r, r0 ) (6.98)

We write in terms of these functions

Ψ(r, r0 ) = cs Ψs (r, r0 ) + ca Ψa (r, r0 ) (6.99)


164 CHAPTER 6. BOSONS AND FERMIONS

with ca and cs constants. Eq. 6.99 is just another way of writing eq. 6.96. The “basis”
states Ψs and Ψa just prove to be more handy. Initially we had two separate wave packets,
but now we start our scattering experiment. When the two particles start to collide, their
wave packets start to overlap. We ask for the probability of finding a particle at the
position r and a particle at the position r0 at a time t after we started the experiment.
Particle 1 can be at r and particle 2 at r0 , or particle 1 at r0 and particle 2 at r, and these
two probabilities are independent. Thus the total probability is given by
¯ ¯2 ¯ ¯2
P (r, r0 , t) = ¯Ψ(r, r0 , t)¯ + ¯Ψ(r0 , r, t)¯ (6.100)

Using eqs. 6.98 and 6.99, it is easy to prove that Ψ(r0 , r) = cs Ψs (r, r0 ) − ca Ψa (r, r0 ) from
which one finds
¯ ¯2 ¯ ¯2
P (r, r0 , t) = |cs |2 ¯Ψs (r, r0 , t)¯ + |ca |2 ¯Ψa (r, r0 , t)¯ (6.101)

Eq. 6.101 is supposed to give the same result independent of the numbers ca and cs ! (Trace
back our reasoning: since we have two indistinguishable particles, any measurement must
give the same result for Ψ12 and Ψ21 , and thus for any linear combination of the two, Ψ).
If this is true then by choosing cs = 1 and ca = 0 or cs = 0 and ca = 1, it follows
¯ ¯ ¯ ¯
¯Ψs (r, r0 , t)¯ = ¯Ψa (r, r0 , t)¯ (6.102)

At the start of our experiment this is indeed true. We have from eq. 6.97
¯ ¯ ¯ ¯ ¯ ¯
¯Ψs (r, r0 , t)¯2 = 1 { ¯Ψ12 (r, r0 , t)¯2 + ¯Ψ21 (r, r0 , t)¯2 +
2
Ψ12 (r, r0 , t)Ψ21 (r, r0 , t) + Ψ∗21 (r, r0 , t)Ψ12 (r, r0 , t)}

¯ ¯ ¯ ¯ ¯ ¯
¯Ψa (r, r0 , t)¯2 = 1 { ¯Ψ12 (r, r0 , t)¯2 + ¯Ψ21 (r, r0 , t)¯2 −
2
Ψ12 (r, r0 , t)Ψ21 (r, r0 , t) − Ψ∗21 (r, r0 , t)Ψ12 (r, r0 , t)}

(6.103)

The Ψ12 , Ψ21 cross-terms are zero at the start t = 0, because we start with well-separated
wave packets. Using eqs. 6.94 and 6.95 one gets terms like φ∗ (r)φ(r0 )ψ(r0 )ψ ∗ (r) in these
cross-terms. Since φ(r) is localized around r = r1 , and ψ(r) is localized around r = r2 ,
their product φ(r)ψ(r) = 0 everywhere. fHowever, the cross-terms do not disappear when
the wave packets start to overlap. For example, suppose the two particles collide without
an interaction potential being present, as, for instance photons or phonons can do; the
wave packets will then just pass through one another without distortion. At a certain time
t1 the wave packets will overlap. This situation is pictured in Fig. 6.7.
For a point r in this overlap region, i.e. r0 = r one has Ψ12 (r, r, t1 ) = Ψ21 (r, r, t1 ) (see
eqs. 6.94 and 6.95), and thus from eq. 6.97

Ψs (r, r, t1 ) = 2Ψ12 (r, r, t1 ) 6= 0
Ψa (r, r, t1 ) = 0 (6.104)

Most certainly, eq. 6.102 does not hold here ! Going back to eq. 6.101 it means that the
probability P (r, r0 , t) depends on the values of ca and cs . For example, choosing cs = 1
and ca = 0 or cs = 0 and ca = 1 certainly gives a very different result in view of eq.
6.104. The outcome of our experiment thus depends upon ca and cs . However, we started
6.6. APPENDIX II. IDENTICAL PARTICLES 165

φ(r ) ψ (r )

1st particle 2nd particle

r1 r r2

Figure 6.7: Two particles in overlapping wave packets.

our discussion with the statement that the two particles are indistinguishable. One of the
consequences of this statement is that we cannot think of a way to fix ca and cs (or α, β
in eq. 6.96).14

So if the numbers ca and cs are truly undetermined , the outcome of an experiment


would be completely unpredictable. Quantum mechanics would be in serious trouble here.
Fortunately as it turns out, nature has fixed the numbers ca and cs for us. Nature found
the following solution

1. cs = 1 and ca = 0. The states are symmetric Ψs (r, r0 , t), the particles are called
bosons.

2. cs = 0 and ca = 1. The states are anti-symmetric Ψa (r, r0 , t), the particles are called
fermions.

This is called the symmetry postulate. Pauli45 has shown that particles of integer spin
(0, 1, 2, ..) are bosons, and particles of half-integer spin ( 12 , 32 , 52 , ...) are fermions.
In principle, using mathematical reasoning only, other (fixed) values of ca and cs could
be possible. In practice, this is almost never observed. The only exception I know are the
anyon particles introduced by Laughlin98 in order to explain the fractional quantum hall
effect. See the footnote in Section 6.1.

6.6.3 Symmetrize The Universe ?


We have discussed a 2-particle system, but a similar type of reasoning can be used for
any number of particles, arriving at the same conclusion: the states are symmetric Ψs
for bosons and anti-symmetric Ψa for fermions. Suppose now we consider an electron on
earth and a second one far away, on the moon for instance. It sounds kind of strange that
we have to form an anti-symmetric state like eq. 6.97 between those two electrons. After
all, if we measure an electron in a wave packet φ(r) on earth, which is in a sense what we
do by simply looking at our computer screen, it is absurd to think that a second electron
14
Usually when we have a problem like this in quantum mechanics, we know how to fix it. For instance,
electrons can have a spin state like cα |αi + cβ |βi, i.e. a linear combination of spin-up and spin-down. This
seems to give a similar problem, but here we know a suitable fix. We send the beam of electrons through
a magnet which splits the beam (Stern-Gerlach experiment). After the magnet we have electrons in state
|αi (spin-up) in one beam, and electrons in state |βi (spin-down) in the other beam. We know of no such
“magnet” to separate indistinguishable particles !!
166 CHAPTER 6. BOSONS AND FERMIONS

in a wave packet ψ(r) on the moon should matter at all (unless you believe in astrology).
The neat way out of this dilemma is eq. 6.103, where we have already argued that in case
the two wave packets do not overlap we get
¯ ¯ n¯ ¯ ¯ ¯ o
¯Ψa (r, r0 , t)¯2 = 1 ¯Ψ12 (r, r0 , t)¯2 + ¯Ψ21 (r, r0 , t)¯2
2
1n o
= |φ(r, t)|2 |ψ(r0 , t)|2 + |ψ(r, t)|2 |φ(r0 , t)|2 (6.105)
2
Note that if r is on earth, the second term in eq. 6.105 gives zero, because obviously the
“moon” wave packet ψ(r) = 0 for points on earth. The probability distribution of eq.
6.100 gives
¯ ¯2 ¯ ¯2
P (r, r0 , t) = ¯Ψa (r, r0 , t)¯ + ¯Ψa (r0 , r, t)¯
= |φ(r, t)|2 |ψ(r0 , t)|2 (6.106)

which means that the probability distribution for the “earth” electron is given by |φ(r, t)|2 ,
independent of that of the “moon” electron |ψ(r0 , t)|2 , as common sense demands. If we
do a position measurement of the “earth” electron we get the expectation value
Z Z Z Z
hΨa |r|Ψa i = 0 3 3 0
rP (r, r , t) d r d r = r |φ(r, t)|2 |ψ(r0 , t)|2 d3 r d3 r0
Z Z Z
= r |φ(r, t)| d r · |ψ(r , t)| d r = r |φ(r, t)|2 d3 r
2 3 0 2 3 0
(6.107)

So indeed the “moon” electron does not have any influence on this measurement. The anti-
symmetric state Ψa of the “earth-moon” electron pair does not produce something which
contradicts common sense, which is reassuring. However it still seems a bit “overdone”.
Note that the probability distribution of eq. 6.106 is identical to

P (r, r0 , t) = |Ψ12 (r, r0 , t)|2

This means that for two electrons so far apart that their wave functions do not overlap,
we can assume that electron 1 is “here” (i.e. around r = r1 ) and electron 2 is “there” (i.e.
around r = r1 ). In other words we can pretend that we can distinguish the two electrons
from one another. Making them indistinguishable by anti-symmetrizing their total wave
function is not wrong, but it produces the same distribution. Note that the same reasoning
does not hold anymore if the cross terms in eq. eq. 6.103 cannot be neglected. This is
the case if the wave packets of the two electrons overlap. The probability distribution of
the two electrons cannot be written anymore as a product of two distribution in which
one only involves the first electron and the other only the second electron. If this is the
case then measuments on one electron cannot be independent of the second electron, and
one has to use the full distribution P (r, r0 , t). The two electrons are then called entangled,
and the anti-symmetric wave function is an example of an entangled state (i.e. one that
cannot be written as a simple product).15
15
Entangled states play a key role in fundamental discussions on quantum mechan-
ics involving the famous Einstein-Podolsky-Rosen (EPR) paradox, for instance. See
http://www.nobel.se/physics/educational/tools/theoretical.html, and modern quantum mechanics
books such as D. J. Griffith, Introduction to Quantum Mechanics, (Prentice Hall, Upper Saddle River,
1995); or R. I. Liboff, Introductory Quantum Mechanics, (Addison Wesley, Reading, 1997).
Entangled states also play an essential role in modern research on “quantum computing”.
6.6. APPENDIX II. IDENTICAL PARTICLES 167

Finally, for those of you who wonder whether it is necessary to include all electrons in
the universe in an anti-symmetric wave function, you may; it is not wrong, but it would
be overdone. If you only anti-symmetrize the wave function of the entangled electrons, i.e.
the ones that are close to one another, such that their wave packets overlap, and pretend
that you can distinguish them from those in the rest of the universe, you get the correct
results. Needless to say that, although the discussion has been for electrons, it also holds
for any kind of fermion or boson (using symmetric states in the latter case, of course).
168 CHAPTER 6. BOSONS AND FERMIONS
Chapter 7

Optics

“Macht doch den zweiten Fensterladen auch auf, damit mehr Licht hereinkomme”, Goethe’s dying
words (“Please open also the second shutter, so more light can come in”).

In the previous two chapters we have looked at “pure” fermion or boson systems of non-
interacting particles. Most of the interesting physics happens of course when particles are
interacting with each other. In this chapter we will look at an “enlightening” example of
fermions interacting with bosons, namely electrons interacting with photons, which gives
rise to all familiar optical processes. Second quantization presents a unified approach for
describing systems of fermions, such as electrons, systems of bosons, such as photons, as
well as the coupling between such systems.1 So, although we are focusing on optics in
this chapter, the general strategy is also applicable to other fermion and boson systems.
We start by reviewing some of the optics we discussed in Chapter 2, but now from a fully
quantum mechanical point of view in which everything, radiation as well as matter, is
described in terms of quantum states and operators. The first part descibes atoms in EM
fields as in Section 2.4. As we shall see, a full quantum mechanical approach allows us to
solve some of the problems we encountered in the semi-classical approach of Chapter 2, see
Section 2.5. In the second part we will consider optical processes in many-fermion systems,
using the homogeneous electron gas of Section 6.3.1. We will introduce Feynman diagrams
for interacting electrons, holes and photons, and discuss their connection to perturbation
theory. In the first appendix the Hamiltonian of an electron in an EM field is derived.
This chapter more or less rounds up the description of non-relativistic quantum electro-
dynamics. A few aspects of relativistic electrons and holes (positrons) are discussed in the
second appendix.

7.1 Atoms and Radiation; the Full Monty


In stone-age notation the Hamiltonian of an atom having N electrons is given by
N · 2
X ¸ N
b atom = b b
p i −Ze2 1X e2
H h + vb = + + (7.1)
2m |b
ri | 2 ri − b
|b rj |
i=1 i6=j

1
I use the phrase “second quantization”. You might also use the phrase “number representation”, or
(if you aspire to be an elementary particle physicist) “quantum fields”. In the present context they are all
synonyms.

169
170 CHAPTER 7. OPTICS

see Section 6.1.3. The first sum at the right hand side is a one-particle operator which
gives the kinetic energy of the electrons and their Coulomb interaction with the nucleus.
The second sum is a two-particle operator which gives the Coulomb repulsion between
the electrons. We assume that the nucleus is at a fixed position at the origin (so R = 0
in comparison to Section 6.1.3). Furthermore we have defined a convenient (orthonormal
and complete) basis set |ki of atomic orbitals, which allows us to rewrite the atomic
Hamiltonian in second quantized form
X 1 X
b atom =
H c†k b
hkl b cl + c†l b
vklmn b c†k b
cm b
cn (7.2)
2
k,l k,l,m,n

according to eqs. 6.34 and 6.35 in Section 6.2.2. The operators b c†k , b
ck create, annihilate
an electron in a state |ki of the basis set. Obviously these are fermion operators, so they
anti-commutate. The Hamiltonian H b atom provides a complete description of the atom.

The Hamiltonian which describes the EM radiation field (in vacuum) is


X½ † 1
¾
b rad =
H b
aK,sb
aK,s + ~ωK (7.3)
2
K,s
X
= a†K,sb
~ωKb aK,s + E0
K,s

according to eq. 5.79 in Section 5.5, where we have used eq. 6.44 to convert to a discrete
set of states, normalized in a box of volume Ω. E0 is the energy of the photon vacuum
state, which can be treated as an unspecified constant. The operators b a†K,s , b
aK,s create,
annihilate a photon of momentum ~K and polarization s. These are boson operators, so
they commutate. The Hamiltonian H b rad provides a complete description of the radiation
field in vacuum. Since electrons and photons clearly are distinguishable, a basic quantum
state for the complete system atom+radiation can be expressed as a product, see Section
1.4

|Ψi = |ψ atom i|φrad i (7.4)

The “full” Hamiltonian operates on this state as

H b
b0 = H +H b rad
³ atom ´ ³ ´
b 0 |Ψi = H
H b atom |ψatom i |φrad i + |ψ atom i Hb rad |φrad i (7.5)

Perhaps one is inclined to think that the description is now complete.

However, the atom and the radiation field interact. According to Section 2.4 this
interaction is represented by the operator
N
X
Vb = −b b =e
µ·E b b
ri · E (7.6)
i=1

in stone-age notation, cf. eqs. 2.41 and 2.42 (note we have set R = 0 again). In Section
2.4 we used in this expression quantum operators bri for the electrons, in combination with
7.1. ATOMS AND RADIATION; THE FULL MONTY 171

a classical electric field E, which made the operator Vb semi-classical. Now we assume that
Vb is a fully quantum mechanical operator in which also the electric field is an operator
Eb according to Section 5.5. We don’t have a stone-age representation for E, b since first
quantization or wave functions do not exist for photons, so to be consistent we must
transform eq. 7.6 into a full second quantized form. Since b ri and E b operate on different
subsystems (electrons and photons, respectively), we may consider them one at a time.
The electronic part is a one-particle operator and its second quantized form is given by
N
X X
b
ri → c†k b
rkl b cl with
i=1 k,l
rkl = hk(1)|r1 |l(1)i (7.7)

according to eq. 6.34. The electric field is given by

b µ ¶1 X ³ ´
b t) = − ∂ A = −i ~ 2 √
E(r, ωK b a†−K,s eiωK t εK,s eiK·r
aK,s e−iωK t − b (7.8)
∂t 2ε0 Ω
K,s

according to eqs. 5.71 and 5.72 in Section 5.5, where we have used the “discretization”
trick of eq. 6.44 again. In order to be consistent this expression has to be modified
slightly. All atomic operators are time independent, cf. eqs. 7.2 and 7.7, which means
that we use the Schrödinger picture. The time dependence in the electric field results
from the Heisenberg picture; to convert it into the Schrödinger picture, we set t = 0, cf.
eq. 2.64 in Section 2.6. Furthermore, in deriving eqs. 2.41 and 2.42 (and thus eq. 7.7)
we have assumed that the electric field is homogeneous, i.e. spatially independent. This
was justified by the fact that the wave length of EM radiation is much larger than the
size of the atom, see Section 2.4. For obvious reasons this is called the long wave length
approximation. The electric field operator now becomes2
µ ¶1 X ³ ´
b = −i ~ 2 √
E ωK b a†−K,s εK,s
aK,s − b (7.9)
2ε0 Ω
K,s

Using eqs. 7.7 and 7.9 in eq. 7.6 gives for the interaction operator
X ³ ´
Vb = β K,s;k,l b c†k b
aK,s b a†−K,s b
cl − b c†k b
cl with (7.10)
K,s;k,l
µ ¶1
~ω K 2
β K,s;k,l = −ie εK,s · rkl
2ε0 Ω

This operator describes the coupling or interaction between the atom and the radiation
field. According to the reasoning below eq. 6.38 in Section 6.2.3, the first term between
brackets in eq. 7.10 describes the annihilation of an electron in state l, accompanied by the
annihilation of a photon in state K,s and the creation of an electron in state k. In other
words it describes the excitation from l to k by absorption of a photon. The corresponding
diagram is given in Fig. 6.3. In a similar way, the second term between brackets in eq.
2 b † = E,
Note that since the electric field is an observable, it must be an Hermitian operator, i.e. E b from
which it follows that εK,s = ε−K,s . (We already know that ωK = c|K| = ω−K ).
172 CHAPTER 7. OPTICS

7.10 descibes the de-excitation from l to k and emission of a photon. It can be represented
by a similar diagram with the arrow on the photon line reversed.

The Hamiltonian H b =H b 0 + Vb , eqs. 7.2, 7.3 and 7.10, gives a complete desciption of the
system, atom plus radiation field. It allows us to descibe all possible optical experiments;
absorption, emission, scattering of light, non-linear optics; you name it. As before, our
b 0 for constructing unperturbed states, use Vb as the perurbation, and
strategy is to use H
the tools of perturbation theory to study optical transitions.

MODEL; A TWO-LEVEL HYDROGEN ATOM

As a demonstration of how the procedure works, we reconsider the hydrogen atom of


Section 2.4 again.3 Because hydrogen contains only one electron, the atomic Hamiltonian
can be simplified. There are no other electrons present and we can neglect the two-particle
terms in eq. 7.2.4 As a basis set we choose the eigenstates of the hydrogen atom, which
makes the one-particle term in eq. 7.2 diagonal, i.e. hkl = δ kl ²k , and
X
b hydrogen =
H c†k b
²k b ck
k

where k = (n, l, m) represents the set of quantum numbers needed to fix the eigenstates
of the hydrogen atom, and ²k = ²nlm = −13.606 n2
eV are the familiar energy levels of the
hydrogen atom. We simplify the discussion even further by selecting just two of these
states, the 1s state, which we call |ii, and the 2p x state, which we call |f i. The simplified
Hamiltonian becomes

b hydrogen → b
H c†i b
h = ²i b c†f b
ci + ²f b cf (7.11)

where ²i = ²1s = −13.606 eV, and ²i = ²2px = −3.402 eV. The two-state basis set |ii and
|f i also allows us to simplify eq. 7.7 to

c†i b
rif b c†f b
cf + rf i b ci with (7.12)
rif = h1s|r|2px i

and the atom-field interaction of eq. 7.7 is then simplified to


X ³ ´
Vb = aK,s b
β K,s;if b †
ci b †
a−K,s b
cf − b †
ci b
cf
K,s
³ ´
+β K,s;f i b c†f b
aK,s b a†−K,s b
ci − b c†f b
ci (7.13)

3
The hydrogen atom is chosen for convenience, since it simplifies the calculations. The method is
generally applicable to atoms and molecules. For solid state materials, see the second part of this chapter.
4
The two-particle term contains two annihilation operators; working on a one-particle state, it always
gives zero. In principle EM radiation can create other electrons (and positrons), which could interact with
the one electron in the hydrogen atom. However, this process happens only at a very high energy; we
neglect it here.
7.1. ATOMS AND RADIATION; THE FULL MONTY 173

7.1.1 Absorption; Fermi’s Golden Rule


Now let us do the following experiment. We prepare the atom in the 1s state, and the
radiation field in a state in which there is one photon present with momentum ~Q and
polarization along the x-axis, i.e. εQ,s = ex . So the complete initial state is given in
second quantization notation by

|Ψini i = |ψ ini,atom i|φini,rad i = |1i i|1Q,x i ≡ |1i ; 1Q,x i (7.14)

We are interested in a final state in which the atom is in the 2px state and the radiation
field is in its ground state, i.e. no photons present.

|Ψf in i = |ψ f,atom i|φf,rad i = |1f i|0i ≡ |1f ; 0i (7.15)

We wish to calculate the transition amplitude between initial and final states
D E
b (t, t0 )|Ψini ; t > t0 where
Ψf in |U (7.16)
b (t, t0 ) = e− ~i (t−t0 )Hb
U b =b
and H b rad + Vb
h+H

As in Section 2.3 we first consider the Born approximation, which neglects all terms beyond
first order perturbation

U b0 (t, t0 ) + U
b (t, t0 ) ≈ U b (1) (t, t0 ) (7.17)
Z t
b0 (t, t0 ) + 1
= U dτ U b0 (t, τ ) Vb U
b0 (τ , t0 )
i~ t0
b0 (t, t0 ) = e− ~i (t−t0 )Hb 0
where U b0 = b
and H b rad
h+H

We follow exactly the same procedure as outlined in Section 2.3. According to eqs. 7.3,
7.5 and 7.11, we have
³ ´ ³ ´
Hb 0 |Ψini i = bh|1i i |1Q,x i + |1i i Hb rad |1Q,x i

= (²i + ~ω Q + E0 )|Ψini i (7.18)

and in a similar way

b 0 |Ψf in i = (²f + E0 )|Ψf in i


H (7.19)

Therefore
D E
b0 (t, t0 )|Ψini = e− ~i (t−t0 )(²i +~ωQ +E0 ) hΨf in |Ψini i = 0
Ψf in |U

since the states are orthogonal. The zero’th order term of eq. 7.17 gives no contribution
in eq. 7.16. Working out the first order term as in Section 2.3 gives
D ¯ ¯ E Z t
¯b ¯ 1 i i
Ψf in ¯U (t, t0 )¯ Ψini = dτ e− ~ (t−τ )(²f +E0 ) Vf in,ini e− ~ (τ −t0 )(²i +~ωQ +E0 )
i~ t0
with Vf in,ini = hΨf in |Vb |Ψini i (7.20)
174 CHAPTER 7. OPTICS

The integral can be done, which yields the transition rate


d ¯¯D ¯
¯b
¯
¯
E¯2 2π
¯ 2
wini→f in = ¯ Ψf in ¯U (t, t0 )¯ Ψini ¯ = 2 |Vf in,ini | δ (ω f i − ω Q ) (7.21)
dt ~
where ωf i = (²f − ²i )/~

under the same conditions as discussed in Section 2.3.

This is Fermi’s golden rule again, cf. eq. 2.32. It will lead to a similar expression
as eq. 2.45 in Section 2.4 for the transition in the hydrogen atom. However, notice the
differences with our earlier derivation

• The EM field is not treated as an external semi-classical perturbation, but it is part


of the quantum description now. The “perturbation” is the coupling Vb between the
atom and the radiation field, eq. 7.13.

• The perturbation operator has no time dependence (which according to Schrödinger


is the correct procedure). All time dependence is in the time evolution operator
Ub (t, t0 ), eq. 7.17.

• The δ-function in eq. 7.21 now truly describes the conservation of energy, ²i +~ωQ =
²f ; i.e. transitions are only allowed between states |Ψini i and |Ψf in i which have the
same energy, cf. eqs. 7.18 and 7.19.

What remains is the calculation of the matrix element Vf in,ini .

Vf in,ini = hΨf in |Vb |Ψini i


X ³ ´
= h1f ; 0| β K,s;if b c†i b
aK,s b a†−K,s b
cf − b c†i b
cf
K,s
³ ´
+β K,s;f i b c†f b
aK,s b a†−K,s b
ci − b c†f b
ci |1i ; 1Q,x i
= h1f ; 0|β Q,x;f i b c†f b
aQ,x b ci |1i ; 1Q,x i
µ ¶1
~ω Q 2
= β Q,x;f i = −ie ex · rf i (7.22)
2ε0 Ω

It is easy to see that only the third term of the right hand side of eq. 7.13 is contributing
to this matrix element. The first two terms contain b cf , and try to annihilate a particle in
state |f i, which is not present. The fourth term creates a photon, instead of annihilating
one. Of the sum over K,s only the one Q,x term yields a non-zero contribution to Vf in,ini ,
since all the photon states of the radiation field are orthogonal (work it out yourself).
Since ex · rf i = xf i = hf |b
x|ii, according to eq. 7.12, we get

e2 ~ω Q
|Vf in,ini |2 = |β Q,x;f i |2 = x|ii|2
|hf |b (7.23)
2ε0 Ω
which gives for the transition rate of eq. 7.21

πe2 ~ωQ
wini→f in = x|ii|2 δ (ω f i − ω Q )
|hf |b (7.24)
~2 ε0 Ω
7.1. ATOMS AND RADIATION; THE FULL MONTY 175


The term ΩQ is the excess energy density that is present in the radiation field (i.e. energy
on top of the vacuum energy) at the start of our “experiment”, since we started with one
photon of energy ~ω Q in a volume Ω. It is not difficult to show that if we start with
nQ photons in this mode, a similar calculation shows that the transition rate simply is
proportional to nQ . For a classical electric field E = E0 cos ωt as we used in Section
2.4, the energy density is 12 ε0 E02 . This means that eq. 7.24 corresponds exactly to the
semi-classical result we found before, cf. eq. 2.45. So all the physical concepts we have
introduced in Section 2.4, oscillator strengths, selection rules, Einstein coefficients, remain
valid.

7.1.2 Spontaneous Emission


However we can do much more than just recovering the results of Section 2.4. For instance
we can calculate the probability amplitude
D E
b (t, t0 )|Ψf in
Ψf in |U (7.25)

where we have prepared at time t0 our atom in the excited state |f i with no photons
present, and probe the probability that is still in this state at a later time t. In other
words, we are interested in the decay of the excited state. We can do a full calculation,
but in this case we already know in general terms what the outcome will be. Although the
physics is a little bit different, mathematically this problem looks like the problem we have
solved in Chapter 4, compare eq. 7.25 to eq. 4.20. Just like in that chapter we can sum
the complete perturbation series in Vb (including all orders), by using the self-energy Σ.
Without going through all the algebra for this particular case, we know from the results
of Chapter 4 that the general result according to eq. 4.39 is5
D E
b (t, t0 )|Ψf in = e− ~i [²f +∆f ](t−t0 ) e− 12 Γf (t−t0 ) for t > t0
Ψf in |U (7.26)

where the level shift ∆f and the decay constant Γf are related to the self-energy Σ ac-
cording to eq. 4.38.6 The probability that the atom is in its excited state is given by
¯D E¯2
¯ b (t, t0 )|Ψf in ¯¯ = e−Γf (t−t0 )
Nf (t) = ¯ Ψf in |U (7.27)

The occupancy of the excited state (which is just another word for this probability) is
decaying exponentially. In this case the decay rate Γf can even be obtained analytically
from the self-energy. Assume that we can do the algebra, and observe that eq. 7.27
corresponds exactly to eq. 2.59 in Section 2.4, because we started from an occupancy
Nf (t0 ) = 1. In other words, we have recovered the Einstein rate equation and therefore
we have found

Γf = Af i (7.28)
5
A famous quote of Feynman is “...the same equations have the same solutions...”. It means that, if
you have two equations which look similar, although they may come from completely different fields in
physics, the solutions also must have a similar form. It sounds trivial, but it is a much used principle in
physics. The strive towards “unification theories” stems from the desire to make different equations look
“similar”. D E
6
The weight factor zf = 1 for the same reason as in eq. 4.38, namely Ψf in |U(t b 0 , t0 )|Ψf in =
hΨf in |Ψf in i = 1.
176 CHAPTER 7. OPTICS

which is the Einstein coefficient for spontaneous decay, cf. eq. 2.60. But we have obtained
it now from quantum mechanical first principles, without needing a phenomenological
model!! Our state |Ψf in i has no photons to start with, but the perturbation Vb of eq. 7.13
couples it to other states, which leads to a probability that the atom emits a photon and
falls to a lower atomic level. Since this occurs without the need for an external field, i.e.
in the radiation field vacuum, this process is truly spontaneous and cannot be switched
off.
We have solved some of the problems mentioned in Section 2.5 by our fully quantum
mechanical approach. The complete Einstein equations can also be derived; it is possible
to calculate the Compton scattering of Section 5.5.3 by this approach, or any other optical
process you might be interested in.7

In addition to a decay, eq. 7.27 also predicts that the atomic level ²f is shifted to ²f +∆f
by the coupling of the atom to the radiation field. Calculating the shift ∆f involves some
subttleties which I won’t discuss now, but it can be done. What is more important, the
shift can be measured experimentally, and theory and experiment agree. From elementary
quantum mechanics we obtained the energy levels of the isolated hydrogen atom, but
now we see that coupling with the radiation field vacuum states shifts these levels.8 The
shifts are small, i.e. ∼ 109 Hz (which is at the ppm level for hydrogen levels), but they
can be measured! The prediction of these shifts is considered to be one of the great
successes of quantum electrodynamics. To be fair, there are also shifts due to relativistic
effects. The “speed” of the electron can no longer be neglected as compared to the speed
of light if one is interested in such small level shifts. Relativistic shifts are an order of
magnitude larger than the shifts due to the coupling with the radiation field, but they can
also be calculated, by using the relativistic Dirac equation instead of the non-relativistic
Schrödinger equation for the hydrogen atom. The relativistic theory lifts some of the
degeneracies that are present in the non-relativistic spectrum of the hydrogen atom, but
e.g. the 2s 1 and 2p 1 levels are still degenerate.9 Coupling to the radiation field splits
2 2
these levels, and the resulting splitting of ∼ 109 Hz, which is called the Lamb shift, can be
measured.

7.2 Electrons, Holes and Photons


“What is the use of a book, thought Alice, without pictures or conversations”, Lewis Caroll, Alice’s
Adventures in Wonderland.

In this section we will consider optical processes related to “free” electrons, using the
homogeneous electron gas of Section 6.3.1. Feynman diagrams are introduced that describe
processes involving electrons, holes and photons, and their connection to perturbation
theory is discussed.
7
If you are interested, see C. Cohen-Tannoudji, J. Dupont-Roc, and G. Grynberg, Atom-Photon Inter-
actions, (Wiley, New York, 1998).
8
Again I want to stress that this is a “spontaneous” effect, i.e. it exists in vacuum. All it requires is
the coupling operator of eq. 7.10, not an external field!
9
See B. H. Brandsen and C. J. Joachain, Quantum Mechanics, (Prentice-Hall, Harlow, 2000), Ch. 8
and 15.
7.2. ELECTRONS, HOLES AND PHOTONS 177

7.2.1 Electrons and Radiation


The hamiltonian b
h of a homogeneous “free” electron gas is given by eq. 6.51.
X
b
h = c†k b
²k b ck
k
X X
= E0,el − ²kbb†kbbk + a†kb
²kb ak (7.29)
|k|≤kF |k|>kF

~2 |k|2
with ²k = (7.30)
2m
where bc†k ,bck refer to general fermion creation, annihilation operators and b ak and bb†k , bbk
a†k , b
refer to the particle and hole operators introduced in Section 6.3.1. E0,el is the energy of
the ground state (vacuum) of the electron gas; we do not need to calculate it explicitly.
The Hamiltonian H b rad of a “free” radiation field is still given by eq. 7.3. H b0 = b h+H b rad
is an “unperturbed” Hamiltonian like the one in eq. 7.5. Again, a perturbation V which b
couples the electronic system to the radiation field needs to be constructed as in the
previous section. However, eq. 7.6 is no good here! That equation was derived assuming
the long wave length limit, i.e. the wave length of the EM radiation is much larger than
the “size” of the electronic system, see also Section 2.4. This is reasonable for a small
atom, but not for a free electron. The latter has an infinite extension, or, if we use states
that are normalized in a box, the electronic wave functions have the same extension as the
EM waves. The correct expression which describes the interaction between the electronic
and the photonic system is given by10
N
e Xb
Vb = bj
rj ) · p
A(b (7.31)
m
j=1

in stone-age representation, where p b i is the momentum operator of the i’th electron and
b ri ) is the vector potential. A derivation of this expression is given in Appendix I. The
A(b
vector potential operator is given by
X
b r) = √1
A(b qbK,s εK,s eiK·br (7.32)
Ω K,s
µ ¶1 ³ ´
~ 2
qbK,s = aK,s + b
b a†−K,s
2ε0 ω K
according to eq. 5.72 in Section 5.5, where we have switched to a “discrete” set of states
again using eq. 6.44, and we have set t = 0, because we wish to work in the Schrödinger
representation. Note that the spatial dependent part eiK·br is considered to be an operator
in electron space!

The expression for the interaction we used in the atomic case, cf. eq. 7.6, looks quite a bit
different from eq. 7.31. However, the “atomic” expression can be derived from the latter,
10
Electrons
PN also carry a spin sj , which has an interaction with the magnetic field. This contributes a
term g2m
ee
bsj · b rj ) to the interaction. This term is usually small, and we neglect it here. See also
B(b
j=1
the footnote in Section 6.3.1.
178 CHAPTER 7. OPTICS

which is also shown in Appendix I. The “atomic” expression of eq. 7.6 only works in the
long wave length limit, i.e. when the spatial dependence of the fields may be neglected.
This is valid for small systems like atoms, but in other cases the more general form of eq.
7.31 has to be used.

As in the previous section, the only way to make progress is to transform eq. 7.31 into a
second quantized expression. Since we are dealing with the homogeneous electron gas, we
use its eigenstates as the basis set for our electronic system, see eqs. 6.40—6.42 in Section
6.3.1. Proceeding as for eq. 7.7, we obtain the second quantized form of the spatially
dependent part of eq. 7.31
N
X X
eiK·brj p
bj → c†k b
π kl b cl with
j=1 k,l
Z
iK·b 1 ~
π kl = hk|e r
b |li =
p d3 r e−ik·r eiK·r ∇eil·r
Ω Ω i
Z
~l
d3 r ei(K−k+l)·r = ~lδ k−K,l (7.33)
Ω Ω

where we have used eq. 6.41 (see also Chapter 4, eq. 4.17). Using eqs. 7.32 and 7.33 in
eq. 7.31 gives the required second quantized form for the interaction operator
X ³ ´
Vb = γ K,s;l b c†l+K b
aK,s b a†−K,s b
cl + b c†l+K b
cl (7.34)
K,s;l
µ ¶1
e ~ 2
with γ K,s;l = ~l · εK,s
m 2ε0 Ωω K
As before, the terms in eq. 7.34 have a physical meaning. The first term describes
absorption of a photon of momentum ~K and polarization s, accompanied by an excitation
of an electron with momentum ~l to momentum ~(l + K); the second term describes an
emission process. We can convert this operator according to the particle-hole formalism
by making the substitutions b cl = b
al if ²l > ²F and b cl = bb†l if ²l ≤ ²F . Note that if ²l ≤ ²F , it
is possible that ²l+K > ²F , if K is large enough (and positive). So the coupling Vb contains
particle-hole terms, as well as particle-particle and hole-hole terms.11
X ³ ´
Vb = aK,s + b
γ K,s;l b a†−K,s b a†l+Kb
al (7.35)
K,s;|l|>kF ,|l+K|>kF
X ³ ´
+ aK,s + b
γ K,s;l b a†−K,s ba†l+Kbb†l (7.36)
K,s;|l|≤kF ,|l+K|>kF
X ³ ´
+ aK,s + b
γ K,s;l b a†−K,s bbl+Kbb†l (7.37)
K,s;|l|≤kF ,|l+K|≤kF
X ³ ´
+ a†−K,s bbl+Kb
aK,s + b
γ K,s;l b al (7.38)
K,s;|l|>kF ,|l+K|≤kF

11
The notation becomes a bit messy, but I will use the “K,s” subscript for photon operators ba†K,s , b
aK,s

(with boson commutation rules) and the “k” subscript for electron operators b ak , b
ak (with fermion anti-
commutation rules).
7.2. ELECTRONS, HOLES AND PHOTONS 179

The four terms, eqs. 7.35—7.38, describe slightly different physical processes. The first term
describes the excitation of an electron, or de-excitation, depending on whether |l + K| > |l|
or |l + K| < |l|. The second term describes the creation of an electron-hole pair, the third
term describes the (de)excitation of a hole, and the fourth term describes the annihilation
of an electron-hole pair. In the following we will consider each of these four processes in
more detail.

7.2.2 Free Electrons and Holes


First we consider whether the following is possible. A photon with energy ~ω and mo-
mentum ~K is absorbed by a particle with energy ²k > ²F and momentum ~k, creating a
particle with energy ²q > ²F and momentum ~q. This event is related to eq. 7.35 and is
shown in Fig. 7.1.

εq
εF

εk

Figure 7.1: Absorption of a photon by an electron.

At first sight the event looks perfectly o.k.; we have discussed it for the atom. The
transition probability can be described by Fermi’s golden rule, eq. 7.21

wini→f in = |Vf in,ini |2 δ ((²q − ²k )/~ − ω) (7.39)
~2
As always the δ-function gives a “conservation of energy”

~ω = ²q − ²k (7.40)

We have to work out the matrix element Vf in,ini . Our initial state is |Ψini i = |1k ; 0; 1K i,
i.e. one electron in state k; zero holes; and one photon in state K. Our final state is
|Ψf in i = |1q ; 0; 0i, i.e. one electron in state q; zero holes; and zero photons. From eq.
7.35, it is straight-forward to show that
X ³ ´
hΨf in |Vb |Ψini i = h1q ; 0; 0| a†−K0 ,s b
aK0 ,s + b
γ K0 ,s;l b a†l+K0 b
al |1k ; 0; 1K i
K0 ,s;|l|>kF ,|l+K|>kF
³ ´
a†−K,s b
aK,s + b
= h1q ; 0; 0|γ K,s;k b a†k+Kbak |1k ; 0; 1K i
µ ¶1
e ~ 2
= γ K,s;k δ k+K,q = ~k · εK,s δ k+K,q (7.41)
m 2ε0 Ωω K
180 CHAPTER 7. OPTICS

It is easy to see that the terms of eqs. 7.36—7.38 do not give a contribution here. The
δ k+K,q in eq. 7.41 is an example of a “selection rule”. In this case it can be interpreted
as the “conservation of momentum”.

~K = ~q−~k (7.42)

When the electron absorbs the photon, its absorbes not only its energy but also its mo-
mentum. As in Fig. 6.3 and or Fig. 3.2, the event can be represented by a diagram. This
is shown in Fig. 7.2.

t0 t
ε k , =k ε q , =q

τ
=ω , = K

Figure 7.2: Absorption of a photon by an electron.

The time in this diagram increases from left to right; the particle comes in from the left,
absorbes the photon, and goes out to the right. This picture of the absorption can directly
be interpreted as a Feynman diagram to calculate the transition probability amplitude.
The corresponding operator is
Z h ³ ´ i
b 1 t b0 (t, τ ) γ K,s;k b b0 (τ , t0 )
(1)
U (t, t0 ) ≡ dτ U a†−K,s b
aK,s + b a†qb
ak U (7.43)
i~ t0
Increasing the time this operator has to be interpreted from right to left, as is usual for
operators. The system propagates unperturbed from time t0 to time τ . Then the event
takes place at the connection point of the lines (the so-called vertex ). It contains the
annihilation of an electron³ in state k, ´and the creation of an electron in state q. The
photon is represented by b a†−K,s . This might look odd at first sight, but it makes
aK,s + b
sense if we consider eqs. 7.32 and 7.34; it is the mode amplitude operators qbK,s of the field
that appear in the perturbation Vb . The “strength” of the interaction event at the vertex
is represented by the factor γ K,s;k . After the interaction the system again propagates
unperturbed from τ to t. As usual we integrate over all possible intermediate times τ .
In this particular diagram the labels (K, k, q) are fixed by the initial and final states. If
any of these would be not fixed, the convention is to sum over the “unfixed” labels. The
calculation has to be completed by sandwiching U b (1) (t, t0 ) between our initial and final
states
b (1) (t, t0 )|1k ; 0; 1K i
h1q ; 0; 0|U (7.44)

This is exactly what is needed to calculate the transition probability according to Fermi’s
golden rule, cf. eqs. 7.39 and 7.41. The diagram thus corresponds directly to the operator
used in eq. 7.41.

Looking at the diagram one can immediately identify the conservation of energy and
momentum by interpreting the diagram as a “circuit”. At the vertex the “sum of all
7.2. ELECTRONS, HOLES AND PHOTONS 181

incoming arrows” must be equal to the “sum of all outgoing arrows”, like the currents in
an electric circuit. This leads to ~ω+²k = ²q and ~K + ~k =~q, which are the conservation
laws of eqs. 7.40 and 7.42. Note the fact that ²q > ²k according to our original assumption,
cf. Fig. 7.1, fixes the direction of the arrow on the photon line. It means that only the
b
aK,s part of the photon operator in eq. 7.43 participates (i.e. the excitation from k to
q is accompanied by absorption, and not by emission). If ²q < ²k then the b a†−K,s part
would be active, which corresponds to emission. The arrow on the photon line would be
reversed, and the circuit rules would give ²k = ²q − ~ω and ~k =~q+~K, which is correct
for emission.

IMPORTANT REMARK

However perfect it looks, the event is in fact forbidden. A free electron cannot absorb a
photon! The reason lies in the fact that the selection rules of eqs. 7.40 and 7.42 cannot
be obeyed simultaneously. An electron wave with a typical energy of 10 eV has a wave
length λ ≈ 4 Å (and thus k = 2π −1
λ ≈ 1.6Å ), as is easily derived from eq. 7.30. A photon
wave with an energy of 10 eV has a wave length λ ≈ 1240 Å (and thus k ≈ 5 × 10−3 Å−1 ),
cf. eqs. 5.64 or 6.56. There is no way that the tiny photon momentum can supply the
large change in the electron momentum, which is required by eq. 7.42, when at the same
time all the photon energy has to be absorbed by the electron, as required by eq. 7.40.12

The reason absorption worked for an “atomic” electron as discussed in the first part
of this chapter, is because an electron that is bound in an atom is not in a momentum
eigenstate, and therefore eq. 7.42 does not hold. The energy is still conserved, cf. eq.
7.21. The photon momentum is absorbed by the atom as a whole (which is the microscopic
origin of “radiation pressure”). It is possible for a free electron to scatter a photon, which
is the Compton scattering discussed in Section 5.5.3. To obey the conservation rules of eq.
5.82, the electron only absorbs part of the energy and momentum of the incoming photon,
the rest is given to the outgoing photon.

The foregoing discussion does not depend upon whether the initial state of the particle
is above or below the Fermi energy, i.e. whether ²k > ²F or ²k ≤ ²F . Suppose the latter
is the case. Then absorption of a photon would create an empty state below and a filled
state above the Fermi level. This forbidden process is shown in Fig. 7.3.
Another way of looking at this process is to say that the photon of energy ~ω and
momentum ~K is annihilated and two “particles” are created. One is a particle (electron)
with energy ²q and momentum ~q and the other is an anti-particle (hole) with energy −²k
and momentum −~k, cf. Table 6.2. In order to see that this process is forbidden one can
again look at the conservation of energy and momentum.

~ω = ²q + (−²k )
~K = ~q + (−~k) (7.45)
12
The only way to satisfy eqs. 7.40 and 7.42 simultaneously is to use high energy (small wave length,
large momentum) photons. However, this would give the electron a relativistic momentum and energy for
which the non-relativistic expression of eq. 6.40 is no longer valid. However, also in a fully relativistic
treatment, absorption of a photon by a free electron is forbidden, see Appendix II.
182 CHAPTER 7. OPTICS

εq
εF

εk

Figure 7.3: Absorption of a photon, creating an electron-hole pair.

These are the same expresssions as eqs. 7.40 and 7.42 and both rules cannot be obeyed
simultaneously, for the same reason as discussed above. Needless to say that if both
²k , ²q ≤ ²F the same reasoning holds, i.e. also a free hole cannot absorb a photon. Of
course in each of these cases the selection rules can be derived from Fermi’s golden rule,
eqs. 7.39 and 7.41 by inserting the operators of eqs. 7.36—7.38 in combination with the
appropriate initial and final states. Before doing that we first adapt our system such that
absorption and emission become allowed processes.

7.2.3 Light Absorption by Electrons and Holes


As we have seen, absorption of a photon by a electron is forbidden, but it is allowed for
an electron that is bound in an atom. The same holds for an electron that is bound in a
crystal. To show this we have to extend the formalism of the previous sections to electrons
in a crystal lattice. In a “free” electron metal the electronic states are still given by eq.
7.30. In crystals it is appropriate to write the momentum as

pi = ~(k + κi ) (7.46)

where k is restricted to lie in the (first) Brillouin zone, and κi is a reciprocal lattice vector.
Note that this is simply a change of notation. We can still cover the whole 3-dimensional
space with pi . In a similar way

pf = ~(q + κf ) (7.47)

with q in the first Brillouin zone, and κf a reciprocal lattice vector. The numbers κi , κf
in a “free” electron metal can also be interpreted as band indices.13 In general, we will use
a combined index like k, i to denote an electron state, where i now gives the band index.
Again we can derive selection rules from Fermi’s golden rule, eq. 7.39.The conservation of
energy is

~ω = ²q,f − ²k,i (7.48)


13
N. W. Ashcroft and N. D. Mermin, Solid State Physics, (Holt-Saunders, Philadelphia, 1976), Ch. 9;
or J. M. Ziman, Principles of the Theory of Solids, (Cambridge University Press, 1979), Ch. 3; or H.
Ibach and H. Lüth, Solid-State Physics, (Springer, Berlin, 1995), Ch. 7.
7.2. ELECTRONS, HOLES AND PHOTONS 183

In a “free electron” metal the relation between the energy ²k,i and ~k, which is called a
dispersion relation, is given simply by

|pi |2 |~k + ~κi |2


²k,i = = (7.49)
2m 2m
(a similar relation holds between ²q,f and ~q). For a “real” material (metal, semiconductor
or insulator), the dispersion relation has a different functional form, determined by the
interaction between the electron and the nuclei that make up the crystal lattice (see the
solid state books cited in the footnote). Calculation of the matrix element, eq. 7.41, gives
for a “free electron”

hΨf in |Vb |Ψini i = γ K,s;k,i δ k+κ i +K,q+κ f


= γ K,s;k,i δ k+K,q δ κ i ,κf (7.50)

The first line is simply a translation of eq. 7.41 using the notation of eqs. eq. 7.46 and
eq. 7.47. The second line involves the following reasoning. We have already argued that
the photon momentum ~K is tiny as compared to the electron momentum; see the remark
in the previous section. So K will also be tiny on the scale of the first Brillouin zone,
and it can only connect points k and q within this zone. The δ k+K,q is interpreted as a
“conservation of momentum”

~K =~q−~k (7.51)

More specifically, it is called the conservation of crystal momentum, since ~k (or ~q) is
called the crystal momentum of the electron. The δ κi ,κ f in eq. 7.50 is a selection rule
that is typical for a “free” electron system. If an electron interacts with the nuclei of the
crystal lattice, this selection rule is broken, and eq. 7.50 is relaxed to

hΨf in |Vb |Ψini i = γ K,s;k,f i δ k+K,q (7.52)

One might also say that interacting with the crystal lattice and constantly being diffracted
by the lattice, the electron looses the momentum ~κi of eq. 7.46 to the lattice. Because
the crystal lattice as a whole is very massive, the resulting recoil from absorbing the
momentum ~κi won’t be observable, and the latter simply disappears. This is why there
are no κi,f ’s in eq. 7.52. The coupling constants γ K,s;k,f i are changed by the interaction
with the lattice however. To calculate it we notice that the whole formalism of the previous
sections can easily be extended to electrons in a crystal lattice. The only thing that changes
in eqs. 7.29—7.38 is that we have to substitute any index of type k by k, i , where i is the
band index. The Hamiltonian for the electron gas still has the form of eq. 7.29, but the
dispersion relation is not given by eq. 7.30 (or eq. 7.49) anymore, since that is exclusively
for “free” electrons, not interacting with the lattice. The spatially dependent part of the
electron-radiation interaction of eq. 7.33 also becomes slightly more complicated, because
the basis states of electrons interacting with a lattice are not simple plane waves anymore,
but Bloch waves, which have a more complicated spatial dependence. One can prove
N
X X
eiK·brj p
bj → c†k,i b
π k,i;l,j b cl,j with
j=1 k,i;l,j
iK·b
r
π kl = hk, i|e b |l, ji = δ k−K,l uk,i;l,j
p (7.53)
184 CHAPTER 7. OPTICS

where uk,i;l,j depend upon the details of the Bloch state, see the solid state books cited in
the footnote. Using eq. 7.53 instead of eq. 7.33 gives instead of eq. 7.34
X ³ ´
Vb = γ K,s;l,ij b c†l+K,i b
aK,s b a†−K,s b
cl,j + b c†l+K,i b
cl,j (7.54)
K,s;l,i,j
µ ¶1
e ~ 2
with γ K,s;l,ij = uk,i;l,j ·εK,s
m 2ε0 Ωω K
Apart from a more complex notation as compared to the “free” electron case, due to the
appearance of a band index, and a change in the numerical values of the energies ²k,i and
the coupling constants γ K,s;l,ij , due to the interaction between the electron and the crystal
lattice, the formalism is unchanged. The calculations proceed as in the previous section,
but now the conservation laws or selection rules of eqs. 7.51 and 7.48 can now be obeyed
simultaneously. The event shown in Fig. 7.2 becomes allowed for electrons in a crystal.14

In a crystal, the lattice is determined by the positions of the nuclei. The presence
of the nuclei therefore allows for transitions that are forbidden for free electrons. In this
sense it is similar to an electron bound in an atom as discussed above. This is generally
true, the presence of a fourth object such as a nucleus, besides the incoming and outgoing
electron or hole and the photon, allows for transitions that are forbidden for a free electron
or hole. This fourth body supplies the momentum required to obey the conservation law.

The absorption of a photon as shown in Fig. 7.3 can be represented by a similar


“billiard-ball” picture as Fig. 7.2. This is shown in Fig. 7.4.

ε q , f , =q

−ε k ,i , − =k

=ω , =K

Figure 7.4: Creation of a particle-hole pair by a photon.

A photon of energy ~ω and momentum ~K is annihilated and a particle of energy


²q,f and momentum ~q, and a hole of energy −²k,i and momentum −~k are created, see
Table 6.2. Most people actually prefer to write the diagram of Fig. 7.4 in a different way,
because one does not like to use too many different sorts of lines. Fig. 7.5 is meant to
represent the same event.
The dashed line of the hole is now substituted by a solid line, but the arrow is reversed
to distinguish it from the particle-line. The signs of the energy and momentum labels on
14
It is still possible that some transitions are forbidden if uk,i;l,j = 0. This is imposed by the symmetry
of the crystal lattice. However, this never results in all transitions being forbidden, so never all uk,i;l,j are
zero, like in the “free” electron case. A systematic study of symmetry and its impact on the selection rules
can be done using “group theory”.
7.2. ELECTRONS, HOLES AND PHOTONS 185

ε q , f , =q

ε k ,i , =k
=ω , =K

Figure 7.5: Creation of a particle-hole pair by a photon.

the hole-line are also changed to ²k,i and a momentum ~k. The idea is that the “−” sign
is now supplied by the direction of the arrow, such that the energy and momentum of the
hole stays the same as in Fig. 7.4. The time in this diagram increases from left to right,
as always. Note therefore that time runs against the arrow in the hole-line (bottom solid
line in Fig. 7.5), whereas it runs with the arrow in the particle-line (top solid line in Fig.
7.5). There are several reasons for choosing this representation. For instance, by looking
at the diagram one can immediately identify the conservation of energy and momentum,
eqs. eq. 7.51 and eq. 7.48, interpreting the diagram as a “circuit”. At the vertex the
“sum of all incoming arrows” must be equal to the “sum of all outgoing arrows”. This
leads to ~ω + ²k,i = ²q,f and ~K + ~k =~q, which are the conservation laws. (One always
implicitly assumes that the crystal momentum ~κ is taken care of by the crystal). The
operator corresponding to this diagram is given by
Z t h ³ ´i
b (1) (t, t0 ) ≡ 1
U b0 (t, τ ) γ K,s;k,f i b
dτ U a†−K,s b
aK,s + b a†q,f bb†k,i U
b0 (τ , t0 ) (7.55)
i~ t0

Note that the middle part now corresponds to the interaction term of eq. 7.36, for which
²k ≤ ²F and ²k+K = ²q > ²F , which corresponds to creation of a a particle-hole pair. To
complete the calculation we have to sandwich this operator between suitable initial and
final states. A bit of reflection shows that a suitable inital state is |Ψini i = |0; 0; 1K i,
i.e. zero electrons; zero holes; and one photon in state K. A suitable final state then
is |Ψf in i = |1q,f ; 1k,i ; 0i, i.e. one electron in state q,f , one hole in state k, i, and zero
photons. The matrix element gives
h ³ ´i
h1q,f ; 1k,i ; 0| γ K,s;k,f i b a†−K,s b
aK,s + b a†q,f bb†k,i |0; 0; 1K i = γ K,s;k,f i δ k+K,q (7.56)

analogous to eq. 7.52. Diagrams like that of Fig. 7.5 are often written without time-labels.
You can easily attach them yourself when required. So t0 goes to the bottom of the photon
line, τ to the vertex, and the ends of the partcicle and hole lines both get the time label t.

Using such diagrams and playing around with the lines and the arrows allows some
quick insight in possible events. For instance, if one flips the solid lines of Fig. 7.5 around
a horizontal axis through the vertex and then rotates them by 90o around the vertex, one
obtains Fig. 7.2. The main difference between Figs. 7.2 and 7.5 is whether the initial state
is above or below the Fermi level, see also Figs. 7.1 and 7.3. The physical interpretation of
186 CHAPTER 7. OPTICS

the diagrams of Figs. 7.2 and 7.5 is a little bit different. The latter describes the creation
of a particle-hole pair. The former simply is absorption of a photon by a particle; one can
also say that a particle in state k, i is annihilated and a particle in state q, f is created.
A third diagram, which is obtained by rotating the solid lines of Fig. 7.5 by 90o around
the vertex, is shown in Fig. 7.6.

ε q , f , =q ε k ,i , =k

=ω , =K

Figure 7.6: Absorption of a photon by a bound hole.

As always the time increases from left to right. So a hole in state q, f is annihilated
(remember, the hole moves in time against the arrow) and one in state k, i is created.
In other words, this diagram represents absorption by a hole. The “circuit” sum rules
of arrows at the vertex again give the familar conservation rules ~ω + ²k,i = ²q,f and
~K + ~k =~q. The operator corresponding to this diagram is given by

Z t h ³ ´i
1
Ub (1) (t, t0 ) ≡ b0 (t, τ ) γ K,s;k,f i b
dτ U a†−K,s bbq,f bb†k,i U
aK,s + b b0 (τ , t0 ) (7.57)
i~ t0

which corresponds to the interaction term of eq. 7.37. Suitable inital and final states are
|Ψini i = |0; 1q,f ; 1K i, and |Ψf in i = |0; 1k,i ; 0i, in the notation used before, and the matrix
element leads to the same result as eq. 7.56.

A fourth possible diagram is given in Fig. 7.7. Since q, f is a particle-line and k, i is


a hole-line in this diagram, one must have ²q,f > ²k,i . Applying the “circuit” rules then
implies that we must have an outgoing photon, i.e. this event describes the emission of a
photon by annihilation of an electron-hole pair. The circuit rules give ~ω + ²k,i = ²q,f and
~K + ~k =~q).

ε k , i , =k

ε q , f , =q

=ω , =K

Figure 7.7: Annihilation of a particle-hole pair creates a photon.


7.2. ELECTRONS, HOLES AND PHOTONS 187

The operator corresponding to this diagram is given by


Z t h ³ ´i
b (1) (t, t0 ) ≡ 1
U b0 (t, τ ) γ −K,s;q,if b
dτ U a†K,s bbk,i b
a−K,s + b b0 (τ , t0 )
aq,f U (7.58)
i~ t0

which corresponds to the interaction term of eq. 7.38. Suitable inital and final states are
a†K,s part
|Ψini i = |1q,f ; 1k,i ; 0i, and |Ψf in i = |0; 0; 1K i. Note that in this case it is only the b
of the photon operator which gives a contribution. This makes sense since it corresponds
to emission. The corresponding selection rule is k = q − K, which agrees with the circuit
rule given above. The four operators of eqs. 7.43 and 7.55—7.58 represent all possible
particle and hole combinations in the perturbation operator, cf. eqs. 7.35—7.38, and the
corresponding four diagrams give all possible first order processes.

7.2.4 Light Scattering by Free Electrons


Not all optical processes are disallowed for free electrons. The first order ones, derived
from Fermi’s golden rule, are, but higher order processes are usually allowed. So let us
return to the free electrons of Sections 7.2.1 and 7.2.2 A bit of reflection on how the
perturbation expansion is constructed, see eq. 2.22 in Chapter 2, with the perturbation
operator of eqs. 7.35—7.38 in mind, shows that all higher order terms can be constructed
in diagrammatic form by linking up the four possible first order diagrams of Figs. 7.2, 7.5,
7.6 and 7.7. Linking two diagrams like Fig. 7.2 gives the diagram shown in Fig. 7.8.

t
ε q , =q
t0 τ'
ε k , =k τ
ε l , =l

t
=K ', =ω K '
=K , =ω K
t0

Figure 7.8: A contribution to electron-photon (or Compton) scattering.

If we translate the lines directly into operators, just as we did for the first order
diagrams, we obtain the second order operator

X µ 1 ¶2 Z t Z τ0 h ³ ´i
b (2) (t, t0 ) ≡
U dτ 0 b0 (t, τ 0 ) γ −K0 ,s0 ;l b
dτ U a†K0 ,s0 b
a−K0 ,s0 + b a†qb
al
i~ t0 t0
l
h ³ ´i
b0 (τ 0 , τ ) γ K,s;k b
U a†−K,s b
aK,s + b a†l b b0 (τ , t0 )
ak U (7.59)

Remember, diagrams are read from left to right when the time is increased, but operators
are read from right to left. Suitable initial and final states are |Ψini i = |1k ; 0; 1K i, and
|Ψf in i = |1q ; 0; 1K0 i, in the notation used before. These fix the initial and final conditions,
188 CHAPTER 7. OPTICS

including the initial time t0 and the final time t. All the labels that are not fixed by these
conditions are summed over (l) or integrated over (τ , τ 0 ). Applying the circuit rules gives
~k+~K = ~l = ~q+~K0
²k + ~ωK = ²l = ²q + ~ω K0 (7.60)
where the left hand side corresponds to applying the circuit rules to the leftmost vertex in
Fig. 7.8, and the right hand side by applying them to the rightmost vertex. This actually
fixes the label l completely, so in the sum of eq. 7.59 only one term survives. Moreover,
eq. 7.60 expresses that the sum of all incoming momenta (energy) is equal to the sum
of all outgoing momenta (energy), so it represents the conservation laws. It is of course
possible to do this without diagrams. One simply calculates the second order expression
of eq. 2.22 in Chapter 2
Z t Z τ0
b (2)
U (t, t0 ) = dτ 0
dτ U b0 (τ 0 , τ )Vb(7.35) U
b0 (t, τ 0 )Vb(7.35) U b0 (τ , t0 )
t0 t0

using the perturbation operator of eq. 7.35. Working out the matrix element between
initial and final states gives the same results as using eq. 7.59, including the selection
rules of eq. 7.60. Diagrams are simply a faster way of arriving at these results.

The diagram of Fig. 7.8 also has a nice physical interpretation. We start at time t0
with an incoming electron in state k and an incoming photon in state K, and at time t
we have an outgoing electron in state q and an outgoing photon in state K0 . In other
words the diagram represents a scattering of an electron and a photon and according to
the selection rules of eq. 7.60, this scattering is elastic. We have seen this process before in
Section 5.5.3, cf. Fig. 5.5, where we meant it to represent Compton scatering. By means
of Fig. 7.8 we have now established that electron-photon scattering is in fact a second
order process (i.e. second order in the fundamental electron-radiation field coupling of
eq. 7.34). We leave it to the reader to show from the conservation laws of eq. 7.60 that
scattering is an allowed process for free electrons, i.e. no other bodies like nuclei need to
be present, which is why we have used free electron labels in this section.

If one wishes to avoid mulitiple time integrals, it is always possible to perform the
calculation in the frequency domain. Fourier transforming eq. 7.59 one obtains
X h ³ ´i
b(2) (ω) ≡
G Gb+ (ω) γ −K0 ,s0 ;l b
a 0
−K ,s0 + b
a†
a†qb
b al
0 K0 ,s0
l
h ³ ´i
b+ (ω) γ K,s;k b
G aK,s + b
a†
a†l b
b b+ (ω)
ak G (7.61)
0 −K,s 0

which is a slightly simpler expression. It is represented by the same diagram as Fig. 7.8,
but without the time labels of course. Sandwiching between initial and final states gives
the conservation laws as eq. 7.60. The end result can then be Fourier transformed back if
one wants to have a look at the time dependence.

7.3 Higher Order Processes; the Quantum Pinball Game


Any kind of optical process can be studied in this manner. The initial state fixes the
number and the kind of particles which are present at the beginning, the final state fixes
7.3. HIGHER ORDER PROCESSES; THE QUANTUM PINBALL GAME 189

the same at the end. Together the two thus describe the kind of process one whishes
to study (absorption, electron-photon scattering, electron-electron scattering etcetera).
Diagrams like Fig. 7.8 can be used to systematically construct the contributions of the
perturbation to whatever order is required. Consider for instance the diagram given in
Fig. 7.9.

Figure 7.9: An 8th order electron-electron scattering diagram.

There are two incoming electron lines at the left, and two outgoing electron lines at
the right, so this is a diagram that participates to electron-electron scattering. All the
stuff in the middle represents intermediate processes in which photons, electrons, and
holes are created after the incoming electrons enter and are annihilated again before the
outgoing electrons leave. This is the true nature of the quantum pinball game!. All such
intermediate processes are relevant because they influence the overall electron-electron
scattering probability. If we compare this to the diagrams drawn in Chapter 4, we see
that the introduction of different kinds of particles (electrons, holes, photons) makes the
perturbation expansion complicated. The perturbation order of the diagram of Fig. 7.9
can be established by counting the number of vertices, i.e. points at which the photon
and electron or hole lines connect. There are eight vertices, so this diagram participates
to the 8th order of the perturbation expansion. It will be clear that to actually calculate
this term involves a much longer expression than eq. 7.59, which contains an 8-fold
integral over time. It contains an operator which consists of a long string of creation
and annihilation operators. As we have seen in Chapters 3 and 4, if we need to go to
high orders it is advantageous to use Green functions and switch to the frequency domain
as in eq. 7.61 in order to avoid the multiple time-integrals. Even then there is a lot of
“administration” involved in keeping track of all the creation and annihilation operators.
There is a systematic way of handling the book keeping of such strings of creation and
annihilation operators. It is based upon so-called “propagators”, which are discussed in
the next chapter.

Why would we wish to calculate such high order perturbation terms? Well, sometimes
calculating the lower orders only does not yield results that are sufficiently accurate, if one
wishes to compare with experimental results. And after all, such a comparison is the only
way to see whether the experimental results are correct.15 In other cases the interaction
15
a sarcastic note, which is typical for this author. It should be the other way around, of course.
190 CHAPTER 7. OPTICS

can be so strong that the lower order perturbation terms gives results that are divergent,
which clearly is useless. The situation is not hopeless however, because we can use the
techniques presented in Chapters 3 and 4 to sum the complete perturbation series over all
orders. However, to be fair, this usually involves approximations. This is the subject of
the final chapter.

7.4 Appendix I. Interaction of an Electron with an EM field


A watertight derivation of the complete Hamiltonian for electrons and photons would go
too far here,16 but the following more or less gives the essential idea. Classically, an
electron with charge −e, moving at a velocity ṙ in an electric field E and a magnetic field
B experiences a Lorentz02 force

F = −eE − e(ṙ × B) (7.62)

and its motion is described by Newton’s law mr̈ = F. It is possible to derive this law of
motion from the following Lagrangian
1
L = m|ṙ|2 + eφ − eA · ṙ (7.63)
2
where φ, A are the usual EM scalar and vector potentials, from which the E and B fields
are derived as
∂A
E = −∇φ − and B = ∇ × A (7.64)
∂t
All fields and potentials depend upon (r, t), but sometimes I leave out these arguments to
shorten the notation. Look at the Lagrange equations of motion in cartesian coordinates.
For the x-degree of freedom one obtains from eq. 7.63
µ ¶ µ ¶
d ∂L ∂L d ∂φ ∂Ax ∂Ay ∂Az
− = (mẋ − eAx ) − e − ẋ − ẏ − ż = 0 (7.65)
dt ∂ ẋ ∂x dt ∂x ∂x ∂x ∂x

The first term on the right hand side gives


dAx
mẍ − e work out
dt
dAx ∂Ax ∂Ax dx ∂Ax dy ∂Ax dz
= + + +
dt ∂t ∂x dt ∂y dt ∂z dt
Using this in eq. 7.65, and collecting terms gives
µ ¶ µ ¶ µ ¶
∂φ ∂Ax ∂Ay ∂Ax ∂Ax ∂Az
mẍ + e − − + eẏ − − eż −
∂x ∂t ∂x ∂y ∂z ∂x
= mẍ + eEx + eẏBz − eżBy = mẍ + eEx + e(ṙ × B)x = 0 (7.66)

where eq. 7.64 has been used. Obviously this corresponds to Newton’s equation of motion
with a force descibed by eq. 7.62. A similar derivation holds for the y- and z-degrees of
16
See R. Loudon, The Quantum Theory of Light, (Clarendon, Oxford, 1981) Ch.8; or the books by
Cohen-Tannoudji cited before.
7.4. APPENDIX I. INTERACTION OF AN ELECTRON WITH AN EM FIELD 191

freedom, which proves that eq. 7.63 is indeed the correct Lagrangian for an electron in an
EM field. If we want to switch to quantum mechanics, we first have to obtain the classical
Hamiltonian. The generalized momenta are given by
∂L
px = = mẋ − eAx (7.67)
∂ ẋ
Substituting x by y (or z) gives the y (or z) component. The Hamiltonian is then given
by

H = ẋpx + ẏpy + żpz − L


|p+eA|2
= − eφ
µ 22m ¶
|p| e e2
= − eφ + (A · p) + |A|2 (7.68)
2m m 2m
expressed in generalized momenta, as it should for a Hamiltonian. Hamilton’s equations
lead to the same equations of motion as eq. 7.66, as you can easily check for yourself.
The first term in the bottom line of eq. 7.68 describes the motion of the electron in an
electrostatic potential φ(r, t). For instance, in the presence of a nucleus of charge Z in the
origin one has
−Ze2
−eφ(r, t) =
|r(t)|
just as in eq. 7.1. If another electron is present at position r0 it contributes a Coulomb
repulsion term
e2
−eφ(r, t) =
|r(t) − r0 |
and all such electrostatic terms are of course additive. The second and third term in
the bottom line of eq. 7.68 describe the interaction of the electron with the radiation
field, represented by the vector potential A(r, t). Eq. 7.68 can be used to obtain the
quantum mechanical Hamiltonian H, b by substituting r(t) → br, p(t) → pb , φ(r, t) → φ(b
r)
b
and A(r, t) →A(b r) as discussed in the text. Note that, as usual, the time disappears in
the operators. Since we work in the Schrödinger picture, the operators are assumed to
be time-independent, and all time dependence goes into the states (or in the evolution
operator, which is the same thing).

Starting from eq. 7.68, it is not difficult to produce the Hamiltonian in stone-age form
for an N -electron system
N · 2
X ¸ N
b = b
p i−Ze2 1X e2
H + + (7.69)
2m |b
ri | 2 ri − b
|b rj |
i i6=j

e X³b ´
N N
e2 X b b ri )
+ ri )·b
A(b pi + A(bri )·A(b (7.70)
m 2m
i i

Comparing to the main text, one can see that eq. 7.69 corresponds to the atomic Hamil-
b atom of eq. 7.1. Eq. 7.70 corresponds to the interaction operator Vb .
tonian H One
192 CHAPTER 7. OPTICS

b rj ) and p
might worry about the order of the operators A(b b j . In the classical expression
of eq. 7.68 they can be interchanged freely, but in quantum mechanics b b j are
rj and p
non-commutating operators. However in this case the order does not matter, if we use the
Coulomb gauge, i.e. ∇ · A =0, see Section 5.5.1. Thus operating on a wave function one
gets

b r)ψ(r) = ~ ~ ~
b · A(b
p ∇ · A(r)ψ(r) = [∇ · A(r)] ψ(r)+ A(r)· [∇ψ(r)]
i i i
~ b r) · p
= A(r)· [∇ψ(r)] =A(b b ψ(r) (7.71)
i
i.e. the two operators commutate in the Coulomb gauge and we can safely write for the
interaction operator17
N N
e X b ri ) + e
2 X
b ri )
b ri )·A(b
Vb = b i ·A(b
p A(b (7.72)
m 2m
i i

The second sum in this expression actually does not give a contribution to the processes
discussed in this chapter. So we have skipped this term altogether to arrive at the ex-
pression of eq. 7.31. (It is a real term however, and in general one has to take it into
account).

7.4.1 Dipole Approximation


For an electron in an atom we did not use eq. 7.72 for the interaction operator, but a
much simpler one, namely
N
X
Vb = e b b
rj ·E (7.73)
j=1

see eq. 7.6. Let us see how we can derive this. Return to the classical expression of eq.
7.68. From classical electrodynamics we know that it is possible to define a so-called gauge
function χ(r, t) and use it to define new vector and scalar potentials by

A0 (r, t) = A(r, t) + ∇χ(r, t)



φ0 (r, t) = φ(r, t) − χ(r, t) (7.74)
∂t
The magnetic and electric fields are not changed by this

B(r, t) = ∇ × A0 (r, t) = ∇ × A(r, t)


∂ ∂
E(r, t) = − A0 (r, t) − ∇φ0 (r, t) = − A(r, t) − ∇φ(r, t) (7.75)
∂t ∂t
since ∇ × ∇χ(r, t) = 0. The transformation of eq. 7.74 is called a gauge transformation.
Since the E and B fields are unchanged by a gauge transformation, we can also use the
potentials φ0 and A0 in our classical Hamiltonian of eq. 7.68 instead of φ and A, since it
gives the same equations of motion, cf. eq. 7.66.
³ ´
17
If you are not using the Coulomb gauge, you should use the symmetric form 1 b ri )+A(b
b i ·A(b
p b ri )·b
pi .
2
7.5. APPENDIX II. RELATIVISTIC ELECTRONS AND HOLES 193

The key element for the atom is the so-called long wave length approximation, where
we assume that the wave length of the EM radiation is much larger than the size of the
atom. In this limit we can treat the radiation field as being homogeneous, i.e. spatially
independent, so A(r, t) ≈ A(t). We now apply two consecutive gauge transformations.
The first one is defined by
Z
e t
χ(r, t) = |A(r, t0 )|2 dt0 (7.76)
m
which gives
Z
e t
A0 (r, t) = A(r, t) + ∇|A(r, t0 )|2 dt0 ≈ A(t)
m
e
φ0 (r, t) = φ(r, t) − |A(r, t)|2 (7.77)
m
where the first line holds because of the long wave length approximation. Using the
transformed potentials φ0 and A0 in eq. 7.68 instead of φ and A gives
µ 2 ¶
|p| e
H= − eφ + (A · p) (7.78)
2m m

i.e. we got rid off the |A|2 term. Now apply a second gauge transformation defined as

χ(r, t) = −r · A(t) (7.79)

and get

A0 (r, t) = A(t) + ∇χ(r, t) = 0



φ0 (r, t) = φ(r, t) − χ(r, t) = φ(r, t) − r · E(t) (7.80)
∂t
Again, using the transformed potentials φ0 and A0 in eq. 7.80 instead of φ and A gives
µ 2 ¶
|p|
H= − eφ + er · E(t) (7.81)
2m
Transforming this into a quantum mechanical operator, we observe that the last term in
this expression gives the interaction operator Vb of eq. 7.73. Note these gauge transfor-
mation tricks only work in the long wave length limit, i.e. when the spatial dependence of
the fields can be neglected. This is valid for small systems like atoms, but in other cases
the more general form of eq. 7.72 has to be used.

7.5 Appendix II. Relativistic Electrons and Holes


I have carefully discussed the direction of the particle and hole lines in Section 7.2, but I
have not talked much about the photon lines, which I have drawn almost vertically. The
reason is that we have been doing non-relativistic quantum mechanics (the Schrödinger
equation is non-relativistic). The speed of light is always much larger than the speed of
the electrons and holes. Photons move very fast, so letting time increase from left to right,
194 CHAPTER 7. OPTICS

ε q , =q
ε q , =q

ε k , =k
ε k , =k
=ω , =K
=ω , =K

(a) (b)

Figure 7.10: Creation of a particle-hole pair; (a) non-relativistic, (b) relativistic.

the photon line has a very high slope, and have to be drawn (almost) vertically, as in Fig.
7.10(a).
In a relativistic theory the electrons and holes can acquire a large speed, which is of
course always smaller than the speed of light, but can be of the same order of magnitude.
In a relativistic diagram, the photon-line has to be drawn more slanted, as in Fig. 7.10(b)
to stress that the photon’s speed is comparable to the “particle” speeds. The relativistic
diagrams (and the perturbation expansion they represent) actually are very similar to
the non-relativistic ones. The only difference is that the electrons and holes have to be
described by the Dirac equation (which gives the appropriate description for relativistic
spin- 12 particles) instead of the Schrödinger equation. Relativistic quantum mechanics is
not part of this course, but I will give some elementary results in this appendix.

In relativistic mechanics the energy of a particle is given by

E 2 = m20 c4 + p2 c2 or
q
E = ± m20 c4 + p2 c2 (7.82)

see eqs. 5.80 and 5.81 in Section 5.5.3. In relativistic quantum mechanics Planck’s and
DeBroglie’s relations are still valid, E = ~ω and p =~k, but now the dispersion relation is
given by eq. 7.82. The solutions of the relativistic wave equations which take over the role
of the Schrödinger equation (the Klein-Gordon equation, or the Dirac equation) still have
plane waves ei(k·r−ωt) as a solution for a free particle. However, according to eq. 7.82, the
energy (or frequency) can be negative as well as positive. If we let the momentum p run
from 0 to ∞, the energy runs from ²q = E = m0 c2 to ∞ but also from ²k = −m0 c2 to
−∞. The energy spectrum then looks like Fig. 7.11.
The region of energy between −m0 c2 and m0 c2 has no allowed energy levels. In other
words, it represents a “band gap”, in the same sense as the band gap between the valence
and conduction bands in a semiconductor. The negative energies can be given a convenient
interpretation, if we associate −²k with the energy of a hole. Note that the latter is positive
if we choose the zero-point of energy as indicated in Fig. 7.11. The particle we are talking
about is a (relativistic) electron, the corresponding hole is called a positron. The band
gap is Eg = 2m0 c2 , which, considering the rest mass of an electron, is ∼ 106 eV. The gap
of an ordinary semiconductor like silicon is slightly lower, Eg ∼ 100 eV (which allows us
to use non-relativistic quantum mechanics for those materials). Thus free space is a very
7.5. APPENDIX II. RELATIVISTIC ELECTRONS AND HOLES 195

εq


m0 c 2

− m0 c 2

−ε k

Figure 7.11: Relativistic particle-hole spectrum.

large band gap semiconductor. As we know, the number of electrons in the universe is
much larger than the number of positrons. Thus the universe is a heavily doped n-type
semiconductor.

Again absorption of a photon by a free electron or positron is a forbidden process. Call


Ei , pi the energy and momentum of the incoming electron (which are related by eq. 7.82)
and Ef , pf the energy and momentum of the outgoing electron. Call E, p the energy and
momentum of the photon, which are related by

E = pc (7.83)

since the photon has a zero rest mass. The conservation of energy and momentum still
follow from Fermi’s golden rule

Ei + E = Ef
pi + p = pf (7.84)

These laws cannot be obeyed simultaneously for a free electron. The proof is straightfor-
ward,18 but it is also instructive to look it graphically. That is done in Fig. 7.12.
The top and bottom thick curves represent E(p) for positive and negative energies
respectively, according to eq. 7.82, where we have used atomic units in which m0 = 1,
c = 1, and ~ = 1. The thin straight line through the origin gives E(p) for a photon,
18
From the first line of eq. 7.84 one obtains Ei2 + Ef2 − 2Ei Ef = E 2 and from £ the second line¤
2
c (p2i
+ p2f − 2pi · pf ) = p2 c2 = E 2 . Equating these two expressions one obtains pi · pf + m20 c2 ∓
³p q ´
p2i + m20 c2 p2f + m20 c2 = 0. One easily sees that the [...] term is smaller than the (...) term, so their
sum or difference cannot be zero, unless pi = pf , but then we have no event at all.
196 CHAPTER 7. OPTICS

2
E
1

-3 -2 -1 0 1 2 3
p
-1

-2

-3

Figure 7.12: Energy-momentum dispersion relations of an electron (top thick curve), a


hole (bottom thick curve), and a photon (thin straight line).

according to eq. 7.83. In order to obey the conservation laws of eq. 7.84 for absorption by
an electron one has to start from a point (pi , Ei ) on the upper curve; let’s say the point
(0, 1). From this point one draws a straight line in the (1, 1) direction to add the photon
energy (remember, atomic units). This is indicated by the top straight line in Fig. 7.12.
Any point on this line now satisfies the right handside of eq. 7.84. In order to satisfy
the left handside of eq. 7.84, the line has to cross the thick curve again. As you can see,
it never does, so eq. 7.84 cannot be satisfied. If you are not completely convinced, the
tangential of the thick curve is

dE
=v
dp

which is the speed of the electron, as you can easily check from eq. 7.82 and eq. 5.81 in
Section 5.5.3. Of course v < c, so the straight line can never cross the thick curves at
more than one point. You can convince yourself by shifting the straight line around as
indicated in Fig. 7.12.

Note that by Fig. 7.12 we have not only shown that absorption of a photon by a free
electron or positron is forbidden, but also electron-positron pair creation by a photon in
free space, emission from a free electron or positron, and electron-positron pair annihila-
tion in free space. As before, in the presence of a fourth object, a nucleus for instance,
these processes become allowed, because this object supplies the momentum (and energy)
required to obey the conservation laws. A second photon can also play that role, which
is the case in Compton scattering, Section 5.5.3. Assuming that nuclei are present, the
“absorption” process shown in Fig. 7.11 is nothing else than the creation of a particle-hole
pair, similar to what is shown in Fig. 7.3. It must be represented by the same diagram
as Fig. 7.5. The diagrams corresponding to pair annihilation and absorption can also be
found in Section 7.2. In the presence of nuclei a high energy electron can emit a photon,
which can create an electron-hole pair, each of which can emit a photon, etcetera. This
7.5. APPENDIX II. RELATIVISTIC ELECTRONS AND HOLES 197

cascade process, in which the energy is distributed over a large number of electrons, holes,
and photons, is shown in Fig. 7.13.

Figure 7.13: Cascade induced by a high energy electron.

The cascade is described by linking a number of the diagrams shown in Section 7.2.
In this case one needs the emission diagram (Fig. 7.2 with the arrow on the photon line
reversed) and the pair creation diagram (Fig. 7.5). The cascade is produced by linking
all possible combinations of these two diagrams. Obviously, time increases going from top
to bottom in this figure; e− stands for an electron, e+ stands for a hole (positron), and γ
stands for a γ-photon. You can draw the arrows yourself. The creation of electron-hole
pairs stops when the energy of the photon is below the band gap, cf. Fig. 7.11, the
remaining process is the emission of low energy radiation by the electrons and holes. A
cascade like this happens when cosmic rays enter the atmosphere. It also happens when
we inject so-called “hot” electrons (i.e. high energy electrons) in semiconductors. There,
the energies are much lower, and the mechanism is slightly different, but the main idea is
similar.19

19
In a solid state, “hot” electrons can also emit other particles, such as phonons or plasmons, to get rid
of their energy. Plasmons can proceed the cascade, but phonons have an energy which is too low. They
latter simply heat up the lattice. The relative efficiency with which the various particles are generated
depends very much upon the energy of the incoming “hot” electrons.
198 CHAPTER 7. OPTICS
Part III

Interacting Particles

199
Chapter 8

Propagators and Diagrams

“...You might say, “My Quetzalcoatl! What tedium—counting beans, putting them in, taking them
out—what a job!” To which the priest would reply, “That’s why we have the rules for the bars and
the dots. The rules are tricky, but they are a much more efficient way of getting the answer than
by counting beans...””, Richard Feynman, QED.

Chapters 2, 3, 4 and 7 all dealed with the following question. Suppose at a time t = t0
we prepare our system in a state |ii, what is the probability that at a later time t = t1
the system will be in a state |f i? The answer to that question is given by the transition
probability
¯D ¯ ¯ E¯2
¯ ¯b ¯ ¯
Wi→f = ¯ f ¯U (t1 , t0 )¯ i ¯ (8.1)

b 1 , t0 ) is the time evolution operator. We handled this expression using pertur-


where U(t
bation theory. In first order (the Born approximation) this gave us Fermi’s golden rule.
Expansion and summation of the whole perturbation series was possible using Fourier
transforms and Green functions. We have encountered several examples for which we cal-
culated these transition probabilities. Unfortunately, but not surprisingly, there are many
physical systems which are too complicated for such “exact” calculations. In particular
this is true for systems comprising many interacting particles.1 We have to change our
tune and, instead of trying to find exact solutions, we must strive to find reasonable ap-
proximations that cover the essential physics of the system. For a many particle system
the tools of the second quantization formalism are especially handy in achieving this goal.
In the first section we will define a very important quantity called the single particle
propagator (also called the single particle Green function; it is related to the Green func-
tions we encountered before). It plays a vital role in interpreting prototype “typical”
experiments. In this chapter this propagator will be applied to cases of increasing com-
plexity. In comparison with the previous chapter, we go back to systems that involve one
kind of particle (or two kinds if you want, a particle and its anti-particle; the two always
go together). We kick off with just a single particle, where we will recover the results of
Chapter 4. Then we will treat an example of a system which consists of many particles
with a very simple interaction. This rounds up the cases that are, in principle, exactly
1
In fact, there is not a single, non-trivial (i.e. in which the particles interact with each other), many-
body system that can be solved exactly, neither in quantum, nor in classical mechanics. The situation is
hopeless but not desperate, since in many cases we can find reasonable approximations.

201
202 CHAPTER 8. PROPAGATORS AND DIAGRAMS

solvable, although already the latter will be far from trivial as we will see. A short dis-
cussion is then given of the very important physical system in which the particles have
pairwise interactions. The last section introduces the so-called spectral function, which
plays a vital role both in the interpretation of experiments and of the formal properties
of the propagator. This is used to give an analysis of two “real” experiments, namely
photoemission and inverse photoemission. The two appendices give an introduction to the
mathematical complexities of many interacting particle systems.

8.1 The Single Particle Propagator


”The time has come,” the Walrus said,
”To talk of many things:
Of shoes—and ships—and sealing-wax—
Of cabbages—and kings—
And why the sea is boiling hot—
And whether pigs have wings.”, Lewis Caroll, The Walrus and the Carpenter.

We “prepare” a specific initial state,

|i(t0 )i = b b (t0 , 0) |0i


c†i U (8.2)

i.e. we start our system at t = 0 in a state |0i, which we propagate to t = t0 . In most of


our applications |0i will be the ground state of the system (the vacuum state), hence the
c†i operate on |0i. We write the
notation. At t0 we create a particle in state |ii by letting b
final state as

|f (t1 )i = b b (t1 , 0) |0i


c†f U (8.3)

with a similar interpretation. The transition probability is of course still given by eq. 8.1.
We can write the matrix element as
D ¯ ¯ E D ¯ ¯ E
¯b ¯ ¯ b† b (t1 , t0 ) b b (t0 , 0)¯¯ 0
f (t1 ) ¯U (t1 , t0 )¯ i(t0 ) = 0 ¯U (t1 , 0) b
cf U c†i U (8.4)

and use

b (t1 , t0 ) = U
U b (t1 , 0) U
b (0, t0 ) = U
b (t1 , 0) U
b † (t0 , 0)

By defining the so-called “Heisenberg” operators as in Section 2.6, eq. 2.63 (dropping the
“H” and “S” subscripts)

b b † (t, 0) b
ck (t) = U b (t, 0)
ck U
b b † (t, 0) b
c† (t) = U b (t, 0)
c† U (8.5)
k k

we then arrive at
D ¯ ¯ E D ¯ ¯ E
¯b ¯ ¯ ¯
f (t1 ) ¯U (t1 , t0 )¯ i(t0 ) = 0 ¯b c†i (t0 )¯ 0
cf (t1 ) b
≡ i~G+ (f, i, t1 − t0 ) ; t1 > t0 (8.6)
8.1. THE SINGLE PARTICLE PROPAGATOR 203

This quantity plays a prominent role in many-particle physics. It is called the single-
particle propagator .2 We used the notation G+ because in view of Chapters 3 and 4, we
are obviously dealing with a matrix element of a Green function operator, see in particular
Sections 3.2.1 and 4.2. Such matrix elements are called Green functions (and rightly so,
since they form a representation of the Green function operator on a basis set), see the
mathematical intermezzo in Section 3.2.1 or eq. 4.20 in Section 4.2. A synonym for the
phrase single-particle propagator therefore is single-particle Green function.3 In view of
Chapter 7 it will be clear that one can also look at more complicated propagators. For
instance Fig. 7.9 in Section 7.3 descibes an event in which the initial state contains two
particles and also the final state contains two. Corresponding “two-particle” propagators
and Green functions exist, but we won’t consider them in this course. Once you know
how to deal with the “single-particle propagator”, it is straight-forward to extend this
knowledge to other propagators. So whenever I use the word “propagator” in these notes,
the “single-particle propagator” is meant.
In Chapter 3 we used a Dyson expansion for the Green function operator to sum
the perturbation series resulting from a perturbing potential Vb . In Chapter 4 we did a
similar expansion for a Green function (matrix element) instead of the operator, using a
perturbation series in terms of a quantity called the self-energy Σ. The latter technique
proves to be more versatile and also applicable to many particle systems. Note that eq.
8.6 is very straight-forward to interpret if we read it from right to left. We start with the
system in the vacuum state |0i. At t = t0 we create a particle by b c†i in the state i and at
a later time t = t1 , we probe by applying bcf the probability amplitude that the particle is
in state f .

8.1.1 A Gedanken Experiment


The physical idea behind the propagator is to mimick the “typical” experiment of Fig.
3.1. At some initial time t0 we send in a particle into the system, which is in state |ki,
having a well-defined energy ²k and momentum ~k. From the point of view of the system
starting in its ground state |0i, at time t0 suddenly a (extra) particle appears, which is
created in |ki by bc†k . At some later time we check the probability of detecting this particle
in state |li (with a well-defined energy ²l and momentum ~l) and the system in its original
state |0i. “Detection” necessarily involves annihilation of the particle, which admittedly
is not very subttle. This “gedanken”-experiment is shown in Fig. 8.1.
This theoretical gedanken experiment is the simplest thing you can do in relation to
the “typical” experiment of Fig. 3.1. If you are an experimentalist, you may wonder how
this “creation of a particle” can be done in practice. The final sections of this chapter give
some examples of real experiments: (inverse) photo-emission. Real experiments are gen-
erally much more complicated than the gedanken experiment of Fig. 8.1. A whole chain
of events is required to create and detect particles. For a start, one needs a source for
creating particles, e.g. an electron gun to produce electrons or a laser to produce photons.
Then the particles should penetrate the boundaries (surfaces) of the system, before they
can propagate through the system. The same is true for emerging particles; these again
h i
2 b (t1 , t0 ) = exp − i (t1 − t0 )H
Since U b and we assume H
b to be time-independent, it is time differences
~
(t1 − t0 ) that appear everywhere.
3
Mattuck’s eqs. 3.3 and 4.29 describe the same thing in words. The full mathematical discussion is
found in his chapter 9.
204 CHAPTER 8. PROPAGATORS AND DIAGRAMS

t0 t1
system system
+
âk âl
εk | 0〉 Uˆ (t1 − t0 ) | 0〉 εl
=k =l

Figure 8.1: A visual interpretation of the propagator i~G+ (l, k,t1 − t0 ).

have to penetrate the boundaries of the system in order to come out. Finally, in order
to be detected, they have to interact with a specially build detector, e.g. a photocell or
a fluorescent screen. If you want to do things really right, all of these events have to be
considered in a quantum mechanical treatment. However with some reasonable approxi-
mations, experiments such as photoemission can be interpreted in terms of properties of
the propagator i~G+ (l, k,t1 − t0 ) only, as is shown later on. Therefore, we will stick to the
propagator and the gedanken experiment it represents.

8.1.2 Particle and Hole Propagators


Note that above we have specified neither the kind of particle nor the vacuum state, so
the formalism is perfectly general. Let us now take a many fermion system, where the
“vacuum” is in fact the ground state of the system. As we have seen in the previous
chapter, we can create particles as well as holes. So we expect to have two kinds of
single-particle propagators. The particle propagator is (see eq. 6.46)
D ¯ ¯ E D ¯ ¯ E
¯ ¯ ¯ ¯
0 ¯b a†k (t0 )¯ 0
al (t1 ) b = 0 ¯b c†k (t0 )¯ 0
cl (t1 ) b ; ²k , ²l > ²F
≡ i~G+ (l, k, t1 − t0 ) ; t1 > t0 (8.7)

A particle is created in state k and we calculate its transition probability to state l. The
energy of both states is above the Fermi level, because it is a particle we are dealing with.

The definition of the hole propagator is a little bit tricky. Analogous to the electron,
a good definition of the hole propagator is (again, see eq. 6.46)
D ¯ ¯ E D ¯ ¯ E
¯ ¯ ¯ † ¯
0 ¯bbl (t1 ) bb†k (t0 )¯ 0 = 0 ¯b
cl (t1 ) b
ck (t0 )¯ 0 ; ²k , ²l ≤ ²F
≡ −i~G− (k, l, t0 − t1 ) ; t1 ≥ t0 (8.8)

The physics is clear; an hole is created in state k and we calculate its transition probability
to state l. The notation in the last line needs a bit of explaining. We associate the
propagator with a Green function G− of which the time argument is negative, t0 − t1 ≤ 0.
Reconsider the definition of Green functions and the small δ limiting procedure presented
in Chapter 3. For negative time arguments, one has to use a

Θδ (−t) = e+δ.t t≤0


= 0 t>0 (8.9)
8.2. A SINGLE PARTICLE OR HOLE 205

in integrals like those of Section 3.2.4, in order to get a converging Fourier integral. The
negative time is where the “−” superscript on G comes from. It leads to a −iδ in the
denominator of eq. 3.26. Furthermore we have put a “−” in front of the G− in eq. 8.8 and
have interchanged the k, l arguments. We are perfectly free to do so, any mathematical
definition is o.k., and as long as we are consistent in its use, the physics should not be
affected. But why haven’t we chosen something more in line with eq. 8.7 and defined
a hole propagator as G+ hole ? The idea is that both definitions eqs. 8.7 and 8.8 can be
combined into one one equation by defining
D ¯ ¯ E
¯ ¯
i~G(l, k, t1 − t0 ) = 0 ¯T {b c†k (t0 )}¯ 0
cl (t1 ) b (8.10)

where

T {b c†k (t0 )} = b
cl (t1 ) b c†k (t0 )
cl (t1 ) b ; t1 > t0
= c†k (t0 ) b
−b cl (t1 ) ; t1 ≤ t0 (8.11)

The notation T {....} stands for the notorious Wick’s time ordering operator ; its general
definition is given by Mattuck’s eq. 9.4. The idea is that it orders the operators it works on
such, that the times increase going from right to left. Whenever you have to interchange
two operators in order to get them into the correct time order, a “−” sign appears (that
is for fermions; for bosons there is no change of sign). This is just a mathematical trick to
combine the electron and hole propagators into one object, there is no deeper meaning.Note
that G(l, k, t1 − t0 ) becomes the particle propagator for t1 > t0 , eq. 8.7, because operating
with bc†k on the vaccuum gives only a non-zero result if ²k > ²F . Similarily it becomes a
hole propagator for t1 ≤ t0 , eq. 8.8, because operating with b cl on the vacuum gives only
a non-zero result if ²l ≤ ²F . The trick to combine electron and hole propagators into one
object can be handy in actual calculations, since we can do all intermediate algebra with
a single object G(l, k, t), and worry about whether we are above or below the Fermi-level
(i.e. whether we have electrons or holes) only in the final stage. Otherwise we have to
drag two versions of propagators along, one for electrons and one for holes. If we are
going to multiply them in a perturbation expansion, as in Chapter 4, we have to keep
track of all the possible combinations. A product of two of them then gives four possible
combinations, etcetera. We will not use this trick in this chapter, and work with separate
particle and hole propagators G+ and G− to keep electrons and holes apart. I had to
discuss the trick now, because it is the generally accepted notation.

8.2 A Single Particle or Hole


8.2.1 Particle Scattering
In this section we will recover the single particle results as obtained in Chapter 4. Let
|ki be a state of the single (free) particle basis set, see eq. 4.16. We consider the particle
propagator of eqs. 8.4, 8.6 and 8.10
D ¯ ¯ E
¯ ¯
i~G+ (l, k, t1 − t0 ) = 0 ¯b a†k (t0 )¯ 0 ;
al (t1 ) b t1 > t0
D ¯ ¯ E
¯ b† b † b ¯
= 0 ¯U (t1 , 0) b al U (t1 , t0 ) b
ak U (t0 , 0)¯ 0 (8.12)
D ¯ ¯ E
¯ b ¯
= 0 ¯b a†k ¯ 0
al U (t1 , t0 ) b
206 CHAPTER 8. PROPAGATORS AND DIAGRAMS
h i £ ¤
Since Ub (t, 0) = exp − i tH b , we have U b (t, 0) |0i = exp − i tE0 |0i with E0 the energy
~ ~
of the vacuum. Usually we leave this energy undetermined in the second quantization
formalism, or, to put it differently, all energies are measured with respect to E0 . Therefore,
with no loss of generality we have set E0 = 0. We can transform to stone-age notation by
noting that b a†k creates a particle in a state |ki, and thus b a†k |0i → |ki, where the left hand
side is in second quantization and the right hand side is in stone-age notation. Eq. 8.12
now becomes
D ¯ ¯ E
¯b ¯
i~G+ (l, k, t1 − t0 ) = l ¯U (t1 , t0 )¯ k ; t1 > t0 (8.13)

which is the familiar expression of Chapter 3. All that we learned in that chapter can be
applied right away. So for instance the unperturbed propagator becomes, according to eq.
3.47
D ¯ ¯ E
¯ b †¯
i~G+0 (l, k, t1 − t0 ) = 0 ¯ b
a U (t
l 0 1 0, t ) b
ak¯ 0
¿ ¯ · ¸¯ À
¯ i ¯
= ¯ b
l ¯exp − (t1 − t0 )H0 ¯¯ k
~
i
= e− ~ (t1 −t0 )²k δ l,k ; t1 > t0 (8.14)

Applying the Fourier transform to the frequency domain we get, according to eq. 4.22 in
Chapter 4

1
i~G+
0 (l, k, ω) = δ l,k (8.15)
~ω − ²k + iδ

If a perturbation is present in the form of scattering centers, we can use the techniques
of Chapter 4. Write down the Dyson expansion for G+ (k, k, ω),4 and solve it formally in
terms of the self-energy Σ(k,ω). One gets according to eq. 4.27

1
G+ (k, k, ω) = (8.16)
~ω − ²k − Σ (k, ω) + iδ

Fourier transforming to the time domain we get, according to eq. 4.39


i 1
i~G+ (k, k, t1 − t0 ) = e− ~ [²k +∆k ](t1 −t0 ) e− 2 Γk (t1 −t0 ) ; t1 > t0 (8.17)

where the level shift ∆k and the (inverse) lifetime τ −1


k = Γk are connected to the real and
imaginary parts of the self-energy Σ(k,²k /~) according to eq. 4.38. As usual, eq. 8.14 is
associated with the propagation of a free particle, and the form of eq. 8.17 is assigned
to the propagation of a quasi-particle. Of course, we could have done the whole analysis
on the basis of Feynman diagrams as in Figs. 4.4, 4.5 and 4.6. For example, eq. 8.14 is
written as

k l
t0 t1 ;
i~G+
0 (l, k, t1 − t0 ) = t1 > t0 (8.18)
4
It is possible to get expressions for the “off-diagonal” elements G+ (l, k, ω) as well. For the moment,
these are not of interest to us.
8.2. A SINGLE PARTICLE OR HOLE 207

8.2.2 The Second Quantization Connection

I will try to explain why propagators and their diagrams are handy in practical calcu-
lations. Let us pretent we didn’t know about Chapter 4 and try to solve our scattering
problem with a fresh start, but now using the second quantization formalism. The scat-
tering potential is obviously a single particle operator, which in second quantization form
looks like

X
Vb = a†mb
Vmnb an (8.19)
m,n

We wish to calculate the propagator

D ¯ ¯ E
¯ ¯
i~G+ (k, k, t1 − t0 ) = a†k (t0 )¯ 0 ;
al (t1 ) b
0 ¯b t1 > t0
D ¯ ¯ E
¯ b ¯
= 0 ¯b a†k ¯ 0
ak U (t1 , t0 ) b (8.20)

which according to eqs. 8.13 and 4.19 in Section 4.2 gives us to the probability that
the particle stays in the same state. We are going to expand the operator Ub (t1 , t0 ) in a
perturbation series, according to eqs. 2.21 and 2.22 in Section 2.2. Look at the second
order term as an example

D ¯ ¯ E
¯ b (2) ¯
ak U (t1 , t0 ) b
0 ¯b a†k ¯ 0
µ ¶2 Z t1 Z τ2 D ¯ ¯ E
1 ¯ b ¯
= dτ 2 dτ 1 0 ¯b ak U0 (t1 , τ 2 ) Vb U
b0 (τ 2 , τ 1 ) Vb U
b0 (τ 1 , t0 ) b
a†k ¯ 0
i~ t0 t0
µ ¶2 X Z t1 Z τ2
1
= Vpq Vlm dτ 2 dτ 1
i~ t0 t0
l,m,p,q
D ¯ ¯ E
¯ b † b † b †¯
ak U0 (t1 , τ 2 ) b
0 ¯b apbaq U0 (τ 2 , τ 1 ) b am U0 (τ 1 , t0 ) b
al b ak ¯ 0 (8.21)

where the third/fourth lines result from inserting eq. 8.19 two times. We now insert
resolutions of identity between each pair of adjacent creation and annihilation operators.
In general, the resolution of identity sums over all possible states having any possible
number of particles

X X
I = |0ih0| + |1i ih1i | + |1i , 1j ih1i , 1j | + .... (8.22)
i i,j

(where 1i indicates that there is one particle in state i, etcetera). However in this case we
can save us a lot of trouble. Reading the string of creation and annihilation operators in
the bottom line of eq. 8.21 from right to left, one observes that each creation operator is
followed by an annihilation operator. After applying such a pair, the number of particles is
not changed, and since we started with zero particles, only the |0ih0| term in the resolution
208 CHAPTER 8. PROPAGATORS AND DIAGRAMS

of identity gives a contribution. In other words


D ¯ ¯ E
¯ b (2) ¯
0 ¯bak U (t1 , t0 ) b a†k ¯ 0
µ ¶2 X Z t1 Z τ2
1
= Vpq Vlm dτ 2 dτ 1
i~ t0 t0
l,m,p,q
D ¯ ¯ E
¯ b b0 (τ 2 , τ 1 ) b b0 (τ 1 , t0 ) b ¯
0 ¯b a†p I b
ak U0 (t1 , τ 2 ) b aq U a†l I b
am U a†k ¯ 0
µ ¶2 X Z t1 Z τ2
1
= Vpq Vlm dτ 2 dτ 1
i~ t0 t0
l,m,p,q
D ¯ ¯ E
¯ b b0 (τ 2 , τ 1 ) b b0 (τ 1 , t0 ) b ¯
0 ¯b a†p |0ih0| b
ak U0 (t1 , τ 2 ) b aq U a†l |0ih0| b am U a†k ¯ 0
X Z t1 Z τ2
= i~ Vpq Vlm dτ 2 dτ 1
l,m,p,q t0 t0
+
G0 (k, p, t1 − τ 2 )G+ +
0 (q, l, τ 2 − τ 1 )G0 (m, k, τ 1 − t0 ) (8.23)

according to the definition of eq. 8.14. Making use of the fact that each unperturbed
propagator leads to a Kronecker-δ, cf. eq. 8.14, eq. 8.23 can be simplified to

1 D ¯¯ b (2) ¯ E XZ ∞ Z ∞
†¯
ak U (t1 , t0 ) b
0 ¯b ak ¯ 0 = dτ 2 dτ 1
i~ −∞ −∞ l
G+
0 (k, k, t1 − τ 2 )Vkl G+
0 (l, l, τ 2 − τ 1 )Vlk G+
0 (k, k, τ 1 − t0 ) (8.24)

Note that we have extended the range of our integration to (−∞, ∞), which we can safely
do because the propagators are zero for negative time arguments. Putting the icing on
the cake, we can get rid of the time integrals by Fourier transforming to the frequency
domain and get the expression
X
G+ + +
0 (k, k, ω)Vkl G0 (l, l, ω)Vlk G0 (k, k, ω) (8.25)
l

This is identical to the bottom line of eq. 4.23 in Section 4.2 !! It can be represented by
a Feynman diagram, see the bottom line in Fig. 4.4. I reproduce it in Fig. 8.2.

l
k k
Vkl Vlk

Figure 8.2: Second order potential scattering of a particle.

IMPORTANT LESSON
The important lesson to be drawn from this section is the following. We will in the future
have to deal with perturbation operators which are expressed in second quantization form,
and which can also be a lot more complicated than that of eq. 8.19. If you want to
calculate a propagator like that of eq. 8.20 (and, believe me, you want to), you need
8.2. A SINGLE PARTICLE OR HOLE 209

to apply the perturbation expansion. Applying it blindly leaves you with a complicated
looking expression involving a string of creation and annihilation operators like eq. 8.21
(and it becomes more complicated the higher you go in the order of the perturbation).
However, the end result of eq. 8.24 or eq. 8.25 looks quite simple; it consists of a string of
unperturbed propagators G+ 0 and matrix elements Vkl . It is even simpler if one starts from
the diagram of Fig. 8.2, because that can be written down immediately. By substituting
each arrow by a G+ 0 according to eq. 8.18, and each dot with a matrix element Vkl , one
immediately obtains eq. 8.24 or eq. 8.25. Summation over the intermediate labels (l) is
assumed, and, in the time domain, integration over all intermediate times. This procedure
is valid in general, also for many particles with complicated interactions!! It takes care
of the “administration” of creation/annihilation operators we worried about in Section
7.3.

8.2.3 Hole Scattering


There is nothing in this formalism which says that we cannot do the same thing with a
single hole, instead of a particle. We simply use the definition of the hole propagator, eq.
8.8.
D ¯ ¯ E
¯ ¯
−i~G− (l, k, t1 − t0 ) = 0 ¯bbk (t0 ) bb†l (t1 )¯ 0 ; t1 ≤ t0
D ¯ ¯ E
¯b ¯
= k ¯U (t0 , t1 )¯ l (8.26)

making use of the fact that bb†l |0i → |li creates a hole in state |li (wave function φl (r)),
according to Table 6.2. In case you might wonder how one introduces a hole experimentally,
there are ways for doing that. In photoemission, for instance, an electron is ejected from
the material, which leaves a hole behind. Using eq. 8.14 one finds for the free hole
propagator
i
−i~G−
0 (l, k, t1 − t0 ) = e
− ~ (t0 −t1 )(−²k )
δ l,k ; t1 ≤ t0 (8.27)

where −²k is the energy level of the hole. Note that in general this energy will be negative
with respect to E0 , the energy of the vacuum, cf. Table 6.2. This figures, since by creating
a hole we are taking energy out of the system. Fourier transforming to the frequency
domain we get
1
G−
0 (l, k, ω) = δ l,k (8.28)
~ω − ²k − iδ
A “−iδ” now appears in the denominator since we have t1 ≤ t0 in eq. 8.27 (see subsection
8.1.2). Writing diagrams we need something to distinguish holes from particles. This is
usually done by reversing the arrow of eq. 8.18

l k
t1 t0
−i~G−
0 (l, k, t1 − t0 ) = ; t1 ≤ t0 (8.29)

Note that the time still increases from left to right, as in eq. 8.18. According to eq. 8.26,
the hole is first created in state |li at time t1 and it is checked at a later time t0 for the
probability (amplitude) for being in state |ki. In time a hole thus propagates against the
210 CHAPTER 8. PROPAGATORS AND DIAGRAMS

direction of the arrow. This is simply a convention for writing down a hole propagator as
a diagram; you should not seek for any deeper meaning.5 In principle, we might as well
have chosen a normal, instead of a reversed arrow, and a dotted, instead of a solid line
to distinguish a hole from an electron propagator, but eq. 8.29 is the standard diagram
convention. The whole analysis on the basis of Feynman diagrams as in Chapter 4 can be
repeated using these reversed arrow diagrams. The second order diagram similar to Fig.
8.2, but now for holes, looks like Fig. 8.3.
l
k k
Vkl Vlk

Figure 8.3: Second order potential scattering of a hole.

It corresponds to
XZ ∞ Z ∞
dτ 2 dτ 1
l −∞ −∞

G−0 (k, k, t1 − τ 2 )Vkl G− −


0 (l, l, τ 2 − τ 1 )Vlk G0 (k, k, τ 1 − t0 ) (8.30)
in the time domain and
X
G− − −
0 (k, k, ω)Vkl G0 (l, l, ω)Vlk G0 (k, k, ω) (8.31)
l
in the frequency domain. Such perturbation terms or diagrams can be summed along the
same lines as Figs. 4.4, 4.5 and 4.6 in Chapter 4. The end result is similar to eq. 8.17
i 1
−i~G− (l, k, t1 − t0 ) = −e− ~ [²k +∆k ](t1 −t0 ) e+ 2 Γk (t1 −t0 ) ; t1 ≤ t0 (8.32)
where the parameters are connected to the self-energy Σ(k,²k /~) in the usual way. In
particular, Im Σ(k,²k /~) turns out to be positive for holes, such that eq. 4.38 and eq. 8.32
again lead to exponential decay for quasi-holes. A complete dictionary for translating the
particle/hole perturbation series into diagrams and vice versa can be found in Mattuck,
table 4.2, p.75. I reproduce it in Table 8.4.
A few words on this table. Mattuck uses diagrams where the time increases from
bottom to top, so everything should be read accordingly (my diagrams read from left to
right). Also he uses units in which ~ = 1, so every i should be substituted by i~ (and
every −i = 1i by i~ 1
). The diagrams (a)-(d) show all possibilities of what can happen
at an interaction. We have covered the possibilites (a) and (d) in the previous and the
present section; they correspond to particle-particle and hole-hole scattering. The other
two possibilities (c) and (d) correspond to particle-hole pair creation and annihilation,
respectively. They are obviously not active for a single particle, but they are of interest in
the many-particle case, see the next section. Following the dictionary of Table 8.4 takes
care of the “administration” of creation/annihilation operators we discussed in Section 7.3.
This table holds for electrons and holes scattered by fixed potentials. For electrons and
holes which have a mutual interaction or which interact with photons, we need a different
table. This will be discussed later on.
5
Feynman states that, if you follow the arrow, “the hole moves backwards in time”, since t1 ≤ t0 . This
sounds very “star trek” like, but in my opinion it is not very helpful. I prefer to follow the time labels and
not the arrows.
8.3. MANY PARTICLES AND HOLES 211

Figure 8.4: Mattuck’s table 4.2, p.75

8.3 Many Particles and Holes


The next step up in complexity is a many-particle system of non-interacting particles (and
holes) as in Section 6.3.1. As one might expect, the homogeneous electron gas is our toy
model for that situation. The Hamiltonian according to eq. 6.51 is given by

X X
b
h = E0 − ²kbb†kbbk + a†kb
²kb ak (8.33)
|k|≤kF |k|>kF

It is possible to study scattering in such a system by introducing atomic scattering poten-


tials. These give a perturbation of the type we have seen before
X
Vb = c†k b
Vkl b cl
k,l
X X
= a†kb
Vkl b al + a†kbb†l
Vkl b
|k|>kF ,|l|>kF |k|>kF ,|l|≤kF
X X
+ Vklbbkb
al + Vklbbkbb†l (8.34)
|k|≤kF ,|l|>kF |k|≤kF ,|l|≤kF

where b c†k are general fermion operators, which can be substituted by particle/hole
ck and b
operators according to the familiar rules. Again one is interested in the result of the
212 CHAPTER 8. PROPAGATORS AND DIAGRAMS

gedanken experiment of Section 8.1.1, for which one has to calculate the propagators of
Section 8.1.2. Of course one can do the perturbation expansion and dilligently calculate
the matrix elements of the resulting strings of creation and annihilation operators. The
foregoing sections, especially Section 8.2.2, let you suspect that a short-cut is possible
in order to obtain an expression for the propagator. Simply draw all possible diagrams
and use Table 8.4 to transform these into expressions which only contain unperturbed
propagators G± 0 and matrix elements Vkl . These expressions can then be evaluated in
the time domain, using eqs. 8.14 and 8.27, or, what is easier in general, in the frequency
domain, using eqs. 8.15 and 8.28.

PROBLEM
This short-cut works fine, and it is absolutely correct, but not for reasons that are imme-
diately obvious. Let me explain what the problem is. The electron propagator was defined
as
D ¯ ¯ E
¯ ¯
i~G+ (l, k, t1 − t0 ) = 0 ¯b a†k (t0 )¯ 0 ;
al (t1 ) b t1 > t0 (8.35)

The state |0i is the vacuum state of the full Hamiltonian H b =b h + Vb , and not the vacuum
state of the unperturbed Hamiltonian b h. The central idea behind the propagator is that
one takes a system in its ground state and then adds an extra particle, cf. Fig. 8.1. For a
single particle this distinction between Hb and bh is irrelevant since Vb |0i = 0, cf. eq. 8.34.
“If there ain’t any particle present, the scattering potential has no effect”.6 However this
is no longer true in the many-particle case, the presence of scattering potentials already
changes the properties of the electron gas, without the external particle introduced by b a†k
in eq. 8.35. The ground state |00 i of the unperturbed Hamiltonian b h is trivial to write
down in the particle-hole formalism since, by definition, it contains neither particles nor
holes

b
h|00 i = E0 |00 i (8.36)

as can be checked by applying eq. 8.33. Unfortunately, this state is not an eigenstate of
the full Hamiltonian.
 
³ ´ X X
b
h + Vb |00 i = E0 + Vkk  |00 i + Vkl |1k ; 1l i (8.37)
|k|≤kF |k|>kF ,|l|≤kF

as can be checked by applying eq. 8.34. The last term in eq. 8.37 mixes in electron-
hole excited states in which we have one electron above, and a hole below the Fermi
level. To construct the propagator of eq. 8.35 we do need the ground state |0i of the full
Hamiltonian. Since we do not appear know this state, this presents a serious problem. In
order to calculate it we could represent the full Hamiltonian on a basis set of states of the
unperturbed Hamiltonian and use relations like eq. 8.37 to determine all matrix elements.
The resulting Hamilton matrix can then be diagonalized to determine its eigenstates. This
route is doable, but is not trivial.7
6
sounds like an expression from J. Cruijff: “om te voetballen heppie een bal nodig”.
7
This is what is usually done in a so-called band structure calculation.
8.3. MANY PARTICLES AND HOLES 213

We proceed using a different route, which is also far from easy, but it is more general
since it applies to all kinds of many interacting particle systems. The trick to construct
the ground state |0i of the full Hamiltonian is to start with the unperturbed Hamiltonian
b
h and its ground state |00 i and switch on Vb as a perturbation adiabatically slow. If we
start at a time t = −∞, and use a perturbation of the form Vb eαt with α a very small
positive number, will have switched on the full Vb at t = 0. In one of the exercises we have
already shown that such an adiabatically slow switching process brings our system from
the ground state of the unperturbed Hamiltonian to the ground state of the perturbed
Hamiltonian. We can write
|0i = N U b (0, −∞)|00 i
b † (0, −∞)U (8.38)
0

where N is some normalization constant. U b is the time evolution operator belonging to


b
the full Hamiltonian, including V e −α|t| , so it incorporates the adiabatic switching on of
the perturbation.8 A partial proof of eq. 8.38 was given in one of the exercises, the general
proof is due to Gell-Mann72 and Low. Defining U bI (0, −∞) = Ub † (0, −∞)U
b (0, −∞) in the
0
“interaction picture” (see Section 2.2, eq. 2.16), determining the normalization constant
and using the properties of the time-evolution operator, one can rewrite eq. 8.35 as
bI (∞, t1 )b
h00 |U aI,k (t1 )UbI (t1 , t0 )b bI (t0 , −∞)|00 i
a†I,l (t0 )U
+
i~G (k, l, t1 − t0 ) = (8.39)
h00 |U bI (∞, −∞)|00 i
where the operators in the interaction picture are
b b † (t, 0) b
aI,k (t) = U b0 (t, 0)
ak U
0
a† (t) = U
b b † (t, 0) b b0 (t, 0)
a† U (8.40)
I,k 0 k

see also eq. 2.17 in Section 2.2. We have made the adiabatic switching symmetric here;
that is, we switch off the perturbation again going from t = 0 to t = ∞ by using Vb e−αt
with α a very small positive number. The details of the derivation of eqs. 8.38—8.40 are
given in the Appendix I. The expression of eq. 8.39 now starts from the ground state
|00 i of the unperturbed Hamiltonian, which is good since we know this state exactly.
The prize we have to pay is that the expression gets a lot more complicated, since all
these evolution operators UbI now appear.9 The idea is that we expand all these evolution
operators as a perturbation series, as in eqs. 2.19 and 2.20 of Section 2.2. It looks
very complicated, because we have several of such UbI terms. Moreover, we have to do a
perturbation expansion in both the numerator and the denominator of the expression of
eq. 8.39.

IMPORTANT MESSAGE
However we are saved by Feynman. Despite the fact that the expression of eq. 8.39 looks
far more complicated than the simple expression of eq. 8.35, we are still allowed to use the
8b b † is included
U0 is the time evolution operator belonging to the unperturbed Hamiltonian; the factor U0
to get rid of a time-dependent phase factor; see one of the exercises.
9
This is a result of the so-called “Free Lunch” theorem, which states: “...There is no such thing as
a free lunch...”. In Dutch this comes close to the “Wet van behoud van ellende”, which is an optimistic
view of the world. As a thermodynamic quantity, “ellende” is closer to entropy than to energy. Only in
adiabatically slow “reversible” processes “ellende” is conserved, otherwise it increases.
214 CHAPTER 8. PROPAGATORS AND DIAGRAMS

same procedure as in the single particle case! That is, draw all possible diagrams as if we
were in the single particle case, and use Table 8.4 to transform these into expressions which
only contain unperturbed propagators G± 0 and matrix elements Vkl . Feynman devised
this procedure on intuitive grounds, but it can be proven that this is the exact result
of the perturbation expansion of the full expression of eq. 8.39! The proof however
is not easy, so I refrain from presenting it here and assume that it is all-right to follow
Feynman’s intuition. A partial proof for the example given in the next section is presented
in Appendix II. As it turns out, there is only one change with respect to the single particle
case. If we only have a single electron (or hole) then all the propagator arrows are going
to the right (or to the left). Of Table 8.4 only the diagrams (a) and (d) participate. In
the many-particle case, we can have particle-hole pair creation or annihilation, induced by
the scattering potential, cf. the second and third terms on the right hand side of eq. 8.34.
This means diagrams (b) and (c) also contribute. The next section gives an example of
how it works.

8.3.1 Atom Embedded in an Electron Gas

We will consider a problem with an Hamiltonian like that of eqs. 8.33 and 8.34. However,
I will simplify it a little bit in order to present a problem that can be solved analytically
and still contains interesting and relevant physics. We take an atom with a single state
of energy ²A > ²F and add it to the electron gas. The atom represents an impurity in a
crystal, for instance. Obviously, the atom’s Hamiltonian in second quantization notation
is
b A = ²Ab
H a†Ab
aA (8.41)

where b a†A , b
aA create/annihilate an electron in the atomic state |Ai. Don’t be alarmed
by the fact that the atom only has one state. The formalism can easily be extended to
include multiple atomic states, but the notation gets more complicated. Often atomic
levels are far apart in energy, such that focussing upon one level at a time is not too bad
an approximation. We embed the atom in the homogeneous electron gas and the total
unperturbed Hamiltonian is
b0 = b
H bA
h+H (8.42)

The atom and the electron gas will of course have an interaction because the atom intro-
duces an extra potential. Since we neglected the (two particle) interactions between the
electrons, the interaction has a one particle form
X X X
Vb = c†k b
Vkl b cl + c†k b
VkA b cA + c†A b
VAk b ck (8.43)
k,l k k

where b c†k are general fermion operators and the matrix elements are given by
ck and b
D ¯ ¯ E Z ½ ¾
¯b ¯ ∗ Ze2
VkA = k ¯Vatom ¯ A = φk (r) − φA (r) d3 r (8.44)
|R − r|

The Coulomb potential from the atomic nucleus Vbatom not only works on the electron in the
atomic state |Ai, but also on the electrons of the gas. The wave functions hr|Ai = φA (r)
8.3. MANY PARTICLES AND HOLES 215

of the single atomic state and hr|ki = φk (r) of a gas electron, certainly overlap and
thus their matrix element is non zero. Note that VkA = VAk ∗ as it should to make the

operator Vb Hermitian. We introduce an extra approximation. Usually the matrix elements


|Vkl | ¿ |VkA | because φA (r) in eq. 8.44 is a function which is localized around the atomic
center at r = R (remember, an atomic function) and φk (r) , φl (r) are extended functions
(free electron functions, plane waves). For simplicity we therefore neglect the term Vkl bc†k b
cl
in eq. 8.43, or, in other words we neglect the fact that the atomic potential can scatter
an electron from state k to l, and focus on the dominant interaction between the atomic
and the free electron states. This is the essential approximation in this model. As stated,
the physics behind it is that of an impurity atom in a metal. The model is called the
Fano-Anderson model. Anderson77 used it to describe an impurity in a solid state.10
Simultaneously, Fano used it to interprete atomic spectra, the mathematics of which is
similar, since one has to describe atomic states embedded in the photon continuum of the
radiation field.

Finally, we write the interaction Vb in terms of particle-hole operators


X ³ ´ X ³ ´
Vb = VkAba†kb
aA + VAkba†Ab
ak + VkAbbkbaA + VAkba†Abb†k (8.45)
|k|>kF |k|≤kF

Note that, since ²A > ²F , we can only create (or annihilate) a particle in the atomic state.
For the states of the electron gas we have to substitute b ck by bak or by bb†k , according to
whether |k| >kF (particles) or |k| ≤kF (holes). Each of the terms in eq. 8.45 has a simple
physical interpretation. The first term describes annihilation of the atomic electron and
creating a gas electron in state |ki; or in other words transferring an electron from the
atom to the gas. The second term then describes the reverse process. The third term
is annihilating the atomic electron and a gas hole; or, in other words, an electron-hole
recombination. The fourth term describes the reverse process, creating an electron in the
atomic state and a hole in the gas, or an electron-hole excitation.

We are now interested in the following electron propagator.

i~G+ (A, A, t2 − t1 ) = h0|b a†A (t1 )|0i


aA (t2 )b (8.46)

where |0i is the vacuum or ground state. At time t1 we create an electron in the atomic
state and at time t2 we want to know the probability amplitude that it is still in the
atomic state. In other words we are considering the decay of the atomic state. As you
know from the exercises, once this problem is solved, the scattering problem can also be
solved in a straightforward way. The cross section for scattering of electrons by the atomic
potential shows a distinct peak, the width of which is determined by the same parameter
that determines the lifetime of the atomic state.

8.3.2 Goldstone Diagrams; Exchange


We will set up the perturbation expansion of this propagator in the spirit of Feynman,
using diagrammatic techniques. We can copy our procedure from Chapter 4. In that case
10
Actually, this is not completely true. Anderson also included a two-particle term in his Hamiltonian.
The model given here is the simplest model of an impurity.
216 CHAPTER 8. PROPAGATORS AND DIAGRAMS

all the single arrows represented free particle propagators G+


0 , which makes sense of one
only has the one particle. In the present case we have to allow for the possibility that
propagation in the intermediate states can also proceed via hole states (which requires
hole propagators G− 0 ). To be more specific we write, working in the frequency domain as
in Chapter 4
X
G+ (A, A, ω) = gn (8.47)
n

where gn represents the term which is n’th order in the perturbation Vb . For instance, the
second order term can then be written in diagram form as

k
A A
VA,k Vk , A
g2 =
A V A,k
k
Vk , A A
+ (8.48)

As usual, in diagrams one assumes a summation over the intermediate label k; the direction
of the arrow now indicates whether we sum over electron states (top) with |k| >kF or over
hole states (bottom) with |k| ≤ kF . The corresponding algebraic expression of these
diagrams is
X
g2 = G+ + +
0 (A, A, ω)VAk G0 (k, k, ω)VkA G0 (A, A, ω)
|k|>kF
X
+ G+ − +
0 (A, A, ω)VAk G0 (k, k, ω)VkA G0 (A, A, ω) (8.49)
|k|≤kF

where G± 0 is given by eqs. 8.15 and 8.28. This is similar to what we found in Chapter 4, eq.
4.23, except that the intermediate propagation can take place via a particle, G+ 0 (k, k, ω),

or a hole, G0 (k, k, ω).

PHYSICAL INTERPRETATION; EXCHANGE


We can Fourier transform eq. 8.49 to the time domain if we want to know its contribution
to the propagator G+ (A, A, t2 − t1 ) of eq. 8.46. The expressions become more compli-
cated because the time integrals reappear (a product in the frequency domain becomes a
convolution in the time domain).
X Z ∞ Z ∞
g2 = dτ 2 dτ 1
|k|>kF −∞ −∞

G+ + +
0 (A, A, t2 − τ 2 )VAk G0 (k, k, τ 2 − τ 1 )VkA G0 (A, A, τ 1 − t1 )
X Z ∞ Z ∞
+ dτ 2 dτ 1
|k|≤kF −∞ −∞

G+ −
0 (A, A, t2 − τ 1 )VAk G0 (k, k, τ 1 − τ 2 )VkA G+
0 (A, A, τ 2 − t1 ) (8.50)
8.3. MANY PARTICLES AND HOLES 217

The correct placement of the time labels in this expression is a bit subbtle (it is correct
however, see Appendix II). It is easier to do it via the diagrams, which stay more or less
the same apart from attaching the time labels. The first diagram of eq. 8.48 becomes

τ1 k τ2
A A
t1 VA,k Vk , A t2
(8.51)

with the time order t1 < τ 1 < τ 2 < t2 and we have to integrate over the intermediate
times τ 1 , τ 2 . Such time-ordered diagrams are also called Goldstone diagrams.11 The
second diagram becomes

τ2
A V A ,k
t1
k
Vk , A A
τ1 t2 (8.52)

The sum of these two diagrams corresponds exactly to eq. 8.50.

Diagrams also have a nice physical interpretation. For instance, one can interprete the
diagram of eq. 8.51 as follows. At time t1 a particle enters the system in state A. At time
τ 1 it is scattered by the potential into the particle state k and at time τ 2 it is scattered
back by the potential to state A, where it leaves the system at time t2 . Obviously the
time order is t1 < τ 1 < τ 2 < t2 . Note that the begin and end labels (A, t1 ) and (A, t2 ) are
fixed, since this is the process we want to consider for the propagator G+ (A, A, t2 − t1 ).
The diagram of of eq. 8.51 is also called a direct diagram in order to distinguish it from
the diagram of eq. 8.52, which is interpreted as follows. A particle enters the system in
state A at time t1 . At time τ 1 a hole in state k is created and also a particle in state
A. At time τ 2 the hole recombines with the incoming particle and the created particle
in state A proceeds to time t2 as the outgoing electron. Note as before that the hole
proceeds from τ 1 to τ 2 , so it is following the arrow backwards. The outgoing particle is
not “the same one” as the incoming one; it has swapped roles with one of the particles of
the electron gas. Such a process is called an exchange and the diagram of eq. 8.52 is called
an exchange diagram. Since we have no way of distinguishing the particles, we can never
tell from the outgoing particles whether exchange has taken place or not. Therefore we
always have to include the possibility of exchange in our calculations. It distinguishes a
many particle system from a single particle one. In the latter one only has direct processes
(represented by direct diagrams), in the former one additionally has exchange processes
(represented by exchange diagrams).

One final note to bring us down to earth again. One should not take the physical
interpretation of the diagrams too literally. In the end they just represent terms in a
perturbation expansion. The diagram of eq. 8.52 has some weird features.

• The time order required by the propagators is t1 < t2 (by assumption), τ 1 < τ 2 ,
τ 1 < t2 and t1 < τ 2 , see also eq. 8.50. There is no relation between t1 and τ 1 , which
11
after the American physicist Jeffrey Goldstone.
218 CHAPTER 8. PROPAGATORS AND DIAGRAMS

means that τ 1 can run from −∞ up to t2 or τ 2 (whatever is smallest). This means


that the creation of the particle-hole pair at τ 1 can take place before the A particle
enters the system at t1 .

• Also there is no relation between t2 and τ 2 , which means that τ 2 can run from τ 1
or t1 (whatever is largest) to ∞. This means that the particle-hole recombination
at τ 2 can take place after the A particle has left the system at t2 .

• Weirdest of all is that for times t, max(t1 , τ 1 ) < t < min(t2 , τ 2 ), there exist two
particles in state A, which seems to contradict the Pauli principle.

The diagrams are correct, since the perturbation terms they represent are correct (see
Appendix II if you don’t believe me). One way out is to say that the diagrams do not
present real physical processes but virtual processes in which virtual particles and holes
are created and annihilated (except at the beginning t1 and the end t2 where we put in
and get out a real particle, respectively). All the virtual processes take place inside the
system. Virtual processes do not have to obey the strict rules of real processes such as
the Pauli principle. The “virtual reality” is however not disconnected from the real world,
since the virtual processes contribute to the propagator G+ (A, A, t2 − t1 ), which, as we
have argued in Section 8.1.1 is tied to real experiments.

8.3.3 Diagram Expansion


We are now in a position to enumerate all diagrams which contribute to the perturbation
expansion of the atomic propagator, eq. 8.47. We work in the frequency domain, because
there the expressions are simpler that in the time domain; compare eqs. 8.49 and 8.50.
The diagrams can be directly translated into algebraic expressions using Table 8.4. The
zero’th order term is the unperturbed propagator G+0 (A, A, ω)

A A
g0 = (8.53)

The first order term gives zero contribution.

VA, A
A A
g1 = =0 (8.54)

Since the begin and end states A are fixed, the only matrix element that can appear at
a†Ab
the dot is VAA . However, there is no term like VAAb aA in our perturbation, cf. 8.43, or
12
to put it differently VAA = 0. A similar reasoning can be applied to all higher order odd
numbered diagrams, which all give zero contribution. For instance, a third order diagram
is
k A
A A
VA,k Vk , A VA, A
g3,1 = =0 (8.55)

Going from left to right, at the first dot we can only go from A to k because only matrix
elements VAk , i.e. between states A and k, are non zero. At the second dot we have to
12 b A of
This makes sense. The “on-site” atomic potential has been included in the atomic Hamiltonian H
eq. 8.41, so obviously it cannot be part of a perturbation.
8.3. MANY PARTICLES AND HOLES 219

go from k to A for the same reason. Finally, at the third dot we end up with 0 because
VAA = 0.

Thus only even numbered diagrams give a contribution. Things are brightening up,
since all even order diagrams can be constructed by linking the second order ones given
in eq. 8.48. An example of a fourth order diagram is given by

k A
A V A, k
V A, k Vk , A
k'
Vk , A A
g4,1= (8.56)

where the intermediate propagation is first via a particle in an intermediate state k. It


is followed by propagation of a particle in A, which recombines with a hole in k0 , and a
second particle in A comes out. As usual, a summation over all intermediate k, k0 states
is implicit. Another fourth order diagram is where all intermediate propagations in k, k0
are via hole lines.

A V A ,k
k
V A,k
Vk , A A
k'
Vk , A A
g4,2= (8.57)

In view of the previous section this diagram represents a double exchange process in which
the incoming particle exchanges twice with a particle in the system. The two remaining
fourth order diagrams are given by

k A k' A
A
g4,3 =
A
k k' A
g4,4 = A (8.58)

Note that the arrow in the middle of the diagrams (i.e. the one between k and k0 ) always
has to carry the label A. This is because only terms of type VAk and VAk occur in the
perturbation, cf. eq. 8.43. The sixth order diagrams can be generated by linking either
of the two diagrams in eq. 8.48 to each of the fourth order diagrams, eqs. 8.56—8.58. The
eighth order diagrams are generated by another link, etcetera, allowing us to construct all
possible diagrams.

8.3.4 Diagram Summation


For this particular model it is not only possible to enumerate all diagrams, but to actually
sum them to infinite order and thus sum the whole perturbation series like we did in
Chapters 3 and 4. We do this by “factorizing” the diagrams. Let us start with the second
220 CHAPTER 8. PROPAGATORS AND DIAGRAMS

order terms of eq. 8.48 and write

A
g2 = ×

k A k
( + A) (8.59)

All the diagrams can be translated directly into G0 ’s and V ’s;Pso the first diagram stands
for G+0 (A, A, ω); cf. eq. 8.53, the second diagram stands for
+
|k|>kF VAk G0 (k, k, ω) VkA
P
G+0 (A, A, ω) and the third diagram for
− +
|k|≤kF VAk G0 (k, k, ω) VkA G0 (A, A, ω), and the
total expression thus correspond to eq. 8.49. The way to transform this symbolic notation
into diagrams again is to let “×” make a connection between its arguments. The “+” is
the usual summation, such that eq. 8.59 produces all the diagrams of 8.48. Using the
same notation all fourth order diagrams can be summed to

A
g4 = g4,1 + g4,2 + g4,3 + g4,4 = ×

k A k
( + A) ×

k' A k'
( + A) (8.60)

Since the k and k0 labels in the last two lines are just dummy labels which are supposed
to sum over all possible states, we may as well write k instead of k0 in the last line, which
gives us

A
g4 = ×

k A k
( + A)2 (8.61)

If you don’t believe this, we can also do the algebra. Consider a fourth order diagram g4,3
again,cf eq. 8.58. It translates into the expression
X
g4,3 = G+ + +
0 (A, A, ω) VAk G0 (k, k, ω) VkA G0 (A, A, ω)
|k|,|k0 |>kF
¡ 0 0 ¢
VAk0 G+ +
0 k , k , ω Vk0 A G0 (A, A, ω) (8.62)

It is clear that this can be factorized as


X
g4,3 = G+ 0 (A, A, ω) VAk G+ +
0 (k, k, ω) VkA G0 (A, A, ω) ×
|k|>kF
X ¡ 0 0 ¢
VAk0 G+ +
0 k , k , ω Vk0 A G0 (A, A, ω) (8.63)
|k0 |>kF

and that the two sums are in fact identical since k and k0 are in fact dummy labels.
8.3. MANY PARTICLES AND HOLES 221

BEWARE OF THE PITFALL

A word of care is at its place here. Such factorizations are not always possible! It is here
because the “arrow in the middle” discussed in the last section has a unique label A; it
corresponds to the G+ 0 (A, A, ω) propagator at the end of the first line of eq. 8.62. If it
were another label over which one had to sum; let’s say l, then the resulting expression
X
G+ + +
0 (A, A, ω) VAk G0 (k, k, ω) Vkl G0 (l, l, ω)
|k|,|k0 |,|l|>kF
¡ 0 0 ¢
Vlk0 G+ +
0 k , k , ω Vk0 A G0 (A, A, ω) (8.64)

would not be factorizable. We have chosen our model such that the perturbation does
not contain any Vkl terms, cf. eq. 8.45, in order to make expressions like the one above
factorizable. A more general perturbation of the type of eq. 8.43 is much harder to do
(although not impossible).

RESUME MAIN TEXT

In our case all the higher order diagrams are constructed by linking second order diagrams,
which means that they are factorizable just like the fourth order one. A general expression
for an even order diagram g2n , analogous to eq. 8.61, is given by

A
g2n = ×

k A k
( + A)n (8.65)

This makes it easy to find a final expression of the propagator of eq. 8.47

X
G+ (A, A, ω) = g2n or
n=0
A, ω
A
= × [1 +

∞ k A k
X
( + A)n ]
n=1

k A k
A A]−1 (8.66)
= × [1 − −

The transition to the last line is possible, since in the end the diagrams
P are just numbers
and between [...] we have simply a geometric series for which [1 + n rn ] = [1 − r]−1 . The
A
final expression can be simplified even further by noting that the final arrow
in the terms between [...] represents the number G+ 0 (A, A, ω) which also can be put outside
222 CHAPTER 8. PROPAGATORS AND DIAGRAMS

the brackets. One obtains

A, ω −1 k k
A
=[ − − ]−1

It is possible to translate this all back into algebraic expressions again and write
−1
X X
G+ (A, A, ω) = [G+ 0 (A, A, ω) − VAk G+0 (k, k, ω) VkA − VAk G−
0 (k, k, ω) VkA ]
−1

|k|>kF |k|≤kF
X |VAk |2 X |VAk |2
= [~ω − ²A + iδ − − ]−1 (8.67)
~ω − ²k + iδ ~ω − ²k − iδ
|k|>kF |k|≤kF

using eqs. 3.13 and 3.21.

8.3.5 Exponential Decay


Now sit back and compare the final expression of eq. 8.67 with the expression of eq. 4.27
we found in Chapter 4. We have a self-energy again ! It can now be defined as
X |VAk |2 X |VAk |2
Σ(A, ω) = + (8.68)
~ω − ²k + iδ ~ω − ²k − iδ
|k|>kF |k|≤kF

which is very much like that of eqs. 4.25 and 4.30. In comparison with those equations,
we see that Σ(A, ω) lacks the zeroth order V (0) and the higher order terms in V . This is
because we have chosen VAA = Vkk0 = 0 in our model; the expression of eq. 8.68 is exact.
Σ(A, ω) now contains a hole part (the second term) as well as a particle part (the first
term).
We can go to the continuum limit analogous to eq. 4.28 by defining
Z
1 1
VkA = √ e−ik·r Vatom (r)φA (r) d3 r= √ VA (k) (8.69)
Ω Ω Ω
where VA (k) is the Fourier transform of Vatom (r)φA (r). An analysis similar to that of
Section 4.2 then gives for the imaginary part of the self-energy
Z |k|>kF
1
Im Σ (A, ω) = −π d3 k |VA (k)|2 δ(~ω − ²k )
(2π)3
Z |k|≤kF
1
+π d3 k |VA (k)|2 δ(~ω − ²k ) (8.70)
(2π)3
Note the restrictions on the integral now; the first term only gives a contribution if ~ω > ²F
and the second term only gives a contribution if ~ω≤²F .13 The real part of the self-energy
is given by
Z 2
1 3 |VA (k)|
Re Σ (A, ω) = Pv d k (8.71)
(2π)3 ~ω − ²k
13
Im Σ(A, ω) is seen to change sign at ~ω=²F . This is caused by the ±iδ in eq. 8.69. This sign change
ensures that, when transforming back to the time domain, holes will also decay exponentially with a decay
time related to Im Σ(A, ω), as do particles.
8.4. INTERACTING PARTICLES AND HOLES 223

where Pv indicates a principal value integral. Having defined the self-energy, the propa-
gator of eq. 8.67 now simply becomes

1
G+ (A, A, ω) = (8.72)
~ω − ²A − Σ(A, ω)

where we got rid of the +iδ because the self-energy is complex anyhow. This equation
is similar to eq. 4.27. We now follow Section 4.4 by the letter. Fourier transforming the
result back to the time domain, we obtain
D ¯ ¯ E
¯ ¯
i~G+ (A, A, t2 − t1 ) = 0 ¯b a†A (t1 )¯ 0
aA (t2 )b
i 1
= zA e− ~ [²A +∆A ](t2 −t1 ) e− 2 ΓA (t2 −t1 ) for t2 > t1 (8.73)

with
µ ¶−1
1 ∂ Re Σ (A, ω)
zA = 1− |ω=²A /~
~ ∂ω
∆A = Re Σ (A, ²A /~)
2zA
ΓA = − Im Σ (A, ²A /~) (8.74)
~
So after a long story we are back to a situation that can be described in terms of a level
shift ∆A and a lifetime τ A = Γ−1A , both of which can be derived from the self-energy. The
weight factor zA is in general not equal to 1 anymore, like it was in the single particle
case. One can prove that 0 ≤ zA ≤ 1; the stronger the scattering, the smaller it is;
usually it is much closer to 1 however. The appearence of a weight factor is another thing
that distinguishes the many particle from the single particle case. Not surprisingly, eq.
8.73 is said to describe the time evolution of a quasi-particle. It has much more of a
“quasi” character than in Chapter 4, because we have incorporated all possible exchange
processes with particles in the electron gas, as explained in Section 8.3.2. So although we
are never sure which particular particle we are looking at, we still can “quasi” do as if we
are handling a single particle, and we call it a quasi-particle. Note that because the energy
²A > ²F by assumption, it is only the first term of eq. 8.70 that gives a contribution. If
²A ≤ ²F , we expect the atomic state to be occupied in the D ¯ground state. ¯ EIn the latter
¯b b† ¯
situation it is appropriate to consider the hole propagator 0 ¯bA (t2 )bA (t1 )¯ 0 . The result
will be similar to eq. 8.73 and describes the time evolution of a quasi-hole.

8.4 Interacting Particles and Holes


Up till now, the many-particle system we have focused on, has been the homogeneous
electron gas comprising non-interacting electrons. In a “real” material, matters are more
complicated.

The inhomogeneous electron gas


“Real” atoms add potentials which, due to their spatial dependence, break the homogene-
ity of the electron gas. We have seen an example in the previous section. Such potentials
224 CHAPTER 8. PROPAGATORS AND DIAGRAMS

have the form


Ze2
V (R − r) = − (8.75)
|R − r|

where R is the position of an atomic nucleus and Z is its charge. In general one has a
number of such potentials (in condensed matter a very large number)
X
V0 (r) = V (R − r) (8.76)
R

One expects the density of electrons (i.e. the density of the electron “gas”) to be higher
close to an attractive Coulomb potential; the electron density becomes spatially depen-
dent, or inhomogeneous. The good news is that such atomic potentials can be treated quite
accurately, although this usually requires a numerical calculation on the computer. Over-
stating it slightly one might say that a one-particle potential, which in second quantization
form reads
X
Vb0 = c†k b
Vkl b cl (8.77)
k,l

can be treated exactly, or at least very accurately. In practice, it is only the limited size
of computational resources that restricts the number of atoms (per unit cell) we can treat.
You have encountered the states of such an inhomogeneous electron gas before, in your
solid state physics course for instance. If the potential of eq. 8.77 has a periodicity, then,
instead of the simple plane waves of eqs. 6.40 and 6.41, we now have so-called Bloch states
|kni, which are eigenstates of
µ 2 ¶
b b
p b
H0 |kni = + V0 |kni = ²kn |kni (8.78)
2me

where n is the band index. The energies ²kn form the so-called “band structure” and
solving eq. 8.78 is called a band structure calculation. In the position representation we
can write a Bloch state as

hr|kni ≡ φkn (r) = eik·r ukn (r) (8.79)

where ukn (r + a) = ukn (r) is a periodic function with a a vector of the periodic lattice.
Everything we have done with the “free” electrons of the homogeneous electron gas with
states |ki and energies ²k we can repeat for the Bloch electrons of the inhomogeneous
electron gas with states |kni and energies ²kn . In conclusion, the inhomogeneous electron
gas might technically be a lot more complicated to handle than the homogeneous electron
gas, but conceptually there are no major qualitative differences.

The interacting electron gas


“Real” electrons interact with each other. A Coulomb repulsion

e2
V (r1 − r2 ) = (8.80)
|r1 − r2 |
8.4. INTERACTING PARTICLES AND HOLES 225

exists between each pair of electrons. As we have seen in previous chapters, in second
quantization such a potential is written as

1 X
Vb = c†l b
Vklmn b c†k b
cm b
cn (8.81)
2
k,l,m,n

Such two-particle interaction terms define the interacting electron gas. Some authors also
use the phrase electron liquid.14 The bad news is that today almost no physical problem
with a realistic two-particle interaction can be treated exactly.15 We have to stumble along
and find workable approximations for the interacting electron gas. Fortunately, in many
cases we can find such approximations. One of the must fruitful techniques in finding
them has been perturbation theory.

8.4.1 Two-Particle Diagrams


We have derived a diagrammatic technique from the mathematical expressions of the prop-
agator and its Dyson expansion for particles without mutual interactions. The interacting
electron gas is mathematically much much more complicated. Following Feynman, we
postulate a diagrammatic technique as a fairly straightforward generalization of the non-
interacting case. Some ideas for how to go about proving that this is a correct technique
for the interacting case are discussed in the appendices and in the references therein.

As we saw in the previous section, if we want to incorporate a one-particle operator


like in eq. 8.77 in a Dyson expansion, we have to connect propagators G0 and matrix
elements Vkl . For instance, in the frequency domain

l k ≡ G+ +
0 (k, k, ω)Vkl G0 (l, l, ω) (8.82)

Of course |k| , |l| > kF , since we are dealing with electrons. Note that we have not
explicitly written Vkl at the dot in the diagram; from now on we write diagrams as concise
as possible. Another possibility would be an electron in state l and a hole in state k
coming in, and recombining at the dot; of course |k| ≤ kF ; |l| > kF .

k ≡ G− +
0 (k, k, ω)Vkl G0 (l, l, ω) (8.83)

Looking at the diagrams of eqs. eq. 8.82 and eq. 8.83 and comparing the expressions of
eqs. 8.77 and 8.81, it will be clear that for a two particle interaction, the diagram needs
14
The interaction between electrons is repulsive. This means that there is no gas/liquid phase transition
in the system as a function of the electron density and/or the temperature. Since one cannot distinguish
between a gas and a liquid phase, the words “gas” and “liquid” are used interchangeably. Unfortunately, the
common practice is inconsistent. If the two-particle interaction is attractive, one can have a condensation-
like phase transition in a quantum system; the occurence of superconductivity at low temperature is a
well-known example of this.
15
Which is one of the reasons why there is still no convincing theory of high-Tc superconductivity. Solve
it, and you can book your trip to Stockholm.
226 CHAPTER 8. PROPAGATORS AND DIAGRAMS

two incoming and two outgoing particles. An example of such a diagram is


m k

n l (8.84)
−1
≡ (i~) 2
G+ +
0 (k, k, ω k )G0 (l, l, ω l ) (i~) Vklmn (i~)2 G+ +
0 (m, m, ω m )G0 (n, n, ω n )

Two electrons come in states m and n, they interact, get annihilated and two electrons
come out in states k and l; |k| , |l| , |m| , |n| > kF . Unfortunately we have to do the book-
keeping for the (i~) factors in detail now; we will talk about the frequency factors later on.
This diagram has been derived from the electron term Vklmnb a†l b
a†kb
amb
an of the interaction.
As before, there are also hole variants of such diagrams; for instance
m

n l (8.85)
−1
≡ (i~) 2
G− +
0 (k, k, ω k )G0 (l, l, ω l ) (i~) Vklmn (i~)2 G+ +
0 (m, m, ω m )G0 (n, n, ω n )

|l| , |m| , |n| > kF ; |k| ≤ kF . Two electrons come in states m en n, one hole comes in in
state k (remember holes move “backwards”, i.e. against the arrows), all get annihilated
and an electron in state l comes out. It is derived from the Vklmn b a†l bbkb
amb
an term in the
interaction. In principle all sorts of combinations of electrons and holes are allowed in
such a diagram, up to
k m

l n (8.86)
−1
≡ (i~) 2
G− −
0 (k, k, ω k )G0 (l, l, ω l ) (i~) Vklmn (i~)2 G− −
0 (m, m, ω m )G0 (n, n, ω n )

derived from Vklmnbblbbkbb†mbb†n . Note that whatever way the diagram is written, the dotted
line represents the interaction matrix element Vklmn , and the labeled arrows are connected
in the order
Vklmn from left to right:
top out, bottom out, top in, bottom in (8.87)
Mattuck writes his diagrams lying down, and the arrows for the particles go from bottom
to top (see his § 4.5). I write the particle arrows from left to right (as do other authors).
8.4. INTERACTING PARTICLES AND HOLES 227

8.4.2 The Homogeneous Electron Gas Revisited


For the interacting homogeneous electron gas there are no atomic potentials (eq. 8.75),
but the electron-electron interaction (eq. 8.80) is present. The perturbation expansion
and the associated diagrams become simpler. This is mainly because the two particle
matrix elements Vklmn are simpler, since the single particle basis states are plane waves.
Z Z
e2
Vklmn = d3 r1 d3 r2 φ∗k (r1 )φ∗l (r2 ) φ (r1 )φn (r2 )
|r1 − r2 | m
Z Z
e2 1
= 2
d3 r1 d3 r2 ei(m−k)·r1 ei(n−l)·r2 (8.88)
Ω |r1 − r2 |

(see Section 6.3.1, in particular eq. 6.41). The integral can be done by defining new
integration variables r = r1 − r2 and R = 12 (r1 + r2 ). The result is

1
Vklmn = vk−m δ k+l,m+n

Z
2 1 4πe2
where vq = e d3 r e−iq·r = 2 (8.89)
r q
2
is the (three-dimensional) Fourier transform of the Coulomb potential er . For a more
detailed derivation, see Mattuck p. 135.16 The δ k+l,m+n can be interpreted like a “con-
servation of momentum” selection rule. It can be incorporated in the diagrams. Defining
q = k − m means we can write m = k − q and, because of the Kronecker δ, n = l + q, so
the non-zero matrix elements are of the form
1
Vklmn = Vk,l,k−q,l+q = vq (8.90)

We can relabel the generic diagram of eq. 8.84 as

m = k −q k

n = l+q l
(8.91)

Since in the homogeneous electron gas the one-particle states are plane waves, the labels
k, l etcetera correspond to the momenta of the particles. The diagram then tells us that the
sum of the momenta of the incoming particles m + n is equal to the sum of the momenta of
the outgoing particles k + l. The diagram thus expresses the conservation of momentum.
Adding the arrow labeled q explicitly gives the momentum which is transferred from the
lower to the upper particle. The diagram can be interpreted in an intuitive way as a
“collision” between two particles where momentum is transferred from one to the other
particle. In the spirit of the previous chapter, one can apply “circuit” rules to this diagram.
At each node the sum of the incoming arrows (read: momentum) has to be equal to the
16
Note that, compared to Mattuck, I have surpressed the spin variables σ1 etcetera. In my case k
implicitly includes the spin variable. It becomes (k, σ 1 ) in Mattuck’s notation.
228 CHAPTER 8. PROPAGATORS AND DIAGRAMS

P
sum of the outgoing arrows (momentum). So at each node momenta = 0, where the
arrows give the direction of momentum flow. This only works for the homogeneous electron
gas because in an inhomogeneous gas the momentum of the electrons is not a conserved
quantity. In an inhomogeneous gas the atomic nuclei (which give the atomic potentials
that cause the inhomogeneities) can absorb part of the electronic momentum in a collisions
between electrons and nuclei.

8.4.3 The Full Diagram Dictionary


Again in the spirit of the previous chapter, one can also attach frequency (or energy) labels
to the diagram of eq. 8.91. Assuming that, like in the previous chapter, the circuit rules
also hold for these frequencies (energies), one then obtains

m = k −q k ,ω k
ωm = ωk − ωq
q,ωq
n = l+q
ωn = ωl + ωq l,ω l
(8.92)

Starting from the diagram representation in the time domain, one can indeed prove by
Fourier transformation to the frequency domain that the circuit rule for the frequency
holds. The conservation of frequency also holds for an inhomogeneous electron gas, since
collisions with the nuclei are elastic.The latter are treated as fixed scattering centers which
do not absorb energy and thus the electronic energy is a conserved quantity. This conser-
vation rule can be used to relate the frequencies in the expressions of eqs. 8.84—8.86.

A full perturbation expansion in terms of two-particle diagrams can lead to pretty


complicated diagram topologies, see e.g. Fig. 7.9 in Section 7.3. Obviously, of diagrams
like those of eqs. 8.84—8.86 there exist a number of particle-hole variants which can be
connected in various ways. Feynman’s rules for transforming diagrams into algebraic
expressions are similar to those of Table 8.4. For two-particle interactions there are a
couple of extra sign rules, which have to do with so-called “fermion loops”. As always,
these are the result of the anti-symmetry of fermion states; if you want to know the details,
look at the advanced books listed in the appendices. Mattuck gives a complete dictionary
for two-particle diagrams in his table 4.3, p. 86. I reproduce this table below.
The same remarks which were made referring to Table 8.4 also apply here. Table 8.5
allows you to write down any perturbation term in the form of Feynman diagrams and
derive an algebraic expression for it in order to evaluate it quantitatively.
8.4. INTERACTING PARTICLES AND HOLES 229

Figure 8.5: Mattuck’s table 4.3, p.86.

8.4.4 Radiation Diagrams


The diagram of eq. 8.84 (or its refined versions, eqs. 8.91 and eq. 8.92) resembles a
diagram we might have considered in the previous chapter on the interaction between
electrons and photons.

m τ1 k

τ 2 > τ1

n τ2 l (8.93)

Two electrons come in from the left and two go out at the right, so this is a diagram that
plays a role in electron-electron scattering. At time τ 1 a photon is emitted by an electron
in state m, transferring the electron to state k. At time τ 2 the photon is absorbed by an
electron in state n, transferring it to state l. As we have seen in one of the exercises, such
a transfer of a photon from one electron to another can be interpreted as an effective two-
particle potential W (r1 − r2 ) operating between the two electrons (with matrix elements
230 CHAPTER 8. PROPAGATORS AND DIAGRAMS

Wklmn in the notation of eq. 8.84). Thus the resulting physics is quite similar to that of the
two-particle Coulomb interaction V (r1 − r2 ) of eqs. 8.80 and 8.84 and the diagrammatic
dictionary is also similar to that of Table 8.5. However, there are also obvious differences.
In eq. 8.93 the photon has a (large, but finite) speed c, which means that τ 2 > τ 1 . The
effective two-particle potential resulting from photon transfer thus depends upon the time
difference τ 2 − τ 1 , i.e. Wklmn (τ 2 − τ 1 ). If we Fourier transform it, it acquires a frequency
dependence Wklmn (ω). In contrast, the Coulomb interaction is instantaneous; it operates
on both particles at the same point τ in time, Vklmn (τ ) = Vklmn δ(τ ). This is why the
dashed line in eqs. 8.84—8.86, which represents the Coulomb interaction, is written as a
vertical line. If we Fourier transform it, it is independent of frequency, Vklmn .
Looking back at the previous chapter, in particular Appendix I, we observe that the
Coulomb interaction V (r1 − r2 ) is part of the scalar potential φ of the electro-magnetic
field, cf. eq. 7.69. The “radiation” potential W (r1 − r2 ) extracted from eq. 8.93 results
from the vector potential A of the electro-magnetic field, cf. eq. 7.70. In “condensed
matter” we are dealing with electrons at faily high densities. Under such circumstances,
the Coulomb terms are always much larger than the radiation terms, and processes such
as shown in eq. 8.93 can be completely neglected. Only in free space, if the electrons
are far apart, the radiation terms might become important. Because of their explicit
time dependence, they give rise to retardation effects, which are familiar from classical
electro-dynamics. In relativistic dynamics there is no such thing as a purely instantaneous
Coulomb interaction as you know. So at relativistic speeds, both Coulomb and radiation
terms have to be included, which forms the most general framework of quantum electro-
dynamics.17 In “condensed matter” circumstances we can separate the two to a very
good approximation. Only the Coulomb terms need to be used to construct the electronic
states; the radiation terms come merely into play if we are explicitly interested in optical
phenomena.

8.5 The Spectral Function


In the next two sections we are making an effort to connect the gedanken experiment of
Section 8.1.1 to a real experiment. The first step is to define a quantity called the spectral
function. We start from the electron propagator

i~G+ (k, k,t2 − t1 ) = hN, 0|b a†k (t1 )|N, 0i Θ(t2 − t1 )


ak (t2 )b (8.94)

where we made it explicit that t2 > t1 . The ground state or vacuum is represented by
|N, 0i, where the N is added to make it explicit that we are dealing with a system of N
electrons. Of course the propagator represents the probability amplitude that, when an
electron is created at time t1 in state k, it will be found in state k at a later time t2 .
We are going to rewrite this expression. The first step is to insert a resolution of identity
between the operators b a†k (t1 ). Since the latter operator adds one particle, it is
ak (t2 ) and b
17
In relativistic electro-dynamics the vector and scalar potentials A and φ form a 4-vector. So both are
needed to be able to apply the Lorentz transformation.
A wonderful exposé of the ideas of quantum electro-dynamics in layman’s terms (without any equations)
is given by Richard Feynman in his book: QED, the strange theory of light and matter (Penguin, 1990,
ISBN 0-14-012505-1); a definite “aanrader”, certainly for a physicist! If you haven’t got it yet, ask it from
“Sinterklaas”!
8.5. THE SPECTRAL FUNCTION 231

clear that the resolution of identity must contain all the states of an N + 1 particle system.
Let’s simply number them |N + 1, ni; n = 0, 1, .... So
X
I= |N + 1, nihN + 1, n| (8.95)
n

All other |N 0 , mihN 0 , m| with N 0 6= N + 1 inserted in eq. 8.94 give zero contribution
because they contain the wrong number of particles. At the same time we can insert
b b † (t, 0)b
ak (t) = U b (t, 0) (which was its definition), and get
ak U
X
i~G+ (k, k,t2 − t1 ) = b † (t2 , 0)b
hN, 0|U b (t2 , 0)|N + 1, ni
ak U
n
b † (t1 , 0)b
hN + 1, n|U b (t1 , 0)|N, 0i Θ(t2 − t1 )
a†k U (8.96)

b (t, 0) = e− ~i tHb for the time


We choose t1 = 0 and t2 = t to simplify our notation, write U
evolution operator, and use
N0
b 0 , mi = Em
H|N |N 0 , mi (8.97)
0
i.e. the states |N 0 , mi are eigenstates of the full Hamiltonian; Em
N is the energy of the
0
m’th state of an N particle system. It is easy to see that eq. 8.96 now transforms into
X i N +1 N ¯¯³ † ´ ¯¯2
i~G+ (k, k,t) = e− ~ t(En −E0 ) ¯ b
ak ¯ Θ(t) where
n0
n
³ ´
a†k
b a†k |N, 0i
= hN + 1, n|b (8.98)
n0

It is custom to rewrite
³ ´ ³ ´
EnN+1 − E0N = EnN+1 − E0N+1 + E0N+1 − E0N (8.99)

because for a large system

E0N+1 − E0N ≡ µN+1 = µ (8.100)

becomes independent of N . The quantity µ is the (lowest possible) energy that is added
to a system when one particle is added. When the number of particles becomes very large
(the so-called “thermodynamic limit”) µ becomes independent of the number of particles;
in thermodynamics it is called the chemical potential.18 For metals at a temperature
T = 0 the chemical potential µ is identical to the Fermi energy ²F .19 We also write

EnN+1 − E0N+1 ≡ ²N+1


n0 (8.101)

These are the excitation energies of an N + 1 particle system. By the same reasoning
as above, for a very large system the excitations of an N + 1 particle system will not
be different from those of an N particle system. The excitation energies will also be
18
Consider a crystal, where the number of particles N > 1023 , typically. Whether we add one extra
electron to 1023 electrons or to 1023 + 1 electrons will not make to much difference. This adds the same
amount of energy µ to the system.
19
In principle the formalism can be extended to a finite temperature, where the term chemical potential
is perhaps more familiar.
232 CHAPTER 8. PROPAGATORS AND DIAGRAMS

independent of the number of particles in the thermodynamic limit. Using eqs. 8.100 and
8.101 in eq. 8.98 yields
X i N +1 ¯³ ´ ¯2
− ~ t(²n0 +µ) ¯ ¯
+
i~G (k, k,t) = e a†k
¯ b ¯ Θ(t) (8.102)
n0
n
The spectral function is defined by
X ¯¯³ † ´ ¯¯2
A+ (k,ω) = ¯ b
ak ¯ δ(~ω−²N+1
n0 ) (8.103)
n0
n
It is also called the spectral density or the spectral density function. Obviously the spectral
function is positive definite, i.e. A+ (k,ω) ≥ 0. We can rewrite eq. 8.102 as
Z ∞
i
i~G+ (k, k,t) = dω e− ~ t(ω+µ) A+ (k,ω)Θ(t) (8.104)
−∞
A similar rewriting can be done for the hole propagator
−i~G− (k, k,t) = hN, 0|bbk (0)bb†k (t)|N, 0i Θ(−t)
Z ∞
i
= dω e ~ t(ω−µ) A− (k,ω)Θ(−t) (8.105)
−∞
with a spectral density for holes defined as
X ¯¯³ † ´ ¯¯2
A− (k,ω) = ¯ bbk ¯ δ(~ω−²N −1
n0 ) (8.106)
n0
n

where we have used E0N − E0N−1 ≡ µN = µ analogous to eq. 8.101 and ²N n0


−1
= EnN −1 −
N−1
E0 are the excitation energies of the N − 1 particle system. “−” signs have been placed
at the relevant places in order to keep everything consistent for holes. Obviously also
A− (k,ω) ≥ 0. The spectral function can be used to investigate the analytical mathematical
properties of propagators. That however is best left to the mathematicians; we are more
concerned with its physical role.

8.5.1 Physical Content


The spectral function plays an important physical role because it is connected to line
shapes in spectra (see also the next section). In order to reproduce the particle and hole
propagators of free, unperturbed particles of eqs. 8.14 and eq. 8.27 one has to use

0 (k,ω) =δ(~ω ∓ (²k − µ)) (8.107)
in eqs. 8.104 and 8.105. The spectral function thus gives a δ-peak at an energy relative
to the Fermi level, i.e. at ²k − µ for ²k > µ (particles) and at µ − ²k for ²k < µ (holes).
In order to reproduce the quasi-particle (or -hole) forms of the propagators, cf. eqs. 8.17,
8.32 and 8.73, the spectral functions to be used in eqs. 8.104 and 8.105 are
zk Γk /2
A± (k,ω) = (8.108)
π [~ω ∓ (²k + ∆k − µ)]2 + (Γk /2)2
So compared to the free particles (holes), the spectral functions for quasi-particles (-holes)
are transformed from δ-peaks into Lorentzian line shapes; see e.g. Fig. 4.14. The peak
maximum is given by ²k +∆k −µ, for ²k +∆k > µ (particles) and µ−²k −∆k for ²k +∆k < µ
(holes), hence the name “level shift” for ∆k , if one compares to the unperturbed case. In
both cases the width of the peak is determined by Γk .
8.6. (INVERSE) PHOTOEMISSION AND QUASI-PARTICLES 233

=ω εq − µ

µ = εF
µ − εk

Figure 8.6: Photoemission: incoming photon of energy ~ω and outgoing electron of energy
²q .

8.6 (Inverse) Photoemission and Quasi-particles


It is the purpose of this section to establish a closer connection between the gedanken
experiment represented by a propagator and real experiments.
¯³ ´ ¯2The interpretation rests
+ ¯ † ¯
on the spectral density A (k,ω). According to eq. 8.98 ¯ b ak ¯ is the probability that,
n0
if one creates an electron in state k and add it to an N -electron system in its ground
state |N, 0i , the N + 1 electron system will be found in its eigenstate |N + 1, ni. The
δ(~ω−²n0 ) factor in eq. 8.103 merely selects the eigenstate for which the energy difference
²n0 corresponds to the frequency ~ω. A similar reasoning holds for A− (k,ω), the creation
of a hole, and probability to find the the N − 1 electron system in a particular one of its
eigenstates. The spectral density A± (k, ω) is actually what is measured in real experiments
such as photoemission or inverse photoemission, as we will show now.

8.6.1 Photoemission
In a photoemission experiment a photon is absorbed by a system that is initially in its
ground state |ii = |N, 0i. We take a photon that has enough energy to excite an electron
to an energy which is high enough, such that the electron becomes free and can leave
the system. Usually this requires a deep-UV or or an X-ray photon. Photoemission
experiments are simply a controlled setup for probing the photoelectric effect, for which
Einstein got his Nobel prize. The basic experiment is shown in Fig. 8.6.
The final state of the system can be described as
a†q |N − 1, ni
|f, ni = |q; N − 1, ni = b
Here q denotes the state of the highly excited, free electron which is now essentially
decoupled from the rest of the system; |N − 1, ni is the state of the system left behind,
234 CHAPTER 8. PROPAGATORS AND DIAGRAMS

which has one hole in it. We keep the label n to distinguish between the possible final
states. The outgoing electron is detected and its energy ²q is measured by the detector.20
Its momentum ~q p can be obtained by measuring the direction of the outgoing electron
and using ~q = 2m²q for its size. The state q of the outgoing free electron can thus
be characterized completely. Obviously the state of the incoming photon is completely
characterized by its energy ~ω and its momentum ~K. We assume that Fermi’s golden
rule holds for the absorption process, which means that the light source must not be
too strong in order for first order perturbation theory to hold. The transition rate for
photoemission wP E then becomes
X ¯¯ ¯2 ¡
¯ ¢
wP E ∝ b
¯hf, n|V |ii¯ δ ~ω−[²q + EnN−1 − E0N ]
n
X ¯¯ ¯2 ¡
¯ ¢
= ¯hN − 1, n|b b
aq V |N, 0i¯ δ ~ω−[²q + EnN−1 − E0N ] (8.109)
n

The δ-function says that the photon energy ~ω must make up for the difference between
the initial state energy E0N , and the final state energy, which is the sum of the energy ²q
of the free electron and the energy EnN−1 of the system left behind. The perturbation
operator Vb describes the absorption process; it has been discussed in Section 7.2, eqs.
7.35—7.38 Since we are not interested in photons right now, we will focus solely on its
electronic part. Since an electron has to be excited from below the Fermi level to above
the Fermi level, the term that contributes must have the form
X
Vb = a†l bb†k
Vlkb (8.110)
|l|>kF ,|k|≤kF

The individual terms in the sum create a hole below the Fermi level and an electron above
it.21 Using eq. 8.110 in eq. 8.109 gives products like b a†kbb†l . Using the anti-commutation
aqb
relations for Fermion operators we can rewrite them as

a†kbb†k = δ qlbb†k − b
aqb
b aqbb†k
a†l b
= δ qlbb† + b
k a†bb† b
aq
l k (8.111)

If we apply this expression in eq. 8.109, the second term of it contributes zero because

b
aq |N, 0i ≈ 0 (8.112)

i.e. a “free” electron in a high energy state has negligible overlap with the system’s ground
state. This is a crucial point. For a system of non-interacting electrons eq. 8.112 is exact,
since the high-energy state |qi is not occupied in the ground state and so there is nothing to
annihilate by baq . For a system of interacting electrons, eq. 8.112 is only an approximation.
20
This is not the full story, since an electron that crosses the surface of a material has to increase its
(electrostatic) potential (if the potential inside a material would not be lower than in vacuum, then all
the electrons would fly off spontaneously). The energy of the detected electron is ²q − V0 , where V0 is the
electrostatic potential level in vacuum, far from the material. In order not too complicate the notation,
we have omitted V0 here. It will merely shift the photoemission spectrum by a constant.
P P b ak and
21
Other possible terms in this operator are a†l b
|l|>kF ,|k|>kF Vlk b ak ; |l|≤kF ,|k|>kF Vlk bl b
P b b †
|l|≤kF ,|k|≤kF Vlk bl b . These terms however do not participate in the absorption as can easily be shown
k
(the second term gives emission, for instance).
8.6. (INVERSE) PHOTOEMISSION AND QUASI-PARTICLES 235

However, the approximation becomes better the higher the energy of the “free” electron.
In the practice of photoemission it works very well. Only the first term in eq. 8.111 is
thus retained and eq. 8.109 becomes
X ¯¯ ¯2
¯
³ ´
¯hN − 1, n|bbk |N, 0i¯ |Vqk | δ ~ω−[²q + ²N−1
† 2
wP E ∝ n0 − µ]
n,|k|≤kF
X X
= A− (k, [² − ²q +µ]/~) |Vqk |2 = VP2E A− (k, [² − ²q + µ]/~)
(8.113)
|k|≤kF |k|≤kF

where we have written ² = ~ω for the energy of the incoming photon and used the definition
of the spectral function, eq. 8.106. If the transition matrix element Vqk is not very
dependent on the states, as is often the case when the energy ²q of the “free” electron is
high enough, we can approximate it by a single number VP E . This is the second crucial
point. We can write
X
wP E ∝= VP2E A− (k, [² − ²q + µ]/~) (8.114)
|k|≤kF

Then the photoemission transition rate wP E is directly proportional to the spectral density
A− . The photoemission signal is proportional to the transition rate, so the spectral density
of a material is directly measured in a photoemission experiment !

8.6.2 Inverse Photoemission


Inverse photoemission is the following experiment. One comes in with an electron at a
relatively high energy ² = ~ω. This electron is captured by the material and transferred
to a lower energy level (decited ?), where at the same time a photon of energy ~ω q = ²q is
emitted. The photon is detected and its energy and momentum are measured. The basic
idea is shown in Fig. 8.7.
The experiment is the inverse of photoemission; hence its name. It is also known under
the name “Bremsstrahlung isochromat spectroscopy” (BIS), i.e. spectrally resolved light
produced by slowing down an electron. It is easy to prove under the same conditions as
above that the transition rate for inverse photoemission wIP E (ω), and thus the inverse
photoemission signal, is proportional to the spectral density A+ .
X
2
wIP E ∝ VIP E A+ (k, [² − ²q − µ]/~) (8.115)
|k|>kF

EXAMPLES
It is instructive to consider some examples. First inverse photoemission from a system
which consists of non-interacting particles. The spectral function then has the form given
by eq. 8.107 and inserted in eq. 8.115 one gets
X
2
wIP E ∝ VIP E δ(² − ²q − ²k ) (8.116)
|k|>kF

A δ-peak appears in the spectrum whenever the energy of the incoming electron minus
that of the outgoing photon, ² − ²q , coincides with an energy-level ²k which is available
236 CHAPTER 8. PROPAGATORS AND DIAGRAMS

=ω q

µ = εF
εk

Figure 8.7: Inverse photoemission: incoming electron of energy ² and outgoing photon of
energy ~ωq .

for a particle in the system. In a similar way one gets for the photoemission spectrum for
a system of non-interacting particles, using eq. 8.107 in eq. 8.114
X
wP E ∝ VP2E δ(² − ²q − (−²k )) (8.117)
|k|≤kF

A δ-peak appears in the spectrum whenever the energy of the incoming photon minus that
of the outgoing electron, ² − ²q , coincides with an energy-level −²k which is available for a
hole in the system. In a system of interacting particles, the spectral function has the quasi-
particle form of eq. 8.108. Instead of δ-functions, the (inverse) photoemision spectrum
now consists of a series of Lorentzian peaks. Each peak is centered around an energy level
²k + ∆k ; it has a width determined by Γk and an integrated intensity given by its weight
zk . In fact, following the discussion in Section 7.2.3, in a real material each energy level
²k,i can be assigned a band index i. Photoemission is then an experimental technique for
measuring the band structure ²k,i of the valence bands (the occupied states). Similarly
inverse photoemission is a technique for measuring the band structure of the conduction
bands (the unoccupied states). Note that the choice of zero-point for the energies does
not play a role, in case you were wondering about it in view of Figs. 8.6 and 8.7. For
instance, choosing ²F = µ as zero-point, we can make the substitutions ²q → ²q − µ and
−²k → µ − ²k in eq. 8.117, without changing wP E ; see also the “loose ends” in Section
6.2.22
22
Again this is not the full story, since the measured energy of the detected electron is ²0q = ²q −V0 , where
V0 is the electrostatic potential level in vacuum, far from the material. So one should use ²q = ²0q + V0
in eq. 8.117. One can write this as ²q = ²0q + µ + [V0 − µ]. The quantity W = V0 − µ is called the
work function; it is the minimum energy required to bring an electron from inside a material to outside at
infinity from the material. It can be measured by photoemission in the following way. The free electrons
8.7. APPENDIX I. THE ADIABATIC CONNECTION 237

8.7 Appendix I. The Adiabatic Connection


8.7.1 The Problem
Consider again the electron propagator
D ¯ ¯ E
¯ ¯
i~G+ (k, m, t2 − t1 ) = Ψ0 ¯b a†m (t1 )¯ Ψ0
ak (t2 ) b (8.118)

For a system of particles that interact with one another, or with atomic potentials, ob-
viously |Ψ0 i is the ground state or vacuum of the full system, i.e. including all interac-
tions. We argued that one has to make a perturbation expansion of the time evolution
operator when working out the operators b ak (t2 ) = Ub † (t2 , 0)b b (t2 , 0) and b
ak U a†m (t1 ) =
b † (t1 , 0)b
U b (t1 , 0). This should be equivalent to a diagram expansion as discussed. How-
a†m U
ever, translating diagrams into algebra again using Table 8.4, one obtains expressions that
contain unperturbed G0 (m, k, t2 − t1 )’s and simple matrix elements only. It is not at all
obvious that this is correct. There are two problems.

1. The trick we used in Section 8.2.2, which is inserting resolutions of identity between
each pair of operators, cf. eq. 8.23, is no good in general. We get matrix elements
like hΨ0 |b
ak | ni, with |ni some n-particle/hole state. The state |Ψ0 i is an eigenstate
of the full Hamiltonian (the ground state), which includes the perturbation, so there
is no simple expression for this matrix element. Whereas in the example treated in
Section 8.3.1 |Ψ0 i could be obtained in principle, albeit after some calculation, this
is not so for the interacting electron gas. The ground state of an interacting electron
system can in general not be obtained exactly.23 It does not make much sense then
to even write down a matrix element like hΨ0 |b ak | ni, since we can not evaluate it.

2. To add to our problems, we used |k| ≤ kF and |k| > kF to distinguish between
electron and hole states. This distinction was based upon the non-interacting elec-
tron gas, where we could fill up the one electron states one by one, starting at the
bottom up to the Fermi level, see Section 6.3.1. In stone-age notation the states of
the non-interacting electron gas can be expressed as

|k1 k2 ....kN i(a) = |k1 ik2 i....|kN i(a) (8.119)

where in the ground state |k1 | ≤ |k2 | ≤ .... ≤ |kN | = kF . This is not an eigenstate
of a full Hamiltonian, when we include atomic potential and/or electron-electron
interactions. If we use states like those of eq. 8.119 as a basis set, the ground state
of the full Hamiltonian is a linear combination of such states, in which also states
with |k| > kF are involved, see e.g. eq. 8.37 in Section 8.3. On forehand it is not
clear what role the Fermi energy or wave number kF plays in an interacting system;
indeed it is not clear that it plays any role at all. In any case, kF will not be “sharp”
in an interacting system, in the sense that the probability of observing an electron
of highest energy are produced from the Fermi level, i.e. using ²k = ²F = µ in eq. 8.117, see also Fig. ??.
Using these expressions ²k and ²q in eq. 8.117 gives a δ(² − ²0q − W ) corresponding to the freed electrons of
highest energy. Since ² and ²0q are measured, the work function W can be obtained. See also A. Zangwill,
Physics at Surfaces, (Cambridge Un. Press, Cambridge, 1988).
23
For a homogeneous interacting electron gas the ground state wave function can be obtained only
after a very lengthy computation using the so-called Quantum Monte Carlo approach. Calculations for
inhomogeneous systems (i.e. which include atomic potentials) are a topic of present research.
238 CHAPTER 8. PROPAGATORS AND DIAGRAMS

in a state with |k| ≤ kF is one and in a state with |k| > kF it is zero. At best this
probability will change with increasing |k| from close to one to close to zero near a
characteristic wave number kF . At worst, the probability will change smoothly and
no clear kF can be distinguished.24

So have we been staging a hoax in assuming that one may use the diagrams of Table
8.4? Well, we have not, but the solution to the problems discussed above involves some
intricate reasoning. It can be found in Mattuck, Appendices B—G. I will walk you through
the reasoning up to and including Appendix E. The ultimate solution lies in the famous
“Wick’s theorem”, which is discussed in Appendix F, and the proof that the diagram-
matic procedure followed in Section 8.3 is correct is found in Appendix G. The following
Appendix II gives a simplified (but non-general) discussion of this case.

8.7.2 The Solution


To facilitate formal operations, the propagator is written as
D ¯ n o¯ E
¯ ¯
i~G(k, m, t2 − t1 ) = Ψ0 ¯T b a†m (t1 ) ¯ Ψ0
ak (t2 ) b (8.120)

where T {...} is Wick’s time ordering operator, cf. eqs. 8.10 and 8.11. It leads to

i~G(k, m, t2 − t1 ) = i~G+ (k, m, t2 − t1 ) t2 > t1



= i~G (k, m, t2 − t1 ) t2 ≤ t1 (8.121)

The main idea behind the time ordering operator is that it allows for a compact expression
of the electron propagator G+ and the hole propagator Gn− into one propagator o G. The
general definition of Wick’s time ordering operator is T A(t b a )B(t
b b )...Z(t b z ) = (−1)p ×
the operators A(tb a ), B(t
b b ), ..., Z(t
b z ) rearranged such, that times ..ti , tj , ... increase from
right to left. In case two times are equal, an annihilation operator comes to the right of
a creation operator of the same time. p is the number of interchanges needed to get all
the operators in the right order and it accounts for the anti-commutation rules of Fermion
operators.

We have a look at the time-dependent perturbation expansion again. Mattuck explains


it in his Appendix B and D; it should look very familiar to you, as we discussed it in the
first chapters. He uses a slightly different notation, which is explained in the following
table

My notation Mattuck’s notation Where


Vb H1 perturbation
UbI (t, t0 ) = Ub † (t, 0)U
b (t, t0 ) e (t, t0 )
U eq. D.1, p.356
0
VbI (t) = U b (t, 0)Vb U

0
b0 (t, 0) e1
H eq. D.5
b b †
aI,k (t) = U0 (t, 0)b b0 (t, 0)
ak U b
ak (t) eq. D.5
24
For “normal” electron systems, which are called Fermi liquids, the situation is actually “at best” and
a clear Fermi wave number can still be distinguished. A discussion of this is presented in Mattuck’s §11.3,
p.209; note especially the figures 11.1-11.3.
8.7. APPENDIX I. THE ADIABATIC CONNECTION 239

Using this table as a tool for translation, you should be able to recognize eq. D.10,
Mattuck p.357 from my lecture notes. The first useful theorem is due to Dyson. It links the
terms of the time-dependent perturbation expansion, which I have introduced in Chapter
2, eq. 2.20, to an expression involving the time ordering operator we have just discussed
µ ¶n Z t Z τn Z τ n−1 Z τ3 Z τ2
b (n) (t, t0 ) = 1
U I dτ n dτ n−1 dτ n−2 ...... dτ 2 dτ 1
i~ t0 t0 t0 t0 t0
VbI (τ n ) VbI (τ n−1 ) VbI (τ n−2 ) ......VbI (τ 2 ) VbI (τ 1 )
µ ¶ Z Z t Z t Z t Z t
1 1 n t
= dτ n dτ n−1 dτ n−2 ...... dτ 2 dτ 1
n! i~ t0 t0 t0 t0 t
n o0
T VbI (τ n ) VbI (τ n−1 ) VbI (τ n−2 ) ......VbI (τ 2 ) VbI (τ 1 ) (8.122)

All the time integrals now run from t0 to t, which is convenient. This theorem can be
proved using induction.

Now we address our main problem. We do know the ground state of the non-interacting
(i.e. the unperturbed) system H b 0 |Φ0 i = E0,0 |Φ0 i, but we don’t know the ground state
of the interacting (i.e. the perturbed) system (H b 0 + Vb )|Ψ0 i = E0 |Ψ0 i. We apply the
adiabatic theorem. Suppose at t = −∞ our system is in the unperturbed ground state
|Φ0 i. Then we turn on the perturbation Vb “adiabatically slow” by Vb e−α|t| , with α a
very small, real and positive number. The adiabatic theorem then states that at t = 0,
the system is in the ground state of the perturbed Hamiltonian
1
bI (0, −∞)|Φ0 i
|Ψ0 i = N − 2 lim U (8.123)
α→0+

The proof of the adiabatic theorem is found by a straightforward generalization of one of


the exercises. N is a normalization factor we will derive below. Now we switch off the
perturbation again, again infinitely slow by Vb e−α|t| . By applying the adiabatic theorem
again, we find that at t = ∞ the system is back in the ground state of the unperturbed
Hamiltonian
1
bI (∞, 0)|Ψ0 i ⇔
|Φ0 i = N 2 lim U
α→0+
− 12 bI (0, ∞)|Φ0 i
|Ψ0 i = N lim U (8.124)
α→0+

b −1 (t1 , t2 ) =
where the last line follows from the properties of the time-evolution operator UI
bI (t2 , t1 ). The normalization factor N is obtained from
U

N hΨ0 |Ψ0 i = N = hΦ0 |Ub † (0, ∞)U bI (0, −∞)|Φ0 i


I
bI (∞, 0)U
= hΦ0 |U bI (∞, −∞)|Φ0 i
bI (0, −∞)|Φ0 i = hΦ0 |U (8.125)

The basic idea is that we can derive the ground state of the perturbed system from that
of the unperturbed system by this adiabatic switching and, via eq. 8.122, we can make a
perturbation expansion of the time-evolution operator UbI (t2 , t1 ). This we can put to use
in our propagator of eq. 8.120.
240 CHAPTER 8. PROPAGATORS AND DIAGRAMS

It remains to rewrite the propagator a bit. Write

b b † (t, 0)b
ak (t) = U b (t, 0)
ak U
b † (t, 0)U
= U b0 (t, 0)Ub † (t, 0)b b0 (t, 0)U
ak U b † (t, 0)U
b (t, 0)
0 0
b † (t, 0)b
= U bI (t, 0)
aI,k (t)U (8.126)
I

using the definitions of the table above. Using eqs. 8.123—8.126 in eq. 8.120 then gives
bI (∞, t2 )b
hΦ0 |U bI (t2 , t1 )b
aI,k (t2 )U bI (t1 , −∞)|Φ0 i
a†I,m (t1 )U
i~G(k, m, t2 − t1 ) =
hΦ0 |U bI (∞, −∞)|Φ0 i

where properties of the time evolution operator like U bI (t1 , 0)U


bI (0, −∞) = U
bI (t1 , −∞) and
b † b
UI (t1 , t2 ) = UI (t2 , t1 ) have been applied. With the time-ordering operator we can give a
more compact expression
bI (∞, t2 )b
U bI (t2 , t1 )b
aI,k (t2 )U bI (t1 , −∞) =
a†I,m (t1 )U
n o
T UbI (∞, t2 )b bI (t2 , t1 )b
aI,k (t2 )U bI (t1 , −∞) =
a†I,m (t1 )U
n o
T U bI (∞, −∞)b a†I,m (t1 )
aI,k (t2 )b

In conclusion, the propagator is given by the expression


n o
hΦ0 |T UbI (∞, −∞)b a†I,m (t1 ) |Φ0 i
aI,k (t2 )b
i~G(k, m, t2 − t1 ) = (8.127)
bI (∞, −∞)|Φ0 i
hΦ0 |U
We now have what we wanted; all the matrix element are between |Φ0 i, the ground state
of the non-interacting (the unperturbed ) system. We know how to handle this. All the
problems we complained about at the beginning of this appendix vanish because we can
work with |Φ0 i, which has the form given in eq. 8.119. The prize that we are paying is
that the expression of eq. 8.127 is much more complex than that of eq. 8.120, due to
bI terms. We can make a series expansion of this operator, the n’th term of which is
the U
given by eq. 8.122. The result is
X∞ µ ¶ Z Z ∞ Z ∞
1 1 n ∞
i~G(k, m, t2 − t1 ) = dτ n ...... dτ 2 dτ 1
n=0
n! i~ −∞ −∞ −∞
n o
b b b
hΦ0 |T VI (τ n ) ......VI (τ 2 ) VI (τ 1 ) b aI,k (t2 )b†
aI,m (t1 ) |Φ0 i
X∞ µ ¶ Z Z ∞ Z ∞
1 1 n ∞
dτ n ...... dτ 2 dτ 1
n! i~ −∞ −∞ −∞
n=0
n o
hΦ0 |T VbI (τ n ) ......VbI (τ 2 ) VbI (τ 1 ) |Φ0 i (8.128)

The expressions for the operators VbI and b


aI,k can be found from the table.25 In the inter-
acting many-particle system the perturbation Vb is formed by two-particle interactions,
25
Note we have to use adiabatic switching on and off so we have to modify Vb by the time factor e−α|t| .
An elegant alternative for this is not to modify Vb and, instead of integrating over real times τ i ; i = 1, .., n,
to integrate slightly below the real axis from τ i = −∞(1 − iα) to ∞(1 − iα); this then automatically adds
the time factors e−α|t| .
8.8. APPENDIX II. THE LINKED CLUSTER EXPANSION 241

which according to Chapter 6, eq. 6.35 can be written as


1 X
Vb = a†l b
Vklmnb a†kb
amb an from which it is easy to prove
2
klmn
1 X
VbI (t) = a†I,l (t)b
Vklmnb a†I,k (t)b
aI,m (t)b
aI,n (t) (8.129)
2
klmn

Perturbing atomic potentials of course have a one-particle form, according to eq. 6.34 in
Chapter 6
X
Vb = a†kb
Vkl b al from which one gets
kl
X
VbI (t) = a†I,k (t)b
Vkl b aI,l (t) (8.130)
kl

Using either of the expressions of eqs. 8.129 or 8.130 in eq. 8.128 one observes that the
calculation amounts to computing a large number of matrix elements of the type
n o
hΦ0 |T ba†I,l (τ n ) ......b
aI,n (τ 1 ) b a†I,m (t1 ) |Φ0 i
aI,k (t2 )b

Rather than doing all the algebra by hand, this can be done in an elegant way using Wick’s
theorem, as explained in Mattuck’s Appendix F. From it follows the diagram expansion
as we have used it in this chapter, using another elegant theorem called the linked cluster
theorem, cf. Mattuck’s Appendix G.

The whole process relies upon the “adiabatic switching on/off” trick by which we
“construct” the ground state |Ψ0 i of our interacting system of particles from the ground
state |Φ0 i of the system of non-interacting particles. This trick usually works fine, but
it is not garanteed to work. If you look back the relevant exercize (or read Mattuck’s
Appendix D), the trick fails if

hΦ0 |Ψ0 i = 0

i.e. when there is no overlap between the ground state of the non-interacting system
and the ground state of the interacting system. This is not usually the case, but it does
happen occasionally. Superconductors are an example where the adiabatic trick fails; see
Mattuck’s Chapter 15. This however is another story, and I am not the man to tell it.

8.8 Appendix II. The Linked Cluster Expansion


In this appendix we will try to evaluate the expression of eq. 8.39 of Section 8.3 step by
step using straight-forward algebra. It reads

bI (∞, t2 )b
h00 |U bI (t2 , t1 )b
aI,A (t2 )U bI (t1 , −∞)|00 i
a†I,A (t1 )U
+
i~G (A, A, t2 − t1 ) = (8.131)
h00 |UbI (∞, −∞)|00 i

The end result will be something called the linked cluster expansion. I will not try to prove
it fully, since that would be too involved even for this rather simple model, but only make
242 CHAPTER 8. PROPAGATORS AND DIAGRAMS

it plausible to you. A full derivation based upon Wick’s theorem can be found in Mattuck’s
appendices F and G for the (non-interacting) many particle case. Proofs for interacting
many particle cases can be found in advanced books on many particle physics.26

8.8.1 Denominator
We first do the denominator h00 |U bI (∞, −∞)|00 i of eq. 8.131. We use the expansion of
eqs. 2.19 and 2.20 in Chapter 2 and get up to second order
Z
b 1 t
h00 |UI (t, t0 )|00 i = h00 |I + dτ 1 VbI (τ 1 )
i~ t0
µ ¶2 Z t Z τ2
1
+ dτ 2 dτ 1 VbI (τ 2 ) VbI (τ 1 ) |00 i (8.132)
i~ t0 t0

where
h i
b † (τ , t0 ) Vb e−α|τ | U
VbI (τ ) = U0
b0 (τ , t0 ) (8.133)

and
X ³ ´ X ³ ´
Vb = a†kb
VkAb a†Ab
aA + VAkb ak + VkAbbkb a†Abb†k
aA + VAkb (8.134)
|k|>kF |k|≤kF

as in eq. 8.45. I prefer to let the time limits t0 , t stand for a while and take the limit to
−∞, ∞ later on. Note that the exponential factors e−α|τ | ensure that the integrals always
converge. The zero’th order term in eq. 8.132 is trivial. It gives h00 |I|00 i = h00 |00 i = 1;
the state is normalized. The unperturbed time-evolution operator simply gives

b0 (τ , t0 ) |00 i = e− ~i (τ −t0 )Hb 0 |00 i = e− ~i (τ −t0 )E0 |00 i


U (8.135)

since according to eqs. 8.33 and 8.41


X X
b 0 = E0 −
H ²kbb†kbbk + a†kb
²kb a†Ab
ak + ²Ab aA (8.136)
|k|≤kF |k|>kF

The first order term of eq. 8.132 becomes


Z t Z
b (1) (t, t0 ) = 1 b b b 1 t
U I dτ 1 e −α|τ 1 |
h0 |U
0 0

(τ ,
1 0t ) V U 0 (τ ,
1 0t ) |00 i = dτ 1 e−α|τ 1 | h00 |Vb |00 i
i~ t0 i~ t0
Z X
1 t
= dτ 1 e−α|τ 1 | VAk h00 |ba†Abb†k |00 i = 0 (8.137)
i~ t0
|k|≤kF

a†Abb†k |00 i = h00 |1A 1k i = 0; the states of the number representation form an
because h00 |b
orthonormal basis set. Note that the time factors resulting from U b0 and Ub † cancel. Note
0
26
Such as, A. L. Fetter and J. D. Walecka, Quantum Theory of Many-Particle Systems, (McGraw-Hill,
Boston, 1971); S. Doniach And E. H. Sondheimer, Green’s Functions for Solid State Physicists, (Imperial
College Press, London, 1998); G. D. Mahan, Many Particle Physics, (Plenum, New York, 2000). J. W.
Negele and H. Orland, Quantum Many-Particle Systems, (Addison-Wesley, Redwood City, 1988).
8.8. APPENDIX II. THE LINKED CLUSTER EXPANSION 243

also that the other three terms in the perturbation Vb give zero since they try to annihilate
a a particle in the ground state (which does not contain any particles).

The second order term can be written as


µ ¶2 Z t Z τ2
b (2) 1 −α|τ 2 | b † (τ 2 , t0 )Vb U
b0 (τ 2 , τ 1 ) Vb U
b0 (τ 1 , t0 ) |00 i
UI (t, t0 ) = dτ 2 e dτ 1 e−α|τ 1 | h00 |U0
i~ t0 t0
µ ¶2 Z t Z τ2
1 −α|τ 2 |
= dτ 2 e dτ 1 e−α|τ 1 |
i~ t0 t0
i
b0 (τ 2 , τ 1 ) Vb |00 ie− ~i (τ 1 −t0 )E0
e ~ (τ 2 −t0 )E0 h00 |Vb U (8.138)
P P
a†Abb†k |00 i = |k|≤kF VAk |1A 1k i we rewrite this as
Using Vb |00 i = |k|≤kF VAkb
µ ¶ Z Z τ2
b (2) (t, t0 ) = 1 2 t
U I dτ 2 e−α|τ 2 | dτ 1 e−α|τ 1 |
i~ t0 t0
i X b0
− ~i (τ 2 −τ 1 )H
e ~ (τ 2 −τ 1 )E0 ∗
VAk VAl h1A 1k |e |1A 1l i
|k|,|l|≤kF
µ ¶2 Z t Z τ2
1 −α|τ 2 |
= dτ 2 e dτ 1 e−α|τ 1 |
i~ t0 t0
X i i
e− ~ (τ 2 −τ 1 )²A e ~ (τ 2 −τ 1 )²k |VAk |2 (8.139)
|k|≤kF

since
i b i
e− ~ (τ 2 −τ 1 )H0 |1A 1l i = e− ~ (τ 2 −τ 1 )(E0 +²A −²l ) |1A 1l i for |l| ≤kF (8.140)

and h1A 1k |1A 1l i = δ k,l . This is easily obtained from eq. 8.136 by noting that H b 0 only
contains number operators, i.e. n bh,l = bbl bbl ; n

be,l = b†
al b
al count the number of holes or

electrons in state l, respectively, and n bA = b aAb aA counts the number of electrons in the
atomic state A. Obviously the state |1A 1l i with |l| ≤kF has one hole in state l and one
electron in the atomic state A.

The final result of eq. 8.139 can be given a nice diagrammatic interpretation in terms
of “unperturbed” propagators. The unperturbed electron propagator is defined as
D ¯ ¯ E
¯ † ¯
i~G+ 0 (l, k, t2 − t1 ) = 0 0 ¯ba I,l (t2 ) b
aI,k (t1 )¯ 00 ; t2 > t1 ; |k|, |l| >kF
D ¯ ¯ E D ¯ ¯ E
¯ b† b0 (t2 , t1 ) b b0 (t1 , 0)¯¯ 00 = e ~i (t2 −t1 )E0 00 ¯¯b b0 (t2 , t1 ) b ¯
= 00 ¯U 0 (t2 , 0) b
al U a†k U al U a†k ¯ 00
D ¯ ¯ E
i
(t2 −t1 )E0 ¯b ¯
=e ~ 1l ¯U0 (t2 , t1 )¯ 1k (8.141)

which is similar to the full propagator of eq. 8.46, but with unperturbed states and
evolution operators. It follows
i i
i~G+
0 (l, k, t2 − t1 ) = e
(t −t1 )E0 − ~ (t2 −t1 )(E0 +²k )
~ 2 e h1l |1k i
i
= e− ~ (t2 −t1 )²k δ l,k ; t2 > t1 ; ²k > ²F (8.142)
244 CHAPTER 8. PROPAGATORS AND DIAGRAMS

Note that this expression is identical to the single particle case of eq. 8.14 ! This is true
in general; the unperturbed propagator in a many-particle system has the same form as
that of a single particle system. A similar ccalculation for the hole propagator gives
D ¯ ¯ E
¯b bb† (t2 )¯¯ 00 ; t2 ≤ t1
−i~G− 0 (l, k, t2 − t1 ) = 00 ¯bI,k (t1 ) I,l
i
= e− ~ (t2 −t1 )²k δ l,k ; t2 ≤ t1 ; ²k ≤ ²F (8.143)
which is again identical to the single particle hole propagator of eqs. 8.26 and 8.27. With
the help of the two propagators of eqs. 8.142 and 8.143, the bottom line of eq. 8.139 can
be rewritten as
Z t Z τ2
b (2) −α|τ 2 |
UI (t, t0 ) = − dτ 2 e dτ 1 e−α|τ 1 |
t0 t0
X
+ −
G0 (A, A, τ 2 − τ 1 )G0 (k, k, τ 1 − τ 2 )|VAk |2 (8.144)
|k|≤kF

This is almost the result we want. The exponential factors e−α|τ | are in fact superfluous
now, because in the definition of Green functions we have already incorporated such expo-
nential factors, see subsections 3.2.4 and 8.1.2; therefore we can discard them. Moreover,
the time boundaries of eqs. 8.142 and 8.143 can also be taken into account by Θ functions,
as in Chapter 3. With that, we can let both integrals run from t0 to t and safely take the
limit t0 → −∞ and t → ∞. The result becomes
Z ∞ Z ∞ X
b (2)
UI (∞, −∞) = − dτ 2 dτ 1 G+ −
0 (A, A, τ 2 − τ 1 )G0 (k, k, τ 1 − τ 2 )|VAk |
2
−∞ −∞ |k|≤kF
(8.145)
This now has a simple diagrammatic representation, if we the dictionary of Table 8.4.

V Ak k VkA
τ1 τ2
b (2) (∞, −∞) =
U A (8.146)
I

The right going arrow represents the unperturbed electron propagator (in state A), the
left going arrow represents the unperturbed hole propagator (in state k) and the dots
(vertices) represent the matrix elements. Note that VAk ∗ = V , see the text following
kA
eq. 8.44. The rule is to sum or integrate over all intermediate states and time labels.
Moreover, for each closed “fermion” loop like in eq. 8.146 that is encountered, a “−” sign
is added. This is one of the rules connecting Feynman diagrams such as eq. 8.146 to
algebraic expressions like eq. 8.145 according to Table 8.4.

After a lengthy discussion, the final expression of eq. 8.132 in diagram form becomes,
up to second order in the perturbation
bI (∞, −∞)|00 i =
h00 |U
V Ak k VkA
I + τ1 τ2 + .....
A (8.147)
8.8. APPENDIX II. THE LINKED CLUSTER EXPANSION 245

The higher order terms can be derived systematically in a similar way. See whether you
can derive all fourth order terms yourself (and argue why there are no third order terms).

8.8.2 Numerator
Unfortunately, eq. 8.147 is just half the job we have to do. We still have to calculate the
numerator of the expression of eq. 8.131.

bI (∞, t2 )b
h00 |U bI (t2 , t1 )b
aI,A (t2 )U bI (t1 , −∞)|00 i
a†I,A (t1 )U (8.148)

This looks like a formidable expression, if we consider expanding each of the evolution
factors U bI as a perturbation series as in eq. 8.132. However, things are somewhat less
horrible than they seem at first sight. First of all, the expression for VbI (τ ) which is
needed in the perturbation expansions is identical to eq. 8.45, if we substitute b a†A by
a†I,A (τ ), etcetera.
b
X ³ ´
VbI (t) = a†I,k (t)b
VkAb a†I,A (t)b
aI,A (t) + VAkb aI,k (t)
|k|>kF
X ³ ´
+ VkAbbI,k (t)b a†I,A (t)bb†I,k (t)
aI,A (t) + VAkb (8.149)
|k|≤kF

This can easily be proven from eqs. 8.40 and 8.133. I have skipped the exponential factor
e−α|τ | now, because we have seen already that its main function is to let the time integrals
in the perturbation series converge; from now on we will assume that they convergence.
The time dependence of these creation and annihilation operators is rather simple
i i
a†I,A (t) = b
b a†A e ~ t²A ; b aA e− ~ t²A
aI,A (t) = b
i i
a†I,k (t) = b
b a†k e ~ t²k ; b ak e− ~ t²k
aI,k (t) = b
bb† (t) = bb† e− ~i t²k ; bbI,k (t) = bbk e ~i t²k (8.150)
I,k k

This can be proved straightforwardly.

b † (t, 0) b
a†I,A (t)|0A i = U
b b0 (t, 0) |0A i
a†A U
0
i b i b i b i
a†A e− ~ tH0 |0A i = e ~ tH0 b
= e ~ tH0 b a†A e− ~ tE0 |0A i
i i b i i b
a†A |0A i = e− ~ tE0 e ~ tH0 |1A i
= e− ~ tE0 e ~ tH0 b
i i i
a†A |0A i
= e− ~ tE0 e ~ t(E0 +²A ) |1A i = e ~ t²A b

using eq. 8.40; see also the discussion around eq. 8.140. I have written the relation for
the state in which there are no electrons in the atomic state A, but it is easy to see that it
also holds for all states with any numbers of particles or holes in the electron gas states.
Therefore it is a property of the operator. The other relations in eq. 8.150 can be proven
in a similar way.

bI in h00 |....|00 i, eq. 8.148, as in eq.


A perturbation expansion of the evolution factors U
8.132, leads to a string of creation and annihilation operators ...b a†I,k (t)b
aI,l (t)... between
246 CHAPTER 8. PROPAGATORS AND DIAGRAMS

the |....| . Let us start with the easy bits first. The zero’th order term is obtained by
setting all UbI = I in eq. 8.148. It gives the unperturbed propagation; compare eq. 8.46

aI,A (t2 )b
h00 |b a†I,A (t1 )|00 i = i~G+
0 (A, A, t2 − t1 ) (8.151)
A A
t1 t2
=

All possible first order terms give zero contributions. For instance look at the term
Z t1
b (1) (t1 , −∞)|00 i 1
aI,A (t2 )b
h00 |b a†I,A (t1 )UI = a†I,A (t1 )
aI,A (t2 )b
h00 |b dτ 1 VbI (τ 1 ) |00 i
i~ −∞
Z t1 X
1
= h00 |b a†I,A (t1 )
aI,A (t2 )b dτ 1 a†I,A (τ 1 )bb†I,k (τ 1 )|00 i
VAkb
i~ −∞ |k|≤kF
Z t1 X
1
= dτ 1 VAk h00 |b
aI,A (t2 )b a†I,A (τ 1 )bb†I,k (τ 1 )|00 i = 0
a†I,A (t1 )b (8.152)
i~ −∞ |k|≤kF

going from the first to the second line one observes that all the other terms in eq. VbI (τ )
cannot contribute, since they try to annihilate a particle that is not present in the un-
perturbed ground state, cf. eq. 8.149. However, looking at the third line of eq. 8.152
one observes that it also must be zero, since b a†I,A (t1 )b
a†I,A (τ 1 ) = 0. Because of the Pauli
principle one cannot create two particles in the same state. One can argue even more
globally whether a matrix element h00 |A|0b 0 i must be zero, where A b is some string of cre-
ation and annihilation operators. Each creation operator increases the number of particles
(or holes) by one and each annihilation operator decreases this number by one. The state
b 0 i must contain zero particles and holes, otherwise h00 |φi = h00 |A|0
|φi = A|0 b 0 i = 0. This
b
means that each creation operator in the A string has to be balanced by an annihilation
operator of the same kind and vice versa in order to give a non-zero overall result. In eq.
8.152 the operator bb†I,k is not balanced by a bbI,k and one of the b a†I,A ’s is not balanced by a
b
aI,A . Therefore the overall result has to be zero. By a similar reasoning it is easy to show
that the other possible first order terms are also zero
Z t2
b (1) (t2 , t1 )b 1
aI,A (t2 )U
h00 |b I a†I,A (t1 )|00 i = h00 |b
aI,A (t2 ) dτ 1 VbI (τ 1 ) b
a†I,A (t1 )|00 i = 0
i~ t1
(8.153)

Z ∞
(1)
b (∞, t2 )b 1
h00 |UI a†I,A (t1 )|00 i = h00 |
aI,A (t2 )b dτ 1 VbI (τ 1 ) b a†I,A (t1 )|00 i = 0
aI,A (t2 )b
i~ t2
(8.154)

If one thinks this through, all perturbation terms which involve an odd number of VbI (τ )
terms in h00 |....|00 i must be zero. It always gives a string of operators in which the
number of ba†A or the number of b aA operators is unbalanced and such a string can never
give a contribution when sandwiched between h00 |....|00 i.
8.8. APPENDIX II. THE LINKED CLUSTER EXPANSION 247

The even order terms are not zero a priori, so we have to work out in detail which of
them give a contributiom. A possible second order term is
(2)
b (t1 , −∞)|00 i
aI,A (t2 )b
h00 |b a†I,A (t1 )UI
(2)
b (t1 , −∞)|00 i
aI,A (t2 )b
= h00 |b a†I,A (t1 )|00 ih00 |UI (8.155)
In the second line we have inserted |00 ih00 |. You may wonder why we are allowed to do
that. We should actually insert a full resolution of identity
X Y Y
I= |nk nA i hnk nA | (8.156)
nk ,nA k k

where one sums over all possible states containing any number of particles and holes.
However, operating with the product b a†I,A (t1 ) on each of these states does not
aI,A (t2 )b
change the number of particles or holes. Taking the inner product h00 |b a†I,A (t1 ) then
aI,A (t2 )b
only gives a non-zero contribution for the state with no particles or holes (i.e. the ground
state), since all the states with a different number of particles and holes are orthogonal
Y
h00 | |nk nA i = 1 if all nk = nA = 0
k
= 0 otherwise (8.157)
We have calculated the two factors in the second line of eq. 8.155 before. Using eqs. 8.151
and 8.145 one gets

aI,A (t2 )b
h00 |b b (2) (t1 , −∞)|00 i = i~G+ (A, A, t2 − t1 ) ×
a†I,A (t1 )UI 0
Z t1 Z ∞ X
− dτ 2 dτ 1 G+ −
0 (A, A, τ 2 − τ 1 )G0 (k, k, τ 1 − τ 2 )|VAk |
2
(8.158)
−∞ −∞ |k|≤kF

Note that we can let the time on the dτ 1 integral run from −∞ to ∞ because G+ 0 and
G−0 are zero anyway if τ 1 > τ 2 . In a similar way one obtains for another possible second
order term
b (2) (∞, t2 )b
h00 |U a†I,A (t1 )|00 i = i~G+
aI,A (t2 )b
I 0 (A, A, t2 − t1 ) ×
Z ∞ Z ∞ X
− dτ 2 dτ 1 G+ −
0 (A, A, τ 2 − τ 1 )G0 (k, k, τ 1 − τ 2 )|VAk |
2
(8.159)
t2 −∞ |k|≤kF

Note that the only difference between the expression of eqs. 8.158 and 8.159 are the limits
on the time integrals. The remaining second order term is

aI,A (t2 )U
h00 |b b (2) (t2 , t1 )b
a†I,A (t1 )|00 i
I
µ ¶2 Z t2 Z τ2
1
aI,A (t2 )
= h00 |b dτ 2 dτ 1 VbI (τ 2 ) VbI (τ 1 ) b
a†I,A (t1 )|00 i
i~ t1 t1
µ ¶2 Z t2 Z τ2
1
= dτ 2 aI,A (t2 )VbI (τ 2 ) VbI (τ 1 ) b
dτ 1 h00 |b a†I,A (t1 )|00 i
i~ t1 t1
µ ¶2 Z t2 Z τ2 X
1
= dτ 2 dτ 1 VAk VkA
i~ t1 t1 |k|>kF

aI,A (t2 )b
h00 |b a†I,A (τ 2 )b a†I,k (τ 1 )b
aI,k (τ 2 )b a†I,A (t1 )|00 i
aI,A (τ 1 )b (8.160)
248 CHAPTER 8. PROPAGATORS AND DIAGRAMS

The last step can be made since all the other terms in VbI try to annihilate a particle
which is not present, or give “double” terms such as ba†Ab
a†A , or lead to a state which has
a non-zero number of particles or holes which gives zero on account of eq. 8.157. We can
apply the same trick as in eq. 8.155 and write
a†I,A (τ 2 )b
aI,A (t2 )b
h00 |b a†I,k (τ 1 )b
aI,k (τ 2 )b a†I,A (t1 )|00 i
aI,A (τ 1 )b
= h00 |b a†I,A (τ 2 )|00 ih00 |b
aI,A (t2 )b a†I,k (τ 1 )|00 i ×
aI,k (τ 2 )b
a†I,A (t1 )|00 i
aI,A (τ 1 )b
h00 |b
= i~G+ +
0 (A, A, t2 − τ 2 ) × i~G0 (k, k, τ 2 − τ 1 ) ×
i~G+
0 (A, A, τ 1 − t1 ) (8.161)
Inserted in eq. 8.160 it gives
b (2) (t2 , t1 )b
aI,A (t2 )U
h00 |b a†I,A (t1 )|00 i
I
Z ∞ Z ∞ X
= i~ dτ 2 dτ 1 |VAk |2
−∞ −∞ |k|>kF

G+
0 (A, A, t2
+
− τ 2 )G0 (k, k, τ 2 − τ 1 )G+
0 (A, A, τ 1 − t1 ) (8.162)
where as before we let the Θ-factors in the Green functions take care of the proper time
limits on the integrals. All the second order contributions are given by the sum of the
final expression of eqs. 8.158, 8.159 and 8.162. However, we are not entirely satisfied
with the time boundaries of the integrals of eqs.8.158 and 8.159, so we do a little bit of
resummation
Z t1 Z ∞ Z ∞ Z ∞
− dτ 2 dτ 1 ... − dτ 2 dτ 1 ...
−∞ −∞ t2 −∞
Z ∞ Z∞ Zt2 Z ∞
= − dτ 2 dτ 1 ... + dτ 2 dτ 1 ... (8.163)
−∞ −∞ t1 −∞

The first double integral of the second line gives


i~G+ (A, A, t2 − t1 ) ×
Z 0∞ Z ∞ X
− dτ 2 dτ 1 G+ −
0 (A, A, τ 2 − τ 1 )G0 (k, k, τ 1 − τ 2 )|VAk |
2
(8.164)
−∞ −∞ |k|≤kF

We are going to rewrite the second double integral of eq. 8.163. One term in it is
Z t2 Z ∞
dτ 2 dτ 1 G+ + −
0 (A, A, t2 − t1 )G0 (A, A, τ 2 − τ 1 )G0 (k, k, τ 1 − τ 2 )
t1 −∞
µ ¶2 Z t2 Z ∞
1 i i
= dτ 2 dτ 1 e− ~ (t2 −t1 )²A e− ~ (τ 2 −τ 1 )²A G−
0 (k, k, τ 1 − τ 2 )
i~ t1 −∞
µ ¶2 Z t2 Z ∞
1 i i
= dτ 2 dτ 1 e− ~ (t2 −τ 1 )²A e− ~ (τ 2 −t1 )²A G−
0 (k, k, τ 1 − τ 2 )
i~ t1 −∞
Z t2 Z ∞
= dτ 2 dτ 1 G+ + −
0 (A, A, t2 − τ 1 )G0 (A, A, τ 2 − t1 )G0 (k, k, τ 1 − τ 2 )
Zt1∞ Z−∞

= dτ 2 dτ 1 G+ + −
0 (A, A, t2 − τ 1 )G0 (A, A, τ 2 − t1 )G0 (k, k, τ 1 − τ 2 )(8.165)
−∞ −∞
8.8. APPENDIX II. THE LINKED CLUSTER EXPANSION 249

From the first to the second line we just used the definition of the propagator, eq. 8.142;
then we reorder the exponentials a little bit, and use the definition again. Note that finally
we let the integrals run from −∞ to ∞; the Θ-factors in the Green functions take care of
the proper time limits. The second double integral of eq. ?? finally becomes
Z ∞ Z ∞ X
i~ dτ 2 dτ 1 |VAk |2
−∞ −∞ |k|>kF

G+
0 (A, A, t2

− τ 1 )G0 (k, k, τ 1 − τ 2 )G+
0 (A, A, τ 2 − t1 ) (8.166)

All the second order contributions are now given by the sum of eqs. 8.162, 8.164 and
8.166. All the integrals run neatly from −∞ to ∞. We can transfer them easily into
diagrams. For instance eq. 8.162 becomes
τ1 k τ2
A A
t1 VA,k Vk , A t2
(8.167)
As usual one sums over all intermediate states k and over all possible times τ 1 and τ 2
(where τ 1 < τ 2 , because of the propagator G+
0 (k, k, τ 2 − τ 1 )). In a similar way eq. 8.164
becomes in diagram form
A A

V Ak k VkA
τ1 τ2
A (8.168)
where use the diagrams of eqs. 8.146 and 8.151. The rule is that whenever you find two
pieces in a diagram that are not connected, like the straight line and the bubble in eq.
8.168, you simply multiply their contributions. A diagram which contains unconnected
pieces is called an unlinked diagram. The diagram in eq. 8.167 is then of course a linked
diagram.This corresponds exactly to eq. 8.164. Finally, eq. 8.166 becomes in diagrammatic
form
τ2
A V A ,k
t1
k
Vk , A A
τ1 t2
(8.169)
The intermediate propagator is now the hole propagator G− 0 (k, k, τ 1 −τ 2 ) with τ 1 < τ 2 . It
is of course a linked diagram again. The upper three diagrams plus the zero order diagram
of eq. 8.151 give all the contributions up to second order of the expression of eq. 8.148.
bI (∞, t2 )b
h00 |U bI (t2 , t1 )b
aI,A (t2 )U bI (t1 , −∞)|00 i
a†I,A (t1 )U

+ + +

= (8.170)
where I omitted the labels on the diagrams to simplify the notation.
250 CHAPTER 8. PROPAGATORS AND DIAGRAMS

8.8.3 Linked Cluster Theorem


We can collect the numerator of eq. 8.170 and the denominator of eq. 8.147 to find the
expression for the propagator of eq. 8.131 up to second order perturbation. In diagram
form it reads
bI (∞, t2 )b
h00 |U bI (t2 , t1 )b
aI,A (t2 )U bI (t1 , −∞)|00 i
a†I,A (t1 )U
i~G+ (A, A, t2 − t1 ) =
h00 |UbI (∞, −∞)|00 i

+ + +

I +
= (8.171)
This is still a complicated expression even up to second order, let alone if we include higher
order terms. However we can get rid of the unlinked diagrams. As noted in connection
with eq. 8.168 such diagrams are simply multiplications between their unconnected parts.
The unconnected straight arrow in the first two diagrams of the numerator in eq. 8.171 can
be factored out. This arrow of course corresponds to the propagator i~G+ 0 (A, A, t2 − t1 )
in eq. 8.164. We can write

× (I + ) + +

I +
(8.172)
One observes that the factor between (...) in the numerator is identical to the denominator
such that the two cancel. This is a very general result ! If one pursues the perturbation
expansion of eq. 8.171 and includes all higher order terms, then a large number of unlinked
diagrams appear in the numerator. They can all be factored out and the end result is

( + + + ..... ) × (I + + ..... )

I + + .....

(8.173)
where the “....” indicate higher order terms. Note that the last factor of the numerator
exacly cancels the denominator ! The final result therefore becomes
i~G+ (A, A, t2 − t1 ) =

+ + + .....
(8.174)
8.8. APPENDIX II. THE LINKED CLUSTER EXPANSION 251

In other words: the propagator only contains linked diagrams. This is the content of what
is known as the linked cluster theorem.

As I promised you in the beginning of this appendix, after some lengthy algebra the
final result looks quite elegant. It resembles the perturbation expansion we made for the
propagator in the single particle system of Chapter 4, see Fig. 4.4. There is actually only
one significant difference. Here the intermediate propagation can proceed via a a particle
or a hole propagator, whereas in Chapter 4 we had only the particle propagator, since there
we only had a single particle. Actually using his remarkable intuition, Feynman thought
of this diagram expansion before all the details of the perturbation expansion algebra were
sorted out. Once you have accepted the diagram way, it gives you fast results. I think
that you will all agree with me that sorting out the algebra of the perturbation expansion
is a lot of work.27 Schwinger and Tomonaga, who obtained the Nobel prize together with
Feynman (in 65 for quantum electrodynamics), did it the hard way by algebra. From
now on we are going to assume that the diagrammatic way works. General proofs that it
works, also in more complicated cases, can be found in the advanced books cited in the
beginning of this section.

27
Probably you will not believe me, but this is only a simple example. Other cases like many interacting
particles, or quantum electrodynamics are much more complicated.
252 CHAPTER 8. PROPAGATORS AND DIAGRAMS
Chapter 9

The electron-electron interaction

“It seems very pretty,” she said when she had finished it, “but it’s rather hard to understand!”
(You see she didn’t like to confess even to herself, that she couldn’t make it out at all.) “Somehow it
seems to fill my head with ideas—only I don’t exactly know what they are!”, Lewis Carrol, Through
the Looking Glass.

9.1 Many interacting electrons


As in the previous chapter, our focus is on the one particle propagator and the gedanken-
experiment it represents. For particles it reads, cf. eq. 8.7,
D ¯ ¯ E
¯ ¯
i~G+ (m, k, t2 − t1 ) = 0 ¯b a†m (t1 )¯ 0
ak (t2 ) b (9.1)

There is however a snake in the grass; the vacuum state |0i is the ground state of the
interacting electron gas and we complained before about not being able to find solutions
(including the ground state) for the interacting electron gas. However, the “adiabatic
connection” and the “linked cluster expansion” discussed in the appendices of the previous
chapter ensure that the diagrammatic techniques we are going to use also work in this
case. From now on we will assume they do. As before the propagator gives the probability
amplitude of finding a particle in a state m at a time t2 , after it has been created in a state
k at a time t1 . It represents a gedanken experiment, but real experiments like (inverse)
photoemission are closely connected to it.

As for the non-interacting electron gas, we approach the problem by rewriting eq. 9.1
in terms of time-evolution operators and making a time-dependent perturbation series
expansion. For the interacting case, the mathematics is much more involved. We will
ignore this completely and handle the diagrams just as in the non-interacting case (which
is correct). The “unperturbed” Hamiltonian H b 0 is taken to be the kinetic energy plus any
atomic potentials present of the type of eq. 8.77.1 The two-particle interaction Vb of eq.
8.81 is now treated as the perturbation. Since this represents a Coulomb repulsion, which
by comparing eqs. 8.75 and 8.80 one expects to be of a similar order of magnitude as the
attraction by the atomic potentials, the perturbation will not be small. On forehand one
1
In a homogeneous electron gas there is only the kinetic energy off course.

253
254 CHAPTER 9. THE ELECTRON-ELECTRON INTERACTION

therefore expects a need to sum over a full perturbation series, rather than considering
just the first few terms of the series. One can only do this in an approximative way (since
otherwise we would have found a full solution for the interacting electron gas, which we
cannot). The basic idea of this chapter is to describe a sequence of such well-defined,
much used approximations of increasing sophistication. We will argue that the strong
two-particle Coulomb repulsion can be taken into account by defining quasi-particles (i.e.
quasi-electrons and quasi-holes). These behave almost as if they were independent particles
with a weak coupling to the rest of the system. This is called the “normal” situation or the
“normal Fermi liquid”.2 The “abnormal” many-particle system in which the properties
cannot be interpreted in terms of “simple” single quasi-particles is then called a “strongly
correlated” system.3 Here we consider “normal” systems only.

A diagram representation of the perturbation series for the one particle propagator
proceeds just as in the previous chapter. Because we are dealing with a two particle
interaction now, we have to link up two-particle diagrams like those of eqs. 8.84—8.86,
which consist of two incoming and two outgoing particles and a (two particle) interaction
matrix element. As before, the one electron propagator is written as
k1 k2
i~G+ (k2 , k1 , t2 − t1 ) ≡ t1 t2

k1 k k' k2
t1 τ τ' t2
= (9.2)
Note that there is one incoming and one outgoing line. This represents the gedanken-
experiment; at t1 we send in a particle in state k1 , and at t2 we consider the probability
amplitude that the particle is in state k2 . A single arrow represents the single particle
propagator of the unperturbed system, i.e. the non-interacting electron gas.
k1 k
t1 τ
= i~G+
0 (k, k1 , τ − t1 )
i
= δ k,k1 e− ~ (τ −t1 )²k1 Θ (τ − t1 ) Θ (²k1 − ²F ) (9.3)
Where the Θ functions take care of the fact that this function is different from zero only
if t1 > τ and ²k1 > ²F (since we are dealing with the electron propagator). Writing down
a diagram like eq. 9.2 one implicitly assumes a summation over all intermediate states k
and k0 and an integration over all intermediate times τ and τ 0 . The interesting part is the
middle part
k k'
τ τ' (9.4)
2
The interaction between electrons is repulsive. This means that there is no gas/liquid phase transition
in the system as a function of the electron density and/or the temperature. Since one cannot distinguish
between a gas and a liquid phase, the words “gas” and “liquid” are used interchangeably. The common
practice is inconsistent; one speaks of an “electron gas” or a “Fermi liquid”; two names for the same thing.
If the two-particle interaction is attractive, one can have a condensation-like phase transition in a quantum
system; the occurence of superconductivity at low temperature is a well-known example of this.
3
Ususally found in solid state materials containing atoms that cause strongly localized atomic shell
effects, such as the high Tc superconductors.
9.2. THE HARTREE APPROXIMATION 255

This is where all the (inter)action is (with the system). Our job to find approximations
for this part in terms of a perturbation series. A number of approximations of increasing
complexity, but also of increasing physical relevance, will be discussed in the following
sections. At the end of this chapter you will be almost up-to-date with the modern state-of-
the-art of normal systems. The famous linked cluster theorem of diagrammatic expansions
states that when expanding eq. 9.2 into two-particle diagrams, only those terms give a
contribution in which all diagrams are linked; in other words, diagrams that contain parts
which are not linked to the rest give no contribution. Also there cannot be any loose lines,
except for the two arrows given in eq. 9.2. Internally, all the arrows inside eq. 9.4 have
to be connected up. Otherwise, we would have other particles going out or coming in
besides the one represented by eq. 9.2.4 We will bare this in mind when constructing our
diagrams.

9.2 The Hartree approximation


“Double, double, toil and trouble, Fire burn, and cauldron bubble”, Shakespeare, Macbeth.

The electron-electron interaction is large. This presents a problem for a perturbation


expansion, in which the electron-electron interaction is used as the perturbation on a non-
interacting system of electrons. The perturbation series will not simply converge after
just a few terms. More horrible even, some of the terms in the expansion turn out to be
infinite and need to be canceled by other infinite terms. To avoid such problems with ill-
behaved series, we adapt a somewhat different strategy here, which consists of two steps.
In the first step we identify a contribution (i.e. a single diagram) which, on physical
grounds, we expect to give an important contribution. In the second step, we then sum
this contribution to infinite order in the perturbation series, which gives us a well-behaved
finite result. This last step we do by solving the Dyson expansion, just as in the foregoing
chapters. In addition, we introduce the brilliant concept of self-consistency, which enables
us to sum in a very elegant way over a series of complicated looking additional diagrams.
The end-result of all this summation is in an “easy-to-understand and user friendly” quasi-
particle form.
We start with a simple approximation, called the Hartree approximation. It is seldom
used in practice anymore nowadays and is mainly of historical relevance, but it allows us
to introduce the concepts and strategies which we will also use in more complicated (and
more accurate) approximations. In the next section the Hartree diagram is introduced,
and the technique of summing it to infinite order in the perturbation series. We dis-
cuss the distinction between time-ordered (Goldstone) and non-time-ordered (Feynman)
diagrams. A physical interpretation is then given by introducing the Hartree potential
and the Hartree self-consistent field equations; the bonus of which is a summation over
an extra class of diagrams. Finally we argue why we still should be looking for a better
approximation.
4
And it would not be a single particle propagator anymore. Two particles coming in, and two going out
would also be a relevant experiment, i.e. a two-particle scattering experiment. One needs a two-particle
propagator for that case. We will not consider such more complicated propagators here, but they can be
studied using the same diagrammatic techniques. Once you get the hang of it, they should not be much
more difficult to understand.
256 CHAPTER 9. THE ELECTRON-ELECTRON INTERACTION

9.2.1 The Hartree (Coulomb) interaction


We kick off with the simplest first-order diagram, which is called the Hartree diagram. In
the time domain it is written as
m k
t1 t2

l
τ
X Z 1
= dτ i~G+
0 (k, k, t2 − τ ) Vklml i~G− +
0 (l, l, τ − τ ) i~G0 (m, m, τ − t1 )
i~
|l|≤kF

= i~G+ 1,H (k, m, t2 − t1 ) (9.5)


where the subscript H stands for “Hartree”. The diagram results from closing the following
diagram upon itself

The diagram has one incoming electron line labeled m, t1 and one outgoing electron line
labeled k, t2 . It is derived from the operator part Vklmlbbl b ambb†l . The so-called bubble
a†kb

is a “hole” line labeled l, τ , where |l| ≤ kF (or ²l ≤ ²F ) labels hole states. Note
again that we sum/integrate over these intermediate states, as required. The hole prop-
agator G− 0 (l, l, τ − τ ) is instantaneous, because the Coulomb interaction “- - - - - - - -
” is instantaneous (see the relativistic intermezzo in the previous chapter). The hole is
created and immediately annihilated again. An equal time hole propagator is defined as
i~G−0 (l, l, τ − τ ) = −1. According to a sign rule which we are not going to discuss, each
“bubble” introduces an extra “−” sign in the equation, so the overall factor given by the
bubble is simply +1.5 We can rewrite eq. 9.5 as
Z
i~G1,H (k, m, t2 − t1 ) = i~ dτ G+
+ +
0 (k, k, t2 − τ ) VH,km G0 (m, m, τ − t1 )
X
where VH,km = Vklml (9.6)
|l|≤kF

is the so-called Hartree potential . This expression has the familiar first order perturbation
form we encountered in Chapter 3, eq. 3.7. We can write the diagram of eq. 9.5 just like
a simple one particle diagram, cf. Figs. 2.2, 2.4, 4.4 and eq. 8.82.
5
Handling signs correctly is one of the most tricky businesses in using the diagrammatic method. I
advice you to be pragmatic and simply follow the prescriptions, given by Mattuck in his table 4.3, p.86.
Deriving these rules is fairly straightforward in principle, but it involves some lengthy algebra.
In any case, one of Murphy’s laws of common experience states that you will always get the sign wrong;
even if you correct for this law.
9.2. THE HARTREE APPROXIMATION 257

m k
t1 t2

m
= t1 VH
k
t2
l
τ

Figure 9.1: The Hartree diagram

The Hartree potential acts like a simple one particle potential, scattering an incoming
particle in state m into an outgoing particle in state k. At t = τ a hole is created in a
state l, which is immediately filled again and, apparently, this process acts as a scattering
potential. To get a physical interpretation of the Hartree potential, we write it in the
position representation, cf. eqs. 6.21, 8.80, 8.81 and 9.6
X
VH,km = Vklml
|l|≤kF
X Z Z e2
= d3 r1 d3 r2 φ∗k (r1 )φ∗l (r2 ) φ (r1 )φl (r2 )
|r1 − r2 | m
|l|≤kF
Z
= d3 r1 φ∗k (r1 )VH (r1 )φm (r1 ) (9.7)

where the Hartree potential in the position representation VH (r1 ) is given by


X Z |φ (r2 )|2
VH (r1 ) = e2 d3 r2 l
|r1 − r2 |
|l|≤kF
Z
ρ (r2 )
= e d3 r2 0 (9.8)
|r1 − r2 |
The physical interpretation is as follows; |l| ≤ kF labels single particle states which are
occupied in the ground state of the electron gas. In the position representation these states
are represented by wave functions φl (r) (which are also called orbitals). PThe total charge
density resulting from these occupied orbitals is given by ρ0 (r) = e |l|≤kF |φl (r)|2 .6
Eq. 9.8 thus represents the electrostatic interaction between this charge density and a
test charge e at position r1 , which is the Coulomb or electrostatic potential caused by the
charge distribution ρ0 (r). It will be clear from eq. 9.7 that this potential (the Hartree
potential) acts as a simple potential in real space.

Intermezzo on Bubbles
This also supplies a physical argument why the “bubble” consists of a “hole” line and not
an “electron” line. A diagram containing
P instead of gives no contribution. The
“electron bubble” would give a |l|>kF in eqs. 9.7 and 9.8, but since these states are
6
ρ0 (r) is the sum over the probabilities of finding an electron at the position r, times the electronic
charge e.
258 CHAPTER 9. THE ELECTRON-ELECTRON INTERACTION

not occupied in the ground state of the electron gas, they cannot give rise to a potential.
Another way of arguing why the “electron bubble” cannot give a contribution is by noting
that a diagram like eq. 9.5 with G+ 0 ’s everywhere, must be derived from the operator
† †
Vklml b
al b
akb
ambal . Operating with this part on the ground state |0i gives zero, since the
right-most operator b al tries to annihilate a particle in a state l that is not occupied in the
ground state, since |l| > kF . The “hole bubble” does not have this problem. The diagram
of eq. 9.5 is then related to the operator Vklmlbbl b ambb†l , with |l| ≤ kF . Since these states
a†kb
are occupied, creating a hole by applying bb†l is no problem. The fact that “hole bubbles”
give a contribution and “electron bubbles” do not, is the physical reason why we have
choose the electron propagator at “equal times” to be zero; i~G+ 0 (l, l, τ − τ ) = 0, and the
hole propagator to be one; −i~G− 0 (l, l, τ − τ ) = 1.

Time ordering; Goldstone vs. Feynman diagrams


So far we have only considered the first order contribution of the Hartree diagram to the
propagator, eqs. 9.5 and 9.6. However the fact that we can treat this diagram conceptually
as simple potential scattering, cf. Fig. 9.1, means that we can use all the tricks of the
previous chapters on the summation of such potential terms to all orders. In diagram form
this looks like
m m τ
k k
m = + t2 +
k t1 t1
t2
t1 t2 VH ,mk

m τ n τ' k + ........
t1 t2
VH ,mn VH ,nk

Figure 9.2: Summing over Hartree diagrams

Fig. 9.1 can be used to substitute the original Hartree diagram. For instance the
second order diagram of the bottom line becomes (without the labels)

Transferring the diagrams into algebra, one obtains (in the time domain)
Z
GH (k, m, t2 − t1 ) = G0 (k, m, t2 − t1 ) + dτ G+
+ + +
0 (k, k, t2 − τ )VH,km G0 (m, m, τ − t1 )
Z Z X
+ dτ dτ 0 G+ 0 + 0 +
0 (k, k, t2 − τ )VH,kn G0 (n, n, τ − τ )VH,nm G0 (m, m, τ − t1 ) + ...
|n|>kF
(9.9)

Where the subscript “H” on the propagator again indicates that we take into account
the Hartree diagrams only, and neglect other contributions (for the moment). We have
9.2. THE HARTREE APPROXIMATION 259

assumed the time order t2 > τ 0 > τ > t1 , etcetera, as usual in a perturbation series. But
we have learned in the previous chapter that these are not the only possibilities in a many
particle system. Diagrams of the following type also contribute

m VH ,mn
t1 τ
n
τ' k
t2
VH ,nk
=
Z Z X
dτ dτ 0 G+ 0 − 0 +
0 (k, k, t2 − τ )VH,kn G0 (n, n, τ − τ )VH,nm G0 (m, m, τ − t1 ) (9.10)
|n|≤kF

where the intermediate state n labels a hole state instead of an electron state. According to
convention this hole must be created at t = τ 0 before it can recombine with the incoming
electron at t = τ (when the hole is created, an outgoing electron is also created and
propagates to t = t2 ). The corresponding substituted Hartree diagram looks like

It might seem that we are in for a lot of complicated drawing if we want to keep track of
the possibility that intermediate states can label holes as well as electrons.7 Fortunately
we do not have to do all of this bookkeeping if we make use of the trick explained in the
previous chapter in Section 8.1. We define a “combined” propagator

G(k, m, τ 0 − τ ) = G+ (k, m, τ 0 − τ ) if τ 0 > τ


= G− (k, m, τ 0 − τ ) if τ 0 ≤ τ (9.11)

which is an electron propagator if τ 0 > τ and a hole propagator if τ 0 ≤ τ . The two second
order diagrams of eqs. 9.9 and 9.10 then can be combined to give
Z Z X
dτ dτ 0 G+ 0 0 +
0 (k, k, t2 − τ )VH,kn G0 (n, n, τ − τ )VH,nm G0 (m, m, τ − t1 )
n

We sum over all intermediate states n and integrate over all intermediate times τ 0 , τ and
sort out the hole/electron character using the definition of eq. 9.11. This convention also
saves us a lot of drawing. We will draw our diagrams like Fig. 9.1 and associate each
intermediate arrow with a G. In fact, while we are at it, we can omit the “+” label on all
the G’s and G0 ’s in eq. 9.9. This allows us to treat the electron and the hole propagator
7
There are only the two possibilities for the second order diagram we just saw. For the n’th order
diagram, there are 2n−1 possibilities. Moreover, it becomes pretty complicated for other (non-Hartree)
diagrams; see, e.g., eq. 9.33, p. 163 of Mattuck.
260 CHAPTER 9. THE ELECTRON-ELECTRON INTERACTION

within the same equation and we kill two birds with one stone.8 The advantage is clear;
we are freed of having to decide whether we use electrons (|n| > kF ) or holes (|n| ≤ kF )
each time we encounter an intermediate state. And if we do it the diagrammatic way, our
main arrows will go from left to right, instead of the turns and kinks appearing as in eq.
9.10. There is also a disadvantage; we loose the distinction between electrons and holes we
had in Fig. 9.1 and eq. 9.10, so the physical picture gets somewhat blurred. We also loose
time order. The diagrams of Fig. 9.1 and eq. 9.10 are “time-ordered”, which means that
going from left to right the time labels appear in their correct order; i.e. t2 > τ 0 > τ > t1
in Fig. 9.1 and t2 > τ ≥ τ 0 > t1 in eq. 9.10. If we agree to associate the diagrams of
Fig. 9.1 with G0 ’s instead of G+ 0 ’s, this simple time order gets lost, because of eq. 9.11,
and the times become merely labels over which to integrate. The time-ordered version
of the diagrams, i.e. with G± 0 ’s, are nowadays commonly called Goldstone diagrams; the
simplified, non-time-ordered diagrams with G0 ’s are called Feynman diagrams. In the
following we will use Feynman diagrams, since their algebra is simpler and we bare in
mind that these are less directly interpreted as a physical event.

Summing the Hartree series


We can now proceed with the perturbation series of eq. 9.9. Transformed into the fre-
quency domain it obtains the familiar form
GH (k, m, ω) = G0 (k, k, ω)δ km + G0 (k, k, ω)VH,km G0 (m, m, ω)
X
+ G0 (k, k, ω)VH,kn G0 (n, n, ω)VH,nm G0 (m, m, ω) + ... (9.12)
n
k or m label states that are eigenstates of the unperturbed Hamiltonian. Going along this
route, we are back to studying the scattering of an incoming particle, a topic we discussed
in depth in the previous chapters.

9.2.2 The Hartree Self-Consistent Field equations


At this point we want to go one step further and introduce the concept of a self-consistent
field. This concept leads to a set of equations that have a clear physical interpretation.
As a bonus, it gives us a tool to sum over a whole new class of diagrams.

Matrix equations
We can write eq. 9.12 as a matrix relation defining the matrix GH (ω) with elements
(GH (ω))km ≡ GH (k, m, ω). Defining the matrices G0 (ω) and V H in a similar way then
gives
GH (ω) = G0 (ω) + G0 (ω)V H G0 (ω) + G0 (ω)V H G0 (ω)V H G0 (ω) + ...
" ∞
#
X n
= G0 (ω) I + (V H G0 (ω))
n=1
= G0 (ω) [I − V H G0 (ω)]−1 which can be rearranged into
GH (ω) = G0 (ω) + GH (ω)V H G0 (ω) (9.13)
8
The dutch expression is: “twee vliegen met één klap slaan”. (Apparently the size of the country
influences the size of the animal featuring in the proverb. I wonder whether the american expression is:
“to kill two buffalos with one gunshot”.)
9.2. THE HARTREE APPROXIMATION 261

which is the familiar Dyson equation again. Remember


1
(G0 (ω))km = G0 (k, k, ω)δ km = δ km (9.14)
~ω − ²0,k ± iδ

b 0 an the ±iδ is for elec-


where ²0,k are the eigenvalues of the unperturbed Hamiltonian H
trons and holes respectively. Defining the obvious diagonal matrix (H 0 )km ≡ H0,km =
²0,k δ km one gets
³ ´
(G0 (ω))km = (~ω − H 0 ± iδ)−1 which gives
km
G0 (ω)−1 = ~ω − H 0 ± iδ (9.15)

The “−1” superscript means a matrix inversion now.9 Multiplying the left- and right-hand
side of eq. 9.13 with G0 (ω)−1 (from the right), rearranging the terms, and using eq. 9.15
finally gives

(~ω − H 0 − V H ± iδ) GH (ω) = I ⇔


GH (ω) = (~ω − H 0 − V H ± iδ)−1 (9.16)

This equation now shows explicitly that the propagator GH in the Hartree approximation
is the Green function (or Green matrix) associated with the Hamiltonian H 0 + V H , in
which the Hartree potential acts as an (extra) potential term.

Transform the basis set


We can make life simpler if we transform to eigenstates of this Hamiltonian. Define a new
basis set of one particle states |mH i, obtained by a linear transformation from the old set
b0 )
|ki (which are the single particle eigenstates of H
X
|mH i = |kiCkm or in wave function notation
k
X
ψ m (r) = φk (r)Ckm (9.17)
k

where, as usual, ψ m (r) = hr|mH i and φk (r) = hr|ki. Define a (column) vector |mH i the
elements of which are |mH i, where mH runs over the whole basis. Eq. 9.17 then becomes
in this notation

|mH i = C T |ki
= C −1 |ki (9.18)
¡ ¢
Note that C T mk = Ckm , as required. The last line follows because we wish our new
basis to be orthonormal, as was our old basis; an elementary theorem in linear algebra
states that the transformation between two orthonormal basis sets has to be unitary, and
thus C T = C −1 . Follow the linear algebra path. If A is a matrix represented on the
basis |ki, and AH is supposed to represent the same quantity on the basis |mH i and if
9
In fact I should have written ~ωI with I the identity matrix, instead of ~ω,but I don’t want to complicate
the notation too much. For the same reason, if you are mathematically strict, read iδI instead of iδ.
262 CHAPTER 9. THE ELECTRON-ELECTRON INTERACTION

the transformation between the bases is given by eq. 9.18, then AH = C −1 AC. We now
choose a very special transformation, namely the one which makes the “Hamiltonian”
H 0 + V H diagonal

C −1 (H 0 + V H ) C = ² (9.19)
where (²)km = ²k δ km is a diagonal matrix of the eigenvalues of H 0 + V H .

Hartree equations
The eigenvalue problem ofPeq. 9.19 can be written asP(H 0 + V H ) C = C². In terms of its
components this becomes m (H 0 + V H )km Cmn = m Ckm ²m δ mn = ²n Ckn . Denoting
the columns of C as vectors cn (i.e. its components are (cn )k = Ckn ), this equation then
becomes
(H 0 + V H ) cn = ²n cn (9.20)
with vector components (cn )k = Ckn
This clearly is an eigenvalue equation where the ²n ’s are the eigenvalues, and the columns
cn of C are the eigenvectors. The eigenvalue equations, eq. 9.20, are called the Hartree equations.
We can put these into wave function form. Using eqs. 8.75—8.79 and 9.7 we write the ma-
trix elements as
Z
H0,km = b 0 φm (r)
d3 r φ∗k (r)H
Z " #
3 ∗ ~2 2 X Ze2
= d r φk (r) − ∇ − φ (r) (9.21)
2me |R − r| m
R
Z
VH,km = d3 r φ∗k (r)VH (r)φm (r) (9.22)
Z
²n = ²n d3 r φ∗k (r)φn (r) (9.23)

where in eq. 9.21 we have ignored a possible band index (see Section ??). Using eqs. 9.17,
9.20 and 9.22 we can write
X XZ
V H cn → VH,km Cmn = d3 r φ∗k (r)VH (r)φm (r)Cmn
Zm m
3
= d r φ∗k (r)VH (r)ψ n (r)

Rewriting all the bits of eq. 9.20 in this form we get the expression
Z " #
3 ∗ ~2 2 X Ze2
d r φk (r) − ∇ − + VH (r) − ²n ψ n (r) = 0
2me |R − r|
R

Hartree self-consistent field equations and orbitals


Since these must hold for any of the basis functions φk (r), this can only be true if
" #
~2 2 X Ze2
− ∇ − + VH (r) − ²n ψ n (r) = 0 (9.24)
2me |R − r|
R
9.2. THE HARTREE APPROXIMATION 263

These are the Hartree equations in wave function form.10 Not surprisingly, the solutions
ψ n (r) are called the Hartree wave functions, or, since these are one particle wave func-
tions, the Hartree orbitals. There is one catch, however! We pretended that the Hartree
potential VH (r) is a simple fixed potential. We know however that it is derived from a
two particle matrix element Vklml , see eq. 9.7. If we decide to switch our representation
to a new basis set, namely the basis of Hartree states |nH i, or in wave function form the
Hartree orbitals ψ n (r), then all the indices of Vklml must refer to this new basis !! In other
words, the correct expression for the Hartree potential becomes
X Z |ψ (r0 )|2
VH (r) = e 2
d3 r0 l 0 (9.25)
|r − r |
|l|≤kF

We now face the following problem; in order to solve eq. 9.24 and find the orbitals ψ n (r),
we must know the Hartree potential VH (r), which contains a sum over the same orbitals
according to eq. 9.25. The problem is not a Catch-22 however. We can solve the non-
linear equation presented by eq. 9.24 by iteration. We start by guessing an initial set of
(0)
functions, for instance the eigenfunctions of H 0 , i.e. ψ n (r) = φn (r). From that we use
(0)
eq. 9.25 to construct a Hartree potential VH (r), which we then use in eq. 9.24 to find
(1) (1)
a new set of functions ψ n (r); use eq. 9.25 to construct a new Hartree potential VH (r),
etcetera. In practice this process almost always converges; after N cycles the “new”
(N) (N−1)
functions are identical to the “old” ones, i.e. ψ n (r) = ψ n (r). We have reached a
steady state which is called the self-consistent field , which means that the orbitals which
determine the electrostatic field derived from the Hartree potential, eq. 9.25, are identical
(i.e. self-consistent) with the solutions of the orbital equation, eq. 9.24.

Hartree propagator
Eqs. 9.24 and 9.25 together are called the Hartree self-consistent field equations. The
physics is pretty obvious; an electron (or hole) is put into a one particle state |nH i (or
orbital ψ n (r)). It experiences a potential VH which is the electrostatic repulsion of the
charge distribution caused by all other electrons in the system, also put in one particle
states |lH i. The particle’s energy in the state |nH i is given by the eigenvalue of the
Hartree equation ²n . The Hartree approximation to the propagator, eq. 9.16 acquires a
very simple form on the basis of these Hartree states. We write
C −1 (~ω − H 0 − V H ± iδ) GH (ω)C = C −1 IC = I ⇔
C −1 (~ω − H 0 − V H ± iδ) CC −1 GH (ω)C = I ⇔
(~ω − ² ± iδ) C −1 GH (ω)C = I ⇔
X
δ km (~ω − ²k ± iδ) GH,mn (ω) = δ kn
m
where in the last line the indices refer to the “new” basis, i.e. the Hartree basis. We thus
obtain
1
GH (k, n, ω) ≡ GH,kn (ω) = δ kn (9.26)
~ω − ²k ± iδ
10
Douglas R. Hartree used these equations to do the first “realistic” calculations on the states, wave
functions and charge densities of atoms. See e.g. Proc. Camb. Phil. Soc. 24, 89 (1928). In his poineering
years he used his pensioned father, W. Hartree, as a human “computer” in order to calculate the necessary
integrals, etcetera. Later on in the 50’s he helpt to build one of the first electronic computers in England.
264 CHAPTER 9. THE ELECTRON-ELECTRON INTERACTION

Compare this to eq. 9.14 ! The propagator in the Hartree approximation GH expressed
in the basis of the Hartree eigenstates (the solutions of the Hartree equations) has the
same mathematical form as the unperturbed propagator G0 in the unperturbed basis
(the eigenstates of H0 ). We can transform to the time domain by Fourier transform and
complex contour integration and find (only the diagonal term is non-zero)

k k
t1 t2

= i~GH (k, k, t2 − t1 )
i
= Θ(t2 − t1 )Θ(²k − ²F )e− ~ ²k (t2 −t1 )
i
− [1 − Θ(t2 − t1 )] [1 − Θ(²k − ²F )] e− ~ ²k (t2 −t1 ) (9.27)

Don’t let
D the
¯ Θ-functions confuse ¯ E you. The third line corresponds to the electron prop-
¯ † ¯
agator 0 ¯b aH,k (t2 ) b
aH,k (t1 )¯ 0 where we create an electron in the Hartree state labeled
D ¯ ¯ E
¯ ¯
k. The fourth line corresponds to the hole propagator 0 ¯bbH,k (t1 ) bb†H,k (t2 )¯ 0 ; compare
also to eqs. 9.3. We have effectively solved the problem posed by Fig. 9.2.

Dressing the propagator


Let us summarize what we have been doing in diagram form. Fig. 9.2 can be reformulated,
using the Dyson expansion of eq. 9.13 as

= + VH

VH =

Figure 9.3: The self-consistent Hartree approximation

The first of these diagram lines corresponds to the Dyson equation, which, following
eq. 9.13, can be solved in closed form and corresponds to summing the Hartree diagrams
to infinite order. The second line is the usual Hartree diagram but with a double line in the
“bubble” instead of a single one. This is the requirement imposed by the “self-consistent
field”; the Hartree potential has be calculated on the basis of the Hartree states, which,
following the derivation after eq. 9.5, is achieved when using the Hartree (hole) propagator
in the “bubble”.

Fig. 9.3 is the typical form that is encountered very often in diagrammatic practice.
Also in more elaborate approximations we encounter a similar pair of equations. The
first line is the Dyson equation, which expresses the propagator (a double line arrow) in
terms of the unperturbed propagator (a single line arrow) and a potential (the circle). Here
the potential is VH ; in a more complicated approximation this will be substituted by a
9.2. THE HARTREE APPROXIMATION 265

(frequency-dependent) self-energy Σ. Solving the Dyson equation corresponds to summing


all the terms that contain VH (or Σ) to infinite order in the perturbation series. The
second line expresses the potential (or self-energy) in terms of the two-particle interaction
(the dashed line) and the propagator. Since we use the full propagator here, the equations
represented by the first and second line have to be solved in a self-consistent way, as
explained above. In view of the foregoing it will be clear that solving the Hartree equations,
9.24 and 9.25, is equivalent to solving the diagram problem!!

Besides fancy pictograms the diagrammatic method also contains some fancy metaphor-
ical language. The unperturbed propagator is called the “bare” propagator and the full
propagator is called the “dressed” propagator. Using the “dressed” propagator in the
second line of Fig. 9.3, instead of the “bare” propagator as in Fig. 9.1, for instance, is
called “clothing” a diagram. In this case it is the Hartree diagram which gets “clothed”.
Writing out the series of Fig. 9.3 out in terms of “bare” propagators only, one observes
that VH incorporates diagrams like

= + + .....

Figure 9.4: The dressed (self-consistent) Hartree potential

“Clothing” or “dressing” a propagator, which is just a fancy term for “applying self-
consistency”, is thus seen to incorporate a whole series of additional diagrams, without
hardly any extra effort.

9.2.3 Pro’s and con’s of the Hartree approximation


We started with the non-interacting electron gas described by a Hamiltonian H b 0 . Its
(electron) propagator has the form given by eq. 9.3. An electron created in a state
k propagates unperturbed with energy ²0,k . In absence of any interactions its lifetime
is infinite; we say that the non-interacting electron gas consists of independent particles,
since the propagation of one electron is not changed by the presence of the others. We then
included all the diagrams of the form of eq. 9.5, the Hartree diagrams. By transforming
to a new basis set, consisting of the so-called Hartree states, the propagator, eq. 9.27
again has an independent particle form. An electron (or hole) created in a Hartree state k
propagates unperturbed with energy ²k ; its lifetime in this state is infinite. Let us call it
the Hartree “quasi-particle”.11 The energy ²k takes into account the average electrostatic
potential caused by the other particles, cf. eq. 9.25, but apart from that, this “quasi-
particle” (electron or hole) behaves just like a regular independent particle.
11
A particle created in an eigenstate of the unperturbed Hamiltonian is called a “particle”; a particle
created in any other state is called a “quasi-particle”. For the moment, this is simply convention.
266 CHAPTER 9. THE ELECTRON-ELECTRON INTERACTION

The electrostatic potential caused by the other particles is large, so we have incorpo-
rated an important physical effect. However, there are two major flaws in the Hartree
approximation

1. Considering the propagator, creating a particle etcetera, we talked as if we could


divide our system into “the particle added” plus “all the other particles”. In other
words we have distinguished a particle from the “others” and reasoned as if this
unique particle determines the propagator. However, we know that this particle is
fundamentally indistinguishable from the “others”. This means that we have no
means to determine whether an exchange of the incoming particle with one of the
particles in the system has taken place. As long as a particle is coming out, the
propagator is o.k.; we are not able to say which one is coming out, since they are
all the same. In order to have a propagator which is consistent with this basic
notion of identical particles, we must at least allow for such an exchange process.
An improvement of the Hartree approximation, which incorporates the exchange of
identical particles, is discussed in the next section.
2. The lifetime of a quasi-particle should be finite (hence the name “quasi-particle”
instead of just “particle”). We know that in a real interacting electron gas there is
no way to inject an electron (or hole) in any state such, that it remains unperturbed in
that state forever. The interactions between the electrons will eventually result in the
electron (or hole) being scattered into other states. Therefore, the probability that
the injected particle will remain in its original state will decay with time. Apparently,
the Hartree approximation does not account for this physical effect. The improved
approximation which we will discuss in the last section of this chapter will.

9.3 The Hartree-Fock approximation


“The Walrus and the Carpenter
Walked on a mile or so,
And then they rested on a rock
Conveniently low:
And all the little Oysters stood
And waited in a row”. Lewis Caroll, The Walrus and the Carpenter.

The Hartree-Fock approximation is still widely used today, not only because it is
the simplest approximation which incorporates the basic quantum mechanics of identical
particles, but also because it serves as a starting point for more complicated approxima-
tions.12 It adds one diagram, the “Fock” or “exchange” diagram to the Hartree procedure.
Apart from some technical complications, the route taken by the Hartree-Fock approxi-
mation is similar to that of the Hartree approximation. We travel this route once more
in the next two sections, passing the exchange interaction, the exchange potential and
the Hartree-Fock self-consistent field equations on the way. We then consider the ho-
mogeneous electron gas once more, for which it is possible to obtain closed analytical
12
The “Fock” or exchange part was introduced by Vladimir A. Fock, a soviet scientist from St. Pe-
tersburg, see e.g. Z. Phys. 61, 126 (1930). Science was a global village long before the www generation;
imagine: in a world without sms’s !
9.3. THE HARTREE-FOCK APPROXIMATION 267

expressions. This enables us to weigh the successes and failures of the Hartree-Fock ap-
proximation in the context of solid state physics. Its main flaw is connected to the physical
concept of screening. Attacking that problem is the main goal of the final section of this
chapter.

9.3.1 The exchange interaction


The particles in an electron gas are indistinguishable, so an incoming particle can be
exchanged with those already in the system without us being able to tell. The exchange
diagram represents such a process

m
t1 τ

k
t2
X Z 1
= dτ i~G+
0 (k, k, t2 − τ ) Vlkml i~G− +
0 (l, l, τ − τ ) i~G0 (m, m, τ − t1 )
i~
|l|≤kF

= i~G+ 1,X (k, m, t2 − t1 ) (9.28)

where the subscript X stands for “exchange”. The diagram results from closing the fol-
lowing diagram upon itself

This diagram has one incoming electron line labeled m, t1 and one outgoing electron line

labeled k, t2 . It is derived from the operator part Vlkml b a†kbbl b


ambb†l . The “snake” line
is again a “hole” line labeled l, τ , where |l| ≤ kF (or ²l ≤ ²F ) labels hole states. The
physical interpretation of the exchange diagram follows the same route as the Hartree
diagram. We can simplify the expression of eq. 9.28 using eq. ??
Z
i~G1,X (k, m, t2 − t1 ) = −i~ dτ G+
+ +
0 (k, k, t2 − τ ) VX,km G0 (m, m, τ − t1 )
X
where VX,km = Vlkml (9.29)
|l|≤kF

is the so-called exchange potential . Compared to the expression for the Hartree term, eq.
9.6, there are two differences.

1. In the exchange potential the labels k and l are exchanged. We will attach a physical
picture to this below.
268 CHAPTER 9. THE ELECTRON-ELECTRON INTERACTION

2. There is an overall “−” in front of the expression. There is no “bubble” sign rule in
this case; see the discussion around eq. 9.6.

Apart from these differences, the expression has the familiar first order perturbation
form again, which means we can represent the exchange diagram as a potential scattering
diagram.
m
t1 τ

m
l = t1 VX
k
t2

k
t2

Figure 9.5: The exchange potential

The exchange potential acts like a one particle potential, which scatters an incoming
particle in state m out into a state k. At time t = τ a hole is created in a state l which
is immediately filled again. To get a physical interpretation of the exchange potential, we
use the position representation once more
X
VX,km = Vlkml
|l|≤kF
X Z Z e2
= d3 r1 d3 r2 φ∗l (r1 )φ∗k (r2 ) φ (r1 )φl (r2 )
|r1 − r2 | m
|l|≤kF
Z Z
= d3 r1 d3 r2 φ∗k (r2 )VX (r2 , r1 )φm (r1 ) (9.30)

where the exchange potential in the position representation VX (r2 , r1 ) is defined as


X φ (r2 )φ∗ (r1 )
l
VX (r2 , r1 ) = e2 l
(9.31)
|r1 − r2 |
|l|≤kF

Non-local exchange potential


This seemingly strange “potential” depends upon two arguments r1 and r2 . The physical
interpretation is as follows; |l| ≤ kF labels the one particle states which are occupied in
the ground state of the electron gas. At position r1 an incoming electron arrives in state
m at time t = τ . At the same an electron-hole pair is created at position r2 . The hole in
state l proceeds to r1 where it recombines with the incoming electron, the created electron
comes out in state k. A shorter way of saying the same thing is: an electron comes in
at r1 in state m and knocks out an electron at r2 in state k. This process is determined
by the “potential” VX (r2 , r1 ) and proceeds via an intermediate hole. We need to extend
our notion of the term “potential”. A normal potential would depend on a single position
in space r1 ; from now on we will call this a local potential. VX (r2 , r1 ) depends on two
positions; it connects two arbitrary points in space; this we will call a non-local potential.
9.3. THE HARTREE-FOCK APPROXIMATION 269

Think of it as a result of our pinball game with identical particles. As soon as our incoming
particle enters the system and becomes part of the electron gas, it is indistinguishable from
the rest. We can measure the probability that a particle comes out, but we cannot tell
which. This “knock-out” process is governed by the (non-local) exchange potential, which
in a pictogram looks like

r2
r1

k
m

Figure 9.6: The exchange potential

Snakes and Oysters


As for the Hartree diagram we can also supply an argument why the exchange diagram
contains a “hole” line
P and not an “electron” line for the intermediate states l. An electron
line would give a |l|>kF in eqs. 9.29 and 9.30, but since these states are not occupied
in the ground state of the electron gas, they cannot gives rise to a potential. Another
way of arguing why the “electron snake” cannot give a contribution is by noting that a
diagram like eq. 9.28 must be derived from the operator Vlkml b a†kb
a†l b
am b
al in the electron
case. Working with this part on the ground state |0i gives zero, since the right-most
operator bal tries to annihilate a particle in a state l that is not occupied in the ground
state, since |l| > kF .

Instead of the diagrams of eq. 9.28 and Fig. 9.5 one usually finds in the literature the
following picture to represent the exchange diagram
m
t1 τ

k
t2

Figure 9.7: The exchange diagram

It means exactly the same as before, but as a drawing it is somewhat simpler. Mattuck
calls it the open oyster diagram.13 The exchange diagram is also called the Fock diagram,
13
Personally I prefer the “snake” diagram of eq. 9.28, since there the hole character of the intermediate
270 CHAPTER 9. THE ELECTRON-ELECTRON INTERACTION

and the exchange potential is also called the Fock potential.

Summing the Hartree-Fock series


The Hartree and the Fock diagram of Figs. 9.1, 9.5 (or 9.7) are the only possibilities to
“close” a two particle diagram in itself such that one particle comes in and one particle goes
out, that give a result which is different from zero. These diagrams (and the algebraic
equations they represent) are equivalent with scattering by one-particle potentials; the
Hartree and the exchange potentials, respectively. Like in the pure Hartree case we can
sum over a series of any such diagrams. A fourth order diagram, for instance, looks like

k1 τ1 k3 τ2
t1

k4
k5 τ3

k2
k6 k7 τ4 k9
t2

k8

Figure 9.8: A fourth order Hartree-Fock diagram

It represents the algebraic term


X Z
1
dτ 1 dτ 2 dτ 3 dτ 4 i~G0 (k9 , k9 , t2 − τ 4 ) Vk9 k8 k7 k8 i~G0 (k7 , k7 , τ 4 − τ 3 )
i~
k2 ,k3 ,k4 ,k5 ,k6 ,k7 ,k8
1 1
Vk6 k7 k5 k6 i~G0 (k5 , k5 , τ 3 − τ 2 ) Vk4 k5 k3 k4 i~G0 (k3 , k3 , τ 2 − τ 1 )
i~ i~
1
Vk k k k i~G0 (k1 , k1 , τ 1 − t1 ) (9.32)
i~ 3 2 1 2

where we have used the equal time expression for the G− 0 ’s. We use the Feynman con-
vention that each intermediate line can represent an electron or hole, and we sort out the
possible ±’s on the G’s in a later stage. Using the definition of Hartree and exchange
potentials, cf. Figs. 9.1, 9.5 and eqs. 9.7-9.8, 9.30-9.31 such diagrams can be simplified.
The fourth order diagram of Fig. 9.8 becomes

k1 τ1 k3 τ2 k5 τ3 k7 τ4 k9
t1 VH ,k1k 3 − VX ,k 3k 5 − VX ,k 5k 7 V H ,k 7 k 9 t2

Figure 9.9: Fourth order Hartree-Fock diagram


line is more clear. However, in the literature, the “oyster” diagram is prefered.
9.3. THE HARTREE-FOCK APPROXIMATION 271

Note the “−” sign in front of VX ; this is consistent with eqs. 9.29 and 9.30. By now
it will be clear that we can use an expansion like in Fig. 9.2, but with the possibility
of putting VH or −VX at each node. It leads to a propagator or Green function in the
frequency domain of the type
GHF (k, m, ω) = G0 (k, k, ω)δ km + G0 (k, k, ω)VH,km G0 (m, m, ω)
− G0 (k, k, ω)VX,km G0 (m, m, ω)
X
+ G0 (k, k, ω)VH,kn G0 (n, n, ω)VH,nm G0 (m, m, ω)
n
X
− G0 (k, k, ω)VH,kn G0 (n, n, ω)VX,nm G0 (m, m, ω)
n
X
− G0 (k, k, ω)VX,kn G0 (n, n, ω)VH,nm G0 (m, m, ω)
n
X
+ G0 (k, k, ω)VX,kn G0 (n, n, ω)VX,nm G0 (m, m, ω) + .... (9.33)
n

“HF ” now stands for Hartree-Fock approximation. In short-hand notation


GHF (k, m, ω) = G0 (k, k, ω)δ km + G0 (k, k, ω)ΣHF (k, m)G0 (m, m, ω)
X
+ G0 (k, k, ω)ΣHF (k, n)G0 (n, n, ω)ΣHF (n, m)G0 (m, m, ω) + ....
n
where ΣHF (k, m) = VH,km − VX,km (9.34)
is called the Hartree-Fock potential or the Hartree-Fock self-energy.

9.3.2 The Hartree-Fock Self-Consistent Field equations


The self-consistent field method, applied to the Hartree-Fock case, allows for a summation
over a whole new class of diagrams. Moreover, the resulting equations can be given a
direct physical interpretation.

Matrix equations
Defining matrices as before, (GHF (ω))km = GHF (k, m, ω) and (ΣHF )km = ΣHF (k, m)
we can “solve” this equation similar to the Hartree case and get
GHF (ω) = (~ω − H 0 − ΣHF ± iδ)−1 (9.35)
This equation now shows explicitly that the propagator GHF in the Hartree-Fock ap-
proximation is the Green function (or Green matrix) associated with the Hamiltonian
H 0 + ΣHF , in which the Hartree-Fock potential acts as the (extra) potential term.

Transform the basis set; the Hartree-Fock equations


As before, we can make life simpler if we transform to eigenstates of this Hamiltonian.
These can be derived in exactly the same way as the Hartree equations in Section 9.2.2, but
now using the Hartree-Fock potential ΣHF (k, m) instead of the Hartree potential VH,km .
The eigenvalue equations, analogous to eq. 9.20, is
X
(H0,km + ΣHF (k, m)) Cmn = ²n Ckn (9.36)
m
272 CHAPTER 9. THE ELECTRON-ELECTRON INTERACTION

or in matrix notation

(H 0 + ΣHF ) cn = ²n cn

where the matrix Cmn defines the Hartree-Fock states |nHF i in terms of the eigenstates
b0
of H
X
|nHF i = |miCmn (9.37)
m

The eigenvalue equations, eq. 9.36 are called he Hartree-Fock equations.

Hartree-Fock self-consistent field equations and orbitals

Defining the Hartree-Fock orbitals as ψ n (r) = hr|nHF i, we can derive, analogous to eq.
9.24, the Hartree-Fock equations in wave function form
" # Z
~2 2 X Ze2
− ∇ − + VH (r) ψ n (r) − d3 r0 VX (r, r0 )ψ n (r0 ) = ²n ψ n (r)
2me |R − r|
R
X Z 0 2
2 3 0 |ψ l (r )|
with VH (r) = e d r
|r − r0 |
|l|≤kF
X ψ (r)ψ ∗ (r0 )
l
and VX (r, r0 ) = e2 l
(9.38)
|r − r0 |
|l|≤kF

using eqs. 9.30, 9.31 and 9.34. Note that the non-local exchange potential leads to an
integral term in this wave equation. The Hartree-Fock equations thus constitute a set of
integro-differential equations which are more complicated (but also more accurate) than
the Hartree equations (eq. 9.24), which are just differential equations. As the latter, the
Hartree-Fock equations have to be solved self-consistently by an iterative procedure to
find the Hartree-Fock orbitals ψ n (r).14

Hartree-Fock propagator

Analogous to eq. 9.26, the Hartree-Fock propagator on the basis set the Hartree-Fock
states has a simple diagonal form

1
GHF (k, n, ω) = δ kn (9.39)
~ω − ²k ± iδ

where ²k are the Hartree-Fock eigenvalues, cf. eq. 9.38. We can now completely copy the
discussion following eq. 9.26 for the Hartree-Fock case. The propagator GHF expressed
in the basis of the Hartree-Fock eigenstates (the solutions of the Hartree-Fock equations)
has the same mathematical form as the unperturbed propagator G0 in the unperturbed
b 0 ). We can transform to the time domain by Fourier transform
basis (the eigenstates of H
14
How this is done in practice should be part of a course in computational (quantum) physics. The
equations look pretty menacing, but numerical solutions can be obtained for not too large a system .
9.3. THE HARTREE-FOCK APPROXIMATION 273

and complex contour integration and find (only the diagonal term is non-zero)

k k
t1 t2

= i~GHF (k, k, t2 − t1 )
i
= Θ(t2 − t1 )Θ(²k − ²F )e− ~ ²k (t2 −t1 )
i
− [1 − Θ(t2 − t1 )] [1 − Θ(²k − ²F )] e− ~ ²k (t2 −t1 ) (9.40)

Diagrams are good for you


Let us summarize the procedure again in diagram form

= + Σ HF

Σ HF = +

Figure 9.10: The Hartee-Fock approximation

The first of these diagram lines corresponds to the Dyson expansion, which, as before,
can be solved in closed form; compare eq. 9.13. We can also solve it using diagrams only.
The algebra goes like

− Σ HF = ⇔

x (I − Σ HF ) = ⇔

= x (I − Σ HF )-1 ⇔

-1 − Σ
= ( HF )-1 (9.41)

This sort of manipulation might seem weird at first sight, but remember that each arrow
represents a propagator G or G0 and the ΣHF represents the Hartree-Fock potential. All
of these quantities depend upon two indices, e.g. k and m, so they are matrices. The
manipulation of diagrams is then only matrix algebra. Going from the third to the last
h i−1
line in eq. 9.41 follows from the general matrix rule A(1 − BA)−1 = (1 − BA)A−1 =
£ −1 ¤−1
A − B . The manipulation of the diagram series is then equivalent to the manipula-
tion of the series of eq. 9.34 using exactly the same tricks as we used in eq. 9.13. Solving
the Hartree-Fock equations, eq. 9.38, is completely equivalent to solving the problem
posed by Fig. 9.10 !!
274 CHAPTER 9. THE ELECTRON-ELECTRON INTERACTION

Dressing the propagator


The second line in eq. 9.40 defines the Hartree-Fock self-energy (or potential) as the
Hartree bubble diagram plus the Fock exchange open oyster diagram. The latter diagrams
contain double lines, as imposed by the “self-consistent field”; i.e. the Hartree-Fock poten-
tial has be calculated on the basis of the Hartree-Fock states, which, following the deriva-
tion above, is achieved when using the Hartree-Fock (hole) propagator. The Hartree-Fock
propagator has a far more eleborate “dress” than the simple Hartree propagator. This
can be seen by expanding the potential Σ in Fig. 9.10 in terms of the “bare” propagator.
It contains terms like

Σ HF = + + + +

+ + .....

Figure 9.11: The self-consistent (dressed) Hartree-Fock potential.

The first two diagrams on the righthand side are the simple “bare” Hartree and Fock
diagrams. The third and fourth diagrams result from inserting an additional Hartree
diagram into a Hartree diagram and a Fock diagram, respectively. The fifth and sixth
diagrams then result from inserting an additional Fock diagram into a Hartree diagram and
a Fock diagram, respectively. One can go on infinitely, creating more and more complicated
higher order diagrams.15 All these terms are included in the “dressed” propagator of Fig.
9.10, which means that by solving the equations represented by this figure in a self-
consistent way (or the Hartree-Fock equations, eq. 9.38) , one indeed includes a large
number of diagrams to infinite order.16

9.3.3 The homogeneous electron gas revisited


Normally the Hartree-Fock equations have to be solved numerically. Analytical work
can be done for the homogeneous electron gas discussed in Section 6.3.1. This give us
important additional physical insight. For the homogeneous electron gas there are no
atomic potentials (eq. 8.75), but the electron-electron interaction (eq. 8.80) is present.
15
The diagrams have been drawn in a way such as to make their connection with the “bare” Hartree
and Fock diagrams most clear. As these are Feynman diagrams, it is only their topology that matters; i.e.
they may be stretched, flipped, or rotated, without altering their numerical value.
16
Note that the fourth order term of Fig. 9.8 and eq. 9.32 is then a relatively simple term of the
propagator which involves only a couple of “bare” Hartree and Fock diagrams.
If you want to have a more exact proof of the statements in this section, have a look at Fetter & Walecka.
9.3. THE HARTREE-FOCK APPROXIMATION 275

The perturbation expansion then becomes much simpler, mainly because the two particle
matrix element Vklmn is much simpler, since the single particle basis states are simple
plane waves
Z Z
e2
Vklmn = d3 r1 d3 r2 φ∗k (r1 )φ∗l (r2 ) φ (r1 )φn (r2 )
|r1 − r2 | m
Z Z
e2 1
= 2
d3 r1 d3 r2 ei(m−k)·r1 ei(n−l)·r2 (9.42)
Ω |r1 − r2 |
(see Section 6.3.1, in particular eq. 6.41). The integral can be done by defining new
integration variables r = r1 − r2 and R = 12 (r1 + r2 ). The result is
1
Vklmn = vk−m δ k+l,m+n

Z
2 1 4πe2
where vq = e d3 r e−iq·r = 2 (9.43)
r q
2
is the (three-dimensional) Fourier transform of the Coulomb potential er . For a more
detailed derivation, see Mattuck p. 135.17 . Given these matrix elements the Hartree-Fock
approximation for the homogeneous electron gas takes on a much simpler form than for
the inhomogeneous case.

The Hartree term


The Hartree potential becomes
X 1 X
VH,km = Vklml = vk−m δ k,m 1 ∝ v0 δ k,m (9.44)

|l|≤kF |l|≤kF

It describes the Coulomb repulsion in of a homogeneous electronic charge distribution,


which is infinite, as one expects of an infinite system. However, this poses no problems,
since it is exactly canceled by adding it to the Coulomb repulsion in the homogeneous com-
pensating positive background charge and subtracting these two terms from the Coulomb
attraction between the homogeneous electronic charge and the background charge. A
more detailed discussion is found in Mattuck, p. 186-187. It makes sense that all these
Coulomb terms added together lead to zero, since the total charge at any point within
a homogeneous distribution is zero.18 . In conclusion, the Hartree terms do not give a
contribution to the energy of the homogeneous electron gas.
17
Note that, compared to Mattuck, I have surpressed the spin variables σ1 etcetera. In my case k
implicitly includes the spin variable. It becomes (k, σ 1 ) in Mattuck’s notation.
18
If it were not zero, it would lead to an infinite Coulomb energy, since the system is infinite (and
homogeneous).

Klaus Fuchs did pioneering work on the physics of electrons embedded in an infinite compensating
background charge. At very low density, the kinetic energy of the electrons becomes negligible as compared
to their Coulomb repulsion and they “crystallize” in a bcc lattice. Such an electron lattice is called a Wigner
lattice. Usually we work at much higher electron densities at which the kinetic energy is non-negligible or
even dominating. The electrons then behave like a gas, albeit a quantum gas (or liquid).

Klaus Fuchs is famous for another reason. Like many others, he participated in the “Manhattan project”
in Los Alamos during the second world war, where the first atomic bombs were constructed. However, he
276 CHAPTER 9. THE ELECTRON-ELECTRON INTERACTION

The Hartree-Fock energies


The exchange potential for the homogeneous electron gas can be calculated along the same
lines
X 1 X
VX,km = Vlkml = δ k,m vl−m (9.45)

|l|≤kF |l|≤kF

We can convert the sum into an integral using eq. 6.44 and use the expression for vq given
by eq. 9.43. The diagonal element is (remember l and k stand for a wave vector)
Z
2e2 1
VX,kk = d3 l (9.46)
(2π)2 l≤kF |l − k|2
where a factor of two is introduced explicitly for the two possible spin states. The integral
is a bit nasty, but it can be done; details can be found in Mattuck p.94-95. The result is
µ ¯ ¯ ¶
e2 kF2 − k 2 ¯¯ kF + k ¯¯
VX,km = δ k,m ln ¯ + kF (9.47)
π 2k kF − k ¯
Apparently the exchange potential is diagonal for the homogeneous electron gas in the
b 0 ! Solving the Hartree-Fock equations, eq.
eigenstates of the unperturbed Hamiltonian H
9.36, now becomes trivial.
X X
(H0,km + ΣHF (k, m)) Cmn = δ k,m (²0,k − VX,kk ) Cmn = ²n Ckn
m m

The solutions are

²k = ²0,k − VX,kk and


Ckn = δ k,n or |kHF i = |ki (9.48)

For the homogeneous electron gas the Hartree-Fock states are identical to the eigenstates
of the unperturbed Hamiltonian ! So, according to eq. 6.41, the Hartree-Fock orbitals are
simply plane waves
1
ψ k (r) = √ eik·r (9.49)

The Hartree-Fock eigenvalues ²k are then given by eqs. 9.48 and 9.47
µ ¯ ¯ ¶
~2 k 2 e2 kF2 − k 2 ¯¯ kF + k ¯¯
²k = − ln ¯ + kF (9.50)
2me π 2k kF − k ¯
It is only for the homogeneous electron gas that the Hartree-Fock equations can be solved
in such a simple way. For the inhomogeneous case, i.e. when atomic potentials are present,
the exchange potential does not have a simple diagonal form; moreover the Hartree term
does not disappear. The Hartree-Fock equations then have to be solved numerically.
was a convinced communist and revealed much of the project’s secrets to the soviets to become the most
notorious scientific spy of all times! The americans did’t catch him; he fled to eastern germany. They
cought a couple of “small fish” instead, the Rosenberg’s, and put them on the electric chair in order to
“defend the interests of the free world”.
9.3. THE HARTREE-FOCK APPROXIMATION 277

-0.4

-0.6

-0.8

π -1
− 2
V X ,k
e
-1.2

-1.4

-1.6

-1.8

-2 k
0 0.2 0.4 0.6 0.8 1 1.2 kF
k

k
Figure 9.12: Exchange energy as function of kF

k
Fig. 9.12 gives a plot of the exchange energy −VX,k ≡ −VX,kk as a function of kF
e2 kF
(in units of π ). Note that it rises steeply at kkF = 1. Its general behaviour can be
understood from eq. 9.46. The wave vectors l that contribute most to the integral are in
a region which is “close” to k. At the same time |l| ≤ kF must hold. If k = |k| increases,
a part of this “close” region is shifted above kF where it cannot contribute to the integral,
so the value of the integral decreases with increasing k. The decrease is highest when k
crosses kF , since then the “closest” region is affected. The decrease is high in absolute
terms, because the integrand has an (integrable) singularity when q ≡ |l − k| = 0.

The size of the exchange contribution in eq. 9.50 to the Hartree-Fock energies depends
upon the Fermi wave vector kF , which depends upon the electron density of the system, cf.
eq. 6.45. Fig. 9.13 shows the Hartree-Fock energies ²k (the energies are in Hartree atomic
2 2
units), compared to the unperturbed eigenvalues ²0,k = ~2mke for kF = 1a−1 0 , which corre-
sponds to an electron density close the average value obtained for the metal aluminium.19
This figuredemonstrates that the exchange energy is an important, and certainly a non-
negligible contribution to the particle energies for metals at ordinary electron densities.

19
Actually, the homogeneous electron gas is not such a bad model for aluminium, since the spatial
variation in electron density is rather small in this metal. The same is true for other simple, i.e. non-
transition metals.
278 CHAPTER 9. THE ELECTRON-ELECTRON INTERACTION

0.5

ε 0,k
k
0 0.2 0.4 0.6 0.8 1 1.2 k F
k

-0.5
εk

-1

Figure 9.13: Hartree-Fock energies ²k and kinetic energies ²0,k for kF = 1a−1
0 .

The Fermi hole


It is possible to give a deeper physical insight into the nature of the exchange term. We
write the exchange potential, eq. 9.31, as
ρd (r1 , r2 )
VX (r1 , r2 ) = e2 with
|r1 − r2 |
X 1 X −il·(r1 −r2 )
ρX (r1 , r2 ) = φ∗l (r1 )φl (r2 ) = e (9.51)

|l|≤kF |l|≤kF

The quantity ρX (r1 , r2 ) is called a “density matrix”. Its “diagonal term” ρX (r1 , r1 ) ≡
ρ(r1 ) corresponds to the normal electron density. As usual, we can convert the sum into
an integral using eq. 6.44; again, the integral is a bit nasty, but it can be done. The result
is
kF2 j1 (kF |r1 − r2 |)
ρX (r1 , r2 ) = = ρX (|r1 − r2 |) (9.52)
2π 2 |r1 − r2 |
where j1 (x) = sin x−x
x2
cos x
is the first-order spherical Bessel function. The function ρX (r1 , r2 )
only depends on |r1 − r2 | ≡ r. It is shown in Fig. 9.14. It can furthermore be shown that
Z
d3 r ρX (r) = 1 (9.53)

The exchange potential VX (r1 , r2 ) = VX (|r1 − r2 |) is obviously also only a function of


|r1 − r2 | ≡ r
ρX (r)
VX (r) = e2 (9.54)
r
9.3. THE HARTREE-FOCK APPROXIMATION 279

0.02

0.015

0.01

0.005

-10 -8 -6 -4 -2 0 2 4 6 8 10

1 j1 (r)
Figure 9.14: ρX (r) = 2π2 r
as function of r

It has the form of a Coulomb potential (or energy), where eρX (r) plays the role of a
charge density. Consider now the Hartree-Fock equations, eq. 9.38, simplified for the
homogeneous electron gas.
Z
~2 e2 ρX (|r − r0 |)
− ∇2 φk (r) − d3 r0 φk (r0 ) = ²k φk (r) (9.55)
2me |r − r0 |

At each possible position of the electron r0 it interacts with a potential VX (|r − r0 |) centered
around this position. This gives an attractive contribution. Since electrons are negative
we can interprete eρX (r) as a positive charge distribution. The intuitive physical picture
for exchange is then as follows. An electron induces a positive charge ρX (r) around itself,
which is called the exchange or Fermi hole (it integrates to 1, cf. eq. 9.53). The exchange
energy can then be interpreted as the interaction of the electron with its own exchange
hole. We derived the “hole” from the exchange term (or diagram); it is called the exchange
hole. Exchange was derived as a consequence of the fact that all Fermion particles are
indistinguishable.20

According to the Pauli exclusion principle, no two fermions can be in the same state
mi . We can make the same argument using the eigenstates of the position operator |ri i
(instead of states |mi i). This means that no two fermions can be at the same position
ri . If one fermion occupies a position ri , other fermions are excluded from that position.
Moreover, since wave functions are in general well-behaved, i.e. continuous, differentiable
etcetera, one expects to find a volume around ri , where the probability of finding another
fermion is diminished. In a more prosaic way, the fermion digs a hole for itself, from
which other fermions are depleted, which is the Fermi hole. The interaction of an electron
with its own hole gives rise to a lowering of its energy. A pictorial representation of the
foregoing discussion is found in Mattuck, p.7-9. He talks about holes being created around
electrons by Coulomb repulsion; you now know that the Pauli exclusion principle is by
itself powerful enough to deplete a region around each electron from other electrons.
20
Note that this is the result of fermion statistics only. The Fermi hole also exist for fermions that have
no Coulomb interaction; in fact, it also exists for fermions that have no interaction al all. The size and
shape of the Fermi hole of course does depend on the interaction between the fermions.
280 CHAPTER 9. THE ELECTRON-ELECTRON INTERACTION

9.3.4 Pro’s and con’s of the Hartree-Fock approximation

”O Oysters,” said the Carpenter,


”You’ve had a pleasant run!
Shall we be trotting home again?’
But answer came there none—
And this was scarcely odd, because
They’d eaten every one. Lewis Caroll, The Walrus and the Carpenter.

In the Hartree-Fock approximation we included all the Hartree diagrams of eq. 9.5
and the exchange (or Fock) diagrams of eq. 9.28. By transforming to a new basis set,
consisting of the Hartree-Fock states, the propagator, eq. 9.39, has a simple independent
particle form again. An electron (or hole) created in a Hartree-Fock state k propagates
unperturbed with energy ²k ; its lifetime in this state is infinite. Besides taking into account
the average electrostatic potential caused by the other particles, the energy ²k also takes
into account the exchange non-local potential. The latter is a direct result of Pauli’s
exclusion principle; the fact that no two electrons can be at the same position means that
each electron creates a “hole” around itself, the exchange or Fermi hole.

The exchange potential is large, so we have incorporated an important physical effect.


Moreover, in the Hartree-Fock approximation the particles are truly indistinguishable and
we have therefore overcome the first flaw of the Hartree approximation, see Section 9.2.3.
The Hartree-Fock approach still has the second flaw however.

2. The lifetime of the Hartree-Fock (quasi-)particle is infinite according to the propaga-


tor of eq. 9.39. As stated in Section 9.2.3, we know that in a real interacting electron
gas an electron (or hole) injected in whatever state will eventually be scattered into
other states. The Hartree-Fock approximation does not account for that. You might
say, so what ? We make approximations in physics all the time and who cares about
lifetimes as long as these are long enough in practice ?

The density of states

There is however more to it. Let us consider the homogeneous electron gas once more and
calculate a quantity called the “density of states” n(²). The latter is defined as n(²)d²
being the “number of states with energies between ² and ² + d²”. For the non-interacting
electron gas calculating n(²) is simple. Consider Fig. 6.4; the number of states with wave
numbers between k and k + dk is given by the volume Υ(k) of the spherical shell between
the radii k and k + dk, divided by the volume V belonging to one k-point.21 . The latter
3
is, according to eq. 6.44, given by V = (2π)Ω . The former is given by the surface of the
sphere s(k) with radius k times its thickness dk. If we put a factor of two in front to

21
Since each state is labeled by k, the number of states is equal to the number of different k’s.
9.3. THE HARTREE-FOCK APPROXIMATION 281

account for the two possible spins we get


Υ(k) Ω Ω
2 = 2 3
s(k)dk = 2 4πk2 dk
V (2π) (2π)3
µ ¶1
Ω 2me dk Ω 2me 1 2me 2 − 1
= 2 ² d² = 2 ² ² 2 d²
2π ~2 d² 2π ~2 2 ~2
µ ¶3
Ω 2me 2 1
= ² 2 d² = n0 (²)d²
2π ~2
where we have used the relation between k and ² given by eq. 6.40. The density of states
n0 (²) for the non-interacting homogeneous electron gas is thus given by
µ ¶3
Ω 2me 2 1
n0 (²) = ²2 (9.56)
2π ~2
It is a perfectly well-behaved density of states with nothing special going on at the Fermi
energy ² = ²F .

Now calculate the density of states of the interacting electron gas in the Hartree-Fock
approximation, where the relation between ² and k is given by eq. 9.50. We find
Υ(k) Ω
2 = 2 s(k)dk
V (2π)3
Ω dk
= 2 3
S(²) d² = nHF (²)d²
(2π) d²
where we defined S(²) = s(k(²)) = 4πk(²)2 . The density of states is
Ω S(²)
nHF (²) = 2 (9.57)
(2π)3 ²0 (k)

whith ²0 (k) = dk . Both ²0 (k) and S(²) can be easily calculated from eq. 9.50. S(²) is
a well-behaved function, but ²0 (k) is not ! Fig. 9.15 shows ²0 (k) for the Hartree-Fock
approximation. At small k the function ²0 (k) is almost linear, like in the non-interacting
electron gas, where ²00 (k) ∝ k. But close to k = kF the function rises steeply and at
k = kF it has a spike which actually goes off to infinity. So ²0 (kF ) = ∞. The reason for
this spike is the same as for the “steep rise” which we discussed for fig. 9.12, namely the
q ≡ |l − k| = 0 singularity of the integrand of eq. 9.46.
This results in
nHF (²F ) = 0 (9.58)
which is a weird, unphysical result ! Remember, the homogeneous electron gas is supposed
to be a model for simple metals. Almost all properties which are characteristic of metallic
behavior (conduction, reflection of light at the surface, etcetera) are caused by electrons
with energies near the Fermi energy. In fact, in most introductory solid state physics
books, the electrons are treated as non-interacting (the “free electron model”), with a
density of states given by eq. 9.54. This naive model does a reasonable, though not
perfect job. Yet the first thing we do to improve the model, i.e. introduce the exchange
interaction via the Hartree-Fock approach, leads to a zero density of states at the Fermi
energy. In other words it does not even give a metal, but an insulator !
282 CHAPTER 9. THE ELECTRON-ELECTRON INTERACTION

7 dε k
6 dk

1
k
0 0.2 0.4 0.6 0.8 1 1.2 kF


Figure 9.15: ²0 (k) = dk in the Hartree-Fock approximation.

9.3.5 Screening
This artefact produced by Hartree-Fock has given the method a very bad name in solid
state physics. It does not deserve this name, since there are other examples where Hartree-
Fock works quite well. The function shown in Fig. 9.15 has a sharp spike at the Fermi wave
number, but at other values it is reasonably well-behaved. Semiconductors, insulators and
finite systems like molecules have, per definition, an energy gap around the Fermi energy.
Thus they have a zero density of states at and around the Fermi energy (called more
properly the chemical potential in those cases). So there the Hartree-Fock artefact, i.e.
the spike at the Fermi energy, is not dramatic.
Unfortunately for metals it is the behavior at and close to the Fermi energy that counts.
Here we have a problem with Hartree-Fock. One of the origins of the problem can be seen
in the Hartree-Fock equations, cf. 9.38, where the potential always contains a factor
|r − r0 |−1 . Its Fourier transform, 4πq −2 , eq. 9.43, leads to the spike in the homogeneous
electron gas due to its singularity at q = 0. The factor |r − r0 |−1 is typical of a Coulomb
interaction in free space. Inside a material however, such interactions are screened. If we
put a charge e in a material at a position r and probe the potential with a test charge e at
r0 the interaction will be much weaker than e2 |r−r0 |−1 , since the charges will polarize their
environment. These induced charges will be have a sign opposite to the original charges
e; they cause a potential which counteracts that of the original charges. In other words
the induced charges will screen the original charges. Read Mattuck’s chapter 0 again on
this point.

In a semiconductor or insulator one would expect the interaction to be more like


(ε|r−r0 |)−1 where ε is the dielectric constant of the material. In an ordinary semiconductor
like silicon, ε ≈ 12, so screening can be a large effect. For a metal the story is slightly
more complicated since the static long range ε = ∞. However, close to the charge, on the
microscopic (nanometer) scale, there is less or no material to polarize and we expect ε to
decrease. On a microscopic scale we expect the dielectric response ε(r, r0 ) to be position
dependent. More specifically, in a metal we expect lim|r−r0 |→0 ε(r, r0 ) = 1 (no screening)
and lim|r−r0 |→∞ ε(r, r0 ) = ∞ (infinite screening). The screened Coulomb interaction in a
9.4. THE RANDOM PHASE APPROXIMATION (RPA) 283

metal, which contains a ε(r, r0 )−1 like factor, should therefore be confined to a region of
r0 that is close to r. In other words, the screened Coulomb interaction in a metal has a
finite range. It leads to much weaker interaction between the electrons as discussed in the
next section. Mathematically spoken, it also gives a smooth Fourier transform, which does
not have a singularity at q = 0. Therefore we expect screening to lift the Hartree-Fock
artefact discussed above.

Moreover, in our propagator “experiment” (or quantum pinball game if you like) we
are not dealing with static charges, but dynamic ones, since we are shooting an extra
particle into our system. The material must have time to respond, and its response to
fast particles will in general be different from its response to slow ones. In other words,
we expect a frequency dependent screening ε(r, r0 , ω), also called the dielectric response
function, to play the central role. The slowness of response of the system to an incoming
particle eventually leads to scattering of the particle and thus a finite lifetime.22 Screening
is the subject of the next section.

9.4 The Random Phase Approximation (RPA)


“What’s in a name? That which we call a rose, by any other name would smell as sweet.”, William
Shakespeare, Romeo and Juliet.

As discussed in the previous section, to get rid of the artefacts produced by the Hartree-
Fock approximation, which are due to the long range of the Coulomb interaction, one has
to introduce a dielectric screening of that interaction. Screening the long range part of
the Coulomb interaction turns the strongly interacting electron gas into a gas of screened
electrons which interact via a much weaker effective short-range potential. The screened
electrons are quasi-particles; one expects that the screening charge that surrounds an
electron cannot follow the electron instantaneously. This leads to scattering and a finite
lifetime of the electron.23 The simplest approximation that takes these physical effects into
account is called the random phase approximation or RPA. This approximation is named
after the french-american couple Jean-Marie Random and An Y. Phase. No, just kidding,
the reasons are purely historical; in the usual formulation there is not a phase in sight, let
alone a random phase. The name unfortunately has been stuck in many particle physics,
so we have no option but to use it.24 In this section we use the diagram approach to
introduce RPA. Application to the homogeneous electron gas illustrates that the physical
effects discussed above are indeed included.
22
Although this is a classical picture and thus not valid, one could say that the material can not give
way to a fast electron fast enough. The electron (and its Fermi hole) must be dragged through an electron
gas which has a certain “viscosity” (although this is again not the right word, it is the slowness of dielectric
response that counts). This gives rise to “losses”, or a decay of the probability of the electron to remain
in its original state, which we describe by a finite lifetime. On a quantum level, these “losses” can be
described as scattering to other states.
23
or rather, of the state the electron occupies. The electron itsself of course does not dissappear; it is
scattered into other states.
24
The name RPA was invented by D. Bohm and D. Pines, see D. Pines, Elementary Excitations in
Solids, (Perseus, Reading, 1999). The diagrammatic approach was introduced by M Gell-Mann69 and K.
A. Brueckner.
284 CHAPTER 9. THE ELECTRON-ELECTRON INTERACTION

9.4.1 The RPA diagram


From the diagrammatic point of view the basic idea is simple; we add the following non-
trivial diagram. The Hartree and the exchange diagrams of eqs. 9.5 and 9.28 are the only
first order diagrams that give a non-zero contribution. They can be linked to yield the
Hartree-Fock approximation for the single particle propagator, see Fig. 9.10.To improve
our approximation, we have to include second order diagrams. Most of these are already
included when doing Hartree-Fock self-consistently, as Fig. 9.11 illustrates. The first
second order diagram not included in (self-consistent) Hartree-Fock is

m τ1 p τ2 k
t1 t2

l
XZ 1
dτ 1 dτ 2 i~G0 (k, k, t2 − τ 2 ) Vknpl i~G0 (l, l, τ 2 − τ 1 ) i~G0 (n, n, τ 2 − τ 1 )
i~
l,n,p
1
i~G0 (p, p, τ 2 − τ 1 ) Vplmn i~G0 (m, m, τ 1 − t1 ) (9.59)
i~
It is called the RPA diagram, or the (electron-hole) pair bubble diagram. The trnslation
between diagrams and propagators G0 and interactions Vklmn can be found in Mattuck’s
table 4.3 or 9.1. We now stick to the Feynman convention that G0 represents either G+ 0
or G− 0 depending upon the time argument (i.e. the arrows can represent electrons or
holes), and we sort out the ± details later (this also determine whether to take |l| ≤ kF ,
or |l| > kF , etcetera). The pair bubble diagram is formed by linking up two first order
diagrams as in the figure below.

The physical picture behind this diagram is as follows. An incoming electron in state m
creates an electron-hole pair in states l and n and is scattered into state p at time t = τ 1 .
At time t = τ 2 the electron-hole pair recombines and the electron is scattered into the
outgoing state k. As always, one has to sum over all possibilities, i.e. one integrates over
all intermediate times and sums over all possible intermediate states.

Topology of Feynman diagrams


To explain the topology once more within the Feynman convention, let us consider the
arrows as rubber bands (with a direction), and the dotted lines as sticks. It is possible
to move the sticks around and stretch the bands in any direction, but do not detach then
from the sticks.25 Interchanging the two sticks in eq. 9.59 for instance changes the time
25
A point where the arrows (rubber bands) and the interaction lines (sticks) meet is called a vertex
(plural: vertices).
9.4. THE RANDOM PHASE APPROXIMATION (RPA) 285

order of τ 1 and τ 2 ; the direction of the arrows (rubber bands) connecting the two sticks
changes with them

(9.60)

If in the diagram of eq. 9.59 we attached a G+ 0 (p, p, τ 2 − τ 1 ) to the line labeled with p,
then for the diagram above we must use a G− 0 (p, p, τ 2 − τ 1 ) for this line. The nice thing
about Feynman diagrams is that we do not need to worry about the time order. We simply
integrate over all time orders; the ± on the propagators G0 will pop up automatically. In
the Feynman convention, the diagram of eq. 9.60 is thus included in that of eq. 9.59. The
disadvantage is that the diagrams are not “time-ordered”, i.e. all the time labels τ 1 , τ 2
can have any value. A couple of restrictions on direction can be placed though. Since τ 2
must lie either to the right or to the left of τ 1 , the pair bubble must have an electron line
and a hole line if we want to close the diagram the way we did. So G± 0 (l, l, τ 2 − τ 1 ) must
be combined with G∓ 0 (n, n, τ 2 − τ 1 ); compare the diagrams of eq. 9.59 and 9.60. More
detail on diagram topology of can be found in Mattuck’s §9.5, p. 160. If you still don’t
believe it, you can do the algebra, see e.g. Mattuck’s §9.8, p. 170. Note that, despite the
fact that a single Feynman diagram represents a number of possible time-orders, we still
talk about it like “a particle comes in, then a pair is created, etcetera”. It is just easier to
talk this way, just don’t take it too literal.

Self-consistent field equations


We know how to proceed now; we link up the RPA diagram together with the Hartree-
Fock diagrams to produce higher order diagrams. An example of an eight order diagram
(eight dotted lines means eight interactions) is given in Fig. 9.16.

Figure 9.16: An eighth order RPA + Hartree-Fock diagram.

We follow the lines set out in Section 9.3 and derive a formal solution of the propagator
in terms of such diagrams, as in Fig. 9.10. The result is shown in Fig. 9.17
The first line again corresponds to the Dyson expansion. The second line defines
the self-energy; it contains the Hartree (bubble) diagram plus the Fock (open oyster)
286 CHAPTER 9. THE ELECTRON-ELECTRON INTERACTION

= + Σ1, RPA

Σ1, RPA = + +

Figure 9.17: The RPA approximation

diagram plus the RPA (pair bubble) diagram. The latter diagrams contain double lines,
which means that the equations represented by these diagrams must be solved in a “self-
consistent” way. By expanding the diagrams by hand, fancy diagrams like

appear in the self-energy, as well as the diagrams on the cover of Mattuck’s book, for
instance. The self-energy Σ1,RP A is not a simple potential anymore, not even a non-local
one like the exchange potential. Besides an incoming state m and outgoing state k it
also depends upon a frequency ω, i.e. Σ1,RP A = Σ1,RP A (k, m,ω). The reason for this can
be found in eq. 9.59. In the time domain the two interactions in this diagram occur at
two different times τ 1 and τ 2 , and the time difference τ 2 − τ 1 appears in the propagators
G0 . This time difference translates into a frequency-dependence when the expression
is Fourier transformed to the frequency domain. Moreover, one finds that Σ1,RP A is a
complex quantity, i.e. it has an imaginary, as well as a real part. As before, the imaginary
part gives rise to a finite lifetime.
9.4. THE RANDOM PHASE APPROXIMATION (RPA) 287

9.4.2 The RPA screened interaction


We can do even better and also include higher order diagrams of the type

and

(9.61)

almost without extra trouble. The physical picture behind these diagram is similar to that
behind eq. 9.59. The incoming electron creates an electron-hole (eh) pair. In the diagrams
above, the eh-pair recombines and creates a new eh-pair. This process occurs a couple
of times before the final recombination sends out the electron in its final state. At any
single time only one eh-pair is present, which recombines before a new one is formed. One
can prove, that these diagrams and the processes they represent are the dominant ones
at high electron densities. The proof is somewhat subtle and can be found in Mattuck’s
§10.4, p.185. Diagrams of the type given in eq. 9.59 can be summed to infinite order.
We will first work through the algebra in diagrammatic form and show that it leads quite
naturally to the concept of “screened interaction”. The trick is to look at the interaction
lines and consider the series

= + + + ....

(9.62)

Attaching arrows to the top and bottom of each of these diagrams turns them into diagrams
which are valid for the self-energy or the propagator. For instance, connecting the top and
bottom of the second diagram on the righthand side gives

=
288 CHAPTER 9. THE ELECTRON-ELECTRON INTERACTION

a diagram which is part of the self-energy Σ1,RP A , see Fig. 9.17. Attaching incoming and
outgoing lines then gives a diagram which is part of the propagator.

The higher order diagrams such as the rightmost one in eq. 9.62 add something new.
These are not included in the self-energy Σ1,RP A of Fig. 9.17, as one can easily see if one
expands Σ1,RP A into diagrams. The idea is to write the series given in eq. 9.62 in iterative
form

= +

(9.63)

Substitution of the double dotted line at the lefthand side by the two terms on the lefthand
side will produce all the diagrams of eq. 9.62. This diagram equation formally resembles
a Dyson equation; compare to the top lines of Figs. 9.2, 9.10 or 9.17. However, the roles
of the interaction (potential) and the propagator lines (the arrows) are interchanged here.
Nevertheless, we can formally solve this diagram equation just like eq. 9.41.

− = ⇔

x (I − ) = ⇔

= x (I − )-1 ⇔

(9.64)
-1
= ( − )-1
(9.65)

Although this might look a little weird about this; all manipulations are just matrix
algebra of interactions or propagators. All the symbols on the righthand side of the last
line can be calculated; they consist of the “bare” electron-electron repulsion (the dotted
line) and the propagators making up the eh-pair bubble. So eq. 9.65 defines a new kind
of interaction (the double dotted line) and gives a closed expression for it. We call it the
9.4. THE RANDOM PHASE APPROXIMATION (RPA) 289

(RPA) screened interaction. It is normally given the symbol W ; mathematically it behaves


just like the “bare” electron-electron repulsion V . In other words it is associated with a
two-particle operator and it has matrix elements Wklmn . Why it has the name “screened
interaction” is best seen from the generic two-particle interaction diagram, analogous to
eq. 8.84 (in the frequency domain)

m k

n l (9.66)
−1
≡ (i~) 2
G+ +
0 (k, k, ω k )G0 (l, l, ω l ) (i~) Wklmn (ω)(i~)2 G+ +
0 (m, m, ω m )G0 (n, n, ω n )

The two incoming electrons in states m and n do not interact via their bare 1r -like inter-
action V ; see eqs. 8.80 and 8.81. Instead, substituting the series of eq. 9.62, we see that
there can be any number of temporary electron-hole pairs in between these two incoming
electrons. These electron-hole pairs screen the interaction. Think of them as being created
in the polarizable medium to shield the harsh 1r e potential. A further difference with the
“bare” diagram of eq. 8.84 is that the screened interaction W depends upon a frequency
ω (whereas the bare interaction obviously does not). The reason for this can be seen from
eq. 9.62. In the time domain each diagram represents a sequence of events; for instance, at
t = τ 1 an eh-pair is formed, and at t = τ 2 6= τ 1 it recombines again. Fourier transforming
to the frequency domain, such time differences transform into a frequency. The formation
of eh-pairs during the interaction represents the response of the system to the incoming
electrons. Obviously this response cannot be instantaneously; there has to be some delay.
It thus seems logical that the response of a system depends upon a frequency ω.

CLOSING REMARK
The time or frequency dependence of the screened interaction means that we should ac-
tually be drawing the diagram of eq. 9.66 slightly “slanted” like the relativistic diagram
of eq. 8.93 in Section 8.4.4. Unfortunately, nobody in solid state physics does so, but re-
member double dotted lines are not instantaneous interactions; there is a time difference
between the top and the bottom of the line. As mentioned in Section 8.4.4, in the rela-
tivistic limit the interaction between two electrons is mediated by “longitudinal photons”.
From the foregoing discussion it is clear that in a condensed medium the interaction is also
mediated by eh-pairs, cf. eqs. 9.62 and 9.66. The electron and hole have a half-integer
spin, so the eh-pair has an integer spin. In other words the eh-pair can be considered
as a boson.26 The diagram of eq. 9.66 is extremely generic; it holds many systems that
consist of interacting fermion particles. The interaction is mediated by boson particles.
Other examples from condensed matter physics are interactions between electrons medi-
ated by phonons; these result in quasi-particles called polarons, for instance. Under certain
26
Actually a second type of boson particle is also involved in the screened interaction. It has to do with
plasma oscillations in the electron gas and is called a plasmon. It is actually included in the eh-pair bubble
diagrams, but it is somewhat hidden. If you want to know something about plasmons, see for instance the
book by Pines cited in a previous footnote.
290 CHAPTER 9. THE ELECTRON-ELECTRON INTERACTION

= + Σ RPA

Σ RPA = +

Figure 9.18: The RPA or GW approximation

conditions phonons can mediate an interaction between electrons which is attractive and
stronger than the (screened) Coulomd repulsion. It leads to Cooper72 pairs which are the
origin of superconductivity. In high energy physics there are numerous examples of this
scheme, fermions interacting via bosons.

9.4.3 The GW approximation


We go back to our first try at formulating the RPA approximation propagator, shown in
Fig. 9.17. We can improve this approximation by including all diagrams of eq. 9.61. As we
have seen in the previous section, all we have to do is to replace the bare electron-electron
interaction V by the screened interaction W . In diagrammatic form this is given in Fig.
9.18.
This, finally, is commonly known as the RPA approximation. As before, the first line
represents the Dyson equation for the propagator G of the interacting electron system in
terms of the self-energy Σ. The second line defines the self-energy in the RPA approxi-
mation. It contains double line arrows, which means that it is defined in terms of G and
not in terms of G0 . In other words, it contains the “dressed” propagator and not the
“bare” propagator. All this means is that the equations represented by the first and sec-
ond line have to be solved self-consistently. The double dotted line represents the screened
interaction W . Note that the RPA approximation formally resembles the Hartree-Fock
approximation, see Fig. 9.10, the only difference being the use of W instead of V . In
the fancy language of diagrams we say that we have “clothed” the interaction and use a
“dressed” instead of a “bare” interaction.27
Note that we have not written down the eh-pair bubble diagram of Fig. 9.17 (the third
diagram on the second line) explicitly. This is actually included in the exchange diagram
of Fig. 9.18 (the second diagram on the second line). The way to see this is to expand
the double dotted line in the latter diagram according to eq. 9.62; the second diagram on
the righthand side then gives the eh-pair bubble diagram of Fig. 9.17.28 One can prove
that when expanding the double dotted line in the Hartree diagram of Fig. 9.18 (the first
diagram on the second line) that only the first term on the righthand side of eq. 9.62 gives
a contribution. This is easily seen from Fig. 9.4, where one observes that expanding the
27
The “dressed” interaction is of course nothing else than the screened interaction W .
28
Apart from the double line arrow instead of a single line. But this is just using G instead of G0 , as
mentioned, or applying self-consistency.
9.4. THE RANDOM PHASE APPROXIMATION (RPA) 291

propagator already includes all eh-pair bubbles in the Hartree diagram, and expanding
the interaction would only lead to double counting of the same terms.29 So all the tricks
of the RPA approximation are contained in the exchange diagram of Fig. 9.18.

The GW self-energy
Let us write out this exchange diagram; this is most easily done in the time domain.

m τ1

τ2 k
ΣX,RP A (k, m,τ 2 − τ 1 ) =
X 1
= i~GRP A (l, l,τ 2 − τ 1 ) Wlkml (τ 2 − τ 1 ) (9.67)
i~
l

Remember there’s a time difference τ 2 − τ 1 between the top and bottom of the double
dotted line, because if we expand, all the eh-pair bubbles have to fit in (or in more
physical terms: the creation/annihilation of the eh-pairs represents the response of the
system and the system needs time to respond). Because of this time difference ΣX,RP A
does not become a “nice” potential like the exchange potential VX of the Hartree-Fock
approximation, cf. eqs. 9.29 and 9.30. In the frequency domain it becomes even more
complicated, since a product in time space becomes a convolution in frequency space
X Z dω0
ΣX,RP A (k, m,ω) = GRP A (l, l,ω − ω0 )Wlkml (ω 0 ) (9.68)

l

Be as it may, the most important term in the RPA approximation can be written symboli-
cally as ΣX,RP A = GW ; hence the RPA approximation is also called the GW approximation.
Apart from the frequency dependence, the rest of the story is similar to Hartree-Fock as
discussed in Section 9.3.2. The Dyson expansion reads

GRP A (k, m, ω) = G0 (k, k, ω)δ km + G0 (k, k, ω)ΣRP A (k, m,ω)G0 (m, m, ω)


X
+ G0 (k, k, ω)ΣRP A (k, n,ω)G0 (n, n, ω)ΣRP A (n, m,ω)G0 (m, m, ω) + ....
n
where ΣRP A (k, m,ω) = VH,km + ΣX,RP A (k, m,ω) (9.69)

Defining the matrices (GRP A (ω))km = GRP A (k, m, ω) and (ΣRP A (ω))km = ΣRP A (k, m,ω)
we can “solve” this equation as before to get

GRP A (ω) = (~ω − H 0 − ΣRP A (ω) ± iδ)−1 (9.70)

This shows that the propagator GRP A in the RPA approximation is the Green function
(or Green matrix) associated with the “Hamiltonian” H 0 + ΣRP A (ω). We transform to
eigenstates of this Hamiltonian analogous to eqs. 9.36 and 9.37.
29
To avoid a double counting of the same diagrams is actually the most difficult bit of the “clothing”
game.
292 CHAPTER 9. THE ELECTRON-ELECTRON INTERACTION

The GW self-consistent field equations


As a result, one derives the GW (self consistent field) equations, which in wave function
form read like30
" # Z
~2 2 X Ze2
− ∇ − + VH (r) ψ n (r,ω) + d3 r0 ΣX,RP A (r, r0 , ω)ψ n (r0 , ω) = ²n (ω)ψ n (r,ω)
2me |R − r|
R
(9.71)

where, similar to the Hartree-Fock case the self-energy ΣX,RP A (r, r0 , ω) in the position
representation is defined by
Z Z
ΣX,RP A (k, m, ω) = d3 r d3 r0 ψ ∗k (r,ω)ΣX,RP A (r, r0 , ω)ψ m (r0 ,ω) ⇔
X
ΣX,RP A (r, r0 , ω) = ψ k (r,ω)ΣX,RP A (k, m, ω)ψ ∗m (r0 ,ω) (9.72)
k,m

Note how the ω-dependence of ΣX,RP A adds an ω-dependence to the GW -“orbitals”


ψ n (r,ω) and their “energies” ²n (ω). Most of this frequency-dependence is boring and
not of interest. Let me try to explain what I mean. Analogous to eq. 9.39, the propagator
in the RPA approximation on the basis of the states which are the solution of eq. 9.71,
has a simple diagonal form
1
GRP A (k, n, ω) = δ kn (9.73)
~ω − ²k (ω) ± iδ

where ²k (ω) are the eigenvalues of eq. 9.71. As always, in the end the propagator in the
time domain is the quantity that we want. Remember, it gives the probability amplitude
that a particle which is created in a state n at some initial time, is found in a state k at
some later time. To get to the time domain we have to Fourier transform eq. 9.73, like we
discussed in the previous chapter and in Section 4; see in particular Section 4.4 and the
discussion following eq. 4.35. We know from this discussion that the poles of the function
GRP A (k, n, ω) determine the behaviour of its Fourier transform. At these poles we have

~ω − ²k (ω) = 0 (9.74)

We now assume that the functions ²k (ω) are simple and well-behaved (monotonic, for
instance). Then eq. 9.74 has only one solution for each ²k ; call this solution ω = ω k .
This assumption is called the quasi-particle ansatz . It can be proven in some cases.
In any case one expects it to be true if the ω-dependence of ΣX,RP A is rather weak.31
This is “normally” the case, and systems for which it is true are called normal Fermi
liquids (or gases). Exceptions are called “strongly correlated systems”. So for a “nor-
mal” system only the points ω = ω k are of interest. The functions ψ k (r,ω k ) are then
called the quasi-particle wave functions or orbitals and the energies ²k (ωk ) are called the
quasi-particle eigenvalues (eq. 9.71 is then called the quasi-particle wave equation). In
general, the self-energy ΣX,RP A is complex (it has an imaginary, as well as a real part);
30
as far as I am concerned you may also call them the RPA equations, since it’s the same thing.
31
If ΣX does not depend on ω at all we are back at the Hartree-Fock case again, where ²k is also
independent of ω, and for sure eq. 9.74 has only one solution.
9.4. THE RANDOM PHASE APPROXIMATION (RPA) 293

we will work this out in somewhat more detail in the next section. This means that the
²k (ω k ) become complex quantities as well. As before, upon Fourier transforming eq. 9.73,
the real part Re ²k (ω k ) determines the energy of the quasi-particle, and the imaginary
part Im ²k (ω k ) determines its lifetime.

In conclusion, quasi-particle wave functions and eigenvalues for normal systems can be
obtained (in the RPA approximation) by solving eq. 9.71 for ωk , which are the solution
of eq. 9.74. This has to be done in a self-consistent way, using the self-energy determined
by eq. 9.67, 9.72 and the screened interaction W given by eq. 9.65. Not surprisingly, in
the general case solving this problem involves some heavy numerical calculations. Also
no surprisingly, for the homogeneous electron gas things become simpler and we can do
some analytical work. This is discussed in the next section. There we also work out the
screened interaction W in somewhat more detail, which so far we have only discussed in
diagrammatic form, cf. eq. 9.65.

9.4.4 The homogeneous electron gas re-revisited


As you will have gathered by now, the homogeneous electron gas is one of the favorite toy
models in many-particle physics. As mentioned in our previous visitation, the main reason
for the popularity of this model is the simplicity of two-particle matrix elements, cf. eq.
9.45. Defining q = k − m means we can write m = k − q and, because of the Kronecker
δ, n = l + q, so the non-zero matrix elements are of the form
1
Vklmn = Vk,l,k−q,l+q = vq ≡ Ṽq (9.75)

Attaching propagators means we can relabel the generic diagram of eq. 8.84 as

m = k −q k

n = l+q l
(9.76)

Since in the homogeneous electron gas the one-particle states are simple plane waves,
cf. eq. 9.49, the labels k, l etcetera correspond to the momenta of the particles. The
diagram then tells us that the sum of the momenta of the incoming particles m + n is
equal to the sum of the momenta of the outgoing particles k + l. The diagram thus
expresses the conservation of momentum. Adding the arrow labeled q explicitly gives the
momentum which is transferred from the lower to the upper particle. The diagram can be
interpreted in an intuitive way as a “collision” between two particles where momentum is
transferred from one to the other particle.32 Adding a bit of phantasy, on may also read the
diagram as an “electronic circuit” with one of Kirchhoff’s laws at work; at each node the
32
Again don’t take this too literally. Whereas in principle it is possible to set up an experiment to
measure such a collision, more often the diagram is used as part of a perturbation expansion (of the
propagator, for instance). It is then a “virtual” collision, which is not measured as an isolated event,
but it does contribute to a possible route the system can take. In the Feynman sense, all possible routes
contribute to the final probability of a process.
294 CHAPTER 9. THE ELECTRON-ELECTRON INTERACTION

total incoming current (read: momentum)


P has to be equal to the total outgoing current
(momentum). So at each node momenta = 0, where the arrows give the direction
of momentum flow. This only works for the homogeneous electron gas because in an
inhomogeneous gas the momentum is not a good quantum number, i.e. the eigenstates
of the GW equations or the Hartree-Fock equations, or whatever approximation you care
to use, are not eigenstates of the single particle momentum operator like they are in the
homogeneous electron gas. In an inhomogeneous gas the fixed atomic nuclei (which give
the atomic potentials that cause the inhomogeneities) absorb momentum.

The Polarization Propagator


For the homogeneous electron gas, the diagram of eq. 9.59 becomes

k τ1 k −q τ2 k
t1 t2
q q
l

l+q
XZ 1
dτ 1 dτ 2 i~G+
0 (k, k, t2 − τ 2 ) Ṽq i~G0 (l, l, τ 2 − τ 1 ) i~G0 (l + q, l + q, τ 2 − τ 1 )
i~
l,q
1
i~G0 (k − q, k − q, τ 2 − τ 1 ) Ṽ−q i~G+
0 (k, k, τ 1 − t1 ) (9.77)
i~
Note that the momentum k of the incoming and the outgoing lines are the same. This
is a consequence of the conservation of momentum; try to make a diagram with different
incoming and outgoing momentum and you will find that it not possible then to satisfy
momentum conservation. We transform this diagram into the frequency domain and
proceed in two steps. As we have seen in Section 9.4.2 a central role is played by the
eh-pair bubble

l
τ1 τ2
l+q
X
= i~Π0 (q,τ 2 − τ 1 ) = i~G0 (l, l, τ 2 − τ 1 ) i~G0 (l + q, l + q, τ 2 − τ 1 ) (9.78)
l

It is called the polarization propagator . The name is not that strange, since we have seen in
Section 9.4.2 that the eh-pairs represent the response, i.e. the polarization of the medium
and they screen the Coulomb field in its “propagation” through the medium. The Fourier
transform of eq. 9.78 is
Z
1 X ∞
−i~Π0 (q,ω) = dω 0 G0 (l, l, ω 0 )G0 (l + q, l + q, ω + ω 0 ) (9.79)
2π −∞ l

as you can easily check yourself. Note that a product in the time domain becomes a
convolution in the frequency domain, as usual. We have all the information to work out
9.4. THE RANDOM PHASE APPROXIMATION (RPA) 295

eq. 9.79 analytically. The ingredients are


1

0 (k, k, ω) = (9.80)
~ω − ²0,k ± iδ
~2 |k|2
with ²0,k =
2me
Replacing the sum by an integral
X Z

→ d3 l (9.81)
(2π)2
l

the integrals can be done (using complex contour integration and trying out all “+” and
“−” combinations on the G0 ’s). However they are a bit nasty; you can find the details in
Mattuck’s §9.8, p.170 and §10.7, p.197. A partial result which we will use later on is
me kF
Π0 (q,ω = 0) = for |q| ¿ kF (9.82)
π
Using the definition of the polarization propagator in eq. 9.77, we can have a look at the
next time-dependent factor

A(τ 2 − τ 1 ) = i~G0 (k − q, k − q, τ 2 − τ 1 )
i~G0 (l, l, τ 2 − τ 1 ) i~G0 (l + q, l + q, τ 2 − τ 1 )
= i~G0 (k − q, k − q, τ 2 − τ 1 )i~Π0 (q,τ 2 − τ 1 )

The Fourier transform of the latter is


Z ∞
1
A(ω) = dω 00 G0 (k − q, k − q, ω − ω 00 )Π0 (q,ω00 ) (9.83)
2π −∞
With this definition we can write eq. 9.77 as
XZ 1 ¯¯ ¯¯2
dτ 1 dτ 2 i~G+
0 (k, k, t2 − τ 2 ) A(τ 2 − τ 1 ) ¯Ṽq ¯ i~G+
0 (k, k, τ 1 − t1 )
q
(i~)2

using V−q = Vq∗ . This expression is a (double) convolution in the time domain. Thus its
Fourier transform becomes a simple product in the frequency domain
X ¯ ¯2
¯ ¯
G+0 (k, k, ω) A(ω) ¯Ṽq ¯ G+
0 (k, k, ω) (9.84)
q

Using eqs. 9.84, 9.83 and 9.79, we can label the diagram of eq. 9.77 in the frequency
domain.
k,ω k − q, ω − ω ' ' k,ω

q,ω ' ' q,ω ' '


l,ω '

l + q, ω '+ω ' ' (9.85)


296 CHAPTER 9. THE ELECTRON-ELECTRON INTERACTION

P
We also have a conservation of frequency, which means that at each node the frequecies
= 0 . Again the arrows indicate the direction of flow; incoming frequencies are counted
as plus, and outgoning frequencies are counted as minus. Unlike the conservation of mo-
menta, the conservation of frequencies also holds for the inhomogeneous electron gas.
Conservation of frequencies is simply a direct consequence of Fourier transforming, so it
is very general !

The dielectric function and the screened interaction


Using the momentum/frequency labeling of diagrams, the diagram algebra we performed
in Section 9.4.2 can now be turned into real algebra. As an example, the double eh-pair
bubble diagram can be labeled as

k,ω k − q, ω − ω ' ' k,ω

q,ω ' ' l,ω '


q,ω ' '
q,ω ' '
l + q, ω '+ω ' ' l,ω '

l + q, ω '+ω ' ' (9.86)

All the diagrams of eqs. 9.61 and 9.62 can thus be formed by linking identical subunits

q,ω ' ' l,ω '

q,ω ' '


l + q, ω '+ω ' '

1 1
= Ṽq (−i~)Π0 (q,ω) Ṽq (9.87)
i~ i~
using eq. 9.79. Eq. 9.62 can then be written in algebraic form as

Wq (ω) = Ṽq + Ṽq (−Π0 (q,ω))Ṽq + Ṽq (−Π0 (q,ω))Ṽq (−Π0 (q,ω))Ṽq + ... (9.88)

This series corresponds to the expansion of eq. 9.63. The terms can be summed to infinite
order and eq. 9.64 then becomes
h i−1
Wq (ω) = Ṽq 1 + Π0 (q,ω)Ṽq (9.89)

This is an expression for the screened interaction Wq (ω) in the homogeneous electron
gas in the RPA approximation. It is caused by the frequency dependent response of the
medium, as represented by the polarization propagator Π0 (q,ω). Carrying the physical
interpretation one step further, we define a dielectric function by

ε(q,ω) = 1 + Π0 (q,ω)Ṽq (9.90)


9.4. THE RANDOM PHASE APPROXIMATION (RPA) 297

so the screened interaction becomes


Ṽq
Wq (ω) = (9.91)
ε(q,ω)
Inside a medium one does not experience the bare Coulomb potential Vq but one that is
dielectrically screened by ε(q,ω). Fourier transforming eq. 9.91 to real space and the time
domain on gets
Z Z
iωt
W (r,t) ∝ dωe d3 qeiq·r Wq (ω)
Z Z
= ε−1 (r − r0 ,t − t0 )V (r0 )d3 r0 dt0
Z Z
e2
= ε−1 (r − r0 ,t − t0 ) 0 d3 r0 dt0 (9.92)
r
Note again that a simple product in momentum/frequency space like eq. 9.91 becomes
a convolution in real/time space. Screening is a non-local, time-dependent effect, as you
undoubtedly know from your electrodynamics course.

Simple screening for metals and isolators


If the perturbing potential does not vary tto much over microscopic distances, and its
frequency is small compared to the natural response frequencies of the system, we may
approximate the dielectric function by a constant ε(q,ω) ≈ εe = ε(q = 0,ω = 0), which is
called the (static) dielectric constant. Such an approximation is not sufficiently accurate
for microscopic potentials, but we can use the idea to get a first look and feel for the physics
involved. In case of metallic screening εe = ∞, so we must do a little better from the
start. We consider a static polarization function, eq. 9.82.33 The bare electron-electron
Coulomb repulsion and its Fourier transform are given by (see Mattuck p. 191; in cgs
units, I am sorry for that)

e2 4πe2
V (r) = ↔ vq = 2 (9.93)
r q
From eqs. 9.82, 9.90 and 9.93, one obtains for the static, ω = 0, and long wave length
q ¿ kF limit
λ2
ε(q,0) = 1 + with
q2
4me e2 kF 3ρ 1
λ2 = = 4me e2 ( ) 3 (9.94)
π π
where ρ is the density of the electron gas. Using this expression for all wave vectors q (al-
though stricktly speaking one should not do that) is called the Thomas-Fermi model for screening.
We find for the screened interaction, eq. 9.91 and its Fourier transform
4πe2 e2 −λr
Wq (ω = 0) = ↔ W (r,ω = 0) = e (9.95)
q 2 + λ2 r
33
The exact expression for Π0 (q,ω) can be obtained, but it is not very enlighting due to its complexity,
see e.g. Fetter & Walecka. For simplicity reasons we therefore stick to the static approximation.
298 CHAPTER 9. THE ELECTRON-ELECTRON INTERACTION

The expression is real space is called the Yukawa potential , which is a Coulomb potential
e2 −λr . This is an example of a screened Coulomb
r multiplied by a shielding factor e
interaction we discussed in Section 9.3.5. The exponential factor reduces the potential
to near zero for distances r much larger than the typical Thomas-Fermi screening length
ls = λ−1 . The latter depends on the density of the electron gas, cf. eq. 9.94; the
dependence however is rather weak. For a low density metal like potassium the Thomas-
Fermi screening length is ls ≈ 1.1Å and for a high density metal like copper ls ≈ 0.6Å.
These are very short lengths indeed, so as we have argued in Section 9.3.5, screening should
have a large effect on all metallic properties. As promised in that section, screening leads
to a Fourier transform Wq that has no singularity at q = 0. One therefore expects that
the Hartree-Fock artefact (the “spike” of ²0 (k) at the Fermi energy which lead to a zero
density of states) will be lifted, as we will see below.

For semiconductors and isolators the Thomas-Fermi model for screening is not valid.
For these cases the static approximation gives a reasonable first ansatz

vq e2
Wq (ω = 0) = ↔ W (r,ω = 0) = (9.96)
εe εe r

where εe is the dielectric constant due to the respons of the electronic system (excluding
nuclear motions or phonons, which are an entirely different matter). As an example, for
silicon εe ≈ 12, indicating that also for semiconductors screening is a large effect, albeit
less important than for metals. εe roughly scales as Eg−1 , where Eg is the band gap of a
material. Even good isolating materials, which have large band gaps, have εe ≈ 2-3, so
even there screening is not negligible.

The self-energy in the Thomas-Fermi screening approximation


Having obtained an expression for the screened interaction, we can set out to calculate
the self-energy ΣX,RP A , see eqs. 9.67 and 9.68. This allows us to solve the GW equations
and obtain the quasi-particle energies and lifetimes, cf. eqs. 9.71—9.74. The self-energy
is diagonal in the basis set of plane waves, i.e. ΣX,RP A (k, m,ω) = δ km ΣX,RP A (k, k,ω).
This is easily observed if one labels the diagram of eq. 9.67, according to the rules of
conservation of momentum and frequency34

k,ω

k − q,
ω −ω' q,ω '

k,ω

Because the self-energy is diagonal, the solutions of the GW equations for the homoge-
neous electron gas are again simple plane waves; read Section 9.3.3 above eq. 9.49. The
GW eigenvalues become ²k (ω) = ²0,k + ΣX,RP A (k, k,ω). The propagator is diagonal and
34
One can also take the expression of eq. 9.68 and use the special rules for matrix elements of the
homogeneous electron gas.
9.4. THE RANDOM PHASE APPROXIMATION (RPA) 299

becomes
1 1
GRP A (k, k, ω) = = (9.97)
~ω − ²k (ω) ± iδ ~ω − ²0,k − ΣX,RP A (k, k,ω) ± iδ
The self-energy according to eq. 9.68 becomes
X Z dω 0
ΣX,RP A (k, k,ω) = GRP A (k − q, k − q,ω − ω 0 )Wq (ω 0 ) (9.98)
q

Making the usual substition, cf. eq. 9.81 and using eq.9.73, 9.75 and 9.91 we get
Z Z
d3 q dω 0 Wq (ω 0 )
ΣX,RP A (k, k,ω) = 3
where (9.99)
(2π) 2π ~(ω − ω ) − ²k−q (ω − ω0 ) ± iδ
0

²l (ω) = ²0,l + ΣX,RP A (l, l,ω) and


vq
Wq (ω) =
ε(q,ω)
Note the occurence of ΣX,RP A both on the righthand and on the lefthand side of eq. 9.99.
This again is a matter of self-consistency; it is an equation that is to be solved iteratively.35
The integrals can in principle be attacked, but they are extremely nasty and we will not
pursue them here. We can get some physical look and feel however by studying a limiting
case. For instance, we can start by setting ΣX,RP A to zero on the righthand side of eq.
9.99. This is equivalent to using G0 instead of GRP A on the righthand side of eq. 9.98,
or, in other words, doing a non self-consistent calculation. We then approximate Wq (ω 0 )
by its static value Wq (0) of eq. 9.95, which uses the Thomas-Fermi model for screening,
and get
Z Z
d3 q 4πe2 dω 0 1
ΣTX,RP
F
A (k, k,ω) ≈ 3 2
(2π) q 2 + λ 2π ~ω 00 − ²0,k−q ± iδ
Z Z
d3 q 4πe2 d3 q 0 4πe2
= − 2 = ¯ ¯ (9.100)
3 2 3 0 2
|k−q|≤kF (2π) q + λ q 0 ≤kF (2π) ¯k − q ¯ + λ
2

where the T F superscript indicates that we have used the Thomas-Fermi model for static
screening. Defining ω 00 = ω − ω 0 , the frequency integral has to be done with +iδ for
electrons, |k − q| > kF , and −iδ for holes, |k − q| ≤ kF . As in the Hartree-Fock case only
the occupied states |k − q| ≤ kF give a contribution, which makes sense since an incoming
particle can only interact with particles in the system that are present (and if present,
they occupy a state). The second line of this equation can then be derived using a contour
integration on the frequency integral (see Mattuck’s Appendix I and the discussion of his
eq. (4.31) and (8.35)). Note that the frequency dependence is now gone on the R righthand
side, as one expects from a static approximation. The remaining integral d3 q0 can be
done, see Mattuck’s p. 93-94, or use MAPLE. The result is the following rather unwieldy
expression
· µ ¶¸
e2 1 ¡ 2 ¢ (k + kF )2 + λ2
ΣTX,RP
F
A (k, k) = − { k − k 2
+ λ 2
ln
π 4k F (k − kF )2 + λ2
· ¸
k + kF k − kF
+kF − λ arctan − arctan } (9.101)
λ λ
35 (1)
For instance, start by setting ΣX,RP A = 0 in the righthand side. This gives a ΣX,RP A on the lefthand
(2)
side. Use this solution in the righthand side to obtain a ΣX,RP A on the lefthand side, etcetera. When
(N +1) (N )
ΣX,RP A = ΣX,RP A we have a self-consistent solution.
300 CHAPTER 9. THE ELECTRON-ELECTRON INTERACTION

-0.2 kF = 1

-0.4

-0.6

-0.8 k F = 0.1
π
Σ X (k ) -1
e2kF
-1.2

-1.4 k F = 0.01

-1.6

-1.8 HF

-2
k
0 0.2 0.4 0.6 0.8 1 k
1.2 F
k

Figure 9.19: The self-energy according to Thomas-Fermi screening

As a consistency test note that we can derive the “bare” Hartree-Fock result of eq. 9.47
by letting the screening go to zero, i.e. limλ→0 ΣTX,RP
F
A (k, k) = −VX,kk . We can make
1
a plot of this function using the expression λ = ( π4 kF ) 2 according to eq. 9.94 (in atomic
e2 kF k
units). The result for ΣTX,RP
F
A (k) in units of π as a function of kF is shown in Fig. 9.19.
The bottom line is the “bare” Hartree-Fock result of Fig. 9.12. The upper curves give
the results for increasing kF (which means increasing electron density, cf. eq. 9.94). Actu-
ally only the topmost curve, i.e. kF = 1, corresponds to a density which is representative
for a real metal, and the lower curves correspond to unrealistically low electron densities.
From the difference between the top and bottom curves one observes that screening indeed
has a huge effect on the “bare” interaction; the topmost curve is almost flat as compared
to the bottom one. The propagator of eq. 9.97 is that of a particle with energy
²T F,k = ²0,k + ΣTX,RP
F
A (k, k) (9.102)
When we plot this particle energy like in Fig. 9.13 for the kF = 1 case, the self-energy term
2 2
gives a correction to the free particle energy ²0,k = ~2mke which is relatively independent of
the wave number k. This is shown in Fig. 9.20.
The bottom curve gives the Hartree-Fock particle energies, the top curve the free
particle energies, and the middle curve the RPA particle energies calculated with Thomas-
Fermi screening. The latter results are much closer to the free particle results again !!
Apparantly, screening results in weakly interacting particles. The resemblance of the RPA
screened particles with free particles gives some justification for using free electron models
in solid state physics !! So what you have learned in your introductory solid state course
indeed did have some meaning.
9.4. THE RANDOM PHASE APPROXIMATION (RPA) 301

0.5

ε 0, k
k
0 0.2 0.4 0.6 0.8 1 1.2 k F
ε TF ,k k

ε HF ,k
-0.5

-1

Figure 9.20: The RPA quasi-particle energies using Thomas-Fermi screening

Last, but not least, I promised you that screening was going to cure the Hartree-Fock
artefact. This is shown in Fig. 9.21.
The top curve is the k-derivative of the Hartree-Fock particle energies, as in Fig.
9.15, which has the spike at k = kF . The bottom curve is the same derivative of the
RPA/Thomas-Fermi particle energies. As one can see, the latter is extremely well-behaved.
Moreover, as ΣTX,RP
F
A (k, k) has a relative weak dependence on k, the derivative of ²T F,k is
dominated by that of ²0,k , cf. eq. 9.102. This means that the density of states, eq. 9.57,
resembles the free particle result of eq. 9.56. The correction due to ΣTX,RP
F
A (k, k) is easily
calculated and free of artefacts. All’s well that ends well.

The self-energy; accurate screening and quasi-particles


I do not want to leave you with the idea that the Thomas-Fermi model we have used
for screening gives very accurate results quantitatively. It does not, although it gives the
correct qualitative impression of what screening does for you. Let us look again critically
at the approximations we have made.

1. We have used eq. 9.82 for the polarization propagator Π0 (q,0) for all q, although
2
it is only valid for q ¿ kF . This lead to the Yukawa potential W (r) = er e−λr of eq.
9.99. If we don’t use this approximation, but do the integral of eq. 9.82 properly to
obtain Π0 (q,0), this leads to a screened potential that falls off slower as a function of
r. In fact, it has a long range oscillating tail that goes like W (r) =e2 r−3 cos(2kF r)
for large r (see Mattuck §10.7).
302 CHAPTER 9. THE ELECTRON-ELECTRON INTERACTION

7 dε k
dk
6

1
k
0 0.2 0.4 0.6 0.8 1 1.2 kF
k

Figure 9.21: k-derivative of the HF (upper) and RPA/Thomas-Fermi (lower) particle


energies

2. We have neglected the frequency dependence of the screened interaction; i.e. we


set Wq (ω) = Wq (0) for all ω. Fourier transforming this approximation to the time
domain one gets Wq (t) = Wq (0)δ(t). In other words we treated the screened in-
teraction as an instantaneous interaction, like the bare interaction Vq . As explained
before, we expect the system to need some time to respond to a propagating charge,
which expresses itself in the dielectric response function ε(q,ω) being frequency de-
pendent. This “sluggishness” in the system’s response hinders the propagating parti-
cle, or, in quantum mechanical terms, it can scatter the particle into a different state.
The propagating particle thus becomes a “quasi-particle” with a finite lifetime. If
we neglect the frequency dependence and make the system respond instanteneously
to the propagating charge, it does no longer hinder the particle. The propagator
GRP A (k, k, ω) = (~ω − ²T F,k ± iδ)−1 then becomes that of a particle with energy
²T F,k and infinite lifetime. This clearly needs to be corrected.

The self-energy ΣX,RP A (k, k,ω) of eq. 9.99 is a complex quantity, because the dielectric
function ε(q,ω) is (it must be in order to describe losses). The frequency dependence of
the self-energy is quite complicated. Note for example that the integrand in the frequency
integral of eq. 9.99 has poles coming from the propagator (from the denominator of eq.
Vq
9.99), as well as poles coming from Wq (ω 0 ) = ε(q,ω 0) .
36 We will not pursue it here, but

merely state that ΣX,RP A (k, k,ω) can be calculated in full (see Fetter & Walecka, for
instance). The propagator of eq. 9.97 is to be Fourier transformed to the time domain
Z ∞
1
GRP A (k, k, t) = dω GRP A (k, k, ω)
2π −∞
Z ∞
1 e−iωt
= dω (9.103)
2π −∞ ~ω − ²0,k − ΣX,RP A (k, k,ω)
36
In a metal there is certainly always (at least) one frequency ωp at which ε(q,ωp ) = 0. It is called the
plasma frequency.
9.4. THE RANDOM PHASE APPROXIMATION (RPA) 303

This is best done by complex contour integration, cf. Sections 4, 4.8.4. For that we need
the poles of the integrand. One can prove on rather formal grounds that the propagator
GRP A (k, k, ωc ) as a function of a complex ω c = ω + iωi is analytical in the upper half of
the complex plane for ~ω > ²F and in the lower half of the complex plane if ~ω < ²F . In
other words its poles must be in the lower half plane when ~ω > ²F and in the upper half
plane when ~ω < ²F .37 This then automatically means that

Im ΣX,RP A (k, k,ω) < 0 for ~ω > ²F


Im ΣX,RP A (k, k,ω) > 0 for ~ω < ²F (9.104)

According to Section 4.8, if we close the contour in the lower half plane we enclose the
poles which lie at ~ω > ²F , and get a non-zero result for t > 0
i
i~GRP A (k, k, t) = zk e− ~ [²0,k +∆k ]t e−Γk t + F (k,t) for ²0,k + ∆k > ²F ; t >(9.105)
0
zk
where ∆k = Re ΣX,RP A (k, k,ω k ) and Γk = |Im ΣX,RP A (k, k,ωk )|
~
µ ¶−1
∂ΣX,RP A (k, k,ω)
and zk = 1− |ω=ωk
∂ω
and ~ω k − ²0,k − ΣX,RP A (k, k,ω k ) = 0 determines the pole’s position. This result
might seem a bit complicated, but start from the free particle case where the self-energy
Σ = 0 per definition (which means ∆k = Γk = 0 and zk = 1). We have i~G0 (k, k, t) =
i
e− ~ ²0,k t ; ²0,k > ²F ; t > 0 which describes the free propagation of an electron. All the
effects of the electron-electron interaction have been worked into the self-energy Σ. As
we observe in eq. 9.105, this results in (1) a shift of the energy level ²0,k → ²0,k + ∆k ;
(2) adding an exponential decay with a characteristic time τ k = Γ−1 k and (3) adding a
weight factor zk . These are the same three features that appeared in the simple scattering
of one electron by interaction with atomic potentials in Section 4. Here they are caused
by scattering due to electron-electron interaction. It seems that any kind of interaction
giving rise to scattering leads to a similar general behaviour for the propagation of the
electron. The seemingly new feature here is the function F (k,t); it takes care of other
possible poles in the integration. For “normal” systems this function is smooth and has
a low weight, so it is not very important, for “strongly correlated” systems it may have a
more complicated structure.

Since the propagator of eq. 9.105 resembles a free particle propagator, modified by
the three factors mentioned above, it is said to describe a quasi-particle (or in this case
a quasi-electron, to be more precise). This term actually only makes sense if Γk is not
too large or |zk | is too small, otherwise the quasi-particle would decay too fast or carry
too small a weight to be observable in practice (this is usually measured experimentally
by (inverse) photoemission, Appendix II of the previous chapter). From eq. 9.104 one ob-
serves that Im Σ(k, k,ω) changes sign at the Fermi level. One can prove that Im Σ(k, k,ω)
37
The propagator must have poles somewhere, since otherwise the contour integration would give zero
according to the residue theorem, cf. section 4.8. Since we do have a particle or hole propagating, the
propagator obviously cannot be zero.
This global statement of where the poles are holds in fact for the exact propagator G(k, k, ω) as well,
see Appendix II of the previous chapter; as well as for all the approximations we have encountered; the
free particle propagator G0 (k, k, ω), eq. 9.14; the Hartree propagator GH (k, k, ω), eq. 9.26 and the
Hartree-Fock propagator GHF (k, k, ω), eq. 9.39.
304 CHAPTER 9. THE ELECTRON-ELECTRON INTERACTION

ε 0,k

V HF ,k
∆k

Figure 9.22: Quasi-particle energies as calculated within RPA.

is a continuous function near the Fermi level. This means that there is a region around
the Fermi level where |Im Σ(k, k,ω)| is small. Γk is then small since it is not difficult to
show that 0 < |zk | ≤ 1.

Numerical examples for the homogeneous electron gas


Fig. 9.22 give the terms that contribute to the quasi-particle energy, as calculated within
RPA with the “exact” frequency dependent self-energy.38 On the x-axis you find kkF as
before; on the y-axis the energy (in Rydberg; if you divide this by 2 you get the atomic
energy unit I used on the y-axis of Figs. 9.20 and 9.13). At positive energy you find
2 2
the free particle values ²0,k = ~2mke . The curves are marked with a typical parameter used
for the homogeneous electron gas called rs ; you can transfer this into a more familiar one
1
by using kF = ( 9π 3 −1
4 ) rs . The curve marked “2” thus gives kF = 0.96, which is close to
the value kF = 1 which we used to plot Figs. 9.20 and 9.13. At negative energies the
quasi-particle RPA corrections ∆k are plotted for the three different values of rs (or kF ).
For comparision, also the curves for the bare Hartree-Fock exchange energy −VX,kk are
given. As remarked in connection with Fig. 9.19, the RPA corrections have a very weak
dependence on the wave number k, in contrast to the Hartree-Fock results. The quasi-
particle energies ²0,k +∆k thus have almost the same dependence on k as the free particles,
in other words the shift ∆k is almost uniform (within say a few percent). The conclusion
drawn from the Thomas-Fermi results still stands: this result is the true justification for
using free electron models in solid state physics, even if the electrons are far from free.
Comparing the RPA shift ∆k to the Thomas-Fermi shift ΣTX,RP F
A (k, k) shown in Fig. 9.19,

38
These and the figures below are taken from: L. Hedin and S. Lundqvist, in Solid State Physics, vol.
23, eds. F. Seitz, D. Turnbull, and H. Ehrenreich (Academic Press, New York, 1969).
9.4. THE RANDOM PHASE APPROXIMATION (RPA) 305

Im Σ(k , k , ω k ) / ε F

RPA

Figure 9.23: The factor |Im ΣX,RP A (k, k,ω k )| which determines the inverse quasi-particle
lifetime, in units of ²F , in the RPA approximation (lower curve).

one observes that ∆k ≈ 3 × ΣTX,RP


F
A (k, k). The Thomas-Fermi result thus underestimates
the shift. This is due to an overestimation of the shielding by Thomas-Fermi, i.e. the
Yukawa potential is too short-ranged, as discussed in the beginning of this section. Yet
it does much better than Hartree-Fock, which gives shifts −VX,kk that are clearly more
inaccurate.

Fig. 9.23 gives the factor |Im ΣX,RP A (k, k,ωk )| = ~(zk τ k )−1 which determines the
quasi-particle lifetime τ k in the RPA approximation, again using the “exact” frequency
dependent self-energy now. Don’t mind the curve marked “PP”; that is just (another)
model calculation. Note that right at the Fermi level, k = kF , |Im ΣX,RP A (k, k,ωk )| = 0 as
expected, and thus τ kF = ∞; the quasi-particle lifetime is infinite. In a fairly large region
around the Fermi level |Im ΣX,RP A (k, k,ω k )| < 0.1 × ²F , which means that the lifetime is
still long enough for the quasi-particle to be observed in an experiment such as (inverse)
photoemission. It is only when k & 1.5kF that |Im ΣX,RP A (k, k,ω k )| starts to rise steeply
and the lifetime thus goes down. At these high energies ~ω k & 2²F , the quasi-particle
model starts to break down. Indeed at such high energies another type of quasi-particle or
collective excitation called “plasmon” is observed. That however is another story, which
we won’t discuss here.

Just to reassure you, in Fig. 9.24 the weight factor zk at the Fermi level, |k| = kF , as
1
−1
a function of rs = ( 9π
4 ) kF is plotted. Only notice the curve marked “RPA”; the other
3

curves and points come from different models. At rs = 2, which is the value for an ordinary
home, garden and kitchen metal such as aluminium, zkF is close to 0.8. Such a high weight
for the quasi-particle term is another justification for the use of the quasi-particle model
or, in other words, the factor F (k,t) in eq. 9.105 is relatively unimportant. This need not
longer be the case for “strongly correlated” systems, where zk can become small and the
factor F (k,t) can show a complicated structure.
306 CHAPTER 9. THE ELECTRON-ELECTRON INTERACTION

Figure 9.24: The weight factor zk at the Fermi level, |k| = kF as a function of rs

“Über allen Gipfeln ist Ruh”’, Goethe.

S-ar putea să vă placă și