Graduate Texts in Physics

Michele Fabrizio

A Course
in Quantum
Many-Body
Theory
From Conventional Fermi Liquids
to Strongly Correlated Systems
Graduate Texts in Physics

Series Editors
Kurt H. Becker, NYU Polytechnic School of Engineering, Brooklyn, NY, USA
Jean-Marc Di Meglio, Matière et Systèmes Complexes, Bâtiment Condorcet, Université
Paris Diderot, Paris, France
Sadri Hassani, Department of Physics, Illinois State University, Normal, IL, USA
Morten Hjorth-Jensen, Department of Physics, Blindern, University of Oslo, Oslo,
Norway
Bill Munro, NTT Basic Research Laboratories, Atsugi, Japan
Richard Needs, Cavendish Laboratory, University of Cambridge, Cambridge, UK
William T. Rhodes, Department of Computer and Electrical Engineering and Computer
Science, Florida Atlantic University, Boca Raton, FL, USA
Susan Scott, Australian National University, Acton, Australia
H. Eugene Stanley, Center for Polymer Studies, Physics Department, Boston
University, Boston, MA, USA
Martin Stutzmann, Walter Schottky Institute, Technical University of Munich,
Garching, Germany
Andreas Wipf, Institute of Theoretical Physics, Friedrich-Schiller-University Jena,
Jena, Germany
Graduate Texts in Physics publishes core learning/teaching material for graduate- and
advanced-level undergraduate courses on topics of current and emerging fields within
physics, both pure and applied. These textbooks serve students at the MS- or
PhD-level and their instructors as comprehensive sources of principles, definitions,
derivations, experiments and applications (as relevant) for their mastery and teaching,
respectively. International in scope and relevance, the textbooks correspond to course
syllabi sufficiently to serve as required reading. Their didactic style, comprehensive-
ness and coverage of fundamental material also make them suitable as introductions
or references for scientists entering, or requiring timely knowledge of, a research field.
Michele Fabrizio

A Course in Quantum
Many-Body Theory
From Conventional Fermi Liquids to
Strongly Correlated Systems
Michele Fabrizio
Department of Physics
International School for Advanced Studies
Trieste, Italy

ISSN 1868-4513 ISSN 1868-4521 (electronic)

Graduate Texts in Physics
ISBN 978-3-031-16304-3 ISBN 978-3-031-16305-0 (eBook)
https://doi.org/10.1007/978-3-031-16305-0

© The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer Nature
Switzerland AG 2022
This work is subject to copyright. All rights are solely and exclusively licensed by the Publisher, whether
the whole or part of the material is concerned, specifically the rights of translation, reprinting, reuse
of illustrations, recitation, broadcasting, reproduction on microfilms or in any other physical way, and
transmission or information storage and retrieval, electronic adaptation, computer software, or by similar
or dissimilar methodology now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication
does not imply, even in the absence of a specific statement, that such names are exempt from the relevant
protective laws and regulations and therefore free for general use.
The publisher, the authors, and the editors are safe to assume that the advice and information in this book
are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or
the editors give a warranty, expressed or implied, with respect to the material contained herein or for any
errors or omissions that may have been made. The publisher remains neutral with regard to jurisdictional
claims in published maps and institutional affiliations.

Cover image: Olga_Kostrova

This Springer imprint is published by the registered company Springer Nature Switzerland AG
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
To my family, especially my wife Laura
Preface

This book collects the Lecture Notes of the Ph.D. course on Many-Body Theory
that I have been teaching for many years now. As such, the book is not intended
to be a compendium of all possible topics in Many-Body Theory, but just aims
to provide Ph.D. students with concepts and tools that I personally consider indis-
pensable to tackle modern issues of Many-Body Theory, most notably strongly
correlated electron systems. For that same reason, the book unavoidably reflects
my own limited research experience.
Besides a core part, Chaps. 1–4, where basic tools are presented in detail, there
are frequent excursions in topics of strongly correlated electron systems, among
which the two final chapters are on Luttinger Liquids and the Kondo Effect. In
addition, the microscopic derivation of Landau’s Fermi Liquid Theory in Chap. 5
includes conventional Fermi Liquids as a special case, but also less conventional
states, like, for instance, the Mott insulators of Sect. 5.6 that display the same
spin and thermal properties of normal metals, which might be relevant to strongly
correlated materials.
All chapters contain very few simple problems, whose solution is not even
shown. In addition, most chapters include sections with applications that are
simply more involved problems whose solution is explicitly derived.

Trieste, Italy Michele Fabrizio

Acknowledgements I am grateful to Prof. Claudio Castellani, whose master’s course on Quantum

Many-Body Theory first sparked my interest in the field, and to Profs. Philippe Noziéres and Erio
Tosatti, who nourished and mentored that interest.

vii
Contents

1 Second Quantization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1 Fock States and Space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Fermionic Operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2.1 Fermi Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.2.2 Second Quantisation of Multiparticle Operators . . . . . . . . . 6
1.3 Bosonic Operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.3.1 Bose Fields and Multiparticle Operators . . . . . . . . . . . . . . . . 10
1.4 Canonical Transformations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.4.1 Canonical Transformations with Charge
Non-conserving Hamiltonians . . . . . . . . . . . . . . . . . . . . . . . . . . 14
1.4.2 Harmonic Oscillators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
1.5 Application: Electrons in a Box . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
1.6 Application: Electron Lattice Models and Emergence
of Magnetism . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
1.6.1 Hubbard Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
1.6.2 Mott Insulators and Heisenberg Models . . . . . . . . . . . . . . . . 26
1.7 Application: Spin-Wave Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
1.7.1 Classical Ground State . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
1.8 Beyond the Classical Limit: The Spin-Wave Approximation . . . . . 32
1.8.1 Hamiltonian of Quantum Fluctuations . . . . . . . . . . . . . . . . . . 34
1.8.2 Spin-Wave Dispersion and Goldstone Theorem . . . . . . . . . 36
1.8.3 Validity of the Approximation
and the Mermin-Wagner Theorem . . . . . . . . . . . . . . . . . . . . . . 37
1.8.4 Order from Disorder . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
2 Linear Response Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
2.1 Linear Response Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
2.2 Kramers-Kronig Relations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
2.2.1 Symmetries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
2.3 Fluctuation-Dissipation Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
2.4 Spectral Representation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53

ix
x Contents

2.5 Power Dissipation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55

2.5.1 Absorption/Emission Processes . . . . . . . . . . . . . . . . . . . . . . . . . 57
2.5.2 Thermodynamic Susceptibilities . . . . . . . . . . . . . . . . . . . . . . . . 57
2.6 Application: Linear Response to an Electromagnetic Field . . . . . . 59
2.6.1 Quantisation of the Electromagnetic Field . . . . . . . . . . . . . . 61
2.6.2 System’s Sources for the Electromagnetic Field . . . . . . . . 63
2.6.3 Optical Constants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
2.6.4 Linear Response in the Longitudinal Case . . . . . . . . . . . . . . 67
2.6.5 Linear Response in the Transverse Case . . . . . . . . . . . . . . . . 74
2.6.6 Power Dissipated by the Electromagnetic Field . . . . . . . . . 76
Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
3 Hartree-Fock Approximation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
3.1 Hartree-Fock Approximation for Fermions at Zero
Temperature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
3.1.1 Alternative Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
3.2 Hartree-Fock Approximation for Fermions at Finite
Temperature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
3.2.1 Saddle Point Solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
3.3 Time-Dependent Hartree-Fock Approximation . . . . . . . . . . . . . . . . . . 96
3.3.1 Bosonization of the Low-Energy Particle-Hole
Excitations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
3.4 Application: Antiferromagnetism in the Half-Filled
Hubbard Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
3.4.1 Spin-Wave Spectrum by Time-Dependent
Hartree-Fock . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
4 Feynman Diagram Technique . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
4.1 Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
4.1.1 Imaginary-Time Ordered Products . . . . . . . . . . . . . . . . . . . . . . 121
4.1.2 Matsubara Frequencies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
4.1.3 Single-Particle Green’s Functions . . . . . . . . . . . . . . . . . . . . . . 127
4.2 Perturbation Expansion in Imaginary Time . . . . . . . . . . . . . . . . . . . . . . 133
4.2.1 Wick’s Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
4.3 Perturbation Theory for the Single-Particle Green’s
Function and Feynman Diagrams . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
4.3.1 Diagram Technique in Momentum and Frequency
Space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
4.3.2 The Dyson Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144
4.3.3 Skeleton Diagrams . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147
4.3.4 Physical Meaning of the Self-energy . . . . . . . . . . . . . . . . . . . 148
4.3.5 Emergence of Quasiparticles . . . . . . . . . . . . . . . . . . . . . . . . . . . 156
4.4 Other Kinds of Perturbations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157
4.4.1 Scalar Potential . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158
4.4.2 Coupling to Bosonic Modes . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159
Contents xi

4.5 Two-Particle Green’s Functions and Correlation Functions . . . . . . 161

4.5.1 Diagrammatic Representation of the Two-Particle
Green’s Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163
4.5.2 Correlation Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165
4.6 Coulomb Interaction and Proper and Improper Response
Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 168
4.7 Irreducible Vertices and the Bethe-Salpeter Equations . . . . . . . . . . . 170
4.7.1 Particle-Hole Channel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171
4.7.2 Particle-Particle Channel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174
4.7.3 Self-energy and Irreducible Vertices . . . . . . . . . . . . . . . . . . . . 175
4.8 The Luttinger-Ward Functional . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 180
4.8.1 Thermodynamic Potential . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181
4.9 Ward-Takahashi Identities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 184
4.9.1 Ward-Takahashi Identity for the Heat Density . . . . . . . . . . 187
4.10 Conserving Approximation Schemes . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189
4.10.1 Conserving Hartree-Fock Approximation . . . . . . . . . . . . . . . 191
4.10.2 Conserving GW Approximation . . . . . . . . . . . . . . . . . . . . . . . . 193
4.11 Luttinger’s Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194
4.11.1 Validity Conditions for Luttinger’s Theorem . . . . . . . . . . . . 197
4.11.2 Luttinger’s Theorem in Presence of Quasiparticles
and in Periodic Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 204
Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 207
5 Landau’s Fermi Liquid Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209
5.1 Emergence of Quasiparticles Reexamined . . . . . . . . . . . . . . . . . . . . . . . 211
5.2 Manipulating the Bethe-Salpeter Equation . . . . . . . . . . . . . . . . . . . . . . 215
5.2.1 A Lengthy but Necessary Preliminary Calculation . . . . . . 216
5.2.2 Interaction Vertex and Density-Vertices . . . . . . . . . . . . . . . . . 221
5.3 Linear Response Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 223
5.3.1 Response Functions of Densities Associated
to Conserved Quantities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 225
5.4 Thermodynamic Susceptibilities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 227
5.4.1 Charge Compressibility . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 227
5.4.2 Spin Susceptibility . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 228
5.4.3 Specific Heat . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 229
5.5 Current-Current Response Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 230
5.5.1 Thermal Response . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 232
5.5.2 Coulomb Interaction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 236
5.6 Mott Insulators with a Luttinger Surface . . . . . . . . . . . . . . . . . . . . . . . . 236
5.7 Luttinger’s Theorem and Quasiparticle Distribution
Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 241
5.7.1 Oshikawa’s Topological Derivation of Luttinger’s
Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 244
xii Contents

5.8 Quasiparticle Hamiltonian and Landau-Boltzmann

Transport Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 248
5.8.1 Landau-Boltzmann Transport Equation
for Quasiparticles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 250
5.8.2 Transport Equation in Presence
of an Electromagnetic Field . . . . . . . . . . . . . . . . . . . . . . . . . . . . 253
5.9 Application: Transport Coefficients with Rotational
Symmetry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 257
Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 262
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 264
6 Brief Introduction to Luttinger Liquids . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 267
6.1 What Is Special in One Dimension? . . . . . . . . . . . . . . . . . . . . . . . . . . . . 267
6.2 Interacting Spinless Fermions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 271
6.2.1 Bosonized Expression of the Non-interacting
Hamiltonian . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 273
6.2.2 Bosonic Representation of the Fermi Fields . . . . . . . . . . . . 275
6.2.3 Operator Product Expansion . . . . . . . . . . . . . . . . . . . . . . . . . . . 279
6.2.4 Non-interacting Green’s Functions
and Density-Density Response Functions . . . . . . . . . . . . . . . 280
6.2.5 Interaction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 284
6.2.6 Interacting Green’s Functions and Correlation
Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 288
6.2.7 Umklapp Scattering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 290
6.2.8 Behaviour Close to the K = 1/2 Marginal Case . . . . . . . . 293
6.3 Spin-1/2 Heisenberg Chain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 303
6.4 The One-Dimensional Hubbard Model . . . . . . . . . . . . . . . . . . . . . . . . . . 305
6.4.1 Luttinger Versus Fermi Liquids . . . . . . . . . . . . . . . . . . . . . . . . 309
Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 311
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 312
7 Kondo Effect and the Physics of the Anderson Impurity Model . . . . . . 313
7.1 Brief Introduction to Scattering Theory . . . . . . . . . . . . . . . . . . . . . . . . . 314
7.1.1 General Analysis of the Phase-Shifts . . . . . . . . . . . . . . . . . . . 319
7.2 The Anderson Impurity Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 320
7.2.1 Non Interacting Impurity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 322
7.2.2 Hartree-Fock Approximation . . . . . . . . . . . . . . . . . . . . . . . . . . . 324
7.3 From the Anderson Impurity Model to the Kondo Model . . . . . . . 325
7.3.1 The Emergence of Logarithmic Singularities
and the Kondo Temperature . . . . . . . . . . . . . . . . . . . . . . . . . . . . 327
7.3.2 Anderson’s Poor Man’s Scaling . . . . . . . . . . . . . . . . . . . . . . . . 330
7.4 Noziéres’s Local Fermi Liquid Theory . . . . . . . . . . . . . . . . . . . . . . . . . . 332
7.4.1 Ward-Takahashi Identity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 335
7.4.2 Luttinger’s Theorem and Thermodynamic
Susceptibilities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 339
Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 343
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 343
Second Quantization
1

The first difficulty encountered in solving a many-body problem is how to deal with a
many-body wavefunction. The reason is that a many-body wavefunction has to take
into account both the indistinguishability of the particles as well as their statistics,
whereas in first quantisation any operator, including the Hamiltonian, does not have
those properties. Therefore, it is desirable to have at disposal an alternative scheme
where the indistinguishability principle as well as the statistics of the particles are
already built in the expression of operators. That is actually the scope of second
quantisation.

1.1 Fock States and Space

Let us take a system of N particles, either fermions or bosons. The Hilbert space
spans a basis of N -body orthonormal wavefunctions which should satisfy both the
indistinguishability principle as well as the appropriate statistics of the particles. The
simplest way to construct this space is as follows.
We start by choosing an orthonormal basis of single-particle wavefunctions:

φa (x) , a = 1, 2, . . . .

Here x is a generalised coordinate that includes both the space coordinate r as well as,
e.g., the z-component, σ , of the spin, which is half-integer for fermions and integer
for bosons. The suffix a is a quantum label and, by definition,
 
d x φa∗ (x) φb (x) = δab , φa∗ (x) φa (y) = δ(x − y) , (1.1)
a

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 1

M. Fabrizio, A Course in Quantum Many-Body Theory, Graduate Texts in Physics,
https://doi.org/10.1007/978-3-031-16305-0_1
2 1 Second Quantization
   
where d x . . . means σ dr . . . , while δ(x − y) ≡ δ(r − r ) δσ σ  with x =
 
(r, σ ) and y = (r , σ ). A generic N -body wavefunction with the appropriate sym-
metry properties can be constructed through the above single-particle states. Since
the particles are not distinguishable, we do not need to know which particle occupies
a specific state. Instead, what we need to know are just the occupation numbers n a ’s,
i.e., the number of particles occupying each single-particle state φa . That number is
either n a = 0, 1 for fermions, because of the Pauli principle, or an arbitrary integer
n a ≥ 0 for bosons. Apart from that, the occupation numbers should satisfy the trivial
particle-conservation constraint

na = N .
a

Since the occupation numbers are the only ingredients we need in order to build up
the N -body wavefunction, we can formally denote the latter as the ket

|n 1 , n 2 , . . .  , (1.2)

which is called a Fock state, while the space spanned by the Fock states is called
Fock space. Within the Fock space, the state with no particles, the vacuum, will be
denoted by |0.
For instance, if the N fermions with coordinates xi , i = 1, . . . , N , occupy the
states a j , j = 1, . . . , N , with a1 < a2 < · · · < a N , namely n a = 1 for a ∈ {a j },
otherwise n a = 0, then the appropriate wavefunction is the Slater determinant
 
 φa1 (x1 ) · · · φa1 (x N ) 
  
1  φa2 (x1 ) · · · φa2 (x N ) 
{n a } (x1 , . . . , x N ) =  .. .. .. .. .. , (1.3)
N!  . ... . 
 
 φa (x1 ) · · · φa (x N ) 
N N

which is therefore the first-quantisation expression of the Fock state |{n a } with
the same occupation numbers. The above wavefunction satisfies the condition of
being antisymmetric if two coordinates are interchanged, namely two columns in
the determinant. Analogously, it is antisymmetric by interchanging two rows, i.e.,
two quantum labels.
On the contrary, if we have N bosons with coordinates xi , i = 1, . . . , N , which
occupy the states a j , j = 1, . . . , M, M ≤ N and a1 < a2 < · · · < a M , with occu-
pation numbers n a j , then the appropriate wavefunction is the permanent

j n j! 
{n a } (x1 , . . . , x N ) = φa1 (x p1 ) . . . φa1 (x pna )
N! p
1
(1.4)
φa2 (x pna ) . . . φa2 (x pna )
1 +1 1 +n a2

. . . φa M (x p N −na ) . . . φa M (x p N ) ,
M +1
1.2 Fermionic Operators 3

where the sum is over all non-equivalent permutations p’s of the N coordinates.1
Indeed, the wavefunction is even by interchanging two coordinates or two quantum
labels.
In conclusion, the space spanned by all possible Slater determinants built with
the same basis set of single-particle wavefunctions constitutes an appropriate Hilbert
(Fock) space for many-body fermionic wavefunctions. Analogously, the space
spanned by all possible permanents is an appropriate Hilbert space for many-body
bosonic wavefunctions.
In the following, we will introduce operators acting in the Fock space. We will
consider separately the fermionic and bosonic cases.

1.2 Fermionic Operators

Let us introduce the creation, ca† , and annihilation, ca , operators which add or remove,
respectively, one fermion in state a. The operator ca† ca first annihilates then creates
a particle in a, which can be done as many times as many particles occupy that state.
Therefore,
ca† ca |n a  = n a |n a , (1.5)
so it acts like the occupation number operator ca† ca ≡ n̂ a . Since by the Pauli principle
n a = 0, 1, then

ca† ca | 0 = 0, ca† ca | n a = 1 =| n a = 1. (1.6)

Analogously, the operator ca ca† first creates then destroys a fermion in state a, which
it cannot do if a is occupied because of the Pauli principle, while it can if it is empty.
Therefore,
ca ca† | 0 =| 0, ca ca† | n a = 1 = 0. (1.7)
Thus, either a is empty or occupied, the following equation holds:

ca ca† + ca† ca | n a  =| n a , n a = 0, 1 , (1.8)

which leads to the operator identity

ca ca† + ca† ca = ca , ca† = 1, (1.9)

where the symbol {. . . } means the anti-commutator. Moreover, since we cannot

create nor destroy two fermions in the same state, it also holds that

ca , ca = ca† , ca† = 0. (1.10)

1 Non-equivalent means for instance that φi (x)φi (y) is equivalent to φi (y)φi (x).
4 1 Second Quantization

Equations (1.9) and (1.10) are the anti-commutation relations satisfied by the fermion
operators with the same quantum label. Going back to (1.6) and (1.7), we readily see
that all are satisfied if

ca† | 0 =| n a = 1, ca† | n a = 1 = 0, ca | 0 = 0, ca | n a = 1 =| 0,

(1.11)
also showing that ca† is the Hermitian conjugate of ca .
Let us now consider the action of the above operators on a Fock state. First, we
need to provide a prescription to build a Fock state by means of the creation operators.
We shall assume that, if
| n a = 1 = ca† | 0,
then by definition and for b = a

cb† | n a = 1 = cb† ca† | 0 ≡| n b = 1, n a = 1. (1.12)

Since the Slater determinant, hence the corresponding Fock state, is odd by inter-
changing two rows, then

| n b = 1, n a = 1 = − | n a = 1, n b = 1 ≡ −ca† cb† | 0. (1.13)

Comparing (1.13) with (1.12), we conclude that

 
ca† cb† = −cb† ca† ⇒ ca† , cb† = 0. (1.14)

The Hermitian conjugate thus implies that also

 
ca , cb = 0. (1.15)

We note that because of (1.10), both (1.14) and (1.15) remain valid even if a = b.
Finally, we need to extract the reciprocal properties of cb† and ca for a = b. We
assume the following result:

ca | n a = 1, n b = 1 = ca ca† cb† | 0 ≡| n b = 1 = cb† | 0. (1.16)

Since | 0 = ca ca† | 0, and ca† cb† = −cb† ca† , it follows that

−ca cb† ca† | 0 = cb† ca ca† | 0,

namely that, for a = b,

 
ca , cb† = 0. (1.17)
1.2 Fermionic Operators 5

All the above anti-commutation relations can be cast in the simple formulas
     
ca , cb† = δab , ca† , cb† = 0, ca , cb = 0, ∀ a, b . (1.18)

If we extend our prescription (1.12) to more than two electrons, we find that any
Fock state has the simple expression
  † n i
| n1, n2, . . .  = ci | 0, (1.19)
i≥1

where the occupation numbers n i = 0, 1.

1.2.1 Fermi Fields

Till now we have defined fermionic operators for a given set of single-particle wave-
functions. Let us now introduce new operators which are independent of that choice.
We define annihilation and creation Fermi fields by
 
σ (r) ≡ (x) = φi (x) ci , (x)† = φi (x)∗ ci† . (1.20)
i i

They have the following anti-commutation relations:

    
(x), (y)† = φi (x) φ j (y)∗ ci , c†j
ij
 (1.21)
= φi (x) φi (y)∗ = δ(x − y),
i

as well as
   
(x), (y) = (x)† , (y)† = 0 , (1.22)

which are indeed independent of the basis. If we change the basis via the unitary
transformation Û acting on the basis set

φi (x) = Ui,α φα (x),
α

with unitary Û , then

  
(x) = φi (x) ci = φα (x) Ui,α ci = φα (x) cα , (1.23)
i i,α α
6 1 Second Quantization

showing that the proper transformation of the fermionic operators is


cα = Ui,α ci . (1.24)
i

Through the Fermi fields we can give a representation of the N -fermion Fock space
not through a basis of single-particle wavefunctions but directly in the space of the
coordinates. Specifically, such Fock space is spanned by the states
1
| x1 , . . . , x N  ≡ √  † (x1 ) . . .  † (x N ) | 0 , (1.25)
N!
which are indeed antisymmetric upon exchanging two coordinates. We note that,
through (1.1) and (1.20),

d x1 . . . d x N | x1 , . . . , x N  x1 , . . . , x N |

1
= d x1 . . . d x N  † (x1 ) . . .  † (x N ) | 0 0 | (x N ) . . . (x1 )
N!
1  † (1.26)
= ca1 . . . ca†N | 0 0 | ca N . . . ca1
N ! a ,...,a
1 N

= ca†1 . . . ca†N | 0 0 | ca N . . . ca1 ≡ I N ,
a1 <a2 ···<a N

where I N is the identity in the N -fermion Hilbert space. In addition, it is straight-

forward to show that the Slater determinant identified by the occupation numbers
{n a } ≡ (n 1 , n 2 . . . ) can be simply written as

{n a } (x1 , . . . , x N ) = x1 , . . . , x N | n 1 , n 2 , . . . 
1 (1.27)
=√ 0 | (x N ) . . . (x1 ) | n 1 , n 2 , . . . .
N!

1.2.2 Second Quantisation of Multiparticle Operators

Let us consider a generic single-particle operator in the first quantisation


N
V̂ = V (xi ) , (1.28)
i=1

where the sum runs over all N particles and V (xi ) is an operator acting both on space
coordinates and spins, namely

V (x) ≡ Vσ σ  (r, p) , (1.29)

where p = −i∇ is the conjugate momentum of r.

1.2 Fermionic Operators 7

Using (1.27) and the indistinguishability of the fermions, the matrix element
between two different Slater determinants can be shown to be

n 1 , n 2 , . . . | V̂ | n 1 , n 2 , . . .
 N   N
∗
≡ d xi {n  } (x 1 , . . . , x N ) V (x i ) {n a } (x 1 , . . . , x N )
a
i=1 i=1
 
N
∗
=N d xi {n  } (x 1 , . . . , x N ) V (x 1 ) {n a } (x 1 , . . . , x N ) (1.30)
a
i=1
 
N
N
= d xi V (x1 ) n 1 , n 2 , . . . |  † (x1 )  † (x2 ) . . .  † (x N ) | 0
N!
i=1
0 | (x N ) . . . (x2 ) (x1 ) | n 1 , n 2 , . . . .

We note that the integral over x2 , . . . , x N can be readily performed through (1.26)

N
d x2 . . . d x N  † (x2 ) . . .  † (x N ) | 0 0 | (x N ) . . . (x2 ) = I N −1 , (1.31)
N!

so that, dropping the identity operator I N −1 ,


n 1 , n 2 , . . . | V̂ | n 1 , n 2 , . . . = n 1 , n 2 , . . . | d x1  † (x1 ) V (x1 ) (x1 ) | n 1 , n 2 , . . . .
(1.32)
It thus follows that the second quantised expression of V̂ is simply
 
V̂ = d x  † (x) V (x) (x) = Vab ca† cb , (1.33)
ab

where
 
Vab ≡ d x φa∗ (x) V (x) φb (x) = dr φa∗ (r, σ ) Vσ σ  (r, p) φb (r, σ  ).
σσ
(1.34)
As an example, let us consider the density operator that, in first quantisation, reads


N
ρ̂σ (r) = δ(r − ri ) δσ,σi ,
i=1

while in second quantisation is


ρ̂σ (r) = dr σ  (r )† δ(r − r ) δσ,σ  σ  (r )
σ (1.35)
= σ (r) σ (r) = (x) (x),
† †
8 1 Second Quantization

which also shows that (x)† is nothing but the operator which creates a particle at
x, while (x) destroys it.
Let us continue and consider a two-particle operator

1 
Û = U (xi , x j ). (1.36)
2
i=j

We can simply repeat the previous steps and find that

n 1 , n 2 , . . . | Û | n 1 , n 2 , . . .
 N
1  ∗
= d xi U (xi , x j ) {n  (x 1 , . . . , x N ) {n a } (x 1 , . . . , x N )
2 a}
i=j i=1
 
N
N (N − 1) ∗
= d xi {n  } (x 1 , . . . , x N ) U (x 1 , x 2 ) {n a } (x 1 , . . . , x N )
2 a
i=1
 
N
N (N − 1)
= d xi U (x1 , x2 )
N!
i=1
n 1 , n 2 , . . . |  † (x1 )  † (x2 )  † (x3 ) . . .  † (x N ) | 0
0 | (x N ) . . . (x3 ) (x2 ) (x1 ) | n 1 , n 2 , . . . ,

which, since,

N (N − 1)
d x3 . . . d x N  † (x3 ) . . .  † (x N ) | 0 0 | (x N ) . . . (x3 ) = I N −2 ,
N!

leads to the second quantised expression


1
Û = d x1 d x2  † (x1 )  † (x2 ) U (x1 , x2 ) (x2 ) (x1 )
2
1  (1.37)
= Uabcd ca† cb† cc cd ,
2
abcd

where

Uabcd = d x d y φa (x)∗ φb (y)∗ U (x, y) φc (y) φd (x) .

Analogously, a generic m-particle operator

1 
Ûm = U (x1 , x2 , . . . , xm )
m!
i 1 =i 2 =··· =i m
1.3 Bosonic Operators 9

translates in second quantisation into


1
Ûm = d x1 . . . d xm  † (x1 ) . . .  † (xm ) U (x1 , . . . , xm ) (xm ) . . . (x1 ) .
m!
(1.38)
We conclude by emphasising the advantages of second quantisation with respect
to first quantisation. In the former, any multiparticle operator depends explicitly on
the particle coordinates. It is only the wavefunction that contains information about
the indistinguishability of the particles as well as their statistics. On the contrary,
in second quantisation, those properties are hidden in the definition of creation and
annihilation operators; hence, the multiparticle operators do not depend anymore on
the particle coordinates. Moreover, in second quantisation, we can also introduce
operators which have no first-quantisation counterpart. For instance, we can define
particle-non-conserving operators which connect subspaces with different numbers
of particles. For instance,
 
∗
ab ca† cb† + ab cb ca
ab

is a Hermitian operator that connects Fock states with particle numbers differing by
two. Those operators are useful when discussing superconductivity.

1.3 Bosonic Operators

As previously done for the fermionic case, we introduce the creation, da† , and its
Hermitian conjugate, the annihilation da , operators which, respectively, create and
destroy a boson in state a. Again the operator da† da counts how many times we can
destroy and create back a boson in state a; hence, it is just the occupation number
n a . However, since the Pauli principle does not hold for bosons, the operator da da†
first adds one boson in state a, next increases n a → n a + 1, and finally destroys one
boson in that same state. This latter process can be done n a + 1 times, being n a + 1
the actual occupation number once one more boson has been added. Therefore,

da† da |n a  = n a | n a , da da† | n a  = 1 + n a | n a  ; (1.39)

Hence, the following commutation relation holds:

 
da , da† = da da† − da† da = 1, (1.40)

where [. . . ] denotes the commutator. Equation (1.39) is satisfied, e.g., by

√ 
da | n a  = n a | n a − 1, da† | n a  = n a + 1 | n a + 1 , (1.41)
10 1 Second Quantization

which we assume hereafter. The permanent, contrary to the Slater determinant, is

invariant upon interchanging two quantum labels. As a consequence, bosonic oper-
ators corresponding to different states commute instead of anti-commuting as the
fermionic ones. Namely, for a = b,
   
da† , db = da† , db† = 0.

Therefore, in general,
     
da , db† = δab , da† , db† = da , db = 0. (1.42)

Moreover, through (1.41) and (1.42), we can write a generic Fock state as

 d † ni
|n 1 , n 2 , . . .  = i | 0 . (1.43)
i ni !

1.3.1 Bose Fields and Multiparticle Operators

The analogous role of the Fermi fields is now played by the Bose fields defined
through
 
(x) = φa (x) da , (x)† = φa (x)∗ da† , (1.44)
a a
which satisfy the commutation relations
     
(x), (y)† = δ(x − y) , (x), (y) = (x)† , (y)† = 0 . (1.45)

Exactly like in the fermionic case, if we define

1
| x1 , . . . , x N  ≡ √ † (x1 ) . . . † (x N ) | 0 , (1.46)
N!

we find that

d x1 . . . d x N | x1 , . . . , x N  x1 , . . . , x N |

1
= d x1 . . . d x N † (x1 ) . . . † (x N ) | 0 0 | (x N ) . . . (x1 ) (1.47)
N!
1  †
= d . . . da†N | 0 0 | da N . . . da1 ≡ I N ,
N ! a ,...,a a1
1 N
1.4 Canonical Transformations 11

as well as that the permanent with occupation numbers {n a } ≡ (n 1 , n 2 . . . ) is just

{n a } (x1 , . . . , x N ) = x1 , . . . , x N | n 1 , n 2 , . . . 
1 (1.48)
=√ 0 | (x N ) . . . (x1 ) | n 1 , n 2 , . . . .
N!
These two equations imply that the second quantised expressions of the operators
are simply the same as in the fermionic case; for instance, a single-particle operator
is

V̂ = d x (x)† V (x) (x) , (1.49)

a two-particle one

1
Û = d x d y (x)† (y)† U (x, y) (y)(x) , (1.50)
2
and an m-particle operator

1
Ûm = d x1 . . . d xm (x1 )† . . . (xm )† U (x1 , . . . , xm ) (xm ) . . . (x1 ) .
m!
(1.51)

1.4 Canonical Transformations

In general, an interacting Hamiltonian that contains besides bilinear also quartic and
higher order terms in creation and annihilation operators cannot be diagonalised. On
the contrary, a bilinear Hamiltonian is diagonalisable by a canonical transformation
that preserves the commutation/anti-commutation properties of the operators.
Since the interaction is commonly analysed perturbatively starting from an appro-
priate non-interacting theory, it is useful to begin with bilinear Hamiltonians and
introduce the canonical transformations that diagonalise them.
The simplest bilinear Hamiltonian is the second quantised expression of a non-
interacting first-quantisation Hamiltonian, which has the general form

H= tab ca† cb , (1.52)
ab

both for fermions and bosons. Since H is Hermitian, then

∗
tab = tba .

If we define a column vector c, i.e., a spinor, with components ca , its Hermitian con-
jugate c† , a row vector, and the Hermitian matrix tˆ with elements tab , the Hamiltonian
can be shortly written as
H = c† tˆ c . (1.53)
12 1 Second Quantization

Since tˆ is Hermitian, there exists a unitary transformation Û , i.e., satisfying Û † Û =

Û Û † = Iˆ, with Iˆ the identity matrix, such that

tˆ = Û † Ê Û ⇒ Ê = Û tˆ Û † , (1.54)

where Ê is diagonal with real elements a. Therefore, if we define a new spinor

operator d through
d ≡ Û c , (1.55)
with elements

da = Uab cb , (1.56)
b
then

H = d † Ê d = †
a da da (1.57)
α
is diagonal. We have now to check whether the above is a canonical transformation.
If the ca ’s are fermionic/bosonic operators, then
     
da , db† = †
Uan Umb cn , cm
†
= †
Uan Unb = δab ,
± ±
nm n

where {. . . }± stands for the anti-commutator (+) and commutator (-), respectively.
Therefore, also the da ’s are fermions/bosons; hence, Û is indeed canonical.
Once the Hamiltonian has been transformed into the diagonal form (1.57), the
problem is solved. Indeed any Fock state constructed through the new basis set with
operators da , namely a wavefunction | {n a } where each state a is occupied by n a
da -particles, is an eigenstate of the Hamiltonian
 

H | {n a } = a na | {n a }. (1.58)
a

It is important to remark that in the case of bosons the eigenvalues a must all be
greater or equal to zero; otherwise, the Hamiltonian would be ill-defined since the
ground state would correspond to putting an infinite number of bosons in the most
negative energy state, thus with infinitely negative total energy. The condition a ≥ 0
implies that tˆ must be semi positive definite.
An equivalent way to implement the canonical transformation (1.55) is through
a unitary operator
† φ̂
U = ei c c
≡ eiϕ , (1.59)
1.4 Canonical Transformations 13

where φ̂ = φ̂ † is Hermitian and thus U † U = 1 unitary. We recall that, given two

operators A and B, then

 (−i)n  
e−i A B ei A = Cn , C0 = B , Cn = A, Cn−1 , (1.60)
n!
n≥0

which can be readily demonstrated by expanding the exponentials, so that, in the

specific case of U † c U
 
C0 = c , C1 = ϕ, c = −φ̂ c , Cn = (−1)n φ̂ n c , (1.61)

namely

 in
U† c U = φ̂ n c = ei φ̂ c , U † c† U = c† e−i φ̂ , (1.62)
n!
n≥0

which coincides with (1.55) if ei φ̂ = Û , and thus e−i φ̂ = Û † . In this case

U † c U = Û c , U † c† U = c† Û † , (1.63)

which also imply

U c U † = Û † c , U c† U † = c† Û . (1.64)
In this formulation, the diagonalization of the Hamiltonian means the following.
Since the eigenstates and eigenvalues of a Hermitian operator do not change under a
unitary transformation, then we are allowed to study, instead of H , the transformed
Hamiltonian H̃ ≡ U H U † , which, through (1.64) and noting that Û tˆ Û † = Ê, can
be written as

H̃ ≡ U H U † = U c† tˆ c U † = U c† U † tˆ U c U † = c† Û tˆ Û † c = c† Ê c,
(1.65)
and therefore, it is diagonal.
Conversely, H = U † H̃ U , so that its eigenstates are simply U † | {n a }, where
| {n a } is a Fock state. Indeed
 

H U † | {n a } = U † H̃ | {n a } = a na U † | {n a } , (1.66)
a

which coincides with (1.58). In this language, the partition function

     
Z ≡ Tr e−β H = Tr U † U e−β H = Tr U e−β H U †
    (1.67)
= Tr e−β U H U = Tr e−β H̃ ≡ Z̃
†
14 1 Second Quantization

is evidently invariant under the unitary transformation, and the thermal average of
any operator A can be simply calculated through the diagonal Hamiltonian H̃ with
A transformed into U A U † ,

1  −β H  1   1  
A = Tr e A = Tr U † U e−β H A = Tr e−β H̃ U A U † .
Z Z Z̃
(1.68)

1.4.1 Canonical Transformations with Charge Non-conserving

Hamiltonians

The above results show that a unitary transformation of the fermionic/bosonic oper-
ators ca preserves the anti-commutation/commutation relations. Indeed, if

ca → U † ca U , ca† → U † ca† U , (1.69)

with any unitary operator U , i.e., such that U † U = U U † = 1, then

   
U † ca U , U † cb† U = U † ca , cb† U = δab ,
± ±
    (1.70)
U † ca U , U † cb U = U † ca , cb U = 0.
± ±

In second quantisation, we have the opportunity to introduce bilinear Hamiltonians

that does not conserve the number of particles:
   
H= tab ca† cb + ab ca cb + ab cb ca = c† tˆ c + cT ˆ c + H .c.
∗ † †

ab
(1.71)
and are relevant for a wide class of physical problems. Again tˆ = tˆ† , while ˆ is
antisymmetric for fermions and symmetric for bosons. Since H is bilinear in single-
particle operators, it can always be diagonalised by a proper unitary transformation
(1.69). However, if we write as before U = eiϕ , now the Hermitian operator ϕ must
be of the form
 
ϕ = c† φ̂ c + cT ψ̂ c + H .c. , (1.72)

with φ̂ = φ̂ † and ψ̂ antisymmetric/symmetric for fermions/bosons. In other words,

ϕ is still bilinear but does not conserve the number of particles. It is always possible
to find φ̂ and ψ̂ such that

U † H U = c† Ê c ≡ H̃ , (1.73)

 that, given a Fock state | {n a }, then

with Ê diagonal with real elements a . It follows
U | {n a } is eigenstate of H with eigenvalue a a n a . As previously discussed, in
the case of bosons a ≥ 0, which poses constraints on tˆ and ˆ .
1.4 Canonical Transformations 15

We just note  that, while | {n a } is eigenstate of the number of particles with

eigenvalue N = a n a , the eigenstate U | {n a } = eiϕ | {n a } of the Hamiltonian

is not, since ϕ does not commute with the number of particle operator a ca† ca .
In the case of fermions, the diagonalization of the Hamiltonian (1.71) can still be
recast to that of a Hermitian matrix. Indeed, if we introduce the new spinors
   
c
ψ≡ , ψ † = c† , cT ,
c†T

the Hamiltonian (1.71) can be rewritten, dropping constant terms, as

 
tˆ/2 ˆ †
H =ψ †
ˆ −tˆT /2 ψ = ψ † Ĥ ψ ,

where now Ĥ is Hermitian and thus has real eigenvalues. We further note that, since
tˆ is Hermitian, then tˆT = tˆ∗ , while, since ˆ is antisymmetric, then ˆ † = − ˆ ∗ , so
that
       
tˆ/2 ˆ † tˆ/2 − ˆ ∗ 0 Iˆ ∗ 0 Iˆ
Ĥ = ˆ = ˆ =− ˆ Ĥ ≡ −σ̂1 Ĥ ∗ σ̂1 ,
−tˆT /2 −tˆ∗ /2 I 0 Iˆ 0

where σ̂1 is a generalised first Pauli matrix, thus σ̂1 Ĥ σ̂1 = − Ĥ ∗ . Therefore, if
 
uλ
χλ = : Ĥ χλ = λ χλ , λ ∈ R ,
vλ

so that Ĥ ∗ χλ∗ = λ χλ∗ , then

Ĥ σ̂1 χλ∗ = σ̂1 σ̂1 Ĥ σ̂1 χλ∗ = −σ̂1 Ĥ ∗ χλ∗ = −λ σ̂1 χλ∗ .

In other words,
 
v∗
σ̂1 χλ∗ = λ∗ ≡ χ−λ
uλ
is the eigenstate with opposite eigenvalue. Therefore, the unitary matrix
     
u λ1 u λ2 v∗λ1 v∗λ2
Û = ... ...
vλ 1 vλ 2 u∗λ1 u∗λ2

brings the Hamiltonian into the diagonal form

 
λ̂ 0
Û † Ĥ Û = ,
0 −λ̂
16 1 Second Quantization

where λ̂ is diagonal with components λi , i = 1, 2, . . . , which also implies that the

canonical transformation
 
d
ψ = Û
d †T
allows to rewrite H in (1.71) as
  λ̂ 0   d    
H = ψ Ĥ ψ = d , d
† †
†T = λi 2di† di − 1 ,
0 −λ̂ d
i

namely in a diagonal form.

1.4.2 Harmonic Oscillators

When dealing with bosons, bilinear Hamiltonians very often reduce to coupled har-
monic oscillators, like, e.g., in the case of lattice vibrations in the harmonic approx-
imation. In those cases, it is more convenient to apply canonical transformations
directly to the conjugate variables rather than to the bosonic creation and annihila-
tion operators. For that, let us consider the prototypical example of a single harmonic
oscillator with Hamiltonian

2 2 g 2
H= p + x , (1.74)
2m 2
 
with x and p conjugate variables, i.e., x, p = i. We consider the transformation

√ 1
x= K X, p=√ P, (1.75)
K

which evidently preserves the commutation relations and therefore is canonical. In

terms of the new variables

2 Kg 2
H= P2 + X . (1.76)
2m K 2
If we fix K such that
2 2
= K g ⇒ K2 = , (1.77)
mK mg
we readily find that

 g ω
H= P2 + X2 ≡ P2 + X2 . (1.78)
2 m 2
1.5 Application: Electrons in a Box 17

Hereafter, we denote the above expression as the ‘canonical form’ of the Hamiltonian.
Indeed, if we now introduce bosonic creation and annihilation operators through

1 i
X = √ a + a† , P = −√ a − a† , (1.79)
2 2

the Hamiltonian becomes

 
1
H = ω a† a + , (1.80)
2

i.e., the standard expression of a harmonic oscillator Hamiltonian with frequency ω.

We now consider the unitary operator
 
α
U = exp − i x p+ px ,
2

whose action on the conjugate variables can be readily found to be

U † x U = eα x , U † x U = e−α p ,

which yields the canonical transformation (1.75) if e2α = K . In other words,

ω
U† H U = p2 + x 2
2
directly brings the Hamiltonian in the canonical form. The eigenstates are therefore
U | n, where | n is the Fock state with n bosons.

1.5 Application: Electrons in a Box

Let us consider electrons in a square box of linear length L with periodic boundary
conditions. As a basis of single-particle wavefunctions, we use simple plane waves
and spin, namely the quantum label a = (k, σ ) with
18 1 Second Quantization

2π
k= nx , n y , nz ,
L
being n i ’s integers, and σ =↑, ↓. Hence, the single-particle wavefunctions of the
basis set are
1
φa (x) = √ eik·r χσ .
L3
The annihilation and creation operators are defined as ck σ and ck† σ , respectively;
hence, the Fermi fields are

1  ik·r 1  −ik·r †
σ (r) = √ e ck σ , σ† (r) = √ e ck σ . (1.81)
L3 k L3 k

The kinetic energy in the first quantisation reads


N
2 2
Hkin = − ∇
2m i
i=1

and is diagonal in the spin. Since

  
1 −ik·r 2 2 2 k 2
dr e − ∇ eip·r = δk p ,
L3 2m 2m

then, on the plane-wave basis, the kinetic energy in the second quantisation becomes

 2 k 2 †  †
Hkin = c k σ ck σ ≡ k c k σ ck σ . (1.82)
2m
kσ kσ

The ground state with N , assumed to be even, electrons is the Fermi sea |F S obtained
by filling with spin-up and spin-down electrons the momentum states from k = 0 up
to the Fermi momentum k F defined through

N =2 ,
|k|≤k F

namely

|F S = ck† ↑ ck† ↓ | 0 ; (1.83)
k: |k|≤k F

hence, the occupation number in momentum space is

n k ↑ = n k ↓ = θ (k F − |k|) ,

where the θ -function is defined through θ (x) = 1 if x ≥ 0, otherwise θ (x) = 0.

1.5 Application: Electrons in a Box 19

Let us add an electron-electron interaction of the general form

1
Hint = U (ri − r j ) .
2
i=j

In second quantisation, it becomes


1
Hint = dx dy σ† (x)σ†  (y) U (x − y) σ  (y)σ (x)
2 
σσ
1  
= ck†1 σ ck†2 σ  ck3 σ  ck4 σ
2L 3 
σ σ ki ,i=1,...,4

1
dx dy e−ik1 ·x e−ik2 ·y eik3 ·y eik4 ·x U (x − y) .
L3

We define the Fourier transform as


U (q) = dr e−iq·r U (r) ,

so that
1  iq·(x−y)
U (x − y) = e U (q) ,
L3 q

and the interaction is finally found to be

1 
Hint = U (q) ck† σ cp+q
†
σ  cp σ  ck+q σ . (1.84)
2L 3 σσ k pq

We showed that the electron density for spin σ in second quantization is

ρσ (r) = σ† (r) σ (r) .

Its Fourier transform is

 
ρσ (q) = dr e−iq·r σ† (r) σ (r) = ck† σ ck+q σ .
k

Let us consider interacting electrons also in the presence of a single-particle potential,

provided, e.g., by the ions,

H pot = dx V (x) ρσ (x) ,
σ
20 1 Second Quantization

so that the total Hamiltonian reads

H = Hkin + Hint + H pot .

Since the Hamiltonian conserves the total number of electrons, the electron density
summed over the spins, i.e., ρ = ρ↑ + ρ↓ must satisfy a continuity equation. Let us
define the Heisenberg evolution of the density through

ρ(r, t) = ei H t ρ(r) e−i H t .

Since the integral of the density over the whole volume is the total number of elec-
trons, N , which is conserved, it follows that
 
∂ρ(r, t) ∂ ∂N
dr = dr ρ(r, t) = ≡ 0.
∂t ∂t ∂t

This condition is satisfied if

∂ρ(r, t)
= −∇J(r, t),
∂t

where J(r, t) is the current density operator. In fact, the integral over the volume of
the left-hand side is also equal to minus the flux of the current out of the surface
of the sample. If the number of electrons is conserved, it means that the flux of the
current through the surface vanishes, which is the desired result. The equation

∂ρ(r, t)
+ ∇J(r, t) = 0 , (1.85)
∂t
is the continuity equation associated to the number of particles, which is a conserved
quantity. Similar equations can be derived for any conserved quantity.

1.6 Application: Electron Lattice Models and Emergence of

Magnetism

Let us consider the electron Hamiltonian in presence of the periodic potential pro-
vided by the ions in a lattice:


N
pi 2   1
H= + V (ri − R) + U (ri − r j )
2m 2 (1.86)
i=1 i R i=j
= Hkin + Hel−ion + Hel−el = H0 + Hel−el ,
1.6 Application: Electron Lattice Models and Emergence of Magnetism 21

where R are lattice vectors. We start by rewriting the Hamiltonian in the second
quantisation. For that purpose, we need to introduce a basis set of single-particle
wavefunctions. Since the Hamiltonian is spin independent, it is convenient to work
with factorised single-particle wavefunctions: φ(r, σ ) = φ(r) χσ . As a basis for the
space-dependent φ(r), we use Wannier orbitals φn,R (r) satisfying
 
dr φn,R1 (r)∗ φm,R2 (r) = δnm δR1 R2 , φn,R (r)∗ φn,R (r ) = δ(r − r ) ,
n,R

as well as
φn,R+R0 (r) = φn,R (r − R0 ) . (1.87)
†
Consequently, we associate to any wavefunction φn,R (r) χσ creation, cn,R,σ , and
annihilation, cn,R,σ , operators and introduce the Fermi fields

  †
σ (r) = φn,R (r) cn,R,σ , σ† (r) = σ (r) .
n,R

Let us start by second quantization of the non-interacting part of the Hamiltonian,

H0 in (1.86):
 
 2 ∇ 2 
H0 = dr σ (r) −
†
+ V (r − R) σ (r)
σ
2m
R
 
= tRnm c† c
1 ,R2 n,R1 ,σ m,R2 ,σ
.
σ nm R1 ,R2

The matrix elements are

  
2 ∇ 2 
tRnm
1 ,R2
= dr φn,R1 (r)∗ − + V (r − R) φm,R2 (r) (1.88)
2m
R

and satisfy
∗
tRnm
1 ,R2
= tRmn
2 ,R1
. (1.89)
By the property (1.87), it follows that
  
2 ∇ 2 
∗
tRnm = dr φn,R1 +R0 (r) −
1 +R0 ,R2 +R0
+ V (r − R) φm,R2 +R0 (r)
2m
R
  
∗ 2 ∇ 2 
= dr φn,R1 (r − R0 ) − + V (r − R) φm,R2 (r − R0 )
2m
R
  
∗ 2 ∇ 2 
= dr φn,R1 (r) − + V (r + R0 − R) φm,R2 (r) = tRnm1 ,R2
,
2m
R
22 1 Second Quantization

since
 
V (r + R0 − R) = V (r − R) .
R R
Let us now introduce the operators in the reciprocal lattice through the canonical
transformation
1  −ik·R
cn,k,σ = √ e cn,R,σ , (1.90)
V R
and its inverse
1  ik·R
cn,R,σ = √ e cn,k,σ , (1.91)
V k
where V is the number of lattice sites and k belongs to the reciprocal lattice, namely
 
eik·R = V δk0 , e−ik·R = V δR0 .
R k

One can check that the above transformation is indeed canonical:

  1  −ik1 ·R1 ik2 ·R2  
† †
cn,k1 ,σ1 , cm,k2 ,σ2
= e e c , c
n,R1 ,σ1 m,R2 ,σ2
V
R1 ,R2
1  −i(k1 −k2 )·R1
= δnm δσ1 σ2 e = δnm δσ1 σ2 δk1 k2 .
V
R1

Let us substitute (1.91) into (1.88):

1     −ik1 ·R1 ik2 ·R2 nm †

H0 = e e tR1 ,R2 cn,k c
1 ,σ m,k2 ,σ
.
V σ nm
R1 ,R2 k1 ,k2

We observe that
 1 
e−ik1 ·R1 eik2 ·R2 tRnm
1 ,R2
= e−ik1 ·R1 eik2 ·R2 tRnm
1 −R0 ,R2 −R0
V
R1 ,R2 R1 ,R2 ,R0
1 
= e−ik1 ·(R1 +R0 ) eik2 ·(R2 +R0 ) tRnm
1 ,R2
V
R1 ,R2 ,R0
 1  −i(k1 −k2 )·R0
= e−ik1 ·R1 eik2 ·R2 tRnm
1 ,R2
e
V
R1 ,R2 R0

= δk1 k2 e−ik1 ·(R1 −R2 ) tRnm
1 ,R2
= V δk1 k2 tknm
1
,
R1 ,R2
1.6 Application: Electron Lattice Models and Emergence of Magnetism 23

where we define
1  −ik·(R1 −R2 ) nm 
tknm ≡ e tR1 ,R2 = e−ik·R tR,0
nm
. (1.92)
V
R1 ,R2 R

Since (1.89) holds, it also follows that

 ∗   ∗  
tknm = eik·R tR,0
nm
= eik·R t0,R
mn
= eik·R t−R,0
mn

R R R
 (1.93)
= e−ik·R tR,0
mn
= tkmn ,
R

namely the matrix tˆk , with elements tkmn , is Hermitian. The Hamiltonian is therefore
 †
H0 = tknm cn,k,σ cm,k,σ . (1.94)
σ nm k

Since tˆk is Hermitian, we can write

tˆk = Û (k)† ˆk Û (k) → tknm = U † (k)ni i,k U (k)im ,

with Û (k)† Û (k) = Iˆ. Therefore, upon applying the canonical transformation

ci,k,σ = U (k)in cn,k,σ ,
n

the non-interacting Hamiltonian acquires a diagonal form

 †
H0 = i,k ci,k,σ ci,k,σ . (1.95)
σ i k

The index i identifies the band, and i,k is the energy dispersion in the reciprocal
lattice: we have thus obtained the band structure.
We can formally write
1  −iR·k
ci,k,σ = √ e ci,R,σ ,
V R

thus introducing a new basis of Wannier functions. Since the Fermi field is invariant
upon the basis choice, then
 1 
σ (r) = φn,R (r) χσ cn,R,σ = √ φn,R (r) χσ eik·R cn,k,σ
R,n
V R,n,k
1 
= √ φn,R (r) χσ eik·R U † (k)ni ci,k,σ
V R,n,i,k
1   
= φn,R (r) χσ eik·(R−R ) U † (k)ni ci,R ,σ ≡ φi,R (r) χσ ci,R ,σ ,
V
R,R ,n,i,k R
24 1 Second Quantization

thus implying the following expression of the new Wannier functions:

1  ik·(R−R ) †
φi,R (r) = e U (k)ni φn,R (r) . (1.96)
V
n,R,k

Going back to the Hamiltonian in the diagonal basis, we can also rewrite it as
 †
H0 = i,k ci,k,σ ci,k,σ
σ i k
1   ik·(R1 −R2 ) †
= i,k e ci,R1 ,σ ci,R2 ,σ (1.97)
V σ
i k R1 ,R2
  †
≡ tRi 1 ,R2 ci,R c
1 ,σ i,R2 ,σ
,
σ i R1 ,R2

namely like a tight-binding Hamiltonian diagonal in the band index. Once we know
the band structure, the ground state of the non-interacting Hamiltonian is simply
obtained by filling all the lowest bands with the available electrons. If the highest
occupied band is full, the model is a band insulator, otherwise is a metal. In particular,
since each band can accommodate 2V electrons, V of spin up and V of spin down,
a necessary condition for a band insulator is to have an even number of available
electrons per unit cell. This is not sufficient since the bands may overlap.

1.6.1 Hubbard Models

Hereafter, we consider the case of a putative metal, in which the highest occupied
bands are partly filled. In many physical situations, the Wannier orbitals φi,R (r) of
(1.96) for those bands, the conduction ones, are quite delocalised; hence, the lattice
vector label R looses its physical meaning. In those cases, although formally exact,
the tight-binding Hamiltonian (1.97) is of little use since the hopping matrix elements
tRi 1 ,R2 are very long ranged.
However, there is a wide class of materials, commonly called strongly correlated
ones, where the tight-binding formalism is meaningful. There, the valence bands
derive from d or f orbitals of transition metals, rare earth or actinides, and the Wan-
nier orbitals keep noticeable atomic character, thus leading to short-range hopping
elements tRi 1 ,R2 in (1.97). Let us consider just the above circumstance and write down
the left-over electron-electron interaction on the Wannier basis. We further assume
that there is only one conduction band well separated from lower and higher ones, so
that we can safely neglect interband transition processes due to interaction and just
project the latter onto the conduction band. For that reason, we hereafter drop the
band index i, so that the Wannier orbitals of the conduction band are simply denoted
as φR (r), and the tight-binding Hamiltonian projected on the same band reads
  †
H0 = tR1 ,R2 cR c
1 σ R2 σ
. (1.98)
σ R1 ,R2
1.6 Application: Electron Lattice Models and Emergence of Magnetism 25

The interaction term is, correspondingly,

1  †
Hint = UR1 ,R2 ;R3 ,R4 cR c† c c
1 σ1 R2 σ2 R3 σ2 R4 σ1
, (1.99)
2σσ
1 2 R1 ,R2 ,R3 ,R4

where

UR1 ,R2 ;R3 ,R4 = drdr φR1 (r)∗ φR2 (r )∗ U (r − r ) φR3 (r ) φR4 (r) . (1.100)

Let us make use of our assumption of well-localised Wannier orbitals, namely that
φR (r) decays sufficiently fast with |r − R|. Within this assumption, the leading
matrix element is when all lattice sites are the same:

UR,R;R,R ≡ U . (1.101)

This term gives rise to an interaction

U  † † 
HU = cRσ cRσ  cRσ  cRσ = U n R↑ n R↓ , (1.102)
2  R σσ R

since cRσ cRσ = 0 by the Pauli principle, where the operator

†
n Rσ = cRσ cRσ ,

counts the number of spin-σ electrons at site R.

The interaction term (1.102), which simply describes an on-site Coulomb repul-
sion, plus the hopping (1.98) yield the so-called Hubbard model [1,2], which is the
prototype of strongly correlated lattice models.
Let us continue and consider in (1.100) two other cases: either (1) R1 = R4 ,
R2 = R3 with R2 nearest neighbour of R1 or (2) R1 = R3 , R2 = R4 still with R2
nearest neighbour of R1 . In case (1), we obtain

U1 n R n R ,
<RR >

where n R = σ n Rσ , < RR > stands for the sum over nearest neighbour sites,
and U1 = UR,R ;R ,R . In case (2), we instead find
  † †
  † †
U2 cRσ cR  σ  cRσ  cR σ = −U2 cRσ cRσ  cR  σ  cR  σ ,
<RR > σ σ  <RR > σ σ 

where U2 = UR,R ;R,R . One can easily show that

 † † 1 
cRσ cRσ  cR  σ  cR  σ = n R n R + 4 SR · SR , (1.103)
2
σσ
26 1 Second Quantization

where the spin-density operator

1 †
SR = cRα σ αβ cRβ ,
2
αβ

being σ = (σ1 , σ2 , σ3 ), with σi ’s the Pauli matrices. Therefore, upon defining V =

U1 − U2 /2 and Jex = −2U2 , the two nearest neighbour interaction terms lead to
 
Hn.n. = V n R n R + Jex SR · SR . (1.104)
<RR > <RR >

Since U1 > U2 > 0, the first term describes a nearest neighbour repulsion, while
the second a spin exchange which tends to align the spin ferromagnetically since
Jex < 0, so called direct exchange. Therefore, although we have started from a spin-
independent interaction, projecting onto the Wannier basis makes a spin interaction
emerge, thus showing in a simple way how magnetism raises out of the charge
Coulomb repulsion.

1.6.2 Mott Insulators and Heisenberg Models

Let us summarise the approximate Hubbard Hamiltonian which we have so far

derived, by further assuming a nearest neighbour hopping:
   †
 
H = −t cRσ cR σ + H .c. + U n R↑ n R↓
σ <RR > R
  (1.105)
+V nR n R + Jex SR · S
R ≡ H0 + Hint .
<RR > <RR >

By construction U > V > 0 and Jex < 0. We consider the case in which the number
of conduction electrons is equal to the number of sites N , i.e., density equal to half-
filling. In the absence of interaction, the hopping forms a band that can accommodate
2N electrons while there are just N of them: the band is therefore half-filled and the
system metallic.
Let us analyse the opposite case of a very large U  t, V , |Jex |. In this case, it is
better to start from the configuration which minimises the Coulomb repulsion U and
treat what is left in perturbation theory. That lowest energy electronic configuration
is the one in which each site is singly occupied. Indeed, the energy cost in having
just an empty site and a doubly occupied one instead of two singly occupied sites is
given by
E(2) + E(0) − 2E(1) = U ,
and it is much larger than the hopping energy t. In this situation, the model describes
an insulator but of a particular kind. Namely, the insulating state is driven by the
strong correlation, while the conventional counting argument instead predicts a metal.
1.6 Application: Electron Lattice Models and Emergence of Magnetism 27

This correlation-induced insulator is called a Mott insulator after Sir Nevil Mott [3].
Therefore, as the strength of U increases with respect to the bandwidth, which is pro-
portional to t, an interaction-driven metal-to-insulator transition, commonly referred
to as a Mott transition, must occur at some critical Uc .
However, the configuration with one electron per site is hugely degenerate, since
the electron can have either spin up or down, thus a degeneracy 2 N that is going to be
split by the other terms in the Hamiltonian. The nearest neighbour interaction V is
not effective in splitting the degeneracy, contrary to the direct exchange that favours
a ferromagnetic ordering.
However, this is not the only source of spin correlations. Indeed, also the hop-
ping term is able to split the degeneracy within second order in perturbation theory.
Specifically, if P is the projector onto the degenerate ground state manifold with
energy E 0 = 0, at second order, the hopping Hamiltonian H0 generates an effective
spin exchange operator, called super-exchange.
 1
Hs−ex = P H0 | n n | H0 P , (1.106)
n
E0 − En

where | n is an intermediate excited state with energy E n . Because of the projector,

the only allowed hopping processes are those where one electron hops to a nearest
neighbour site and then hops back to the initial site. It follows that the intermediate
states allowed in the sum are just those where one site is doubly occupied and a
nearest neighbour one empty, thus E n = U independent of n and

1  1
Hs−ex = − P H0 | n n | H0 P = − P H0 H0 P
U n U
2    
t † † † †
=− cRσ cR  σ c R c
 σ  Rσ  + c c c c
R σ Rσ Rσ  R σ  (1.107)
U  
σ σ <RR >
2t 2   † †
=− cRσ cR σ cR  σ  cRσ  ,
U  
σ σ <RR >

which, through (1.103) and since n R = 1, is simply

4t 2   1

Hs−ex = SR · SR − . (1.108)
U 4
<RR >

Therefore, the second-order perturbation theory in the hopping gives rise to an anti-
ferromagnetic spin super-exchange. In conclusion, the effective Hamiltonian at large
U describes localised spin-1/2 (each site occupied by one electron) coupled by the
overall spin exchange (dropping constant terms)

H H eis = J SR · SR , (1.109)
<RR >
28 1 Second Quantization

where
4t 2 4t 2
J = Jex += −2U2 + .
U U
The spin exchange may be either ferromagnetic or antiferromagnetic depending on
the strength of the direct-exhange constant with respect to the super-exchange one.
The effective Hamiltonian (1.109) thus corresponds just to a spin-1/2 Heisenberg
model.

1.7 Application: Spin-Wave Theory

When the hopping involves further neighbour sites and more than a single band, as
it is commonly the case for magnetic ions, the corresponding Heisenberg model has
the general expression

H= JR,R SR · SR . (1.110)
RR
The Hamiltonian to be Hermitian requires that JR,R = JR ,R ∈ R, while transla-
tional symmetry that

JR,R = JR−R0 ,R −R0 , ∀ R0 . (1.111)

For instance, taking R0 = R or R0 = R, and using the property JR,R = JR ,R ,

JR,R = JR−R ,0 = J0,R −R = JR −R,0 . (1.112)

We can introduce Fourier transforms defined through

 1  iq·R
S(q) = e−iq·R SR , SR = e S(q) , (1.113)
N q
R

where N is the number of lattice sites, and note that

1  −iq·R iq ·R 1  1  −iq·R iq ·R

Jq,q = e e JR,R = e e JR,R
N 
N N 
RR R0 RR
1  1  −iq·R iq ·R
= e e JR−R0 ,R −R0
N N 
R0 RR
1  −i(q−q )·R0 1  −iq·R iq ·R
= e e e JR,R
N N 
R0 RR
1  −iq·(R−R ) 
= δq,q e JR,R = δq,q e−iq·R JR,0 ≡ δq,q J (q) .
N
RR R
(1.114)
1.7 Application: Spin-Wave Theory 29

It follows that the Hamiltonian in Fourier space is simply

1 
H= J (q) S(−q) · S(q) . (1.115)
N q

Moreover, since SR = S†R ,


S(q)† = eiq·R SR = S(−q) , (1.116)
R

and, since JR,R ∈ R,

  
J (q)∗ = eiq·R JR,0 = J (−q) = eiq·R JR−R,0−R = eiq·R J0,−R
R R R
 
−iq·R
= eiq·R
J−R,0 = e J R,0 = J (q) ,
R R
(1.117)
namely J (q) = J (−q) ∈ R. In conclusion,

1 
H= J (q) S(q)† · S(q) . (1.118)
N q

1.7.1 Classical Ground State

Let us first assume that the spin SR is a classical vector satisfying

SR · SR = S 2 , ∀ R , (1.119)

so that the Hamiltonian becomes

1 
H= J (q) S(q)∗ · S(q) . (1.120)
N q

The classical ground state is actually identified by a vector Q and a classical spin
configuration

S(Q) = S(−Q)∗ = 0 , S(q) = 0 , ∀ q = (Q, −Q) . (1.121)

To prove that, let us assume it is indeed the case, and first identify what Q is. Under
that assumption, and if Q = −Q, where the equality holds apart from a reciprocal
lattice vector G, then
 
1 ∗ −iQ·R
SR = S(Q) e iQ·R
+ S(Q) e . (1.122)
N
30 1 Second Quantization

In order to verify (1.119), We have to impose that

S(Q) · S(Q) = S(−Q) · S(−Q) = 0 ,

N2 2 (1.123)
S(Q)∗ · S(Q) = S(−Q) · S(Q) = S ,
2
which can be satisfied by taking any two orthogonal real unit vectors, u1 and u2 ,
thus ui · u j = δi j , and writing

SN SN
S(Q) = u1 − i u2 , S(−Q) = S(Q)∗ = u1 + i u2 . (1.124)
2 2
With that choice
SR = S u1 cos Q · R + S u2 sin Q · R . (1.125)
If instead Q = −Q + G, i.e., 2Q = G, then S(Q) = S(−Q) ∈ R, so that, since

1
Q·R = G · R ≡ π n R (G) , (1.126)
2
with n R (G) integer, then

1 1
SR = S(Q) cos Q · R = S(Q) (−1)n R (G) . (1.127)
N N
Equation (1.119) is now satisfied if

S(Q) = N S u , u · u = 1. (1.128)

In both cases, 2Q = G and 2Q = G, the energy of the classical configuration is

1  
E(Q) = J (Q) S(Q)∗ · S(Q) + J (−Q) S(−Q)∗ · S(−Q) = N S 2 J (Q) ,
N
(1.129)
which is minimised by Q such that

J (Q) = min J (q) . (1.130)

q

We note that the energy does not depend on the choice of u1 and u2 , if 2Q = G, or u
if 2Q = G, which reflects the invariance of the Hamiltonian under spin O(3) rotations
and, if 2Q = G, an additional U (1) symmetry related to S(Q) being complex.
Now suppose we choose a combination of two Q, Q1 and Q2 , assuming, just for
simplicity, that both satisfy 2Qi = G i . In this case,

1 1
SR = S(Q1 ) cos Q1 · R + S(Q2 ) cos Q2 · R . (1.131)
N N
1.7 Application: Spin-Wave Theory 31

The equation (1.119) is fulfilled if

S(Q1 ) = N S cos φ u1 , S(Q2 ) = N S sin φ u2 , ui · u j = δi j , (1.132)

in which case the energy of such configuration reads

 
E Q1 , Q2 , φ = N S 2 cos2 φ J Q1 + sin2 φ J Q2 . (1.133)

It is easy to realise that this energy is always higher than choosing the single Q that
minimises J (q), thus proving the claim (1.121).
We end by observing that the average over the Brillouin zone of J (q), i.e.,
1 
J (q) ≡ JR,R = 0 , (1.134)
N q

is just the on-site exchange, which is zero. Therefore, if the average over the Brillouin
zone vanishes, then J (q) must have both positive and negative values; hence, the
minimum, i.e., J (Q1 ), is necessarily negative and so is the classical energy.

1.7.1.1 Symmetry Considerations

The lattice Hamiltonian is generally invariant under a space group S, i.e., under the
symmetry transformations g ∈ S. It follows that, if J (q) is minimised by Q1 , then

J g(Q1 ) = J (Q1 ) (1.135)

namely is minimum for the whole set of vectors obtained by applying any g ∈ S
to Q1 . In general g(Q1 ) for all g ∈ S generates a finite number of inequivalent Qi ,
i = 1, n ∗ , called the star of Q1 . We emphasise that the Hamiltonian (1.110) implicitly
assumes, besides spin O(3), also inversion symmetry, in which absence additional
terms would be present, as, e.g., SR ∧ SR · SR . It follows that S at least includes
inversion, Q1 → −Q1 , besides the identity.
If n ∗ > 1, the classical configuration choosing any of the Qi , e.g., Q1 , breaks the
space group symmetry. Specifically, such classical configuration will be invariant
under a subgroup S  ⊆ S that contains all g  ∈ S  such that g  (Q1 ) = Q1 , apart from
a reciprocal lattice vector.
If the star of Q contains just Q, which also implies that −Q ≡ Q apart from a
reciprocal lattice vector G, the classical configuration still seems to break translation
symmetry, which belongs to the space group, since
G·R
SR = N S u cos Q · R = N S u cos (1.136)
2
is not invariant under translation unless G = 0, namely Q = 0 =  equal to the 
point. However, if we combine the translation TR0 such that
G · (R + R0 ) G·R
cos = − cos , (1.137)
2 2
with time reversal, T (SR ) = −SR , we do recover full translational symmetry.
32 1 Second Quantization

In conclusion, any classical configuration that minimises the energy is unavoid-

ably not invariant under spin O(3) symmetry, time reversal symmetry, and, eventually,
under the space group symmetry of the Hamiltonian, which entails the existence of
a whole manifold of different classical configurations with the same energy.

1.8 Beyond the Classical Limit: The Spin-Wave Approximation

In reality SR is not a classical vector, but a quantum operator whose components

satisfy the commutation relations
 
SiR , S jR = i δR,R i jk SkR , (1.138)

with i jk the antisymmetric tensor and taking  = 1, which translates, for the Fourier
transformed operators, into
 
Si (q) , S j (q ) = i i jk Sk (q + q ) . (1.139)

Our aim here is to add the quantum mechanics implicit in the non-trivial commutation
relations under the assumption that the quantum fluctuations do not alter completely
the classical ground state. For that, we associate to the chosen classical configuration,
assuming for simplicity that J (q) is minimum for q = Q such that 2Q = G, the
vacuum | 0 of the quantum fluctuations. Specifically, if we choose the classical
state such that
Siclass (q) = N S δi,z δq,Q , (1.140)
namely taking u = (0, 0, 1), we define | 0 through

0 | Si (q) | 0 = Siclass (q) = N S δi,z δq,Q . (1.141)

We thus write
Si (q) = Siclass (q) + δSi (q) , (1.142)
where δSi (q) is an operator that describes the quantum fluctuation corrections, while
Siclass (q) just a c-number, and satisfies

0 | δSi (q) | 0 = 0 . (1.143)

√
We adopt a semiclassical approach, and thus assume that δ S(q) ∼ O( N ), unlike
Szclass (Q) ∼ N , and proceed consistently. For instance,
   
Sx (q) , S y (q ) = δSx (q) , δS y (q )
= i Sz (q + q ) = i δq+q ,Q N S + δSz (q + q )  i δq+q ,Q N S + · · · ,
(1.144)
1.8 Beyond the Classical Limit: The Spin-Wave Approximation 33

where . . . mean corrections subleading in N . It follows that, at leading order in N ,

the above commutation relation is fulfilled if
√ √
δSx (q)  N S xq , δS y (q)  N S pq−Q , (1.145)

† †
where xq = x−q and pq = p−q are conjugate variables satisfying
 
xq , pq† = i δq,q , (1.146)

and such that the expectation values on the vacuum,

0 | xq | 0 = 0 | pq | 0 = 0 . (1.147)

It follows that, if we write these conjugate variables in terms of bosonic operators,

i.e.,
1  †
 i  †

xq = √ bq + b−q , pq = − √ bq − b−q , (1.148)
2 2
then | 0 is their vacuum.
However, we have to enforce two more commutation relations, specifically
   
Sx (q) , Sz (q ) = −i S y (q + q ) , S y (q) , Sz (q ) = i Sx (q + q )
(1.149)
which imply that
    √  
Sx (q) , Sz (q ) = δSx (q) , δSz (q ) = N S xq , δSz (q )
√
= −i δS y (q + q ) = −i N S pq+q −Q ,
    √   (1.150)
S y (q) , Sz (q ) = δS y (q) , δSz (q ) = N S pq−Q , δSz (q )
√
= i δSx (q + q ) = i N S xq+q ,

namely
   
xq , δSz (q ) = −i pq+q −Q , pq−Q , δSz (q ) = i xq+q . (1.151)

The solution for δSz (q) with the condition (1.143) can be readily shown to be

1  † 
δSz (q) = − xk xk+q−Q + pk† pk+q−Q − δq,Q , (1.152)
2
k

since
1
0 | xq† xq | 0 = 0 | pq† pq | 0 = δq,q . (1.153)
2
34 1 Second Quantization

In conclusion, at leading order in an expansion for large N , we can represent the spin
operators as
√
Sx (q) = N S xq ,
√
S y (q) = N S pq−Q ,
(1.154)
1  † 
Sz (q) = N S δq,Q − xk xk+q−Q + pk† pk+q−Q − δq,Q .
2
k

We note that the expansion also looks like an expansion for large S. This is not surpris-
ing. Indeed, if we normalise the spin operators, SiR → siR /S, then the commutation
relations for the operators siR , which are O(1) in S, read
  1
siR , s jR = i i jk skR −−−→ 0 , (1.155)
S S→∞

showing that the spin operators become classical vectors in the large S limit.2

1.8.1 Hamiltonian of Quantum Fluctuations

Now we substitute the expressions in (1.154) in the Hamiltonian

1 
H= J (q) S(q)† · S(q) , (1.158)
N q

2 There are actually more rigorous ways to implement the mapping from spin operators to bosonic

annihilation and creation ones. However, we preferred the less rigorous approach presented, since
it is more flexible and can be adopted in more general cases than just spin models. However, for
completeness, we here mention one of those rigorous mappings, known as the Holstein-Primakoff
transformation.
Suppose that one of the equivalent classical configurations corresponds to the spin SR at site R
polarised along the z-direction. One can readily demonstrate that the following way of writing spin
operators in terms of bosonic one does preserve the spin commutation relations:

SRz = S − dR† dR , SR+ = 2S − dR† dR dR , SR− = dR† 2S − dR† dR , (1.156)

where the classical configuration is the vacuum of the bosonic operators dR and dR† . The square
roots in (1.156) assure that it is not possible to create more than 2S bosons at any given site, so that,
correctly, SRz  = −S, . . . , S. The spin-wave theory is recovered by assuming that the number of
excited bosons is negligible with respect to S, namely that the actual quantum ground state is close
√
to the classical one. In that case 2S − dR† dR  2S, and thus
√ √
SRz = S − dR† dR , SR+  2S dR , SR− = 2S dR† . (1.157)
Substituting those expressions in the Hamiltonian reproduces the spin-wave theory we discuss.
1.8 Beyond the Classical Limit: The Spin-Wave Approximation 35

consistently within an expansion for large

√ N , or, more correctly, large S. We note
that Sx (q) and S y (q) are of order O S , Sz (Q) of order O(S) plus a correction
O(1), while Sz (q) for q = Q of order O(1). Therefore, the Hamiltonian up to order
O(S) is simply
 
1  J (Q)
H= J (q) Sx (q)† Sx (q) + S y (q)† S y (q) + Sz (Q)† Sz (Q) + O S 0
N q N
 
S J (q) xq† xq + J (q + Q) pq† pq
q
 † 
+ J (Q) N S 2 − J (Q) S xq xq + pq† pq − 1
q
     
= E class + N S J (Q) + S J (q) − J (Q) xq† xq + J (q + Q) − J (Q) pq† pq
q
     
= J (Q) N S S + 1 + S J (q) − J (Q) xq† xq + J (q + Q) − J (Q) pq† pq .
q
(1.159)
Note the appearance of the quantum spin magnitude S(S + 1) that corrects the large
S classical limit, i.e., S 2 .
We need to diagonalise the Hamiltonian, which looks like the Hamiltonian of
coupled harmonic oscillators. For that, we recall that the Hamiltonian of a set of
harmonic oscillators in the canonical basis is equal to

1   
H= ωq x†q xq + p†q pq , (1.160)
2 q

with ωq = ω−q , so that, upon introducing the bosonic creation and annihilation
operators through

1  †
 i  †

xq = √ aq + a−q , pq = − √ aq − a−q , (1.161)
2 2

then
  
1
H= ωq aq† aq + , (1.162)
q
2

which is the known diagonal form. In order to bring H in (1.159) to the canonical
form (1.160), we apply the transformation

 1
xq = K q xq , pq =  pq , (1.163)
Kq

which preserves the commutation relations, and find

   J (q + Q) − J (Q) †

H = J (Q) N S S + 1 + S J (q) − J (Q) K q x†q xq + pq pq .
q
Kq
36 1 Second Quantization

We fix K q so that
  J (q + Q) − J (Q) J (q + Q) − J (Q)
J (q) − J (Q) K q = ⇒ K q2 = .
Kq J (q) − J (Q)
(1.164)
With that choice H does acquire a canonical expression

1   
H = J (Q) N S S + 1 + ωq x†q xq + p†q pq
2 q
   (1.165)
1
= J (Q) N S S + 1 + ωq a q a q +
†
,
q
2

with frequency
  
ωq = 2S J (q) − J (Q) J (q + Q) − J (Q) , (1.166)

which is the so-called spin-wave dispersion.

1.8.2 Spin-Wave Dispersion and Goldstone Theorem

Let us first assume that Q =  = 0. In this case, upon expanding in Taylor series for
small q, assuming for simplicity an isotropic system,

J (q + Q) − J (Q)  γ q 2 , (1.167)

where γ > 0 since J (Q) is the minimum of J (q). It follows that

  
ωq −−−→ 2 q γ J (0) − J (Q) ≡ v q , (1.168)
q→0

namely the excitations show an acoustic dispersion, with velocity v becoming mass-
less at q = 0. If Q =  = 0, which is the case of a ferromagnet, then

ωq −−−→ 2 γ q 2 , (1.169)
q→0

the energy still vanishing at q = 0, but now quadratically. The difference between
the two cases is that in the ferromagnetic one, Q = 0, the classical ‘order parameter’,
i.e., S(0), is actually a conserved operator of the Hamiltonian, the total spin. On the
contrary, for Q = 0, S( Q) is not a conserved quantity.
The vanishing of the energy at q → 0 is just a consequence of the Goldstone
theorem, which states that whenever the ground state is not invariant under a contin-
uous symmetry, in this case the spin SU (2) symmetry, quantum counterpart of the
1.8 Beyond the Classical Limit: The Spin-Wave Approximation 37

classical O(3), and the Hamiltonian is short-ranged, then the excitation spectrum
has to become massless at q = 0. The reason is that S(0) is just the generator of
the SU (2) rotations, which transform a state with S( Q) directed, e.g., along the
z-direction, into a state where it is directed along another direction. Since all these
states have the same energy, by symmetry, then there is no energy cost in such global
rotation. We mention that despite one can continuously rotate the direction of S( Q),
states with different directions are orthogonal in the thermodynamic limit N → ∞,
however small is the rotation, which is the essence of a symmetry breaking.
We further note that ωq in (1.166) for Q = 0 also vanishes when q → Q. The
reason is that the magnetic configuration breaks the translational symmetry, so that
Q folds to the  point. Even more, in case the star of Q contains other momenta
distinct from Q, the spin-wave dispersion vanishes at all g(Q) = Q, with g in the
coset S/S  .

1.8.3 Validity of the Approximation and the Mermin-Wagner

Theorem

We need now to check the validity of the initial assumption that the ground state with
quantum fluctuation is not much different from the classical one, i.e., the vacuum | 0.
The new ground state is actually the vacuum of the bosonic operators aq in (1.165),
which are different from the original ones, bq in (1.148). One way to assess this
assumption is to evaluate the expectation value of the order parameter and compare
it with its classical counterpart. We note that

1  †
Sz (Q) = N S − xq xq + pq† pq − 1 
2 q
1 
=NS− K q x†q xq + K q−1 p†q pq − 1  (1.170)
2 q
 
1  Kq 1
=NS− + −1 .
2 q 2 2 Kq

Therefore, the quantum fluctuations reduce the value of the order parameter.3 We
expect that the correction does not kill completely the classical order parameter

3 IfQ = 0, then K q = 1 and the quantum fluctuations have no effect on the value of the order
parameter. This is so since S(0) is conserved.
38 1 Second Quantization

unless the sum is divergent. Therefore, let us analyse

 
1  Kq 1
= + −1
2N q 2 2 Kq
   
1  J (q + Q) − J (Q) J (q) − J (Q)
= + −2 .
4N q J (q) − J (Q) J (q + Q) − J (Q)
(1.171)
Convergence problems may arise only if Q = 0 and in the points where the denom-
inators vanish, which are actually when the spin-wave energy ωq vanishes, as, e.g.,
q = 0. Therefore, the convergence of the sum is equivalent to the following d-
dimensional integral close to the origin:
 
1 1
Id = dd q ∼ dd q . (1.172)
ωq q

This integral is singular in d = 1, while converges for any d > 1. It follows that,
however large S is, namely however close we are to the classical limit, the ground state
of the quantum Hamiltonian in one dimension cannot have a finite expectation value
of S(Q) if Q = 0. This result is consistent with the Mermin-Wagner theorem stating
that a continuous symmetry cannot be spontaneously broken at zero temperature,
T = 0, i.e., in the ground state, in one dimension.
Let us consider now the effect of a finite temperature, T > 0, i.e., of thermal
fluctuations. In this case,

1! † † "
x†q xq  = aq + a−q aq + a−q = 2 n ωq + 1
2
ωq (1.173)
2
= βω + 1 = coth = p†q pq  .
e q −1 2 T

It follows that
 
1  ωq J (q + Q) − J (Q)
(T ) = coth
4N q 2T J (q) − J (Q)
 
ωq J (q) − J (Q)
+ coth −2 .
2T J (q + Q) − J (Q)

We observe that, since coth ωq /2T grows like T for large T , also (T ) grows and
at some temperature unavoidably cancels the classical value, even if Q = 0. This
simply means that the symmetry will be restored above a critical temperature Tc that
can be estimated through
(Tc )  S . (1.174)
1.8 Beyond the Classical Limit: The Spin-Wave Approximation 39

We also note that the integral Id is now replaced by

  
1 ωq 1 1
Id (T ) = dd q coth T dd q T dd q , (1.175)
ωq 2T ωq2 q2

which is singular in d = 1, 2. It follows that in d ≤ 2 the equilibrium state at any

T = 0 cannot break the SU (2) symmetry. Once again this is consistent with the
Mermin-Wagner theorem stating that a continuous symmetry cannot be broken spon-
taneously at any finite temperature in d ≤ 2.

1.8.4 Order from Disorder

If the ordering wave vector is not invariant under the space group S, it means that
there is a set Qi , i = 1, . . . , n ∗ , such that J (Qi ) is the same for all Qi . In this case, the
classical configuration is actually not unique, since we can build any configuration
that is a combination of all S(Qi ) and get the same classical energy.
Let us consider the simplest case of n ∗ = 2, thus a star that contains just two wave
vectors, Q1 and Q2 , which implies that there is an element of the space group, g,
such that
g Q1 = Q2 , (1.176)
and vice versa. For simplicity, we still assume 2Qi = 0, i = 1, 2, apart from a recip-
rocal lattice vector. We take as classical configuration

SR = cos φ S u1 cos Q1 · R + sin φ S u2 cos Q2 · R , (1.177)

with φ ∈ [0, π/2], and ui · u j = δi j , which guarantees that SR · SR = 1. It follows

that
S(q) = N S cos φ u1 δq,Q1 + N S sin φ u2 δq,Q2 , (1.178)
so that the classical energy

1   
E= J (q) S(q)† · S(q) = N S 2 J (Q1 ) cos2 φ + J (Q2 ) sin2 φ = N S 2 J (Q1 )
N q
(1.179)
is independent of φ, u1 and u2 . To better understand the freedom we actually have
at the classical level, we note that we can always write

u+v u−v
cos φ u1 = sin φ u2 = , (1.180)
2 2
with u and v two arbitrary unit vectors such that u · v = cos 2φ. It follows that
the arbitrariness consists of two O(3), one related to the global spin O(3), and
40 1 Second Quantization

another that represents an internal symmetry. We further note that under the symmetry
transformation g

g (S(q)) = N S cos φ u1 δq,Q2 + N S sin φ u2 δq,Q1 , (1.181)

so that g exchanges the two components of the classical configuration.

In order to represent the quantum spin operators as an expansion from the classical
case, we have to slightly modify the approach of the previous section. First of all,
we define another unit vector u3 ≡ u1 ∧ u2 , so that the three form a basis set. Then,
we define the spin components in such basis, i.e., SiR ≡ SR · ui , i = 1, 2, 3, which
evidently satisfy the standard commutation relations
 
SiR , S jR = i δR,R i jk SkR , (1.182)

or, in momentum space,

 
Si (q) , S j (q ) = i i jk Sk (q + q ) . (1.183)

Denoting the classical spin configuration as

Sclass (q) = N S cos φ u1 δq,Q1 + N S sin φ u2 δq,Q2 , (1.184)

namely

Siclass (q) = N S cos φ δi,1 δq,Q1 + N S sin φ δi,2 δq,Q2 , (1.185)

which is simply a c-number, we write, as before, the quantum operators

Si (q) = Siclass (q) + δSi (q) , i = 1, 2, 3 , (1.186)

where δSi (q) describes the quantum fluctuations and has vanishing expectation value
on the ‘vacuum’, which is identified with the classical state. It follows that
   
S3 (q) , S1 (q ) = δS3 (q) , δS1 (q ) = i S2 (q + q )
= i N S sin φ δq+q ,Q2 + i δS2 (q + q ) ,
   
S2 (q) , S3 (q ) = δS2 (q) , δS3 (q ) = i S1 (q + q )
= i N S cos φ δq+q ,Q1 + i δS1 (q + q ) ,
   
S1 (q) , S2 (q ) = δS1 (q) , δS2 (q ) = i S3 (q + q ) = i δS3 (q + q ) .
(1.187)
1.8 Beyond the Classical Limit: The Spin-Wave Approximation 41

One can readily check that the above commutation relations are satisfied if we assume
the following equivalences:
√
S3 (q)  N S x(q) ,
√
S1 (q)  cos φ N S δq,Q1 − cos φ (q − Q1 ) + sin φ N S p † (Q2 − q) ,
√
S2 (q)  sin φ N S δq,Q2 − sin φ (q − Q2 ) − cos φ N S p † (Q1 − q) ,
(1.188)
where
1  † 
(q) = p (k) p(k + q) + x † (k) x(k + q) − δq,0 , (1.189)
2
k

while
 x(q) =x † (−q) and p(q) = p † (−q) are conjugate variables, i.e., they satisfy
x(q), p † (q ) = i δq,q , and are defined for q = 0, Q 1 − Q 2 . The reason is that
S(0) is the generator of the global SU (2) rotations, therefore connects equivalent
states with the order parameter pointing in a different direction. Similarly, the unitary
operator
U (γ ) = e−iγ S3 (Q1 −Q2 ) = e−iγ S3 (Q2 −Q1 ) , (1.190)
since Q1 − Q2 ≡ Q2 − Q1 , simply shifts φ by γ ; hence, it is a global transformation
that connects states with different values of φ.
At leading order, the equations (1.188) become
√
S3 (q)  N S x(q) ,
√ √
S1 (Q2 − q)  sin φ N S p † (q) , S2 (Q1 − q)  − cos φ N S p † (q) ,
S1 (Q1 )  cos φ N S − cos φ (0) , S2 (Q2 )  sin φ N S − sin φ (0) ,
(1.191)
and substituted in the Hamiltonian lead, up to second order in the quantum fluctua-
tions,
1  
H= J (q) S(q)† · S(q)  N S 2 J (Q1 ) + S J (q) x † (q) x(q)
N q q
 
+S cos φ J (q + Q1 ) + sin φ J (q + Q2 ) p † (q) p(q)
2 2

q
 
− S J (Q1 ) x † (q) x(q) + p † (q) p(q) − 1
q

= N S J (Q1 ) + N S J (Q1 )
2

 # 
+S cos2 φ J (q + Q1 ) + sin2 φ J (q + Q2 ) − J (Q1 ) p † (q) p(q)
q
  $
+ J (q) − J (Q1 ) x † (q) x(q) .
(1.192)
42 1 Second Quantization

The eigenvalues of the normal modes can be obtained as we did earlier and are simply
  
ω(q, φ) = 2S cos2 φ J (q + Q1 ) + sin2 φ J (q + Q2 ) − J (Q1 ) J (q) − J (Q1 ) ,
(1.193)
so that, if E class = N S 2 J (Q1 ) < 0 is the classical energy, the quantum one, includ-
ing the zero-point contribution, is

1 
E(φ) = E class + N S J (Q1 ) + ω(q, φ) ≡ E class + δ E quant (φ) .
2 q
(1.194)
We note that the zero-point contribution to the total energy depends on φ, and thus,
the minimum energy corresponds to specific values of that angle. This phenomenon
is known under the name order from disorder [4] and occurs when the classical
lowest energy state has an accidental degeneracy that is unavoidably lifted by short-
wavelength fluctuations, both quantum, as we are discussing here, and thermal.
We further observe that, if g(Q1 ) = Q2 , then J g(q) = J (q) by symmetry and
J g(q) − Q1 = J (q − Q2 ), so that

ω g(q), φ = ω(q, π/2 − φ) , (1.195)

and thus E(φ) = E(π/2 − φ) is symmetric around π/4, which is therefore either
a minimum or a maximum. Let us therefore expand the energy around π/4, when
cos 2φ  0 can be taken as expansion parameter. We can write

ω(q, φ) = ω(q, π/4)2 + cos 2φ (q)

(q) (q)2 (1.196)
 ω(q, π/4) + cos 2φ − cos2
2φ ,
2ω(q, π/4) 8 ω(q, π/4)3

where
  
(q) = 2S 2 J (q + Q1 ) − J (q + Q2 ) J (q) − J (Q1 ) (1.197)

is odd under g, i.e.,  g(q) = −(q). Therefore,

1  1   1 
ω(q, φ) = ω(q, φ) + ω g(q), φ = ω(q, π/4)
2 q 4 q 2 q
 (q)2 (1.198)
1
− cos2 2φ ,
q ω(q, π/4)
16 3

which shows that the zero-point energy is actually maximum at π/4 and minimum at
φ = 0, π/2, which are therefore the optimal values selected by quantum fluctuations.
1.8 Beyond the Classical Limit: The Spin-Wave Approximation 43

The final result is that the huge degeneracy of the classical ground state,
related to the internal O(3) symmetry, is resolved by quantum fluctuations, thus
an order from disorder effect, which favour either the ordering at the wave vector
Q1 , or that at Q2 = g Q1 ), but disfavour any linear combination of them. This result
is not surprising, since the physical symmetries involved are the spin SU (2) and the
transformation g of the space group, which, since g 2 = 1, is effectively a Z 2 symme-
try corresponding to the exchange Q1 ↔ Q2 , i.e., an Ising one. In other words, the
underlying symmetry, which is broken in the ground state, is actually SU (2) × Z 2 .
In order to let the Z 2 degrees of freedom emerge more clearly, let us go back to
the state at fixed φ that derives from the classical configuration

S S
SR = u + v cos Q1 · R + u − v cos Q2 · R , (1.199)
2 2
where u · u = v · v = 1, and

u · v = cos 2φ ≡ σ ∈ [−1, 1] . (1.200)

The parameter σ ∈ [−1, 1] plays the role of an Ising order parameter. Through
(1.196) we can write ω(q, φ) ≡ ω(q, σ ), so that ω g(q), σ ) = ω(q, −σ ), and the
total energy becomes

1  
E(σ ) = E class + N S J (Q1 ) + ω(q, σ ) + ω(q, −σ ) , (1.201)
4 q

which explicitly looks like a double well potential, as one does expect in an Ising
model as a function of the average magnetisation. That suggests, e.g., an Ising transi-
tion at finite temperature in two dimensions, corresponding to the symmetry lowering
C4 → C2 , despite the spins are always disordered at any T = 0 [5].
Finally, to understand the order-from-disorder phenomenon in simple terms, let
us imagine a particle of mass m moving in a two-dimensional potential well V (x, y),
which has a flat minimum along a curve (x(s), y(s)) parametrised by s, i.e.,

V x(s), y(s) = Vmin = min V (x, y) . (1.202)

x,y

The classical ground state corresponds to the particle localised at any point
(x(s), y(s)) along the curve; thus, it is degenerate and parametrised by the variable
s. Inclusion of thermal and/or quantum fluctuations requires, at first approximation,
to study the potential within the harmonic approximation, namely to calculate the
Hessian at any of the points (x(s), y(s)). Evidently, along the curve (x(s), y(s))
the potential is flat, and thus has no curvature. Therefore, the Hessian at a given s
has one vanishing eigenvalue in the tangential direction of the curve (x(s), y(s)),
and the other eigenvalue K (s) > 0 in the normal direction, which corresponds to a
frequency of oscillations

K (s)
ω(s) = . (1.203)
m
44 1 Second Quantization

It follows that both zero-point quantum fluctuations and thermal ones prefer the
values of s with minimum frequency, i.e., minimum curvature K (s).
Problems

1.1 Various commutation relations—Calculate the commutators

     
Ca,nm = ca , cn† cm , Cab,nm = ca† cb , cn† cm , Ca,nmlp = ca , cn† cm
†
cl c p ,
   
Da,nm = ca† , cn† cm , Da,nmlp = ca† , cn† cm
†
cl c p ,

where the operators ca are either fermionic or bosonic ones.

1.2 Energy of the Fermi sea and f -sum rule—Consider the Hamiltonian

 2 k 2 † 1  †
H= ckσ ckσ + U (q) ckσ ck†  +qσ  ck σ  ck+qσ ,
2m 2V
kσ kk q σ σ 

in a plane-wave basis, and calculate the formal expression of its expectation value
over the Fermi sea wavefunction (1.83). Then, through the Heisenberg equation of
motion,
∂ρ(r)  
i = ρ(r), H ,
∂t
and the continuity equation relating the density to the current density J(r), i.e.,

∂ρ(r)
+ ∇ · J(r) = 0 ,
∂t

calculate the expression of the Fourier transform of the current J(q), and the following
commutator
 
ρ(q, t), J(−q, t) , (1.204)

which is commonly known as the f -sum rule.

1.3 A simple Hamiltonian to diagonalise—Consider the Hamiltonian

H = − c1† c1 − c2† c2 − t c1† c2 + c2† c1 − c1† c2† + c2 c1 .

Find the Hermitian operator

 
ϕ = c† φ̂ c + cT ψ̂ c + H .c. ,
References 45

with c† = c1† , c2† , such that, if U = eiϕ , then


2
†
U† H U = i ci ci ,
i=1

and calculate the eigenvalues i. Determine under which conditions, in the case of
bosons, i ≥ 0, i = 1, 2.

1.4 Heisenberg antiferromagnet in a magnetic field—Consider the antiferromag-

netic Heisenberg model

H=J SR · SR  , (1.205)
<RR >

where J > 0, and the sum runs over nearest neighbour bonds on a hypercubic lattice
in d ≥ 3 dimensions. Calculate the spin-wave spectrum and the spin-wave contribu-
tion to the specific heat.
Add to the Hamiltonian (1.205) with symmetry breaking along the z-direction a
magnetic field along the x-direction with Fourier components Bq , i.e., a term
 √ 
δH = − B−q Sqx  − S N B−q xq .
q q

Diagonalise the Hamiltonian in the presence of this term and calculate the ground
state energy.

1.5 Heisenberg ferromagnet—Calculate the spin-wave spectrum of the Heisen-

berg ferromagnetic Hamiltonian

H = −J SR · SR ,
<RR >

with J > 0, and on a hypercubic lattice in d dimensions.

References
1. P.W. Anderson, Phys. Rev. 124, 41 (1961). https://doi.org/10.1103/PhysRev.124.41
2. J. Hubbard, Proc. R. Soc. Lond. A 276, 238–257 (1963)
3. N.F. Mott, Proc. Phys. Soc. Sect. A 62(7), 416 (1949). https://doi.org/10.1088/0370-1298/62/7/
303
4. J. Villain, R. Bidaux, J.P. Carton, R. Conte, J. Phys. France 41(11), 1263 (1980). https://doi.org/
10.1051/jphys:0198000410110126300
5. P. Chandra, P. Coleman, A.I. Larkin, Phys. Rev. Lett. 64, 88 (1990). https://doi.org/10.1103/
PhysRevLett.64.88
Linear Response Theory
2

In order to access the physical properties of a system, one has to act on it with
some external probe. This amounts to add to the unperturbed Hamiltonian Ĥ0 a
time-dependent perturbation of the general form
ˆ
V̂ (t) = dr Â(r) v(r, t) , (2.1)

where v(r, t) represents the external probe that couples to the Hermitian operator
Â(r). Our scope is to study the effects of V̂ (t) on some measurable quantity described,
e.g., by an operator B̂(r), namely to calculate
 
B(r, t) ≡ Tr ρ̂(t) B̂(r) ,

being ρ̂(t) the time-dependent density matrix in presence of the perturbation.

2.1 Linear Response Functions

We assume that the perturbation is switched on at time t → −∞. Initially, the system
is in thermal equilibrium, so that the density matrix

lim ρ̂(t) = ρ̂0 , (2.2)

t→−∞

with
1 −β Ĥ0 1  −β E n
ρ̂0 = e = e |φn φn | . (2.3)
Z0 Z0 n

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 47

M. Fabrizio, A Course in Quantum Many-Body Theory, Graduate Texts in Physics,
https://doi.org/10.1007/978-3-031-16305-0_2
48 2 Linear Response Theory


Here Ĥ0 |φn  = E n |φn  and Z 0 = n exp (−β E n ). Therefore, the time evolution
of the density matrix in the presence of V̂ (t) is given by

1  −β E n
ρ̂(t) = e |φn (t)φn (t)| , (2.4)
Z0 n

where
∂  
i|φn (t) = Ĥ0 + V̂ (t) |φn (t) (2.5)
∂t
is the Shrœdinger equation which determines the evolution of the eigenstates of the
unperturbed Hamiltonian in presence of the perturbation. The meaning of (2.4) is
that initially the system is described by a statistical ensemble of sub-systems, each
in a given eigenstate of Ĥ0 and weighed by its Boltzmann factor. After we switch on
the perturbation, |φn  ceases to be an eigenstate of the perturbed Hamiltonian, so it
acquires a non-trivial time evolution.
Through (2.5) one readily finds the equation of motion for the density matrix

∂  
i ρ̂(t) = Ĥ0 + V̂ (t), ρ̂(t) . (2.6)
∂t
We introduce the Dirac, also called interaction, representation of the density matrix
as
ρ̂ D (t) = ei Ĥ0 t/ ρ̂(t) e−i Ĥ0 t/ ,
which satisfies
∂    
i ρ̂ D (t) = − Ĥ0 , ρ̂ D (t) + ei Ĥ0 t/ Ĥ0 + V̂ (t), ρ̂(t) e−i Ĥ0 t/
∂t  
= V̂D (t), ρ̂ D (t) , (2.7)

where
ˆ ˆ
V̂D (t) = dr ei Ĥ0 t/ Â(r) e−i Ĥ0 t/ v(r, t) = dr Â(r, t) v(r, t) ,

being Â(r, t) the Heisenberg evolution of Â(r) with the unperturbed Hamiltonian.
(0) (1)
We solve (2.7) perturbatively in v(x, t), i.e., ρ D (t) = ρ D (t) + ρ D (t) + · · · , where
(n)
ρ D (t) contains n-powers of the perturbation. Obviously,

(0) (n)
lim ρ̂ D (t) = ρ̂0 = ρ̂ D ⇒ lim ρ̂ D (t) = δn0 ρ̂0 . (2.8)
t→−∞ t→−∞

We will limit our analysis to the linear response; hence, we just need the first-order
term that satisfies
∂ (1)  
(0)
i ρ̂ D (t) = V̂D (t), ρ̂ D (t) ,
∂t
2.1 Linear Response Functions 49

with solution
ˆ t  
(1) i
ρ̂ D (t) = − dt  V̂D (t  ), ρ̂0 . (2.9)
 −∞
Therefore, at linear order,
   
B(r, t) = Tr ρ̂(t) B̂(r) = Tr ρ̂ D (t) ei Ĥ0 t/ B̂(r) ei Ĥ0 t/
     
(1)
= Tr ρ̂ D (t) B̂(r, t) = Tr ρ̂0 B̂(r, t) + Tr ρ̂ D (t) B̂(r, t)
 
(1)
= B0 (r) + Tr ρ̂ D (t) B̂(r, t) ,

where we used the fact that

   
Tr ρ̂0 B̂(r, t) = Tr ρ̂0 B̂(r) = B0 (r)

is the unperturbed expectation value. We thus find that the variation of the latter is
simply given by
ˆ  
i t
B(r, t) − B0 (r) = − dt Tr V̂D (t  ), ρ̂0 B̂(r, t)

 −∞
ˆ ˆ   
i t
=− dt  dr Tr ρ̂0 B̂(r, t), Â(r , t  ) v(r , t  )
 −∞
ˆ ˆ    (2.10)
i ∞       
=− dt dr θ (t − t ) Tr ρ̂0 B̂(r, t), Â(r , t ) v(r , t )
 −∞
ˆ ∞ ˆ
≡ dt  dr χ B A (r, r , t − t  ) v(r , t  ),
−∞

with the linear response function defined through

i  
χ B A (r, r , t − t  ) = − θ (t − t  ) B̂(r, t), Â(r , t  ) , (2.11)


where . . .  means thermal average with the unperturbed density matrix of operators
evolved in time with Ĥ0 , and χ B A (r, r , t − t  ) only depends on the time difference
since the Schrœdinger equation is time-translationally invariant. Indeed,
1   

 B̂(t) Â(t  ) =Tr e−β H ei H t/ B̂ e−i H t/ ei H t / Â e−i H t /
Z
1   

= Tr e−β H ei H (t−t )/ B̂ e−i H (t−t )/ Â =  B̂(t − t  ) Â(0)
Z
1   

= Tr e−β H B̂ ei H (t −t)/ Â e−i H (t −t)/ =  B̂(0) Â(t  − t) .
Z
Equation (2.10) shows that, at linear order, the variation of any measurable quantity
is obtained through the linear response function (2.11) which is only related to
expectation values on the unperturbed system.
50 2 Linear Response Theory

2.2 Kramers-Kronig Relations

Let us now study the analytical properties of the response function in the frequency
domain. In the following, we drop the space-coordinate dependence of the response
function, which is not relevant for what we are going to demonstrate. The response
of an operator  in the presence of an external probe that couples to B̂ is therefore
i  
χ AB (t) = − θ (t)  Â(t), B̂  , (2.12)

where
Ĥ0 t Ĥ0 t
Â(t) = ei  Â e−i  .
The response function (2.12) vanishes for t < 0, which is a consequence of causality.
We introduce the Fourier transform through
ˆ ∞
dω −iωt
χ AB (t) = e χ AB (ω) , (2.13)
−∞ 2π

as well as its analytical continuation in the complex frequency plane χ AB (z). If we

assume, as it is always the case, that χ AB (z) does not diverge exponentially for
|z| → ∞, we can regard (2.13) as the result of a contour integral
˛
dz −i zt
χ AB (t) = e χ AB (z) ,
2π
where the contour is in the upper half-plane for t < 0 and in the lower one for t > 0.
The integral thus catches all poles and branch cuts lying inside the contour. Since
χ AB (t) = 0 for t < 0, it follows that

•> Analytic property of the response function

As a consequence of causality χ AB (z) is analytic in the upper half-plane.

Let us now consider the contour drawn in Fig. 2.1. Since there are no poles
enclosed by the contour, the integral
˛
χ AB (z)
dz = 0. (2.14)
C ω−z
On the other hand, the above integral is also equal to the line integral along the lower
edge, hence
ˆ ω− ˆ ∞ ˛
χ AB (ω ) χ AB (z)
0= + dω 
+ dz
−∞ ω+ ω−ω z=ω+ exp(iθ ): θ ∈[π,0] ω−z
 ˆ 0  
χ AB (ω )
= dω −i dθ χ AB ω + eiθ ,
ω − ω π
2.2 Kramers-Kronig Relations 51

Fig. 2.1 Integration contour Im(z)

in (2.14)

2ϵ

ω Re(z)

ffl
the symbol . . . denoting the Cauchy principal value of the integral. In the limit
→ 0, the above expression simplifies into

χ AB (ω )
dω + i π χ AB (ω) = 0 , (2.15)
ω − ω

which implies that

dω Re χ AB (ω )
Im χ AB (ω) = , (2.16)
π ω − ω
dω Im χ AB (ω )
Re χ AB (ω) = − , (2.17)
π ω − ω

known as the Kramers-Kronig relations. Therefore, because of causality, the real

and imaginary parts of the response function are not independent of each other. It is
possible to rewrite both expressions as
ˆ ∞ dω Im χ AB (ω )
χ AB (ω) = , (2.18)
−∞ π ω − ω − iη

with η an infinitesimal positive number.

2.2.1 Symmetries

Let us introduce back the space dependence, so that

i  
χ AB (r, r , t − t  ) = − θ (t − t  )  Â(r, t), B̂(r , t  )  . (2.19)

Since both operators are Hermitian, it follows that

i  
χ AB (r, r , t − t  )∗ = θ (t − t  )  B̂(r , t  ), Â(r, t)  = χ AB (r, r , t − t  ) .

(2.20)
By definition
ˆ

χ AB (r, r , ω) = dt eiωt χ AB (r, r , t) ,
52 2 Linear Response Theory

therefore, through (2.20), we find that

ˆ
χ AB (r, r , ω)∗ = dt e−iωt χ AB (r, r , t)∗ = χ AB (r, r , −ω) .

This implies that

1 
Re χ AB (r, r , ω) = χ AB (r, r , ω) + χ AB (r, r , −ω)
2
is even in frequency, while

1  
Im χ AB (r, r , ω) = χ AB (r, r , ω) − χ AB (r, r , −ω)
2i
is odd.

2.3 Fluctuation-Dissipation Theorem

Let us introduce other types of functions. The first is the so-called structure factor,
defined through
1
S AB (r, r , t) =
 Â(r, t) B̂(r ) . (2.21)

In addition, we introduce the dissipation function

 1   1 
χ AB (r, r , t) =
 Â(r, t), B̂(r )  = S AB (r, r , t) − S B A (r , r, −t) ,
2 2
(2.22)
whose meaning will be explained in the following section, as well as the fluctuation
one
1    
FAB (r, r , t) = Â(r, t), B̂(r )  = S AB (r, r , t) + S B A (r , r, −t) .
2 2
(2.23)
One readily verifies that the former is related to the response function through

 i  
χ AB (r, r , t) = χ AB (r, r , t) − χ B A (r , r, −t) ,
2
which in the frequency domain reads

 i 
χ AB (r, r , ω) = χ AB (r, r , ω) − χ B A (r , r, −ω) . (2.24)
2
In particular,
χ A A (r, r, ω) = −Im χ A A (r, r, ω) . (2.25)
2.4 Spectral Representation 53

Through the definition (2.21), we find that


1 t t
S B A (r , r, −t) = Tr e−β Ĥ0 e−i Ĥ0  B̂(r ) ei Ĥ0  Â(r)
Z0

1 (t−iβ) (t−iβ)
= Tr e−β Ĥ0 ei Ĥ0  Â(r) e−i Ĥ0  B̂(r )
Z0
= S AB (r, r , t − iβ) .

Therefore,
ˆ ˆ
S B A (r , r, −ω) = dt e−iωt S B A (r , r, t) = dt eiωt S B A (r , r, −t)
ˆ
= dt eiωt S AB (r, r , t − iβ) = e−βω S AB (r, r , ω) ,

namely,
1  

χ AB (r, r , ω) = S AB (r, r , ω) 1 − e−βω ,
2
  
FAB (r, r , ω) = S AB (r, r , ω) 1 + e−βω .

2
In other words, the following relation holds:

 βω 
FAB (r, r , ω) =  coth χ AB (r, r , ω) , (2.26)
2

known as ‘fluctuation-dissipation theorem’. Indeed, if  = B̂ and r = r , FA A (r, r,

t = 0) is an estimate of the fluctuations of Â. On the other hand,
ˆ ˆ 
dω dω βω
FA A (r, r, t = 0) = FA A (r, r, ω) =  coth χ A A (r, r, ω) ,
2π 2π 2
(2.27)
which relates fluctuations to dissipation.

2.4 Spectral Representation

The spectral representation of the response functions gives instructive information

about their physical meaning. Let us start from the structure factor (2.21), which can
be written as (in the following we do not explicitly indicate the space dependence)

1  −β E n
S AB (t) = e n|ei Ĥ t/ Â e−i Ĥ t/ B̂|n
Z n
1  −β E n i(E n −E m )t/
= e e n| Â|mm| B̂|n .
Z nm
54 2 Linear Response Theory

The Fourier transform in frequency thus reads

2π  −β Ei  
S AB (ω) = e i | Â | f   f | B̂ | i δ ω + E i − E f . (2.28)
Z
if

The meaning is now self-evident. The matrix element  f | B̂ | i is the transition

amplitude of the excitation from the initial state | i into the final one | f  induced by
the operator B̂, while i | Â | f  describes the reverse process but now induced by Â.
The excitation followed by the relaxation process is weighed by the Boltzmann factor
of the initial state and contributes to S AB (ω) only if the energy difference E f − E i is
equal to ω. Thus S AB (ω) is a spectral function that measures the transition amplitude
for excitations induced by B̂ and de-excitation induced by  with a given energy ω.
Through (2.26) we also find that

 π  −β Ei    
χ AB (ω) = e 1 − e−βω i | Â | f   f | B̂ | i δ ω + E i − E f
Z
if
π   −β Ei   
= e − e−β E f i | Â | f   f | B̂ | i δ ω + E i − E f ,
Z
if
(2.29)
 (ω) is the transition amplitude for | i →| f  induced by B̂
which means that χ AB
and | f  →| i induced by  weighted by the occupation probability

e−β Ei
pi =
Z
of the initial state | i minus the one

e−β E f
pf =
Z
of the final state | f . We note that
   
pi − p f = pi 1 − p f − p f 1 − pi ,

namely it is the probability of i being occupied and f empty, minus the opposite.
 (ω) measures the absorption minus the emission probability of
In other words, χ AB
energy ω, namely the total absorption probability.
Finally, one can analogously derive the spectral representation of the response
function χ AB ,
i 1   
χ AB (t) = − θ (t) n | Â | m m | B̂ | n ei(E n −E m )t/ e−β E n − e−β E m .
 Z nm

To obtain the Fourier transform, one has to evaluate the integral

ˆ ∞
−i dt eiωt ei(E n −E m )t/ .
0
2.5 Power Dissipation 55

Since the perturbation has been assumed to be switched on at very early times,
namely at time difference t → ∞, a meaningful regularisation of the above integral
is
ˆ ∞

−i dt eiωt ei(E n −E m )t/ e−ηt/ = ,
0 ω − (E m − E n ) + iη
where η/ > 0 is the switching rate of the perturbation, and is taken to be an infinites-
imal positive number. As a result, we find that

1  e−β E n − e−β E m
χ AB (ω) = n | Â | m m | B̂ | n . (2.30)
Z nm ω − (E m − E n ) + iη

We note that η correctly makes the function analytic in the upper half-plane.

2.5 Power Dissipation

Till now we have formally introduced several response functions. In this section and
in the following ones, we are going to show how those functions emerge in real
experiments.
Let us first analyse the power dissipated in presence of the perturbation. Given
our starting assumption about the time evolution of the density matrix (2.4), it is
clear that the entropy defined through the phase space occupied by the statistical
ensemble remains constant and equal to the thermal equilibrium one, S0 . Therefore,
the system free energy is

F(t) = U (t) − T S0 = (U (t) − U0 ) + F0 ,

so that
∂ F(t) ∂U (t)
= .
∂t ∂t
On the other hand, since
 
U (t) = Tr ρ̂(t) Ĥ0 ,
it follows that
    
∂U (t) i i
= − Tr Ĥ0 + V̂ (t), ρ̂(t) Ĥ0 = − Tr ρ̂(t) Ĥ0 , Ĥ0 + V̂ (t)
∂t  
  
i
= − Tr ρ̂(t) Ĥ0 , V̂ (t) .

(2.31)
We assume that the perturbation has the general form

V̂ (t) = Â J v J (t) ,
J
56 2 Linear Response Theory

so that
  
∂U (t) i 
=− v J (t) Tr ρ̂(t) Ĥ0 , Â J . (2.32)
∂t 
J
We further note that
  ˆ
A J (t) = Tr ρ̂(t) Â J = dt  χ J J  (t − t  ) v J  (t  ) ,
J

where
i  
χ J J  (t − t  ) = − θ (t − t  )  Â J (t), Â J  (t  )  .

Therefore, by recalling that the operators evolve with the unperturbed Hamiltonian,
we find that
ˆ ˆ  
 ∂  
i dt χ J J  (t − t ) v J  (t ) = dt  δ(t − t  )  Â J (t), Â J  (t)  v J  (t  )
∂t
J J
ˆ i   
− dt  θ (t − t  )  Â J (t), Ĥ0 , Â J  (t  )  v J  (t  )

J
    
=  Â J (t), Â J  (t)  v J  (t) +  Â J (t), Ĥ0  .
J

After inserting into (2.32), we finally get

 ˆ
∂U (t) ∂
=− dt 
χ J J  (t − t  ) v J (t) v J  (t  )
∂t 
∂t
JJ
i   
−  Â J (t), Â J  (t)  v J (t) v J  (t) .
  JJ

The last term vanishes since the commutator is odd by interchanging J with J  , while
v J (t) v J  (t) is even. Hence,
 ˆ
∂U (t) ∂
=− dt  χ J J  (t − t  ) v J (t) v J  (t  ) . (2.33)
∂t 
∂t
JJ

Let us write

1
v J (t) = v J e−iωt + v∗J eiωt , (2.34)
2
and define the power dissipated within a cycle, W , through
ˆ
ω 2π/ω ∂U (t)
W = dt . (2.35)
2π 0 ∂t
2.5 Power Dissipation 57

By performing the integral and by means of (2.33) and (2.34), we obtain for W the
expression
ω  ∗   ω 
W =i v J v J  χ J J  (ω) − χ J  J (−ω) = v∗J χ J J  (ω) v J  , (2.36)
4 
2 
JJ JJ

where we use (2.24). The power dissipated during a cycle is proportional to what
we denoted as the dissipation response function, thus explaining its name. Indeed,
as we showed in the previous section, χ J J  (ω) measures the probability of energy
absorption during the process, hence its appearance in (2.36) it is not unexpected.
We further note that, since W > 0, it follows that ω χ J J  (ω) is a positive-definite
quadratic form. In particular,

ω χ J J (ω) = −ω Im χ J J (ω) > 0 ,

namely the imaginary part of χ J J (ω) is positive for ω < 0 and negative otherwise.

2.5.1 Absorption/Emission Processes

The power dissipation is related to the absorption minus emission probability. How-
ever, there are other measurements where only absorption or emission is revealed.
For instance, one can shot on a sample with a beam of particles, either photons,
neutrons, electrons, etc..., and measure the absorption probability of an energy ω.
If the coupling between the beam and the sample is represented by an operator Â,
the Fermi golden rule tells us that the absorption rate per unit time of an ensemble
at thermal equilibrium is

2π  −β Ei  2 
 
PA (ω) = e  f | Â|i δ ω + E i − E f = S A A (ω) , (2.37)
Z
if

which enlightens the meaning of the structure factors.

2.5.2 Thermodynamic Susceptibilities

Let us consider again a perturbation of the form

ˆ
V̂ (t) = dr  J (r) v J (r, t) .
J

Let us further assume that the only time-dependence of the external probes v J (r, t)
comes from a very slow switching rate that just sets the proper regularisation of
time-integrals as in (2.30). Therefore, for times far away from the time at which the
58 2 Linear Response Theory

perturbation is switched on, the external probes become constant in time, v J (r, t) →
v J (r), and the thermodynamic averages lose any time-dependence. In this limit,
ˆ ∞ ˆ
A J (r) − A0J (r) = dt  dr χ J J  (r, r , t − t  ) v J  (r )
J −∞
ˆ
→ dr χ J J  (r, r , ω = 0) v J  (r ) .
J

Let us now consider a generically perturbed Hamiltonian of the form

ˆ
Ĥ = Ĥ0 + dr  J (r ) v J (r ) ,
J

which therefore admits stationary eigenstates. The perturbed free energy  turns out
to be a functional of the static external fields v J (r ), i.e., F = F v J (r ) . Standard
thermodynamics tells us that

δF
 Â J (r)  = ,
δv J (r)

which is in general different from its expectation value, A0J (r) ≡  Â J (r)0 , in the
absence of external fields. When the latter are very small, one finds at linear order
that
ˆ 
 δ2 F
 Â J (r)  −  Â J (r)0 = dr v J  (r )
δv J (r) δv J  (r  )
J v=0
ˆ (2.38)
  
≡− dr κ J J  (r, r ) v J  (r ) ,
J

where κ J J  (r, r ) are the thermodynamic susceptibilities. Comparing (2.38) with

(2.5.2) one obtains that

κ J J  (r, r ) = −χ J J  (r, r , ω = 0) , (2.39)

which relates thermodynamic susceptibilities to the response functions at zero fre-

quency. Notice that thermodynamic stability implies that κ J J  (r, r ) is positive def-
inite.
2.6 Application: Linear Response to an Electromagnetic Field 59

2.6 Application: Linear Response to an Electromagnetic Field

Let us consider a system made of interacting electrons and of immobile positive ions
with charge density e ρions (r) such that
ˆ ˆ ˆ
dr  ρ(r)  = dr  σ† (r) σ (r)  = dr ρions (r) ,
σ

where σ (r) is the Fermi field for spin-σ electrons, which guarantees overall charge
neutrality. We assume that at equilibrium the electron density is homogeneous so
that  ρ(r) 0 = n, with n the average density. The system is coupled to the electro-
magnetic field (EMF), whose source is provided by the same system’s charges as
well as by an external time- and space-dependent source that we assume classical.
Hereafter, we decompose any space-dependent vector v(r) as

v(r) = v|| (r) + v⊥ (r) ,

where the longitudinal component v|| (r) is curl free, i.e., ∇ ∧ v|| (r) = 0, and the
transverse one v⊥ (r) divergence-less, i.e., ∇ · v⊥ (r) = 0. With this convention, the
Hamiltonian that describes the system plus the EMF in the Coulomb gauge and in
CGS units, taking  = 1, reads
ˆ   1 ˆ
1
H= dr E ⊥ (r) · E ⊥ (r) + B(r) · B(r) − dr J ⊥ext (t, r) · A⊥ (r)
8π c
ˆ 
1  e e 2
+ dr σ† (r) − i ∇ + A⊥ (r) + A ext (t, r) σ (r)
2m σ c c
2 ¨    
e 1
+ dr dr ρ(r) − ρion (r) ρ(r 
) − ρ (r 
)
|r − r |
ion
2
ˆ   ˆ
−e dr φext (t, r) ρ(r) − ρion (r) − dr M(r) · B(r) .
(2.40)
In the Hamiltonian (2.40), J ⊥ ext (t, r) is an external transverse current, source of an
external transverse vector potential defined through the conventional wave-equation
for the EMF, i.e.,
 
∂2
− c2 ∇ 2 A⊥ext (t, r) = 4π c J ⊥ext (t, r) , (2.41)
∂t 2

while A|| ext (t, r) and φext (t, r) are, respectively, the longitudinal component of the
external vector potential and the external scalar potential. We recall that only their
combination
1 ∂ A|| ext (t, r)
E || ext (t, r) ≡ − − ∇ φext (t, r) (2.42)
c ∂t
60 2 Linear Response Theory

has physical meaning, E || ext (t, r) being just the longitudinal component of the exter-
nal electric field.1 The system operator M(r) is instead the magnetisation density
contributed by the electron spins, defined by

M(r) = ge μ B α† (r) Sαβ β (r) , (2.43)
αβ

where μ B = e/2mc and ge  2 are the electron Bohr magneton and spin gyromag-
netic ratio, and S is the spin vector matrix.

1 That statement is just a consequence of gauge invariance. Specifically, let us consider the

Schrœdinger equation
i | ψ̇ = H | ψ ,
and apply on both sides a unitary transformation
 ˆ 
 
U (t) ≡ exp − ie dr ϕ(t, r) ρ(r) − ρion (r) ,

so that
∂      
i U (t) | ψ̇ = i U (t) | ψ − i U̇ (t) U † (t) U (t) | ψ = U (t) H U † (t) U (t) | ψ .
∂t
˙ = H∗ | , where
Therefore, if we define |  ≡ U (t) | ψ, that wavefunction satisfies i | 
ˆ
 
H∗ = U (t) H U † (t) + i U̇ (t) U † (t) = U (t) H U † (t) + e dr ϕ̇(t, r) ρ(r) − ρion (r) .

Since
U (t) σ (r) U † (t) = eie ϕ(t,r) σ (r) ,
as one can readily demonstrate, H∗ has the same expression as H apart from

φext (t, r) → φext (t, r) − ϕ̇(t, r) , A ext (t, r) →A ext (t, r) + c ∇ϕ(t, r) ,

which implies that the combination

1
E ext (t, r) =− Ȧ ext (t, r) − ∇φext (t, r)
c
remains invariant, as well as any gauge invariant operator, like the density ρ(r) or its corresponding
current; see (2.53). In other words, given a gauge invariant operator A such that U (t) A U † (t) = A,

ψ | A | ψ =  | U (t) A U † (t) |  =  | A |  ,

so that we can work with either wavefunctions | ψ or | , provided the transformation leaves
E ext (t, r) invariant.
2.6 Application: Linear Response to an Electromagnetic Field 61

It is worth remarking that, since we have explicitly included the mutual

Coulomb interaction among the system charges, the longitudinal component of
the electric field that enters the Hamiltonian refers exclusively to the external
source and we assume is a purely classical field. With that choice, the total
longitudinal electric field, which we shall denote as ‘internal’ field, is not an
independent dynamical quantity, but is defined through

E || (r, t) = E || ext (r, t) + E || sys (r) , (2.44)

where E || sys (r, t) is the longitudinal field generated by the system charges
through the Gauss law
 
∇ · E || sys (r) = −4π e  ρ(r)  − ρion (r) . (2.45)

On the contrary, the transverse fields A⊥ (r), E ⊥ (r) and B(r) ≡ ∇ ∧ A⊥ (r)
are, so far, genuine quantum fields, not to be confused with the external ones,
whose dynamics is defined through the commutation relations between A⊥ (r)
and E ⊥ (r), which we now introduce.

2.6.1 Quantisation of the Electromagnetic Field

The transverse E ⊥ (r) is actually conjugate to A⊥ (r), specifically,

  
∂i ∂ j  
Ai⊥ (r) , E j⊥ (r ) = −i 4π c δi j − δ r − r , (2.46)
∇2

or, in Fourier components,

  
 qi q j
Ai⊥ (q) , E j⊥ (−q ) = −i 4π c δ q,q δi j − . (2.47)
q2

We introduce the polarisation vectors  s (q), s = 1, 2, which are orthogonal unit

vectors satisfying

q 
2
qi q j
 1 (q) ∧  2 (q) = ,  s (−q) = (−1)s  s (q) , is (q) js (q) = δi j − .
q
s=1 q2
62 2 Linear Response Theory

The transverse vector potential and electric field can thus be written as

1  iq·r 1   iq·r
2
A⊥ (r) = √ e A⊥ (q) = − √ e (−i)s  s (q) xs (q) ,
V q V s=1 q

1  iq·r 1   iq·r
2
E ⊥ (r) = √ e E ⊥ (q) = 4π c √ e (−i)s  s (q) ps (q) ,
V q V s=1 q
(2.48)
where, since both A⊥ (r) and E ⊥ (r) are Hermitian, xs (−q) = xs (q)† and ps (−q) =
ps (q)† , which, through (2.46), satisfy
 
xs (q) , ps† (q ) = i δss  δq,q ,

namely they are conventional conjugate variables. Using that representation, and
defining, accordingly,

1   iq·r
2
J ⊥ext (r) = − √ e (−i)s  s (q) Js ext (q) ,
V s=1 q

we find that
ˆ 
1
HE M F ≡ dr E ⊥ (r) · E ⊥ (r) + B(r) · B(r)
8π
ˆ
1
− dr J ⊥ext (t, r) · A⊥ (r)
c
 
1 
2
q2 †
= 4π c2 ps† (q) ps (q) + xs (q) xs (q)
2 q
4π (2.49)
s=1

1 
2 
− Js ext (t, q) xs† (q)
c q
s=1
  
1 
2 2
1
= ωq bsq
†
bsq + − Js ext (t, q) xs† (q) ,
q
2 c q
s=1 s=1

where ωq = c q, and we define

 
2π c  †
 q  †

xs (q) ≡ bsq + bs−q , ps (q) ≡ −i bsq − bs−q ,
q 8π c

†
in terms of the photon annihilation, bsq , and creation, bsq , operators with polarisation
s.
2.6 Application: Linear Response to an Electromagnetic Field 63

We observe that, through (2.49), the following Heisenberg equations of motion

readily follow:

q2 1
ṗs (q) = − xs (q) + Js ext (t, q) , ẋs (q) = 4π c2 ps (q) ,
4π c

so that, since q ∧ q ∧  s (q) = −q 2  s (q),

1   iq·r
2
Ȧ⊥ (r) = − √ e (−i)s  s (q) 4π c2 ps (q) = −c E ⊥ (r) ,
V s=1 q
 
1   iq·r
2
q2 1
Ė ⊥ (r) = 4π c √ e (−i)  s (q) −
s
xs (q) + Js ext (t, q)
V s=1 q 4π c
= c ∇ ∧ ∇ ∧ A⊥ (r) − 4π J ⊥ext (t, r) = c ∇ ∧ B(r) − 4π J ⊥ext (t, r) ,
(2.50)
thus the well-known Maxwell equations

1 ∂ A⊥ (r) 1 ∂ E ⊥ (r) 4π
E ⊥ (r) = − , ∇ ∧ B(r) = + J ⊥ext (t, r) . (2.51)
c ∂t c ∂t c

2.6.2 System’s Sources for the Electromagnetic Field

Through the continuity equation,2 the electron current is defined by

 
−e ρ̇(r) = −i − e ρ(r) , H ≡ −∇ · J(r) ,

2 Let us consider a conserved quantity Q, and the corresponding density ρ Q (r), so that
ˆ
Q = dr ρ Q (r) ≡ ρ Q (q = 0) ,

where ρ Q (q) is the Fourier transform of ρ(r). Since Q is conserved

  ˆ ˆ  
Q̇ = −i Q , H = 0 = dr ρ̇ Q (r) = −i dr ρ Q (r) , H .

Since the commutator of ρ Q (r) with the Hamiltonian does not vanish, the only possibility for the
integral over the whole volume to vanish is that
 
ρ̇ Q (r) = −i ρ Q (r) , H ≡ −∇ · J Q (r) ,

which defines the current J Q (r) associated to ρ Q (r), under the assumption that the flux of J Q (r)
through the surface of the system vanishes. The equation

ρ̇ Q (r) + ∇ · J Q (r) = 0 (2.52)

is the continuity equation corresponding to the conserved quantity Q.
64 2 Linear Response Theory

which, given the Hamiltonian (2.40), leads to

     
1  e
J(r) = −e σ† (r) − i ∇ σ (r) + i ∇ σ† (r) σ (r) + A(t, r) ρ(r)
2m σ mc

e
≡ −e j (r) + A(t, r) ρ(r) ,
mc
(2.53)
where A(t, r) = A⊥ (r) + A|| ext (t, r), and j (r) is the current density in absence of
the EMF.
The total charge current, defined through

δH
J tot (r) ≡ −c , (2.54)
δ A(r)

is contributed also from the spin magnetisation. Indeed,

ˆ ˆ   ˆ  
− drM(r) · B(r) = − drM(r) · ∇ ∧ A(r) = − dr A(r) · ∇ ∧ M(r) ,

thus
δ J tot (r) = c ∇ ∧ M(r) .
Finally, also the external transverse current contributes, so that,

e
J tot (r) = −e j (r) + A(t, r) ρ(r) + c ∇ ∧ M(r) + J ⊥ext (t, r) ,
mc
(2.55)
or, decomposed in longitudinal and transverse components,

e
J ⊥tot (r) = −e j ⊥ (r) + A⊥ (r) ρ(r) + c ∇ ∧ M(r) + J ⊥ext (t, r) ,
mc

e
J tot (r) = −e j (r) + A ext (t, r) ρ(r) .
mc
(2.56)
With the above definitions, the Hamiltonian can be rewritten as
ˆ 
1
H= dr E ⊥ (r) · E ⊥ (r) + B(r) · B(r)
8π
ˆ  
1 
+ dr σ† (r) − ∇ 2 σ (r)
2m σ
¨    
e2 1
+ dr dr ρ(r) − ρion (r)  ρ(r ) − ρion (r )
2 |r − r |
ˆ 
1 e2
− dr A(r) · − e j (r) + c ∇ ∧ M(r) + J ⊥ext (t, r) − A(r) ρ(r)
c 2mc
ˆ  
−e dr φext (t, r) ρ(r) − ρion (r)
2.6 Application: Linear Response to an Electromagnetic Field 65
ˆ  
≡ H0 − e dr φext (t, r) ρ(r) − ρion (r)
ˆ 
1 e2
− dr A(r) · − e j (r) + c ∇ ∧ M(r) + J ⊥ext (t, r) − A(r) ρ(r)
c 2mc
= H0 + δ H ,
(2.57)

so that, through the Heisenberg equation of motion for E ⊥ (r), we obtain the Maxwell
equation
1 ∂ E ⊥ (r) 4π
∇ ∧ B(r) = + J ⊥tot (t, r) . (2.58)
c ∂t c

2.6.3 Optical Constants

Hereafter, all quantum fields and density operators are to be considered as their
expectation values calculated within linear response theory.
Within the linear response, in frequency and momentum space and assuming
translational invariance, we write, see (2.53),

J(ω, q) ≡ σ̂ (ω, q) E(ω, q) , (2.59)

where E(ω, q) is the internal electric field, and σ̂ (ω, q) is the optical conductiv-
ity tensor with components σi j (ω, q), i, j = x, y, z. If we further assume spatial
isotropy, then

qi q j qi q j
σi j (ω, q) = 2
σ (ω, q) + δi j − σ⊥ (ω, q) ,
q q2

where σ (ω, q) and σ⊥ (ω, q) are, respectively, the longitudinal and transverse com-
ponents of the optical conductivity, which are the first two optical constants we shall
be interested in. It follows that

J ⊥ (ω, q) = σ⊥ (ω, q) E ⊥ (ω, q) , J (ω, q) = σ (ω, q) E (ω, q) . (2.60)

We next define a ‘displacement’ field through

∂ D(r) ∂ E(r)
≡ + 4π J tot (t, r) . (2.61)
∂t ∂t
We note that, after taking the divergence of both sides, thus selecting the longitudinal
components, and using the continuity equation, the Gauss law, as well as the fact
66 2 Linear Response Theory

that ρion (r) is time independent,

∂∇ · D (r) ∂∇ · E (r)
= + 4π ∇ · J(t, r)
∂t ∂t
∂∇ · E sys (r) ∂∇ · E ext (r) ∂ρ(r)
= + − 4π
∂t ∂t ∂t
∂    ∂∇ · E ext (r)
= ∇ · E sys (r) + 4π e ρ(r) − ρion (r) +
∂t ∂t
∂∇ · E ext (r)
= ,
∂t

which is satisfied if we assume that D (r) is just the external E ext (r).
In frequency and momentum space (2.61) reads

4π
D(ω, q) ≡ E(ω, q) + i J(ω, q) , (2.62)
ω
or, through (2.59),

4π
D(ω, q) = 1 + i σ̂ (ω, q) E(ω, q) . (2.63)
ω

Now, we define the ‘dielectric constant’ tensor through

D(ω, q) ≡ ˆ (ω, q) E(ω, q) , (2.64)

whose components, as before, can be written as


qi q j qi q j
i j (ω, q) = 2
(ω, q) + δi j − ⊥ (ω, q) ,
q q2

in terms of the longitudinal, (ω, q), and transverse, ⊥ (ω, q), components. Since
D (ω, q) ≡ E ext (ω, q), it follows that

E ext (ω, q) = (ω, q) E (ω, q) , (2.65)

which relates to each other external and internal longitudinal components of the
electric field.
Finally, comparing (2.64) with (2.63), we find the following relation between
transverse and longitudinal dielectric constants and optical conductivities:

4π 4π
⊥ (ω, q) =1+i σ⊥ (ω, q) , (ω, q) = 1 + i σ (ω, q) . (2.66)
ω ω
2.6 Application: Linear Response to an Electromagnetic Field 67

2.6.4 Linear Response in the Longitudinal Case

Suppose that the external source generates only a longitudinal electric field, which
is represented in the gauge such that

E ext (t, r) = −∇ φext (t, r) ⇒ E ext (ω, q) = −i q φext (ω, q) . (2.67)

Through the Hamiltonian (2.57), the perturbation is in this case

ˆ  
δ H = −e dr φext (t, r) ρ(r) − ρion (r) . (2.68)

We recall that at equilibrium ρ0 (q) = 0 but at q = 0, when its value is n times the
volume V . Therefore, in presence of the external potential, a finite value of ρ(q) for
q = 0 is already the deviation from equilibrium, and thus, in linear response theory,

ρ(ω, q) = −e χ (ω, q) φext (ω, q) ,

where χ (ω, q) is the Fourier transform of the density-density response function

i  
χ (t, q) = − θ (t)  ρ(t, q) , ρ(−q, 0)  . (2.69)
V
Through (2.65), we can also write

ρ(ω, q) = −e χ (ω, q) φext (ω, q) = −e χ (ω, q) (ω, q) φ(ω, q)

(2.70)
≡ −e χ∗ (ω, q) φ(ω, q) ,

where φ(ω, q) is the internal scalar potential, and

χ∗ (ω, q) = χ (ω, q) (ω, q)

is the so-called proper response function as opposed to χ (ω, q), which is denoted
as the improper response function. Since φ(ω, q) = φext (ω, q) + φsys (ω, q) where
the system field is obtained by the Gauss law

4π e
φsys (ω, q) = φ(ω, q) − φext (ω, q) = − ρ(ω, q)
q2
  4π e2
= 1− (ω, q) φ(ω, q) = χ∗ (ω, q) φ(ω, q)
q2

1 4π e2
= −1 φext (ω, q) = χ (ω, q) φext (ω, q) ,
(ω, q) q2
68 2 Linear Response Theory

we arrive at the following equations:

4π e2 1 4π e2
(ω, q) = 1 − χ∗ (ω, q) , = 1 + 2 χ (ω, q) ,
q2 (ω, q) q
χ∗ (ω, q) (2.71)
χ (ω, q) = 2
.
4π e
1− χ∗ (ω, q)
q2

We note that, since ρ(q → 0) is the total number N of electrons, which is conserved,
i.e., N (t) = N , then
   
V lim χ (t, q) = −i θ (t) lim  ρ(q, t), ρ(−q)  = −i θ (t)  N (t), N 
q→0 q→0
 
= −i θ (t)  N , N  = 0 .

Accordingly,
lim χ (ω, q) = lim χ∗ (ω, q) = 0 , (2.72)
q→0 q→0

which derive from the conservation of N .

2.6.4.1 Consequences of Gauge Invariance

Equivalently, we can represent the external longitudinal field only through a vector
potential, i.e.,

1 ∂ A ext (t, r)
E ext (t, r) = −∇φext (t, r) ≡ − ,
c ∂t
implying
qc
A ext (ω, q) =−
φext (ω, q) ,
ω
in which case, see (2.57), the perturbation is
ˆ 
1 e2
δH = − dr A ext (t, r) · − e j (r) − A ext (t, r) ρ(r) . (2.73)
c 2mc

The Fourier transform χi j (ω, q) of current-current linear response function

i  
χi j (t, q) = − θ (t)  ji (q, t) , j j (−q, 0)  , i, j = x, y, z , (2.74)
V
can also be written as

qi q j qi q j
χi j (ω, q) = χ (ω, q) + δi j − 2 χ⊥ (ω, q) . (2.75)
q2 q
2.6 Application: Linear Response to an Electromagnetic Field 69

It follows that, in linear response,

e
j (ω, q) = χ (ω, q) A ext (ω, q) . (2.76)
c

Considering J(r) in (2.53), its expectation value at first order in A ext , which
amounts to ρ(r) → n, is

e2 n
J (ω, q) = − χ (ω, q) + A ext (ω, q)
c m

e2 q n
= χ (ω, q) + φext (ω, q)
ω m

e2 n
=− χ (ω, q) + E ext (ω, q)
iω m

e2 n
=− χ (ω, q) + (ω, q) E (ω, q) ≡ σ (ω, q) E (ω, q) .
iω m
(2.77)
The continuity equation, eρ̇ = ∇ · J implies that

e ω ρ(ω, q) = −e2 ω χ (ω, q) φext (ω, q)


e2 q 2 n (2.78)
= −q · J (ω, q) = − χ (ω, q) + φext (ω, q) ,
ω m

leading to the equivalence


n
ω2 χ (ω, q) = q 2 χ (ω, q) + , (2.79)
m

which is merely a consequence of gauge invariance. In particular, if ω = 0, since

χ (0, q) is finite being a thermodynamic susceptibility, (2.79) implies that
n
χ (0, q) = − , ∀q. (2.80)
m
Moreover, through (2.77), we find that the longitudinal component of the optical
conductivity

e2 n e2 ω2
σ (ω, q) = − χ (ω, q) + (ω, q) = − χ (ω, q) (ω, q)
iω m iω q 2
(2.81)
e2 ω2
=− χ∗ (ω, q) .
iω q 2
70 2 Linear Response Theory

2.6.4.2 Screening of a Static Long-Wavelength Electric Field

Let us assume to change the chemical potential and compensate for the change of
electron number by a change of ionic charge so as to maintain charge neutrality. In
this case, the perturbation is simply
ˆ ˆ
δ H = −μ dr ρ(r) − e dr δφion (r) ρ(r) ,

where δφion (r) satisfies ∇ 2 δφion (r) = −4π e δρion (r), and is the additional poten-
tial generated by the change of the ionic charge. The induced field generated by
the change in electron density satisfies instead ∇ 2 δφels (r) = 4π e δρ(r), so that the
internal potential is
μ μ
φ(r) = + δφion (r) + δφels (r) = + δφsys (r)
e e
and is time independent. Therefore, only the ω = 0 component of the response func-
tion is involved and we obtain, through the proper response,

δρ(q) = −e χ∗ (0, q) φ(0, q) . (2.82)

In the limit q → 0, since we assumed that charge neutrality is preserved then

δφsys (q) = δφion (q) + δφels (q) −−−→ 0 ,

q→0

and, being δρ(q → 0) = δ N just the change in electron number, we find

δ N = −χ∗ (0, q → 0) δμ ,

or, equivalently,
δN
κ≡ = − lim χ∗ (0, q) , (2.83)
δμ q→0

where κ ≥ 0 is the uniform electron compressibility. Through (2.71) we find that

4π e2
(0, q ∼ 0) = 1 + κ, (2.84)
q2

so that the internal field felt by the electrons gets screened with respect to a long-
wavelength static external one as

q2
E (0, q ∼ 0) = E ext (0, q ∼ 0) .
q 2 + 4π e2 κ
2.6 Application: Linear Response to an Electromagnetic Field 71

Note that at ω = 0 and in the limit q → 0 longitudinal and transverse components

of the field coincide. Therefore, in the above equation, we can safely omit the label
and write
q2
E(0, q ∼ 0) = E ext (0, q ∼ 0) . (2.85)
q 2 + 4π e2 κ

We now summarise through all the above results the behaviour of the longitudinal
optical constants separately for insulators and metals.

2.6.4.3 Longitudinal Response of an Insulator

In an insulator, the chemical potential μ lies inside the gap, so that an infinitesimal
variation δμ does not yield any variation δ N in the electron number, thus κ = 0 =
−χ∗ (0, q → 0). In particular, since χ∗ (ω, q → 0) → 0, see (2.72), then

lim lim χ∗ (ω, q) = lim lim χ∗ (ω, q) = 0 ,

ω→0 q→0 q→0 ω→0

namely χ∗ (ω, q) is analytic at ω = q = 0. Since −χ∗ (0, q ∼ 0) > 0 is the thermo-

dynamic susceptibility to a slowly varying chemical potential variation δμ(q ∼ 0),
we can conclude that, for small ω and q,

χ∗ (ω, q)  −κ  q 2 , κ  > 0 , (2.86)

so that

lim (0, q) = lim ⊥ (0, q) ≡ (0, 0) = 1 + 4π e2 κ  > 1 , (2.87)

q→0 q→0

namely the static long-wavelength limit of the dielectric constant is finite and greater
than one, and, through (2.85),

1
E(0, 0) = E ext (0, 0) , (2.88)
1 + 4π e2 κ 
thus the internal field is reduced with respect to the external one but not totally
suppressed.
Moreover, (2.81) implies that the DC conductivity, i.e., Re σ (ω → 0, 0), van-
ishes, as expected in an insulator.

2.6.4.4 Longitudinal Response of a Metal

Let us now consider a metal. We remark that, from the point of view of the longitudinal
response, it makes no difference whether the metal is normal or superconducting.
In a metal, the chemical potential lies in the conduction band and so κ = 0. It
follows that

4π e2
lim (0, q) = lim 1 + 2 κ → ∞ , (2.89)
q→0 q→0 q
72 2 Linear Response Theory

Fig. 2.2 Hypothetical

scenario in which the
conduction electrons are
rigidly shifted by x

and thus, through (2.85), we conclude that a static long-wavelength external field is
totally screened inside the metal. We further note, through (2.72), that

lim lim χ∗ (ω, q) = 0 , lim lim χ∗ (ω, q) = −κ < 0 ,

ω→0 q→0 q→0 ω→0

hence that χ∗ (ω, q) is not analytic at ω = q = 0, which is a key property of metals.

Since in a metal the energy scale associated with q is v F q, where v F is the Fermi
velocity, the absence of analyticity implies that we cannot find the behaviour of
χ∗ (ω, q) for ω  v F q through that at v F q  ω.
In order to proceed, we note that at q = 0 a change in the density of conduction
electrons corresponds to a rigid shift of the latter with respect to the positive back-
ground of the ions plus all bound electrons, core and valence ones; see Fig. 2.2. In
that limit, the electrons thus behave as a macroscopic object that can be regarded as
classical plasma. If that shift is driven by an external time-dependent field E ext (t),
as shown in the figure, and assuming that the conduction electrons have an effective
mass m ∗ , then the classical equation of motion reads

m ∗ ẍ(t) = −4π e σ (x) − e E ext (t) ,

where σ (x) = n ∗ e x is the surface charge with n ∗ the density of conduction electrons.
In Fourier space,
e
−m ∗ ω2 x(ω) = −4π e2 n ∗ x(ω) − e E ext (ω) ⇒ x(ω) = − E ext (ω) ,
4π n ∗ e − m ∗ ω2
2

which implies that the internal field

  4π n ∗ e2
E(ω) = 4π σ x(ω) + E ext (ω) = − E ext (ω) + E ext (ω)
4π n ∗ e2 − m ∗ ω2
m ∗ ω2 1
=− E ext (ω) = E ext (ω)
4π n ∗ e2 − m ∗ ω2 ω2p
1− 2
ω
1
≡ E ext (ω) ,
(ω, 0)
(2.90)
2.6 Application: Linear Response to an Electromagnetic Field 73

where
4π n ∗ e2
ω2p ≡ (2.91)
m∗
is the so-called plasma frequency, namely that

ω2p 4π e2
(ω, 0) = 1 − = 1 − lim χ∗ (ω, q) , (2.92)
ω2 q→0 q2

and thus
n∗ q 2
χ∗ (ω, q ∼ 0) = . (2.93)
m ∗ ω2
Equations (2.90) and (2.92) imply that an infinitesimally small external field with
frequency ω  ω p produces a huge internal one. That corresponds to the frequency
being in resonance with internal collective excitations, the plasmons. We remark that
the above result is valid provided at ω  ω p only the conduction electrons are free
to move, while valence electrons are still bound, which is justified if ω p is small
enough with respect to the gap between valence and conduction band.
If that is the case, through (2.81), we find that

1 ωp
2
e2 ω2
σ (ω, q ∼ 0) = − χ ∗ (ω, q ∼ 0) = − .
iω q 2 4π iω

We recall that ω = ω + i0+ , so that

1 2 1 ω2p 1 ωp
2
σ (ω, 0) = i ωp = δ(ω) + i , (2.94)
4π ω + i0+ 4 4π ω

It follows that Re σ (ω, 0) is a δ-peak at zero frequency, so-called Drude peak, with
weight ω2p /4, the Drude weight.
When there are channels that dissipate current, provided, for instance, by impu-
rities or by electron-electron and electron-phonon umklapp scattering, one conven-
tionally assumes a Drude-Lorentz expression

1 2 1 ω2p 1 γ ω2p 1 ω
σ (ω, 0) = i ωp = + i ,
4π ω + iγ 4 π ω +γ
2 2 4 π ω + γ2
2
(2.95)
so that the real part becomes a Lorentzian.
74 2 Linear Response Theory

2.6.5 Linear Response in the Transverse Case

In the presence of a transverse electromagnetic field, the perturbation we must add

to the Hamiltonian is
ˆ ˆ
e e2
δH = dr A⊥ (r) · j (r) + dr ρ(r) A⊥ (r) · A⊥ (r)
c
ˆ 2mc2
  1 ˆ
− dr A⊥ (r) · ∇ ∧ M(r) − dr A⊥ (r) · J ⊥ext (t, r) ,
c

where, we remark, A⊥ (r) is the internal field, contributed both by the external source
J ⊥ ext (t, r) as well as by the electrons. Specifically, A⊥ (r) satisfies the equation

∂2
− c2 ∇ 2 A⊥ (t, r) = 4π c J ⊥ (t, r) + 4π c2 ∇ ∧ M(r) + 4π c J ⊥ext (t, r) ,
∂t 2
(2.96)
where, see (2.53),

e
J ⊥ (t, r) = −e j ⊥ (t, r) + ρ(t, r) A⊥ (t, r) .
mc

In linear response, and after Fourier transform in frequency and momentum,

e
j ⊥ (ω, q) = χ⊥ (ω, q) A⊥ (ω, q) , (2.97)
c
where χ⊥ (ω, q) is the transverse component of the current-current response function,
see (2.74) and (2.75), and so
 
e2 n e2 n
J ⊥ (ω, q) = − χ⊥ (ω, q) + A⊥ (ω, q) = − χ⊥ (ω, q) + E ⊥ (ω, q)
c m iω m
iω
≡ σ⊥ (ω, q) E ⊥ (ω, q) = σ⊥ (ω, q) A⊥ (ω, q) .
c
(2.98)
Similarly,
M(ω, q) = −i χσ (ω, q) q ∧ A⊥ (ω, q) , (2.99)
where χσ (ω, q) is the spin density-density linear response function multiplied by
g 2 μ2B , so that the Fourier transform of ∇ ∧ M(r) becomes in linear response
−q 2 χσ (ω, q) A⊥ (ω, q), so that (2.96) reads in frequency and momentum
 
c2 q 2 − ω2 A⊥ (ω, q) = 4π i ω σ⊥ (ω, q) A⊥ (ω, q) − 4π c2 q 2 χσ (ω, q) A⊥ (ω, q)
+ 4π c J ⊥ext (ω, q) .

Since
 
c2 q 2 − ω2 A⊥ext (ω, q) = 4π c J ⊥ext (ω, q)
2.6 Application: Linear Response to an Electromagnetic Field 75

defines the external transverse field, we can write

c2 q 2 − ω2
A⊥ (ω, q) = A⊥ext (ω, q)
c q − ω − 4π i ω σ⊥ (ω, q) + 4π c2 q 2 χσ (ω, q)
2 2 2

c2 q 2 − ω2
= A⊥ext (ω, q) ,
c2 q 2 − ω2 ⊥ (ω, q) + 4π c2 q 2 χσ (ω, q)
(2.100)
which relates the internal vector potential to the external one, and where we use
(2.66). The root with respect to ω of the denominator in (2.100) determines the
dispersion of the light inside the system, which in the vacuum is simply c q 
v F q. For that reason, we can safely assume that ω  v F q, in which case, since the
magnetisation is conserved, χσ (ω, q)  χσ (ω, 0) = 0, and ⊥ (ω, q)  ⊥ (ω, 0) =
(ω, 0) = (ω, 0), so that

c
ω  q. (2.101)
(ω, 0)

It follows that

• In an insulator, and for ω small compared to the gap, (0, 0) > 1, see (2.87),
so that light propagates, though with generally a smaller velocity than c. The
insulator is therefore transparent in the visible.
• In a metal, (2.92) implies that (ω, 0) is negative for ω < ω p , where ω p is gen-
erally in the ultraviolet. Therefore, light cannot propagate as long as ω < ω p , so
that the metal is opaque in the visible range. Only when ω exceeds ω p , light starts
to propagate. The reflectivity is therefore close to one at small ω and bends down
around the plasma frequency.

Concerning the response to a uniform q = 0 field, i.e., to a time-dependent electric

field, the transverse optical constants become equal to the longitudinal ones, which
we earlier discussed.

2.6.5.1 Response to a Static Non-uniform Field

If ω → 0 but q = 0, though small, i.e., in presence of a static non-uniform magnetic
field, (2.100) becomes

c2 q 2
A⊥ (ω → 0, q) = A⊥ext (0, q) .
c2 q 2 − 4π i ω σ⊥ (ω, q) + 4π c2 q 2 χσ (0, q)

In this case, χσ (0, q)  −κσ , where κσ > 0 is the uniform magnetic susceptibility,
while

n
lim i ω σ⊥ (ω, q) = −e χ⊥ (0, q) +
2
.
ω→0 m
76 2 Linear Response Theory

We recall that, as a consequence of gauge invariance, χ (0, q) = −n/m. It would be

tempting to assume that
n
χ⊥ (0, q  0)  χ (0, 0) + α q 2 = − + α c2 q 2 , (2.102)
m

with α > 0 since −χ⊥ (0, q  0) > 0 is a thermodynamic susceptibility that is sup-
posedly maximum at q = 0, so that

c2 q 2
A⊥ (0, q ∼ 0) = A⊥ext (0, q ∼ 0)
c2 q 2 + 4π e2 α c2 q 2 − 4π c2 q 2 κσ (2.103)
1
= A⊥ext (0, q ∼ 0) ,
1 + 4π e2 α − 4π κσ

which implies that the magnetic field propagates freely inside the system, and is
enhanced with respect to the external one if κσ > e2 α and reduced otherwise, cor-
responding, respectively, to paramagnetic and diamagnetic behaviour.
However, the assumption (2.102) is valid only in insulators and normal metals.
In a superconducting metal,

n − nc n
χ⊥ (0, q → 0) = − >− , (2.104)
m m
where n c ≤ n is the density of the superconducting fraction, the condensate. In this
case,

c2 q 2
A⊥ (0, q ∼ 0) = A⊥ext (0, q ∼ 0) ,
c2 q 2 + ωc2 − 4π c2 q 2 κσ

where ωc = 4π n c e2 /m is the plasma frequency of the condensate. It follows that a
magnetic field cannot penetrate in a superconductor and decays from the surface over
a distance λ L ∼ c/ωc , the so-called London penetration depth. A superconductor is
therefore a perfect diamagnet, a phenomenon known as Meissner effect.

2.6.6 Power Dissipated by the Electromagnetic Field

In presence of an external longitudinal field, the power dissipated according to the

general formula is
 e 2 ω    2  
W =  A|| ext (q, ω)2 χ  (ω, q) = − e ω  A|| ext (q, ω)2 Im χ|| (ω, q)
c 2 || c 2
 e 2 ω  2
−  
E || ext (q, ω) Im χ|| (ω, q) ,
ω 2
Problems 77


where χ|| (ω, q) is the dissipative current-current function which is equal to minus
the imaginary part of the linear response function. On the other hand,

ω σ|| (ω, q) ω σ|| (ω, q)
Im χ|| (ω, q) = Im −i = − 2 Re
e || (ω, q)
2 e || (ω, q)

ω ω   1 ω2 1
= − 2 Re || (ω, q) − 1 = Im
e 4πi || (ω, q) 4π e2 || (ω, q)
⎛ ⎞
ω ⎜ σ|| (ω, q) ⎟ ω 1
=− Re ⎝ ⎠=− 2 Re σ|| (ω, q) ,
e2 4πi e | || (ω, q)|2
1+ σ|| (ω, q)
ω
    
which, since  E || ext  =  ||   E || , implies that

ω 1    
W =− Im  E || ext (ω, q)2 = 1 Re σ|| (ω, q)  E || (ω, q)2 .
8π || (ω, q) 2
(2.105)
In the case of a transverse field
 e 2 ω
W⊥ = − |A⊥ (ω, q)|2 Im χ⊥ (ω, q) ,
c 2
where
ω ω2
Im χ⊥ (ω, q) = − 2
Re σ⊥ (ω, q) = − Im ⊥ (ω, q) .
e 4π e2
Therefore,

ω2 1
W⊥ = 2
Re σ⊥ (ω, q) |A⊥ (ω, q)|2 = Re σ⊥ (ω, q) |E ⊥ (ω, q)|2
2c 2 (2.106)
ω
= Im ⊥ (ω, q) |E ⊥ (ω, q)| .
2
8π
If q = 0, longitudinal and transverse responses become equal, so that

ω 1    
W = W⊥ = − Im  E ext (ω, 0)2 = 1 Re σ|| (ω, 0)  E(ω, 0)2 .
8π || (ω, 0) 2
(2.107)
Problems

2.1 Linear response of free electrons—Consider a free electron Hamiltonian

 †
H0 = k ckσ ckσ ,
kσ
78 2 Linear Response Theory

in the presence of a Zeeman splitting due to a slowly varying magnetic field oriented
along the z-direction B(q, t). The perturbation is

1
V (q, t) = μ B g B(q, t) Sz (−q),
V
where V is the volume, μ B the Bohr magneton, g  2 the electron gyromagnetic
ratio, and Sz (q) the spin-density operator at momentum q defined through

1  † †

Sz (q) = ck↑ ck+q↑ − ck↓ ck+q↓ .
2
k

By linear response theory, the expectation value of Sz (q) is

ˆ
 Sz (q) (t) = g μ B dt  χ (q, t − t  ) B(q, t  ) ,

where
i 1  
χ (q, t − t  ) = − θ (t − t  )  Sz (q, 0), Sz (−q) 
 V
is the magnetic response function per unit volume.

• Find the formal expression of

ˆ
χ (q, ω) = dt eiωt χ (q, t).

• Calculate the uniform magnetic susceptibility, namely

χ = − lim χ (q, ω = 0),

q→0

at zero temperature in terms of the density of states

1 
N( ) ≡ δ( − k ).
V
k

• Prove that
χ (q = 0, ω) = 0,
and discuss why.
Problems 79

Fig. 2.3 Geometry of the

tunnelling problem (2)
T
I
µ
L
R

2.2 Tunnelling between two leads—Consider two disconnected metallic leads, a

right one (R) and a left one (L), see Fig. 2.3, described by the unperturbed Hamiltonian
   μ  
H0 = k c†R kσ c R kσ + c†L kσ c L kσ + c†R kσ c R kσ − c†L kσ c L kσ
2
kσ kσ
   μ
† †
≡ k c R kσ c R kσ + c L kσ c L kσ + (N R − N L )
2
kσ
(2.108)
where μ = eV is the bias difference between the leads. At some given time, a tun-
nelling between the two leads is switched on, by the perturbation

V =− T ( k, p) c†R kσ c L pσ + H .c. ,
kσ

where we assume that the tunnelling amplitudes T ( k , p) depend only on the ener-
gies. Therefore, the fully perturbed Hamiltonian is

H = H0 + V . (2.109)

• Using the Heisenberg equation of motion with the Hamiltonian (2.109), calculate
the expression of the operator of the particle current flowing from the L to the R
lead, which is defined through

∂  i  
I = N R − NL = − (N R − N L ) , H .
∂t 
• Calculate at the leading non-zero order in the tunnelling, actually second order,
the expectation value I (V ) of the current operator, assuming that the density of
states of both leads is N ( ).
Hartree-Fock Approximation
3

In this chapter, we describe the simplest approximation to tackle interacting fermions:

the Hartree-Fock approximation. This technique is variational and can be applied both
at zero and at finite temperature. Essentially, within the Hartree-Fock approximation,
the effects of the particle-particle interaction are described by an effective single-
particle potential that is self-consistently determined. This is also the reason why the
Hartree-Fock approximation is a Mean Field theory.

3.1 Hartree-Fock Approximation for Fermions at Zero

Temperature

The Hartree-Fock (HF) approximation at zero temperature consists in searching

for a Slater determinant that minimises the expectation value of the total energy.

Since in general the ground state wavefunction is not a single Slater determinant,
the HF approach is variational; hence, the HF energy is an upper bound to the true
ground state energy.
The trial HF wavefunction is by definition a Slater determinant, namely
 
 φ1 (x1 ) · · · φ1 (x N ) 

1  φ2 (x1 ) · · · φ2 (x N ) 
 H F (x1 , . . . , x N ) = √  . .. .. .. .. , (3.1)
N !  .. ... . 

 φ N (x1 ) · · · φ N (x N ) 

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 81

M. Fabrizio, A Course in Quantum Many-Body Theory, Graduate Texts in Physics,
https://doi.org/10.1007/978-3-031-16305-0_3
82 3 Hartree-Fock Approximation

where N is the electron number, and the single-particle wavefunctions φi (x) are to
be determined variationally. However, they are assumed to belong to a complete set
of orthonormal wavefunctions

d x φi (x)∗ φ j (x) = δi j . (3.2)

The Hamiltonian in first quantisation is


N
1 
H= T (xi , pi ) + U (xi , x j ) ,
2
i=1 i= j

where
2 2
T (x, p) = − ∇ + V (x)
2m
is the non-interacting single-particle contribution that includes the kinetic energy as
well as a potential.
The expectation value of the Hamiltonian over the wavefunction (3.1) is

 H F | Ĥ | H F 
EH F = . (3.3)
 H F | H F 

One has to impose that the variation of E H F vanishes upon varying the trial wave-
function, keeping it still of the form of a Slater determinant. This amounts to change
one of the single particle wavefunctions, namely

φi (x) → N i (φi (x) + δφi (x)) , (3.4)

with Ni a normalisation constant. Clearly, if the variation δφi (x) has a finite overlap
with any of the φk (x)’s already present in (3.1), the Slater determinant does not
change; hence, such variation is irrelevant. Therefore, the only meaningful possibility
is that
δφi (x) = η φ j (x) , (3.5)
where η is an infinitesimal quantity and φ j belongs to the set of wavefunctions (3.2)
but does not appear in (3.1), namely j > N . The normalisation Ni in (3.4) is given
by

−2   
Ni = d x φi (x)∗ + η∗ φ j (x)∗ φi (x) + η φ j (x) = 1 + |η|2 ,

and thus Ni 1 at linear order in η. This implies that also the normalisation of the
Slater determinant  H F + η δ H F remains one at linear order in η; hence,

δ E H F = η∗ δ H F | H |  H F  + η  H F | H | δ H F 
   
= Re η Re δ H F | H |  H F  − Im η Im δ H F | H |  H F  = 0 .
3.1 Hartree-Fock Approximation for Fermions at Zero Temperature 83

Since η is an arbitrary infinitesimal number, this implies that

   
Re δ H F | H |  H F  = Im δ H F | H |  H F  = 0 ,

namely
δ H F | H |  H F  = 0 . (3.6)
Let us rephrase (3.6) in second quantisation. We associate to any of the wave-
functions φi (x) an annihilation and a creation operator, ci and ci† , respectively. In
second quantisation, the wavefunction (3.1) corresponds to the Fock state

N
| H F  = ci† | 0 ,
i=1

while the variation

|δ H F  = c†j ci | H F  ,
with j > N while i ≤ N . The Hamiltonian in second quantised form is
 1 
Ĥ = ti j ci† c j + Ui jkl ci† c†j ck cl , (3.7)
2
ij i jkl

where the parameters are the matrix elements over the basis set (3.2). What we need
to solve is therefore the equation

 H F | ci† c j H |  H F  = 0, i ≤ N , j >N.

For that we need the following two equalities, which can be easily derived:

tkl  H F | ci† c j ck† cl |  H F  = t ji ,
kl
1 
Uklmn  H F | ci† c j ck† cl† cm cn |  H F 
2
klmn

1 
N N
= U jmmi + Um jim − U jmim − Um jmi = U jmmi − U jmim ,
2
m=1 m=1

where the last expression comes from the symmetry relation Ui jkl = U jilk . Equation
(3.6) thus implies that


N
t ji + U jmmi − U jmim = 0 , (3.8)
m=1
84 3 Hartree-Fock Approximation

for any j > N and i ≤ N . In other words, the Slater determinant which minimises
the total energy is constructed by single-particle wavefunctions φi (x)’s that have
matrix elements obeying (3.8). Let us suppose we find instead a set of wavefunctions
φi ’s satisfying

N
t ji + U jmmi − U jmim = i δi j . (3.9)
m=1
This set automatically satisfies also (3.8), hence it does solve our variational problem.
In first quantisation, (3.9) reads

N 
  
T (x, p) φi (x) + dy U (x, y) φm (y)∗ φm (y) φi (x) − φm (y)∗ φm (x) φi (y) = i φi (x) ,
m=1
(3.10)
which is the standard Hartree-Fock set of equations.
Notice that there might be several Slater determinants built up using N of the
wavefunctions solving (3.9) and which satisfy the HF variational principle. Among
them, one has to choose the Slater determinant with the minimum total energy for
N electrons, which can be easily found to be

 1   
N N
E H F (N ) =  H F | H |  H F  = tii + Uimmi − Uimim
2
i=1 i,m=1
(3.11)
 1   
N N
= i − Uimmi − Uimim .
2
i=1 i,m=1

If for N − 1 particles the Hartree-Fock single-particle wavefunctions stay approxi-

mately invariant, then

−1 N −1
 1   
N
E H F (N − 1) tii + Uimmi − Uimim
2
i=1 i,m=1
 1   
N N
 
= tii (1 − δi N ) + Uimmi − Uimim 1 − δi N 1 − δm N
2
i=1 i,m=1
= E H F (N ) −  N ,
(3.12)
showing that the Hartree-Fock single-particle energies correspond approximately to
the ionisation energies.

3.1.1 Alternative Approach

The Hartree-Fock equations (3.10) are complicated non-linear integral-differential

equations. This is especially true if one does not impose any constraint on the form of
3.1 Hartree-Fock Approximation for Fermions at Zero Temperature 85

the variational Slater determinant dictated for instance by the symmetry properties of
the Hamiltonian, what is called unrestricted Hartree-Fock approximation. However,
very often one expects that the true ground state has well-defined properties under
symmetry transformations that leave the Hamiltonian invariant. For instance, if the
Hamiltonian is translationally and spin-rotationally invariant, one may expect that
the true ground state is an eigenstate of the total spin and of the total momentum.
For this reason, one would like to search for a variational wavefunction within the
subspace of Slater determinants which are eigenstates of the total spin and momen-
tum. This amounts to impose symmetry constraints on the general form (3.1) of the
variational wavefunction. Yet, this is not a simple task if one keeps working within
first quantisation. The second quantisation approach to the Hartree-Fock approxi-
mation that we describe in the following has the big advantage to allow an easy
implementation of such symmetry constraints.
Suppose that our Hamiltonian is written in a basis of single-particle wavefunctions
{φα (x)} which is more convenient to work with, for instance Block waves, and reads
 1 
H= tαβ cα† cβ + Uαβγ δ cα† cβ† cγ cδ . (3.13)
2
αβ αβγ δ

Our scope is to find the basis {φi (x)} which solves the Hartree-Fock equations (3.9).
The Hartree-Fock wavefunction (3.1), being a Slater determinant, should be the
ground state of a single-particle Hamiltonian, which we define as Ĥ H F . We write
such a Hamiltonian in the following general form:

Ĥ H F = h αβ cα† cβ , (3.14)
αβ

where we have introduced a set of unknown variational parameters satisfying h αβ =

h ∗βα for the Hamiltonian to be Hermitian. The ignorance about the basis set {φi (x)} is
reflected in the ignorance about the h αβ ’s. The Hartree-Fock wavefunction satisfies

Ĥ H F |  H F  = E |  H F , (3.15)

namely is the lowest energy eigenstate of Ĥ H F . We define

αβ =  H F | cα† cβ |  H F  , (3.16)

the expectation values of all bilinear operators on the wavefunction, which are there-
fore functional of the variational parameters h αβ . Since

E =  H F | Ĥ H F |  H F  = h αβ αβ , (3.17)
αβ

it follows that
∂E  ∂ αβ
= μν + h αβ . (3.18)
∂h μν ∂h μν
αβ
86 3 Hartree-Fock Approximation

On the other hand, since |  H F  is an eigenstate of the Hamiltonian Ĥ H F , the

Hellmann-Feynman theorem holds,1 according to which

∂E ∂ Ĥ H F
=  H F | | H F  = μν . (3.19)
∂h μν ∂h μν

By comparing (3.18) with (3.19), we thus conclude that

 ∂ αβ
h αβ = 0, (3.20)
∂h μν
αβ

for any h μν .
On the other hand, the expectation value of the original Hamiltonian (3.13) on
the Hartree-Fock wavefunction is readily found to be

E H F =  H F | Ĥ |  H F 
 1   
(3.21)
= tαβ αβ + αβ γδ Uαγ δβ − Uαγβδ ,
2
αβ αβγ δ

which is also a functional of the h αβ ’s. Minimisation of E H F requires as a necessary

condition that
∂ EH F  ∂ αβ   
= tαβ + γ δ Uαγ δβ − Uαγβδ = 0. (3.22)
∂h μν ∂h μν
αβ γδ

The Hartree-Fock wavefunction, hence the variational parameters h μν , must there-

fore satisfy both (3.20) and (3.22) for any h μν , which thus implies that
  
h αβ = tαβ + γ δ [h] Uαγ δβ − Uαγβδ . (3.23)
γδ

1The Hellmann-Feynman theorem states that, if | ψ is the normalised eigenstate with eigenvalue
E of a Hamiltonian Ĥ that depends on some parameter λ, then

∂E ∂ Ĥ
= ψ | | ψ .
∂λ ∂λ
This theorem can be easily proved. Indeed

∂E ∂ Ĥ ∂ψ ∂ψ
= ψ | | ψ +  | Ĥ | ψ + ψ | Ĥ | 
∂λ ∂λ ∂λ ∂λ
∂ Ĥ ∂ψ ∂ψ
= ψ | | ψ + E  | ψ + ψ | 
∂λ ∂λ ∂λ
∂ Ĥ ∂ψ | ψ ∂ Ĥ
= ψ | | ψ + E = ψ | | ψ ,
∂λ ∂λ ∂λ
since |ψ is normalised for any λ.
3.1 Hartree-Fock Approximation for Fermions at Zero Temperature 87

Since the average values γ δ [h] are themselves functional of the h’s, the above set of
equations is actually a self-consistency equation. These equations are fully equivalent
to solving the Hartree-Fock non-linear differential equations that are found in first
quantisation with one major advantage. Indeed, the symmetry properties of |  H F 
are completely determined by the parameters h αβ . In turn, this implies that we can
implement any desired symmetry operation Ô by imposing that
 
Ô, Ĥ H F = 0 ,

which leads to simple conditions to be imposed on the h αβ . For instance, if the

quantum number α = (a, σ ) includes an orbital index a and spin σ , and we would
like to enforce full spin-rotational symmetry, then we must assume that

h (aσ )(bσ ) = δσ σ h ab

does not depend on the spin indices. If we expect that spin SU (2) symmetry is
lowered down to U (1), namely the rotation around the magnetisation axis, which
we can always assume to be also the quantisation axis, then we have to impose that,
while h (aσ )(bσ ) are finite and depend on σ , h (a↑)(b↓) and h (a↓)(b↑) are identically
zero.
We note that (3.23) could be also regarded as the definition of h αβ in terms of new
variational parameters γβ that, only at the end, must be fixed in such a way that

αβ =  H F | cα† cβ |  H F , (3.24)

the expectation value being done on a chosen eigenstate of

   
Ĥ H F = cα† cβ tαβ + γδ Uαγ δβ − Uαγβδ , (3.25)
αβ γδ

which depends parametrically on the ’s. Moreover, it follows that the variational
energy
 
EH F =  H F | Ĥ |  H F 
1   
=  H F | Ĥ H F |  H F  − αβ γ δ Uαγ δβ − Uαγβδ
2
αβγ δ
   1   
=E h − αβ γ δ Uαγ δβ − Uαγβδ (3.26)
2
αβγ δ

also becomes functional of the ’s. Through (3.19) and (3.23), we find that

∂E  ∂ E ∂h αβ   
= = αβ U αγ δβ − U αγβδ ,
∂ γδ ∂h αβ ∂ γ δ
αβ αβ
88 3 Hartree-Fock Approximation

which also implies

∂E   
− αβ Uαγ δβ − Uαγβδ = 0 ,
∂ γδ αβ

or, equivalently,

δ 1   
E[ ] − αβ αβ Uαα β β − Uαα ββ = 0. (3.27)
δ γδ 2
αβα β

The equations (3.23), (3.25), (3.26), and (3.27) suggest two alternative ways of
stating the Hartree-Fock variational problem, which are both easy to implement:

Given the interacting Hamiltonian

 1 
H= tαβ cα† cβ + Uαβγ δ cα† cβ† cγ cδ , (3.28)
2
αβ αβγ δ

the Hartree-Fock variational wavefunction |  H F  is the ground state with

energy E of the single-particle Hamiltonian
   
Ĥ H F = tαβ cα† cβ + cα† cβ γδ Uαγ δβ − Uαγβδ , (3.29)
αβ αβγ δ

where the parameters αβ have to be determined self-consistently by imposing

either of the following two conditions:

Condition 1.
αβ =  H F | cα† cβ |  H F  . (3.30)
Condition 2.

δ 1   
E[ ] − αβ γδ Uαγ δβ − Uαγβδ = 0.
δ μν 2
αβγ δ
(3.31)
3.2 Hartree-Fock Approximation for Fermions at Finite Temperature 89

3.2 Hartree-Fock Approximation for Fermions at Finite

Temperature

A variational Hartree-Fock approximation can be defined also at finite temperature

T (we take k B = 1).
Consider again the interacting Hamiltonian
 1 
H= tαβ cα† cβ + Uαβγ δ cα† cβ† cγ cδ , (3.32)
2
αβ αβγ δ

represented in a given basis of single-particle wavefunctions, and define the free-

energy functional
   
F[ρ, T ] ≡ U [ρ] − T S[ρ] = Tr ρ H + T Tr ρ ln ρ , (3.33)

where U [ρ] is the internal energy, S[ρ] the von Neumann entropy, and ρ a den-
sity matrix, i.e., a linear Hermitian
 and positive-definite operator in the many-body
Hilbert space such that Tr ρ) = 1. The true free energy F(T ) of the system described
by the Hamiltonian (3.32) satisfies

F(T ) ≡ min F[ρ, T ] . (3.34)

ρ

However, even though we do know what is the density matrix that minimises F[ρ, T ],
namely, the Boltzmann distribution

e−β H
ρmin ≡   ,
Tr e−β H

that knowledge is useless for any practical purpose, since ρmin is a complicated
multiparticle operator, which can be dealt with only at very high-temperature through
an expansion in β = 1/T → 0.
Alternatively, we could search for a minimum in a restricted subspace S of all
density matrices. In that way, we can only find an upper limit to the true free energy,
namely
F(T ) ≤ min F[ρ, T ] . (3.35)
ρ∈S

Equation (3.35) is the desired variational principle at T = 0.

Since we can easily deal only with density matrices of non-interacting systems, we
can take the subspace S as that of all Boltzmann distributions ρ H F of non-interacting
Hamiltonians, namely

e−β HH F 
ρH F ≡   , HH F = h αβ cα† cβ . (3.36)
Tr e−β HH F αβ
90 3 Hartree-Fock Approximation

Therefore, the optimisation procedure amounts to search for the parameters h αβ that
minimise F[ρ, T ] in (3.33). We note that
     
F ρ H F , T = Tr ρ H F H + T Tr ρ H F ln ρ H F
      
= Tr ρ H F H H F + T Tr ρ H F ln ρ H F + Tr ρ H − H H F
    
= FH F (T ) + Tr ρ H F H − H H F = FH F (T ) + H − H H F H F ,
(3.37)
where
 
FH F (T ) = −T ln Z H F , Z H F = Tr e−β HH F (3.38)

is the true free energy of the system described by H H F , and . . .  H F is the thermal
expectation value on the density matrix ρ H F .
Since H H F in (3.36) depends on the variational parameters h αβ , so do ρ H F and
FH F (T ). Therefore, the minimum must first of all be a saddle point, i.e., satisfy
 
∂ F ρH F , T
= 0, (3.39)
∂h αβ

and, to be a minimum rather than a maximum, its Hessian, i.e., the tensor with
components
 
∂ 2 F ρH F , T
Hαβ,γ δ ≡ , (3.40)
∂h αβ ∂h γ δ
has to be positive definite.

3.2.1 Saddle Point Solution

Let us start by solving the saddle point equation (3.39). We note that

1   
V ≡ H − HH F = Uαβγ δ cα† cβ† cγ cδ + tαβ − h αβ cα† cβ , (3.41)
2
αβγ δ αβ

so that we must first learn how to calculate  V  H F . Since H H F is quadratic in the

fermionic operators, it can be diagonalised by some unitary transformation

cα = Uαi ci ,
i

after which

HH F = i ci† ci .
i
3.2 Hartree-Fock Approximation for Fermions at Finite Temperature 91

Any many-body eigenstate of H H F , e.g., |n with energy E n , can be written as a

single Slater determinant, i.e., a Fock state with |n ≡ {n i }. Through (3.36), the
following property holds:

1   
 cα† cβ† cγ cδ  H F = †
Uiα U †jβ Uγ k Uδl Tr e−β HH F ci† c†j ck cl
ZHF
i jkl
1  
= †
Uiα U †jβ Uγ k Uδl e−β E n n | ci† c†j ck cl | n
ZHF n
i jkl
1  †
= Uiα U †jβ Uγ k Uδl
ZHF
i jkl (3.42)
  
−β E n
e δil δ jk − δik δ jl n | ci† ci | n n | c†j c j | n
n
 †  
= Uiα U †jβ Uγ k Uδl δil δ jk − δik δ jl f i (T ) f j (T )
i jkl

=  cα† cδ  H F  cβ† cγ  H F −  cα† cγ  H F  cβ† cδ  H F

≡ αδ (T ) βγ (T ) − αγ (T ) βδ (T ) ,

where2
1
f i (T ) = βi
e +1
is the Fermi distribution function, and we have introduced the thermal expectation
values over the Hartree-Fock density matrix

αβ (T ) =  cα† cβ  H F .

2 In the previous equation, we make use of the following trivial result. If

 
H= i ci† ci = i n i ,
i i

then
e−β H e−β i n i e−β i n i
ρ=   = ρi , ρi =   = ,
Tr e−β H i Tr e−β i n i 1 + e−β i

the last expression deriving from the Pauli principle, according to which n i can be either zero or
one. Therefore, e.g.,
     
 n i n j  = Tr ρ n i n j = Tr ρi n i Tr ρ j n j = f i (T ) f j (T ) ,

since
   e−β i n i e−β i 1
Tr ρi n i = −β i
= = ≡ f i (T ) .
n i =0,1
1+e 1 + e−β i eβ i + 1
92 3 Hartree-Fock Approximation

c†1 c2 c†3 c4 = c†1 c2 c†3 c4 + c†1 c2 c†3 c4

= c†1 c2 c†3 c4 + c†1 c4 c†2 c3

c†1 c†2 c3 c4 = c†1 c†2 c3 c4 + c†1 c†2 c3 c4

= c†1 c4 c†2 c3 c†1 c3 c†2 c4

Fig. 3.1 Graphical way to determine the sign of the product of the contractions and the type of each
of them shown for two cases. One has to join the two operators, one annihilation, e.g., ci , and one
creation, e.g., c†j , which are contracted together by an arrow line directed from the annihilation to
the creation. If the line goes from left to right the contraction means  ci c†j , while, if the line goes
from the right to the left, the contraction means  c†j ci . The sign of each product of contractions
is (−1)n , where n is the number of crossings of the lines, which is zero in all shown cases but in
the last one, when n = 1, the green dot in the figure

The previous result is a special case of the following general result:

The thermal average of a product of n creation and n annihilation operators

with a bilinear non-interacting Hamiltonian H0 , namely

1  
 ci†1 ci†2 . . . ci†n c j1 c j2 . . . c jn 0 = Tr e−β H0 ci†1 ci†2 . . . ci†n c j1 c j2 . . . c jn
Z0

= (−1)  ci1 c j P 0  ci2 c j P 0 . . .  ci†n c j P 0 ,
P † †
1 2 n
P
  (3.43)
where j P1 , j P2 , . . . , j Pn = P( j1 , j2 , . . . , jn ) is a permutation of the n j-
indices, and P is the order of the permutation.
The most general case where the operators are not ordered in such a way that
creation operators appear on the left can be straightforwardly derived from the
above result. Essentially one has to consider all possible contractions between
a creation, ci† , and an annihilation, c j , operator. If the former is on the left of
the latter, the contraction means

 ci† c j 0 ;

otherwise, it means
 c j ci† 0 = δi j −  ci† c j 0 .
The sign of a given product of contractions is plus if the number of fermionic
hops one has to perform so to bring each pair of operators to be contracted close
together is even, and minus otherwise. A simple graphical implementation of
those rules is shown in Fig. 3.1.
3.2 Hartree-Fock Approximation for Fermions at Finite Temperature 93

We mention that the above rule is required to demonstrate the so-called Wick’s
theorem, which we shall discuss later.
Coming back to our original task, we need to solve the saddle point equation (3.39),
namely
 
∂ F ρH F , T ∂ FH F (T ) ∂ V  H F
= + =0. (3.44)
∂h αβ ∂h αβ ∂h αβ
Let us start from the first term on the right-hand side. We note that

∂ FH F (T ) ∂ ln Z H F (T ) T ∂ZHF
= −T =− . (3.45)
∂h αβ ∂h αβ Z H F ∂h αβ

On the other hand, if E n are the eigenvalues of the many-body eigenstates | n of

Ĥ H F , then

∂ZHF  ∂e−β E n
=
∂h αβ n
∂h αβ
 ∂ En  ∂ Ĥ H F
= −β e−β E n = −β e−β E n n | | n (3.46)
n
∂h αβ n
∂h αβ

= −β e−β E n n | cα† cβ | n = −β Z H F αβ (T ) ,
n

where we made use of the aforementioned Hellmann-Feynman theorem. Therefore,

through (3.45) and (3.46), we find that

∂ FH F (T )
= αβ (T ) , (3.47)
∂h αβ

a simple extension of the Hellmann-Feynman theorem at finite temperature.

Considering now the last term on the right-hand side of (3.44), we find, by means
of (3.42), that before the derivative it reads

1     
V̂  H F = αβ (T ) γδ (T ) U αγ δβ − U αγβδ + tαβ − h αβ αβ (T ) ,
2
αβγ δ αβ

hence

∂ V  H F  ∂
γ δ (T )
=− αβ (T ) + tγ δ − h γ δ
∂h αβ ∂h αβ
γδ
 ∂   (3.48)
γ δ (T )
+ μν (T ) Uγ μνδ − Uγ μδν .
∂h αβ
γ δμν
94 3 Hartree-Fock Approximation

The sum of (3.47) and (3.48) yields

 
 ∂   
γ δ (T )
tγ δ − h γ δ + μν (T ) Uγ μνδ − Uγ μδν = 0, (3.49)
∂h αβ μν
γδ

which must hold for every pair of indices (αβ). The solution of this equation thus
reads
  
h γ δ = tγ δ + μν (T ) Uγ μνδ − Uγ μδν
μν
1     (3.50)
= tγ δ + Uγ μνδ − Uγ μδν Tr e−β ĤH F cμ
†
cν .
ZHF μν

Therefore, the single-particle Hamiltonian H H F that minimises F[ρ H F , T ], hence

providing an upper bound to the exact free energy, is defined through variational
parameters h αβ that have to be determined self-consistently through (3.50). We note
that this equation is just the extension of the Hartree-Fock Hamiltonian (3.14) at finite
temperature, where the variational parameters correspond now to thermal averages
and not anymore to ground state averages.
Correspondingly, if we instead write

   
HH F = tαβ + γδ Uαγ δβ − Uαγβδ cα† cβ , (3.51)
αβ γδ

as functional of the parameters αβ , so that

 
F ρ H F , T = FH F (T ) + V̂  H F
1   
(3.52)
= FH F (T ) − αβ γδ Uαγ δβ − Uαγβδ
2
αβγ δ

also becomes functional of the αβ ’s, then the saddle point corresponds to imposing
 
Tr e−β HH F cα† cβ
αβ =  cα† cβ  H F =   , (3.53)
Tr e−β HH F

or, equivalently, to the solution of

 
∂ F ρH F , T
= 0, (3.54)
∂ αβ

which are just the finite temperature generalisation of (3.30) and (3.31).
3.2 Hartree-Fock Approximation for Fermions at Finite Temperature 95

However, we still need to determine under which conditions the above saddle
point is a minimum. Through (3.49), we readily find that the elements of the Hessian
matrix are
    
∂ 2 F ρH F , T ∂ ∂ γ δ (T )    
= tγ δ − h γ δ + μν (T ) U γ μνδ − U γ μδν .
∂h αβ ∂h ηξ ∂h ηξ ∂h αβ μν
γδ

The derivative with respect to h ηξ of ∂ γ δ (T )/∂h αβ vanishes since the term in

square brackets is zero at the saddle point. It follows that

  ⎡ ⎤
∂2 F ρH F , T  ∂ γ δ (T ) ∂h γ δ  ∂ μν (T )  
= ⎣− + Uγ μνδ − Uγ μδν ⎦
∂h αβ ∂h ηξ ∂h αβ ∂h ηξ μν
∂h ηξ
γδ
∂ ηξ (T )  ∂ γ δ (T ) ∂ μν (T )  
=− + Uγ μνδ − Uγ μδν
∂h αβ ∂h αβ ∂h ηξ
γ δμν

∂2 F H F (T )
 ∂ 2 FH F (T ) ∂ 2 FH F (T )  
=− + Uγ μνδ − Uγ μδν .
∂h αβ ∂h ηξ ∂h αβ ∂h γ δ ∂h μν ∂h ηξ
γ δμν

We note that
∂ 2 FH F (T )
− ≡ καβ,ηξ
HF (3.55)
∂h αβ ∂h ηξ
are the components of the thermodynamic susceptibility tensor κ̂ H F of the Hartree-
Fock Hamiltonian, which is positive definite. Therefore, the Hessian components are
defined through
  
Hαβ,ηξ = καβ,ηξ
HF
+ καβ,γ
HF
δ Uγ μνδ − Uγ μδν κμν,ηξ ,
HF

γ δμν

thus, in matrix notations,

Ĥ = κ̂ H F + κ̂ H F Û κ̂ H F , (3.56)

which also defines the tensor Û . It follows that the saddle point is a minimum if Ĥ
in (3.56) is positive definite, which is surely the case if Û κ̂ H F is a weak correction
to the identity matrix.
We end noticing that the Hartree-Fock Hamiltonian H H F represents non-
interacting electrons in the presence of a fictitious external field that, through (3.50),
is indeed the self-consistent field generated by the same electrons. This is the reason
why the Hartree-Fock approximation is essentially a mean-field approximation.
96 3 Hartree-Fock Approximation

3.3 Time-Dependent Hartree-Fock Approximation

Suppose we have solved the time-independent Hartree-Fock problem, namely, we

have found a set of single-particle wavefunctions φi (x) that diagonalise

N 
 
ti j + Uikk j − Uik jk = i δi j . (3.57)
k=1

The Hartree-Fock wavefunction is just the Fock state

N
| H F  = ci† | 0 , (3.58)
i=1

where ci† creates a fermion in state φi . The original Hamiltonian in the Hartree-Fock
basis is given by
 1 
H= ti j ci† c j + Ui j,kl ci† c†j ck cl . (3.59)
2
ij i jkl

Let us suppose to perturb the Hamiltonian by a time-dependent perturbation


V (t) = Vi j (t) ci† c j (3.60)
ij

and study the response of the system to linear order. Therefore, the full Hamiltonian
H (t) = H + V (t) is time-dependent and the variational principle now involves the
time-dependent Shrœdinger equation. Namely, we are now going to search within
the subspace of time-dependent Slater determinants for a | H F (t) that satisfies

⎡ ∂ ⎤
 H F (t) | i − H (t) |  H F (t)
⎢ ∂t ⎥
δ⎣ ⎦=0 . (3.61)
 H F (t) |  H F (t)

We assume that the time-dependent Slater determinant is build by N single-particle

normalised wavefunctions φα (x, t). By analogy with the conventional derivation of
the time-independent Hartree-Fock equations, one finds that

∂
i φα (x, t) = [T (x, p) + V (x, p, t)] φα (x, t)
∂t
N 
  2
+ dy U (x, y) φβ (y, t) φα (x, t) − φβ (y, t)∗ φβ (x, t) φα (y, t) ,
β=1
(3.62)
3.3 Time-Dependent Hartree-Fock Approximation 97

where T (x, p) is the single-particle term of the Hamiltonian, V (x, p, t) the pertur-
bation, and U (x, y) the interaction. We assume that the φα (x, t)’s are related to the
φi (x)’s by the time-dependent unitary transformation
 
φα (x, t) = Uiα (t) φi (x) , Uiα (t) = d x φi (x)∗ φα (x, t) ,
i

which, exploiting the invariance of the Fermi fields, imply the following transforma-
tions of the corresponding annihilation operators:
 
ci = Uiα (t) cα , ci† = Uiα (t)∗ cα† . (3.63)
α α

In terms of the matrix elements, the time-dependent equations (3.62) read

∂  
i Uiα (t) = ti j + Vi j (t) U jα (t)
∂t
j


N   
+ Ukβ (t)∗ Ulβ (t) U jα (t) Uikl j − Uik jl ,
β=1 jkl
 (3.64)
∂ 
i Uiα (t)∗ = − t ji + V ji (t) U jα (t)∗
∂t
j


N   
− Ukβ (t) Ulβ (t)∗ U jα (t)∗ U jlki − Ul jki .
β=1 jkl

Through (3.63), the time-dependent Hartree-Fock wavefunction is uniquely identi-

fied by the expectation values


N
i j (t) =  H F (t) | ci† c j |  H F (t) = Uiα (t)∗ U jα (t) , (3.65)
α=1

which, calculated on the time-independent wavefunction (3.58), are simply

(0)
ij = δi j n i , (3.66)

with n i = 1 if i ≤ N and n i = 0 otherwise. Through (3.64) and (3.65), we obtain

the following equations for the i j ’s:
∂      
i i j (t) = − tki + Vki (t) k j (t) + t jk + V jk (t) ik (t)
∂t
k
  
+ − kl (t) m j (t) Umkli − Ukmli (3.67)
klm
 
+ il (t) km (t) U jkml − U jklm .
98 3 Hartree-Fock Approximation

We solve the above equation up to the first order in the perturbation V (t). One can
readily check that the zeroth order time-independent term is indeed given by (3.66).
The first-order terms instead satisfy the equations
 N 

∂   
(1) (1)
i i j (t) = − tki + Ukmmi − Umkmi k j (t)
∂t
k m=1
 n 

  
(1)
+ t jk + U jmmk − U jmkm ik (t)
k m=1
     
(1)
+ ni − n j ml (t) U jmli − U jmil + V ji (t) n i − n j .
lm
(3.68)
Through (3.57) we finally obtain that, for any pair of indices i and j, the following
equation must be satisfied:

∂     
(1) (1)
i −  j + i i j (t) = n i − n j V ji (t) + kl (t) U jkli − U jkil ,
∂t
kl
(3.69)
which is the desired result. We note that (3.69) is just the equation of motion of
non-interacting electrons described by the diagonalised Hartree-Fock Hamiltonian

HH F = i ci† ci , (3.70)
i

at linear order in a self-consistent time-dependent potential

   
(1)
V∗ (t) = V∗ i j (t) ci† ci , V∗ i j (t) ≡ V ji (t) + kl (t) U jkli − U jkil ,
ij kl
(3.71)
once again showing the mean-field character of the approximation. Indeed, since

∂c j   
i = c j , H H F + V∗ (t) =  j c j + V∗ jk (t) ck ,
∂t
k
∂ci†   
i = ci† , H H F + V∗ (t) = −i ci† + V∗ ki (t) ck† ,
∂t
k

then

∂ ci† c j  ∂ i j (t)    
i ≡ i =  j − i i j (t) + V∗ jk (t) ik (t) − V∗ ki (t) k j (t) .
∂t ∂t
k
3.3 Time-Dependent Hartree-Fock Approximation 99

(0)
At zeroth order in V , ij = δi j n i , while consistently at first order,

(1)
∂ i j (t)   (1)
 
i =  j − i i j (t) + V∗(1)jk (t) (0)
ik (t) − V∗(1)
ki (t)
(0)
k j (t)
∂t
k
  (1)  
=  j − i i j (t) + n i − n j V∗(1)ji (t) ,
(3.72)
where
  
(1) (1)
V∗ ji (t) = V ji (t) + kl (t) U jkli − U jkil .
kl
Indeed, (3.72) coincides with (3.69). This observation allows us to easily calculate
linear response functions within the time-dependent Hartree-Fock approximation.
Let us consider again the non-interacting Hamiltonian H H F perturbed by V∗ (t). The
linear response function of H H F is by definition
 
χiHj,kl
F
(t) = −i θ (t)  ci† (t) c j (t) , ck† cl 

and is generally a tensor. Its Fourier transform in frequency can be readily obtained
through (3.72), and it is

ni − n j
χiHj,kl
F
(ω) = δil δ jk ≡ δil δ jk χiHj F (ω) , (3.73)
ω −  j + i

where ω has an infinitesimal imaginary part η > 0 that enforces causality. It follows
that, in linear response,

(1)

i j (ω) = χiHj,kl
F
(ω) V∗(1) (1)
kl (ω) = χi j (ω) V∗ ji (ω)
HF

kl
  
(1)
= χiHj F (ω) V ji (ω) + kl (t) U jkli − U jkil ,
kl

namely

  
(1)
δi j,kl − χiHj F (ω) U jkli − U jkil kl (ω) = χiHj F (ω) V ji (ω) ,
kl

(1)
which, upon defining the vector, (ω) and V (ω) with components i j (ω) and
Vi j (ω), respectively, and the tensors χ̂ H F (ω) and Q̂(ω) ≡ Iˆ − χ̂ H F (ω) Û with com-
ponents, respectively,
 
χiHj,kl
F
(ω) = δil δ jk χiHj F (ω) , Q i j,kl (ω) = δi j,kl − χiHj F (ω) U jkli − U jkil ,
100 3 Hartree-Fock Approximation

can be shortly written as

Q̂(ω) (ω) = χ̂ H F (ω) V (ω) ,

namely,
 −1
(ω) = Q̂(ω)−1 χ̂ H F (ω) V (ω) = Iˆ − χ̂ H F (ω) Û χ̂ H F (ω) V (ω)
(3.74)
≡ χ̂ t D−H F (ω) V (ω) ,

which defines the linear response functions within the time-dependent Hartree-Fock
approximation in terms of the simple Hartree-Fock one χ̂ H F (ω).
We have till now worked at zero temperature, T = 0. It is straightforward to extend
all previous results at T = 0 and find that the only difference is the expression of
χiHj F (ω) in (3.73), which at finite temperature becomes

n i (T ) − n j (T )
χiHj F (ω) = , (3.75)
ω −  j (T ) + i (T )

where i (T ) and  j (T ) are eigenvalues of the Hartree-Fock Hamiltonian at T = 0,

and
1
n i (T ) = −β (T ) .
e i +1
We finally observe that the time-dependent Hartree-Fock approximation, being based
on the variational principle directly applied to the Schrœding equation, does not
spoil the symmetries of the interacting Hamiltonian, evidently preserved during the
time evolution, unlike what may happen within the simple Hartree-Fock approxi-
mation. For that reason, the time-dependent Hartree-Fock approximation is the sim-
plest symmetry-conserving approximation scheme, and it is historically named as
the random-phase approximation (RPA).

3.3.1 Bosonization of the Low-Energy Particle-Hole Excitations

We note from (3.69) that the only expectation values that are affected at linear order
by the perturbation are those where either n i = 1 and n j = 0, or viceversa, namely
where one of the index refers to an occupied state within (3.58) and the other to an
unoccupied one. Let us denote by greek letters α, β, · · · > N the unoccupied states,
hereafter dubbed ‘particles’, and by roman letters a, b, · · · ≤ N the occupied ones,
named ‘holes’. We denote by
†
bαa = cα† ca , (3.76)
the particle-hole creation operator, which destroys one electron, i.e., creates one
hole, inside the Slater determinant, and creates it outside; as well as its Hermitian
conjugate
bαa = ca† cα . (3.77)
3.3 Time-Dependent Hartree-Fock Approximation 101

The commutators
   
†
bαa , bβb = bαa
†
, bβb =0

vanish, while
 
†
bαa , bβb = δαβ ca† cb − δab cβ† cα

has expectation value on (3.58)

   
†
 H F | bαa , bβb |  H F  = δαβ δab n a − n α = δαβ δab ,

as if the particle-hole creation and annihilation operators were bosonic particles. If we

assume that, at linear order in V (t), the Hartree-Fock wavefunction is very slightly
modified with respect to its time-independent value; in other words, we make a
similar assumption as we did in deriving spin-wave theory, then we can approximate
the commutator by its average value, thus obtaining
     
† †
bαa , bβb = δαβ δab , bαa , bβb = bαa
†
, bβb = 0, (3.78)

which are indeed bosonic commutation relations. We observe that in this approxima-
tion the bosonic vacuum is actually the time-independent Hartree-Fock wavefunc-
tion, whereas, in the spin-wave approximation, it is the classical spin configuration.
Since
aα (t) =  H F (t) | bαa |  H F (t) ,
Equation (3.69) corresponds to the following equations of motion for the bosonic
operators:

∂
i − α + a bαa = Vαa (t)
∂t
     (3.79)
†
+ bβb Uαbβa − Uαbaβ + bβb Uαβba − Uαβab .
βb

We may now ask the following question: What would be a bosonic Hamiltonian
leading to the above equation of motion? The answer can be readily found and reads
  †   
α − a bαa bαa + †
bαa bβb Uαbβa − Uαbaβ
αa αβab
1     
†
+ †
bαa bβb Uαβba − Uαβab + bβb bαa Uabβα − Uabαβ (3.80)
2
αβab
 
+ Vαa (t) bαa
†
+ Vaα (t) bαa .
αa
102 3 Hartree-Fock Approximation

It is now clear that the external field simply probes particle-hole excitations, which
have their own dynamics provided by the Hamiltonian
  †   
Ht D−H F = α − a bαa bαa + †
bαa bβb Uαbβa − Uαbaβ
αa αβab
1     
†
+ †
bαa bβb Uαβba − Uαβab + bβb bαa Uabβα − Uabαβ .
2
αβab
(3.81)
Due to the presence of the last two terms, the true ground state of (3.81) is not the
vacuum, namely the Hartree-Fock wavefunction, but another state

| t D−H F  = ei A |  H F  ,

with A an Hermitian operator quadratic in the bosonic operators. The above wave-
function is actually an improvement of the Hartree-Fock one that includes zero-point
fluctuations of the particle-hole excitations.
The above results can be rederived, still in the spirit of the spin-wave theory,
without even invoking the time-dependent variational principle. If we assume, to be
checked a posteriori, that the Hartree-Fock wavefunction is not strongly modified by
the inclusion of quantum fluctuations, we can assume that the commutators
   
ca† cα , cβ† cb = δαβ ca† cb − δab cβ† cα , cα† ca , cb† cβ = δab cα† cβ − δαβ cb† ca ,
   
ca† cb , cc† cd = δbc ca† cd − δad cc† cb , cα† cβ , cγ† cδ = δβγ cα† cδ − δαδ cγ† cβ

can be effectively substituted by their expectation values on the Hartree-Fock wave-

function, i.e.,

ca† cb  ca† cb  H F = δab , cα† cβ  cα† cβ  H F = 0 ,

so that
   
ca† cα , cβ† cb δαβ δab , cα† ca , cb† cβ −δαβ δab ,
   
ca† cb , cc† cd 0, cα† cβ , cγ† cδ 0,

which justifies the identification with bosonic operators. The non-interacting Hamil-
tonian plus the component of interaction that enters in the Hartree-Fock calculation
define the Hartree-Fock Hamiltonian, which can be represented as
  †
HH F EH F + α − a bαa bαa , (3.82)
αa

where E H F is the total energy of the Hartree-Fock Slater determinant, i.e., the vacuum
of the bosons, and the other term is just the cost of a particle-hole excitation.
3.3 Time-Dependent Hartree-Fock Approximation 103

However, the interaction includes also matrix elements that couple holes with
particles. The only matrix elements that may have non-vanishing effect when applied
either on the right or on the left to the Hartree-Fock wavefunction are those that
contain two particles and two holes, namely

1 
δ Hint Ubαβa cb† cα† cβ ca + Uαbaβ cα† cb† ca cβ
2
abαβ

+ Ubαaβ cb† cα† ca cβ + Uαbβa cα† cb† cβ ca


+ Uαβba cα† cβ† cb ca + Uabβα ca† cb† cβ cα

1   
= 2 Uαbβa − Uαbaβ cα† cb† cβ ca + Uαβba cα† cβ† cb ca
2
abαβ

+ Uabβα ca† cb† cβ cα .

We have to express δ Hint in terms of bosons. For the first term that is simple, i.e., it
remains the same
  
δ H1 int = Uαbβa − Uαbaβ bαa †
bβb . (3.83)
abαβ

The second term is more delicate to express in terms of the bosonic particle-hole
† †
operators, since one can at will pair α with a or b, thus getting either bαa bβb or
† †
−bαb bβa . Although the two terms correspond to the same fermionic operator, they
are independent from the bosonic point of view; Hence, they both have to be taken
into account, leading to

1   
†
δ H2 int = Uαβba − Uαβab bαa
†
bβb + H .c. . (3.84)
2
abαβ

Indeed, the sum of (3.82), (3.83), and (3.84) is just Ht D−H F of (3.81). Correspond-
ingly, the linear response functions obtained within the time-dependent Hartree-Fock
approximation are simply the linear response functions of the bosonic Hamiltonian
(3.81) upon consistently representing the perturbation in terms of bosons. Specifi-
cally,
  
V (t) = Vi j (t) ci† c j Vaα (t) ca† cα + Vαa (t) cα† ca
ij aα
  (3.85)
→ Vaα (t) bαa + Vαa (t) bαa
†
,
aα

where, since V (t) must be Hermitian, then Vαa (t) = Vaα (t)∗ so that the bosonic
representation is Hermitian, too.
104 3 Hartree-Fock Approximation

We remark that the bosonization of the particle-hole excitation has been con-
structed strictly at zero temperature, and therefore cannot be used at T = 0. Indeed,
the basic assumption we made was
 
bαa , bαa
†
 ca† ca  H F −  cα† cα  H F = n a (T ) − n α (T ) ,

which does reproduce the correct bosonic commutation relation only at T = 0, when
n a (0) = 1 and n α (0) = 0, but not at T = 0, when n a (T ) − n α (T ) is strictly less than
one, and vanishes for T → ∞.

3.4 Application: Antiferromagnetism in the Half-Filled

Hubbard Model

Let us consider the simplest version of a Hubbard model on a hypercubic lattice in

d-dimensions:
  †  
H = −t cR σ cR σ + H .c. + U n̂ R ↑ n̂ R ↓ , (3.86)
<RR > σ R

where
†
n̂ R σ = cR σ cR σ ,
and < RR > means that the sum runs over nearest neighbour bonds, i.e., that R and
R are nearest neighbour sites. We already showed that for very large U /t and one
electron per site, i.e., at half-filling, that model describes a Mott insulator with a Neèl
magnetic order in d > 1 at zero temperature and in d > 2 at finite temperature below
a critical TN . Let us now try to recover that result within the Hartree-Fock approxi-
mation. We use directly the T = 0 version, since the zero temperature Hartree-Fock
approximation is just a special case.
The first step is to identify the variational parameters. The interaction, being
on-site, leads to local variational parameters

†
R, σ σ =  cR σ cR σ  H F . (3.87)

We search for a Hartree-Fock Hamiltonian that describes a Neèl magnetic order with
the antiferromagnetic order parameter along the z-direction, namely we assume that
the spin SU (2) symmetry is lowered down to a U (1) symmetry that describes just
spin-rotations around the z-axis. As a consequence, only the diagonal variational
parameters R, σ σ ≡ n Rσ in (3.87) are allowed by the remaining U (1) symmetry.
Since the average number of electrons per site is σ n R σ = 1, we can use the
following parametrisation:

1 1
nR ↑ = + m (−1) R , n R ↓ = − m (−1) R , (3.88)
2 2
3.4 Application: Antiferromagnetism in the Half-Filled Hubbard Model 105

where, if the vector R = a(n 1 , n 2 , . . . , n d ), a being the lattice spacing, then R =

d
i=1 n i . In other words, the expectation values of components of the magnetisation
density are assumed to be

1 
Sz R = n R ↑ − n R ↓ = m (−1) R , Sx R = S y R = 0 .
2
The Hartree-Fock Hamiltonian is therefore
    
†
H H F = −t cR σ cR σ + H .c. + U n̂ R ↑ n R ↓ + n̂ R ↓ n R ↑
<RR > σ R
      U  
†
= −t cR σ cR σ + H .c. − U m (−1) R n̂ R ↑ − n̂ R ↓ + n̂ R ↑ + n̂ R ↓ .
2
<RR > σ R R

The last term is proportional to the total number of electrons N , which is conserved.
Therefore, that term, actually equal to U N /2, is just a constant that can be dropped
leading to the Hartree-Fock Hamiltonian
  †
   
H H F = −t cR σ cR σ + H .c. − U m (−1) R n̂ R ↑ − n̂ R ↓ .
<RR > σ R
(3.89)
Let us rewrite this Hamiltonian in momentum space. Since

(−1) R = eiQ·R ,

where Q = π(1, 1, . . . , 1)/a, one readily finds that

 †
 
HH F = k ckσ ckσ − U m ck† ↑ ck+Q ↑ − ck† ↓ ck+Q ↓ , (3.90)
kσ k

where, if k = (k1 , k2 , . . . , kd ), the bare dispersion is simply


d
k = −2t cos ki a .
i=1

We observe that, since Q ≡ −Q apart from a reciprocal lattice vector, then (k +

Q) + Q ≡ k, which implies a one-to-one correspondence between k and k + Q.
Moreover, it trivially holds that k+Q = −k . Therefore, if we define a Magnetic
Brillouin Zone (MBZ) as the volume that encloses all k-points such that k ≤ 0, the
rest of the Brillouin zone includes all partners k + Q, with k+Q > 0. It follows that
the volume of the MBZ is half that of the Brillouin zone, which also implies that at
half-filling the Fermi volume is just the MBZ, i.e., it includes all k’s with k ≤ 0,
and the Fermi surface is thus defined through k F : k F = 0. In Fig. 3.2, we show
the case of a square lattice.
106 3 Hartree-Fock Approximation

Fig. 3.2 Brillouin zone, the ky

outer square, magnetic

ce
Brillouin zone, the inner

fa
ur
rotated yellow square, and

is
Fermi surface, in red, of a

rm
Fe
tight-binding Hamiltonian MBZ
with nearest neighbour kx
hopping at half-filling and on
a square lattice

Since the Hartree-Fock Hamiltonian breaks translational symmetry—the new unit

cell is twice as large—the actual Brillouin zone is just the MBZ. Because of the one-
to-one correspondence between k and k + Q, we are allowed to define new fermionic
operators with k ∈ M B Z by

a k σ = ck σ , bk σ = ck+Q σ ,

and spinor operators

ak σ
ψk σ = . (3.91)
bk σ
The Hamiltonian (3.90) in terms of those spinors can be written as
   † 
HH F = k ψk† σ τ3 ψk σ − U m ψk ↑ τ1 ψk ↑ − ψk† ↓ τ1 ψk ↓ ,
k∈M B Z σ k∈M B Z
(3.92)
where τi ’s, i = 1, 2, 3, are the Pauli matrices acting in the spinor basis. We observe
that3
k Um
k τ3 ∓ U m τ1 = −E k τ3 − ±i τ2
Ek Ek
= −E k τ3 e±2i θk τ2 = −E k e∓i θk τ2 τ3 e±i θk τ2 ,

3 We recall that the Pauli matrices σi , i = 1, 2, 3 satisfy

! "  
σi , σi = δi j , σi σ j = δi j + i i jk σk , σi σ j σi = δi j σi − 1 − δi j σ j ,

where i jk is the antisymmetric tensor. It follows that σi2n = 1 while σi2n+1 = σi , and thus that
 (iθ)n  (iθ)2n  (iθ)2n+1
eiθ σi = σin = + σi = cos θ + i σi sin θ .
n! (2n)! (2n + 1)!
n≥0 n≥0 n≥0

Moreover, since σi = σi† = σi−1 is Hermitian and unitary, for i = j the following result holds:

e−iθ σi σ j = σ j σ j e−iθ σi σ j = σ j e−iθ σ j σi σ j = σ j eiθ σi .

3.4 Application: Antiferromagnetism in the Half-Filled Hubbard Model 107

where
#
Ek = k2 + U 2 m 2 , (3.93)
and
k Um
cos 2θk = −
, sin 2θk = .
Ek Ek
Therefore, if we define two new spinors through

αk ↑ αk ↓
φk ↑ = = eiθk τ2 ψk ↑ , φk ↓ = = e−iθk τ2 ψk ↓ , (3.94)
βk ↑ βk ↓

the Hamiltonian becomes

   
HH F = − E k φk† σ τ3 φk σ = − E k αk† σ αk σ − βk† σ βk σ ,
k∈M B Z σ k∈M B Z σ
(3.95)
namely it acquires a diagonal form. Since E k > 0 the ground state with a number of
electrons equal to the number of sites N is simply obtained by filling completely the
α-band (notice that the MBZ contains N /2 k-points, hence each band can accom-
modate 2 N /2 = N electrons). The Hartree-Fock Hamiltonian thus describes a band
insulator with valence and conduction bands separated by a gap of minimal value
2U m. It follows that

 φk† σ τi φk σ  H F = δi3  αk† σ αk σ − βk† σ βk σ  H F

     Ek (3.96)
= δi3 f − E k − f E k = δi3 tanh ,
2T
with f (x) the Fermi distribution function.
We still have to impose the self-consistency condition
1  1 
m= (−1) R  n̂ R ↑ − n̂ R ↓  H F =  ψk† ↑ τ1 ψk ↑ − ψk† ↓ τ1 ψk ↓  H F
2V 2V
R k∈M B Z
1 
=  ψk† ↑ eiθk τ2 τ1 e−iθk τ2 ψk ↑ − ψk† ↓ e−iθk τ2 τ1 eiθk τ2 ψk ↓  H F
2V
k∈M B Z
1     
=  φk† ↑ cos 2θk τ1 + sin 2θk τ3 φk ↑ − φk† ↓ cos 2θk τ1 − sin 2θk τ3 φk ↓  H F
2V
k∈M B Z
1  Ek
= sin 2θk tanh ,
V 2T
k∈M B Z

which corresponds to the self-consistency equation

U  1 Ek  
m 1− tanh ≡ m K U , m, T = 0 , (3.97)
2V Ek 2T
k

which has a trivial solution m = 0, corresponding to a non-magnetic state, and a

non-trivial one, namely the real root of the term in parenthesis, which we denote as
108 3 Hartree-Fock Approximation

F(m,T)
T  TN T > TN

T < TN

0 m

Fig. 3.3 Sketch of the Hartree-Fock free energy F(m, T ) as function of m at different temperatures.
For T < TN , blue curve, F(m, T ) has two opposite minima and a maximum at m = 0. Increasing T ,
the two minima approach each other till they merge into a single minimum with vanishing curvature
at m = 0 when T = TN . Above TN , the minimum, with finite curvature, remains at m = 0. This is
the conventional behaviour of a second-order phase transition
   
K U , m, T . We note that K U , m, T actually depends on m 2 , so that when a root
exists with m > 0, there is an equivalent one at −m < 0, and they both describe the
desired Néel state. Specifically, m > 0 implies that at sites with R even/odd the spin
is up/down, and the reverse for m < 0. Moreover, one can easily realise that, if that
root exists, it also corresponds to the global minimum of the variational free energy,
i.e.,
  
F ρ H F , T = −2T ln 1 + eβ E k + V U m 2 ,
k
whose saddle point equation with respect to m is exactly (3.97), as expected, while
the saddle point at m = 0 is a maximum. We also note that, if a non-trivial solution
with m = 0 exists at T = 0, since tanh E k /2T decreases with increasing T , and
vanishes for T → ∞, above some temperature one cannot find anymore a root, so
that only the trivial solution m = 0 remains. The situation is depicted in Fig. 3.3 and
resembles what happens, e.g., in an Ising model. Indeed, once within Hartree-Fock
one chooses from the start a symmetry-breaking axis, what remains of the SU (2)
symmetry is the U (1) rotations around that axis, but also the Z 2 symmetry ↑ ↔ ↓
that is broken below TN . However, since the choice of the symmetry-breaking axis
is arbitrary, the actually symmetry that is broken is the original SU (2).
We can gain a rough idea of the non-trivial solution of (3.97) by assuming a
constant density of states in the interval [−D, D] and zero outside, namely,

1  
δ  − k ) ρ0 θ (D − ) θ ( + D) ,
V
k

where ρ0 = 1/2D,with D ∼ 2dt. With that assumption and at zero temperature, the
root of K U , m, T is the solution of

U ρ0 D 1
1= d $ ,
2 −D 2 + U 2m2
3.4 Application: Antiferromagnetism in the Half-Filled Hubbard Model 109

which exists for any positive U and reads

D 1
m=± sinh−1 .
U U ρ0

In particular, for U  D,

D 1
m ± exp − . (3.98)
U U ρ0

This result implies that the Hubbard model at half-filling on a hypercubic lattice
and within the Hartree-Fock approximation is always an antiferromagnetic insulator
whatever is the value of the Hubbard U > 0.4 Note that in the opposite limit of U 
D, m ∼ ±D ρ0 = ±1/2, which is the expected result since, for large U , electrons
localise, one per site, hence behave like local spin-1/2 moments.
We can  also determine
 the Neèl temperature, which is the temperature TN at
which K U , 0, TN = 0, i.e., at which a non-trivial solution of (3.97) exists at m = 0,
namely

U  1 k U ρ0 D d 
1= tanh tanh . (3.99)
2V k 2TN 2 −D  2TN
k
Upon expanding √ (3.97) for T smaller but very close to TN , and m  1, one finds
that m(T ) ∼ ± TN − T , the standard mean-field behaviour. The Néel temperature
defined by the solution of (3.99) is proportional to U m(T = 0), namely

1
TN ∼ D exp − ,
U ρ0

for small U , but TN ∼ U at large U . The latter results contradict what we previously
found in the large U limit by mapping the Hubbard model onto a Heisenberg model.

4 This conclusion actually reflects the nesting property of the non-interacting Fermi surface, i.e.,

the set of points k F : k F = μ, where μ is the chemical potential and is zero in our case. A Fermi
surface has nesting when there exists
 a single
 momentum Q, defined apart from a reciprocal lattice
vector, such that k+Q − μ = − k − μ for a set of momenta W with non-zero measure around
the Fermi surface. In our case, nesting holds throughout the whole Brillouin zone. When the Fermi
surface has nesting, the non-interacting value of the thermodynamic susceptibility κ(Q) to a static
field with momentum Q,
1  f (k+Q − μ) − f (k − μ) 1  1 − 2 f (k − μ)
χ(Q) ∼ − ∼−
V (k+Q − μ) − (k − μ) V −2(k − μ)
k k∈W

diverges logarithmically at T = 0 when the density of states at the Fermi energy is finite. It follows
that the non-interacting model may become unstable to a modulation at momentum Q in presence
of interaction. For instance, within the Hartree-Fock approximation, that occurs when the Hessian
(3.56) becomes negative definite, which is always the case when χ(Q) diverges and the matrix Û
in the channel transferring momentum Q has a negative eigenvalue.
110 3 Hartree-Fock Approximation

There, within spin-wave theory, we showed that TN ∼ J ∼ t 2 /U , hence vanishes for

large U . The origin of the disagreement comes from the fact that the Hartree-Fock
approximation is valid only at weak interaction compared to the electron bandwidth.
Essentially, the Mott phenomenon that dominates at large U /t escapes any descrip-
tion using a single Slater determinant, which is only able to mimic a band insulator.

3.4.1 Spin-Wave Spectrum by Time-Dependent Hartree-Fock

Rigorously speaking, the Hartree-Fock approximation only allows accessing static

equilibrium properties. Therefore, it is not legitimate to interpret the spectrum of the
Hartree-Fock Hamiltonian as an approximation of the actual spectrum. For instance,
we showed that the Heisenberg antiferromagnet, which corresponds to the large U -
limit of the Hubbard model at half-filling, has gapless spin-wave excitations, which
are simply a consequence of Goldstone’s theorem and thus reflect conservation laws.
On the contrary, the Hartree-Fock Hamiltonian describes a band insulator and thus
has a gap for all excitations.
We already mentioned that the time-dependent Hartree-Fock approximation is
supposed not to spoil the Hamiltonian symmetries. Therefore, the half-filled Hubbard
model is the ideal example to check that assumption.
We define
1  †
S(q) ≡ ckα σ αβ ck+q,β ,
2
k
 
the Fourier transform at momentum q of the spin density SR = SxR , S yR , SzR ,
with σ = (σ1 , σ2 , σ3 ), where σi are the Pauli matrices in the spin-up and spin-down
space. At q = 0, Sa (0) are the generators of the spin SU (2) symmetry, thus commute
with the Hamiltonian,
 
Sa (0) , H = 0 , a = x, y, z .

Therefore,
Sa (q, t) ≡ ei H t Sa (q) e−i H t
becomes time independent at q = 0, i.e., Sa (0, t) = Sa (0). It follows that the spin-
spin linear response functions, whose only non-zero components by translation and
spin SU (2) invariance are
 
χa (t, q) = −iθ (t)  Sa (q, t) Sa (−q)  ,

at q = 0,
   
χa (t, 0) = −iθ (t)  Sa (0, t) Sa (0)  = −iθ (t)  Sa (0) Sa (0)  = 0 ,
(3.100)
3.4 Application: Antiferromagnetism in the Half-Filled Hubbard Model 111

must vanish.
Within the Hartree-Fock approximation, we decided to break spin SU (2) symme-
try along the z-direction, but we could have equally well chosen any other direction
and get right the same results. The rotation of the symmetry-breaking axis from the
chosen one, i.e., z, is accomplished by the generators Sx (0) and S y (0), or, equiva-
lently, their rising/lowering combinations
   †
S + (q) = †
ck↑ ck+q↓ , S − (q) = †
ck↓ ck+q↑ = S + (−q) ,
k k

at q = 0. It follows that, e.g., the expectation value of H over the wavefunction

S + (q) | φ H F , where | φ H F  is the Hartree-Fock Slater determinant with symmetry
breaking along z, must vanish for q → 0. In terms of the Fourier transform in time
χ± (ω, q) of the corresponding response function
 
χ± (t, q) = −iθ (t)  S + (q, t) S − (−q)  ,

that implies the existence of a pole in χ± (ω, q) for ω = ω(q) with ω(q → 0) → 0,
which is just the spin-wave dispersion in momentum space. We will therefore check
whether this condition as well as (3.100) are indeed satisfied by the time-dependent
Hartree-Fock approximation.
We define four components spinors
   
k↑ k↑
k = , k = ,
k↓ k↓

which are related to each other through (3.94), namely

k = e−iθk σ3 τ1 k , k† = †k eiθk σ3 τ1 .

For q = |q|  |Q|, the following spin-density operators in terms of those spinors
become
1  † † 1 
S(q) = ckα σ αβ ck+qβ k† σ k+q
†
2 2
kαβ k∈MBZ
1 
= †k eiθk τ2 σ3 σ e−iθk+q τ2 σ3 †k+q ,
2
k∈MBZ
1 † † 1 
S(q + Q) = ckα σ αβ ck+Q+qβ k† σ τ1 k+q
†
2 2
kαβ k∈MBZ
1 
= †k eiθk τ2 σ3 σ τ1 e−iθk+q τ2 σ3 †k+q .
2
k∈MBZ
112 3 Hartree-Fock Approximation

Hereafter, unless necessary, we omit to specify that the momenta are within the MBZ,
and thus all summations over momenta are implicitly assumed to be restricted within
the MBZ.
In the spirit of the time-dependent Hartree-Fock approximation, we need to keep
only particle-hole excitations with respect to the Hartree-Fock state, i.e., terms like
† τ1  and † τ2 . Therefore, for small q,5

i 
Sz (q) − q· ∇ k θk †k τ2 k+q
2
k

and vanishes at q = 0, thus automatically guaranteeing the validity of (3.100) for

a = z.
On the contrary, for a = x, y,

1  † iθk τ2 σ3
Sa (q) = k e σa e−iθk+q τ2 σ3 †k+qβ
2
k
1  †
= k σa e−i(θk+q +θk )τ2 σ3 †k+qβ
2
k
i   
− sin θk+q + θk †k σa σ3 τ2 k+q
2
k
1   
= a3b sin θk+q + θk †k σb τ2 k+q ,
2
k
1  † iθk τ2 σ3
Sa (q + Q) = k e σa τ1 e−iθk+q τ2 σ3 †k+qβ
2
k
1  †
= k σa τ1 e−i(θk+q −θk )τ2 σ3 †k+qβ
2
k
1  
cos θk+q − θk †k σa τ1 k+q ,
2
k

5 Ifwe deal with two sets of Pauli matrices, σi and τi , i = 0, 1, 2, 3, assuming that σ0 and τ0 are
the identity matrices, it is straightforward to demonstrate that, as before,

ei θ τi σ j = cos θ + i sin θ τi σ j ,
 2n  2n+1
since it is still true that τi σ j = 1 and τi σ j = τi σ j . Moreover, also in this case, σk τn is
a unitary Hermitian operator and therefore
 
e−iθ σi τ j σk τn = σk τn σk τn e−iθ σi τ j σk τn = σk τn exp iθ σk τn σi τ j σk τn ,

which can be easily calculated using the properties of the Pauli matrices.
3.4 Application: Antiferromagnetism in the Half-Filled Hubbard Model 113

so that
  
S + (q) = i sin θk+q + θk †k↑ τ2 k+q↓
k
√   
≡i 2 sin θk+q + θk Pk;q ,
k
   (3.101)
+
S (q + Q) = cos θk+q − θk †k↑ τ1 k+q↓
k
√   
≡ 2 cos θk+q − θk X k;q ,
k
where we have introduced the following operators:
% %
1 † † 1 †
X k;q = φ τ1 φk+q ↓ , X k;q = φ τ1 φk ↑ ,
2 k↑ 2 k+q ↓
% %
1 † † 1 †
Pk;q = φ τ2 φk+q ↓ , Pk;q = φ τ2 φk ↑ .
2 k↑ 2 k+q ↓

According to the time-dependent Hartree-Fock approximation, we can assume that

  i  
X k;q , Pk† ;q = δk+q,k +q φk† ↑ τ3 φk ↑ + δk,k φk† +q ↓ τ3 φk+q ↓
2 
 X k;q , Pk† ;q  H F = i δk,k δq,q ,

which shows that X and P † are like conjugate variables. The energy cost of the
spin-flip excitations are described by the free Hamiltonian
  
† †
H0 = ωk;q X k;q X k;q + Pk;q Pk;q , (3.102)
kq

where
ωk;q = E k+q + E k
is simply the energy required to destroy an α particle at k + q and create a β one at
k.
Following the prescriptions of time-dependent Hartree-Fock, we still need to
express the Hubbard interaction in terms of these spin-flip bosonic excitations. Since
† † † † + −
n̂ R ↑ n̂ R ↓ = cR ↑ cR ↑ cR ↓ cR ↓ = −cR ↑ cR ↓ cR ↓ cR ↑ = −SR SR ,

then
  U  +
Hint = U n̂ R ↑ n̂ R ↓ = −U SR+ SR− = − S (q) S − (−q) ,
V q
R R
114 3 Hartree-Fock Approximation

or, considering all momenta restricted within the MBZ,

U  + 
Hint − S (q) S − (−q) + S + (q + Q) S − (−q − Q) .
V q

The dynamical behaviour of S + (q) and S − (−q) = S + (q)† is determined just by

replacing the spin-density operators with the bosonic representation (3.101), which
leads to

2U       †
Hint − sin θk+q + θk sin θk+q + θk Pk;q Pp;q
V q
kp
(3.103)
    †
+ cos θk+q − θk cos θk+q − θk X k;q X p;q .

The full Hamiltonian for the spin-flip excitations is therefore

  
† †
H= ωk;q X k;q X k;q + Pk;q Pk;q
kq
2U       †
− sin θk+q + θk sin θp+q + θk Pk;q Pp;q (3.104)
V q
kp
    †
+ cos θk+q − θk cos θp+q − θk X k;q X p;q .

One may diagonalise this Hamiltonian to obtain the full spin-flip excitation spectrum.
Here, however, we just aim to show that (3.104) does contain the spin waves. For
that, we add a perturbation
 √   
V (t) = V (q, t) S − (−q) −i 2 †
sin θk+q + θk V (t, q) Pk;q ,
q kq

in presence of which the equations of motion read

∂ X k;q   2U   
= ωk;q Pk;q − sin θk+q + θk sin θp+q + θk Pp;q
∂t V p
√  
− i 2 sin θk+q + θk V (t, q) , (3.105)
∂ Pk;q   2U   
= −ωk;q X k;q + cos θk+q − θk cos θp+q − θk X p;q .
∂t V p

In absence of V (t), the expectation values of X and P vanish, while they become
finite when the perturbation is present. Therefore, we take the expectation value
of both right- and left-hand sides of the two equations and write, for simplicity,
3.4 Application: Antiferromagnetism in the Half-Filled Hubbard Model 115

 X k;q  ≡ X k;q , and similarly for P. In that way, we can directly obtain the linear
response function χ± (ω, q) we are interested in.
Taking the Fourier transform in frequency, we obtain
  2U   
−iω X k;q (ω) = ωk;q Pk;q (ω) − sin θk+q + θk sin θp+q + θk Pp;q (ω)
V p
√  
− i 2 sin θk+q + θk V (ω, q) ,
  2U   
−iω Pk;q (ω) = −ωk;q X k;q (ω) + cos θk+q − θk cos θp+q − θk X p;q (ω) .
V p

We define
2U    √
X (ω, q) ≡ cos θp+q − θk X p;q (ω) U 2 S + (ω, q + Q) ,
V p
2U    √
P(ω, q) ≡ sin θp+q + θk Pp;q (ω) −i U 2 S + (ω, q) ,
V p

and the effective field

i
V∗ (ω, q) ≡ V (ω, q) − √ P(ω, q) = V (ω, q) − U S + (ω, q) ,
2
so that
  √  
−iω −ωk;q X k;q (ω) −i 2 sin θk+q + θk V∗ (ω, q)
= ,
ωk;q −iω Pk;q (ω) cos θk+q − θk X (ω, q)

namely
  √  
X k;q (ω) 1 iω −ωk;q −i 2 sin θk+q + θk V∗ (ω, q)
= 2 .
Pk;q (ω) ω − ωk;q
2 ωk;q iω cos θk+q − θk X (ω, q)

If we use the above equation to calculate X (ω, q) and P(ω, q), we finally find,
recalling the definition of V∗ (ω, q),

1 + c (ω, q) (ω, q) X (ω, q) √ (ω, q)

= −i 2 V (ω, q) ,
−(ω, q) 1 + s (ω, q) P(ω, q) s (ω, q)
(3.106)
where we define
2U  sin 2θk
(ω, q) ≡ iω ,
V
k
ω2 − ωk;q
2

2U    ωk;q
c (ω, q) ≡ cos2 θk+q − θk 2 ,
V
k
ω − ωk;q
2

2U    ωk;q
s (ω, q) ≡ sin2 θk+q + θk 2 .
V
k
ω − ωk;q
2
116 3 Hartree-Fock Approximation

We shall analyse the behaviour of the solution for small q. We note, using (3.97) at
zero temperature, that

2U  2E k U  1 ω2
c (ω, 0) = = − 1 +
V
k
ω2 − 4E k2 V
k
Ek ω2 − 4E k2
U  1
= −1 − ω2  2  ≡ −1 − ω2 A(ω) ,
V
k
E k 4E k − ω 2

where A(ω) is positive at small ω. Similarly,

2U  sin 2θk
(ω, 0) ≡ iω = −iω 2U m A(ω) ,
V
k
ω2 − 4E k2
2U  2E k
s (ω, 0) ≡ sin2 2θk 2 = −4U 2 m 2 A(ω) .
V
k
ω − 4E k2

At leading order in the small q, we can write

 
1 + c (ω, q) = 1 + c (ω, 0) + c (ω, q) − c (ω, 0) −ω2 A(ω) + q 2 B(ω) ,

where B(ω) is positive at small ω, and finite at ω = 0, while we can safely take
(ω, q) and s (ω, q) at q = 0. Therefore, (3.106) becomes

−ω2 A(ω) + q 2 B(ω) −iω 2U m A(ω) X (ω, q)

iω 2U m A(ω) 1 − 4U 2 m 2 A(ω) P(ω, q)
√ −iω 2U m A(ω)
= −i 2 V (ω, q) ,
−4U 2 m 2 A(ω)

with solution
√
X (ω, q) = 2 U S + (q + Q)
√ 2m ω A(ω)
=− 2U   V (ω, q) ,
1 − 4U m A(ω) B(ω) q 2 − ω2 A(ω)
2 2

√
P(ω, q) = −i 2 U S + (ω, q)
√ 4U m 2 A(ω) B(ω) q 2
=i 2 U   V (ω, q) ,
1 − 4U 2 m 2 A(ω) B(ω) q 2 − ω2 A(ω)

which implies that

4U m 2 A(ω) B(ω) q 2
χ± (ω, q) = −   . (3.107)
1 − 4U 2 m 2 A(ω) B(ω) q 2 − ω2 A(ω)
3.4 Application: Antiferromagnetism in the Half-Filled Hubbard Model 117

Indeed χ± (ω, q) vanishes at q = 0, consistently

 with (3.100)
 for a = x, y, and, for
small q, has a pole at ω2 = ω2 (q) ≡ 1 − 4U 2 m 2 A(0) B(0) q 2 , an acoustic mode
in agreement with Goldstone’s theorem.
For U  t, E k U m + k2 /2U m, and the root of K (U , m, 0) in (3.97) yields

1 t2
|m| − 2d .
2 U
In that same limit,

t2 t2
1 − 4U 2 m 2 A(0) → 8d , B(0) → 2 , (3.108)
U2 U2
so that
1 J q2
χ± (ω, q) → − ,
2 d J 2 q 2 − ω2
where J = 4t 2 /U is the antiferromagnetic exchange of the Heisenberg model cor-
responding to the large-U limit of the half-filled Hubbard model, and therefore
ω2 (q) → d J 2 q 2 . We recall that within the spin-wave approximation one finds
  
ω2SW (q) = 4S 2 J (q) − J (Q) J (q + Q) − J (Q) ,

where, in a d-dimensional hypercubic lattice with nearest neighbour exchange,

d
J (q) = J i=1 cos qi . For the present case of S = 1/2, one can easily realise that
ω2SW (q → 0) → d J 2 q 2 , remarkably the same result of the time-dependent Hartree-
Fock approximation.
We further note that, if we set ω = 0 and then send q → 0,

4U m 2 A(0)
χ± (0, q → 0) = − ≡ −κ± , (3.109)
1 − 4U 2 m 2 A(0)

where κ± is the thermodynamic susceptibility to a magnetic field perpendicular to

the symmetry-breaking axis. It follows that κ± is positive and finite, contrary to κz
that vanishes for q → 0 irrespective whether we first send ω → 0 and then q → 0
or vice versa. That is a well-known behaviour of antiferromagnets.
In conclusion, we have explicitly demonstrated in the case of a half-filled Hubbard
model that the time-dependent Hartree-Fock approximation yields the linear response
functions consistent with conservation laws.
118 3 Hartree-Fock Approximation

Problems

3.1 BCS mean-field theory of superconductivity—Let us consider again the half-

filled repulsive Hubbard model (3.86), thus U > 0, now in presence of a Zeeman field
that splits spin-up electrons from spin-down electrons. The Hamiltonian is written
as
  †
  1 1
H = −t cR σ cR σ + H .c. + U n̂ R ↑ − n̂ R ↓ −
σ
2 2
<RR > R
 
−μ n̂ R ↑ − n̂ R ↓ .
R
(3.110)
We previously showed that the Hartree-Fock approximation at μ = 0 yields an anti-
ferromagnetic state that breaks spin-SU (2) symmetry and is characterised by a stag-
gered magnetisation that can be oriented in any direction as well as by a finite mag-
netic susceptibility to a Zeeman field perpendicular to the staggered magnetisation.
In the case of the Hamiltonian (3.110) where the Zeeman field is along the z-axis,
the staggered magnetisation must lie within the x − y plane, which corresponds to
an order parameter
  (−1) R
≡ (−1) R cos φ SxR + sin φ S yR = e−iφ  cR↑
† †
cR↓  + eiφ  cR↓ cR↑  ,
2
(3.111)
with arbitrary phase φ reflecting the residual spin-U (1) symmetry at μ = 0. In addi-
tion to the staggered in-plane magnetisation, the Hartree-Fock state has a finite uni-
form magnetisation along z.
Let us consider now the unitary transformation
†
cR↑ → cR↑ , cR↓ → (−1) R cR↓ ,

under which
1 1 1 1
n̂ R ↑ − → n̂ R ↑ − , n̂ R ↓ − → −n̂ R ↓ + ,
2 2 2 2
so that the half-filling condition
   
0= N −V = n̂ R ↑ + n̂ R ↓ − 1 → n̂ R ↑ − n̂ R ↓ ≡ Sz
R R

transforms into the condition of vanishing magnetisation along z, while the Zeeman
term
     
−μ n̂ R ↑ − n̂ R ↓ → −μ n̂ R ↑ + n̂ R ↓ − 1 ≡ μ N − V ,
R R
Problems 119

becomes a chemical potential term, and the Hamiltonian (3.110) turns into
  †
  1 1
H = −t cR σ cR σ + H .c. − U n̂ R ↑ − n̂ R ↓ −
σ
2 2
<RR > R
 
−μ n̂ R ↑ + n̂ R ↓ − 1 ,
R
(3.112)
namely into the attractive Hubbard model in the subspace of zero magnetisation
and with finite chemical potential.6 Since the spectrum is invariant under a unitary
transformation, the Hartree-Fock state transforms into a state characterised by a finite
expectation value

(−1) R
= e−iφ  cR↑
† †
cR↓  + eiφ  cR↓ cR↑  → e−iφ  cR↑
† †
cR↓  + eiφ  cR↓ cR↑  ,
2
(3.114)
which corresponds to an order parameter breaking charge U (1) symmetry. This
state describes therefore a superconductor below a critical temperature Tc which
corresponds to the Neél temperature TN of the repulsive case.

• Show that this conclusion remains true even if further neighbour hopping terms
are added to the nearest neighbour one in (3.112), yielding to a generic dispersion
(k) in momentum space, provided the density of states at the chemical potential

1   
ρ0 ≡ δ (k) − μ
V
k

is finite.

One way to convince oneself that the Hartree-Fock state indeed describes a super-
conductor is to prove that it has Meissner effect, see Sect. 2.6.5. For that,

6 We note that at μ = 0 the Hamiltonian (3.112) is invariant under the particle-hole transformation
† †
cR↑ → (−1) R cR↓ , cR↓ → −(−1) R cR↑ , (3.113)
which leaves the spin operators invariant but transforms N − V → −(N − V ). If the equilibrium
state is also invariant under that transformation, then N − V ≡ −(N − V ), thus N = V , i.e., the
model is at half-filling for μ = 0. On the contrary, if μ = 0, the Hamiltonian is not invariant and
thus N = V . In reality, the Hamiltonian (3.112) at μ = 0 is invariant, besides spin SU (2), also
under a charge SU (2) symmetry with generators
1  1 
Iz ≡ n Rσ − , I+ ≡ †
(−1) R cR↑ cR↓ , I − ≡ I +† ,
2 2
Rσ R

which implement the transformation (3.113). A finite chemical potential term −μ I z lowers that
charge SU (2) symmetry into the common charge U (1).
120 3 Hartree-Fock Approximation

• Apply the time-dependent Hartree-Fock approximation and show that the trans-
verse current-current response function vanishes for ω = 0 and q → 0, corre-
sponding to n = n c in (2.104), thus to a perfect diamagnetic behaviour.

In a lattice model with dispersion (k), the consequence of gauge invariance (2.79)
changes.

• Find that equation calculating

∂ ∂ ∂ ∂    
i i χ (t − t , q) = i i − i θ (t − t )  ρ(t, q) , ρ(−q, t )  ,
∂t ∂t ∂t ∂t
where
 †
ρ(q) = ckσ ck+qσ ,
kσ
recalling that
∂ρ(q)  
i = ρ(q) , H ,
∂t
and taking the Fourier transform in frequency.
• Show that within the time-dependent Hartree-Fock approximation gauge invari-
ance is satisfied, i.e., that equation holds, despite the Hartree-Fock solution breaks
global charge U (1) symmetry. To simplify the calculation, consider the Hamilto-
nian (3.112) at μ = 0 so to use the results of Sect. 3.4.
Feynman Diagram Technique
4

Time-dependent Hartree-Fock is a relatively simple way to access excitation prop-

erties thus to go beyond Hartree-Fock theory. A more systematic scheme, which
includes time-dependent Hartree-Fock as an approximation, is to perform pertur-
bation theory. However, the perturbation expansion in a many-body problem is not
as straightforward as in a single-body case but can be simplified substantially by
means of the so-called Feynman diagram technique. In this chapter we present this
technique at finite temperature in the so-called Matsubara frequency formalism.
The relevant references of all this chapter are the books by Abrikosov, Gorkov, and
Dzyaloshinskii [1] and by Noziéres [2], as well as the seminal work by Luttinger
and Word [3].

4.1 Preliminaries

We first need some preliminary definitions and results, which we are going to present
in this section.

4.1.1 Imaginary-Time Ordered Products

We introduce the imaginary time evolution of an operator A(x) through

A(x, τ ) = e H τ A(x) e−H τ , (4.1)

where H is the fully interacting Hamiltonian and τ ∈ [0, β], where β = 1/T is the
inverse temperature (we use  = 1 and K B = 1).

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 121
M. Fabrizio, A Course in Quantum Many-Body Theory, Graduate Texts in Physics,
https://doi.org/10.1007/978-3-031-16305-0_4
122 4 Feynman Diagram Technique

Given two operators A(x) and B(y), we define their time-ordered product through
 
Tτ A(x, τ ) B(y, τ  ) ≡ −θ (τ − τ  ) A(x, τ ) B(y, τ  ) ∓ θ(τ  − τ ) B(y, τ  ) A(x, τ ) ,
(4.2)
where Tτ denotes the time-ordered product, and the minus sign applies when one
or both the operators are bosonic-like, namely contain bosons or an even product of
fermionic operators, while the plus sign when both operators are fermionic-like, i.e.,
contain an odd product of fermionic operators.
Analogously, given n operators, Ai (xi , τi ), i = 1, . . . , n, we can define their time-
ordered product
 
Tτ A1 (x1 , τ1 ) . . . An (xn , τn ) , (4.3)
as the product in which operators with earlier times appear on the right of those
at later times, and the sign is plus if the number of permutations needed to bring
fermionic-like operators in the correct time-ordered sequence is even, and minus
otherwise.
The imaginary time Green’s function is defined by the expectation value of time-
ordered products of operators. Specifically, for two operators A(x) and B(y), the
Green’s function is
 
G AB (τ − τ  ; x, y) = −Tτ A(x, τ ) B(y, τ  ) 
= −θ (τ − τ  ) A(x, τ ) B(y, τ  ) ∓ θ(τ  − τ ) B(y, τ  ) A(x, τ ) .
(4.4)

We note that, because of time-translation invariance, the Green’s function only

depends on the time difference τ − τ  ∈ [−β, β].
Similarly, the multi-operator Green’s function is defined as
 
G A1 ,...,An (τ1 , . . . , τn ; x1 , . . . , xn ) = −Tτ A1 (x1 , τ1 ) . . . An (xn , τn )  . (4.5)

4.1.2 Matsubara Frequencies

Let us consider the two-operator Green’s function G AB (x, y; τ ), with τ ∈ [−β, β]

being the time difference. Since the time domain is bounded, we can introduce a
discrete Fourier transform, with frequencies ωn = (2π/2β) n = π n T , with n an
integer. We define, dropping for simplicity the spatial dependence,
 β
1
G AB (iωn ) = dτ eiωn τ G AB (τ ),
2 −β

and, consequently,

G AB (τ ) = T e−iωn τ G AB (iωn ).
n
4.1 Preliminaries 123

Since the trace is invariant under cyclic permutations, it follows that

1  
B(0) A(τ ) = Tr e−β H B e H τ A e H − τ
Z
1  
= Tr e−β H eβ H e H τ A e−H τ e−β H B = A(β + τ ) B(0),
Z
as well as
 β
1
G AB (iωn ) = dτ eiωn τ G AB (τ )
2 −β
 β  0
1 1
=− dτ eiωn τ A(τ ) B(0) ∓ dτ eiωn τ B(0) A(τ )
2 0 2 −β
 β  0
1 1
=− dτ eiωn τ A(τ ) B(0) ∓ dτ eiωn τ A(β + τ ) B(0)
2 0 2 −β
 β  β
1 1 −iωn β
=− dτ eiωn τ A(τ ) B(0) ∓ e dτ eiωn τ A(τ ) B(0)
2 0 2 0

1  β
= 1 ± e−iωn β dτ eiωn τ G AB (τ ). (4.6)
2 0

Since exp (−iωn β) = (−1)n then a non-zero Fourier transform requires for bosonic-
like operators even integers n = 2m, and for fermionic-like odd n = 2m + 1. We thus
define the bosonic Matsubara frequencies

m = 2m π T , (4.7)

and the fermionic Matsubara frequencies

m = (2m + 1) π T , (4.8)

so that the bosonic- and fermionic-like Fourier transform of G AB (τ ) become, respec-

tively,
 β  β
G AB (m ) = dτ eim τ G AB (τ ) , G AB (m ) = dτ eim τ G AB (τ ) , (4.9)
0 0

hence just require the knowledge of the Green’s functions for positive τ .
124 4 Feynman Diagram Technique

4.1.2.1 Connection Between Bosonic Green’s Functions and Linear

Response Functions
Let us assume that both A and B are observable quantities, which implies they are
hermitian, hence bosonic-like operators. Then, for positive τ , we have

1  −β E n
G AB (τ ) = −A(τ ) B(0) = − e n | e H τ A e−H τ B | n
Z n
1  −β E n −(E m −E n )τ
=− e e n | A | m m | B | n .
Z nm

Since
 β  
1
dτ eil τ e−(E m −E n )τ =  e−(E m −E n )β − 1 ,
0 il − E m − E n

we finally get

1   −β E n  1
G AB (il ) = e − e−β E m  n | A | m m | B | n .
Z nm il − E m − E n
(4.10)
Comparing this expression with the Fourier transform in frequency of the linear
response function χ AB (ω), see (2.30), we easily realize that the two coincide if we
analytically continue G AB (il ) on the real axis, il → ω + iη. In other words the
knowledge of G AB (il ) means that we know the value of the function G AB (z) of
the complex variable z on an infinite set of points on the imaginary axis zl = il .
This is sufficient to determine G AB (z) in the complex plane given the supplementary
condition that, since G AB (z) = χ AB (z), it has to be analytic in the upper half-plane.

4.1.2.2 Useful Formulas

There is a very useful trick to perform summations over Matsubara frequencies. Let
us start from the fermionic case. Suppose we have to calculate

T F(in ) ,
n

where we assume that F(z) vanishes faster than 1/|z| for |z| → ∞ and has poles
but on the imaginary axis. We note that the Fermi distribution function of a complex
variable z,
1
f (z) = ,
eβ z + 1
has poles at
π
z n = i (2n + 1) = in ,
β
4.1 Preliminaries 125

Fig. 4.1 Integration contour

that run anticlockwise Im(z)
Im(z)
avoiding the imaginary axis

Re(z)
Re(z)

namely right at the Matsubara frequencies, with residue −T . Let us consider the
integral along the contour shown in Fig. 4.1,

dz
I = f (z) F(z) .
2πi
This integral can be calculated through all poles of the integrand in the region enclosed
by the contour that includes the imaginary axis, shaded area in Fig. 4.1. Since this
area is enclosed clockwise, the integral yields
   
I =− Res f (z n ) F(z n ) = T F(in ) ,
zn n

which is just the sum we want to calculate. On the other hand, I can be equally
calculated by catching all poles of the integrand in the area which does not include
the imaginary axis, the non-shaded region, which, being enclosed anti-clockwise
gives
  
I = f (z ∗ ) Res F(z ∗ ) .
z ∗ = poles of F(z)

Therefore
   
T F(in ) = f (z ∗ ) Res F(z ∗ ) . (4.11)
n z ∗ = poles of F(z)

If F(z) has branch cuts the calculation is slightly more complicated since we have
to deform the contour so as to avoid them. In this case, instead of catching poles, we
have to integrate along branch cuts. For instance, suppose that F(z) has a branch cut
along the horizontal axis z = x + iω, with x ∈ [−∞, ∞], as shown in Fig. 4.2. Let
us consider the non-shaded area enclosed inside the contour depicted in Fig. 4.2. In
this area the integrand is analytic, and thus the contour integral vanishes. On the other
hand, this contour integral is also equal to the contribution of the poles inside the
126 4 Feynman Diagram Technique

Fig. 4.2 Integration contour Im(z)

Im(z)
in the presence of a branch
cut

Re(z)
Re(z)

shaded area, which includes the imaginary axis, plus the integral along the contour
that encloses the branch cut. Therefore
   
dx
T F(in ) = − f (x + iω) F(x + iω + i0+ ) − F(x + iω − i0+ ) .
n
2πi
(4.12)
Note that the above equation accounts also for the circumstance in which the branch
cut does not extend over the whole axis z = x + iω, since the term in square brackets
does vanishes only along the branch cut.
When F(z) has both poles, at z = z n , and branch cuts, at z = x + iωm , with
x ∈ [−∞, ∞], the result of the summation over the Matsubara frequencies can be
readily found by the previous two examples, and is
     
T F(in ) = f z n Res F z n
n n
  dx     
− f x + iωm F x + iωm + i0+ − F x + iωm − i0+ .
m
2πi
(4.13)

In the case in which the summation is performed over bosonic frequencies, we can
proceed similarly once we recognise that the poles of the Bose distribution function

1
b(z) = βz
,
e −1

coincide with the bosonic Matsubara frequencies,

π
z n = i 2n ≡ i n ,
β

and have residue T .

4.1 Preliminaries 127

4.1.3 Single-Particle Green’s Functions

Among the expectation values of time-ordered products, an important role in the

diagrammatic technique is played by the so-called single-particle Green’s functions.

4.1.3.1 Fermionic Case

If σ (x, τ ) is the imaginary-time evolution of the Fermi field, then the single-particle
Green’s functions is defined through
 
†
G σ σ  (τ ; x, y) = −Tτ σ (x, τ ) σ  (y) 
(4.14)
= −θ (τ )  σ (x, τ ) σ†  (y)  + θ (−τ )  σ†  (y) σ (x, τ )  .

In principle, we could also define anomalous Green’s functions, as opposed to

the above normal ones, as the expectation values of the time-ordered products
†
σ (x, τ ) σ  (y) or σ (x, τ ) σ  (y). However, those expectation values can be finite
†

only if the U (1) symmetry that derives from the conservation of the number of par-
ticles,

σ (x) → eiφ σ (x) , σ (x)

†
→ e−iφ σ (x) ,
†
(4.15)

is spontaneously broken, which occurs in superconductors, if the fermions are

charged, or in superfluids, if they are neutral. Hereafter, we shall discard such pos-
sibility, and thus assume that only normal single-particle Green’s functions can be
different from zero.
If we perform a spectral representation of G σ σ  (τ ; x, y) in (4.14) and calculate
the Fourier transform in the fermionic Matsubara frequency i, we readily find that

1   −β E n  n | m m | σ†  (y) | n
σ (x) |
G σ σ  (i; x, y) = e + e−β E m  .
Z nm i − E m − E n

Let us take σ = σ  and x = y, which correspond to the so-called local Green’s

function, and introduce the real and positive spectral function

1   −β E n  2
Aσ (, x) = e + e−β E m m | σ (x)
†
| n δ ( − E m + E n ) ,
Z nm

through which, after continuation in the complex plane i → z


1
G σ σ (z; x, x) = d Aσ (, x) .
z−
As function of the complex frequency, G(z) has generally branch cut singularities
along the real axis. Indeed
 
G σ σ z =  + i0+ ; x, x − G σ σ z =  − i0+ ; x, x = −2π i Aσ (, x).
128 4 Feynman Diagram Technique

What is the physical meaning of the spectral function? Let us rewrite Aσ (, x) in the
following equivalent way:

 e−β E n 2
Aσ (, x) = m | σ (x)
†
| n δ ( − E m + E n )
nm
Z
(4.16)
 e−β E n 2
+ m | σ (x) | n δ ( + E m − E n ) ,
nm
Z

which shows that Aσ (, x) is just the probability of adding, first term in the right
hand side, or removing, second term, a particle at position x with spin σ . For that
reason, Aσ (, x) is commonly known as the single-particle local density of states.
Finally we note that
  e−β E n 2 2
d Aσ (, x) = m| σ (x)|n
†
+ m| σ (x)|n
nm
Z
 e−β E n  
= n | σ (x) σ (x)
†
| n + n | σ (x)
†
σ (x) | n
n
Z
 e−β E n  
= n | σ (x), σ (x)
†
| n = 1 ,
n
Z

namely that the integral of the spectral function is normalised to one.

Hereafter, we assume to work in the grand canonical ensemble, so that the Hamil-
tonian H contains a chemical potential term −μ N . In the limit of zero temperature,
T = 0, the sum over | n in (4.16) reduces to just the ground state | 0 (not to be con-
fused with the vacuum), which contains N0 electrons and has energy E 0 . It follows
that (4.16) becomes

 2
Aσ (, x) = m | σ (x)
†
| 0 δ ( − E m + E 0 )
m

2
+ m | σ (x) | 0 δ ( + E m − E 0 ) ,

and, since E m − E 0 is positive by definition, Aσ (, x) at  > 0 is the density of

states for adding one electrons, while at  < 0 for removing one electron. The local
single-particle density of states is, e.g., measurable in tunnelling microscopy, see
Fig. 4.3.
In a periodic system, and considering a metal with a partially filled conduction
band and unbroken spin SU (2), we can define the single-particle Green’s function
through the creation and annihilation operators corresponding to the conduction band
Bloch waves, namely,
 
†
G σ σ  (k, p; τ ) = − Tτ ckσ (τ ) cpσ   = δkp δσ σ  G(τ, k) ,
4.1 Preliminaries 129

Fig. 4.3 Sketch of a tunnelling microscopy experiment. A tip at a certain distance from the sample
surface is kept at a different electrochemical potential with respect to the sample below, the potential
drop being V . If V > 0, electrons from the surface of the sample flow to the tip, and the reverse
if V > 0. The flowing current is proportional to the probability of removing an electron in the
former case, thus to the local single-particle density of states at negative energy  = −e V , while
to the probability of adding an electron in the latter case, i.e., the local single-particle density of
states at positive energy  = −e V > 0

the δ-functions deriving from translational and spin-SU (2) symmetries, where
 
†
G(τ, k) = − Tτ ckσ (τ ) ckσ , ∀ σ = ↑, ↓ . (4.17)

The Fourier transform in fermionic Matsubara frequencies i has the spectral decom-
position
1   −β E n  2 1
G(i, k) = e + e−β E m m | ckσ†
| n , (4.18)
Z nm i − E m + E n

which, upon defining the real and positive density of states for adding and removing
a particle at momentum k, i.e.,
1   −β E n  2 
A(, k) = e + e−β E m m | ckσ †
| n δ  − E m + E n ≥ 0 ,
Z nm
(4.19)
satisfying, as before,

d A(, k) = 1 , (4.20)

can be simply written as the Hilbert transform


A(ω, k)
G(i, k) = dω . (4.21)
i − ω
Therefore

ω
Re G(i, k) = Re G(−i, k) = − dω 2 A(ω, k) ,
 + ω2
 (4.22)
A(ω, k)
Im G(i, k) = −Im G(−i, k) = − dω 2 ,
 + ω2
130 4 Feynman Diagram Technique

which also imply through (4.20) that

  
1 1   
† 
Re G(i, k) −−−−→ − dω ω A(ω, k) = − ckσ , H , ckσ ,
||→∞ 2 2
 (4.23)
1 1
Im G(i, k) −−−−→ − dω A(ω, k) = − .
||→∞  

Moreover, we note that for  > 0,


1 1 
− Im G(i, k) = dω A(ω, k) (4.24)
π π  + ω2
2

is the overlap integral between the density of states A(ω, k) and a Lorentzian centred
at ω = 0 and with width . In an insulator with a hard gap, A(ω, k) = 0 for ω ∈
[−ω1 , ω2 ], where ω1 and ω2 are both positive and their sum is the gap. In this case,
the overlap integral (4.24) is vanishingly small for   min(ω1 , ω2 ) and strictly
zero at  = 0. In a metal or, more generally, in a system with gapless single-particle
excitations, A(ω, k) is finite and smooth around ω = 0, and A(0, k) is different
from zero in a bona fide metal, and zero in a pseudo-gapped anomalous one. Here,
therefore, the overlap integral (4.24) is finite and smooth at small , eventually
vanishing as a power law at  = 0 in the case of a pseudo gap.
We can straightforwardly continue G(k, i) in the complex frequency plane,
i → z ∈ C,

1
G(z, k) = d A(, k) , (4.25)
z−

or impose particular analytic properties. For instance, if we define


1
G ± (z, k) = d A(, k) , (4.26)
z −  ± iη

with η > 0 infinitesimal, then G + (z, k) is analytic in the upper half plane, and thus
is a causal function commonly named retarded Green’s function, while G − (z, k) is
analytic in the lower half plane, an anti-causal function known as advanced Green’s
function. It follows that G(z, k) has generally branch cuts on the real axis, which are
shifted below in G + (z, k) and above in G − (z, k). Specifically, if  ∈ R,

G( + i0+ , k) − G( − i0+ , k) = G + (, k) − G − (, k) = −2πi A(, k).

(4.27)
In other words, since G − (, k) = G + (, k)∗ ,

1
Re G + (, k) = Re G − (, k) , Im G + (, k) = −Im G − (, k) = − A(, k) .
π
(4.28)
4.1 Preliminaries 131

Fig.4.4 Sketch of an ARPES experiment. A photon with momentum q, and thus frequency ω = cq,
hits the sample and provoke the emission of an electron. The detector allows extracting energy and
momentum of the photoemitted electron. From energy and momentum of the photon and the emitted
electron, one can trace back with controlled assumptions to the original energy  and momentum p
of that electron in the system. The intensity of the signal is therefore proportional to the probability
of removing that electron, and thus to the density of states A(, k) times the Fermi distribution
function f ()

Throughout these lecture notes, whenever we write G(, k) with real  we refer to the
retarded component G + (, k), unless otherwise specified. The advantage of dealing
with G + (, k) is that, being causal, its real and imaginary parts are related to each
other by the Kramers-Krœnig equations.
We note that, because of translational symmetry,

1 
A(, x) = A() = A(, k) . (4.29)
V
k

The single-particle density of states A(, k) for removing an electron can be mea-
sured, e.g., by Angle-Resolved Photoemission Spectroscopy (ARPES), as sketched
in Fig. 4.4.
As an example, which will turn useful in what follows, let us consider non-
interacting electrons described by the Hamiltonian
 †
H0 = k ckσ ckσ ,
kσ

where, as discussed previously, k is measured with respect to the chemical potential

μ, which is the Fermi energy at T = 0. In this case, it is easy to show that

1
G 0 (i, k) = , (4.30)
i − k

while the spectral function is simply

A0 (, k) = δ( − k ) . (4.31)

132 4 Feynman Diagram Technique

4.1.3.2 Bosonic Case

If (x) is the Bose field for spinless bosons, the corresponding . single-particle
Green’s function reads
 
G(x, y; τ ) = −Tτ (x, τ ) (y)†  = −θ(τ )(x, τ ) (y)†  − θ(−τ )(y)† (x, τ )  .

Let us again assume that the model is translationally invariant and introduce the
bosonic operators aq and aq† , as well as the conjugate variables
     
1 † 1 †
xq = aq + a−q , pq = −i aq − a−q .
2 2
In most situations, the object that appears in the perturbation theory is not the single-
particle Green’s function but the x − x time-ordered product, namely
 
D(τ, q) = − Tτ xq (τ ) x−q (0)  . (4.32)

The reason why the two operators have opposite momentum is again translational
symmetry, which implies that only momentum-zero operators can have finite expec-
tation value.
As before, let us consider a bosonic non-interacting Hamiltonian

H0 = ωq aq† aq ,
q

with ωq = ω−q ≥ 0. Then, for positive τ ,

1 1 †
D0 (q, τ ) = − aq (τ ) aq† (0) − a−q (τ ) a−q (0)
2 2
1  1
= − e−ωq τ 1 + b(ωq ) − eωq τ b(ωq ) ,
2 2
where
1
b(ωq ) = aq† aq  = βωq
,
e −1
is the Bose distribution function, and
1
aq aq†  = 1 + b(ωq ) = − −βωq
.
e −1
Since
 β 1  ∓βω
dτ eim τ e∓ωq τ = e q −1 ,
0 im ∓ ωq
one readily finds that
 
1 1 1 ωq
D0 (im , q) = − =− . (4.33)
2 im − ωq im + ωq 2m + ωq2
4.2 Perturbation Expansion in Imaginary Time 133

4.2 Perturbation Expansion in Imaginary Time

Let us suppose that the full Hamiltonian

H = H0 + V ,

where H0 is a single particle Hamiltonian that can be exactly diagonalised, while

V is a perturbation that makes H not solvable anymore, e.g., the electron-electron
interaction.

We consider the Heisenberg evolution operator e−H (τ −τ ) from τ  to τ , and write
it as
   
e−H (τ −τ ) = e−H0 τ S(τ, τ  ) e H0 τ ⇒ S(τ, τ  ) = e H0 τ e−H (τ −τ ) e−H0 τ .
(4.34)

Note that, trivially,

S(τ, τ ) = 1 , S(τ, τ1 ) S(τ1 , τ  ) = S(τ, τ  ) , S(τ, τ  ) S(τ  , τ ) = S(τ, τ ) = 1 ,

namely, S(τ  , τ ) = S −1 (τ, τ  ).

We observe that

∂ S(τ, τ  )    
= e H0 τ H0 e−H (τ −τ ) e−H0 τ − e H0 τ H e−H (τ −τ ) e−H0 τ
∂τ
   
= −e H0 τ V e−H (τ −τ ) e−H0 τ = −e H0 τ V e−H0 τ e H0 τ e−H (τ −τ ) e−H0 τ
≡ −V (τ ) S(τ, τ  ) , (4.35)

where V (τ ) is the perturbation evolved in imaginary time with the unperturbed

Hamiltonian. The reason why we introduced S(τ, τ  ) is because, through (4.35), it
admits a very simple perturbative expansion. Suppose therefore that

S(τ, τ  ) = S (n) (τ, τ  ) , (4.36)
n=0

where S (n) (τ, τ  ) contains n powers of V . By definition, the zeroth order term corre-
sponds to H = H0 , and thus, through (4.34), S (0) (τ, τ  ) = 1, ∀ τ, τ  . It follows that
the boundary condition S(τ, τ ) = 1 is already satisfied by the zeroth order term, so
that S (n) (τ, τ ) = 0 for any n > 0. The (4.35) for the nth term is, consistently,

∂ S (n) (τ, τ  )
= −V (τ ) S (n−1) (τ, τ  ) , S (n) (τ, τ ) = 0 ,
∂τ
with solution
 τ
S (n) (τ, τ  ) = − dτ1 V (τ1 ) S (n−1) (τ1 , τ  ) . (4.37)
τ
134 4 Feynman Diagram Technique

Since we know S (0) (τ, τ  ) = 1, we can iteratively find all S (n) (τ, τ  ). At first order,
n = 1,
 τ  τ
S (1) (τ, τ  ) = − dτ1 V (τ1 ) S (0) (τ1 , τ  ) = − dτ1 V (τ1 ) . (4.38)
τ τ

At second order, n = 2,
 τ  τ  τ1
(2)  (1) 
S (τ, τ ) = − dτ1 V (τ1 ) S (τ, τ ) = dτ1 dτ2 V (τ1 ) V (τ2 )
 τ 
τ τ  τ1  ττ  τ2
1 1
= dτ1 dτ2 V (τ1 ) V (τ2 ) + dτ2 dτ1 V (τ2 ) V (τ1 )
2 τ τ 2 τ τ
 τ  
1
= dτ1 dτ2 Tτ V (τ1 ) V (τ2 ) .
2 τ
Iterating the procedure, one finds that
 τ  
(−1)n
S (n) (τ, τ  ) = dτ1 dτ2 . . . dτn Tτ V (τ1 ) V (τ2 ) . . . V (τn ) .
n! τ

It follows that
  
τ
S(τ, τ  ) = Tτ e− τ  dτ1 V (τ1 ) , (4.39)

whose meaning is the following. One has first to expand the exponential, and after
move the time-ordering operator inside the integration.
Suppose we have to calculate the multi-operator Green’s function (4.5), and further
assume that, e.g., τ1 ≥ τ2 ≥ · · · ≥ τn−1 ≥ τn , then

G A1 ,...,An (τ1 , . . . , τn ) = − A1 (τ1 ) . . . An (τn ) 

1
= Tr e−β H eτ1 H A1 e−τ1 H eτ2 H A2 e−τ2 H . . . e−τn−1 H eτn H An e−τn H
Z
1
= Tr e−β H0 S(β, τ1 ) eτ1 H0 A1 e−τ1 H0 S(τ1 , τ2 ) eτ2 H0 A2 e−τ2 H0 . . .
Z

. . . S(τn−1 , τn ) eτn H0 An e−τn H0 S(τn , 0)

1
= Tr e−β H0 S(β, τ1 ) A1 (τ1 ) S(τ1 , τ2 ) A2 (τ2 ) . . . S(τn−1 , τn ) An (τn ) S(τn , 0) ,
Z

where, unlike in the first line, in the last one the operators are evolved with the
unperturbed Hamiltonian H0 , and the expectation value is over the same Hamiltonian
H0 . We readily see that all times remain ordered. Indeed, e.g., S(τ1 , τ2 ) contains all
times ∈ [τ2 , τ1 ] and therefore must be on the right of A1 (τ1 ) but on the left of A2 (τ2 ).
In other words, since

S(β, 0) = S(β, τ1 ) S(τ1 , τ2 ) . . . S(τn , 0) ,

4.2 Perturbation Expansion in Imaginary Time 135

then, for β > τ1 > τ2 · · · > τn > 0,

S(β, τ1 ) A1 (τ1 ) S(τ1 , τ2 ) A2 (τ2 ) . . . S(τn−1 , τn ) An (τn ) S(τn , 0)

 
= Tτ S(β) A1 (τ1 ) . . . An (τn ) ,

so that, for a generic time order,

G A1 ,...,An (τ1 , . . . , τn ) = − Tτ (A1 (τ1 ) . . . An (τn )) 

1  
= Tr e−β H0 Tτ S(β, 0) A1 (τ1 ) . . . An (τn ) ,
Z

where, we emphasise again, the time evolution of the Ai ’s operators in the last
equation is controlled by the non-interacting Hamiltonian. Since, by definition,

e−β H = e−β H0 S(β, 0) ,

then
 
      Tr e−β H0 S(β, 0)
−β H −β H0 −β H0  
Z = Tr e = Tr e S(β, 0) = Tr e = Z 0  S(β, 0)  .
Tr e−β H0

We therefore conclude that

 Tτ (S(β, 0) A1 (τ1 ) . . . An (τn )) 
G A1 ,...,An (τ1 , . . . , τn ) = − , (4.40)
 S(β, 0) 

where the thermal averages as well as the imaginary-time evolution are done with the
non-interacting Hamiltonian H0 . This expression is now suitable for an expansion
in V , which is the reason why we introduced the operator S(β, 0).

4.2.1 Wick’s Theorem

Upon expanding S(β, 0) in powers of the perturbation V , the calculation of any

Green’s function reduces to evaluate the average value of a time-ordered product of
Fermi or Bose fields with a non interacting Hamiltonian. It is therefore essential to
know how to perform this calculation.
Suppose we have to evaluate the expectation value

−Tτ (x1 , τ1 ) (x2 , τ2 ) . . . (xn , τn ) †
(xn , τn ) . . . †
(x1 , τ1 )  , (4.41)

over the non-interacting Hamiltonian H0 . For any time-ordering this amounts to

average a product of creation and annihilation operators. We already know that this
is the sum of the products of all possible contractions of an annihilation with a
136 4 Feynman Diagram Technique

Fig. 4.5 Wick’s theorem applied in two exemplary cases. We represent a contraction as an oriented
line from the annihilation operator, e.g., (2), where 2 stems for space and time coordinates, to
the creation operator, e.g., † (1) to which (2) is contracted. Each contraction now represent a
non-interacting Green’s function, in the case of (2) contracted with † (1) is G 0 (2, 1). The sign
of each term is −1 from the definition of the expectation value, times (−1)n R , where n R is the
number of lines that go from left to right, and, finally, times (−1)n c , where n c is the number of
crossing, the green dot in the bottom panel

creation operator. Suppose that the operator (xi , τi ) is contracted with † (xj , τ j ).
If all other times but τi and τ j remain the same, there are two cases: if τi ≥ τ j we
have to evaluate the contraction
 (xi , τi ) †
(xj , τ j )  ,

while, if τi ≤ τ j we need to interchange the two operators in the time-ordering,

which leads to a minus sign and thus to
− †
(xj , τ j ) (xi , τi )  .

Both cases can be represented by a single quantity, namely,


−G 0 τi − τ j ; xi , xj .

This argument can be extended to all other contractions. As we did when dis-
cussing the Hartree-Fock approximation, it is more convenient to express the general
rules graphically, as in Fig. 4.5 for two simple cases. The rule is very simple.

•> Wick’s theorem

The expectation value (4.41) is the sum of all possible ways to contract pairs of anni-
hilation and creation operators. In this case, the contraction is just a non-interacting
Green’s function. The sign of each product of Green’s functions can be found by
drawing oriented lines from each annihilation operator to the creation operator which
it is contracted to. In that case, the sign is simply −(−1)n R (−1)n c , where n R is the
number of lines that go from left to right, and n c the number of crossings.
4.2 Perturbation Expansion in Imaginary Time 137

This is the just the so-called Wick’s theorem.

In the case of bosons Wick’s theorem does not hold rigorously. Let us consider,
e.g., a periodic system, a non-interacting bosonic Hamiltonian

H0 = ωq bq† bq ,
q

and the corresponding Bose fields

1  iq·r 1  −iq·r †
(r) = √ e bq , † (r) = √ e bq .
V q V q

We note that the expectation value over the non-interacting system

  
 bq†1 bq†2 bq3 bq4  = 1 − δq1 q2 δq1 q4 δq2 q3 + δq1 q3 δq2 q4 n q1 n q2

+ δq1 q2 δq1 q4 δq2 q3 n q1 n q1 − 1
=  bq†1 bq4   bq†2 bq3  +  bq†1 bq3   bq†2 bq4 

− δq1 q2 δq1 q4 δq2 q3 n q1 n q1 + 1 ,

where n q = b(ωq ) is the Bose distribution function, differs from simple pairwise
contractions of creation and annihilation operators because of the last term where
all four operators have the same momentum. That circumstance does not occur for
fermions because of Pauli principle, but it does for bosons, thus invalidating Wick’s
theorem.
Indeed, assuming τ1 > τ2 > τ3 > τ4 ,
 
−  Tτ † (r1 , τ1 ) † (r2 , τ2 )  (r3 , τ3 )  (r4 , τ4 ) 
1 
=− 2
e−i(q1 ·r1 +q2 ·r2 −q3 ·r3 −q4 ·r4 ) eωq1 τ1 +ωq2 τ2 −ωq3 τ3 −ωq4 τ4  bq†1 bq†2 bq3 bq4 
V q1 q2 q3 q4

= −G 0 (τ4 − τ1 , r4 − r1 ) G 0 (τ3 − τ2 , r3 − r2 )
− G 0 (τ4 − τ2 , r4 − r2 ) G 0 (τ3 − τ1 , r3 − r1 )
1  −iq·(r1 +r2 −r3 −r4 ) ωq (τ1 +τ2 −τ3 −τ4 ) 
− 2 e e nq nq + 1 , (4.42)
V q

where
 
G 0 (τ − τ  , r − r ) = − Tτ (r, τ ) † (r , τ  ) 
1  iq·(r−r )  
=− e  Tτ bq (τ ) bq† (τ  )  .
V q
138 4 Feynman Diagram Technique

The two products of Green’s functions in (4.42) are those expected from Wick’s
theorem, while the last term in that equation is the deviation from that theorem.
However, in the thermodynamic limit, V → ∞, that deviation vanishes since it is
of order 1/V . It follows that Wick’s theorem does hold also for bosons in periodic
systems and in the thermodynamic limit. In that case, the rule is the same as for
fermions, with the only difference that the sign of each product of n Green’s functions
is simply (−1)n+1 .
There is however a caveat. In some cases, Bose condensation occurs at a given q,
usually q = 0, as we assume. In the condensed phase, n 0 =  b0† b0  = ρc V , where
ρc is the finite density of the condensate, thus yielding a deviation from Wick’s
theorem that remains finite also in the thermodynamic limit. Nonetheless, a way
out still exists. Since there is a macroscopic occupation at q = 0, one can safely
√
replace the operator b0 by a c-number, b0 ∼ eiφc V ρc , and similarly its hermitian
conjugate, so that, e.g.,

√ 1  iq·r √
(r)  eiφc ρc + √ e bq ≡ eiφc ρc +  (r) .
V q=0

Under that assumption, Wick’s theorem does work for  (r) and its hermitian con-
jugate in the thermodynamic limit.

4.3 Perturbation Theory for the Single-Particle Green’s

Function and Feynman Diagrams

Let us consider the Hamiltonian for free electrons

 †
H0 = k ckσ ckσ , (4.43)
kσ

with spin-independent non-interacting Green’s function

 
G 0 (τ − τ  , r − r ) = − Tτ σ (r, τ ) σ† (r , τ  )  , (4.44)

in presence of the interaction


1
Hint = dr dr σ (r)
† † 
σ  (r ) U (r − r ) σ  (r

) σ (r) ,
2 
σσ

which plays the role of the perturbation V . The operator S(β, 0) is built through
  
1 
dτ Hint (τ ) = dr dr dτ σ (r, τ )
† † 
σ  (r , τ ) U (r − r ) σ  (r

, τ) σ (r, τ )
2
σσ
 
1 
= dr dr dτ dτ  σ† (r, τ ) †  
σ  (r , τ ) U (r − r ) δ(τ − τ  ) σ  (r

, τ ) σ (r, τ )
2 
σσ
4.3 Perturbation Theory for the Single-Particle Green’s Function … 139

1 †
≡ dx dy σ (x)
†
σ  (y) U (x − y) σ  (y) σ (x) ,
2 
σσ

where we have introduced the space-time coordinates x = (r, τ ), y = (r , τ  ) and,

by definition,
U (x − y) = U (r − r ) δ(τ − τ  ) ,
so that
   
1 †
S(β, 0) = Tτ exp − d x1 d x2 σ (x 1 )
†
σ  (x 2 ) U (x 1 − x2 ) σ (x2 ) σ (x 1 ) .
2 
σσ

The single-particle Green’s function can be calculated perturbatively through

1  
G(x, y) = −  Tτ S(β, 0) σ (x) σ (y)
†
, (4.45)
 S(β, 0) 

where the operators evolve with H0 , the expectation value is over the non-interacting
states, and we assume unbroken spin SU (2) symmetry, implying that the Green’s
function is diagonal in spin and independent of it.
Let us perform the calculation up to first order in perturbation theory, namely
approximating

S(β, 0)  S (0) (β, 0) + S (1) (β, 0)


1 †
=1− d x1 d x2 σ (x 1 )
†
σ  (x 2 ) U (x 1 − x2 ) σ  (x 2 ) σ (x 1 ) .
2 
σσ

We need to calculate up to first order both numerator and denominator in (4.45).

About the numerator, by Wick’s theorem we find
 
− Tτ S(β) σ (x) σ† (y)  = G 0 (x, y)
  
1 † † †
+ d x1 d x2 U (x1 − x2 ) Tτ σ (x) σ (y) α (x 1 ) β (x 2 ) β (x 2 ) α (x 1 ) 
2
αβ
 
= G 0 (x, y) 1 +  S (1) (β, 0) 

1
+ d x1 d x2 U (x1 − x2 ) δσ α G 0 (x, x1 ) G 0 (x1 , y) G 0 (x2 , x2 )
2
αβ

− δσ α δαβ G 0 (x, x1 ) G 0 (x1 , x2 ) G 0 (x2 , y)


1
+ d x1 d x2 U (x1 − x2 ) δσβ G 0 (x, x2 ) G 0 (x2 , y) G 0 (x1 , x1 )
2
αβ

− δσβ δβα G 0 (x, x2 ) G 0 (x2 , x1 ) G 0 (x1 , y) .

140 4 Feynman Diagram Technique

The term G 0 (x, y)  S (1) (β, 0)  is dubbed disconnected, which implies that the con-
tractions factorise in separate terms that do not have common variables. We note that
the last two integrals are actually equal since we can exchange x1 ↔ x2 , so that
   
−  Tτ S(β, 0) σ (x) σ† (y)  = G 0 (x, y) 1 +  S (1) (β, 0) 

+ d x1 d x2 U (x1 − x2 ) 2 G 0 (x, x1 ) G 0 (x1 , y) G 0 (x2 , x2 ) − G 0 (x, x1 ) G 0 (x1 , x2 ) G 0 (x2 , y) ,

where the factor two comes from the summation over the free internal spin. The
denominator is simply 1 +  S (1) (β, 0)  so that, up to first order, it cancels the dis-
connected term from the numerator and the final result reads

G(x, y) = G 0 (x, y) + d x1 d x2 U (x1 − x2 ) 2 G 0 (x, x1 ) G 0 (x1 , y) G 0 (x2 , x2 )

− G 0 (x, x1 ) G 0 (x1 , x2 ) G 0 (x2 , y) .

(4.46)

Let us give a graphical representation of this result. We represent the fully inter-
acting Green’s function as a bold line directed from the coordinate of to that of
† , the non-interacting one as a regular line directed as before, and the interaction

as a wavy line with four legs, see Fig. 4.6. An incoming vertex represent a creation
operator, while an outgoing one an annihilation operator. With these notations the
Green’s function up to first order in the interaction can be represented as in Fig. 4.7.
The conventions are that any internal coordinate is integrated, the spin is conserved
along a Green’s function as well as at any interaction-vertex, and a wavy line is
the interaction U . There are two first order diagrams. The tadpole one has a plus
sign and a factor two, while the other diagram a minus sign. We note that both dia-
grams are fully connected; the disconnected terms cancel out with the denominator
S(β, 0). One can go on and calculate the second order corrections to infer the rules
for constructing diagrams. Here we just quote the final answer.

Fig. 4.6 Graphical representation of the Green’s functions and the interaction

Fig. 4.7 Graphical representation of the Green’s function up to first order. Note that sign and the
loop multiplicity 2 L are implicit in the diagram definition, and thus are not indicated explicitly
4.3 Perturbation Theory for the Single-Particle Green’s Function … 141

Fig. 4.8 Graphical representation of all second order corrections to the Green’s function

The nth order diagrams for the single particle Green’s functions are all fully
connected and topologically inequivalent diagrams which can be constructed
with n interaction lines and 2n + 1 non-interacting Green’s functions, with
external points at x and y. All internal coordinates are integrated. The sign of
each diagram is (−1)n (−1) L , where L is the number of internal loops. Spin is
conserved at each vertex, hence each loop implies a spin sum, thus a prefactor
2 L . Finally, if a Green’s function is connected by the same interaction line it
has to be interpreted as the limit of τ → 0− , since, in the interaction, creation
operators are on the left of the annihilation ones.

One easily realises that the following rules reproduce the two first order corrections
we just derived.
In Fig. 4.8 we draw all topologically inequivalent diagrams at second order in
perturbation theory. Let us follow the above rules to find the expression of the last
two, (i) and (l). The former has no loops, hence its sign is simply (−1)n = (−1)2 = 1.
The spin is the same for all lines hence its expression is

 
4
(h) = d xi U (x1 − x3 ) U (x2 − x4 )
i=1
G 0 (x, x1 ) G 0 (x1 , x2 ) G 0 (x2 , x3 ) G 0 (x3 , x4 ) G 0 (x4 , y) .
142 4 Feynman Diagram Technique

Diagram (l) has a loop, hence (−1)n (−1) L = (−1)2 (−1)1 = −1. The internal loop
implies a spin summation, hence

 
4
(i) = −2 d xi U (x1 − x3 ) U (x2 − x4 )
i=1
G 0 (x, x1 ) G 0 (x1 , x2 ) G 0 (x2 , y) G 0 (x3 , x4 ) G 0 (x4 , x3 ) .

4.3.1 Diagram Technique in Momentum and Frequency Space

Since there is time-translation invariance, and the Hamiltonian is also space-

translation invariant, it is more convenient to consider the Fourier transform of the
Green’s function G(in , k), and derive its perturbative expansion. We further need to
introduce the Fourier transform of the interaction line as well as we need to indicate
a direction to this line, which is the direction along which momentum and frequency
flow, see Fig. 4.9. Notice that the frequency carried by the interaction is the difference
between two fermionic Matsubara frequencies, hence it is a bosonic one, namely an
even multiple of π T . Moreover, since U (x − y) = U (r − r ) δ(τ − τ  ), the Fourier
transform is
 
U (iω, q) = dτ dr e−iq·r eiωτ δ(τ ) U (r) = U (q) ,

independent of frequency. Conversely

 1 
U (x) ≡ U (r) δ(τ ) = T eiq·r e−iωτ U (iω, q) , (4.47)
ω
V q

and, seemingly,
 1   
G 0 (x − y) = T eik·(r−r ) e−i(τ −τ ) G 0 (i, k) . (4.48)

V
k

Fig. 4.9 Graphical

representation of the
interaction in Fourier space,
where  and   are fermionic
Matsubara frequencies,
while ω a bosonic one
4.3 Perturbation Theory for the Single-Particle Green’s Function … 143

Moving to the perturbation expansion, since the spatial and time coordinates of each
internal vertex are integrated out, momentum and frequency are conserved at each
vertex, namely the sum of the incoming values is equal to that of the outgoing ones.1
Therefore the rules are:

The nth order diagrams for the single particle Green’s functions with momen-
tum k and frequency i are all fully connected and topologically not equivalent
diagrams which can be constructed with n interaction lines and 2n + 1 non-
interacting Green’s functions, with two external lines G 0 (i, k) and n internal
frequencies and momenta. At each vertex, the sum of momenta and that of
frequencies of the incoming lines must be equal to the those of outgoing lines.
All internal momenta ki and frequencies i , i = 1, . . . , n, are summed, i.e.,


n
1  
T ... .
V 
i=1 ki i

The sign of each diagram is (−1)n (−1) L , where L is the number of internal
loops. Spin is conserved at each vertex, hence each loop implies a factor 2.
Finally, if a Green’s function is connected by the same interaction line it has
to be interpreted as the inverse Fourier transform at τ = 0− , i.e.
 −  +
T G 0 (i, k) e−i 0 = T G 0 (i, k) ei 0 .
 

1 Before integration over the internal coordinates, an nth diagram Dn (x, y, x1 , y1 , . . . , xn , yn ) has
2n internal coordinates, xi = (τi , ri ) and yi = (τi , ri ), i = 1, . . . , n, for each interaction line, and
two external ones, x = (τ, r) and y = (0, 0), where we exploit space and time translation invariance.
Therefore, it is composed by 2n + 1 Green’s functions G 0 , and n interactions U (xi − yi ). The
Fourier transform of Dn (x, y, x1 , y1 , . . . , xn , yn ) is
  
n
D(i, k) = d(x − y) eiτ e−iq·r d xi dyi Dn (x, y, x1 , y1 , . . . , xn , yn ) . (4.49)
i=1

If each of the n + 1 Green’s functions is written as in (4.48), and each of the interaction as in (4.47),
we have 2n + 1 space-time integrations and (2n + 1) + n frequency-momentum summations. It
follows that just n internal frequency-momentum summations remains, the space-time integration
yielding frequency-momentum conservation at each vertex.
144 4 Feynman Diagram Technique

Fig. 4.10 Diagrams for the single-particle Green’s function up to first order in Fourier space

For instance, the diagrams up to first order are those in Fig. 4.10, and their expres-
sion reads
 1  +
G(in , k) = G 0 (in , k) + 2 G 0 (in , k)2 U (0) T G 0 (im , p) eim 0
m
V p
 1  +
− G 0 (in , k)2 T U (k − p) G 0 (im , p) eim 0 .
m
V p

Using the contour in Fig. 4.1, we can write

 +  + 1
T G 0 (im , p) eim 0 = T eim 0
m m
im − p
dz + 1 
= f (z) ez 0 = f p ,
2πi im − p

+
where the exponential ez 0 guarantees a decay faster than 1/z for |z| → ∞, and thus
allows using Cauchy’s theorem. Therefore, the diagrams in Fig. 4.10 reads explicitly

1   
G(in , k) = G 0 (in , k) + 2 G 20 (in , k) 2 U (0) − U (k − p) f p .
V p

4.3.2 The Dyson Equation

The graphical representation of the single-particle Green’s function perturbative

expansion already represents a substantial simplification. However, the number of
diagrams grow exponentially with the order n, see Table 4.1. Therefore, if the only
way to proceed were to calculate order by order all diagrams, the Feynman diagram
technique would be of little practical use. In reality, the technique is extremely useful
since it gives hints how to sum up the perturbative series. Let us discuss how that
works for the single-particle Green’s function.
Among all possible diagrams for the single particle Green’s function, we can
distinguish two classes. The first includes all diagrams that, by cutting an inter-
nal Green’s function line, transform in two lower-order diagrams for the Green’s
function. These kind of diagrams are called single-particle reducible. For instance
diagrams (a), (b), (c) and (d) in Fig. 4.8 are single-particle reducible. The other class
4.3 Perturbation Theory for the Single-Particle Green’s Function … 145

Table 4.1 Number of n # of diagrams

topologically inequivalent
diagrams up to order n = 7 1 2
2 10
3 74
4 706
5 8162
6 110410
7 1708391

contains all diagrams which are not single-particle reducible, also called single-
particle irreducible. For instance both first order diagrams as well as the second
order ones (e) to (j) are irreducible. Let us define as the single-particle self-energy
σ (i, k) the sum of all irreducible diagrams without the external legs. For instance,
the self-energy diagrams up to second order are drawn in Fig. 4.11, where the self-
energy is represented by a rounded box. It is not difficult to realise that, in terms of
the self-energy, the perturbation expansion can be rewritten as in Fig. 4.12, which
has the formal solution
G 0 (in , k) 1
G(in , k) = = . (4.50)
1 − G 0 (in , k) (in , k) in − k − (i, k)

This is the so-called Dyson equation for the single-particle Green’s function. We
note that, because of (4.22)

Re (i, k) = Re (−i, k) , Im (i, k) = −Im (−i, k) . (4.51)

Fig. 4.11 Diagrams for the single-particle self-energy up to second order

146 4 Feynman Diagram Technique

Fig. 4.12 Dyson equation for the single-particle Green’s function

Moreover, through (4.23), and, since

   † 1  
†
ckσ , H , ckσ = k + U (0) − δσ σ  U (k − p)  cpσ  cpσ   ,
V  pσ

then

Im (i, k) −−−−→ 0 ,
||→∞
1  
†
Re (i, k) −−−−→ U (0) − δσ σ  U (k − p)  cpσ  cpσ   ≡ HF (k) ,
||→∞ V pσ
(4.52)

which defines HF (k).

There are therefore two possible ways of doing perturbation theory. The simplest
is just to calculate directly the Green’s function, e.g., up to order n. The second is to
calculate up to order n the self-energy, and insert its expression in the Dyson equation
(4.50). This provides an approximate Green’s function that contains all orders in
perturbation theory. Therefore, (4.50) represents a way to sum up perturbation theory,
and it is in reality physically more appropriate. Indeed, we showed that the Green’s
function analytically continued in the complex plane has a branch cut on the real axis
that is just the single-particle density-of-states, and which is strongly contributed by
the interaction, since the non-interacting Green’s function has a simple pole on the
real axis. It is easy to convince oneself that calculating the Green’s function up to
any order does not give access to that branch cut, unlike calculating the self-energy
at that same order.
Similarly to the Green’s function, we can continue the self-energy (i, k) to the
complex plane, (z, k) with z ∈ C. The straight continuation has generally a branch
cut on the real axis because G(z, k) has it. Otherwise, as in (4.26) we can define
retarded, + (z, k), and advanced, − (z, k), components analytic in the upper and
lower half-plane, respectively. As for the Green’s function,(, k) with  ∈ R refers
to the retarded component, i.e., (, k) ≡ + (, k) =  z →  + iη, k with η >
0 infinitesimal. Since that is causal, we can write

dω Im (ω, k)
(, k) = − , (4.53)
π  − ω + iη
4.3 Perturbation Theory for the Single-Particle Green’s Function … 147

Fig. 4.13 Skeleton diagrams

for the single-particle
self-energy up to second
order. Differently from
Fig. 4.11, now all internal
Green’s functions are the
fully interacting ones

so that

dω Im (ω, k)
Re (, k) = − − . (4.54)
π −ω

4.3.3 Skeleton Diagrams

The self-energy perturbative series can be recast in a simpler form. Indeed, some
diagrams in the series are actually self-energy corrections of the internal Green’s
functions, like the first four second order diagrams in Fig. 4.11. Therefore, we can
discard all such diagrams simply regarding the internal Green’s functions as fully-
interacting ones. The rules for constructing diagrams remain the same: each interac-
tion yields a factor −1, each loop (−2), and equal-time Green’s functions are to be
evaluated at τ = 0− . However, in the skeleton expansion the order of a diagram only
refers to the number of interaction lines explicitly drawn, ignoring that the internal
Green’s functions includes all orders in interactions. The skeleton diagram of the
self-energy up to second order are shown in Fig. 4.13.
The skeleton expansion implies that the self-energy can be regarded as a functional
of the interacting
 Green’s
 function and of the interaction, that we shortly denote as
U , thus  =  G, U . If follows that the Dyson equation (4.50) can be also written
as the self-consistency equation

1
G=   . (4.55)
G −1
0 −  G, U
 
Therefore, should we know  G, U , we would be able to calculate G solving(4.55).
Moreover, as we will discuss in the following sections, the knowledge of  G, U
also allows calculating linear response function, thus providing the complete physical
characterisation of the system, essentially its ‘exact’ solution.
However, the explicit expression of the self-energy functional  is unknown.
Nonetheless, one can assume an approximate expression, appr ox G, U , and with
148 4 Feynman Diagram Technique

that calculate the Green’s function solving the self-consistency equation (4.55) and
all linear response functions. If that approximation captures the main physical ingre-
dients, the outcomes provide sensible description of the system. For instance, if one
assumes a self-energy functional that includes just the two first order diagrams in
Fig. 4.13, that is equivalent to the Hartree-Fock approximation, as we shall explicitly
show later. Explicitly, those first order diagrams, which involve equal-time Green’s
functions, read explicitly
1   1  
U (0) − δσ σ  U (k − p) G(τ = 0− , k) = †
U (0) − δσ σ  U (k − p)  cpσ  cpσ  
V 
V 
pσ pσ
= HF (k) ,

where HF (k) was earlier defined in (4.52), and explain the reason of the subscript
HF. In particular, the tadpole diagram is known as Hartree term, while the other as
Fock one.

4.3.4 Physical Meaning of the Self-energy

According to (4.27) and (4.28), and with our convention G(, k) ≡ G + (, k) and
(, k) ≡ + (, k), the single-particle density of states (DOS) is

1 1 η − Im (, k)
A(, k) = − Im G(, k) =  2  2 .
π π
 − k − Re (, k) + η − Im (, k)

Since A(, k) > 0, and η is infinitesimal, it follows that Im (, k) < 0. For non-
interacting electrons (, k) = 0 and the spectral function becomes δ( − k ) for
η → 0+ . When the interaction is finite, and thus we can safely set η = 0,

1 −Im (, k)
A(, k) =  2  2 . (4.56)
π
 − k − Re (, k) + Im (, k)

Equation (4.56) resembles a set of Lorentzian functions, each centred at  = ∗ (k),

where ∗ (k) are the solutions of the equation

∗ (k) = k + Re  ∗ (k), k ,

and with width −Im (, k) calculated at  = ∗ (k). Indeed, let us assume  close
to a root ∗ (k), so that
 
 − k − Re (, k)  Z ∗ (k), k)−1  − ∗ (k) ,

where
∂Re (, k)
Z (, k)−1 ≡ 1 − , (4.57)
∂
4.3 Perturbation Theory for the Single-Particle Green’s Function … 149

Fig. 4.14 A second-order

diagram contributing to the
imaginary part of the
self-energy

will be hereafter denoted as quasiparticle residue for reasons that will be clarified
later. It follows that
 
Z ∗ (k), k) ∗ k
A( ∼ ∗ (k), k) =   ,
π 2
 − ∗ (k) +  2 k ∗
  
indeed a Lorentzian, with weight Z ∗ (k), k) and width  ∗ k = −Z ∗ (k), k)
Im  ∗ (k), k .
Therefore, a weak interaction at first shifts the position of the δ-peak of the non-
interacting DOS, k → ∗ (k), and yields a finite broadening. The latter reflects the
fact that a particle (hole), i.e., an electron outside (inside) the Fermi surface and
thus with k > 0 (k < 0), can decay because of interaction into a particle (hole)
plus several particle-hole pairs provided momentum is conserved. That process is
accounted for by Im  , k that can be regarded as the decay rate of a particle
with momentum k into a composite multi-particle object with same momentum and
energy ; an interpretation which can be justified by analysing the structure of the
self-energy diagrams that contribute to the imaginary part.
Let us consider, for instance, the second-order diagram drawn in Fig. 4.14, Assum-
ing a short range interaction, U (q)  U , ∀ q, its expression is

1  
(i, k) = −2 U 2 T 2
δk+p1 ,k0 +k1
V2  ω
k0 k1 p1 1 1

1 1 1
i1 − k1 iω1 − p1 i + iω1 − i1 − k0

where the δ-function imposes the momentum conservation and the factor 2 comes
from the loop spin summation. Let us first sum over i1 , which yields
 
 1 1 f k 1 f i + iω1 − k0
T = −
1
i1 − k1 i + iω1 − i1 − k0 i + iω1 − k1 − k0 i + iω1 − k0 − k1
    1
= f k1 − f − k0
i + iω1 − k0 − k1
    1
= − 1 − f  k 1 − f k 0 ,
i + iω1 − k0 − k1
150 4 Feynman Diagram Technique

where we used the fact the i + iω1 is a bosonic frequency, and thus eiβ(+ω1 ) = 1,
and that f (−x) = 1 − f (x), where f (x) is the Fermi distribution function.
Now let us sum over iω1 ,
 
 1 1 f p 1 f k0 + k1 − i
T = +
ω1
iω1 − p1 i + iω1 − k0 − k1 i + p1 − k0 − k1 k0 + k1 − i − p1
    1
= f p1 + b k0 + k1 .
i + p1 − k0 − k1

Here we used the fact that, since i is fermionic, then

 −1  −1
 
f k0 + k1 − i = eβ(k0 +k1 −i) + 1 = − eβ(k0 +k1 ) + 1 = −b k0 + k1 .

Therefore,

2U 2  1
(i, k) = δk+p1 ,k0 +k1
V 2 i + p1 − k0 − k1
k0 k1 p1
      
1 − f k1 − f k0 f p1 + b k0 + k1 .

We note that, since,

1 − f (x)
eβx = ,
f (x)

then

1 f (x) f (y) f (x) f (y)

b(x + y) = =   = ,
eβx eβ y − 1 1 − f (x) 1 − f (y) − f (x) f (y) 1 − f (x) − f (y)

and thus
 
   f k0 f k1
f p1 + b k0 + k1 = f p1 +  
1 − f k0 − f k1
   
 f k0 f k1  
= f p1 +   1 − f p1 + f p1
1 − f k0 − f k1
1      
=   f p1 1 − f k0 1 − f k1
1 − f k0 − f k1
    
+ 1 − f p1 f k0 f k1 ,
4.3 Perturbation Theory for the Single-Particle Green’s Function … 151

implying that

2U 2  1
(i, k) = δk+p1 ,k0 +k1
2
V k k p i + p1 − k0 − k1
0 1 1
          
f p1 1 − f k0 1 − f k1 + 1 − f p1 f k0 f k1 .
(4.58)
Now, we send i →  + i0+ and just consider the imaginary part,
 
1 
Im = −π δ  + p1 − k0 − k1 ,
 + i0+ + p1 − k0 − k1

so that

U2  
Im (, k) = −2π 2
δk+p1 ,k0 +k1 δ  + p1 − k0 − k1
V k0 k1 p1
          
f p1 1 − f k0 1 − f k1 + 1 − f p1 f k0 f k1 ,

which is correctly negative, and just represents the Fermi golden rule describing
the decay of an electron at momentum k and energy   into either
 a hole at p1
and two particles at k0 and k1 , the term f (p1 ) 1 − f (k0 ) 1 − f (k1 ) , or a
particle at p1 and two holes at k0 and k1 , the term 1 − f (p1 ) f (k0 ) f (k1 ). This
process requires momentum conservation, i.e., k = k0 + k1 − p1 , as well as energy
conservation,  = k0 + k1 − p1 . Upon defining the dimensionless function

1 1    
Sk (0 , 1 , ω1 ) ≡ δ 0 − k0 δ 1 − k1 δ ω1 − p1 δk+p1 ,k0 +k1 ,
ρ03 V 2
k0 k1 p1
(4.59)

where
1  
ρ0 = δ k ,
V
k

is the non-interacting single-particle density of states at the chemical potential, we

can write


Im (, k) = −2π U 2 ρ03 d0 d1 dω1 Sk (0 , 1 , ω1 ) δ  + ω1 − 1 − 0
          
f ω1 1 − f 0 1 − f 1 + 1 − f ω1 f 0 f 1 .
(4.60)
152 4 Feynman Diagram Technique

We are interested in (4.60) evaluated at small  and temperature T . Let us first

take T = 0, thus f (x) = θ (−x), in which case  > 0 implies 0 > 0, 1 > 0 and
ω1 < 0 (two particles and one hole), while  < 0 the opposite, i.e., 0 < 0, 1 < 0
and ω1 > 0 (two holes and one particle). For   0, the energy conservation
 = 0 + 1 − ω1 = 0 + 1 + |ω1 | ,
implies that all variables 0 , 1 and −ω1 are not only positive but also smaller than
, so that, after changing ω1 → −ω1 ,
 

Im (, k) = −2π U 2 ρ03 d0 d1 dω1 Sk (0 , 1 , −ω1 ) δ  − ω1 − 1 − 0 .
0

Since the integration range already vanishes as  → 0, at leading order we can make
the approximation Sk (0 , 1 , −ω1 )  Sk (0, 0, 0), in which case
 

Im (, k)  −2π U 2 ρ03 Sk (0, 0, 0) d0 d1 dω1 δ  − ω1 − 1 − 0
0
2
= −2π U 2 ρ03 Sk (0, 0, 0) ,
2
(4.61)
vanishes quadratically as  → 0. Equation (4.62) holds provided Sk (0, 0, 0) is not
singular, which we assume for now and later discuss its validity.
Accordingly, if we are interested in the leading term in T , we can safely set  = 0
in (4.60) and still approximate Sk (0 , 1 , ω1 )  Sk (0, 0, 0), so that, at T  0, and
recalling that 1 − f (x) = f (−x)
 ∞

Im (0, k)  −2π U ρ0 Sk (0, 0, 0)
2 3
d0 d1 dω1 δ ω1 − 1 − 0
−∞
     
f ω1 f − 0 f − 1 + f − ω1 f 0 f 1
 ∞
  
= −4π U ρ0 Sk (0, 0, 0)
2 3
d0 d1 f − 0 − 1 f 0 f 1
−∞
π2 2
= −2π U 2 ρ03 Sk (0, 0, 0) T ,
2
vanishes quadratically as T → 0.2 In conclusion, for small  and T , and provided
Sk (0, 0, 0) in (4.59) is not singular,
 
Im (0, k)  −π U 2 ρ03 Sk (0, 0, 0)  2 + π 2 T 2 , (4.62)

vanishes for both  → 0 and T → 0.

2 If we define
1
F(x) = ,
ex + 1
4.3 Perturbation Theory for the Single-Particle Green’s Function … 153

Fig. 4.15 Graphical

representation of a generic
self-energy diagram that
contributes to the imaginary
part

More generally, the diagrams which contribute to the imaginary part of the self-
energy can be represented as in Fig. 4.15, where a particle with momentum k and
Matsubara frequency i decays at the left box, which represents a generic matrix
element, into a particle/hole, the line (i0 , k0 ), plus n ≥ 1 particle-hole pairs, the
pair of lines (ii , ki ) − (iωi , pi ), which recombine at the right box into the original
particle/hole. Momentum and frequency conservation imply


n
 
n

 = 0 + i − ωi , k = k0 + ki − pi .
i=1 i=1

As before, if we sum over all the independent internal frequencies by a contour

integral, and assume that the leading terms are the poles of the Green’s function,
thus neglecting the frequency dependence of the matrix element, and finally send
i →  + i0+ , the imaginary part at T = 0 is finite if, at  > 0, ki , i = 0, . . . , n,
are all particles, while p j , j = 1, . . . , n, all holes, i.e., ki > 0 and p j < 0. On the
contrary, if  < 0, ki , i = 0, . . . , n, are all holes, while p j , j = 1, . . . , n, all particles,
i.e., ki < 0 and p j > 0. Energy conservation requires


n

 = k0 + ki − pi ,
i=1

where all terms on the right hand side have the same sign as . Therefore, if the latter is
small in absolute value, each term on the right hand side is equally small in absolute
value, i.e., ki < ||, i = 0, . . . , n, and p j < ||, j = 1, . . . , n. Proceeding as

then
 ∞  ∞
   3 π2 2
d0 d1 f − 0 − 1 f 0 f 1 = T 2 dx dy F(−x − y) F(x) F(y) = T 2 ζ (2) = T ,
−∞ −∞ 2 2

being ζ (n) the Riemann ζ -function.

154 4 Feynman Diagram Technique

before, we may realise that the phase space for the decay process vanishes at least
as  2n , or, at  = 0 but T = 0, as T 2n .
Let us now discuss whether Sk (0, 0, 0) in (4.59) is indeed finite. Let us consider,
for simplicity, k = (k), where k = |k|. In this case, (k) = 0 when k = k F , the
Fermi momentum. In units such that the volume of the unit cell v = 1,

1   
Sk (0 , 1 , ω1 ) = dk0 dk1 dp1 δ 0 − (k0 ) δ 1 − (k1 ) δ ω1 − ( p1 )
(2π )2d ρ03

δ k + p1 − k0 − k1 .

Since 0 ∼ 1 ∼ ω1 ∼ 0, then k0 ∼ k1 ∼ p1 ∼ k F and thus k0 + k1 − p1 ≤ 3k F ,

so that momentum conservation k = k0 + k1 − p1 implies that Sk (0 , 1 , ω1 ) is
finite only if |k| ≤ 3k F , otherwise it vanishes. We note that Sk (0 , 1 , ω1 ) does not
depend on the direction of k but only on its magnitude k, so that, if k = kv 1 with v 1
a unit vector, then

dv 1
Sk (0 , 1 , ω1 ) = Sk (0 , 1 , ω1 ) ≡ Sk (0 , 1 , ω1 ) , (4.63)
d
where d is the d-dimensional solid angle. Let us define k ≡ κ1 k F , and

k0 :  k0 = 0 , k0 ≡ −k0 v 2 ≡ −k F κ2 v 2 ,

k1 :  k1 = 1 , k1 ≡ −k1 v 3 ≡ −k F κ3 v 3 ,

p 1 :  p 1 = ω1 , p1 ≡ p1 v 4 ≡ k F κ4 v 4 ,

with all v i unit vectors, through which we can write, assuming the density of states
smooth around the chemical potential,
 
4
(2π )d 
Sk (0 , 1 , ω1 )  dv i δ κ1 v 1 + κ2 v 2 + κ3 v 3 + κ4 v 4
4d k dF i=1
 4 

1
= dr dv i eiκi r·vi
4d k dF i=1
 ∞ 
4
1
≡ r d−1 dr Id (κi r ) .
3d k dF 0 i=1

The integral over r is convergent at r = 0 but may be divergent at r → ∞. For that,

we need to know how Id (κi r ) behaves at r → ∞. Noticing that in d = 1 the integral
over a unit vector v i is just the sum over vi = ±1, we find

sin κi r
I3 (κi r ) = 3 ,
κi r
 1  π
I2 (κi r ) = 2 J0 κi r −−−→ √ cos κi r − ,
r →∞ κi r 4

I1 (κi r ) = 1 cos κi r .
4.3 Perturbation Theory for the Single-Particle Green’s Function … 155

!4
In d = 3, i=1 I3 (κi r ) ∼ 1/r 4 and the integral converges even if we set all κi = 1,
thus 0 = 1 = ω1 = 0. In other words, Sk (0, 0, 0) in d = 3 is finite.
In d < 3, the integral with all κi = 1 diverges logarithmically in d = 2, where
I24 (r ) ∼ 1/r 2 , and linearly in d = 1, where I14 (r ) ∼ constant. In both cases, the
different κi = 1 play a crucial role in cutting off the singularity. Since close to the
Fermi surface, e.g.,
  0
0 = (k0 )  v F k0 − k F = v F k F κ2 − 1 ⇒ κ2  1 + ,
vF k F

and similarly for all other momentum, taking, for simplicity, k = k F , so that κ1 = 1,
in d = 1, 2 the product of the cosine functions yields, at leading order,

  
 π 1   1
4
0 + 1 − ω1
cos κi r −  cos 1 − κ2 − κ3 + κ4 r = cos r
4 8 8 vF k F
i=1
 
1 
= cos r ,
8 vF k F

explicitly showing that the finite  cuts off the large r singularity. As a result, in
d = 1 and d = 2 the  2 behaviour in (4.62) is replaced, respectively, by − 2 ln ||
and ||.
In conclusion, we have shown that, order by order in perturbation theory and at
T = 0,
⎧
⎪
⎨ −
2 d = 3,
lim Im (, k) ∼  ln  d = 2 ,
2 (4.64)
→0 ⎪
⎩
− d = 1,
and thus vanishes. However, through (4.54), the one-dimensional behaviour
Im (, k) ∼ −  implies that

∂Re (, k) dω Im (ω, k)
=−  , (4.65)
∂ π −ω
2

diverges as ln || → −∞ as  → 0, corresponding to a vanishing quasiparticle

residue Z ( → 0, k), see (4.57). That is actually one of several singularities that
appear in perturbation theory, symptoms that the latter does not converge in d = 1
however small the interaction strength is. Notice that, in d = 2, 3 the integral in
(4.65) converges as  → 0, implying a finite quasiparticle residue Z ( → 0, k).
Hereafter, we thus proceed assuming d > 1, so that perturbation theory is well
behaved. However, even in that case, the observation that order by order in pertur-
bation theory Im( → 0, k) ∼ − 2 , with log-corrections in d = 2, does not guar-
antee that the whole perturbation series converges and the result indeed vanishes
like − 2 . For that reason, we make the assumption that in d > 1 the perturbation
series has a finite convergence radius, and, further, that the interaction strength is
156 4 Feynman Diagram Technique

within that radius. Later, when discussing Landau’s Fermi liquid theory, we shall
relax the latter assumption. Therefore, for now on we take for granted that, if each
term in the perturbative expansion of Im (, k) vanishes, e.g., as − 2 at small ,
also the whole series vanishes similarly. Correspondingly, as discussed above the
quasiparticle residue Z (, k) is finite as  → 0 for any k.
We conclude by highlighting another important side result that derives from (4.58)
before taking the analytic continuation on the real axis. Indeed, if we consider the
imaginary part of that equation for positive Matsubara frequency  > 0,

2U 2  
Im (i, k) = − 2
δk+p1 ,k0 +k1  2
V k0 k1 p1 p1 − k0 − k1 +  2
          
f p1 1 − f k0 1 − f k1 + 1 − f p1 f k0 f k1 ,

and note that the Lorentzian function

1  
 −−−→ δ p1 − k0 − k1 ,
π p − k − k 2
+ 2 →0
1 0 1

so that

lim Im (i, k) = lim Im ( + iη, k) = 0 , (4.66)

→0 →0

and, in particular, Im (i, k) ∼ − for small . As before, this result remains true
order by order in perturbation theory, so that, if the latter converges and the strength
of interaction is within the convergence radius, we can conclude that (4.66) holds
for the whole perturbation series.

4.3.5 Emergence of Quasiparticles

Assuming, e.g., d = 3, we just showed that Im ( → 0, k)  −γ (k)  2 , so that

the single-particle density-of-states in (4.56)
1 −Im (, k)
A(, k) =  2
π
 − k − Re (, k) + Im2 (, k)
Z (0, k) γ∗ (k)  2
−−→  ≡ Z (0, k) Aqp (, k) ,
→0 π 2
 − ∗ (k) + γ∗ (k)2  4
(4.67)

where Aqp (, k) will be denoted as the quasiparticle density of states, Z (, k) the
quasiparticle residue in (4.57), and
 
∗ (k) ≡ Z (0, k) k − Re (0, k) , γ∗ (k) ≡ Z (0, k) γ (k) . (4.68)
4.4 Other Kinds of Perturbations 157

Aqp (, k) in (4.67) typically vanishes at the chemical potential, i.e.,  = 0, unless
on the surface defined through ∗ (k) = 0. Since Z (0, k) is finite, that occurs when

k F + Re (0, k F ) = 0 , (4.69)

which defines the interacting Fermi surface. As a consequence,

1 γ∗ (k)  2 
Aqp ( → 0, k)   −−−−→ δ  − ∗ (k) , (4.70)
π  − ∗ (k) 2 + γ∗ (k)2  4 k→k F

looks like the density of states of non interacting particles, the quasiparticles, with
dispersion ∗ (k) in momentum space. Strictly speaking, the quasiparticle at energy 
and momentum k has a finite decay rate γ∗ (k)  2 , which however vanishes faster than
its energy approaching the Fermi surface. We can thus state that the quasiparticles
are coherent excitations, namely their lifetime diverges at the Fermi surface.
It follows that the physical electron DOS

A( → 0, k → k F ) = Z (0, k) Aqp ( → 0, k → k F ) = Z (0, k F ) δ  − ∗ (k) ,

also becomes a δ-function centred at the chemical potential for k → k F , though

with finite weight Z (0, k). Since the integral over  of A(, k) is unity, Z (0, k F )
is positive and less than one. The rest 1 − Z (0, k F ) of the spectral weight must be
concentrated at finite energy, so that

A(, k → k F ) = Z (0, k F ) δ  − ∗ (k) + Ainc (, k) , (4.71)

where Ainc (, k) carries the rest of spectral weight and is commonly denoted as
incoherent background to distinguish it from the quasiparticle coherent δ-peak. Cor-
respondingly, through (4.25),
 
A(, k) Z (0, k F ) Ainc (, k)
G(z, k) = d −−−−→ + d
z −  k→k F z − ∗ (k) z−
Z (0, k F )
≡ + G inc (z, k) ≡ G coh (z, k) + G inc (z, k) .
z − ∗ (k)
(4.72)

Therefore, the Green’s function for k moving towards the Fermi surface develops
a coherent part G coh (z, k), i.e., a simple pole on the real axis at the quasiparticle
energy ∗ (k) with finite quasiparticle residue, Z (0, k F ), thus its name, besides an
incoherent component G inc (z, k) that has instead a branch cut on the real axis.

4.4 Other Kinds of Perturbations

Before proceeding deep in analysing the perturbation theory in the electron-electron

interaction, let us briefly discuss the diagrammatic techniques for other two kinds of
perturbations: a scalar potential and the coupling to bosonic modes.
158 4 Feynman Diagram Technique

4.4.1 Scalar Potential

Let us suppose to have non-interacting electrons identified by a quantum label a,

with non-interacting Green’s functions

G 0 ab (τ ; x, y) = δab G 0 a (τ ; x, y) ,

in presence of the scalar potential


V = dx a (x) Vab (x)
†
b (x). (4.73)
ab

The S-operator is now

  
 β 
S(β, 0) = Tτ exp − dτ dx a (x, τ ) Vab (x)
†
b (x, τ )
ab 0

  

≡ Tτ exp − dx a (x) Vab (x)
†
b (x) ,
ab

where, as before, we have introduced the four-dimensional coordinate x = (τ, x).

Upon expanding S(β, 0) up to first order and keeping only connected diagrams, one
can readily obtain the Green’s function expansion
 
†

G ab (x, y) = δab G 0 a (x, y) + dz Vcd (z) Tτ a (x) b (y) c (z)
†
d (z) conn
cd

= δab G 0 a (x, y) + dz Vab (z) G 0 a (x, z) G 0 b (z, y). (4.74)

If we represent graphically the perturbation as in Fig. 4.16a, then the Green’s function
(4.74) up to first order can be drawn as the first two diagrams in Fig. 4.16b. Higher
order terms can be simply obtained by inserting other potential lines, as the second
order term in Fig. 4.16b. All diagrams have a positive sign. The Dyson equation can
be easily read out:

G ab (x, y) = δab G 0 a (x, y) + dz G 0 a (x, z) Vac (z) G cb (z, y), (4.75)
c

and is drawn in Fig. 4.16b.

4.4 Other Kinds of Perturbations 159

(a) a b
Vab(x) = x
(b) a b
a b ab a a c b
= + +
ab a a c b
= +

Fig. 4.16 a Graphical representation of the scalar potential perturbation. b Perturbation expansion
of the Green’s function and Dyson equation. The bold blue line represents the Green’s function at
all orders in perturbation theory

4.4.2 Coupling to Bosonic Modes

Let us imagine now that our system of electrons is coupled to bosonic modes
described by the free Hamiltonian

Hbos = ωq aq† aq .
q

These modes could represent phonons, photons, or any other bosonic excitation. The
coupling term is assumed to be

V = g(q) xq ρ(−q) , (4.76)
q

where g(q)∗ = g(−q) is the coupling constant, the phonon coordinate is


1  †

xq = aq + a−q ,
2

in terms of bosonic annihilation and creation operators, and

 †
ρ(q) = ckσ ck+qσ ,
σ k

is the electron density operator.

We can perform perturbation theory in (4.76) using Wick’s theorem both for
the electrons and for the bosons. The latter amounts to contract an xq with x−q
which leads to the free propagator D0 (iω, q) of (4.33). Therefore, if we represent
the electron-boson coupling as in Fig. 4.17, in which a dotted vertex line represents
the boson-coordinate, by contracting two vertices one recovers an effective electron-
electron interaction, see also Fig. 4.17, which is mediated by the bosons and given
by
2 2 ωq
g(q) D0 (iω, q) = − g(q) . (4.77)
ω2 + ωq2
160 4 Feynman Diagram Technique

i + iω, k + q i ,k i + iω, k + q i ,k
g(q)

xq (iω)
2
= g(q) D0 (iω, q)
x−q (−iω)

g(−q)
i ,k i + iω, k + q i ,k i + iω, k + q

Fig.4.17 On the left: electron-boson vertices at q and −q. The external dotted vertex line represents
boson coordinates. On the right: the effective electron-electron interaction after contracting the
bosons

Fig. 4.18 Lowest orders in

the diagrammatic = +
+
perturbation expansion of
D(iω, q). The
fully-interacting D(iω, q) is
represented by a bold dashed
line, while D0 (iω, q) by a
+
thiner dashed line. The solid
lines are electron Green’s
functions + +
Such effective interaction is retarded, namely depends on the frequency, and attrac-
tive. Specifically, if we move on the real axis, then

2 2 ωq
g(q) D0 (iω, q) −−−−−−→ − g(q) ,
iω→ω+iη ωq2 − ω2 − iω η

where, as usual, η > 0 is infinitesimal, which is attractive if ω2 < ωq2 and repulsive
otherwise.
The rules for constructing diagrams are therefore the same as for the electron-
electron interaction, with the additional complication that interaction is frequency
dependent.
One may also investigate the effects of the electron-boson coupling on the boson
Green’s function D(iω, q). The perturbation expansion is shown up to second order
in Fig. 4.18. Two kinds of diagrams can be identified. The first class includes dia-
grams which can be divided into two lower order diagrams by cutting a D0 (iω, q)-
line, like the first second order diagram shown in Fig. 4.18. These diagrams are
called reducible. The other class includes all other diagrams which are therefore irre-
2
ducible. If we define, similarly to Dyson’s equation, a self-energy g(q) (iω, q)
that includes all irreducible diagrams, shown up to second order in Fig. 4.19, we can
4.5 Two-Particle Green’s Functions and Correlation Functions 161

= + + +

= +
Fig. 4.19 Upper panel: Boson self-energy, represented by a filled circle, up to second order. Lower
panel: Dyson equation for D(q)

easily derive the Dyson equation, also shown in the same figure, whose solution is
D0 (iω, q) ωq
D(iω, q) = 2
=− 2
1 − g(q) D0 (iω, q) (iω, q) ω 2
+ ωq2 − g(q) (iω, q)
ωq
−−−−−−→ 2
. (4.78)
iω→ω+iη ω2 − ωq2 + g(q) (ω, q)

We note that, when continued on the real axis, the real solutions of
2 
ω∗2 (q) = ωq2 − g(q) Re  ω∗ (q), q) ,

represent the boson dispersion relation modified by the coupling to electrons, which
2 
also provide a decay rate to the bosonic mode proportional to g(q) Im  ω∗ (q), q).

4.5 Two-Particle Green’s Functions and Correlation Functions

Now we move back to the case of an instantaneous electron-electron interaction, and

proceed in analysing the perturbative expansion.
In that case, the interacting Hamiltonian is
 † 1  † †
H= k ckσ ckσ + U (q) cpα ck+qβ ckβ cp+qα . (4.79)
2V
kσ kpq αβ

The Heisenberg imaginary time evolution of an annihilation operator is

∂ckσ (τ )   1  †
− = ckσ (τ ) , H (τ ) = k ckσ (τ ) + U (q) cp+qα (τ )cpα (τ )ck+qσ (τ ) ,
∂τ V pqα

through which one readily find the equation of motion of the Green’s function
  
∂ ∂ †
− G(τ, k) = − − Tτ ckσ (τ ) ckσ 
∂τ ∂τ
   
∂ † †  ∂ckσ (τ ) † 
=− − θ (τ ) ckσ (τ ) ckσ  + θ (−τ ) ckσ ckσ (τ ) = δ(τ ) − Tτ ckσ
∂τ ∂τ
1   
† †
= δ(τ ) + k G(τ, k) − U (q) Tτ cp+qα (τ ) cpα (τ ) ck+qσ (τ ) ckσ .
V pqα
162 4 Feynman Diagram Technique

This equation can be written as

   
∂ 1 
−  − k G(τ  , k) = δ(τ  ) − †
U (q) Tτ cp+qα (τ  ) cpα (τ  ) ck+qσ (τ  ) ckσ
†
.
∂τ V pqα
(4.80)
The non-interacting Green’s function satisfies on the contrary
 
∂
− − k G 0 (τ  , k) = δ(τ  ) .
∂τ 

If we multiply both sides of (4.80) by G 0 (τ − τ  , k) and integrate over τ  , we obtain

  
  ∂
dτ G 0 (τ − τ , k) G(τ  , k) = G 0 (τ, k)
−  − k
∂τ
  
1 
− U (q) dτ  G 0 (τ − τ  , k) Tτ cp+qα
†
(τ  ) cpα (τ  ) ck+qσ (τ  ) ckσ
†
,
V pqα

and upon integrating by part the left hand side, one obtains

G(τ, k) = G 0 (τ, k)
  
1 
− U (q) dτ  G 0 (τ − τ  , k) Tτ cp+qα
†
(τ  ) cpα (τ  ) ck+qσ (τ  ) ckσ
†
.
V pqα
(4.81)

Therefore the single-particle Green’s function can be expressed in terms of a two-

particle Green’s function. Namely, let us define the two-particle Green’s function
  
K σ1 σ2 ;σ3 σ4 τ1 p1 , τ2 p2 ; τ3 p3 , τ4 p4 = −Tτ cp1 σ1 (τ1 )cp2 σ2 (τ2 ) cp†3 σ3 (τ3 )cp†4 σ4 (τ4 )  , (4.82)

where, by momentum conservation,

p1 + p2 = p3 + p4 ,

in terms of which

G(k, τ ) = G 0 (k, τ )

1  
− U (q) dτ G 0 (τ − τ  , k) K σ α;ασ τ  k + q, τ  p; τ  + 0+ p + q, 0 k ,
V pqα
(4.83)

†
where, since in (4.81) the operator cq+qα (τ  ) appears on the left, we have set its time
 +
to τ + 0 to enforce that.
4.5 Two-Particle Green’s Functions and Correlation Functions 163

4.5.1 Diagrammatic Representation of the Two-Particle Green’s

Function

In absence of interaction,
  
K σ1 σ2 ;σ3 σ4 τ1 p1 , τ2 p2 ; τ3 p3 , τ4 p4 = −Tτ cp1 σ1 (τ1 )cp2 σ2 (τ2 ) cp†3 σ3 (τ3 )cp†4 σ4 (τ4 ) 0
= −δσ1 σ4 δσ2 σ3 δp1 p4 δp2 p3 G 0 (τ1 − τ4 , p1 ) G 0 (τ2 − τ3 , p2 )
+ δσ1 σ3 δσ2 σ4 δp1 p3 δp2 p4 G 0 (τ1 − τ3 , p1 ) G 0 (τ2 − τ4 , p2 ) .

The interaction in perturbation theory has two effects. First it turns each G 0 into the
fully-interacting G, and next it couples together the two Green’s functions. For that
reason, we can formally write

K σ1 σ2 ;σ3 σ4 τ1 p1 , τ2 p2 ; τ3 p3 , τ4 p4 = −δσ1 σ4 δσ2 σ3 δp1 p4 δp2 p3 G(τ1 − τ4 , p1 )G(τ2 − τ3 , p2 )
+ δσ1 σ3 δσ2 σ4 δp1 p3 δp2 p4 G(τ1 − τ3 , p1 )G(τ2 − τ4 , p2 )
 
4
+ dτi G(τ1 − τ1 , p1 ) G(τ2 − τ2 , p2 ) G(τ3 − τ3 , p3 ) G(τ4 − τ4 , p4 )
i=1

σ1 σ2 ;σ3 σ4 τ1 p1 , τ2 p2 ; τ3 p3 , τ4 p4 , (4.84)

which is represented graphically to Fig. 4.20. In Matsubara frequency, and taking

into account spin, momentum and frequency conservation, we can write the four-leg
interaction vertex  as

σ1 σ2 ;σ3 σ4 i + iω k + q, i  k ; i  + iω k + q, i k , σ1 + σ2 = σ3 + σ4 ,
(4.85)

where  and   are fermionic Matsubara frequencies, ω a bosonic one, and the spin
condition implies that the sum of the spin z-components of the incoming electrons,
σ1 + σ2 , must be equal to the sum of the outgoing electrons, σ3 + σ4 . The skeleton
diagrams for  up to second order are shown in Fig. 4.21, where, e.g., 1 stands for
frequency i + iω, momentum k + q and spin σ1 .
In conclusion the (4.83) of the single particle Green’s function in terms of the
two-particle one can be expressed as function of the single particle Green’s function
itself and the interaction vertex as shown in Fig. 4.22. Notice that the interaction
vertex acts as the bare interaction, so it carries a (–1) sign. Since there is a loop in

1 3 1 3 1 1 3 3
K(1, 2; 3, 4) = - + +
4 2 4 2 4 4 2 2

Fig. 4.20 Graphical representation of the two-particle Green’s function. Here, all numbers are short
notation that include the imaginary time, momentum and spin of each vertex. The coloured box
represent the four-leg vertex (1 , 2 ; 3 , 4 ). Notice that all Green’s functions are fully interacting
ones
164 4 Feynman Diagram Technique

1 3

= + - + +
4 2

- - - -

+ + + -

Fig. 4.21 Lowest order skeleton expansion of the interaction vertex (1, 2; 3, 4) in skeleton dia-
grams. The external vertices, 1, 2, 3 and 4, where each number stands for frequency, momentum and
spin, are drawn as narrow arrow lines, while the internal interacting Green’s functions as bold arrow
lines. We explicitly show the sign of each diagram that can be found from the first two recalling
that each interaction and loop bring an minus sign

= + -

Fig. 4.22 Dyson equation in terms of the interaction vertex 

the third diagram of Fig. 4.22, the overall sign is plus. This result also provides an
expression for the single-particle self-energy shown in Fig. 4.23. Figure 4.23 implies
the following analytic equation

1  
(i, k) = 2 U (0) − U (k − k ) n(k )
V 
k
1  
+ 2
T2 U (q) G(i + iω, k + q) G(i  , k ) G(i  + iω, k + q)
V k q σ  ω

σ,σ  ;σ  ,σ i + iω k + q, i  k ; i  + iω k + q, i k , (4.86)

where
  0+
n(k ) ≡  ck†  σ ck σ  = T ei G(i  , k ) , (4.87)

 
is the interacting momentum distribution. Since  =  G, U , the above equation
implies that also  can be expressed as a functional of G and U .
4.5 Two-Particle Green’s Functions and Correlation Functions 165

=+ - +
Fig. 4.23 Single-particle self-energy in terms of the interaction vertex 

4.5.2 Correlation Functions

Let us consider the density operators,

  †
  †
A(q) = 0A k α, k + q β) ckα ck+qβ , B(q) = 0B k α, k + q β) ckα ck+qβ ,
k αβ k αβ

which have bosonic character, and the expectation value of their time-ordered prod-
uct, which we shall denote as the correlation function χ AB instead of the G AB Green’s
function of (4.4),
1
χ AB (τ, q) ≡ −
Tτ (A(q, τ ) B(−q))
V
1      
† †
=− 0A k α, k + q β) 0B p + q γ , p δ)  Tτ ckα (τ )ck+qβ (τ )cp+qγ cpδ 
V
k αβ p γ δ
1   A  
= 0 k α, k + q β) 0B p + q γ , p δ) K βδ;γ α τ k + q, 0 p; 0+ p + q, τ + k ,
V
k αβ p γ δ

where we use the definition (4.82) of the two-particle Green’s function, and times 0+
and τ + are assumed infinitesimally larger than 0 and τ , since in the definition of the
density operators the creation operator is to the left of the annihilation one. Through
the expression of the two-particle Green’s function in terms of the interaction vertex,
see (4.84) and also Fig. 4.20,

δq,0
χ AB (τ, q) = −  A(0)   B(0) 
V
1   
+ 0A k α, k + q β) 0B k + q β, k α) G(τ, k + q) G(−τ, k)
V
k αβ
1   
+ 0A k α, k + q β) 0B p + q γ , p δ)
V
kp αβγ δ
 
4
dτi G(τ − τ1 k + q) G(−τ2 , p) G(τ3 , p + q) G(τ4 − τ, k)
i=1

β,δ;γ ,α τ1 k + q, τ2 p; τ3 p + q, τ4 k ,
166 4 Feynman Diagram Technique

Fig. 4.24 Graphical representation of the correlation function χ AB (iω, q). All Green’s functions
are fully-interacting ones

Fig. 4.25 Top panel: definition of interacting A-density vertex, in orange, as opposed to the non-
interacting one, light yellow triangle. Bottom panel: diagrammatic representation of the correlation
function χ AB in terms of the interacting A-density vertex. Note that the B-density vertex is the
non-interacting one. We could alternatively use the interacting B-density vertex, defined just by the
same equation as in the top panel with A replaced by B, in which case, the A-density vertex would
be the non-interacting one while the B-density vertex the interacting one

where, e.g.,
 
 A(0)  ≡ 0A k α, k α) G(k, 0− ) ,
kα

is just the expectation value of the density operator A(q) at q = 0. The first term con-
tributes only at q = 0, and can be generally absorbed into the definition of χ AB (τ, 0).
With that prescription, the Fourier transform in Matsubara frequencies of the corre-
lation function has the diagrammatic representation of Fig. 4.24.
In the figure, the triangular vertices represent the matrix elements 0A and 0B
of the corresponding density operators, which we shall denote as non-interacting A-
and B-density vertices. We can correspondingly define interacting A- and B-density
vertices as shown in Fig. 4.25. The explicit dependence upon frequency, momentum
and spin variables of the density-vertex is shown in Fig. 4.26.
In the specific case of the charge-density,
 †
ρ(q) = ckσ ck+qσ ,
kσ

the non-interacting vertex, which we hereafter denote simply as 0 without any

superscript, is simply unity, i.e., 0 = 1, while the interacting one, consistently
denoted as , is graphically shown in Fig. 4.26. Comparing the graphical repre-
4.5 Two-Particle Green’s Functions and Correlation Functions 167

Fig. 4.26 Explicit dependence upon frequency, momentum and spin variables of the, e.g., A-
density-vertex. The lines are just external legs carrying the shown frequency, momentum and,
eventually, spin

Fig. 4.27 Top panel: graphical representation of the interacting charge-density vertex  in terms
of interacting Green’s function G and the four-leg vertex . Bottom panel: graphical representation
of the self-energy  in terms of G and , cf. with Fig. 4.23

sentation of  with that of the self-energy in Fig. 4.23, we find the alternative
representation shown also in Fig. 4.27, whose analytic expression, cf. (4.86), reads

(i, k) = U (0) n
1   
− T G(i − iω, k − q) U (q)  i − iω k − q σ, i k σ ; iω, q ,
V qσ ω
(4.88)

where, see (4.87),

1  1   i0+
n= n(k) = T e G(i, k) ,
V V 
kσ kσ

is the total electron density.

4.5.2.1 Non-interacting Values

In the absence of interaction only the first term in Fig. 4.24 survives with the Green’s
function lines being the non-interacting ones. Since there is a loop, according to what
we said before, the sign is (−1) L−1 = (−1)1−1 = 1 hence
1   A 
χ0 AB (iω, q) = T 0 k α, k + q β) 0B k + q β, k α)
V 
kαβ
G 0 (i + iω, k + q) G 0 (i, k)
1    A   1 1
= T Tr 0 k, k + q) 0B k + q, k) ,
V n
i + iω − k+q i − k
k
(4.89)
168 4 Feynman Diagram Technique

where the trace is over the spin variables. By means of (4.11) we can easily perform
the sum over frequencies in (4.89), which gives
 1 1  1  1
T = f k+q − iω + f k
 i + iω − k+q i − k k+q − iω − k iω − k+q + k
 
f k − f k+q
= ,
iω − k+q + k

where we used the fact that, if iω is bosonic, then f ( ± iω) = f (). The final result
is therefore

1   A   f k − f k+q
χ0 AB (iω, q) = Tr 0 k, k + q) 0 k + q, k)
B
.
V iω − k+q + k
k
(4.90)
The linear response function is simply obtained by iω → ω + iη, with η an infinites-
imal positive number.

4.6 Coulomb Interaction and Proper and Improper Response

Functions

Let us now consider the case in which the electron-electron interaction is the Coulomb
repulsion
4π e2
U (q) = ,
q2
which is singular for q → 0. Strictly at q = 0, the repulsion is cancelled by the
neutralising ionic charge, which implies, for instance, that the Hartree term in (4.88)
vanishes.
In order to cure that singularity, it is convenient to recast perturbation theory in a
different manner, which naturally brings to identify the proper and improper response
functions we earlier introduced. In Fig. 4.28 we draw the diagrammatic skeleton
expansion up to first order of the density-density correlation function χ (iω, q),
which is also the improper response function once analytically continued on the
real frequency axis from above.
Already at first order we can distinguish two kinds of diagrams: those which can
be cut into two by cutting an interaction line, as the last diagram, and those which
cannot, all the others, which we define as irreducible with respect to the interaction.

Fig. 4.28 Diagrammatic skeleton expansion of the density-density correlation function up to first
order. The black dots represent the non-interacting charge density vertex, 0 , which is just the
identity
4.6 Coulomb Interaction and Proper and Improper Response Functions 169

We define χ∗ (iω, q) the sum of all irreducible diagrams, in the above sense. We can
formally write χ∗ (iω, q) as

1   
χ∗ (iω, q) = T G(i, k) ∗ i k σ, i + iω k + q σ ; iω, q G(i + iω, k + q) ,
V 
kσ
 (4.91)
which defines the proper charge density-vertex ∗ i k σ, i + iω k + q σ ; iω, q ,
shown graphically in the upper panel of Fig. 4.29. In terms of χ∗ (iω, q) the pertur-
bation expansion of χ (iω, q) can be recast as in the lower panel of Fig. 4.29, which
has the formal solution3

χ∗ (iω, q) χ∗ (iω, q)
χ(iω, q) = ≡
4π e2  (iω, q)
1 − 2 χ∗ (iω, q)
q

1   ∗ i k σ, i + iω k + q σ ; iω, q
= T G(i, k) G(i + iω, k + q) ,
V 
 (iω, q)
kσ
(4.92)

proving that χ∗ (iω, q) is actually the proper density-density response function that
we have previously introduced, which defines the longitudinal dielectric constant
 (iω, q). Since

1   
χ(iω, q) = T G(i, k)  i k σ, i + iω k + q σ ; iω, q G(i + iω, k + q) ,
V 
kσ

in terms of the interacting improper charge density-vertex, comparing the above

equation with the last equation in (4.92), we find a relationship between improper
and proper charge density-vertices

 1 
 i k σ, i + iω k + q σ ; iω, q = ∗ i k σ, i + iω k + q σ ; iω, q .
 (iω, q)
(4.93)

It follows that the expression (4.88) of the self-energy in terms of the improper charge
density-vertex  changes into

(i, k) =  H F − U (0) ρion

1   
− T G(i − iω, k − q) W (iω, q) ∗ i − iω k − q σ, i k σ ; iω, q
V qσ ω

≡  H F − U (0) ρion +  xc (i, k) , (4.94)

3 Notice both the interaction and χ∗ (iω, q), which is a loop, bring a minus sign, hence the sign is
plus.
170 4 Feynman Diagram Technique

χ∗ ρ∗

χ ρ

= χ∗ + χ∗ χ

Fig. 4.29 Upper panel: graphical representation of χ∗ (iω, q) in terms of the proper charge density-
vertex ∗ i k σ, i + iω k + q σ ; iω, q , blue triangle with label ρ∗ . Note that the non-interacting
charge density-vertex 0 = 1. Lower panel: graphical representation of the equation that relates
χ(iω, q) with χ∗ (iω, q). We also show in the first line the representation of χ(iω, q) in terms of
the interacting charge density-vertex , green triangle with label ρ∗

W
ionic
potential
G ρ∗
= − −

Fig. 4.30 Graphical representation of the self-energy in terms of the interacting Green’s function
G, the screened Coulomb interaction W and the proper charge density-vertex ∗

where the Hartree term cancels the contribution from the interaction with the positive
ionic charges, and

4π e2 4π e2
W (iω, q) = = , (4.95)
q 2 (iω, q) q 2 − 4π e2 χ∗ (iω, q)

is the screened Coulomb interaction.

The graphical representation of (4.94) is shown in Fig. 4.30. The last contribution
in (4.94) is named exchange-correlation self-energy  xc (i, k), and includes all
corrections beyond the simple Hartree potential.

4.7 Irreducible Vertices and the Bethe-Salpeter Equations

Let us consider again the interaction vertex (1, 2; 3, 4), whose lower order skeleton
diagrams
 are shown in Fig. 4.21. Here we assume  that 1 ≡ i + iω, k + q, σ1 ,
2 ≡ − i − iω + i, −k − q + Q, σ2 , 3 ≡ − i + i, −k + Q, σ3 and 4 ≡
i, k, σ4 , with the constraint σ1 + σ2 − σ3 − σ4 = 0 by conservation of the z-
component of the total spin. We denote the vertices 1 and 2 that correspond to creation
operators as particles, while 3 and 4 that correspond to annihilation operators as holes.
It follows that (1, 2; 3, 4) can be regarded as a particle-hole scattering amplitude,
namely the sum of all interaction processes that annihilate the particle-hole (p-h) pair
4.7 Irreducible Vertices and the Bethe-Salpeter Equations 171

1 3 1 4

4 2 2 3
particle-hole particle-particle

Fig. 4.31 Interaction vertex (1, 2; 3, 4) represented in the particle-hole channel, left, and particle-
particle one, right

1 3 1 3 1 3

= Γp-h
0 + Γp-h
0

4 2 4 2 4 2

Fig. 4.32 Bethe-Salpeter equation for the interaction vertex  in the particle-hole channel

3&2 and create the pair 1&4, see left picture in Fig. 4.31. Each pair is characterised
by the frequency transfer iω, momentum transfer q and z-component of the spin
σ1 − σ4 = σ3 − σ2 , in short 1 − 4 = 3 − 2. Equivalently, (1, 2; 3, 4) can be seen
as a particle-particle scattering amplitude, where the particle-particle (p-p) pair 4&3
is annihilated and the pair 1&2 created, see right picture in Fig. 4.31. Each pair, also
known as Cooper pair, is characterised by the total frequency i, total momentum
Q and z-component of the spin σ1 + σ2 = σ4 + σ3 , in short 1 + 2 = 4 + 3. We
emphasise that, even though the two scattering process are different, (1, 2; 3, 4) is
the same.

4.7.1 Particle-Hole Channel

The lowest order skeleton expansion of (1, 2; 3, 4) regarded as a p-h scattering

amplitude was already shown in Fig. 4.21. In that figure we can distinguish two
class of diagrams. Indeed, we note that the diagrams highlighted in red can be cut
into two by cutting a particle line, from left to right, and a hole one, from right to
left. In that case the frequency, momentum and z-component of the spin differences
between the two lines is exactly 1 − 4 = 3 − 2. We denote such class of diagrams
as reducible in the particle-hole channel. All other diagrams are instead irreducible
in the particle-hole channel. We name the sum of the irreducible diagrams at all
p-h
orders as 0 , in terms of which  can be obtained by solving the equation shown
in Fig. 4.32, which we shortly write as

0 (1, 2 ; 3 , 4) R(1 , 2 ; 3 , 4 ) (1 , 2; 3, 4 )
p-h p-h
(1, 2; 3, 4) = 0 (1, 2; 3, 4) +
1 ,2 ,3 ,4
p-h p-h
≡ 0 + 0  R , (4.96)
172 4 Feynman Diagram Technique

p-h
where, regarding , 0 and R as tensors,  indicates the tensor product. We define

R(1 , 2 ; 3 , 4 ) ≡ δ1 3 δ2 4 G(1 ) G(4 ) , (4.97)

 
where, since by construction 1 − 4 = 1 − 4, if 4 = i  , k , σ4 , then 1 = i  +
iω, k + q, σ1 , with σ1 − σ4 = σ1 − σ4 . The lowest order skeleton expansion of
p-h
0 contains the diagrams in Fig. 4.21 excluded the four highlighted reducible ones.
Equation (4.96) is the Bethe-Salpeter equation in the particle-hole channel.
We can also exploit spin SU (2) symmetry. We introduce the Pauli matrices in the
spin space,
     
01 0 −i 1 0
σ̂1 = , σ̂2 = , σ̂3 = ,
10 i 0 0 −1

including the identity σ̂0 = Iˆ, and note that spin SU (2) symmetry implies that
1   
σ̂a σ σ σ̂b σ σ σ1 ,σ2 ;σ3 ,σ4 = δab a , a, b = 0, . . . , 3 , (4.98)
4 σ σ σ σ 1 4 2 3
1 2 3 4

as well as 1 = 2 = 3 . We therefore define the scattering amplitude in the S = 0

p-h scattering channel through
1   
S ≡ σ̂0 σ1 σ4
σ̂0 σ2 σ3
σ1 ,σ2 ;σ3 ,σ4
4 σ1 σ2 σ3 σ4
1  1 
= δσ1 σ4 δσ2 σ3 σ1 ,σ2 ;σ3 ,σ4 = ↑,↑;↑,↑ + ↓,↓;↓,↓ + ↑,↓;↓,↑ + ↓,↑;↑,↓
4 σ1 σ2 σ3 σ4
4
1 
= ↑,↑;↑,↑ + ↑,↓;↓,↑ , (4.99)
2

since ↑,↑;↑,↑ = ↓,↓;↓,↓ and ↑,↓;↓,↑ = ↓,↑;↑,↓ again by spin-SU (2).

Similarly, we define the scattering amplitudes in the S = 1 p-h scattering channel
a , a = 1, 2, 3, where, e.g.,
1    1 
3 ≡ σ̂3 σ σ σ̂3 σ σ σ1 ,σ2 ;σ3 ,σ4 = ↑,↑;↑,↑ − ↑,↓;↓,↑ .
4 σ σ σ σ 1 4 2 3 2
1 2 3 4
(4.100)

Since, as we mentioned, a , a = 1, 2, 3, are all equal, then

1 = 2 = 3 ≡  A , (4.101)

where  A =  S . Wenote that 3 corresponds to the Sz = 0 component of the p-h spin
triplet, while + ≡ 1 + i 2 /2 and − ≡ 1 − i 2 /2 represent the Sz = 1 and
Sz = −1 components, respectively. We emphasise that

 S/A ≡  S/A i + iω k + q, −i − iω + i − k − q + Q; −i + i − k + Q, i k ,

do not depend anymore on the spin variables.

4.7 Irreducible Vertices and the Bethe-Salpeter Equations 173

We can similarly define spin-singlet and spin-triplet combinations of the irre-

p-h p-h
ducible vertex, respectively, 0S and 0 A , and note that, since R in (4.97) does not
depend on spin, the Bethe-Salpeter equations can be shortly written as
p-h
 p-h
↑,↑;↑,↑ = 0 ↑,↑;↑,↑ + 0 ↑,σ ;σ,↑  R  σ,↑;↑,σ
σ
p-h p-h p-h
= 0 ↑,↑;↑,↑ + 0 ↑,↑;↑,↑  R  ↑,↑,↑,↑ + 0 ↑,↓;↓,↑  R  ↓,↑;↑,↓ ,
p-h
 p-h
↑,↓;↓,↑ = 0 ↑,↓;↓,↑ + 0 ↑,σ ;σ,↑  R  σ,↓;↓,σ
σ
p-h p-h p-h
= 0 ↑,↓;↓,↑ + 0 ↑,↑;↑,↑  R  ↑,↓,↓,↑ + 0 ↑,↓;↓,↑  R  ↓,↓;↓,↓
p-h p-h p-h
= 0 ↑,↓;↓,↑ + 0 ↑,↑;↑,↑  R  ↑,↓,↓,↑ + 0 ↑,↓;↓,↑  R  ↑,↑;↑,↑ ,

where now the tensor product  does not include the spin labels. One can easily
show that
1 
p-h p-h
↑,↑;↑,↑ + ↑,↓;↓,↑ =  S = 0 S + 2 0 S  R   S ,
2
1  (4.102)
p-h p-h
↑,↑;↑,↑ − ↑,↓;↓,↑ =  A = 0 A + 2 0 A  R   A ,
2
namely that, unsurprisingly, the Bethe-Salpeter equations decouple into the two sym-
metry channels.
We can proceed further and, using the same notations, calculate the charge density-
density correlation function of Fig. 4.24 with the non-interacting vertices replaced
the charge-density ones, i.e., the identity, and find
       
χ= Tr R + Tr R  σ,σ  ;σ  ,σ  R = 2 Tr R + 4 Tr R   S  R .
σ σσ
(4.103)

Similarly, if we define the spin density operators as

 † 
σa (q) ≡ ckσ σ̂a σ σ  ck+qσ  , a = 1, 2, 3 ,
k σσ

the spin density-density correlation functions

 
χab (τ, q) ≡ − Tτ σa (τ, q) σb (0, −q)  = δab χσ (τ, q) , (4.104)

the last equation following from spin SU (2) symmetry, the Fourier transform in
Matsubara frequency can be readily calculated and reads
  
χσ = 2 Tr R + 4 Tr R   A  R . (4.105)
174 4 Feynman Diagram Technique

Fig. 4.33 Skeleton expansion up to second order of  in the particle-particle scattering channel

4.7.2 Particle-Particle Channel

In Fig. 4.33 we show the skeleton expansion up to second order of  in the particle-
particle scattering channel. Note that the diagrams are equal to those in Fig. 4.21,
but are simply drawn differently. Furthermore, one easily realises that  is odd
interchanging 3 with 4, or 1 with 2. That is simply consequence of the Pauli exclusion
principle, and leads to the following symmetry properties of the interaction vertex:

(1, 2; 3, 4) = (2, 1; 4, 3) = −(1, 2; 4, 3) = −(2, 1; 3, 4) . (4.106)

As before, we can distinguish diagrams reducible in the particle-particle channel,

namely that can be cut into two lower-order diagrams cutting two particle lines
with total frequency, momentum and spin equal to 1 + 2 = 3 + 4. Those are the
two diagrams highlighted in red. All other diagrams are irreducible in the particle-
p-p
particle channel. We define the sum of all order irreducible skeleton diagrams as 0 ,
through which  satisfies the following equation, graphically shown in Fig. 4.34,

0 (1, 2; 3 , 4 ) S(1 , 2 ; 3 , 4 ) (1 , 2 ; 3, 4)
p-p p-p
(1, 2; 3, 4) = 0 (1, 2; 3, 4) +
1 ,2 ,3 ,4
p-p p-p
≡ 0 + 0  S , (4.107)

with the meaning of  the same as before, and

S(1 , 2 ; 3 , 4 ) ≡ δ1 4 δ2 3 G(1 ) G(2 ) , (4.108)

 
where, since 1 + 2 = 1 + 2, if 1 = i  + i, k + Q, σ1 , then 2 = − i  , −k ,
σ2 , with σ1 + σ2 = σ1 + σ2 . Equation (4.107) is the Bethe-Salpeter equation in
the particle-particle channel. We emphasise that, even though  is the same, the
p-h p-p
irreducible vertices 0 and 0 are different.
4.7 Irreducible Vertices and the Bethe-Salpeter Equations 175

1 4 1 4 1 4

= Γp-p
0 + Γp-p
0

2 3 2 3 2 3

Fig. 4.34 Bethe-Salpeter equation for the interaction vertex  in the particle-particle channel

Also a Cooper pair can  be a spin-singlet or a spin-triplet. Accordingly, if we

redefine, as before, 1 ≡ i + iω, k + q without the spin quantum number, and
similarly for all other variables 2, 3, 4, and show the spins as subscripts, the interaction
vertex in the spin-singlet channel reads

1  
 S=0 (1, 2; 3, 4) = ↑,↓;↓,↑ 1, 2; 3, 4) − ↑,↓;↑,↓ 1, 2; 3, 4)
2  (4.109)
 
+ ↓,↑;↑,↓ 1, 2; 3, 4) − ↓,↑;↓,↑ 1, 2; 3, 4) ,

while those in the S = 1 channel with Sz = 0, ±1 as

1  
 S=1,Sz =0 (1, 2; 3, 4) = ↑,↓;↓,↑ 1, 2; 3, 4) + ↑,↓;↑,↓ 1, 2; 3, 4)
2 
 
+ ↓,↑;↑,↓ 1, 2; 3, 4) + ↓,↑;↓,↑ 1, 2; 3, 4) ,

 S=1,Sz =+1 (1, 2; 3, 4) = ↑,↑;↑,↑ 1, 2; 3, 4) ,

 S=1,Sz =−1 (1, 2; 3, 4) = ↓,↓;↓,↓ 1, 2; 3, 4) ,
(4.110)
and they must be all equal by spin SU (2) symmetry and different from
 S=0 (1, 2; 3, 4). We can similarly define the irreducible vertex in the spin-singlet
and spin-triplet channels, and, as before, the Bethe-Salpeter equations decouple into
different equations for each distinct channel. With this notation, (4.106) becomes

σ1 ,σ2 ;σ3 ,σ4 (1, 2; 3, 4) = σ2 ,σ1 ;σ4 ,σ3 (2, 1; 4, 3) = −σ1 ,σ2 ;σ4 ,σ3 (1, 2; 4, 3)
= −σ2 ,σ1 ;σ3 ,σ4 (2, 1; 3, 4) ,

and implies that  S=0 (1, 2; 3, 4) is even under 3 ↔ 4 or 1 ↔ 2, while  S=1,Sz

(1, 2; 3, 4) odd, as expected.

4.7.3 Self-energy and Irreducible Vertices

In Fig. 4.35 we show the same skeleton diagrams of the self-energy as in Fig. 4.13,
but rotated and with explicitly shown external legs. Each diagram is a functional of
the interacting Green’s function. Suppose we calculate the variation of the diagrams
with respect to a Green’s function variation. That amounts to consider an internal
Green’s function, label its initial and final points, e.g., as 3 and 2, respectively, cut the
176 4 Feynman Diagram Technique

1 1
1

Σ = +
4
4 4
1
1

+ +
4
4

Fig. 4.35 Skeleton diagrams of the self-energy up to second order rotated by 90◦ with respect to
those of Fig. 4.13, with the external legs explicitly indicated

line and multiply by δG(3, 2), i.e., the Green’s function variation. That procedure is
shown explicitly for two diagrams in Fig. 4.36. It is easy to realise that, repeating this
procedure for all diagrams and all internal Green’s functions, we obtain all skeleton
p-h
diagrams for the irreducible vertex 0 in the particle-hole channel, also shown in
Fig. 4.36 up to second order. We remark that, while in the self-energy diagrams 1
and 4 have same frequency, momentum and spin, and so 2 and 3 have, the variation
δG(3, 2) is generally off diagonal in those variables, and so the external legs 1 and
4 become. In other words, under the transformation G(3, 2) → G(3, 2) + δG(3, 2),
with

δG(3, 2) ≡ δG i  + iω, k + q, σ3 ; i  , k , σ2 ,

the self-energy ceases to be diagonal in frequency, momentum and spin,

(1, 4) ≡ δ1,4 (i, k) → (1, 4) + δ(1, 4)


≡ δ1,4 (i, k) + δ i + iω, k + q, σ1 ; i, k, σ4 ,

and thus

δ(1, 4) δ i + iω, k + q, σ1 ; i, k, σ4 p−h
=  ≡ 0 (1, 2; 3, 4)
δG(3, 2) δG i  + iω, k + q, σ3 ; i  , k , σ2
(i + iω k + q σ1 , i  k σ2 ; i  + iω k + q σ3 , i k σ4 .
p−h
= 0
(4.111)

We have shown such equality up to second order, but it is actually true at any order
in perturbation theory. Figure 4.37 is a graphical representation of the functional
derivative of the self-energy with respect to the Green’s function. We must imagine
to take any of the internal Green’s functions, extract out of the box that identifies
4.7 Irreducible Vertices and the Bethe-Salpeter Equations 177

1 3 1
3

δG(3, 2)

2 2
4 4
1
1
3 3

δG(3, 2)

2
2 4
4

1 3
Γp-h
0 = + - - - + + + -
4 2

Fig.4.36 Top panel: graphical representation of the functional derivative of the self-energy skeleton
diagrams with respect to the Green’s function. Bottom panel: skeleton diagrams of the irreducible
p-h
vertex 0 up to second order

Fig. 4.37 Graphical 1

representation of the 3
functional derivative of the
self-energy with respect to
the Green’s function

2
4

the self-energy and cut it. Evidently, what remains is a four leg vertex. However,
since self-energy insertions do not appear by definition in the skeleton diagrams,
that vertex must be irreducible in the particle-hole channel 1 − 4 = 3 − 2, and thus
p-h
is just 0 .
We mentioned that δG(3, 2) needs not to satisfy the symmetries of the Hamil-
tonian. Therefore, we can also imagine a variation of  with respect to a Green’s
function that does not conserve the electron number. For that we define the anomalous
Green’s functions
 
F (1, 2) ≡ − Tτ c(1) c(2)  = −F (2, 1)

= −θ (τ1 − τ2 )  ck1 σ1 (τ1 ) ck2 σ2 (τ2 )  + θ (τ2 − τ1 )  ck2 σ2 (τ2 ) ck1 σ1 (τ1 )  ,
 
F (1, 2) ≡ − Tτ c† (1) c† (2)  = −F (2, 1)

= −θ (τ1 − τ2 )  ck†1 σ1 (τ1 ) ck†2 σ2 (τ2 )  + θ (τ2 − τ1 )  ck†2 σ2 (τ2 ) ck†1 σ1 (τ1 ) 
 
= F (1, 2)∗ −  ck†1 σ1 (τ1 ) , ck†2 σ2 (τ2 )  , (4.112)

as opposed to the normal one G(1, 2), and the corresponding anomalous self-
energies, both shown in the top of Fig. 4.38. Those allow us to introduce a Green’s
178 4 Feynman Diagram Technique

1 2 1 2
F(1, 2) = F(1, 2) =
1 2 1 2
Δ(1, 2) = Δ(1, 2) =

a b a=4 & b=3

δF (a, b) = or
a=3 & b=4
1 1 1
1 1 a
a
b a
δΔ = + + +
b a
b
2 2
b
2 2 2
Fig. 4.38 Top panel: graphical representation of the anomalous Green’s functions F (3, 4) and
F (3, 4). Bottom panel: the skeleton diagrams up to second order of the self-energy variation obtained
by changing an internal Green’s function in the diagrams of Fig. 4.35 with δ F (a, b). That can be
done only in the first and third diagram. The internal vertices can be a = 4 and b = 3, or viceversa,
in which case the diagram gets an additional minus sign since δ F (3, 4) = −δ F (4, 3)

function matrix
     
− Tτ c(1) c† (2)  − Tτ c(1) c(2)  G(1, 2) F (1, 2)
Ĝ(1, 2) ≡ = ,
− Tτ c† (1) c† (2)  − Tτ c† (1) c(2)  F (1, 2) −G(2, 1)
(4.113)

and a self-energy matrix

 
ˆ (1, 2) (1, 2)
(1, 2) ≡ , (4.114)
(1, 2) −(2, 1)

functional of Ĝ, which are related to each other also through Dyson’s equation

Ĝ(1, 2) = Ĝ 0 (1, 2) + ˆ
Ĝ 0 (1, 3) (3, 4) Ĝ(4, 2) , (4.115)
3,4

where
   
G 0 (1, 2) 0 G 0 (τ1 − τ2 , k1 ) 0
Ĝ 0 (1, 2) = = δk1 ,k2 δσ1 ,σ2 ,
0 −G 0 (2, 1) 0 −G 0 (τ2 − τ1 , k1 )

is the non-interacting Green’s function.

4.7 Irreducible Vertices and the Bethe-Salpeter Equations 179

With those notations, we consider the variation of the anomalous self-energy

obtained upon replacing an internal normal Green’s function of the diagrams in
Fig. 4.35 with an anomalous Green’s function variation δ F (a, b) =  F (a, b), with
infinitesimal , and changing accordingly the directions of the normal Green’s func-
tions. The endpoints a and b can be either a = 4 and b = 3 or viceversa. Since
δ F (a, b) = −δ F (b, a), the two cases belong to the single variation with respect to
δ F (4, 3), though with opposite signs. Cutting in two δ F (a, b) we recover all dia-
grams in Fig. 4.33 irreducible in the p-p channel, i.e., all diagrams but the highlighted
two. The same holds at any order in the skeleton expansion. Therefore, similarly to
(4.111),

δ(1, 2) p-p
= 0 (1, 2; 3, 4) , (4.116)
δ F (4, 3) G

where the functional derivative on the left-hand-side must be evaluated with all
internal lines being normal, consistently with our original assumption that the global
charge U (1) symmetry is unbroken.
Since the irreducible vertex is unique, unlike the irreducible ones, hereafter we
p-h
shall only deal with 0 in the p-h channel. For that reason, we will refer to it simply
at 0 , without any superscript.

4.7.3.1 Proper Interaction Vertex

In the case in which the interaction is the Coulomb repulsion, we better use the
formalism described in Sect. 4.6. Therefore, given the self-energy (4.94), whose
diagrammatic representation is shown in Fig. 4.30, the irreducible interaction vertex
is defined through

δ δ xc
0 = = U (q) + .
δG δG
It is easy to realise that the second terms includes only skeleton diagrams that are
irreducible with respect to cutting not only a particle-hole pair of lines with fre-
quency and momentum transferred iω and q, respectively, but also an interaction
line. Therefore,

δ xc
≡ 0 ∗ , (4.117)
δG
defines the proper interaction vertex, which is related to the proper density-vertex
(4.93) as shown in Fig. 4.39.

Fig. 4.39 Bethe-Salpeter

equation relating the proper
interaction vertex 0 ∗ to the
ρ* = − ρ* Γ0 *
proper density-vertex ∗
180 4 Feynman Diagram Technique

4.8 The Luttinger-Ward Functional

The skeleton expansion shows that the self-energy [G, U ] is functional of the
Green’s function as well as of the specific form of interaction, in short U . We previ-
ously demonstrated that
δ(1, 4)
= 0 (1, 2; 3, 4) ,
δG(3, 2)
exists and is symmetric, i.e.,
δ(2, 3)
= 0 (2, 1; 4, 3) = 0 (1, 2; 3, 4) .
δG(4, 1)
Those conditions guarantee the existence of a functional [G, U ] such that

δ[G, U ] δ 2 [G, U ]
= (1, 4) , = 0 (1, 2; 3, 4) = 0 (2, 1; 4, 3) .
δG(4, 1) δG(4, 1) δG(3, 2)
(4.118)

This functional, known as Luttinger-Ward functional, has the simple expression in

perturbation theory
   1
[G, U ] = T ei η G(i, k)  (n) (i, k)

2n
kσ n≥1
  (4.119)
i η
≡ T e (i, k) ,
kσ 

where  (n) (i, k) includes all nth order skeleton diagrams, and satisfies
 
δ[G, U ] = T ei η (i, k) δG(i, k) . (4.120)
kσ 

The parameter η > 0 is infinitesimal and can be set equal to zero only after per-
forming the summation over Matsubara frequencies. The reason is that ei η regu-
larises that summation since (i, k) include terms that do not decay faster than
1/ for || → ∞. Indeed, while G(i, k) ∼ 1/i, Re  (n) (i, k) is generally finite
as || → ∞. Therefore the sum over  does not converge unless we explicitly take
ei η into account.4 In Fig. 4.40 we show the skeleton expansion up to second order

4 More rigorously,
   
T ei η (i, k) = T cos( η) (i, k) + (−i, k)
 >0
  
+i T sin( η) (i, k) − (−i, k) .
>0
4.8 The Luttinger-Ward Functional 181

1 1
Φ[G, U ] = − +
2 2

1 1
+ −
4 4

Fig. 4.40 Skeleton expansion up to second order of the Luttinger-Ward functional [G, U ]. Note
the pre-factors of each diagram

of the Luttinger-Ward functional, explicitly indicating sign and pre factor. We can
easily realise that, since 2n is the number of Green’s function of an nth order skele-
ton diagram, and the functional derivatives with respect to each of the 2n Green’s
functions yield topologically equivalent skeleton diagrams, that cancels the 1/2n and
reproduces the skeleton expansion of the self-energy.

4.8.1 Thermodynamic Potential

Suppose we multiply the interaction U by a parameter λ ∈ [0, 1], so that [G, U ] →

[G λ , λ U ], where G λ is the Green’s function parametrised by λ. Since the skeleton
diagrams that define λ(n) contain n interaction lines, and

∂λn n
= n λn−1 = λn ,
∂λ λ

The first term converges, since it involves either Re G Re  (n) or Im G Im  (n) , both of which are
even in , see (4.22) and (4.51), and decay at least as 1/ 2 , see (4.23) and (4.52). Therefore, in
that term we can safely set η = 0 even before performing the sum. For η → 0, the second term
is dominated by the odd function Im G Re  (n) , which decays as 1/ as || → ∞, see (4.23) and
(4.52), so that
     
1 sin( η)
iT sin( η) (i, k) − (−i, k) −−→ 2 Re  (n) (i∞, k) T ,
η0 2n 
>0 n≥0 >0

where

 sin( η)  sin (2k + 1)π T η 1
T = = ,
 (2k + 1) π 4
>0 k≥0

is finite, and independent of η, only if the sum is done before sending η → 0.

182 4 Feynman Diagram Technique

then
∂[G λ , λ U ] 1   i η 
= T e G λ (i, k) λ(n) (i, k)
∂λ Gλ 2λ 
kσ n≥1
(4.121)
1   i η
= T e G λ (i, k) λ (i, k) ,
2λ  kσ

where the derivative is done at fixed G λ . Through (4.80),

 
1  ∂ T   −iτ
− − k G λ (τ, k) = e G 0 (i, k)−1 G λ (i, k)
2 ∂τ 2 
kσ kσ
δ(τ ) λ   
† † +
= + U (q) Tτ ckσ (0) cp+qσ  (τ + 0 ) cpσ  (τ ) ck+qσ (τ ) ,
2 2V 
kpq σ σ

so that, for τ = −η → −0+ ,

T   i η T   i η  
e G 0 (i, k)−1 G λ (i, k) = e G λ (i, k)−1 + λ (i, k) G λ (i, k)
2  2 
kσ kσ
T   i η
= e λ (i, k) G λ (i, k)
2 
kσ
λ  † †
= U (q)  ckσ cp+qσ  cpσ  ck+qσ  ≡  Hint (λ)  .
2V  kpq σ σ

Comparing the above equation with (4.121) and using the Hellman-Feynman theorem
we conclude that
∂[G λ , λ U ] 1   i η 1
= T e G λ (i, k) λ (i, k) =  Hint (λ) 
∂λ Gλ 2λ 
λ
kσ
 ∂ H (λ)  ∂ F(λ)
= = , (4.122)
∂λ ∂λ
where F(λ) is the free-energy of the interacting system as fixed λ.
We define a functional
 
    iη  
F Gλ, λ U ≡ T e ln G λ (i, k) − G 0 (i, k)−1 G λ (i, k) +  G λ , λ U .
kσ 

If G λ is the solution of Dyson’s equation

G λ (i, k)−1 = G 0 (i, k)−1 − λ (i, k) ,

then
 
∂ F G λ, λ U δ
= G −1 −1
λ − G0 + = G −1 −1
λ − G 0 + λ = 0 .
∂G λ λU δG λ
4.8 The Luttinger-Ward Functional 183

In other
 words,
 the Green’s function solution of Dyson’s equation is the saddle point
of F G, U .
It follows that
   
d F Gλ, λ U ∂ F Gλ, λ U ∂G λ ∂[G λ , λ U ] ∂[G λ , λ U ] ∂ F(λ)
= + = = ,
dλ ∂G λ λU ∂λ ∂λ Gλ ∂λ Gλ ∂λ

and thus, upon integration over λ from 0 to 1,

   
F G, U − F G 0 , 0 = F(1) − F(0) = F − F0 , (4.123)

where G = G λ=1 is the interacting Green’s function, F(λ = 1) = F the interacting

free-energy, and F(λ = 0) = F0 the non-interacting one. Since5
    iη
F G0, 0 ≡ T e ln G 0 (i, k) = F0 ,
kσ 

we finally obtain that the free-energy of the interacting system

 
    iη  
F = F G, U = T e ln G(i, k) − G 0 (i, k)−1 G(i, k) +  G, U , (4.124)
kσ 

with G solution of the Dyson equation

δ[G, U ]
G −1 = G −1
0 − .
δG

5 The free-energy of the non-interacting system is

    
F0 = −T ln Z 0 = −T ln 1 + e−βk = −T ln 1 + e−βk .
kσ kσ

On the other hand, using the standard tricks to perform the sum over Matsubara frequencies, and
since
 
ln G 0 ( ± i0+ , k) = − ln  − k ± i0+ = − ln  − k ∓ i π θ k − ) .

   ∞
d G 0 ( + i0+ , k)   k
T eiη ln G 0 (i, k) = − f () eη ln + = d f () eη
kσ  kσ −∞ 2πi G 0 ( − i0 , k) kσ −∞
  k     
= −T d ln 1 + e−β eη = −T ln 1 + e−βk = F0 ,
kσ −∞ kσ

which is desired result.

184 4 Feynman Diagram Technique

Fig. 4.41 Graphical τ1

representation of (4.127) τ1
k + q, σ1
τ, q
ρQ
k, σ2
τ2
τ2

4.9 Ward-Takahashi Identities

Let us suppose that our model has a set of conserved quantities


Q = dr ρ Q (r) ,

with the Fourier transform of the density having the general expression
 †
ρ Q (q) = ckσ 0Q (k σ, k + q σ  ) ck+qσ  . (4.125)
k σσ

The conservation of Q entails the existence of a continuity equation

∂ρ Q (r)   ∂ρ Q (q)  
− = ρ Q (r) , H ≡ −i ∇ · J Q (r) , − = ρ Q (q) , H ≡ q · J Q (q) ,
∂τ ∂τ
(4.126)
which defines the corresponding current J Q (r) or its Fourier transform J Q (q).
Let us consider the following expectation value, graphically shown in Fig. 4.41,
 
†
−  Tτ ρ Q (τ, q) ck+qσ 1
(τ 1 ) ckσ2 (τ 2 ) 


= dτ1 dτ2  Q τ2 k σ2 , τ1 k + q σ1 ; τ, q G(τ2 − τ2 , k) G(τ1 − τ1 , k + q) ,
(4.127)

where  Q is the interacting Q-density vertex, and apply −∂/∂τ . The derivative acts
either directly on the charge density operator, in which case we can use the continuity
equation (4.126), or on the θ -functions that define the time-ordered product. The latter
is, dropping for simplicity all unnecessary indices,
   
ρ Q (τ ) c† (τ1 ) c(τ2 ) ∂τ θ(τ − τ1 ) θ(τ1 − τ2 ) + c† (τ1 ) ρ Q (τ ) c(τ2 ) ∂τ θ(τ1 − τ ) θ(τ − τ2 )
   
+ c† (τ1 ) c(τ2 ) ρ Q (τ ) ∂τ θ(τ1 − τ2 ) θ(τ2 − τ ) − ρ Q (τ ) c(τ2 ) c† (τ1 ) ∂τ θ(τ − τ2 ) θ(τ2 − τ1 )
   
− c(τ2 ) ρ Q (τ ) c† (τ1 ) ∂τ θ(τ2 − τ ) θ(τ − τ1 ) − c(τ2 ) c† (τ1 ) ρ Q (τ ) ∂τ θ(τ2 − τ1 ) θ(τ1 − τ )
   
= δ(τ − τ1 ) θ(τ − τ2 )  ρ Q (τ ) , c† (τ ) c(τ2 ) + θ(τ1 − τ ) δ(τ − τ2 ) c† (τ1 ) ρ Q (τ ) , c(τ ) 
   
− δ(τ − τ2 ) θ(τ − τ1 )  ρ Q (τ ) , c(τ ) c† (τ1 ) − θ(τ2 − τ ) δ(τ − τ1 ) c(τ2 ) ρ Q (τ ) , c† (τ ) 
     
= δ(τ − τ1 )  Tτ ρ Q (τ ) , c† (τ ) c(τ2 )  + δ(τ − τ2 ) Tτ c† (τ1 ) ρ Q (τ ) , c(τ )  .
4.9 Ward-Takahashi Identities 185

It follows, through (4.126), that


∂ 
− dτ1 dτ2  Q τ2 k σ2 , τ1 k + q σ1 ; τ, q G(τ2 − τ2 , k) G(τ1 − τ1 , k + q)
∂τ


= q · dτ1 dτ2  Q τ2 k σ2 , τ1 k + q σ1 ; τ, q G(τ2 − τ2 , k) G(τ1 − τ1 , k + q)
  
†
+ δ(τ − τ1 )  Tτ ρ Q (τ, q) , ck+qσ 1
(τ ) c kσ2 (τ2 ) 
  
†
+ δ(τ − τ2 ) Tτ ck+qσ 1
(τ1 ) ρ Q (τ, q) , c kσ2 (τ ) , (4.128)

where  Q is the interacting Q-current density vertex. One can readily find that the
equal-time commutators
   
†
ρ Q (τ, q) , ck+qσ 1
(τ ) = †
ckσ (τ ) 0Q k σ, k + q σ1 ,
σ
   Q
ρ Q (τ, q) , ckσ2 (τ ) = − 0 k σ2 , k + q σ ck+qσ (τ ) ,
σ

so that (4.128) becomes


∂ 
− dτ1 dτ2  Q τ2 k σ2 , τ1 k + q σ1 ; τ, q G(τ2 − τ2 , k) G(τ1 − τ1 , k + q)
∂τ


= q · dτ1 dτ2  Q τ2 k σ2 , τ1 k + q σ1 ; τ, q G(τ2 − τ2 , k) G(τ1 − τ1 , k + q)
 
+ δ(τ − τ1 ) G τ2 − τ, k) 0Q k σ2 , k + q σ1
Q 
− δ(τ − τ2 ) 0 k σ2 , k + q σ1 G τ − τ1 , k + q ,

or, in Matsubara frequencies,


iω  Q i k σ2 , i + iω k + q σ1 ; iω, q G(i, k) G(i + iω, k + q)

= q ·  Q i k σ2 , i + iω k + q σ1 ; iω, q G(i, k) G(i + iω, k + q)
  
− 0Q k σ2 , k + q σ1 G(i + iω, k + q) − G(i, k) .
(4.129)
Dividing both sides by the product of the two Green’s function, (4.129) is also
equivalent to
 
iω  Q i k σ2 , i+iω k+q σ1 ; iω, q −q ·  Q i k σ2 , i + iω k + q σ1 ; iω, q
  
= 0Q k σ2 , k + q σ1 G(i + iω, k + q)−1 − G(i, k)−1
  
= 0Q k σ2 , k + q σ1 G 0 (i + iω, k + q)−1 − G 0 (i, k)−1
 
Q
− 0 k σ2 , k + q σ1 (i + iω, k + q) − (i, k) .
(4.130)
186 4 Feynman Diagram Technique

i + iω, k + q, σ1
Σ
iω ρQ −q· JQ = iω ρQ −q· JQ + ρQ − ρQ

i , k, σ2 Σ

Fig. 4.42 Graphical representation of the Ward-Takahashi identity (4.131). The darker triangles
with bold contours are fully interacting density vertices, while the other non-interacting ones. The
incoming and outgoing arrow lines just indicate the external vertices. All incoming lines carry the
same frequency i + iω, momentum k + q, and spin σ1 , the reason why we show them only in the
first vertex. Similarly, all outgoing lines carry the same frequency i, momentum k, and spin σ2 ,
only shown in the first vertex. The self-energies, red ellipses, carry the same frequency, momentum
and spin of the line they are attached to

Q Q
In absence of interaction,  Q = 0 ,  Q = 0 ,  = 0, so that

Q Q
iω 0 k σ2 , k + q σ1 − q · 0 k σ2 , k + q σ1
  
= 0Q k σ2 , k + q σ1 G 0 (i + iω, k + q)−1 − G 0 (i, k)−1 ,

which allows us to rewrite (4.130) as

 
iω  Q i k σ2 , i + iω k+q σ1 ; iω, q − q ·  Q i k σ2 , i + iω k + q σ1 ; iω, q
Q Q
= iω 0 k σ2 , k + q σ1 − q · 0 k σ2 , k + q σ1
  
− 0Q k σ2 , k + q σ1 (i + iω, k + q) − (i, k) ,
(4.131)

which is the Ward-Takahashi identity, graphically shown in Fig. 4.42.

Let us study the consequences the Ward-Takahashi identity considering for sim-
plicity the case Q = N , the number of electrons, so that ρ Q = ρ the charge density,
0Q the identity, and 0Q = ∂k /∂k for small q times the identity in spin. In this
case, the interacting density-vertices are also diagonal in spin, and (4.131) reads, for
small ω and q,
 
iω  i k σ, i + iω k + q σ ; iω, q − q ·  i k σ, i + iω k + q σ ; iω, q
 
 iω  i k σ, i k σ ; iω, q − q ·  i k σ, i k σ ; iω, q
   
∂k
= iω − q · − (i + iω, k + q) − (i, k)
∂k
 
∂(i, k) ∂  
 iω 1 − −q· k + (i, k) ,
∂i ∂k
(4.132)
4.9 Ward-Takahashi Identities 187

It is common to denote the ω-limit as the limit sending first q = |q| to zero, and next
ω to zero. The reverse case is accordingly denoted as q-limit. We thus find
  ∂(i, k)
ω i, k, σ ) = lim lim  i k σ, i + iω k + q σ ; iω, q = 1 − ,
ω→0 q→0 ∂i
  ∂  
q i, k, σ ) = lim lim  i k σ, i + iω k + q σ ; iω, q = k + (i, k) .
ω→0 q→0 ∂k
(4.133)

Those equations play an important role in Landau’s Fermi liquid theory, as we shall
see.

4.9.1 Ward-Takahashi Identity for the Heat Density

The Hamiltonian H is trivially associate to a conserved quantity, the total energy E.

However, H is not a one-body operator, and therefore the previous derivation of the
Ward-Takahashi identity cannot be used in this case.
Let us assume that the Hamiltonian is local, namely can be written as

H = dr ρ E (r) .

It follows that, by definition,

∂    
σ (r)
− = σ (r) , H = σ (r) , ρ E (r) ,
∂τ
∂    
σ (r)
†
− = σ (r) ,
†
H = σ (r) ,
†
ρ E (r) ,
∂τ
and, therefore,
  ∂  
σ (r) ∂ σ (r)
†

ρ E (r) , σ (r ) = δ(r − r ) , ρ E (r) , † 
σ (r ) = δ(r − r ) .
∂τ ∂τ
(4.134)

Those equations can actually be used to find the heat density operator ρ E (r). Since
the total energy is conserved, there exists a continuity equation

∂ρ E (r)
− + i ∇ · J E (r) = 0 ,
∂τ
with J E (r) a multi-particle operator just like ρ E (r) is. We consider the same function
as in (4.127), but now in real space and imaginary time,
 
I E (x; y, z) = − Tτ ρ E (x) σ† (y) σ (z)  , (4.135)
188 4 Feynman Diagram Technique

where x = (τ, r) ≡ (x0 , x), and similarly for y and z. Proceeding as before, and
using (4.134) we find

∂  
− I E (x; y, z) = − Tτ − i ∇ · J E (x) σ† (y) σ (z) 
∂ x0
  
  †
− δ x0 − y0 Tτ σ (z) ρ E (x) , σ (y)
  
  
− δ x0 − z 0 Tτ ρ E (x) , σ (z) σ† (y)
 
   ∂ σ† (x) 
= −i ∇ · I E (x; y, z) − δ x0 − y0 δ x − y Tτ σ (z)
∂ x0
 
   ∂ σ (x) † 
− δ x0 − z 0 δ x − z Tτ σ (y)
∂ x0
∂ ∂
= −i ∇ · I E (x; y, z) + δ(x − y) G(z, y) + δ(x − z) G(z, y) ,
∂ y0 ∂z 0

since the imaginary-time derivatives acting on the θ -functions that define G cancels
out, and thus

∂ ∂ ∂
− I E (x; y, z) + i ∇ · I E (x; y, z) = δ(x − y) G(z, y) + δ(x − z) G(z, y) .
∂ x0 ∂ y0 ∂z 0

Introducing the heat and heat-current density-vertices, the above equation becomes
in frequency and momentum space
 
iω  E (i k σ, i + iω k + q σ ; iω, q) − q ·  E (i k σ, i + iω k + q σ ; iω, q)
G(i, k) G(i + iω, k + q)

= i G(i, k) − i  + ω G(i + iω, k + q) ,

which, proceeding as before, leads to

iω  E (i k σ, i + iω k + q σ ; iω, q) − q ·  E (i k σ, i + iω k + q σ ; iω, q)


= i G(i + iω, k + q)−1 − i + iω G(i, k)−1

= −i (k + q) − (k) + iω (k)
 
− i (i + iω, k + q) − (i, k) + iω (i, k) . (4.136)

Equation (4.136) is the Ward-Takahashi identity for the heat density. Expanding for
small ω and q, we get

iω  E (i k σ, i k σ ; iω q) − q ·  E (i k σ, i k σ ; iω q)
∂(i, k) ∂k ∂(i, k)
= iω (k) + (i, k) − i − i q · + ,
∂i ∂k ∂k
4.10 Conserving Approximation Schemes 189

so that
 ω ∂(i, k)
 E (i, k, σ ) = (k) + (i, k) − i ,
∂i
 q ∂k ∂(i, k)
(4.137)
 E (i, k, σ ) = i + .
∂k ∂k

Like (4.133), those equations are important in Landau’s Fermi liquid theory.
In particular, (4.137) allows calculating the interacting specific heat. We note that
the internal energy is defined through

∂ ln Z (β) ∂  
U (β) = − = β F(β) .
∂β ∂β

Suppose we change H → λ H , then it follows that Z (β) → Z (λβ), and thus

δ F(λβ) δ ln Z T ∂  
= −T =− Tr e−λβ H = U (λβ) .
δλ δλ Z ∂λ
The specific heat at fixed λ is defined through

∂U (λβ) ∂U (λβ) ∂U (λβ) ∂U (λβ) ∂ 2 F(λβ)

cV = = −β 2 = −β 2 = −λ β = −λ β ,
∂T ∂β ∂β ∂λ ∂λ2

so that the actual value at λ = 1 reads

1 ∂ 2 F(λβ)
cV = − .
T ∂λ2 |λ=1

The second derivative of the free energy with respect to λ is just the q-limit of the
heat density-heat density response function χ E (iω, q), so that

1
cV = − lim lim χ E (0, q) . (4.138)
T q→0 ω→0

We shall use both (4.137) and (4.138) to calculate the specific heat of a Landau’s
Fermi liquid.

4.10 Conserving Approximation Schemes

We already mentioned that, since the exact self-energy functional [G, U ] is

generally unknown, the common approach is to use an approximate expression
appr ox [G, U ], and then solve the self-consistency Dyson’s equation

G −1 = G −1
0 − appr ox [G, U ] ,
190 4 Feynman Diagram Technique

to get the corresponding approximation G appr ox of the actual Green’s function. The
question we here address is how to extend that approximation scheme to calculate
linear response functions without spoiling conservation laws.
Let us go back to (4.131), which, in shorthand notations, can be written as
Q Q Q Q
iω  Q − q ·  Q = iω 0 − q · 0 − 0   +   0 , (4.139)

while (4.129) as, see (4.97),

iω  Q  R − q  Q  R = −0Q  G + G  0Q . (4.140)

In this representation, both equations also account for the case in which there are
additional quantum numbers, e.g., band, orbital or reciprocal lattice vector indices,
and the Green’s functions becomes matrices with finite off-diagonal elements. By
definition, see Fig. 4.25, and using the Bethe-Salpeter equation in the particle-hole
channel, see Fig. 4.32,
Q Q Q Q Q
 Q ≡ 0 + 0  R   = 0 + 0  R  0 + 0  R    R  0
= 0Q +  Q  R  0 ,
(4.141)
and similarly for the current density-vertex, which are just the Bethe-Salpeter equa-
tion for the density-vertices. Through (4.141) we can rewrite (4.139) as
   
Q Q Q Q
iω 0 +  Q  R  0 − q · 0 +  Q  R  0 = iω 0 − q · 0
Q Q
− 0   +    0 ,

namely, through (4.141),

   
iω  Q  R − q ·  Q  R  0 = − 0Q  G + G  0Q  0
Q Q
= −0   +   0 .

That equation corresponds analytically to

1    Q     
T 0 k σ2 , k + q σ1 G(i  + iω, k + q) − G(i  , k )
V   
k σ1 σ2 

0 i  + iω k + q σ1 , i k σ2 ; i + iω k + q σ1 , i  k σ2
  
= 0Q k σ2 , k + q σ1 (i + iω, k + q) − (i, k) . (4.142)

If we regard
 
Q 
0 k σ2 , k + q σ1 G(i  + iω, k + q) − G(i  , k ) ≡ δG i  + iω k + q σ1 , i  k σ2 ,
 
Q 
0 k σ2 , k + q σ1 (i + iω, k + q) − (i, k) ≡ δ i + iω k + q σ1 , i k σ2 ,
4.10 Conserving Approximation Schemes 191

as variations of G and , which, as we mentioned, are generally non diagonal in

frequency, momentum and spin, then (4.142)

1    
T δG i  + iω k + q σ1 , i  k σ2
V   
k σ1 σ2 

0 i  + iω k + q σ1 , i k σ2 ; i + iω k + q σ1 , i  k σ2

= δ i + iω k + q σ1 , i k σ2 ,

is just the definition of 0 as functional derivative of  with respect to G. In turn,

that result implies that an irreducible vertex obtained as the functional derivative of
 with respect to G preserves the Ward-Takahashi identities, hence does not spoil
conservation laws.
Therefore, if we approximate [G, U ] with appr ox [G, U ], then the irreducible
vertex is accordingly

δappr ox [G, U ]
0 appr ox = . (4.143)
δG
A conserving approximation scheme consistent with all conservation laws consists
in solving the Bethe-Salpeter for the reducible vertex

 = 0 + 0  R   ,

with 0 = 0 appr ox and R = Rappr ox = G appr ox G appr ox , and thus obtain an

approximation appr ox for the interaction vertex. A generic response function must
be calculated through
   
χ AB = Tr 0A  Rappr ox  0B + Tr 0A  Rappr ox  appr ox  Rappr ox  0B . (4.144)

In that way, we are guaranteed that the linear response theory is consistent with all
conservation laws.

4.10.1 Conserving Hartree-Fock Approximation

As a first example, we shall consider the Hartree-Fock approximation within the

diagrammatic technique.
For that, we consider an interacting Hamiltonian
 1 
H= tab ca† cb + Uacdb ca† cc† cd cb ,
2
ab abcd
192 4 Feynman Diagram Technique

Fig. 4.43 Self-energy

functional in the
Hartree-Fock approximation d c

Σab [G] = a b + a d c b

written in a generic single-particle basis of wavefunctions. Let us approximate the

skeleton expansion of the self-energy with the two first-order diagrams, as shown in
Fig. 4.43. That amounts to the following the self-energy functional
  +   +
ab = Uacdb T ei 0 G dc (i) − Uacbd T ei 0 G dc (i) ,
cd  cd 

where
 +
T ei 0 G dc (i) =  cc† cd  ≡ cd . (4.145)

The Green’s function is a matrix and satisfies the Dyson equation
   
Ĝ(i)−1 = i − tab − ab = i − tab − Uacdb − Uacbd cd ,
ab
cd

which is diagonalised by diagonalising

  
tab + cd (Uacdb − Uacbd ) = Ĥ H F ,
ab
cd

where Ĥ H F is nothing but the matrix representation of the Hartree-Fock Hamil-

tonian. Namely, the self-consistency requirement (4.145) is just the Hartree-Fock
self-consistency equation.
In order for the Hartree-Fock approximation to be a consistent scheme we need
to approximate 0 as

δab
0 (a, c; d, b) = = Uacdb − Uacbd ,
δG dc

and use it to solve the Bethe-Salpeter equation for the reducible vertex , see
Fig. 4.44. One can readily verify that the correlation functions calculated with the
above  coincide with the time-dependent Hartree-Fock, which is therefore a con-
sistent approximation.
4.10 Conserving Approximation Schemes 193

a d a d
Γ0 (a, c; d, b) = + -
b c b c

a d a d
Γ(a, c; d, b; iω, q) = + -
b c b c

a d a d
+ -
b b c
c

Fig. 4.44 Top panel: irreducible vertex 0 in the particle-hole channel within the Hartree-Fock
approximation. Bottom panel: Bethe-Salpeter equation for the reducible interaction vertex  in
the Hartree-Fock approximation. In this approximation the interaction vertex only depends on the
frequency iω and momentum q transferred, namely on the frequency and momentum difference
between the two internal lines

Fig. 4.45 Top panel: W (iω, q)

self-energy functional in the
GW approximation. Bottom
i + iω, k + q
panel: Dyson’s equation for
the screened interaction
ΣGW (i , k) = -
W (q)

W (iω, q) U (q)
= +

4.10.2 Conserving GW Approximation

Figure 4.30 is an exact representation of the self-energy in terms of the Green’s

function, the screened Coulomb interaction W (iω, q), and the proper charge density-
vertex ∗ . An approximation that allows solving the self-consistent Dyson equation
and yet taking into account the screening of the Coulomb interaction that is missing
in the above Hartree-Fock approximation, is to assume consistently ∗ = 0 = 1.
This is the so-called GW approximation, graphically shown in Fig. 4.45.
In the GW approximation one has to solve two coupled equations, since W (iω, q)
is also functional of G, specifically

U (q) U (q)  
W (iω, q) = ,  (iω, q) = 1 − T G(i + iω, k + q) G(i, k) .
 (iω, q) V 
kσ

In order to make the GW approximation a conserving scheme, the irreducible vertex

0 in the particle-hole channel must be defined through the functional derivative of
GW with respect to G, which we show in Fig. 4.46, and used to solve the Bethe-
Salpeter equations and calculate linear response functions.
194 4 Feynman Diagram Technique

1 3

0 =+ - + +
4 2
Fig. 4.46 Irreducible vertex 0 in the GW approximation. Note that the first contribution comes
from the Hartree diagram in the self-energy, which does exists although it is cancelled by the
interaction with the positive ions

4.11 Luttinger’s Theorem

We end this chapter presenting the so-called Luttinger theorem [4], which is often
invoked in the context of strongly-correlated materials.
Luttinger’s theorem refers to any conserved quantity, though it is commonly dis-
cussed for the total electron number and assuming translational symmetry. Since
the latter symmetry is not necessary, we here present a derivation [5] that does not
assume it since we shall later use Luttinger’s theorem in the context of Anderson
impurity models, see Chap. 7, which lack translational invariance.
We consider a generic system of interacting electrons with annihilation operators
cα corresponding to a complete basis of single-particle wavefunctions labelled by
α = 1, . . . , K , with K → ∞ in the thermodynamic limit. The Hamiltonian admits a
set of conserved quantities Q, represented by hermitian matrices Q̂ with components
Q αβ defined in such a way that the eigenvalues are integers. Q αβ = δαβ corresponds
to the total number N of electrons, while all other independent Q’s are represented by
traceless matrices Q̂. To avoid any issue related to the discontinuity at zero imaginary
time of the Green’s functions, we use instead of N the deviation N − K /2 of the
electron number with respect to half-filling, so that we can write the expectation
value of any conserved quantity as
1    1    
Q= Q βα  cβ† cα  −  cα cβ†  = Tr Ĝ(τ = 0− ) Q̂ + Tr Ĝ(τ = 0+ ) Q̂
2 2
αβ
       
T + + 
= ein 0 + e−in 0 Tr Ĝ(in ) Q̂ = T cos n 0+ Tr Ĝ(in ) Q̂
2 n n
  
=T Tr Ĝ(in ) Q̂ , (4.146)
n

where Ĝ(in ) = Ĝ(−in )† is the Green’s function matrix in the Matsubara frequen-
cies n = (2n + 1) π T , and the last equality is valid since the sum involves only the
 trace even in n , and since that decays at least as 1/n we can
component of the 2

safely take cos n 0+ = 1.

According to Dyson’s equation,

Ĝ −1 (in ) = in Iˆ − Ĥ0 − (i

ˆ n) , (4.147)

with Iˆ the identity matrix, and Ĥ0 the non-interacting Hamiltonian, including the
chemical potential term, represented in the chosen basis. (i ˆ
ˆ n ) = (−i n ) is the
†
4.11 Luttinger’s Theorem 195

self-energy matrix that accounts for all interaction effects. We can thus write (4.146)
as
 ∂  
Q = −T Tr ln Ĝ(in ) Q̂ + I L (Q) , (4.148)
n
∂in

where
  
ˆ n)
∂ (i
I L (Q) = T Tr Ĝ(in ) Q̂ , (4.149)
n
∂in

is the Luttinger integral for the conserved quantity Q. Hereafter, we use simply I L
for the case Q̂ = Iˆ, i.e., for the number of electrons.
We note that at particle-hole symmetry6 I L (Q) vanishes identically for all Q’s
that are odd under a particle-hole transformation, thus also the electron number
ˆ n ) = 0.
operator. Seemingly, I L (Q) = 0 in absence of interaction, where (i
Let us instead analyse I L (Q) in more general circumstances. By definition, if
[G] is the Luttinger-Ward functional, then, see (4.120) and (4.119),
  
δ[G] = T ˆ n ) δ Ĝ(in ) ,
ein η Tr (i (4.150)
n

with η > 0 that must be sent to zero after performing the summation, and where
  1   
[G] = T ein η ˆ (m) (in ) ≡ T
Tr Ĝ(in )  ein η (in ) ,
n
2m n
m≥1
(4.151)

6 The Hamiltonian has particle-hole symmetry if there is a one-to-one correspondence between the
fermionic operators with quantum number, e.g., α = 1, . . . , K /2 and those with quantum number
α + K /2, assuming even K , such that the Hamiltonian remains invariant under the transformation
†
P := cα ↔ cα+K /2 . It follows that, for generic α, β = 1, . . . , K defined modulus K , i.e.,  + K ≡
 for 1 ≤  ≤ K ,
   
G αβ (τ − τ  ) = − T cα (τ ) cβ† (τ  )  − †
→ − T cα+K  
/2 (τ ) cβ+K /2 (τ )  = −G β+K /2 α+K /2 (τ − τ ) ,

so that G αβ (i) = −G β+K /2 α+K /2 (−i). Similarly, αβ (i) = −β+K /2 α+K /2 (−i). There-
fore, I L (Q) in (4.149), for instance with Q̂ = Iˆ, transforms as

 ∂βα (in )  ∂α+K /2 β+K /2 (−in )

IL = T G αβ (i) −
→T G β+K /2 α+K /2 (−i) = −I L ,
∂in ∂in
n αβ n αβ

thus I L = 0. One can readily prove that the same holds for all Q’s odd under particle-hole transfor-
mation, namely such that Q αβ = Q β−K /2 α−K /2 , but not for those even, Q αβ = −Q β−K /2 α−K /2 .
196 4 Feynman Diagram Technique

where ˆ (m) (in ) is the sum of all mth order skeleton diagrams. Through (4.150) and
(4.151) it readily follows that
 
δ[G]  ∂ Ĝ(in )  ∂(in )
≡T ˆ
Tr (in ) =T , (4.152)
δi n
∂in n
∂in

where we set η = 0 before performing the sum since the function decays faster than
1/n for n → ±∞. Equation (4.152) allows us to rewrite I L (Q) of (4.149) for
Q̂ = Iˆ simply as
 ∂ I L (in )
IL = T , (4.153)
n
∂in

where
 
ˆ n ) Ĝ(in ) − (in ) .
I L (in ) = Tr (i (4.154)

In other words, it is always possible to represent the Luttinger integral as the sum-
mation over Matsubara frequencies of the derivative of a function. It follows that the
total number of electrons can be written as7

K  ∂     ˆ n)
∂ (i

N= −T Tr ln Ĝ(in ) + T Tr Ĝ(in )
2 n
∂in n
∂in
K  ∂    ∂ I L (in )
= −T Tr ln Ĝ(in ) + T
2 n
∂in n
∂in
 ∞    ∞ d ∂ I (i)
K d ∂ L
−−−→ − Tr ln Ĝ(i) + .
T →0 2 −∞ 2π ∂i −∞ 2π ∂i

Since Ĝ(−i) = Ĝ(i)† and, similarly, I L (−i) = I L (i)∗ , if we define, through

the polar decomposition of Ĝ(i), the matrix
 
δ̂() ≡ arg Ĝ(i) = Im ln Ĝ(i) , (4.155)

then, for T → 0,
 ∞    ∞ d ∂Im I (i)
K d ∂ L
N= − Tr δ̂() + .
2 0+ π ∂ 0 + π ∂

7 Since n = (2n + 1)π T , it becomes a continuous variable for T → 0 such that

  ∞
d
T F(in ) −−−→ F(i) .
T →0 −∞ 2π
n
4.11 Luttinger’s Theorem 197

Since Ĝ(i) → Iˆ/i for  → ∞, then

π   π
δ̂() −−−→ − Iˆ , Tr δ̂() −−−→ − K ,
→∞ 2 →∞ 2
while, through its definition (4.154), we realise that Im I L (i) → 0 for  → ∞, so
that we finally find that
 ∞
K d  
N= + Tr Ĝ(i)
2 −∞ 2π
 ∞    ∞ d ∂ I (i)
K d ∂ L
= − Tr ln Ĝ(i) + (4.156)
2 −∞ 2π ∂i −∞ 2π ∂i
1   1
= K + Tr δ̂(0+ ) − Im I L (i0+ ) .
π π
This expression is exact. It is still not Luttinger’s theorem but a kind of generalisation
of it, and it is remarkable as it shows that a quantity requiring integration over all
frequencies can be alternatively calculated by just knowing the Green’s function at
the single point  = 0+ , namely, just a boundary term. In reality, Luttinger’s theorem
statement is that Im I L (i0+ ) = 0 in (4.156), which is not to be expected a priori.
Hereafter, we analyse in detail the proof of that theorem, and highlight under
which conditions it is valid, and what are its physical consequences.

4.11.1 Validity Conditions for Luttinger’s Theorem

The Luttinger-Ward functional [G] is invariant if the Matsubara frequency of each

internal Green’s function is replaced, see (4.147), by in Iˆ + iω Q̂ for any conserved
Q, where ω = 2π T is the first bosonic Matsubara frequency.8 Therefore,

 Q [G]    Q Ĝ(i)

0= =T ˆ n)
Tr (i ,
iω n
iω

8 Suppose we change the interacting Hamiltonian H = H + H into the non-hermitian one H −

 0 int
iω Q = H0 − iω Q + Hint . It follows that
 β  β
S(β, 0) = dτ e H0 τ Hint e−H0 τ → dτ e(H0 −iω0 Q)τ Hint e−(H0 −iω0 Q)τ = S(β, 0) ,
0 0

is invariant since both H0 and Hint commute with Q. Therefore also  Hint  =
 S(β, 0) Hint / S(β, 0)  is invariant under H0 → H0 − iω0 Q. Through (4.122), we conclude
that the Luttinger-Ward functional
 1
dλ
[G] =  Hint (λ) 
0 λ
is also invariant under H0 → H0 − iω0 Q, which however transforms (4.147) into

Ĝ −1 (in ) = in Iˆ − Ĥ0 + i ω Q̂ − (i
ˆ n ) = Ĝ −1 in + i ω Q̂ .

Therefore, [G] must be invariant under such Green’s function transformation.

198 4 Feynman Diagram Technique

with

 Q Ĝ(i) Ĝ(in + iω Q̂) − Ĝ(in )

≡ ,
iω iω

the finite difference of Ĝ(i). For Q̂ = Iˆ, we thus find

  
Ĝ(in + iω) − Ĝ(in )
0=T ˆ n)
Tr (i
n
iω
  
ˆ n + iω) − (i
(i ˆ n)
= −T Tr Ĝ(in ) (4.157)
n
iω
 I L (in + iω) − I L (in )
≡T ≡ I L ,
n
iω

which just means that the convergence of the series allows the change of variable
in + iω → in that makes I L trivially vanish. It is tempting to assume that I L ,
i.e., the sum over n of the finite difference, coincides with I L in (4.153), i.e., the
sum over n of the derivative, in the limit T → 0, thus ω → 0. That is actually
what Luttinger assumed, in which case I L = 0 follows, and thus Im I L (i0+ ) = 0 in
(4.156). However, that apparently reasonable assumption is not at all guaranteed, as
we now discuss.
For that, let us analyse in detail the differences between the two series

 
ˆ n + iω) − (i
(i ˆ n − iω)
I L ≡ T Tr Ĝ(in ) = 0,
n
2iω
  ˆ n)
∂ (i

IL ≡ T Tr Ĝ(in ) ,
n
∂in

in the limit of zero temperature, thus also ω → 0, noticing that the invariance of the
Luttinger-Ward functional implies I L = 0. Let us start from the latter. Since Ĝ(i)
ˆ
and (i) have discontinuous antihermitian components at  = 0, I L must be dealt
with care in the T → 0 limit, since the functions that are summed may be on different
sides of the imaginary axis. Therefore, we can write

  
ˆ n + iω) − (i
(i ˆ n − iω)
I L =T Tr Ĝ(in )
2iω
n≥1∨n≤−2
1  
+ ˆ
Tr Ĝ(iπ T ) (3iπ ˆ
T ) − (−iπ T)
4πi
1  
+ ˆ
Tr Ĝ(−iπ T ) (iπ ˆ
T ) − (−3iπ T) ,
4πi
4.11 Luttinger’s Theorem 199

so that the two summations n ≥ 1 and n ≤ −2 only involve functions on the same
side of the imaginary axis. At this stage, it is tempting to straight take the T → 0
limit and conclude that
    0−  
∞ d ˆ
∂ (i) d ˆ
∂ (i)
I L −−−→ Tr Ĝ(i) + Tr Ĝ(i)
T →0 0+ 2π ∂i −∞ 2π ∂i
1   
+ ˆ + ) − (i0
Tr Ĝ(i0+ ) + Ĝ(i0+ )† (i0 ˆ + )† ,
4πi

which is however correct only at leading order in T . At the same leading order, I L
simply reads
    0−  
∞ d ˆ
∂ (i) d ˆ
∂ (i)
I L −−−→ Tr Ĝ(i) + Tr Ĝ(i) .
T →0 0+ 2π ∂i −∞ 2π ∂i

Since I L = 0, it follows that

1 1   
IL = − Im I L (i0+ )  − ˆ + ) − (i0
Tr Ĝ(i0+ ) + Ĝ(i0+ )† (i0 ˆ + )† ,
π 4πi
(4.158)

which, we stress again, is an equality only at leading order in T . Therefore, if

  
S(i) ≡ Tr Ĝ(i) + Ĝ(i)† ˆ
(i) ˆ
− (i)†
, (4.159)

is finite as  → 0+ , we can definitely conclude that I L = 0, and therefore that Lut-

tinger’s theorem is not valid.
On the contrary, one can readily prove that S(i → 0) = 0 in the perturbative
regime. Indeed, if we define
& ˆ ˆ ˆ ˆ
(i) − (i)†
(−i) − (−i)†
Ẑ (i)†−1 Ẑ (i)−1 ≡ Iˆ − = Iˆ − ,
2i −2i
(4.160)

where Ẑ (i) = Ẑ (−i) and Ẑ (i → 0) is the quasiparticle residue (4.57) at zero
energy, we do know that perturbatively Ẑ (0) = Ẑ (0)† is positive definite, so that
 
ˆ
(i) ˆ
− (i)†
−−→ 2 Iˆ − Ẑ (0)−1 i ,
→0

and thus S(i) vanishes as  → 0. However, S(i → 0) = 0, though necessary for

I L = 0, is not a sufficient condition. The reason is that the right hand side of (4.158)
200 4 Feynman Diagram Technique

is just the leading term of an expansion in T . Its vanishing means that each term
of the expansion goes to zero as T → 0, which does not guarantee that the whole
series vanishes, too. In other words, while we can safely state that, in the regime
where perturbation theory is valid, S(i → 0) = 0 does imply that I L = 0 hence
Luttinger’s theorem is valid, we cannot exclude that the theorem is violated when
perturbation theory breaks down.
As an example explicitly showing why S(i → 0) = 0 is a necessary but not
sufficient condition for Luttinger’s theorem to hold, let us assume the hypothetical
case in which
 
I L (i) =  ln 1 − (i) , (4.161)

where  ∈ R and (i) = (−i)∗ is perturbative, i.e., it can be expanded in pow-

ers of the interaction and vanishes when the latter does, correctly yielding I L = 0
in the non-interacting case. Accordingly, (i) must be analytic and thus admit a
Taylor expansion in powers of ,

(i)  (0) +  (0) i + O  2 .

Expanding Im I L (iπ T ) of (4.161) at leading order in T , we get

1   (0)
IL = − Im I L (iπ T )  T −−−→ 0 .
π 1 − (0) T →0

Similarly, all higher order terms vanish, too. Since (i) is perturbative in the inter-
action, we come to the conclusion that order by order in perturbation theory I L → 0
for T → 0, which is what happens in reality, as earlier discussed. This result remains
valid so long as (0) < 1, which represents the convergence radius of the series.
When (0) > 1, perturbation theory breaks down and
  
Im I L (i) =  Im ln 1 − (i) −−−→ −π  sign  (0) ,
→0+

jumps to a finite value.

4.11.1.1 Quasiparticle Hamiltonian

Even though S(i → 0) = 0, see (4.159), does not guarantee the validity of Lut-
tinger’s theorem, let us try to uncover its physical meaning.
By definition, the single-particle density of states at the chemical potential A is

1  
A = − lim Tr Ĝ(i) − Ĝ(i)† ,
→0+ 2πi
4.11 Luttinger’s Theorem 201

so that, if

1  
Â(i) ≡ − Ĝ(i) − Ĝ(i)† = Â(i)† = − Â(−i) ,
2πi
then
 
A = lim Tr Â(i) .
→0+

Through Â(i), we can write

 
ˆ
(i) ˆ
− (i)†
= 2i + Ĝ(i)−1 Ĝ(i) − Ĝ(i)† Ĝ(i)†−1

= 2i − 2πi Ĝ(i)−1 Â(i) Ĝ(i)†−1 ,

and thus S(i) in (4.159) becomes

   
S(i) = 2i Tr Ĝ(i) + Ĝ(−i) − 2πi Tr Ĝ(i)−1 + Ĝ(i)†−1 Â(i) .

We define, through (4.160),

& &
Ĝ qp (i)−1
≡ Ẑ (i)† Ĝ(i)−1
Ẑ (i) = i Iˆ − (i)
ˆ , (4.162)

where
&  &
1
ˆ
(i) ˆ
= (i)† ˆ
= (−i) ≡− Ẑ (i)† Ĝ(i)−1 + Ĝ(i)†−1 Ẑ (i)
2& &
1  
= ˆ
Ẑ (i)† 2 Ĥ0 + (i) ˆ
+ (i) †
Ẑ (i) ,
2
(4.163)

is a K × K hermitian matrix, and thus has real eigenvalues ∗ () = ∗ (−),  =
1, . . . , K . Therefore, if we further define

1   
Âqp (i) ≡ − Ĝ qp (i) − Ĝ qp (i)† = Ĝ qp (i) Ĝ qp (i)†
2πi & π &
 1 (4.164)
= = Ẑ (i)−1 Â(i) Ẑ (i)†−1 ,
ˆ
π  2 + (i) 2

ˆ
which is diagonal in the basis that diagonalises (i) with elements

1 
Aqp  (i) = ,
π  + ∗ ()2
2
202 4 Feynman Diagram Technique

then
 
ˆ
S(i) = 2i Tr Ĝ(i) + Ĝ(−i) + 4πi Tr (i) Âqp (i)

  
K (4.165)
∗ () 
= 2i Tr Ĝ(i) + Ĝ(−i) + 4πi .
π  2 + ∗ ()2
=1

Since the first term on the right hand side of (4.165) vanishes for  → 0, the necessary
condition for Luttinger’s theorem to hold is therefore

  
K
∗ () 
ˆ
lim Tr (i) Âqp (i) = lim = 0. (4.166)
→0+ →0+ π  + ∗ ()2
2
=1

In the thermodynamic limit, K → ∞, ∗ () defines a continuous spectrum where 

runs in a d-dimensional space, with d the spatial dimension of the system times the
number of internal degrees of freedom. For instance, in the periodic case,  labels
the momentum within the Brillouin zone, the band index and the spin. The condition
ˆ
(4.166) depends on the analytic properties of the matrix function (i).

•> Analytic assumption

For instance, if we assume that

ˆ
(i) is analytic at  = 0, at least to leading order , (4.167)

then the necessary condition for Luttinger’s theorem to hold is satisfied.


In this case, ∗ ( → 0)  ∗ (0) + O  2 , where ∗ (0) ≡ ∗ are the eigenval-
ues of
&  &
Ĥ∗ ≡ Ẑ (0) Ĥ0 + (0)
† ˆ Ẑ (0) . (4.168)

Accordingly, the quasiparticle Green’s function and density of states at the chemical
potential are

1
Ĝ qp (i) −−→ ,
→0 i Iˆ − Ĥ∗
       (4.169)
Aqp = lim Tr Âqp (i) = Tr δ Ĥ∗ = δ ∗ ,
→0+

4.11 Luttinger’s Theorem 203

and correspond to those of free particles, the ‘quasiparticles’, described by the ‘quasi-
particle’ Hamiltonian Ĥ∗ with eigenvalues ∗ . We will see in Sect. 5.1 that (4.169)
is the starting point of Landau’s Fermi liquid theory.
ˆ
We remark that also a non-analytic (i) may satisfy (4.166), as it happens for
interacting electrons in one dimension, which we discuss in Chap. 6. Those systems
do not sustain quasiparticles in the sense of (4.169), and yet Luttinger’s theorem is
valid. Conversely, since S(i → i0+ ) = 0 is not sufficient for Luttinger’s theorem
to hold, we must also conclude that ‘quasiparticles’ may exist even when Luttinger’s
theorem is violated.
Let us assume the existence of quasiparticles, i.e., of (4.167). Through (4.155),
and since
& 
Im ln det Ẑ (i) Ẑ (i) = 0 ,
†

we readily find that

−
δ̂() −−−→ tan−1 ,
→0+ − Ĥ∗
 
is diagonal with elements −π θ ∗ = −π + π θ − ∗ in the basis that diago-
nalises Ĥ∗ . It follows that (4.156) becomes

1  +  1
N=K+ Tr δ̂(0 ) − Im I L (i0+ )
π π
K   1 
K
1   1
=K+ − π + π θ − ∗ − Im I L (i0+ ) = θ − ∗ − Im I L (i0+ ) ,
π π π
=1 =1
(4.170)

which represents the general statement (4.156) of Luttinger’s theorem when quasi-
particles exist. We note that the first term of the last equation is integer, as N is
integer at T = 0, which implies that

1
− Im I L (i0+ ) = L ∈ Z ,
π

in other words that the Luttinger integral (4.149) for Q̂ = Iˆ is quantised in integer
values. We note that this condition is compatible with the example (4.161) provided
 is integer.
Therefore, when perturbation theory is valid, and thus quasiparticles exist, then


K

N= θ − ∗ , (4.171)
=1
204 4 Feynman Diagram Technique

which is the conventional statement of Luttinger’s theorem, while, when perturbation

theory breaks down and yet quasiparticles still exist, then


K

N= θ − ∗ + L , L ∈ Z, (4.172)
=1

namely, Luttinger’s counting formula (4.171) misses an integer number of electrons.

4.11.2 Luttinger’s Theorem in Presence of Quasiparticles and in

Periodic Systems

In a periodic system invariant under spin SU (2) symmetry, we have the possibility
to further elaborate on the meaning of ‘quasiparticle’. In this case, Ĝ(i) is diagonal
in momentum and spin with elements G(i, k) independent of spin, and thus (i) ˆ
is diagonal, too, with elements ∗ (, k) equal for spin σ = ↑ and ↓, now defined,
see (4.163), as
1 '  
∗ (, k) = ∗ (, k)∗ = (i, k) = Z (i, k)∗ Z (i, k) 2(k) + (i, k) + (i, k)∗ .
2
(4.173)
Correspondingly, the quasiparticle, Aqp , and physical electron, A, density of states
at the chemical potential are, in units of the volume V , see (4.169),
1   1  
Aqp = δ ∗ (k) , A= Z (0, k) δ ∗ (k) , (4.174)
V V
kσ kσ

where ∗ (k) = ∗ ( → 0, k).

We already know that (4.166) and (4.167) imply that, if a manifold k = k F L
exists such that ∗ ( → 0, k F L ) = 0, then ∗ ( → 0, k F L )   2 . We observe that
∗ ( → 0, k F L ) = 0 occurs

Fermi Surface if (k F ) + (0, k F ) = 0 while 0 < Z (0, k F ) < 1, which

defines a conventional Fermi surface k = k F through the roots of G(0, k)−1
in momentum space. The Fermi surface contribution to the physical electron
DOS (4.174) is finite since Z (0, k F ) = 0.
Luttinger Surface if (k F ) + (0, k L ) = 0 but
' 2i
lim Z (i, k L )∗ Z (i, k L ) = lim
→0+ →0+ 2i − (i, k L ) + (i, k L )∗
∼ lim  2 = 0 , (4.175)
→0

which implies (i, k L ) ∼ 1/i and, correspondingly, G(i, k L ) → 0 as  →

0. Therefore, (4.175) defines the so-called Luttinger surface, i.e., the manifold
Problems 205

of roots k = k L of G(0, k) in momentum space, whose existence is due

to a singular self-energy and thus signals the breakdown of perturbation
theory. Remarkably, even though the Luttinger surface contribution to the
quasiparticle DOS, Aqp in (4.174), is finite, its contribution to the physical
electron DOS vanishes.

Therefore, under the analyticity assumption (4.167), Fermi and Luttinger surfaces
are both described by the single equation ∗ ( → 0, k F L ) = 0. Moreover, if pertur-
bation theory is valid, there are quasiparticles, only Fermi surfaces exist within the
Brillouin zone, and, see (4.171),
 
N= θ − ∗ (k) , (4.176)
kσ

which is the standard perturbative statement that the volume fraction of the quasi-
particle Fermi volume, i.e., the manifold of k : ∗ (k) < 0, with respect to the whole
Brillouin zone is equal to the electron filling fraction.
When perturbation theory breaks down without breaking translational and spin
SU (2) symmetries, and Luttinger surfaces appear inside the Brillouin zone, the more
general formula must be used, i.e.,
 
N= θ − ∗ (k) + L , L ∈ Z , (4.177)
kσ

and thus the volume fraction of the quasiparticle Fermi volume no more accounts
for the electron filling fraction.

Problems
4.1 Electron decay rate due to phonons—Consider the Hamiltonian of free elec-
trons coupled to phonons
 †
 †
 ωq  
H= k ckσ ckσ + g(q) x(q) ck+qσ ckσ + x † (q) x(q) + p † (q) p(q) ,
q
2
kσ qkσ

where x(q) = x † (−q) and p(q) = p † (−q) are the Fourier transforms of the phonon
coordinate and its conjugate momentum, satisfying
 
x(q) , p † (q ) = i δq,q .

Consider the Fock diagram  F (i, k) for the self-energy in Fig. 4.47. Find the
expression of the electron decay rate due to phonons contributed by this diagram,
i.e., −Im  F ( + i0+ , k), at finite temperature.
206 4 Feynman Diagram Technique

Fig. 4.47 Fock contribution 2

to the self-energy due to the g(q) 1 1
−
exchange of phonons. The 2 iω − ωq iω + ωq
effective interaction is the
phonon propagator times
2
g(q)

i ,k i − iω, k − q i ,k

4.2 Conserving Hartree-Fock approximation—Consider the Hamiltonian

 † U  † †
H= k − μ ckσ ckσ + ckσ cp+qσ  cpσ  ck+qσ , (4.178)
2V 
kσ kpq σ σ

with chemical potential μ = 0, and U > 0. Approximate the self-energy functional

with the Hartree-Fock skeleton diagrams, see Fig. 4.43, forcing translation and spin
SU (2) symmetries.

• Write, but not solve, the Hartree-Fock self-consistency equation.

• Use the conserving Hartree-Fock approximation to calculate the irreducible inter-
action vertex in the particle-hole channel and, through the Bethe-Salpeter equa-
tion, the reducible interaction vertices in the S = 0 and S = 1 particle-hole chan-
nels. Show that the latter are only function of the transferred frequency iω and
momentum q, thus  S=0,1 =  S=0,1 (iω, q).
• After the analytic continuation iω → ω + i0+ , search for possible poles on the
real frequency axis of the reducible interaction vertices  S=0,1 (ω, q) in the limit
q  k F , where, e.g.,

∂k   ∂ f k
k+q − k  · q ≡ vk · q , f k − f k+q  − vk · q .
∂k ∂k

You should find that only in the spin-singlet channel there is a pole on the real
frequency axis, which corresponds to a collective particle-hole excitation whose
spectrum is that of an acoustic mode, so called Landau’s zero sound.
• Using the reducible interaction vertices  S=0,1 (iω, q) calculate the charge com-
pressibility and spin susceptibility through the corresponding linear response
function at ω = 0 in the limit q → 0. You will notice that the spin susceptibility
diverges at a finite value of U∗ and change sign. Discuss its meaning.
References 207

References
1. A. Abrikosov, L. Gorkov, I. Dzyaloshinskii, Methods of Quantum Field Theory in Statistical
Physics (Dover, New York, 1975). See Sect. 19.4
2. P. Noziéres, Theory of Interacting Fermi Systems (CRC Press, Boca Raton, 1998)
3. J.M. Luttinger, J.C. Ward, Phys. Rev. 118, 1417 (1960). https://doi.org/10.1103/PhysRev.118.
1417
4. J.M. Luttinger, Phys. Rev. 119, 1153 (1960). https://doi.org/10.1103/PhysRev.119.1153
5. J. Skolimowski, M. Fabrizio, Phys. Rev. B 106, 045109 (2022). https://doi.org/10.1103/
PhysRevB.106.045109
Landau’s Fermi Liquid Theory
5

Landau’s Fermi liquid theory is by now a paradigm of interacting fermions, even

strongly interacting ones as 3 He or heavy fermions. For instance, it explains why the
Drude-Sommerfeld theory, which essentially deals with non-interacting electrons
that scatter off randomly distributed impurities, and the semiclassical Boltzmann
equations for Bloch wave-packets describe so well the thermodynamic and transport
properties of metals.
Originally, Landau [1,2] derived his celebrated theory of Fermi liquids by mak-
ing the audacious assumption that the many-body excited states of non-interacting
electrons evolve smoothly into the excited states of interacting electrons when the
interaction is slowly turned on, and that despite the absence of any energy scale that
could justify an adiabatic switching.
The non-interacting excited states | {δn kσ }0 are uniquely identified by the vari-
ations δn kσ of the occupation number of Bloch states with momentum k and spin
σ with respect to the equilibrium values f (k ), the Fermi distribution function, and
have excitation energy
  
δ E 0 {δn kσ } = k δn kσ .
kσ

Landau’s hypothesis implies that each | {δn kσ }0 is in one-to-one correspondence

with an interacting excited state, which allows labelling the latter as its non-
interacting counterpart, thus | {δn kσ }. However, δn kσ cannot refer anymore to the
occupation numbers of the physical electrons, but to different entities that Landau
dubbed quasiparticles.  
The excitation energy δ E {δn kσ } of the interacting excited state | {δn kσ } is
evidently different from the non-interacting one. However,
 for small deviations from
equilibrium, i.e. δn kσ  1, one can expand δ E {δn kσ } in Taylor series,

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 209
M. Fabrizio, A Course in Quantum Many-Body Theory, Graduate Texts in Physics,
https://doi.org/10.1007/978-3-031-16305-0_5
210 5 Landau’s Fermi Liquid Theory

   1 
δ E {δn kσ }  ∗ (k) δn kσ + f k σ,k σ δn kσ δn k σ + · · · , (5.1)
2  
kσ kk σσ

and stop the expansion at second order. The reason why one has to keep the second
order term is because it has essentially the magnitude of the first order one, since,
for small deviations from equilibrium, k is close to the Fermi surface and thus ∗ (k)
is as small as δn kσ . Starting from the energy functional (5.1), Landau was able to
derive many properties, specifically the low temperature linear response functions in
the long wavelength and small frequency limits, and from those the thermodynamic
susceptibilities, the transport properties, and the collective modes, which compare
extremely well with those observed in experiments. We emphasise that, since adia-
batic evolution may only occur in same-symmetry subspaces—different symmetry
states are allowed to cross in energy—Landau’s Fermi liquid theory gives only access
to linear response functions of density operators associated with conserved quanti-
ties. For instance, the non-interacting state | {δn kσ }0 is characterised by a deviation
of the total physical electron number

δN = δn kσ ,
kσ

and, by definition, that must be the same as the deviation of the total quasiparticle
number in the interacting state | {δn kσ }. In other words, the quasiparticle carries the
same conserved quantum numbers of the physical electron, i.e., the same charge and
spin.
In reality, Landau’s adiabatic hypothesis is nothing but the assumption that per-
turbation theory is valid. Indeed, right after the development of diagrammatic many-
body perturbation theory, Landau’s Fermi liquid theory was given a microscopic
justification [3–7] just assuming the perturbative result (4.64). However, the discov-
ery of strongly correlated metals which, at the verge of a Mott transition, display
anomalous properties, partly at odds with conventional Fermi liquids, concurrently
with striking phenomena, like high-temperature superconductivity in doped Mott
insulator copper-oxides, elicited a closer inspection of Landau’s Fermi liquid theory.
That resulted mostly in searching for new states of matter alternative to Fermi liquids,
see, e.g., [8–14], though that is not an exhaustive list, but, partly, also in reexamin-
ing [15,16] the basic assumption that, for Landau’s Fermi liquid theory to hold true,
perturbation theory must be valid, which is evidently not the case in those materials.
Indeed, we already observed, when discussing Luttinger’s theorem in Sect. 4.11, that
quasiparticles exist also at Luttinger’s surfaces, see Sect. 4.11.2, which appear only
when perturbation theory has broken down. That already gives a strong indication
that Landau’s Fermi liquid theory remains valid beyond the perturbative regime.
In this chapter, we derive microscopically Landau’s Fermi liquid theory through
the diagrammatic technique under a more general assumption that includes the valid-
ity of perturbation theory as a special case. The derivation closely follows the con-
ventional one [3,4] with some new developments [15,16]. For simplicity, throughout
this chapter, we deal with interacting electrons in three dimensions, although the final
5.1 Emergence of Quasiparticles Reexamined 211

results can be straightforwardly extended in two dimensions. Moreover, we assume

that none of the symmetries of the Hamiltonian is broken, namely, translation, charge
U (1), and spin SU (2) symmetries, so that the interacting Green’s function G(i, k)
is diagonal in momentum and spin, and independent of the latter.

5.1 Emergence of Quasiparticles Reexamined

In Sect. 4.3.4, we have demonstrated, see (4.64), that order by order in perturbation
theory
Im (, k) −−−→ −γ(k) 2 , (5.2)
→00

where (, k) ≡ ( + i0+ , k) is the retarded self-energy, i.e., the causal analytic
continuation of (i, k) in the complex frequency plane and calculated on the real
frequency axis.
Moreover, in Sect. 4.3.5, we have shown that, under the assumption that the per-
turbative result (5.2) holds true for the whole perturbation series, one can prove the
existence of ‘quasiparticles’, namely of coherent fermionic excitations with disper-
sion ∗ (k) = k in momentum space and whose decay rate (, k) vanishes at the
chemical potential and on the Fermi surface k = k F . Specifically, the quasiparticle
dispersion and decay rate are defined by the equations
   
∗ (k) = Z ∗ (k), k k + Re  ∗ (k), k , k F : ∗ (k F ) = 0 ,
(5.3)
(, k) = −Z (, k) Im (, k)  Z (, k) γ(k) 2 ,

where, see (4.57),

−1
∂Re (, k)
Z (, k) = 1 − , (5.4)
∂
 
is the quasiparticle residue. Specifically, Z 0, k F < 1 is the weight of a quasiparticle
in the physical electron at the Fermi surface and at the chemical potential.
However, the simple requirement that a coherent ‘quasiparticle’ does exist, namely
that its lifetime grows indefinitely as  → 0, just implies

(, k) = −Z (, k) Im (, k) −−→ γ∗ (k) 2 → 0 , (5.5)

→0

which also occurs if, e.g., Im (0, k) is finite but Z (, k) ∼ 2 , completely at odds
with the perturbative result (4.64). In other words, the condition (5.5) is actually more
general than (5.2), which is just a special case. We emphasise that the continuously
vanishing (, k) implies a finite and smooth single-particle density of states (DOS)
around  = 0, see (4.67), as long as γ∗ (k) > 0, while a gapped DOS if γ∗ (k) = 0,
signals of an interaction-driven non-symmetry breaking Mott insulator, which may
occur only at half-filling in the present single-band model. To maintain full generality,
212 5 Landau’s Fermi Liquid Theory

in what follows we keep using γ∗ (k) without specifying whether it is finite or zero,
with the prescription that, if γ∗ (k) = 0, then γ∗ (k) 2 in (5.5) is replaced by an
infinitesimal constant η > 0.
In principle, one could even imagine that (, k) vanishes not analytically as
 → 0. However, recalling that Re (, k) and Im (, k) are related to each other
by the Kramers-Krönig equations, that would most likely correspond to a perturbation
theory ill-defined already at weak coupling,1 as it happens in one dimension, see
Chap. 6.

•
> Analytic assumption
Hereafter, we exclude that possibility, although it might be realised even beyond one
dimension, and assume that

(, k) is everywhere analytic at small  but, eventually, right at  = 0 . (5.6)

As a matter of fact, the above statement is the real frequency counterpart of

the statement (4.167) that we already showed in Sect. 4.11.2 to yield well-defined
quasiparticles. Indeed, what follows is exactly the same we discussed in Sect. 4.11.2
but now seen on the real axis.
We thus assume the analytic conjecture (5.6) and the analytic behaviour (5.5), at
least to leading order in , which does include the weak coupling perturbative regime
as a special case, and try to draw all its consequences, which coincide with the results
of Sect. 4.11.2. In terms of (, k) and Z (, k), we can write the single-particle DOS
as, cf. (4.67),

1 −Im (, k)
A(, k) =  2  2
π
 − k − Re (, k) + Im (, k)
(5.7)
Z (, k) (, k)
= ≡ Z (, k) Aqp (, k) ,
π (, k)2 + (, k)2

where Aqp (, k), see also (4.164), is the ‘quasiparticle’ DOS as opposed to the
physical electron one, A(, k), and, see also (4.173),
 
(, k) = (i →  + i0+ , k) = Z (, k) k + Re (, k) −  . (5.8)

1 In two dimensions, see (4.64), Im ( → 0, k) ∼ 2 ln ||, which, despite being weakly not ana-
lytic, does not yield singularities in perturbation theory.
5.1 Emergence of Quasiparticles Reexamined 213

Let us at first concentrate on Aqp (, k). We note that, because of (5.5),

1 (, k) 1 γ∗ (k) 2
Aqp (, k) = −−→ ,
π (, k)2 + (, k)2 →0 π ( → 0, k)2 + γ∗ (k)2 4
(5.9)
typically vanishes unless ( → 0, k) = 0, in which case Aqp ( → 0, k) diverges.
Let us, therefore, define the ‘quasiparticle’ energy dispersion through
      
∗ (k) :  ∗ (k), k = Z ∗ (k), k k + Re  ∗ (k), k − ∗ (k) = 0 .
(5.10)

The equation ∗ (k) = 0, or, equivalently, (0, k) = 0, defines surfaces k =

k F L within the Brillouin zone, exactly the Fermi, k = k F , and Luttinger,
k = k L , surfaces of Sect. 4.11.2. Moreover, at k = k F L , (5.8) as well as our
analytical conjecture (5.6) suggest that (, k F L ) vanishes linearly in . Indeed,
we can envisage two different analytic scenarios:

Fermi Surface: k F L + Re (0, k F L ) = 0, so that

 −1
Re G  → 0, k F L = Z (0, k F L )−1  , ( → 0, k F L ) = − .

This is the case of conventional Fermi liquids, hereafter denoted as (F),

where k F L ≡ k F belongs to the Fermi surface. At the Fermi surface, the
Green’s function has a simple pole at  = 0 with residue Z (0, k F ), or,
equivalently, G(0, k F )−1 = 0.
Luttinger Surface: The quasiparticle residue Z ( → 0, k F L ) vanishes
faster than Re ( → 0, k F L ) diverges, which implies
 2
   kF L   2
Re   → 0, k F L  , Z  → 0, k F L   2 ,
  +  kF L
2

so that
lim (, k F L ) =  .
→0
In this case, hereafter denoted as (L), k F L = k L defines a Luttinger sur-
face [17], i.e., the location within the Brillouin zone of the zeros of the
Green’s function at  = 0, G(0, k L ) = 0.

By our main assumption (5.6), (, k) has a regular Taylor expansion at small 
and k  k F L , at least to leading order, in which case
   
  → 0, k  k F L  (0, k) ∓  ≡ ∓  − ∗ (k) , (5.11)
214 5 Landau’s Fermi Liquid Theory

where the minus and plus signs refer, respectively, to the cases (F) and (L) above. It
follows that, for small  and close to the Fermi or Luttinger surfaces, the quasiparticle
DOS (5.9) as function of a small  becomes

1 γ∗ (k)2  
Aqp (, k)   2  δ  − ∗ (k) , (5.12)
π
 − ∗ (k) + γ∗ (k)2 4

both close to a Luttinger surface as well as a Fermi surface. The precise meaning
of the last equality in (5.12) is that as  → 0 the quasiparticle DOS always vanishes
unless  = ∗ (k) and k moves towards the Fermi or Luttinger surfaces. In the sense
of a distribution, the above statement corresponds to the following equivalence at
small temperature T ,
 
∂ f () ∂ f ∗ (k)  
d − Aqp (, k) F() ≡ − F ∗ (k) . (5.13)
∂ ∂∗ (k)

However, the two cases (F) and (L) differ substantially from the point of view of
the the physical electron DOS (5.7), which for k close to k F L can be written as, cf.
(4.71),

Z (, k) γ∗ (k)2
A(, k)   2 + Ainc (, k) , (5.14)
π
 − ∗ (k) + γ∗ (k)2 4

where Ainc (, k) is the finite energy incoherent background that carries the rest of
spectral weight. Indeed, in case (F), the quasiparticle residue Z (0, k F ) is finite and
thus the physical electron DOS is also characterised by a peak at the quasiparti-
cle energy ∗ (k), which becomes a δ-function with weight Z (0, k F ) for k moving
towards the Fermi surface. On the contrary, in case (L)

2
Z (, k L )  ,
(k L )2 + 2

vanishes quadratically as  → 0, and so does A(, k). In other words, the physical
electron DOS vanishes
  at the Luttinger surface, and yet coherent quasiparticles exist.
The fact that Z , k → 0 just means that the quasiparticles, though they do exist,
have zero weight in the physical electron excitation.
We remark that, if the system is a symmetry-invariant Mott insulator, then only a
Luttinger surface can exist. The physical electron DOS A(, k) = 0 for all  within
the insulating gap, and yet Aqp (, k) is δ-peaked at the quasiparticle energy ∗ (k), a
rather striking physical scenario. In that case, which must correspond to half-filled
density N = V , with V the number of sites, the generalised Luttinger’s theorem
(4.177) determines unequivocally the value of L, and suggests that a Luttinger sur-
face can only accommodate N = V electrons, whatever is the enclosed volume. As
5.2 Manipulating the Bethe-Salpeter Equation 215

a consequence, the Fermi pockets that must appear upon doping the Mott insula-
tor only count for the doping away from half-filling. In other words, if v E P and
v H P , respectively, are the volume fractions of electron and hole Fermi pockets with
respect to the Brillouin zone volume, then the electron filling fraction ν = N /2V
in the absence of Luttinger surfaces, i.e., when perturbation theory is valid and thus
Luttinger’s theorem holds, reads
   
ν = vE P + 1 − vH P = 1 + vE P − vH P , (5.15)

while, when a Luttinger surface is present, thus perturbation theory breaks down and
Luttinger’s theorem is violated,

1  
ν= + vE P − vH P . (5.16)
2

5.2 Manipulating the Bethe-Salpeter Equation

We know that the Bethe-Salpeter equation is important to calculate the interaction

vertex and, from that, all linear response functions. In the particle-hole channel, that
equation reads explicitly, see (4.96),

σ1 ,σ2 ;σ3 ,σ4 i1 + iω k1 + q, i2 k2 ; i2 + iω k2 + q, i1 k1 )

= 0 σ1 ,σ2 ;σ3 ,σ4 i1 + iω k1 + q, i2 k2 ; i2 + iω k2 + q, i1 k1 )
1   
+ T 0 σ1 ,σ ;σ,σ4 i1 + iω k1 + q, i k; i + iω k + q, i1 k1 )
V 
k1 σσ i1
G(i + iω, k + q) G(i, k)

σ,σ2 ;σ3 ,σ i + iω k + q, i2 k2 ; i2 + iω k2 + q, i k) ,

which, as in Sect. 4.7, we shortly write as

 = 0 + 0 R , (5.17)

with implying the sum of all internal variables, i.e., frequencies, spins and
momenta, and the kernel

R(i + iω k + q, i k) ≡ G(i + iω, k + q) G(i, k) . (5.18)

We recall that in each interaction vertex, the z-component of the spin is conserved,
which implies that σ1 − σ4 = σ3 − σ2 = σ − σ  .
216 5 Landau’s Fermi Liquid Theory

5.2.1 A Lengthy but Necessary Preliminary Calculation

Since our goal is to calculate low temperature linear response functions at long
wavelength and small frequency, we hereafter analyse the behaviour of  at a very
small frequency, ω, and momentum, q = |q|, transferred, as well as at very low
temperature T .
Let us start by studying, as representative of the Bethe-Salpeter equation, the
Matsubara frequency summation

Ck (iω, q) ≡ T R(i + iω k + q, i k) F(i)

 (5.19)
=T G(i + iω, k + q) G(i, k) F(i) ,


with the test function F(i) = F(−i)∗ smooth around  = 0 and not singular for
 → ±∞. Our scope here is to understand whether the kernel R is analytic at ω =
q = 0 so that the value of Ck (iω, q) for q → 0 and ω → 0 is independent of the
order of the limits.
We recall that in a system with gapless single-particle excitations, G(i, k) has
generally a discontinuous imaginary part crossing  = 0, see Sect. 4.1.3, but other-
wise is a non-singular function. Taking that property into account, we write, assuming
the bosonic Matsubara frequency ω > 0,

Ck (iω, q) = T G(i + iω, k + q) G(i, k) F(i)
>0

+T G(i + iω, k + q) G(i, k) F(i)
<−ω (5.20)

+T G(i + iω, k + q) G(i, k) F(i)
−ω<<0

≡ Ck (iω, q) + Ck(2) (iω, q) + Ck(3) (iω, q) ,

(1)

so that the functions in each separate sum have no discontinuity in the range of
summation. If we take ω = q = 0 from the start
 
Ck (0, 0) = T G(i, k) G(i, k) F(i) + T G(i, k) G(i, k) F(i)
>0 <0

= Ck (0, 0) + Ck(2) (0, 0) ,

(1)

while, if we first send ω → 0 and then send q → 0, so-called q-limit,

q
Ck ≡ lim Ck (0, q)
q→0
 
= lim T G(i, k + q) G(i, k) F(i) + T G(i, k + q) G(i, k) F(i)
q→0
>0 <0
(1) (2)
= Ck (0, 0) + Ck (0, 0) .
5.2 Manipulating the Bethe-Salpeter Equation 217

In the opposite ω-limit, i.e., sending first q → 0 and then ω → 0

Ckω ≡ lim Ck (iω, 0)

ω→0
 
= lim T G(i + iω, k) G(i, k) F(i) + T G(i + iω, k) G(i, k) F(i)
ω→0
>0 <−ω

+T G(i + iω, k) G(i, k) F(i)
−ω<<0

= Ck(1) (0, 0) + Ck(2) (0, 0) + lim Ck(3) (iω, 0) .

ω→0
(5.21)
(3)
Therefore, a non-analytic behaviour may arise only because Ck (iω, q) could remain
(3)
finite in the ω-limit. Naïvely, having assumed F(i) smooth at small , Ck (iω, q),
being limited to frequencies −ω <  < 0, would seem to vanish as ω → 0. That is
surely the case of insulators with gaps to all kinds of excitations, where the Green’s
functions are also smooth. However, in a system with gapless excitations, the product
of the two Green’s functions might have a singularity at ω = 0 that compensates the
(3)
vanishing frequency range of the summation, thus yielding a finite Ck (iω, 0) as
ω → 0. In other words, at small ω, q and T , the kernel R for  ∈ [−ω, 0], which we
hereafter name   , might be singular and non analytic at ω = q = 0, as opposed to
R for  ∈ / [−ω, 0], which we denote as R inc and is instead analytic at ω = q = 0.
Our aim here is to explicitly calculate .

By standard tricks, we can rewrite (5.20) as


dz
Ck (iω, q) = − f (z) G(z + iω, k + q) G(z, k) F(z) ,
2πi

with the contour shown in Fig. 5.1 that encircles clockwise the two branch cuts z = 
and z =  − iω, with real  ∈ [−∞, ∞].

Fig. 5.1 Contour used in the

integration (5.22), red and
blue arrow lines. The contour
goes along two branch cuts
at z =  and z = −iω + 
with real  ∈ [−∞, ∞]. For
each Green’s function,
G(z + iω, k) and G(z, k),
we indicate its retarded (+)
or advanced (−) character
above or below the branch
cuts
218 5 Landau’s Fermi Liquid Theory

The singular contribution to R may derive from the green region in that figure
where Im z ∈ [−ω, 0], thus the integral over the two lines in blue that reads

∞ d  
(3)
Ck (iω, q) = − f () F() G + (, k + q) G − ( − iω, k) − G + ( + iω, k + q) G − (, k) ,
−∞ 2πi

where G + (, k) = G( + i0+ , k) and G − (, k) = G( − i0+ , k) = G + (, k)∗ are,
respectively, the retarded and advanced components of the Green’s functions, and
similar definitions and properties hold for the self-energy as well. Since ω is small,
we can write the term in square brackets as

G + (, k + q) G − ( − iω, k) − G + ( + iω, k + q) G − (, k)

 
= G + (, k + q) G − ( − iω, k) − G + (, k + q) G − ( − iω, k) 
→+iω

∂ 
 −iω G + (, k + q) G − ( − iω, k) ,
∂
where

 
G + (, k + q) G − ( − iω, k) = G − ( − iω, k) − G + (, k + q)
1
.
iω − k+q + k − + (, k + q) + − ( − iω, k)

At leading order in ω and q,

G − ( − iω, k) − G + (, k + q)  2π i A(, k) ,

while, recalling our definitions of (, k) in (5.5), (, k) in (5.8) and Z (, k) in
(5.4), and, expanding in ω and q,

iω − k+q + k − + (, k + q) + − ( − iω, k)

= iω − k+q + k − Re + (, k + q) + Re + ( − iω, k)

− i Im + (, k + q) − i Im + ( − iω, k)

∂k ∂Re + (, k)

 iω Z (, k)−1 − q · + − 2i Im + (, k)
∂k ∂k

∂k ∂Re + (, k)

 Z (, k)−1 iω − q · Z (, k) + + 2i (, k) .
∂k ∂k
5.2 Manipulating the Bethe-Salpeter Equation 219

Therefore

G + (, k + q) G − ( − iω, k) − G + ( + iω, k + q) G − (, k)

⎛ ⎞

∂ ⎜ iω Z (, k) ⎟
⎜ ⎟
 −2π i ⎜ A(, k) ⎟
∂ ⎝ ∂k ∂Re + (, k) ⎠
iω − q · Z (, k) + + 2i (, k)
∂k ∂k

so that, upon integration by part, and noting that ∂ F()/∂ can be neglected with
respect to the singular ∂ f ()/∂ at low T , we finally find through (5.13) that
∞ ∂ f ()
(3)
Ck (iω, q)  d − A(, k) F()
−∞ ∂
iω Z (, k)
∂k ∂Re + (, k)
iω − q · Z (, k) + + 2i (, k)
∂k ∂k
∞ ∂ f ()
= d − Aqp (, k) F() Z (, k)2
−∞ ∂
iω Z (, k)
∂k ∂Re + (, k)
iω − q · Z (, k) + + 2i (, k)
∂k ∂k

∞ ∂ f ()  
 d − δ  − ∗ (k) F() Z (, k)2
−∞ ∂
iω Z (, k)
∂k ∂Re + (, k)
iω − q · Z (, k) + + 2i (, k)
∂k ∂k (5.22)
 
∂ f ∗ (k)  2 iω  
− Z ∗ (k), k F ∗ (k)
∂∗ (k) iω − q · v k
 
∂ f ∗ (k)  2 iω
− Z ∗ (k), k F(0) ,
∂∗ (k) iω − q · v k
the last almost equalities deriving from −∂ f ()/∂  δ() at low T , which  picks out

only the small  component of the quasiparticle DOS, thus Aqp (, k)  δ  − ∗ (k) ,
and further implies that ( → 0, k)  0 because of (5.5). Moreover, through the
definition (5.10) we can readily show that
     
d  ∗ (k), k ∂ , k ∂∗ (k) ∂ ∗ (k), k
0= =  +
dk ∂ =∗ (k) ∂k ∂k
 
∂∗ (k)   ∂k ∂Re + , k
=− + Z ∗ (k), k + 
∂k ∂k ∂k =∗ (k) ,
220 5 Landau’s Fermi Liquid Theory

namely
 
  ∂k ∂Re + , k ∂∗ (k)
Z ∗ (k), k +  = ≡ vk , (5.23)
∂k ∂k =∗ (k) ∂k

which is the quasiparticle group velocity.

In conclusion, coming back to our original task (5.19), we find that a singular part
 of the kernel R, namely that in the Matsubara frequency interval −ω <  < 0,

does exist and can be written, in the sense of distribution, as
 
δ,0 ∂ f ∗ (k) iω

(i + iω k + q, i k) ≡ − Z (k)2 , (5.24)
T ∂∗ (k) iω − q · v k

having defined
 
Z (k) ≡ Z ∗ (k), k . (5.25)
Indeed, we note that in the ω-limit
 
δ,0 ∂ f ∗ (k)
 ω (i, k) ≡ lim lim 
  (i + iω k + q, i k) = − Z (k)2 ,
ω→0 q→0 T ∂∗ (k)
(5.26)
sing
yields a finite contribution to Ck (iω → 0, 0) despite the vanishing range of the
sum. On the contrary, in the opposite q-limit,

  (i + iω k + q, i k) = 0 ,
q (i, k) ≡ lim lim  (5.27)
q→0 ω→0

as expected, which explicitly shows that   is non analytic at ω = q = 0. The com-


ponent Rinc of the kernel R is instead, by definition, analytic at ω = q = 0, thus
ω = R
R q = R inc , the last equality due to our assumption of small q and ω. It thus
inc inc
follows that

Rω =  inc
ω + R ω
= inc ,
ω + R q = R
q + R
Rq =  inc . (5.28)
inc

For later convenience, we define

 (i + iω k + q, i k) − 
(i + iω k + q, i k) ≡   ω (i, k)
 
δ,0 ∂ f ∗ (k) q · vk (5.29)
=− Z (k)2 ,
T ∂∗ (k) iω − q · v k

and consistently
   
R= inc = 
+R −
ω + Rinc + 
 ω ≡  + Rinc . (5.30)
5.2 Manipulating the Bethe-Salpeter Equation 221

With those notations

 ω (i, k) , ω (i, k) = 0 ,
q (i, k) = − (5.31)

so that

R ω = ω + Rinc
ω q
= Rinc , R q = q + Rinc = q + Rinc . (5.32)

We emphasise that, unlike the complex kernel R,  only depends on two unknown
quantities: the quasiparticle dispersion ∗ (k) and the quasiparticle residue Z (k). In
the case of a Luttinger surface, Z (k → k L ) → 0 and thus  → 0, too, which, one
might erroneously conclude, should put an end to the story, including the circum-
stance that the Luttinger surface appears in a Mott insulator. However, the key to the
derivation of Landau’s Fermi liquid theory is the ability to absorb Z (k) into effective
quantities that are well behaved even though Z (k → k L ) → 0.

5.2.2 Interaction Vertex and Density-Vertices

The Bethe-Salpeter equation (5.17) can, therefore, be written as

 
 = 0 + 0 R  = 0 + 0  + Rinc 
  (5.33)
= 0 +  R 0 = 0 +   + Rinc 0 ,

so that, since 0 is analytic at ω = q = 0,2 through (5.32),

 ω = 0 + 0 Rω  ω = 0 + 0 Rinc  ω = 0 +  ω Rinc 0 .

We can express 0 in terms of 

 −1  −1
0 =  1+ R  = 1+ R ,

or in terms of  ω , i.e.,
 −1  −1
0 =  ω 1 + Rinc ω = 1 + ω Rinc ω ,

which implies, e.g., that

 −1  −1
 1+ R  = 1 + ω Rinc ω

2 By definition, 0 does not include the kernel R, i.e., a particle line and a hole one differing by
frequency iω and momentum q, and therefore, cannot have any singularity in that channel. However,
the analyticity at ω = q = 0 does not exclude singularities in the other variables.
222 5 Landau’s Fermi Liquid Theory

namely,
   
1 + ω Rinc  = ω 1+ R  ,

which corresponds to a new definition of the Bethe-Salpeter equation

 
 = ω + ω R − Rinc  = ω + ω  . (5.34)

We note that we have been able to absorb the unknown 0 and Rinc into a single
unknown quantity, the interaction vertex  ω .
Through (5.34) we find that
 −1
ω =  1+  ,

which, since R ω +  = Rinc +  = R, also implies

   −1  −1
1 + Rω ω = 1 +   1+  + Rω  1+ 
   −1
= 1+ R  1+  .
(5.35)

Let us consider now the interacting A density-vertex  A that is defined, see Fig. 4.25,
by the equation
 A = 0A + 0A R , (5.36)
where 0A is the non-interacting value. As before, we can express the latter in terms
of the interacting  A ,
 −1  −1
0A =  A 1+ R  = 1+ R A . (5.37)

Alternatively, we can take the ω-limit

 A ω = 0A + 0A Rω ω ,

and again solve for 0A and get, making use of (5.35),
 −1    −1
0A =  A ω 1 + Rω ω = A ω 1+  1+ R 
 −1  
= 1+ R 1+  A ω .
(5.38)
5.3 Linear Response Functions 223

Comparing (5.37) with (5.38) we find

 −1    −1
A 1+ R  = A ω 1+  1+ R  ,

namely
 
A = A ω + A ω   = A ω 1+  . (5.39)

Just like for the interacting vertex, we have been able to absorb Rinc into the ω-limit
of the vertex.
We can straightforwardly repeat all the above calculations focusing on the q-limit,
i.e., using  q in place of  ω and  A q in place of  A ω . In that case, the (5.34), (5.38)
and (5.39) are replaced, respectively, by

 = q + q  ,
   −1
 −1  
0A =  A q  
1+ 1+ R  = 1+ R 1+ 
 A q ,
 
A = A q + A q 
  = A q   .
1+
(5.40)

5.3 Linear Response Functions

We can now represent a generic linear response function in terms of the above quanti-
ties. Let us consider the response function χ AB (ω + iη, q), with η > 0 infinitesimal,
which we shortly write as
   
χ AB ≡ Tr  A R 0B = Tr 0A R  B . (5.41)

Using (5.39), the last equation in (5.38), and recalling that R =  + R ω , we can write
 
χ AB = Tr  A ω 1+  R 0B
     
= Tr  A ω Rω 0B + Tr  A ω  0B + Tr  A ω   R 0B
 
= χωAB + Tr  A ω  1+ R 0B
   −1  
= χωAB + Tr  A ω  1+ R 1+ R 1+  B ω
 
= χωAB + Tr  A ω  1+  B ω
   
= χωAB + Tr  A ω   B ω + Tr  A ω    B ω ,
(5.42)
224 5 Landau’s Fermi Liquid Theory

where
   
χωAB = Tr  A ω Rω 0B = Tr  A ω Rinc 0B , (5.43)

is the ω-limit of the response function.

We define
 
δ,0 ∂ f ∗ (k) q · vk
(i + iω k + q, i k) = −Z (k) 2
T ∂∗ (k) iω − q · v k (5.44)
≡ Z (k) Rqp (i + iω k + q, i k) ,
2

and note that, for small q

  1 1
T Rqp (i + iω k + q, i k)  T
 
i + iω − ∗ (k + q) i − ∗ (k)

≡T G qp (i + iω, k + q) G qp (i, k) ,

(5.45)
is just the sum over Matsubara frequencies of the product of two non-interacting
Green’s functions with k → ∗ (k), thus the Green’s functions G qp (4.169) of the
quasiparticles.
In addition, we define the quasiparticle interaction vertex

Aσ1 ,σ2 ;σ3 ,σ4 (k, k ; iω, q) ≡ Z (k) Z (k ) σ1 ,σ2 ;σ3 ,σ4 (i + iω k + q, i k ; i + iω k + q, i k) ,
(5.46)
where i → ∗ (k)  0 and i → ∗ (k )  0, and the quasiparticle density-vertices
for small q
 
λ A i k σ, i + iω k + q σ  ; iω, q) = Z (k)  A i k σ, i + iω k + q σ  ; iω, q) ,
(5.47)
both shown in Fig. 5.2, through which we can write (5.42) as
   
χ AB = χωAB + Tr λ A ω Rqp λ B ω + Tr λ A ω Rqp A Rqp λ B ω .
(5.48)
We can repeat step-by-step the above derivation using (5.40) instead of (5.39) and
(5.38), in which case we find
   
q
χ AB = χ AB + Tr λ A q R qp λ B q + Tr λ A q R qp A R qp λ B q ,
(5.49)
where

qp (i + iω k + q, i k) = 1
R  (i + iω k + q, i k)

Z (k)2
  (5.50)
δ,0 ∂ f ∗ (k) iω
=− .
T ∂∗ (k) iω − q · v k
5.3 Linear Response Functions 225

Z(1) Z(3)
1 3 1 3

A =
4 2 4 2
Z(4) Z(2)

Z(1)
1 1

λ = Λ
Z(2)
2 2

Fig. 5.2 Top panel: definition of the quasiparticle interaction vertex A in terms of the physical
electron one  and the quasiparticle residue Z . The numerals indicate the frequency, momentum,
and spin of each external leg. Bottom panel: same as top one but for the quasiparticle triangular
density-vertex λ in terms of the physical electron one 

In both cases (5.48) and (5.49), all the complexity of the interacting linear response
function has been notably reduced at small ω, q and T . That simplification has
been achieved by simply exploiting the analytic properties of the kernel R, which is
remarkable.

5.3.1 Response Functions of Densities Associated to Conserved

Quantities

If A or B are conserved quantities, so that ρ A (0) or ρ B (0) are just numbers, then in
(5.48) χωAB = 0, and
   
χ AB = Tr λ A ω Rqp λ B ω + Tr λ A ω Rqp A Rqp λB ω .
(5.51)
We note that Rqp , see (5.44), forces all internal frequencies equal to the quasiparticle
energies, which, in turn, are vanishingly small because of the Fermi distribution
derivative. Therefore, the Ward-Takahashi identity (4.131) in the ω-limit implies, if
A is conserved, that3

  ∂(, k)  
 A ω (, k) = 0A k σ2 , k σ1 1 −  = 0A k σ2 , k σ1 Z (k)−1 .
∂ =∗ (k)

3 Since  (k) → 0, namely k is close to the Fermi or Luttinger surface, then Im  is either vanishing
∗
in case (F), or negligible with respect to Re  in case (L). Therefore, in both case the derivative of
the self-energy can be replaced by the derivative of its real part.
226 5 Landau’s Fermi Liquid Theory

namely
 A ω = 0A Z −1 , λ A ω = Z  A ω = 0A , (5.52)
and, similarly, if B is conserved

λ B ω = 0B . (5.53)

If both A and B are conserved, the linear response function simplifies even more
   
χ AB = Tr 0A Rqp 0B + Tr 0A Rqp A Rqp 0B , (5.54)

and explicitly reads

 
1  ∂ f ∗ (k) q · vk
χ AB (ω, q) = − 0A (k σ2 , k + q σ1 ) 0B (k + q σ1 , k σ2 )
V ∂∗ (k) iω − q · v k
kσ1 σ2
   
1  ∂ f ∗ (k) ∂ f ∗ (k ) q · vk q · v k
+ 2
V kk ∂∗ (k) ∂∗ (k ) iω − q · v k iω − q · v k

0A (k σ4 , k + q σ1 ) Aσ1 ,σ2 ;σ3 ,σ4 (k, k ; ω, q)0B (k + q σ3 , k σ2 )
σ1 σ2 σ3 σ4
T 
= G qp (i + iω, k + q) G qp (i, k) 0A (k σ2 , k + q σ1 ) 0B (k + q σ1 , k σ2 )
V 
kσ1 σ2

T2  
+ G qp (i + iω, k + q) G qp (i, k) G qp (i + iω, k + q) G qp (i , k )
V2 kk 

0A (k σ4 , k + q σ1 ) Aσ1 ,σ2 ;σ3 ,σ4 (k, k ; ω, q)0B (k + q σ3 , k σ2 ) .
σ1 σ2 σ3 σ4
(5.55)
Remarkably, the linear response functions in the case of conserved quantities just
depend on the quasiparticle interaction vertex A and quasiparticle energy ∗ (k), and
not on Z (k); the complexity of the interacting response function has been reduced
just to two unknown functions at the Fermi or Luttinger surfaces.
Similar to (4.99)–(4.101), we can define the interaction vertex A in the S = 0 and
S = 1 quasiparticle-quasihole channels, respectively,

1 
A S (k, k ; ω, q) = A↑,↑;↑,↑ (k, k ; ω, q) + A↑,↓;↓,↑ (k, k ; ω, q) ,
2
1  (5.56)

A A (k, k ; ω, q) = A↑,↑;↑,↑ (k, k ; ω, q) − A↑,↓;↓,↑ (k, k ; ω, q) .
2

We finally observe that in the case of a Luttinger surface, finite quasiparticle inter-
action vertex and density-vertices despite the vanishing quasiparticle residue imply
diverging physical electron interaction vertex and interacting density-vertices. That is
5.4 Thermodynamic Susceptibilities 227

certainly the case of quasiparticle density-vertices corresponding to conserved quan-

tities, which through (5.52) are equal to the corresponding non-interacting vertices
of the physical electrons. It is, therefore, remarkable that a theory full of singularities
can nonetheless yield a regular quasiparticle description.

5.4 Thermodynamic Susceptibilities

If A and B are both conserved quantities, through (5.55) the thermodynamic suscep-
tibility κ AB reads

 
q 1  ∂ f ∗ (k)
κ AB ≡ −χ AB = − 0A (k σ2 , k σ1 ) 0B (k σ1 , k σ2 )
V ∂∗ (k)
kσ1 σ2
   
1  ∂ f ∗ (k) ∂ f ∗ (k )
− 2 (5.57)
V kk ∂∗ (k) ∂∗ (k )

0A (k σ4 , k σ1 ) Aσ1 ,σ2 ;σ3 ,σ4 (k, k )0B (k σ3 , k σ2 ) ,
q

σ1 σ2 σ3 σ4

where
Aσ1 ,σ2 ;σ3 ,σ4 (k, k ) = lim lim Aσ1 ,σ2 ;σ3 ,σ4 (k, k ; ω, q) ,
q
q→0 ω→0

is the q-limit of the quasiparticle interaction vertex. In what follows we derive the
explicit expressions of the most relevant susceptibilities characterising Landau’s
Fermi liquid.

5.4.1 Charge Compressibility

If A = B = N , the number of particles, the non-interacting vertex 0 (k σ, k σ  ) =

δσσ , and the corresponding susceptibility, i.e., the charge compressibility κ, is, see
(5.56),

 
1  ∂ f ∗ (k)
κ=−
V ∂∗ (k)
kσ
   
1   ∂ f ∗ (k) ∂ f ∗ (k )
Aσ,σ ;σ ,σ (k, k )
q
− 2
V kk σσ ∂∗ (k) ∂∗ (k )
       
2  ∂ f ∗ (k) 4  ∂ f ∗ (k) ∂ f ∗ (k )
A S (k, k ) ≡ 2 ρqp 1 − A S ,
q
=− − 2 
V ∂∗ (k) V kk ∂ ∗ (k) ∂ ∗ (k )
k
(5.58)
228 5 Landau’s Fermi Liquid Theory

where ρqp ≡ Aqp (0) is the quasiparticle local density of states at the chemical poten-
tial as opposed to the non-interacting one ρ0 , with, see also (4.169),
 
1  ∂ f ∗ (k) 1   
Aqp () = −  δ  − ∗ (k) , (5.59)
V ∂∗ (k) V
k k

and
   
2  ∂ f ∗ (k) ∂ f ∗ (k )
ρ−1 A S (k, k )
q
AS ≡ qp 
2
V kk ∂ ∗ (k) ∂ ∗ (k )
 (5.60)
2     q
 ρ−1
qp 2
δ ∗ (k) δ ∗ (k ) A S (k, k ) ,
V kk

is the average over the Fermi or Luttinger surface of A S (k, k ). We remark that the
q

compressibility κ0 of non-interacting electrons is just 2ρ0 , so that

κ ρqp  
= 1 − AS . (5.61)
κ0 ρ0

We expect that a repulsive interaction slows down quasiparticle motion yielding a

reduced quasiparticle bandwidth, or, equivalently, increased quasiparticle effective
mass, and thus ρqp > ρ0 . In addition, we also expect that the interacting compress-
ibility κ is smaller than the non-interacting one κ0 . It follows that
ρ0
AS > 1 − > 0.
ρqp

As long as A S < 1, the system is compressible and hence metallic. The repulsion
driven Mott metal-to-insulator transition is, therefore, signalled by

1 
AS = A↑,↑;↑,↑ + A↑,↓;↓,↑ → 1 .
2
Since the more localised the charge, the smaller A↑,↑;↑,↑ is because of Pauli exclusion
principle, the approach to a Mott transition actually corresponds to A↑,↑;↑,↑ → 0
and A↑,↓;↓,↑ → 2. Therefore, in a Mott insulator that has a Luttinger surface, thus
a finite ρqp , A↑,↑;↑,↑ = 0, A↑,↓;↓,↑ = 2, and so A S is strictly equal to one.

5.4.2 Spin Susceptibility

If we take, instead

1  † †

A = B = Sz = ck↑ ck↑ − ck↓ ck↓ ,
2
k
5.4 Thermodynamic Susceptibilities 229

of the total spin, the non-interacting vertex 0 z (k σ, k σ  ) =

S
i.e., the z-component

δσσ δσ↑ − δσ↓ , and simply following the derivation above, the corresponding spin
susceptibility χ is finite at low temperature and reads
 
χ = ρqp 1 − A A , (5.62)

where A A is now the average over the Fermi or Luttinger surface of A A (k, k ).
q

Evidently, because of spin SU (2) symmetry, the spin susceptibilities to fields along
any other direction different from z is exactly the same as (5.62). It follows that
χ ρqp  
= 1 − AA , (5.63)
χ0 ρ0

where χ0 = ρ0 is the non-interacting spin susceptibility.

Contrary to the charge compressibility, we expect that χ > χ0 when the interaction
is repulsive. Moreover, approaching a Mott transition, and in light of the previous
discussion
1  1
AA = A↑,↑;↑,↑ − A↑,↓;↓,↑  − A↑,↓;↓,↑ → −1 ,
2 2
so that χ → 2ρqp , i.e., twice the value expected for non-interacting quasiparticles.
Strikingly, this result remains true even inside a Mott insulator that sustains a Lut-
tinger surface, and which, despite the charge gap, should display a conventional Pauli
paramagnetic susceptibility; what is commonly known as a spin-liquid insulator.

5.4.3 Specific Heat

In Sect. 4.9.1, we derived the Ward-Takahashi identity (4.137) for the heat density,
which, in the ω-limit and, as discussed earlier, for real and small frequency  reads

∂(, k)  
 E ω (, k, σ) = (k) + (, k) −  = (k) + Re (, k) − 
∂
∂Re(, k) ∂Im(, k)
+ i Im (, k) +  1 − −i
∂ ∂
∂Im(, k)
= Z (, k)−1 (, k) +  + i (, k) − i Z (, k) 
∂
 
−1 −1 E ω
 Z (, k) (, k) +  ≡ Z (, k) λ (, k, σ) ,
(5.64)
230 5 Landau’s Fermi Liquid Theory

the last almost equality deriving from the fact that both in case (F) and (L) the
neglected imaginary
 terms
 vanishes at least as 2 . Since Rqp in (5.54) implies  =
∗ (k) and  ∗ (k), k = 0, then the specific heat through (5.57) becomes
 
1 q 2  2 ∂ f ∗ (k)
cV = − χ E E (0, q) = − ∗ (k)
T TV ∂∗ (k)
k
   
4   ∂ f ∗ (k) ∂ f ∗ (k )

A S (k, k )
q
− ∗ (k) ∗ (k ) 
TV 2

∂ ∗ (k) ∂ ∗ (k )
kk
(1) (2)
≡ cV + cV .
(5.65)
(1)
The first term cV can be written as

2 ∞ ∂ f () 2π 2
(1)
cV = d A∗ () 2 −  T ρqp . (5.66)
T −∞ ∂ 3

c(2) expanding A S (k, k ) at first order in ∗ (k) ∗ (k ),

q
The second contribution V requires

q  q 
i.e., A S (k, k ) − A S k F L , k F L ∼ ∗ (k) ∗ (k ) times known, see, e.g., [18], loga-
rithmic corrections that yield c(2) (1)
V ∼ T ln T , subleading with respect to cV ∼ T .
3

Therefore, at leading order in T ,

2π 2
cV = T ρqp , (5.67)
3
which is the same expression of the specific heat c0 V of non-interacting electrons,
with the non-interacting DOS ρ0 replaced by the quasiparticle one ρqp . It is common
to quantify the strength of correlations in a Fermi liquid through the so-called Wilson
ratio, defined as
χ c0 V
RW ≡ = 1 − AA . (5.68)
χ0 cV
The Wilson ratio RW = 1 in absence of interaction, and grows with the strength
of interaction reaching RW = 2 at the Mott transition, and even inside the Mott
insulator if it has a Luttinger surface. That is also striking since it implies that such
Mott insulator has a Pauli-like paramagnetic susceptibility, linear in temperature
specific heat, and all that despite the charge gap.

5.5 Current-Current Response Functions

The densities ρ A of conserved quantities are associated via the continuity equation to
the current densities J A . Although the latter are not densities of conserved quantities,
we can still derive a simple expression of their response functions, which, in presence
5.5 Current-Current Response Functions 231

of Coulomb repulsion, correspond to the proper ones, i.e., the response to the internal
fields

1  
χi A j B (τ , q) = −  Tτ Ji A (τ , q) J j B (0, −q)  ,
V
where Ji A and Ji B , i = x, y, z, are the components of J A and J B , respectively. We
remark that, since we are considering small q and normal, i.e., non superconducting,
Fermi liquids, we can safely assume that the proper longitudinal and transverse
current-current response functions are coincident, and thus focus just on the proper
longitudinal response.
For that, we use the alternative expression (5.49) of a generic response function,
namely
   
q Aq
χi A j B = χi A j B +Tr λi qp
R λj
Bq Aq
+Tr λi qp
Rqp λ B q ,
R A j
(5.69)
where now, through (4.131), the interacting current density-vertices satisfy in the
q-limit the equation

Qq Qq ∂(, k)
λ Q q = Z (, k) 0 + 0 ,
∂k

for Q = A, B. We note that, in the response function (5.69), the kernel R qp imply
that  = ∗ (k). In our specific case, the only conserved quantities are the total charge,
spin, and energy. Therefore, for the charge current vertex, we find

  ∂k ∂(, k) ∂∗ (k)

λq = Z ∗ (k), k +  = = vk , (5.70)
∂k ∂k =∗ (k) ∂k

for the spin, e.g., the component a = x, y, z,

q σa ∂∗ (k) σa
λa = = vk , (5.71)
2 ∂k 2
with σa the Pauli matrices, and finally, for the heat current, through (4.137)

  ∂k ∂(i, k) ∂∗ (k)

λ E q = ∗ (k) Z ∗ (k), k + = ∗ (k) = ∗ (k) v k .
∂k ∂k ∂k
(5.72)
The next question concerns the q-limit of the current-current response functions
(5.69). In the case of the charge current that is not a real issue, since gauge invari-
ance implies that the q-limit of the current-current response function cancels the
diamagnetic term. In that way, we can readily derive the expression of the ω-limit
232 5 Landau’s Fermi Liquid Theory

of the conductivity, which, assuming space isotropy or cubic symmetry, and real
frequency, iω → ω + iη with η > 0 infinitesimal, is simply

e2    
σ (ω) = σ⊥ (ω) = i
q
Tr λi ∗ω
R λi
q
+ Tr λi
q ∗ω Aω R
R ∗ω λq
ω + iη i
  
e2 2  ∂ f ∗ (k)
=i vk · vk −
ω + iη 3V ∂∗ (k)
k
    (5.73)
4  ∂ f ∗ (k) ∂ f ∗ (k )
+ vk · vk  AωS (k, k )
3V 2 
∂∗ (k) ∂∗ (k )
kk
i
= D∗ δ(ω) + ,
ω

with the Drude weight

  
2e2  ∂ f ∗ (k)
D∗ = vk · vk −
3V ∂∗ (k)
k
   
4e2  ∂ f ∗ (k) ∂ f ∗ (k )
+ v k · v k AωS (k, k )
3V 2 kk ∂∗ (k) ∂∗ (k )
   (5.74)
2e2  ∂ f ∗ (k)
= vk · vk −
3V ∂∗ (k)
k
2 
   
4e ∂ f ∗ (k) ∂ f ∗ (k )
+ v k · v k f S (k, k ) ,
3V 2 kk ∂∗ (k) ∂∗ (k )

having defined

AωS (k, k ) ≡ f S (k, k ) , AωA (k, k ) ≡ f A (k, k ) . (5.75)

Assuming instead a fictitious longitudinal and static vector potential opposite for
the two spin directions, we could, similarly as in the charge case, gauge it away
through a unitary transformation, and thus conclude that also the q-limit of the spin
current-spin current response function vanishes, which allows calculating the latter
with ease.

5.5.1 Thermal Response

The thermal response is less simple to analyse, since the conserved quantity, the
total energy, is a many-body operator. First, let us discuss how thermal response
may arise. We shall do that in an unusual way. Specifically, we assume the system is
initially at equilibrium with a bath at temperature
 T , but
 later, the bath changes and
thus its temperature T → T + δT (t, r), with δT (t, r)  T and so slowly varying
5.5 Current-Current Response Functions 233

in space and time that it is still possible to describe the system density matrix just
replacing T with T + δT (t, r), thus

1 ρ E (r)
ρ̂(t) = exp − dr .
Z T + δT (t, r)

Therefore
ρ E (r) ρ E (r) δT (t, r) 1  
 − ρ E (r) = ρ E (r) + φ E (t, r) ρ E (r) ,
T + δT (t, r) T T2 T

with the dimensionless field

δT (t, r)
φ E (t, r) ≡ − , (5.76)
T
so that the density matrix
 
1 1   e−β H +δ H (t)
ρ̂(t)  exp − dr ρ E (r) + φ E (t, r) ρ E (r) =    ,
Z T
Tr e−β H +δ H (t)

with the perturbation

1 
δH = dr φ E (t, r) ρ E (r) = φ E (t, q) ρ E (−q) ,
V q

with the convention

A(q) = dr e−iq·r A(r) ,

for both φ E (t, q) and ρ E (q). At φ E = 0, the heat density at finite q acquires a finite
expectation value that, in linear response, reads

ρ E (ω, q) = χ E E (ω, q) φ E (ω, q) ,

where χ E E (ω, q) is the Fourier transform of

i  
χ E E (t, q) = − θ(t)  ρ E (t, q) , ρ E (0, −q)  ,
V
calculated on the initial state at temperature T . We observe that the change in total
energy δ E due to a static δT (r) in the uniform limit δT (r) → δT can be calculated
through

δT q
δ E = lim lim ρ E (ω, q) = lim lim χ E E (ω, q) φ E (ω, q) = −V χ ,
q→0 ω→0 q→0 ω→0 T EE
234 5 Landau’s Fermi Liquid Theory

yielding the known expression of the specific heat

1 δE 1 q
cV = = − χE E .
V δT T
On the other end, by the continuity equation
 
i ρ̇ E (t, q) = ρ E (t, q) , H = q · J E (t, q) , ω ρ E (ω, q) = q · J E (ω, q) .

so that
ω
J  E (ω, q) = χ E E (ω, q) q φ E (ω, q) . (5.77)
q2
We note that since χ E E (0, q) is a thermodynamic susceptibility, and thus finite, the
heat current vanishes at ω = 0, and therefore, the heat current-heat current response
function must vanish too, exactly like for the charge and spin. That includes evidently
also the q-limit.
However, we still need to relate (5.77) to the current-current response. For that,
we can readily show that

∂ ∂ 1 ∂ ∂    
−i  i χ E E (t − t  , q) = − i  i − i θ(t − t  )  ρ E (t, q) , ρ E (t  , −q) 
∂t ∂t V ∂t ∂t
1  
 
= δ(t − t )  q · J E (q) , ρ E (−q) 
V
  
− i θ(t − t  )  q · J E (t, q) , q · J E (t  , −q) 
1  
= δ(t − t  )  q · J E (q) , ρ E (−q)  + q 2 χ E E (t − t  , q) ,
V
(5.78)
where χ E E is the longitudinal component of the heat current-heat current response
function. We further note that, at T = 0,

1   1  
0 | q · J E (q) , ρ E (−q) | 0 = 0 | ρ E (q) , H , ρ E (−q) | 0
V V
1 E0
= 0 | ρ E (q) H ρ E (−q) | 0 − 0 | ρ E (q) ρ E (−q) | 0
V V
≡  E (q) S E (q) ,

where
1 0 | ρ E (q) H ρ E (−q) | 0 E0
 E (q) ≡ − ,
V 0 | ρ E (q) ρ E (−q) | 0 V
is the excitation energy per unit volume of an heat-density fluctuation at momentum
q and, in a Fermi liquid, vanishes linearly in q as q → 0, while

S E (q) = 0 | ρ E (q) ρ E (−q) | 0 ,

5.5 Current-Current Response Functions 235

is the heat-density structure factor, which also vanishes linearly in a Fermi liquid
with gapless excitations. Therefore, at small q,  E (q) S E (q) ∼ q 2 , and we can write

1  
0 | q · J E (q) , ρ E (−q) | 0 = q 2 N E (q) .
V

with N E (0) finite, so that, upon Fourier transform (5.78), we find

 
ω 2 χ E E (ω, q) = q 2 χ E E (ω, q) + N E (q) , (5.79)

which is the counterpart of the similar equation for the charge density, in which case
N E (q) → n/m, the diamagnetic term. It thus follows that, through (5.76),

ω   q φ (ω, q)
E
J  E (ω, q) = χ E E (ω, q) q φ E (ω, q) = χ E E (ω, q) + N E (q)
q2 ω
i   ∇δT (ω, r)
= χ E E (ω, q) + N E (q) dr e−q·r .
ω T

Let us take the limit q → 0 assuming that that ∇δT (ω, r) → ∇δT (ω) remains finite
and space independent, thus

1 1  
J  E (ω, 0) = − χ E E (ω, 0) + N E (0) ∇δT (ω) ≡ K (ω) ∇δT (ω) ,
V iω T
(5.80)
which defines the thermal conductivity K (ω → 0). Since, as we earlier demon-
q
strated, χ E E + N E (0) = 0, through (5.69), we readily find

  
1 i 2  ∂ f ∗ (k)
K (ω) = ∗ (k)2 v k · v k −
T ω + iη 3V ∂∗ (k)
k
   
4  ∂ f ∗ (k) ∂ f ∗ (k )
+ ∗ (k) ∗ (k ) v k · v k f S (k, k )
3V 2 
∂∗ (k) ∂∗ (k )
kk
i
= K∗ δ(ω) + ,
ω
(5.81)
thus a thermal Drude weight K ∗ ∼ T .
We note that the sign in (5.80) is opposite with respect to the usual expression
J E = −V K ∇T . The reason lies in the way we have added the thermal perturba-
tion. Namely, we have assumed that the system is initially in equilibrium at uniform
temperature T , and later on, a temperature drop across the opposite sides of the
sample is applied as a perturbation. On the contrary, the common way to measure
heat conductivity is in a sample that, at equilibrium, is already in contact with two
baths, one at higher temperature than the other. To appreciate the difference, we can
consider the equivalent case of an electrochemical potential drop. If the system is ini-
tially at equilibrium with constant electrochemical potential −e μ, thus Hamiltonian
236 5 Landau’s Fermi Liquid Theory

H − e μ N , and a δμ(t, r) is switched on, such that ∇δμ(t) = ∇μ(t) is constant in

space, then for ∇μ(t) → ∇μ, namely, in the ω-limit, we would find

J = V σ ∇μ . (5.82)

Therefore, in this set up, electrons flow towards the side with larger chemical poten-
tial, and thus J is parallel to ∇μ. On the contrary, should we assume a sample that,
at equilibrium, had an electrochemical potential drop, electrons would flow from the
side with larger μ towards that with lower one, thus J antiparallel to ∇μ.

5.5.2 Coulomb Interaction

The whole derivation above outlined in Landau’s Fermi liquid Theory is based on the
non-analytic properties of the kernel R in (5.18). A long-range Coulomb interaction
yields an additional singularity at small q of all improper vertices, both interaction
and density ones. However, if we concentrate just on the proper vertices, we can repeat
step-by-step the whole derivation, which implies that the density-density response
functions (5.55) of densities associated with conserved quantities remain the same
if we regard them as proper density-density response functions.
For instance, the proper density-density response function is still given by (5.55),
and reads

 
2e2  ∂ f ∗ (k) q · vk
χ∗ (ω, q) = −
V ∂∗ (k) ω − q · vk
k
  
2  ∂ f  (k) ∂ f  (k )

4e ∗ ∗ q · vk q · v k
+ 2 A S (k, k ; ω, q) ,
V 
∂∗ (k) ∂∗ (k ) ω − q · vk ω − q · v k
kk
(5.83)
while the improper one is, as usual,

χ∗ (ω, q) χ∗ (ω, q)
χ(ω, q) = ≡ .
4πe 2  (ω, q)
1− χ∗ (ω, q)
q2

5.6 Mott Insulators with a Luttinger Surface

With the inclusion of Coulomb repulsion, we can now discuss more realistically what
a Mott insulator with a Luttinger surface implies for transport properties, and more
importantly, whether the resulting behaviour is physically consistent, e.g., compati-
ble with gauge invariance.
5.6 Mott Insulators with a Luttinger Surface 237

The longitudinal current-current proper response function can be written either

using (5.48) or (5.49), thus
   
χ∗ = χω∗ + Tr λω Rqp λω + Tr λω Rqp A Rqp λω
    (5.84)
q
= χ∗ + Tr λ
q qp
R
q
λ + Tr λ
q qp
R A qp
R
q
λ ,

where
q · λω q · vk
λω =
q
, λ = −e ,
q q
the last equation deriving from the Ward-Takahashi identity.
In a lattice model where a Mott transition at half-filling may occur, gauge invari-
ance (2.79) implies, similar to (5.78), that, for small q

1  
ω 2 χ∗ (ω, q) = q 2 χ∗ (ω, q) +  q · J(q) , ρ(−q)  ,
V
where
 
†
 †
−e k+q − k ckσ ck+qσ −−−→ q · J(q) , ρ(−q) = −e ck+qσ ckσ .
q→0
kσ kσ

Therefore

e2   
† 1  
k+q + k−q − 2k  ckσ ckσ  −−−→  q · J(q) , ρ(−q)  ≡ q 2 T ,
V q→0 V
kσ

where
e2   qi q j ∂ 2 k
3
†
T ≡ 2 ∂k ∂k
 ckσ ckσ  ,
V q i j
kσ i, j=1

is the diamagnetic term of the lattice model. It follows that gauge invariance entails

q2  
q
χ∗ (ω, q) + T = χ∗ (ω, q) , χ∗ = −T , (5.85)
ω2
while, through (5.84), the optical conductivity read

i   i
σ|| (ω, q) = e2 χ∗ (ω, q) + T −−−−→ σ||ω = D∗ .
ω + iη ω−limit ω + iη

In the Mott insulator, D∗ = 0 and that implies χω∗ = −T , which is not surpris-
ing since the non-analyticity at ω = q = 0 typical of metal must disappear in the
insulator.
238 5 Landau’s Fermi Liquid Theory

If we now take the ω-limit and use the second equation in (5.84), or the q-limit
and use instead the first equation, we find that
   
χω∗ = −T + Tr λ
q qp
R ω q q
λ + Tr λ R qp
ω
Aω qp
R ω q
λ ,
   
= −T = χω∗ + Tr λω λω + Tr λω λω .
q
χ∗ q
Rqp q
Rqp Aq q
Rqp

Therefore, in a Mott insulator where χω∗ = −T , and recalling that

 
∂ f ∗ (k)
qp
R ω
= −Rqp
q
=− ≡ −f,
∂∗ (k)

the following two equations are fulfilled:

   
f  λ + Tr λ f  Aω f  λ ,
q q q q
0 = −Tr λ
    (5.86)
0 = Tr λω f  λω + Tr λω f  Aq f  λω .

On the other hand, the density-density proper response function reads

   
χ∗ = Tr λω Rqp λω + Tr λω Rqp A Rqp λω , (5.87)

where λω = −e by the Ward-Takahashi identity. The charge compressibility is, there-

fore
   
κ = −χ∗ = −Tr λω Rqp λω − Tr λω Rqp λω
q q q
Aq Rqp q
   
= −Tr λω f  λω − Tr λω f  Aq f  λω ,

and must vanish in the Mott insulator, yielding an additional equation

   
0 = Tr λω f λω + Tr λω f Aq f λω , (5.88)

which, compared with the second equation in (5.86), and recalling (5.85), implies
that
ω ω
λω = λω = −e , (5.89)
q q
in a Mott insulator, besides the already discussed property A S = 1.
5.6 Mott Insulators with a Luttinger Surface 239

The first equation in (5.86) reads instead, since Aω = f, the Landau f-parameter,
      
1  ∂ f ∗ (k) 2  ∂ f ∗ (k ) 
0= − q · vk q · vk − q · v k f S (k, k )
V ∂∗ (k) V  ∂∗ (k )
k k
      
1  ∂ f ∗ (k) 2  ∂ f ∗ (k ) 
⇒0= − vk · vk − v k f S (k, k )
V ∂∗ (k) V  ∂∗ (k )
k k
  
1  ∂ f ∗ (k)
= − vk · vk ,
V ∂∗ (k)
k
(5.90)
where the last two equations follow from the fact that in the ω-limit longitudinal and
transverse response coincide, and thus the Ward-Takahashi identity actually implies
λω = −e v k , thus the second and third equations. In particular, the latter equation
states that the flux of v k out of the Luttinger surface vanishes.

The density-density proper response function in (5.87) becomes in the ω-limit

 
2e2  ∂ f ∗ (k)  q · v k 2
χω∗−
V ∂∗ (k) ω
k
   
4e2  ∂ f ∗ (k) ∂ f ∗ (k ) q · v k q · v k q2
+ 2 f (k, k) ≡ Tqp ,
∂∗ (k )
S
V kk ∂∗ (k) ω ω ω2
(5.91)
where Tqp is actually the quasiparticle contribution to the diamagnetic term T , and
must vanish in the Mott insulator. Indeed, in the same ω-limit, the longitudinal
conductivity, which can be also written as
ω   ω
σ|| (ω, q) = || (ω, q) − 1 = − 2 χ∗ (ω, q) ,
4π i iq

becomes
i i
σω = Tqp ≡ D∗ , (5.92)
ω + iη ω + iη
which does imply that Tqp = 0 in the Mott insulator. We observe that Tqp = 0 from
(5.91) is equivalent to the first equation in (5.90).
Therefore, a Mott insulator with a Luttinger surface sustaining well-defined quasi-
particles is perfectly consistent with gauge invariance, and just requires two simple
conditions to be verified

      
1  ∂ f ∗ (k) 2  ∂ f ∗ (k ) ) = 0 ,
− vk · vk − v k f (k, k AS = 1 .
∂∗ (k )
S
V ∂∗ (k) V 
k k
(5.93)
240 5 Landau’s Fermi Liquid Theory

As earlier discussed, in the Mott insulator, we expect A↑,↑;↑,↑ = 0, and thus also
f ↑,↑;↑,↑ = 0. That implies f A = − f S , and therefore, if the charge Drude weight
(5.74) vanishes because of (5.93), the spin Drude weight Dσ
      
2μ2B  ∂ f ∗ (k) 2  ∂ f ∗ (k ) 
Dσ = − vk · vk − v k f A (k, k )
3V ∂∗ (k) V  ∂∗ (k )
k k
      
2μ2B  ∂ f ∗ (k) 2  ∂ f ∗ (k ) 
 − vk · vk + v k f S (k, k )
3V ∂∗ (k) V  ∂∗ (k )
k k
  
4μ2B  ∂ f ∗ (k)
= − vk · vk ,
3V ∂∗ (k)
k
(5.94)
is finite and consistent with χ = 2 ρqp , which is indeed remarkable.
Let us now discuss more in detail the thermal properties. We already showed that
also in a Mott insulator quasiparticles at a Luttinger surface yield a leading linear in
temperature specific heat
2π 2
cV =
T ρqp . (5.95)
3
We here rewrite for convenience the thermal Drude weight of (5.81),

  
1 2  ∂ f ∗ (k)
K∗ = ∗ (k)2 v k · v k −
T 3V ∂∗ (k)
k
   
4  ∂ f ∗ (k) ∂ f ∗ (k )
+ ∗ (k) ∗ (k ) v k · v k f S (k, k ) ,
3V 2 kk ∂∗ (k) ∂∗ (k )

since it highlights that, just like in the case of specific heat, the leading in temperature
contribution derives just from the first term in the square brackets. Therefore, in
leading order
  
1 2  ∂ f ∗ (k)
K∗  ∗ (k)2 v k · v k − , (5.96)
T 3V ∂∗ (k)
k

and again is consistent with the specific heat (5.95).

In conclusion, a Mott insulator that displays a Luttinger surface is expected to
have gapless quasiparticles, which do not contribute to charge properties, thus the
insulator is correctly incompressible and has vanishing Drude weight, and yet yield
the spin and thermal properties expected in a conventional Fermi liquid. This striking
scenario does not seem to contradict fundamental properties, like gauge invariance,
and therefore, it may be in principle realised. The obvious caveat is that the existence
5.7 Luttinger’s Theorem and Quasiparticle Distribution Function 241

of a Mott insulator that has a Luttinger surface and does not break any symmetry is
highly hypothetical. However, the discussion of this section proves that at least such
possibility is contemplated within Landau’s Fermi liquid theory.

5.7 Luttinger’s Theorem and Quasiparticle Distribution

Function

Under the assumption (5.6), which corresponds to the assumption (4.167) in

Sect. 4.11.1.1, the electron density is obtained by the generalised Luttinger’s theorem

N 1   i0+ ∂ ln G(i, k) L
=− T e +
V V ∂i V
kσ i
(5.97)
1  ∞ d ∂ Im ln G( + i0+ , k) L
= f () + ,
V −∞ π ∂ V
kσ

with integer L and for T → 0. We define the real frequency counterpart of (4.155),

−ImG σ ( + i0+ , k)
δσ (, k) ≡ π + Im ln G σ ( + i0+ , k) = tan−1 ∈ [0, π] ,
−ReG σ ( + i0+ , k)
(5.98)
where δσ (−∞, k) = 0, δσ (∞, k) = π, and, for convenience, we have indicated the
spin label though the Green’s function is independent of it, so that
N 1  ∞ d ∂ δσ (, k) L
= f () + . (5.99)
V V −∞ π ∂ V
kσ

In the perturbative regime L = 0, and we can legitimately regard

∞ d ∂ δσ (, k) ∞ d ∂ f ()
n kσ ≡ f () =− δσ (, k) , (5.100)
−∞ π ∂ −∞ π ∂

as the quasiparticle distribution function in momentum space, not to be confused with

†
the electron distribution function  ckσ ckσ . When perturbation theory breaks down,
thus L = 0, (5.100) is not valid anymore. However, we here show that any variation
of the actual quasiparticle distribution function with respect to a thermodynamic
variable λ Q conjugate to a conserved quantity Q is given by

∂n kσ ∞ d ∂ ∂ f ()
=− δσ (, k) . (5.101)
∂λ Q −∞ π ∂λ Q ∂

This result can be anticipated without even proving it. Indeed, since L must be an
integer, it cannot vary continuously upon varying λ Q . Therefore, ∂ L/∂λ Q = 0, and
thus (5.101) follows. However, the explicit proof is a useful exercise.
242 5 Landau’s Fermi Liquid Theory

In our case of interest, the conserved quantities are the charge, the spin, and the
total energy, thus the corresponding λ Q are, respectively, the chemical potential μ,
the Zeeman splitting field μ Z , and β = 1/T . For simplicity, we hereafter consider
the chemical potential case. Let us, therefore, calculate

∞ ∞  
∂n kσ d ∂ f () ∂δσ (, k) d ∂ f () ∂σ (, k)
=− = Im G σ (, k) 1 − ,
∂μ −∞ π ∂ ∂μ −∞ π ∂ ∂μ
(5.102)
where G(, k) ≡ G( + i0+ , k) and similarly for (, k), and we recall that, under
changing μ → μ + δμ

∂σ (, k)
G σ (, k)−1 →  − k + δμ − σ (, k) − δμ .
∂μ
On the other hand, by definition of the irreducible interaction vertex and moving
back to Matsubara frequencies

∂σ (i, k) 1     ∂G σ (i , k )

= T 0 i k σ, i k σ  ; i k σ  , i k σ ,
∂μ V   
∂μ
kσ 

or, in shorthand notation,

∂σ ∂G σ ∂σ
= 0 = −0 G2 1− .
∂μ ∂μ ∂μ

Therefore, in sequence
  ∂σ ∂σ  −1
1 − 0 G2 = −0 G2 ⇒ = − 1 − 0 G2 0 G2
∂μ ∂μ
∂σ
⇒ = − G2 ,
∂μ
thus

∂σ (i, k) 1    
1− =1+ T  i k σ, i k σ  ; i k σ  , i k σ G σ (i , k )2
∂μ V    kσ 
=1+ q
R = q ,
q

(5.103)
where the last equation derives from the fact that adding a chemical potential is
equivalent to adding a static and a long-wavelength potential and then take the limit
of infinite wavelength, thus just the q-limit.
Let us now recall the Ward-Takahashi identity (4.131), see Fig. 4.42, which yields
for the charge density-vertex and in the ω-limit

∂σ (i, k)
1− = ω , (5.104)
∂i
5.7 Luttinger’s Theorem and Quasiparticle Distribution Function 243

We recall (5.39) for the charge density-vertex

 = ω + ω  ,

which reads in the q-limit

    ∂σ
q = 1 +  q q ω = 1 +  q q 1−
∂i
  ∂σ
= Z −1 1 + Aq q
Rqp =1− ,
∂μ

so that, see (5.102),

 
∂δσ (, k) ∂σ (, k)
= −Im G σ (, k) 1 −
∂μ ∂μ
   
−1 1  q  ∂ f ∗ (k )
= −Im G σ (, k) Z (, k) 1+ A   (k, k )
V   σ,σ ;σ ,σ ∂∗ (k )
kσ
 
1  q ∂ f ∗ (k )
= π Aqp (, k) 1 + Aσ,σ  ;σ  ,σ (k, k ) ,
V   ∂∗ (k )
kσ

and thus, recalling (5.13),

1  ∂n kσ 1  ∞
d ∂ f () ∂δσ (, k)
=−
V ∂μ V −∞ π ∂ ∂μ
kσ kσ
   
1  ∂ f ∗ (k) 1  q 
 ∂ f ∗ (k )
=− 1+ A   (k, k ) ,
V ∂∗ (k) V   σ,σ ;σ ,σ ∂∗ (k )
kσ kσ
(5.105)
which is just the charge compressibility (5.58), and thus proves that (5.101) is true
for λ Q = μ. One can straightforwardly prove the same result also for a Zeeman field
λ Q = μ Z . The case λ Q = β is slightly more complicated but can be worked as well.
For that, it is important to notice the following equation:
 
∂σ (, k) 1  ∂ f ∗ (k )
= Z (, k)−1 
Aσ,σ ;σ ,σ (k, k ) ,
q
d (5.106)
∂T V  ∂T
kσ

which can be derived by inspection of the perturbative expansion of the self-energy

in terms of skeleton diagrams.
244 5 Landau’s Fermi Liquid Theory

Therefore, even though in general

∞ d ∂ δσ (, k)
n kσ = f () −−−→ δσ (0− , k) , (5.107)
−∞ π ∂ T →0

unless in the perturbative regime, upon varying the chemical potential of each spin
species, μσ = μ → μ + δμσ , or the temperature T  0,

 1 ∂δσ (0− , k) 1 ∂δσ (0− , k)

δn kσ = δμσ + δT , (5.108)

π ∂μσ π ∂T
σ

is instead true irrespective of whether perturbation theory is valid or breaks down,

provided (5.6) holds. In other words, the variation of the quasiparticle distribution
function is just the variation of the phase of the Green’s function G σ (0− , k) in units
of π.
Let us unveil the meaning of the above result in the case of a Mott insulator of
Sect. 5.6, in which case

 δn kσ  1 ∂δσ (0− , k)
δN = δμ= δμ = 0 ,
δμ π ∂μ
kσ kσ
 δn k↑ δn k↓  1 ∂δ↑ (0− , k) ∂δ↓ (0− , k)
δM = − δμ Z = − δμ Z = 0 ,
δμ Z δμ Z π ∂μ Z ∂μ Z
k k
 δn kσ  1 ∂δσ (0− , k)
δE = ∗ (k) δT = ∗ (k) δT = 0 .
δT π ∂T
kσ kσ
(5.109)
Since the system has a single-particle gap, δσ (0− , k) is equal, e.g., to π inside the Lut-
tinger surface of spin-σ quasiparticles and 0 outside, and that must remain true even
under the perturbations δμ, δμ Z and δT . It follows that the vanishing compressibility,
δ N /δμ = 0 in (5.109), implies that the volume enclosed by the Luttinger surface
does not change upon small variations of the chemical potential. On the contrary,
the relative volumes of spin-up and spin-down quasiparticles are free to change in a
Zeeman field, provided the sum of the two volumes remains constant. Similarly, the
finite specific heat, δ E/δT = 0 in (5.109), implies that the shape changes at fixed
volume upon varying temperature.

5.7.1 Oshikawa’s Topological Derivation of Luttinger’s Theorem

In periodic systems, one can derive Luttinger’s theorem in a much simpler and
physically more transparent way [19]. For simplicity, we here consider a two dimen-
sional lattice model with a torus geometry, see Fig. 5.3, and, in order to avoid any
complication related to the difference between internal and external electromagnetic
5.7 Luttinger’s Theorem and Quasiparticle Distribution Function 245

Fig. 5.3 A two-dimensional

system with a torus
geometry. A magnetic field,
concentrated in the centre of y
the torus, is adiabatically Φ(t)
switched on till the flux x R
threading the torus reaches
the flux quantum
0 = 2π/q

fields, we assume a fictitious flux (t) that threads the torus and is coupled to fic-
titious spin-dependent charges −qσ , with either q↑ = q↓ = q or q↑ = −q↓ = q to
discuss separately spin and charge response. We further assume that (t) slowly
grows in time from 0 at t = 0 to a unit flux quantum 0 = 2π/q at t = τ .4 That
amounts to add a fictitious vector potential A(t) = (t)/L x , where, see Fig. 5.3,
L x = 2π R = N x a, with N x(y) the number of unit cells along x(y), and a the lattice
spacing. The corresponding Hamiltonian reads
  1 
H (t) = d x d y σ† (x, y) − i ∂x + qσ A(t) − 2 ∂ y2 σ (x, y)
2m σ

+ d x d y σ† (x, y) V (x, y) σ (x, y) + Hint ,
σ
(5.110)
where V (x, y) = V (x + na, y + ma) is the periodic potential, and Hint the trans-
lationally invariant interaction. During the switching time, (t) drives currents into
the system, and the initial state, assumed to be the ground state | 0  of H [0], evolves
according to
t
i
dt  H [(t  )]
| (t) = e−  0 | 0  ≡ U (t) | 0  .
After the flux has reached 0 , the system flows to a stationary state described by the
wavefunction
i
| ∗ (t) = e−  H [0 ](t−τ ) | (τ ) . (5.111)
At  = 0, the primitive lattice translation operator along x is defined through Tx [0] =
e−i Px a/ , where Px is the x-component of the lattice momentum operator. Since
H [(t)] commutes with Tx [0], if the initial ground-state wavefunction is eigenstate
of Px with eigenvalue Pi x , then

4 In presence of a static and uniform flux , the momentum quantisation in the gauge in which the

vector potential A  x ⊥ B reads, cfr. (5.110) with qσ → e/c,

2π 2π e 
kx = nx −
→ nx + , nx ∈ Z .
Nx a Nx a c Nx a
We note that, if  = 0 ≡ 2πc/e the magnetic flux quantum, the quantisation is invariant, simply
n x → n x + 1.
246 5 Landau’s Fermi Liquid Theory

i
Tx (0) | ∗ (t) = e−  H [0 ](t−τ )
U (τ ) Tx (0) | 0  = e−i Pi x a/ | ∗ (t) ,

thus | ∗ (t) is also eigenstate with same eigenvalue. Note that lattice translation
implies that
 Mi x
Pi x = 2π , Mi x ∈ Z , (5.112)
a Nx
is quantised. However, at finite , the actual translation operator Tx [] = Tx [0] is
gauge dependent. Since a static flux  can be reabsorbed by a gauge transformation

 
H [] = UG []† H [0] UG [] , UG [] = exp i dr qσ x ρσ (r) ,
 Lx σ

then
 a 0 q σ
Tx [0 ] = UG† [0 ] Tx [0] UG [0 ] = Tx [0] exp − i Nσ
σ
 Lx
2π  qσ
= Tx [0] exp − i Nσ ,
Nx σ q

and thus
2π  qσ
−i
Tx [] | ∗ (t) ≡ e−i P f x a/ | ∗ (t) = e
Nσ
Nx σ q Tx [0] | ∗ (t)
2π  qσ
−i
e−i Pi x a/ | ∗ (t) .
Nσ
=e Nx σ q

In conclusion
2π  qσ
P f x = Pi x + Nσ ≡ Pi x + Px , (5.113)
Nx a σ q

and it is still quantised as in the absence of flux because 0 is a unit flux quantum.
In other words, the adiabatic threading of a unit flux quantum transforms the initial
ground state with lattice momentum Pi x along x into a final state with different lattice
momentum P f x .
Let us now assume to be in the perturbative regime, and thus that Landau’s adia-
batic principle holds true. In that case, the above fictitious charges qσ are
 inherited
by quasiparticles, and the final state just corresponds to ∗ (k) → ∗ kσ , where

qσ 2π
kσ = k + , (5.114)
q Nx a

so that the lattice momentum, assumed to vanish in the ground state at  = 0, changes
for q↑ = q↓ into
2π
Px = N Fqp , (5.115)
Nx a
5.7 Luttinger’s Theorem and Quasiparticle Distribution Function 247

where N Fqp is twice the number of k-points within the Fermi volume. Therefore,
comparing (5.113) for q↑ = q↓ = q, namely

2π   2π
Px = N↑ + N↓ = N, (5.116)
Nx a Nx a

with (5.115), we conclude that N = N Fqp , which is just the conventional statement
of Luttinger’s theorem.
Let us consider now the non-symmetry breaking half-filled Mott insulator with
a Luttinger surface discussed in Sect. 5.6. Here, we cannot invoke Landau’s adia-
batic principle, so that we cannot anticipate what is the momentum acquired by the
quasiparticles at the Luttinger surface when q↑ = q↓ , which we can write as

2π
Px = N Lqp , (5.117)
Nx a

in terms of an unknown N Lqp . On the other hand, (5.116) implies, for N = N x N y

that
2π
Px = Nx N y , (5.118)
Nx a
and thus N Lqp = N x N y . Moreover, since Px is a multiple of the primitive recipro-
cal lattice vector 2π/a, it is equivalent to zero, which is consistent with the system
being insulating, and thus having a charge gap. Indeed, since H [0] and H [0 ] have
the same spectrum, and if the ground state is unique, we can always design an adi-
abatic switching protocol such that the initial state evolves into itself, hence the
acquired momentum must strictly vanish.
On the other hand, we could follow a different route and repeat the same hypo-
thetical experiment taking, e.g. q↑ = q but q↓ = 0, in which case we would find,
since N↑ = N x N y /2,

2π N↑ π 0 N y even ,
Px = Ny = N y ≡ π (5.119)
a Nx N y a a N y odd ,

which, for odd N y , may look in contradiction with the previous physical argument,
but in reality it is not so. The reason is that the Luttinger surface entails the existence
of quasiparticles that, as we showed in Sect. 5.6, do not contribute to the charge
response but yields conventional thermal and magnetic properties. Therefore, while
for q↑ = q↓ the flux insertion affects the charge, whose fluctuations are suppressed
in a Mott insulator, for q↑ = q and q↓ = 0 the spin degrees of freedom are heavily
involved and can freely absorb a π/a momentum without any cost in energy.
Let us finally consider the metallic case in which Fermi pockets and Luttinger
surfaces coexist. Here, we may reasonably argue that the quasiparticles at the Fermi
248 5 Landau’s Fermi Liquid Theory

surfaces carry the same quantum numbers of the physical particles. It follows that
equating (5.116) with the sum of (5.115) and (5.117)

2π 4π 2π 2π

Px = N= Nx N y ν ≡ N Fqp + N Lqp
Nx a Nx a Nx a Nx a
2π 2π
= N Fqp + Nx N y ,
Nx a Nx a

where the equivalence holds apart from a reciprocal lattice

 vector. Since the hole
and electron Fermi pockets contain, respectively, N x N y 1 − v H P and N x N y v E P
k-points, then

4π 4π 1 4π 1

Nx N y ν ≡ Nx N y v E P + 1 − v H P + ≡ Nx N y v E P − v H P + ,
Nx a Nx a 2 Nx a 2

which is just (5.16).

5.8 Quasiparticle Hamiltonian and Landau-Boltzmann

Transport Equation

Since G qp in (5.45) can be regarded as the non-interacting quasiparticle Green’s

function, (5.55) looks like the linear response function within the conserving Hartree-
Fock approximation of Sect. 4.10.1. It follows that the quasiparticle interaction vertex
A must satisfy a Bethe-Salpeter equation like that in Fig. 4.44, namely

A= f + f Rqp A, (5.120)

with f the Hartree-Fock irreducible interaction vertex

f = f σ1 ,σ2 ;σ3 ,σ4 (k + q, k ; k + q, k) . (5.121)

Note that, because of spin SU (2), σ1 + σ2 = σ3 + σ4 in (5.121). Therefore, if we

consider the quasiparticle Hamiltonian

 †
H∗ ≡ ∗k dkσ dkσ
kσ
1  
+ Uσ1 ,σ2 ;σ3 ,σ4 (k + q, k ; k + q, k) dk+qσ
†
d † d  d ,
2V 1 k σ2 k +qσ3 kσ4
 σ σ σ σ
kk q 1 2 3 4
(5.122)
†
where dkσ and dkσ are the quasiparticle annihilation and creation operators, and the
interaction U such that

f σ1 ,σ2 ;σ3 ,σ4 (k + q, k ; k + q, k) = Uσ1 ,σ2 ;σ3 ,σ4 (k + q, k ; k + q, k)

− Uσ1 ,σ2 ;σ4 ,σ3 (k + q, k ; k, k + q) ,
5.8 Quasiparticle Hamiltonian and Landau-Boltzmann Transport Equation 249

then all previous results imply that the conserving Hartree-Fock approximation yields
reliable linear response functions of the physical electrons.
Within the Hartree-Fock approximation, the energy ∗ (k) of the quasiparticle at
momentum k and spin σ must therefore be identified with

1 
∗ (k) = ∗k + f σ,σ ;σ ,σ (k, k ; k , k) n 0k σ , (5.123)
V  
kσ

where
 
n 0k σ =  dk† σ dk σ 0 = f ∗ (k) , (5.124)
is the equilibrium distribution function. The total energy thus reads
 1 
E0 = ∗k n 0kσ + f σ,σ ;σ ,σ (k, k ; k , k) n 0kσ n 0k σ .
2V  
kσ kk σσ

An excited state with quasiparticle distribution n kσ = n 0kσ + δn kσ has instead total

excitation energy

   1 
δ E {δn kσ } = ∗k n kσ + f σ,σ ;σ ,σ (k, k ; k , k) n kσ n k σ
2V
kσ kk σσ 
 1 
− ∗k n 0kσ − f σ,σ ;σ ,σ (k, k ; k , k) n 0kσ n 0k σ
2V
kσ kk σσ 
 1 
= ∗ (k) δn kσ + f σ,σ ;σ ,σ (k, k ; k , k) δn kσ δn k σ ,
V
kσ kk σσ 
(5.125)
which is just Landau’s energy functional (5.1) after defining

f σ,σ ;σ ,σ (k, k ; k , k) ≡ f kσ,k σ .

The parameters f in (5.121) are just the Landau’s f -parameters in a more general
spin SU (2) formulation. Since R∗ω = 0, see (5.44), it follows that

Aω = lim lim Aσ1 ,σ2 ;σ3 ,σ4 (k, k ; ω, q) = f σ1 ,σ2 ;σ3 ,σ4 (k, k ; k , k) , (5.126)
ω→0 q→0

namely, the f -parameter is just the ω-limit of the quasiparticle interaction vertex,
related through (5.46), see also Fig. 5.2, to the physical electron vertex . Similarly
to (5.56) we can define f is the S = 0 and S = 1 channels, specifically

1 
f S (k, k ) = f ↑,↑;↑,↑ (k, k ; k , k) + f ↑,↓;↓,↑ (k, k ; k , k) ,
2
1  (5.127)

f A (k, k ) = f ↑,↑;↑,↑ (k, k ; k , k) − f ↑,↓;↓,↑ (k, k ; k , k) ,
2
250 5 Landau’s Fermi Liquid Theory

which are just the functions defined in (5.75). The Bethe-Salpeter equation (5.120)
becomes evidently diagonal in the S = 0 and S = 1 channels.
We end remarking that, since δn kσ = δ Im ln G σ (0− , k)/π, Landau’s energy
functional is actually a functional of the phases of the Green’s functions at zero
energy.

5.8.1 Landau-Boltzmann Transport Equation for Quasiparticles

The Ward-Takahashi identities imply that, when the physical electron Hamiltonian
is perturbed, H → H + δ H A (t), with

  †   
δ H A (t) = V A (t, q) ck+qσ 1
0A k + q σ1 , k σ2 ckσ2 = V A (t, q) ρ A (−q) ,
kq σ1 σ2 q

where, by definition,

1 
V A (t, q) = dr e−iq·r V A (t, r) , V A (t, r) = eiq·r V A (t, q) ,
V q
(5.128)
if ρ A (q) is associated to a conserved quantity, i.e., ρ A (0) commutes with H , then
that corresponds to perturbing the quasiparticle Hamiltonian (5.122) with
  †  
δ H∗A (t) = V A (t, q) dk+qσ 1
0A k + q σ1 , k σ2 dkσ2 . (5.129)
kq σ1 σ2

In other words, external fields couple to physical electron densities associated with
conserved quantities as they do with the corresponding quasiparticles densities. Said
differently, quasiparticles carry the same conserved quantum numbers as the physical
electrons.
In our case, the only conserved quantities besides the total energy are the total
electron number N , the total spin as well as its component Sa along an arbitrary
quantisation axis a = x, y, z. The physical electron densities associated with N and
2S = 2(Sx , S y , Sz ) are
 †
 †
ρ(q) = ckσ ck+qσ , σ(q) = ckα σ αβ ck+qβ ,
kσ kαβ

 
where σ = σ1 , σ2 , σ3 , with σi , i = 1, 2, 3, the Pauli matrices. Without loss of
generality, we can specify the quantisation axis as the z-one, and consider instead
 †
ρσ (q) = ckσ ck+qσ , σ = ↑, ↓ ,
k
5.8 Quasiparticle Hamiltonian and Landau-Boltzmann Transport Equation 251

so that the perturbation

  †
δ H (t) = Vσ (t, q) ρσ (−q) −−−−−−−→ δ H∗ (t) = Vσ (t, q) dk+qσ dkσ ,
quasiparticles
qσ kqσ

which implies that H∗ → H∗ (t) with

 †
H∗ (t) = H∗ + Vσ (t, q) dk+qσ dkσ . (5.130)
kqσ

We can now simply apply the time-dependent Hartree-Fock approximation of

Sect. 3.3 to calculate the equation of motion of the expectation value n kσ (t, q) =
†
 dkσ (t) dk+qσ (t)  at linear order in the perturbation, which implies also at linear
order in the deviation from equilibrium

δn kσ (t, q) = n kσ (t, q) − δq0 n 0kσ ,

and get

∂δn kσ (t, q)    
= −i ∗ (k + q) − ∗ (k) δn kσ (t, q) − i n 0kσ − n 0k+qσ Vσ (t, q)
∂t
  1 
− i n 0kσ − n 0k+qσ f kσ,kσ δn k σ (t, q)
V  
kσ
 
∂ f ∗ (k) 1 
 −i v k · q δn kσ (t, q) + i v k · q Vσ (t, q) + f kσ,kσ δn k σ (t, q) ,
∂∗ (k) V  
kσ
(5.131)
the last equation valid at small q. Upon defining the Wigner distribution

1  iq·r
n kσ (t, r) = e n kσ (t, q) , n kσ (t, q) = dr e−iq·r n kσ (t, r) ,
V q

under the assumption

  that q  k  k F L , which also implies that, at equilibrium,
n 0kσ (t, r) = f ∗ (k) /V , and, through (5.128), (5.131) becomes, in real space and
time,

∂δn kσ (t, r)
+ v k · ∇δn kσ (t, r)
∂t  
1 ∂ f ∗ (k) 
− v k · ∇ Vσ (t, r) + f kσ,kσ δn k σ (t, r) = Ikσ (t, r) ,
V ∂∗ (k)
k σ 
(5.132)
where we also added a collision integral Ikσ (t, r) defined as the probability per unit
time that a quasiparticle with momentum k and spin σ is created in an infinitesimal
volume around r and infinitesimal time interval around t minus the probability that
it is destroyed. The collision integral is due to scattering off impurities and lattice
252 5 Landau’s Fermi Liquid Theory

vibrations, as well as to interaction effects not accounted for by the Hartree-Fock

approximation, specifically, the imaginary part of the quasiparticle self-energy yield-
ing the quasiparticle decay rate.
We observe that (5.132) is the conventional Boltzmann transport equation

∂n kσ (t, r) ∂∗ kσ (t, r) ∂n kσ (t, r)

+ ∇n kσ (t, r) · − · ∇∗ kσ (t, r) = Ikσ (t, r) ,
∂t ∂k ∂k
(5.133)
linearised in the deviation from equilibrium, where

∗ kσ (t, r) = ∗ (k) + f kσ,k σ δn k σ (t, r) + Vσ (t, r) ≡ ∗kσ (t, r) + Vσ (t, r) ,
k σ 

is the quasiparticle energy in presence of the perturbation and of other excited quasi-
particles. Alternatively, we can define a deviation from local equilibrium

1   1     
δn kσ (t, r) = n kσ (t, r) − f ∗kσ (t, r) = δn kσ (t, r) − f ∗kσ (t, r) − f ∗ (k)
V V
 
1 ∂ f ∗ (k) 
 δn kσ (t, r) − f kσ,kσ δn k σ (t, r) ,
V ∂∗ (k)  
kσ
(5.134)
through which (5.132) acquires a more compact expression
 
∂δn kσ (t, r) 1 ∂ f ∗ (k)
+ v k · ∇δn kσ (t, r) − v k · ∇Vσ (t, r) = Ikσ (t, r) .
∂t V ∂∗ (k)
(5.135)
If the temperature T (t, r) smoothly varies in space and time, with average T , the
quasiparticles are coupled to δT (t, r) = T (t, r) − T just like physical electrons are,
which implies that

δE δT (t, r)
= ∗ kσ (t, r) → ∗ kσ (t, r) 1− ,
δn kσ (t, r) T

so that

∇T (t, r) ∇T (t, r)
∇∗ kσ (t, r) → ∇∗ kσ (t, r) − ∗ kσ (t, r)  ∇∗ kσ (t, r) − ∗ (k) ,
T T
 (5.136)

where the last expression derives from the implicit assumption that δT (t, r)  T .
Therefore, the linearised Boltzmann equation becomes
 
∂δn kσ (t, r) 1 ∂ f ∗ (k) ∗ (k)
+ v k · ∇δn kσ (t, r) + v k · ∇T (t, r)
∂t V ∂∗ (k) T
  (5.137)
1 ∂ f ∗ (k)
− v k · ∇Vσ (t, r) = Ikσ (t, r) .
V ∂∗ (k)
5.8 Quasiparticle Hamiltonian and Landau-Boltzmann Transport Equation 253

Equation (5.137) with Ikσ (t, r) = 0 reproduces all thermal and transport properties
we earlier calculated through the response functions.
By definition, the collision integral satisfies
 
Ikσ (t, r) = 0 , ¯ ∗ kσ (t, r) Ikσ (t, r) = 0 ,
kσ kσ

where the last equation follows from the conservation of local quasiparticle energy.
Therefore, in absence of any external perturbation and at linear order,

 ∂  
0= Ikσ (t, r) = δn kσ (t, r) + ∇ · v k δn kσ (t, r) ,
∂t
kσ kσ kσ
 ∂  
0= ¯ ∗ kσ (t, r) Ikσ (t, r)  ∗ (k) δn kσ (t, r) + ∇ · v k ∗ (k) δn kσ (t, r) ,
∂t
kσ kσ kσ

which, according to the continuity equations for charge and heat, provide the defini-
tion of charge and heat currents
 
J(t, r) = v k δn kσ (t, r) = v k δn kσ (t, r) ,
kσ kσ
 (5.138)
J E (t, r) = v k ∗ (k) δn kσ (t, r) ,
kσ

where
 
2   ∂ f ∗ (k )
vk ≡ vk − f S (k, k ) v k . (5.139)
V  ∂∗ (k )
k

5.8.2 Transport Equation in Presence of an Electromagnetic Field

We earlier mentioned that, by construction, Landau’s Fermi liquid theory gives only
access to the proper response functions when the electron-electron interaction is the
Coulomb repulsion, i.e., the response to the internal field including the potential
generated via Gauss’s law by the same electrons. That simply amounts in (5.137) to
the replacement

e2 
Vσ (t, r) −
→ Vσ (t, r)+ dr   δρ(t, r ) , δρ(t, r ) = δn k σ (t, r ) ,
r − r 

k σ 
(5.140)
after which we can deal with both proper and improper linear response to the longi-
tudinal component of the electromagnetic field.
In order to extend the transport equation in the presence of a transverse electro-
magnetic field, we can exploit the semiclassical approximation, see, e.g., [7]. Let us
254 5 Landau’s Fermi Liquid Theory

consider the Boltzmann equation (5.133) before linearisation. In the presence of a

vector potential A(t, r), the conjugate momentum is not anymore k but
e
K=k− A(t, r) , (5.141)
c
Therefore, the quasiparticle distribution function in the semiclassical limit can be
still parametrised by n Kσ (t, r). Nevertheless, it is more convenient to define the
occupation density in an equivalent way

n Kσ (t, r) ≡ n kσ (t, r) = n K+ ec A(t,r) σ (t, r), (5.142)

since, in presence of A, the quasiparticle excitation energy, (5.133), changes simply

into
∗ kσ (t, r) → ∗ K+ ce A(t,r) σ (t, r) = ∗ kσ (t, r).
From the above equation, it follows that

∂ ∗ K+ ec A(t,r) σ (t, r) ∂∗ kσ (t, r)

= = v kσ (t, r) ,
∂K ∂k
is the quasiparticle group velocity in presence of the field, and

∂ ∗ K+ ec A(t,r) σ (t, r)  e ∂ Ai (t, r) ∂Vσ (t, r)  ∂n k σ (t, r)

= vkσ i (t, r) + + f kσ,k σ ,
∂r c ∂r ∂r  
∂r
i kσ

with Vσ (t, r) defined in (5.140).

Through the definition (5.142), it follows that

∂n Kσ (t, r) ∂n kσ (t, r) e ∂n kσ (t, r) ∂ A(t, r)

= + · ,
∂t Kr ∂t kr c ∂k ∂t

∂n Kσ (t, r) ∂n kσ (t, r)  e ∂n kσ (t, r) ∂ Ai (t, r)

= + ,
∂r K ∂r k c ∂ki ∂r
i

∂n Kσ (t, r) ∂n kσ (t, r)
= .
∂K r ∂k r

Putting everything together, we find the following transport equation for charged
electrons:
∂n kσ (t, r) e ∂n kσ (t, r) ∂ A(t, r)
Ikσ (t, r) = + ·
∂t c ∂k ∂t
∂n kσ (t, r)  e ∂n kσ (t, r) ∂ Ai (t, r)
+v kσ (t, r) · + vkσ j (t, r)
∂x c ∂ki ∂r j
ij
5.8 Quasiparticle Hamiltonian and Landau-Boltzmann Transport Equation 255

∂n kσ (t, r) ∂Vσ (t, r)  ∂n k σ (t, r)

− · + f kσ k σ
∂k ∂r  
∂r
kσ
 e ∂n kσ (t, r) ∂ Ai (t, r)
− vkσ i (t, r) .
c ∂k j ∂r j
ij

We note that the two terms with the ij can be written as

 ∂n kσ (t, r) ∂ Ai (t, r) ∂ A j (t, r)   ∂n (t, r)

kσ
vkσ j (t, r) − = − vkσ (t, r) ∧ B(t, r) · ,
∂ki ∂r j ∂ri ∂k
ij

where B(t, r) = ∇ ∧ A(t, r) is the magnetic field. Since the external transverse
electric field is
1 ∂ A(t, r)
E(t, r) = − ,
c ∂t
the final expression of the transport equation at uniform temperature reads
∂n kσ (t, r)
Ikσ (t, r) = + vkσ (t, r) · ∇n kσ (t, r)
∂t
∂n kσ (t, r) 
− · ∇ Vσ (t, r) + f kσ,k σ n k σ (t, r)
∂k
k σ 
∂n kσ (t, r) e   ∂n (t, r)
kσ
−e · E(t, r) − vkσ (t, r) × B(t, r) · .
∂k c ∂k
(5.143)
Let us now expand (5.143) at linear order in the deviation from equilibrium

n kσ (t, r) = n 0kσ + δn kσ (t, r) .

First, we assume an ac electromagnetic field acting as a perturbation, in which case

  ∂n (t, r) 1  ∂ f ∗ (k)
kσ
v kσ (t, r) × B(t, r) ·  v k ∧ B(t, r) · v k = 0.
∂k V ∂∗ (k)
This shows that an ac magnetic field does not contribute to the linearised transport
equation, that becomes
 
∂δn kσ (t, r) 1 ∂ f ∗ (k)
Ikσ (t, r) = + v k (t, r) · ∇δn kσ (t, r) − e v k · E(t, r)
∂t V ∂∗ (k)
 
1 ∂ f ∗ (k) 
− v k · ∇ Vσ (t, r) + f kσ,k σ n k σ (t, r)
V ∂∗ (k)   kσ
∂δn kσ (t, r)
= + v k · ∇δn kσ (t, r)
∂t  
1 ∂ f ∗ (k)
+ v k · − ∇Vσ (t, r) − e E(t, r) .
V ∂∗ (k)
(5.144)
256 5 Landau’s Fermi Liquid Theory

In presence of a dc magnetic field B(r), a subtle issue arises in connection with the
Lorentz’s force term
e  ∂n (t, r)
kσ
Lorentz’s force = − v kσ (t, r) ∧ B(r) · .
c ∂k
Indeed, a dc field, unlike an ac one, can be assumed as an integral part of the unper-
turbed Hamiltonian and, if large, can be taken as a zeroth order term. This requires
to expand at linear order vkσ (r) and ∂n kσ (r)/∂k,

∂ ∗ kσ (r)  ∂ f kσ,k σ
v kσ (r) =  vk + δn k σ (t, r) ,
∂k  
∂k
kσ
 
∂n kσ (t, r) 1 ∂ f ∗ (k) ∂δn kσ (t, r)
 vk + ,
∂k V ∂∗ (k) ∂k

leading to the linearised transport equation, including back the temperature gradient
term,

∂δn kσ (t, r) e  ∂δn (t, r)

kσ
Ikσ (t, r) = + v k · ∇δn kσ (t, r) − v k ∧ B(r) ·
∂t   c ∂k
1 ∂ f ∗ (k) 
+ v k · − ∇Vσ (t, r) − e E(t, r)
V ∂∗ (k)
 
1 ∂ f ∗ (k) ∗ (k)
+ v k · ∇T (t, r) .
V ∂∗ (k) T
(5.145)
The electric current is defined as, see (5.138) and (5.139),
  
J(t, r) = −e v k δn kσ (t, r) = −e v k δn kσ (t, r) ≡ J kσ (t, r) ,
kσ kσ kσ

whose component i, in absence of the collision term and for Vσ = 0, satisfies the
equation of motion

∂ Ji (t, r)  ∂δn kσ (t, r) 

= −e v ik = −∇ · v ik J kσ (t, r)
∂t ∂t
kσ kσ
 
e  ∂v ik e2  ∂ f ∗ (k)
−  jk J jkσ (t, r) Bk − v ik v k · E(t, r) ,
c ∂k V ∂∗ (k)
kσ kσ

where  jk is the antisymmetric tensor and repeated indices are summed. The above
equation defines the cyclotron effective mass tensor m̂ ∗ (k), whose inverse has com-
ponents
5.9 Application: Transport Coefficients with Rotational Symmetry 257

  ∂v ik  
m̂ ∗ (k)−1 ≡ , Tr m̂ ∗ (k)−1 = ∇ k · v k .
ij ∂k j
For a uniform ac electric field, and assuming v k = k/m ∗ , the equation simplifies into
the conventional equation of the electric current in presence of a uniform magnetic
field,
 
∂ J(t) e e2  ∂ f ∗ (k)
=− J(t) ∧ B − v k v k · E(t)
∂t m∗ c V ∂∗ (k)
kσ
e
=− J(t) ∧ B + D∗ E(t) ,
m∗ c
where D∗ is the Drude weight (5.74), which can be written in d = 2, 3 dimensions
and assuming space isotropy as
   
2e2  ∂ f ∗ (k) 2e2  ∂ f ∗ (k)
D∗ = − vk · vk = − · vk
dV ∂∗ (k) dV ∂k
k k
2e2    2e2     
= f ∗ (k) ∇ k · v k = f ∗ (k) Tr m̂ ∗ (k)−1 ,
dV dV
k k

yielding the conventional result D∗ = ne2 /m ∗ for v k = k/m ∗ , where n is the elec-
tron density. In the Mott insulator of Sect. 5.6, the Drude weight D∗ = 0, which is
equivalent to a vanishing trace of the inverse cyclotron mass tensor averaged over
the volume enclosed by the Luttinger surface.

5.9 Application: Transport Coefficients with Rotational

Symmetry

In this section, we consider the simple case where perturbation theory is valid and
there is just a single Fermi surface. The aim is to derive from the linearised Boltzmann
equation (5.137) the expression of the transport coefficients of a conventional Fermi
liquid.
Equation (5.145) in presence
 of an internal electric field E(t, r) becomes, once
summed over σ, thus for σ δn kσ (t, r) ≡ δn k (t, r),
 
∂δn k (t, r) 2 ∂ f ∗ (k) ∗ (k)
+ v k · ∇δn k (t, r) + v k · ∇T (t, r)
∂t V ∂∗ (k) T
 
2e ∂ f ∗ (k)
− v k · E(t, r) = Ik (t, r) .
V ∂∗ (k)
(5.146)
258 5 Landau’s Fermi Liquid Theory

We recall the definitions (5.138) for electric and heat current densities
 
J(t, r) = −e v k δn kσ (t, r) , J E (t, r) = ∗ (k) v k δn kσ (t, r) .
kσ kσ
(5.147)
We want to find their expressions in presence of uniform internal electric field and
temperature gradient. For that, we suppose that the model is, to a good approximation,
O(3) rotational invariant. Therefore ∗ (k) = ∗ (k) depends just on k = |k|, and
the Landau f -parameters are invariant under O(3) rotations, so that, close to the
quasiparticle Fermi surface,

∂∗ (k) k S/A

 S/A  
vk =  vF , f k,k = f P cos θk,k , (5.148)
∂k k
 
where P cos θk,k are Legendre polynomials, with θk,k the angle between k and
k .
We recall , since it will be useful in what follows, that

d k     1  
P cos θk,k Y m k =δ  Y m k , (5.149)
4π 2 +1

where k is the solid angle that represents a unit vector parallel to k, and Y m ( ) ≡
Y m (θ, φ) are the spherical harmonics. Moreover, one can easily demonstrate, since
∗ (k) = ∗ (k), that

  
1  ∂ f ∗ (k)     ∂ f () d k  
− F ∗ (k) G k = d ρqp () F() − G k
V ∂∗ (k) ∂ 4π
k
 
2 2 ∂ 2 ρ () F()
π T qp d k  
 ρqp F(0) +  G k ,
6 ∂2 =0 4π
(5.150)
where ρqp () is the quasiparticle density of states, ρqp = ρqp (0), and 4π is the total
solid angle in three dimensions.
Exploiting the O(3) symmetry and the fact that all physical quantities depend on
momenta close to the quasiparticle Fermi surface, we write
 
1 ∂ f ∗ (k) 
δn k (t, r) = − δn m (t, r; ∗ (k)) Y m ( k) ,
V ∂∗ (k)
m
 
1 ∂ f ∗ (k) 
δn k (t, r) = − δn m (t, r; ∗ (k)) Y m ( k) .
V ∂∗ (k)
m

We further assume that the collision integral is due to elastic scattering off non-
magnetic impurities, thus, according to Fermi golden rule, it can be written as
5.9 Application: Transport Coefficients with Rotational Symmetry 259

2π     
Ikσ (t, r) = Wkk n k σ (t, r) 1 − n kσ (t, r) − n kσ (t, r) 1 − n k σ (t, r)
V k
 
δ ∗kσ (t, r) − ∗k σ (t, r)

2π     
=− Wkk n kσ (t, r) − n k σ (t, r) δ ∗kσ (t, r) − ∗k σ (t, r) .
V  k

We remark that the energy conservation must involve the local quasiparticle energies.
The reason is that
 the collision
 integral must vanish at local equilibrium n kσ (t, r) =
0
n kσ (t, r) ≡ f ∗kσ (t, r) , which indeed it does. Therefore

2π     
Ikσ (t, r) = − Wkk δn kσ (t, r) − δn k σ (t, r) δ ∗kσ (t, r) − ∗k σ (t, r)
V
k
2π     
− 0
Wkk n kσ (t, r) − n k0 σ (t, r) δ ∗kσ (t, r) − ∗k σ (t, r)
V
k
2π     
=− Wkk δn kσ (t, r) − δn k σ (t, r) δ ∗kσ (t, r) − ∗k σ (t, r)
V
k
2π     
− Wkk δn kσ (t, r) − δn k σ (t, r) δ ∗ (k) − ∗ (k ) ,
V
k
(5.151)
only depends on the deviations from local equilibrium, and the last equation is the
linearised expression. We also write, consistently with the O(3) symmetry,
    
Wkk = W ∗ (k) P cos θkk , (5.152)

so that, using (5.149), the collision integral (5.151) can be written as

 
 ∂ f ∗ (k)  1
Ikσ (t, r) = δn m (t, r; ∗ (k)) Y m( k) ,
σ
∂∗ (k) τ (∗ (k))
m
(5.153)
where
 
1   W ∗ (k)
  = 2π ρqp (∗ (k)) W0 ∗ (k) − , (5.154)
τ ∗ (k) 2 +1
260 5 Landau’s Fermi Liquid Theory

and vanishes at = 0. We introduce the euclidean combinations of spherical har-

monics

3
Y1z = Y10 = cos θ,
4π
1 3
Y1x = √ (Y1−1 − Y11 ) = sin θ cos φ,
2 4π
i 3
Y1y = √ (Y1−1 + Y11 ) = sin θ sin φ,
2 4π

that transform like the components of a unit vector Y 1 , so that (5.148) can be written
as
k 4π
vk = v F ≡ v F Y 1 ( k ).
k 3
We are interested in the transport coefficients, which are calculated through the
charge and heat currents in presence of uniform electric field E(t) and temperature
gradient ∇T (t). In that case, δn k (t, r) = δn k (t) becomes independent of r and the
Boltzmann equation (5.146) simplifies into
 
∂δn k (t) 2 ∗ (k) ∂ f ∗ (k) 4π
Ik (t) = + v F Y 1 ( k ) · ∇T (t)
∂t V T ∂∗ (k) 3
 
2e ∂ f ∗ (k) 4π
− v F Y 1 ( k ) · E(t).
V ∂∗ (k) 3

or, equivalently, after Fourier transform in frequency,

 1 
−   δn m (ω; ∗ (k)) Y m( k) = −iω δn m (ω; ∗ (k)) Y m( k)
m
τ ∗ (k) m
2e 4π 2 ∗ (k) 4π
− vF Y 1( k) · E(ω) + vF Y 1( k ) · ∇T (ω) .
V 3 V T 3
We note that only the =1 components are  different from
 zero. Therefore, if we
define two vectors δn1 = δn 1x , δn 1y , δn 1z and δn1 = δn 1x , δn 1y , δn 1z in real
spherical harmonics, then the Boltzmann equation has the simple expression

δn1 (ω; ∗ (k)) 1 ∗ (k) 4π 1 4π

i   = −ω δn1 (ω∗ (k)) + 2i v F ∇T (ω) − 2i e v F E(ω).
τ1 ∗ (k) V T 3 V 3

We now set ω = 0, so that

    1 ∗ (k) 4π   1 4π
δn1 ∗ (k) =2 τ1 ∗ (k) v F ∇T − 2 e τ1 ∗ (k) vF E .
V T 3 V 3
(5.155)
5.9 Application: Transport Coefficients with Rotational Symmetry 261

The current densities are, therefore, at linear order

 4π   
J = −e v k δn k = −e v F Y 1 k δn k
3
k k
  
e vF 4π  ∂ f ∗ (k)
=− − δn1 (∗ (k))
4π 3 ∂∗ (k)
k
2e v 2F ∂ f () 
= d ρqp () τ1 () − eE− ∇T
3 ∂ T (5.156)
2e v 2F ρqp τ1 π2 T C
= eE− ∇T ,
3 3
 2e v 2F ∂ f () 
JE = ∗ (k) v k δn k = − d ρqp () τ1 ()  − eE− ∇T
3 ∂ T
k
2e v 2F ρqp τ1 π 2 T
= ∇T − e C T E ,
3 3

where τ1 = τ1 (0) and

 
∂ ln ρqp () τ1 ()
C= 
∂ =0 .
The expression of the charge conductivity σ is defined as

2 e2 v 2F ρqp τ1
σ= , (5.157)
3
through which we can write

π2 T C
J =σ E−σ ∇T ,
3e
(5.158)
π2 T 2 C π2 T
JE = −σ E+σ ∇T ,
3e 3e2
that provide the expressions of the transport coefficients, namely the proportionality
constants between the currents and the electric field or temperature gradient.
Through (5.158), we can now evaluate other measurable quantities besides elec-
trical conductivity. However, as discussed at the end of Sect. 5.5.1, in common
experiments the temperature drop exists at equilibrium, and that simply amounts to
invert the sign of ∇T in (5.158), thus

π2 T C
J =σ E+σ ∇T ,
3e
(5.159)
π2 T 2 C π2 T
JE = −σ E−σ ∇T ,
3e 3e2
262 5 Landau’s Fermi Liquid Theory

Therefore, if a temperature drop exists across a finite sample, electrons from the
hot side will start flowing towards the cold one. This charge flow will stop once
the opposite charge accumulated on the opposite sides generates an electric field
that prevents further charge flowing. We can determine the value of the electric field
generated as the one that cancels J in (5.159), namely

π2 T C
E=− ∇T . (5.160)
3e
This value substituted in (5.159) provides the expression of the heat current that
keeps flowing at zero charge current

π2 T π2 T 2 C 2
J E = −σ 1− ∇T ≡ −K ∇T ,
3e2 3

where K is the thermal conductivity. We note that

K π2
lim = 2,
T →0 Tσ 3e
is a universal constant, which is known as the Wiedemann-Franz law.
Another property that can be measured is the thermoelectric power. As we men-
tioned, in an open circuit of length L subject to a temperature drop L ∇T , a
potential drop −E L is generated to stop current flowing, see (5.160), which reads

−E L −E π2 T C
Q= = = , (5.161)
L ∇T ∇T 3e
and is the thermopower or Seebeck coefficient.
Finally, suppose that the temperature gradient is zero. A constant charge current
will induce a heat current which, solving (5.158) with ∇T = 0 defines the so-called
Peltier coefficient
JE
= = T Q. (5.162)
J

Problems

5.1 Bosonization of quasiparticles’ Fermi surface vibrations—Recall Landau’s

energy functional and transport equation without external field and with vanishing
collision integral:
 1 
E= dr ∗ (k) δn kσ (r) + dr f kσ,k σ  δn kσ (r) δn k σ  (r) ,
2
kσ kk σσ 
 
∂δn kσ (t, r) 1 ∂ f ∗ (k) 
= −v k · ∇δn kσ (t, r) + f kσ,kσ  v k · ∇δn k σ  (t, r) ,
∂t V ∂∗ (k)   kσ
Problems 263

and consider excitations of the form of a space-dependent transformation in momen-

tum space

k → kσ (r) = k − δukσ (r) ,

 
∗ (k) → ∗ kσ (r) ≡ ∗ (k) + δ∗σ (k, r)  ∗ (k) − v k · δukσ (r) .

It follows that
    
V δn kσ (r) = f ∗ kσ (r) − f ∗ (k)
   
∂ f ∗ (k) 1 ∂ 2 f ∗ (k)
 δ∗σ (k, r) + δ∗σ (k, r)2 .
∂∗ (k) 2 ∂∗ (k)2

After defining in generic dimensions d

 
v k · δukσ (r) ≡ (2π)d v k  u kσ (r) ≡ (2π)d v(k) u kσ (r) , (5.163)

and recalling that

1     dS 1   dS 1  
δ ∗ (k) . . . =   ... = ... ,
V SF (2π) d v k  SF (2π) d v(k)
k

where S d S . . . denotes a surface integral, and S F is the quasiparticles’ Fermi
surface

• rewrite Landau’s energy functional and transport equation in terms of the new
variables u kσ (r).

Assume to promote u kσ (r) to a field operator defined on the quasiparticles’

  Fermi
surface, the Landau’s energy functional to its Hamiltonian H {u kσ (r)} and the
transport equation to its Heisenberg equation of motion

∂u kσ (r)   
≡ −i u kσ (r) , H {u k σ (r )} .
∂t
• Show that such identification is consistent if
  i   vk
u kσ (r) , u k σ (r ) = − δσσ δ S F L k − k   · ∇ r δ(r − r ) ,
(2π)d v k 

where δ S F L (k − k ) is a δ-function on the Fermi or Luttinger surface.

• Define a new field φkσ (r) through

1 v
u kσ (r) ≡  k  · ∇ φkσ (r) ⇒ ∇ φkσ (r) = δukσ (r) ,
(2π) d v k 

and find the expression of its commutator with u kσ (r).

264 5 Landau’s Fermi Liquid Theory

The operator that yields the variation in the number of quasiparticles with respect to
its equilibrium value N∗ is defined as
 
 1  ∂ f ∗ (k)
δ N̂∗ ≡ dr δn kσ  − dr v k · δukσ (r)
V ∂∗ (k)
kσ kσ

= dr d S u kσ (r) .
σ SF

• Assume a state | δ N∗  eigenstate of δ N̂∗ with eigenvalue δ N∗ , i.e., such that

δ N̂∗ | δ N∗  = δ N∗ | δ N∗  .

Show that e−iφkσ (r) | δ N∗  is still eigenstate of δ N̂∗ and calculate the eigenvalue.

The above quantum field operators correspond to the bosonization of the

quasiparticle-quasihole excitations in the time-dependent Hartree-Fock approxima-
tion to the quasiparticle Hamiltonian, see Sect. 3.3.1. This representation, reminis-
cent of the abelian bosonization in one dimension that we discuss in Chap. 6, was
introduced in [20], and further elaborated in [21,22].

5.2 Transport coefficients in presence of a magnetic field—Repeat the same cal-

culations as in Sect. 5.9 but now in presence of a static and uniform magnetic field
B and in a Hall bar geometry where E and ∇T are in plane and B out of plane. In
this case, the electric and heat currents satisfy a matrix equation

J = σ̂ E + α̂ ∇T ,
J E = α̂ E + K̂ ∇T ,

where σ̂, α̂ and K̂ are 2 × 2 matrices in the space of the two planar coordinates.
Calculate those matrices.

References
1. L. Landau, Zh. Eskp. Teor. Fiz. 30, 1058 (1956). [Sov. Phys. JETP 3, 920 (1957)]
2. L. Landau, Zh. Eskp. Teor. Fiz. 32, 59 (1957). [Sov. Phys. JETP 5, 101 (1957)]
3. P. Nozières, J.M. Luttinger, Phys. Rev. 127, 1423 (1962). https://doi.org/10.1103/PhysRev.127.
1423
4. J.M. Luttinger, P. Nozières, Phys. Rev. 127, 1431 (1962). https://doi.org/10.1103/PhysRev.127.
1431
5. A. Abrikosov, L. Gorkov, I. Dzyaloshinskii, Methods of Quantum Field Theory in Statistical
Physics (Dover, New York, 1975). See Sect. 19.4
6. P. Noziéres, Theory of Interacting Fermi Systems (CRC Press, Boca Raton, 1998)
7. D. Pines, P. Nozières, The Theory of Quantum Liquids (CRC Press, Boca Raton, 1989). https://
doi.org/10.4324/9780429492662
References 265

8. P. Anderson, Mater. Res. Bull. 8(2), 153 (1973). https://doi.org/10.1016/0025-5408(73)90167-

0
9. C.M. Varma, P.B. Littlewood, S. Schmitt-Rink, E. Abrahams, A.E. Ruckenstein, Phys. Rev.
Lett. 63, 1996 (1989). https://doi.org/10.1103/PhysRevLett.63.1996
10. X.G. Wen, Phys. Rev. B 65, 165113 (2002). https://doi.org/10.1103/PhysRevB.65.165113
11. T. Senthil, S. Sachdev, M. Vojta, Phys. Rev. Lett. 90, 216403 (2003). https://doi.org/10.1103/
PhysRevLett.90.216403
12. P.A. Lee, N. Nagaosa, X.G. Wen, Rev. Mod. Phys. 78, 17 (2006). https://doi.org/10.1103/
RevModPhys.78.17
13. S. Hartnoll, A. Lucas, S. Sachdev, Holographic Quantum Matter (MIT Press, Cambridge, 2018)
14. T. Andrade, A. Krikun, K. Schalm, J. Zaanen, Nat. Phys. 14(10), 1049 (2018). https://doi.org/
10.1038/s41567-018-0217-6
15. M. Fabrizio, Phys. Rev. B 102, 155122 (2020). https://doi.org/10.1103/PhysRevB.102.155122
16. M. Fabrizio, Nat. Commun. 13(1), 1561 (2022). https://doi.org/10.1038/s41467-022-29190-y
17. I. Dzyaloshinskii, Phys. Rev. B 68, 085113 (2003). https://doi.org/10.1103/PhysRevB.68.
085113
18. A.V. Chubukov, D.L. Maslov, S. Gangadharaiah, L.I. Glazman, Phys. Rev. Lett. 95, 026402
(2005). https://doi.org/10.1103/PhysRevLett.95.026402
19. M. Oshikawa, Phys. Rev. Lett. 84, 3370 (2000). https://doi.org/10.1103/PhysRevLett.84.3370
20. F.D.M. Haldane, Luttinger’s Theorem and Bosonization of the Fermi Surface (2005)
21. D.V. Else, R. Thorngren, T. Senthil, Phys. Rev. X 11, 021005 (2021). https://doi.org/10.1103/
PhysRevX.11.021005
22. X.G. Wen, Phys. Rev. B 103, 165126 (2021). https://doi.org/10.1103/PhysRevB.103.165126
Brief Introduction to Luttinger Liquids
6

In Sect. 4.3.4 we showed that perturbatively Im( → 0, k) → || in one dimension,

see (4.64), which implies through (4.65) that the quasiparticle residue vanishes as
 → 0, and thus that conventional Landau’s Fermi liquid theory in not applicable
whatever the interaction strength is. Nonetheless, we can still uncover in great detail
the properties of interacting electrons in one dimension, commonly refereed to as
Luttinger Liquids. That is actually the content of the present Chapter, where we briefly
discuss Luttinger Liquids and abelian bosonization. More details can be found in the
books Bosonization and Strongly Correlated Systems, authors A.O. Gogolin, A.A.
Nersesyan and A.M. Tsvelik [1], and Quantum Physics in One Dimension, author
T. Giamarchi [2].

6.1 What Is Special in One Dimension?

Let us imagine to apply the conserving Hartree-Fock approximation to calculate

linear response functions, considering, for instance, the Hamiltonian on an L site
chain
 † U  † †
H= (k) ckσ ckσ + ckσ c p+qσ c pσ ck+qσ , (6.1)
2L 
kσ kqσσ
with an on-site interaction U , either positive or negative, and a generic band structure
as in Fig. 6.1. At U = 0, the ground state is obtained by filling all energy states up
to the Fermi energy, or, equivalently, occupying all momenta |k| ≤ k F , see Fig. 6.2,
where

 
kF  kF dk kF N π
= 2L = 2L = N ⇒ kF = π ≡ n,
σ −k F 2π π 2L 2
k=−k F

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 267
M. Fabrizio, A Course in Quantum Many-Body Theory, Graduate Texts in Physics,
https://doi.org/10.1007/978-3-031-16305-0_6
268 6 Brief Introduction to Luttinger Liquids

Fig. 6.1 Generic

one-dimensional energy
dispersion (k) in
momentum space. The
Brillouin zone is defined for
k ∈ [−π/a, π/a], where a is
(k)
the lattice spacing

π π
− k
a a

with n the electron density. If we force spin SU (2) symmetry and translation invari-
ance, the Hartree-Fock Green’s function is the same as for non-interacting electrons,
apart from a renormalisation of the chemical potential, which is irrelevant if we work
at fixed density n. In other words,

1
G 0 (i, k) = .
i − (k)

For weak interaction compared to the Fermi energy, the low energy physics is con-
trolled by the excitations
  around the two Fermi points,
 ±k
 F , see Fig. 6.2, where
(k  k F )  v F k − k F and (k  −k F )  −v F k + k F , so that we can write
⎧
⎪
⎪ 1  ≡ G 0R (i, k − k F ) , k − kF kF ,
⎪
⎪ i − v F k − k F
⎨
G 0 (i, k) 
⎪
⎪
⎪
⎪
⎩ 1  ≡ G 0L (i, k + k F ) , k + kF kF .
i + v F k + k F
(6.2)
The labels R and L indicate electrons close to k F and −k F , respectively, convention-
ally named right (R) and left (L) moving electrons. With that notation, the particle-
hole bubble at zero transferred frequency and momentum transferred q = 2k F ,

1  
χ0 (0, 2k F ) = T G 0 (i, k + k F ) G 0 (i, k − k F )
L k
   
1   1  f − vF k − f vF k
 T G 0R (i, k) G 0L (i, k) = −
L 
L 2v F k
|k| k F |k| k F

 kF dk vF k 1 1 vF k F
− tanh − ln ,
−k F 2π 2T 2v F k 2πv F T
6.1 What Is Special in One Dimension? 269

Fig. 6.2 One-dimensional

Fermi sea

(k)
F

Fermi sea

π π
− −kF kF
a k a

diverges as T → 0. The reason of such singular behaviour is the nesting property of

the one-dimensional Fermi surface. In general, a d-dimensional Fermi surface has
nesting if there are parallel portions of it, thus connected by a single nesting vector Q
defined apart from a reciprocal lattice vector. Whenever that occurs, the non interact-
ing particle-hole bubble at momentum transferred Q + q, with |q| |Q|, diverges
at small temperature and frequency transferred ω at least as ln max(T , v F |q|, ω).
Similarly, we can consider the particle-particle bubble at zero total frequency and
momentum,

1  
S0 (0, 0) = T G 0 (i, k + k F ) G 0 (−i, −k − k F )
L 
k
1  
 T G 0R (i, k) G 0L (−i, −k)
L 
|k| k F
1   1 vF k F
=− T G 0R (i, k) G 0L (i, k)  ln ,
L 
2πv F T
|k| k F

which also diverges logarithmically. That actually occurs in any dimensions and for
any Fermi surface provided the non-interacting local density of states is finite at zero
energy, and it is known as the Cooper singularity. In one dimension, the Cooper sin-
gularity and the nesting at q = 2k F coexist, and that has very peculiar consequences,
that are already clear within the conserving Hartree-Fock approximation.
In that approximation, the irreducible vertices in the particle-hole singlet, S, and
triplet, T , channels can be readily found through (4.99) and (4.100), and read

p-h U p-h U
0S = , 0T = − ,
2 2
whereas those in the singlet and triplet particle-particle channels are, see (4.109) and
(4.110),
p-p p-p
0S=0 = U , 0S=1 = 0 .
270 6 Brief Introduction to Luttinger Liquids

Correspondingly, the interaction vertex that transfers frequency iω = 0 and momen-

tum q = 2k F between the particle and the hole can be obtained through the Bethe-
Salpeter equation (4.102)

U 1 U 1
 S (0, 2k F ) = , T (0, 2k F ) = − .
2 1 − U χ0 (0, 2k F ) 2 1 + U χ0 (0, 2k F )
(6.3)
Since χ0 (0, 2k F ) ∼ ln T , the spin-triplet vertex at 2k F becomes singular at any
U > 0, which indicates instability towards a magnetic ordering modulated with wave
vector 2k F . Conversely, for any U < 0,  S (0, 2k F ) is singular, now signal of a
charge-density wave instability at 2k F .
Similarly, the interaction vertex where the two incoming lines in a spin-singlet
configuration with zero total frequency and momentum is, still through the Bethe-
Salpeter equation,
U
 S=0 (i, P) = ,
1 + U S0 (0, 0)
and is singular for any U < 0, reflecting the conventional superconducting instability
for attractive interactions.
In conclusion, however small U is, and whatever its sign is, the system is unstable
and would like to spontaneously break a continuous symmetry, either the translational
symmetry if 2k F is incommensurate, or spin SU (2), or global charge U (1). However,
a continuous symmetry cannot be broken in one-dimension because of the Mermin-
Wagner theorem, see Sect. 1.8.3, which implies that quantum fluctuations are strong
and prevent symmetry breaking. As a consequence, we anticipate that long-range
order is replaced by a quasi long-range one where the correlation functions decay
power-law at long distances/times with anomalous interaction-dependent exponents,
which is the first distinctive feature of one-dimensional interacting electrons.
Besides that, there is another peculiar property in one dimension. At generic
transferred real frequency ω, with an infinitesimal positive imaginary part that we
do not indicate unless necessary, and momentum q, the vertices in (6.3) are

U 1 U 1
 S (ω, q) = , T (ω, q) = − ,
2 1 − U χ0 (ω, q) 2 1 + U χ0 (ω, q)

where
    
dk f (k) − f (k + q)
χ0 (ω, q) = .
2π ω − (k + q) + (k)
For small |q| k F and T = 0,

dk   vk q 1 q q 1 vF q 2
χ0 (ω, q)  δ (k) = − = ,
2π ω − vk q 2π ω − vF q ω + vF q π ω − v 2F q 2
2

(6.4)
6.2 Interacting Spinless Fermions 271

so that

U ω 2 − v 2F q 2 U ω 2 − v 2F q 2
 S (ω, q)  , T (ω, q) = − ,
2 ω − v 2F q 2 − U v F q 2 /π
2 2 ω − v 2F q 2 + U v F q 2 /π
2

This result implies that, at small |q|, the particle-hole excitations are exhausted by
two acoustic modes, a charge mode and a spin one with velocities, respectively,

U U
vρ = v F 1+ , vσ = v F 1− . (6.5)
2πv F 2πv F

These two modes correspond to two linearly dispersive bosons, completely control
the long wavelength behaviour, even beyond the weak coupling conserving Hartree-
Fock approximation, as we shall see, and are actually the building blocks of Bosoniza-
tion. We further note that these bosons carry separately charge and spin degrees of
freedom, reflecting the so-called spin-charge separation in one dimension, which we
will also discuss in what follows.

6.2 Interacting Spinless Fermions

Let us begin considering a one dimensional model of spinless fermions described by

a non interacting Hamiltonian of the form

H0 = (k) ck† ck , (6.6)
k

with an energy dispersion (k) in momentum space of the form in Fig. 6.1, measured
with respect to the chemical potential. The electrons mutually interact through

1 
Hint = V (q) ρ(q) ρ(−q) , (6.7)
2L q

where L is the number of sites, V (q) = V (−q) the interaction, and


ρ(q) = ck† ck+q , (6.8)
k

the Fourier transform of the density. In perturbation theory, as discussed earlier, one
finds singularities that, in this case, indicate instabilities towards spontaneous break-
down of translational symmetry with wave vector 2k F or towards superconductivity,
which, as we mentioned, are prevented by quantum fluctuations. We will see in detail
what that means.
272 6 Brief Introduction to Luttinger Liquids

For that, we adopt an approach substantially similar to the one we applied in

Sects. 1.7 and 3.3.1. We start from the non interacting Hamiltonian (6.6). The ground
state with N particles is just the Fermi sea (FS) obtained by occupying momentum
eigenstates up to the Fermi momentum k F , which for spinless fermions is simply
k F = π n, with n the average density, i.e., the number of electrons per site, thus the
occupation number in momentum space
 
ck† ck FS = θ k F − |k| ≡ n 0 (k) .

With Sect. 1.7 in mind, we shall regard the Fermi sea, see Fig. 6.2, like a classical
ground state characterised by the macroscopic ‘classical order parameter’
 
≡ n 0 (k) − n 0 (k) = N , (6.9)
|k|≤k F |k|>k F

as well as the ‘vacuum‘ of the quantum fluctuations brought about by the interaction
(6.7). Such ‘semiclassical‘ approximation is valid provided the order parameter (6.9)
is only weakly affected by interaction. That implies, in the first place, that the effective
strength of interaction, to be specified later, is much smaller that  F of Fig. 6.2, or,
equivalently, that a macroscopic number of states deep inside, |k| k F , and outside,
|k| k F , the Fermi sea are totally unaffected by interaction. We can therefore define
new operators corresponding to the right-moving (R) and left-moving (L) fermions
 
ck ∼+k F −−→ c Rk , c†Rk c Rk FS = n 0R (k) = θ k F − k ,
  (6.10)
ck ∼−k F −−→ c Lk , c†Lk c Lk FS = n 0L (k) = θ k + k F ,

without running the risk of overcounting states. Correspondingly, we introduce the

densities of right and left moving fermions,
 
ρ R (q) ≡ c†Rk c Rk+q , ρ L (q) ≡ c†Lk c Lk+q , (6.11)
k k

which are unambiguously defined only for |q| k F . For that, whenever we shall
need to sum over q, we will add a short-distance cutoff through the function e−α|q|
where 0 < α  1/k F  a, with a the lattice spacing.
The density operators (6.11) creates particle-hole excitation over the Fermi sea,
or destroy them if they are already present. If we regard ρ R (q) and ρ L (q) as the spin-
density operators in Sect. 1.7, then the counterpart of the spin-wave approximation
is replacing commutators with their expectation values over the ‘vacuum‘, i.e., the
Fermi sea. Therefore, through (6.10) and since q kF ,
6.2 Interacting Spinless Fermions 273

   
ρ R (q) , ρ R (−q  ) = c†Rk c Rk+q−q  − c†Rk+q  c Rk+q
k

 c†Rk c Rk+q−q  − c†Rk+q  c Rk+q FS
k

   ∂n 0R (k)
= δq,q  n 0R (k) − n 0R (k + q)  q δq,q  −
∂k
k k

dk   qL
= q δq,q  L δ k − k F = δq,q  ,
2π 2π
(6.12)
for the right-moving density,
   
ρ L (q) , ρ L (−q  ) = c†Lk c Lk+q−q  − c†Lk+q  c Lk+q
k

 c†Lk c Lk+q−q  − c†Lk+q  c Lk+q FS
k

   ∂n 0R (k)
= δq,q  n 0L (k) − n 0L (k + q)  q δq,q  −
∂k
k k

dk   qL
= −q δq,q  L δ k + k F = −δq,q  ,
2π 2π
(6.13)
for the left-moving one, while
 
ρ R (q) , ρ L (−q  ) = 0 .

Therefore, in this approximation ρ R (q) and ρ L (q) behave as independent bosonic

operator. We√also note that the right hand sides of (6.12) and (6.13) imply that
ρ R(L) (q) ∼ L while the ‘classical order parameter‘ (6.9) is of order L at fixed
density n = N /L, consistently with the semiclassical approximation.

6.2.1 Bosonized Expression of the Non-interacting Hamiltonian

For q k F , the density (6.8) is just

ρ(q) = ρ R (q) + ρ L (q) . (6.14)

The continuity equation provides the expression of the current, specifically, through
(6.6) and (6.8) and still assuming q kF
274 6 Brief Introduction to Luttinger Liquids

     ∂(k) †
i ρ̇(q) = q J (q) = ρ(q) , H0 = (k + q) − (k) ck† ck+q  q c c
∂k k k+q
k k
 
 v F q ρ R (q) − ρ L (q) ,

where v F > 0 is the Fermi velocity at +k F , while −v F that at −k F . Since ρ R (q)

and ρ L (q) commute with each other, it follows that
   
i ρ̇ R (q) = ρ R (q) , H0 = v F q ρ R (q) , i ρ̇ L (q) = ρ L (q) , H0 = −v F q ρ L (q) ,
(6.15)
which implies that ρ R (0) ≡ N R and ρ L (0) ≡ N L are separately conserved and their
corresponding currents read

J R (q) = v F ρ R (q) , JL (q) = −v F ρ R (q) . (6.16)

If we consider an equilibrium state with a finite current, as shown in Fig. 6.3, then
 kRF  kL F
dk L kRF dk L kL F
ρ R (0) = N R = = , ρ L (0) = N L = = ,
0 2π 2π 0 2π 2π
(6.17)
with N = N R + N L number of fermions, and current
 
J = vF N R − NL . (6.18)

Given the commutation relations (6.12) and (6.13), (6.15) can be reproduced by
πv F   
H0 = ρ R (q) ρ R (−q) + ρ L (q) ρ L (−q) − μ R N R − μ L N L
L q
πv F 2πv F   
= N R2 + N L2 + ρ R (q) ρ R (−q) + ρ L (q) ρ L (−q) − μ R N R − μ L N L
L L
q>0
πv F 2πv F   
= N2 + J2 + ρ R (q) ρ R (−q) + ρ L (q) ρ L (−q) − μ N − μ J J ,
2L L
q>0
(6.19)
where
μ R = vF k R F , μL = vF k L F , (6.20)
guarantee through (6.17) that the ground state has the desired number of right and
left moving fermions, while

μ R + μL μ R − μL
μ= , μJ = .
2 2v F
6.2 Interacting Spinless Fermions 275

Fig. 6.3 Excited state with

finite current due to
k R F = k L F

(k)

π π
− −kLF kRF
a k a

6.2.2 Bosonic Representation of the Fermi Fields

In real space the right- and left-moving densities are defined trough

1  iq x †
ρ R(L) (x) = e ρ R(L) (q) ≡  R(L) (x)  R(L) (x) , (6.21)
L q

which also defines the corresponding Fermi fields. Equation (6.12) thus implies that
  
1  i ∂ dq iq(x−y) −α|q|
ρ R (x) , ρ R (y) = q eiq(x−y) e−α|q| = − e e
2πL q 2π ∂x 2π
i ∂ 1 α i ∂
=− =− δα (x − y)
2π ∂x π (x − y)2 + α2 2π ∂x
i ∂2 x−y
=− tan−1 ,
2π ∂x2 2 α
(6.22)
where δα (x − y) is, for α → 0, the representation of a δ-function through a
Lorentzian. Similarly,
  i ∂
ρ L (x) , ρ L (y) = δα (x − y) . (6.23)
2π ∂x
We further note, through (6.21), that
   
 R (x) , ρ R (y) = δ(x − y)  R (x) ,  L (x) , ρ L (y) = δ(x − y)  L (x) ,
   
 R (x) , N R =  R (x) ,  L (x) , N L =  L (x) ,
(6.24)
276 6 Brief Introduction to Luttinger Liquids

while, by definition,
   
 R (x) ,  R† (y) =  L (x) ,  L† (y) = δ(x − y) ,
   
 R (x) ,  R (y) =  L (x) ,  L (y) = 0 , (6.25)
   
 R (x) ,  L† (y) =  R (x) ,  L (y) = 0 .

Equation (6.24) suggest that the Fermi fields act on the densities like the unit trans-
lation operator T = ei p does on the coordinate x, i.e., through [x, p] = i,
 
e−i p x ei p = x − 1 ⇒ ei p , x = ei p .

Therefore, let us assume

 R (x) = N R ei φ R (x) ,  L (x) = N L e−i φ L (x) , (6.26)

with N R(L) normalisation constants, and try to find the real φ R (x) and φ L (x) that
allow fulfilling (6.24) and (6.25). The simplest condition to fulfil is that right and
left moving fields anticommute with each other, namely,

eiφ R (x) e∓iφ L (y) = − e∓iφ L (y) eiφ R (x) , ∀ x, y . (6.27)

We recall that, if two operators A and B commute with [A, B], then the following
equations hold

e A e B = e B e A e−[B,A] , e A+B = e A e B e−[A,B]/2 = e B e A e−[B,A]/2 ,

(6.28)
through which (6.27) can be satisfied if
 
φ R (x) , φ L (y) = −i π , ∀ x, y . (6.29)

It is easy to realise that (6.24) are instead fulfilled if

   
φ R (x) , ρ R (y) = −i δ(x − y) , φ R (x) , N R = −i ,
    (6.30)
φ L (x) , ρ L (y) = i δ(x − y) , φ L (x) , N L = i .

It follows, through (6.22) and (6.23), that

1 ∂φ R (x) 1 ∂φ L (x)
ρ R (x) = , ρ L (x) = , (6.31)
2π ∂x 2π ∂x
6.2 Interacting Spinless Fermions 277

where, including also (6.29),

  x−y
φ R (x) , φ R (y) = 2i tan−1  iπ sign(x − y) ,
α
  x−y
φ L (x) , φ L (y) = −2i tan−1  −iπ sign(x − y) , (6.32)
  α
φ R (x) , φ L (y) = −i π , ∀ x, y ,

are the commutation relations between the φ-fields. We note that

 
d x ∂φ R(L) (x)
N R(L) = d x ρ R(L) (x) = ,
2π ∂x
is a conserved quantity. Therefore, the vacuum expectation value
N R(L)
0 | φ R(L) (x) | 0 = x = k R(L)F x . (6.33)
2πL
We can now easily demonstrate that, for x = y,
 
 R (x),  R (y) = N R2 eiφ R (x) eiφ R (y) + eiφ R (y) eiφ R (x)
 
= N R2 eiφ R (x) eiφ R (y) 1 + e− iφ R (x),iφ R (y)

= N R2 eiφ R (x) eiφ R (y) 1 + eiπ sign(x−y) = 0,

and similarly for left movers.

However, we still need to prove that the representation (6.26) with fields satisfying
(6.32) fulfils the first equation in (6.25). Let us write, recalling (6.33),

φ R (x) = k R F x + φ(+) (−) (+) (−)

R (x) + φ R (x) , φ L (x) = k L F x + φ L (x) + φ L (x) ,
(6.34)
where φ(+) include creation operators and φ(−) annihilation ones, and, since φ is
hermitian, then φ(+) = φ(−) † . Therefore, upon defining
 
φ(+)
R(L) (x) , φ (−)
R(L) (y) ≡ FR(L) (x − y) , (6.35)

where F only depends on x − y because of translation symmetry, then,

     
φ R (x) , φ R (y) = φ(+)
R (x) , φ(−)
R (y) + φ(−)
R (x) , φ(+)
R (y)
x−y
= FR (x − y) − FR (y − x) = 2i tan−1 ,
     α
(+) (−) (−) (+)
φ L (x) , φ L (y) = φ L (x) , φ L (y) + φ L (x) , φ L (y)
x−y
= FL (x − y) − FL (y − x) = −2i tan−1 ,
α
278 6 Brief Introduction to Luttinger Liquids

which are satisfied if

   
FR (x − y) = ln α + i (x − y) , FL (x − y) = ln α − i (x − y) . (6.36)
 
Let us continue and prove that  R (x),  R† (y) = δ(x − y). We observe, exploiting
(6.28), that

(+) (−) (+) (−)

 R (x) = N R eiφ R (x) = N R eik R F x eiφ R (x) eiφ R (x) e−[iφ R (x),iφ R (x)]/2
(+) (−) √ (+) (−)
= N R eik R F x eiφ R (x) eiφ R (x) e FR (0)/2 = N R eik R F x α eiφ R (x) eiφ R (x) ,

and thus
√ (+) (−)
 R† (y) = N R e−ik R F x α e−iφ R (y) e−iφ R (y) .
It follows that, repeatedly using (6.28),
(+) (−) (+) (−)
 R (x)  R† (y) = N R2 α eik R F (x−y) eiφ R (x)
eiφ R (x)
e−iφ R (y)
e−iφ R (y)

(+) (+) (−) (−)

 (+) (−)

= N R2 α eik R F (x−y) e iφ R (x)
e −iφ R (y) iφ R (x)
e e−iφ R (y)
e− −iφ R (y) , iφ R (x)
 (+) (+)
 (−) (−)
 
= N R2 α eik R F (x−y) e−FR (y−x) ei φ R (x)−φ R (y) ei φ R (x)−φ R (y)
 
α
= N R2 : ei φ R (x)−φ R (y) : , (6.37)
α − i (x − y)
(+) (−) (+) (−)
 R† (y)  R (x) = N R2 α eik R F (x−y) e −iφ R (y)
e−iφ R (y) iφ R (x)
e e iφ R (x)
 (+) (+)
  (−) (−)

i φ R (x)−φ R (y)
= N R2 ik R F (x−y)
αe e −FR (x−y)
e ei φ R (x)−φ R (y)
 
α φ R (x)−φ R (y)
= N R2 : ei :,
α + i (x − y)

where : (. . . ) : means the operators inside are normal ordered, i.e., creations opera-
tors are on the left of annihilation ones. For left moving fermions, the above equations
read
 
α
 L (x)  L† (y) = N L2 : e−i φ L (x)−φ L (y) : ,
α + i (x − y)
  (6.38)
α
 L† (y)  L (x) = N L2 : e−i φ L (x)−φ L (y) : .
α − i (x − y)

Therefore,
   
α α
†
 R (x) ,  R (y) = N R2 : ei φ R (x)−φ R (y) : +
α + i (x − y) α − i (x − y)
    1 1
= N R2 : ei φ R (x)−φ R (y) : − i α −
(x − y) − iα (x − y) + iα
6.2 Interacting Spinless Fermions 279
 
 N R2 : ei φ R (x)−φ R (y) : 2π α δ(x − y) = N R2 2π α δ(x − y) ,
 
 L (x) ,  L† (y) = N L2 2π α δ(x − y) ,

   
since : ei φ R (x)−φ R (y) := 1 and : e−i φ L (x)−φ L (y) := 1 for x = y, which implies
that N R2 = N L2 = 1/(2πα), and thus

1 1
 R (x) = √ eiφ R (x) ,  L (x) = √ e−iφ L (x) . (6.39)
2π α 2π α

6.2.3 Operator Product Expansion

Equations (6.21) and (6.31) imply, e.g., for right moving electrons, that

1 ∂φ R (x) 1
e−iφ R (x) eiφ R (x) .
?
ρ R (x) = =  R† (x)  R (x) =
2π ∂x 2π α
At first sight, the bosonized expressions of the Fermi fields should simply yield
1/2πα, which is a wrong result. The issue is that the exponential operators are
delicate to dealt with. The proper way to proceed is through point splitting, which
defines the so-called Operator Product Expansion (OPE). For that, we note that

1 ∂φ R (x) 1 † 
ρ R (x) = = lim  R (x)  R (y) +  R† (y)  R (x)
2π ∂x x→y 2

1
= lim e−iφ R (x) eiφ R (y) + e−iφ R (y) eiφ R (x)
x→y 4πα
   
1 : e−i φ R (x)−φ R (y) : : ei φ R (x)−φ R (y) :
= lim + .
x→y 4π α − i(x − y) α + i(x − y)

The limit x → y cannot be straightly done. The reason is that in our approach α
quantifies the indeterminacy of the coordinates. It follows that we are allowed to
send x to y only after we send the indeterminacy α → 0. With such prescription
   
1 ∂φ R (x) 1 : e−i φ R (x)−φ R (y) : : ei φ R (x)−φ R (y) :
ρ R (x) = = lim lim +
2π ∂x x→y α→0 4π α − i(x − y) α + i(x − y)
 
i 1 ∂φ R (x) 1 ∂φ R (x)
= lim 1−i (x − y) − 1+i (x − y)
4π x→y x−y ∂x x−y ∂x

1 ∂φ R (x)
= ,
2π ∂x
(6.40)
280 6 Brief Introduction to Luttinger Liquids

which is the desired equivalence. Similarly, one can readily verify with that same
prescription that

∂ R (x) i ∂ R† (x) ∂ R (x)

−i  R† (x) = lim  R (y) −  R† (y)
∂x x→y 2 ∂x ∂x
   
 −i φ R (x)−φ R (y) 
1 ∂ :e : : ei φ R (x)−φ R (y) :
= − lim +
x→y 8π ∂x x−y x−y
(6.41)
2
1 ∂φ R (x)
 ,
4π ∂x
2
∂ L (x) 1 ∂φ L (x)
i  L† (x)  ,
∂x 4π ∂x

thus  R (x) and  L (x) have Dirac-like dispersion. The almost equivalence means
we have dropped an infinite positive term that, strictly speaking, cancels the infinite
negative energy of the linearised spectrum to yield a finite energy.

6.2.4 Non-interacting Green’s Functions and Density-Density

Response Functions

The non-interacting Hamiltonian (6.19) can be written, through (6.20), as

     
H0 = π v F d x ρ R (x)2 + ρ L (x)2 − v F d x k R F ρ R (x) + k L F ρ L (x)
     
vF ∂φ R (x) 2 ∂φ L (x) 2 vF ∂φ R (x) ∂φ L (x)
= dx + − dx kRF + kL F ,
4π ∂x ∂x 2π ∂x ∂x

so that, through (6.30),

 
i φ̇ R (x, t) = φ R (x, t) , H0 = −i 2π v F ρ R (x, t) + i v F k R F
∂φ R (x, t)
= −i v F − kRF ,
∂x
 
i φ̇ L (x, t) = φ L (x, t) , H0 = i 2π v F ρ L (x, t) − i v F k L F
∂φ L (x, t)
= i vF − kL F ,
∂x

namely,

∂ ∂ ∂ ∂
+ vF φ R (x, t) = v F k R F , − vF φ L (x, t) = −v F k L F ,
∂t ∂x ∂t ∂x
(6.42)
6.2 Interacting Spinless Fermions 281

with solution
   
φ R (x, t) = v F k R F t + φ R x − v F t , φ L (x, t) = −v F k L F t + φ R x + v F t .
(6.43)
We observe that (6.33) implies that
 
0 | φ R (x, t) | 0 = v F k R F t + 0 | φ R x − v F t | 0 = k R F x ,
 
0 | φ L (x, t) | 0 = −v F k L F t + 0 | φ L x + v F t | 0 = k L F x ,

and therefore the expectation value is constant in time, as it should since it is asso-
ciated to a conserved quantity.

6.2.4.1 Green’s Functions

Using (6.37) and (6.38), we readily find the non-interacting Green’s functions

G 0R (x, t) = −i θ(t)  R (x, t)  R† (0, 0) + i θ(−t)  R† (0, 0)  R (x, t)

 
1 1 1
= − i θ(t) + i θ(−t) : ei(φ R (x−v F t)−φ R (0)) :
2π α − i(x − v F t) α + i(x − v F t)
1 eik R F x
= ,
2π x − v F t + iα sign(t)
(6.44)
and, similarly,

G 0L (x, t) = −i θ(t)  L (x, t)  L† (0, 0) + i θ(−t)  L† (0, 0)  L (x, t)

 
1 1 1
= − i θ(t) + i θ(−t) : e−i(φ L (x−v F t)−φ L (0)) :
2π α + i(x + v F t) α − i(x + v F t)
1 e−ik L F x
= .
2π −x − v F t + iα sign(t)
(6.45)
Through G 0R (x, t) we can calculate the non-interacting momentum distribution of
right moving fermions, n 0R (k),
 
d x e−i(k−k R F )x
n 0R (k) = −i d x e−ikx G 0R (x, t = 0− ) = . (6.46)
2πi x − iα

We can turn the integral over x into a contour integral in the complex plane z =
x + i y, where the contour runs along the real axis and closes in the upper half plane
for k < k R F and in the lower one otherwise. Only in the former case the contour
integral yields a finite result, since the function to be integrated has a single
 pole at
z = iα with residue 1, thus in the upper half plane. Therefore n 0R (k) = θ k R F − k ,
which is the correct result. An alternative way to get the same result, which will turn
282 6 Brief Introduction to Luttinger Liquids

useful in the interacting case, is to calculated

 
∂n 0R (k) −ikx dx − x
=− dx x e G 0R (x, t
=0 )=− e−i(k−k R F )x
∂k 2π x − iα
 
d x −i(k−k R F )x dx 1
=− e −iα e−i(k−k R F )x
2π 2π x − iα

d x −i(k−k R F )x  
−−−→ − e = −δ k − k R F .
α→0 2π
(6.47)
For later convenience, we define the vacuum expectation value

β R ,β L (x, t) ≡ eiβ R φ R (x,t)+iβ L φ L (x,t) e−iβ R φ R (0,0)−iβ L φ L (0,0) . (6.48)

We observe that, upon repeatedly using (6.28) and (6.29)

eiβ R φ R (x,t)+iβ L φ L (x,t) e−iβ R φ R (0,0)−iβ L φ L (0,0)

= e−iπ β R β L eiβ R φ R (x,t) eiβ L φ L (x,t) e−iβ R φ R (0,0) e−iβ L φ L (0,0)
= eiβ R φ R (x,t) e−iβ R φ R (0,0) eiβ L φ L (x,t) e−iβ L φ L (0,0) ,

so that, through (6.37) and (6.38),

β R ,β L (x, t) = eiβ R φ R (x,t) e−iβ R φ R (0,0) eiβ L φ L (x,t) e−iβ L φ L (0,0)

β 2R β L2
α α
= eik R F β R x+ik L F β L x .
α − i(x − v F t) α + i(x + v F t)
(6.49)

6.2.4.2 Density-Density Correlation Functions

The density-density correlation function that is simplest to calculate is the long
wavelength one, |q| k R(L)F . We first note that

0 | φ R (x, t) φ R (y, 0) | 0 = 0 | φ R (x − v F t) φ R (y) | 0

= k 2R F x y + 0 | φ(−) (x − v F t) φ(+)
R (y) | 0
R 
(+) (−)
= k 2R F x y − 0 | φ R (y) , φ R (x − v F t) | 0
    
= k 2R F x y − FR v F t − x + y = k 2R F x y − ln α + i v F t − x + y
π  
= k 2R F x y + i − ln x − y − v F t + iα ,
2  
π
0 | φ R (x, t) φ R (y, 0) | 0 = k 2L F x y − i − ln x − y + v F t − iα ,
2
6.2 Interacting Spinless Fermions 283

so that
1∂2
0 | ρ R (x, t) ρ R (y, 0) | 0 = 0 | φ R (x, t) φ R (y, 0) | 0
4π ∂x∂ y 2
2
k2 1 1
= R F2 − 2 , (6.50)
4π 4π x − y − v F t + iα
2
k 2L F 1 1
0 | ρ L (x, t) ρ L (y, 0) | 0 = − .
4π 2 4π 2 x − y + v F t − iα
It follows that the Fourier transforms at q = 0
 2
1 1
0 | ρ R (q, t) ρ R (−q, 0) | 0 = − d x e−iq x
4π 2 x − v F t + iα
q −iv F qt
= θ(q) e ,
2π
 2
1 1
0 | ρ R (−q, 0) ρ R (q, t) | 0 = − 2 d x e−iq x
4π x − v F t − iα
|q| −iv F qt
= θ(−q) e ,
2π
 2
1 1
0 | ρ L (q, t) ρ L (−q, 0) | 0 = − 2 d x e−iq x
4π x + v F t − iα
|q| iv F qt
= θ(−q) e ,
2π
 2
1 1
0 | ρ L (−q, 0) ρ L (q, t) | 0 = − 2 d x e−iq x
4π x + v F t + iα
q iv F qt
= θ(q) e ,
2π
so that
  q
χ R (q, t) = −i θ(t) 0 | ρ R (q, t) , ρ R (−q, 0) | 0 = −i θ(t) e−iv F qt ,
  2π
q
χ L (q, t) = −i θ(t) 0 | ρ L (q, t) , ρ L (−q, 0) | 0 = i θ(t) eiv F qt ,
2π
and thus
|q| 1 1
χ R (q, ω) = χ L (q, ω) = − .
2π ω − v F |q| + iη ω + v F |q| + iη
(6.51)
Therefore, the density-density response function in the long-wavelength limit is
simply

|q| 1 1
χ0 (q, ω) = χ R (q, ω) + χ L (q, ω) = − ,
π ω − v F |q| + iη ω + v F |q| + iη
(6.52)
284 6 Brief Introduction to Luttinger Liquids

which is the correct expression (6.4) one explicitly derives without passing through
bosonization. The non-interacting compressibility is therefore
1
κ0 = −χ0 (q → 0, 0) = . (6.53)
2πv F
Let us now consider the density-density correlation function at q ∼ k R F + k L F ,
namely
 
χ0k R F +k L F (x, t) = −i θ(t) 0 |  R† (x, t)  L (x, t) ,  L† (0, 0)  R (0, 0) | 0
1  
= −i θ(t) 2 2 1,1 (x, t) − −1,−1 (−x, −t)
4π α
ei(k R F +k L F )x 1 1
= −i θ(t) − c.c. .
4π 2 x − v F t + iα x + v F t − iα
(6.54)
The Fourier transform right at q = k R F + k L F is
1 1 1
χ0 (k R F + k L F , t) = −θ(t) + .
4π v F t − iα v F t + iα

One easily realises that χ(k R F + k L F , ω) diverges logarithmically for ω → 0, as

discussed in Sect. 6.1.

6.2.5 Interaction
Let us now take into account the interaction (6.7), namely
1 
Hint = V (q) ρ(q) ρ(−q) .
2L q

Consistently with the ‘semiclassical’ approximation, we will keep only the interac-
tion scattering processes that correspond to low-energy particle-hole excitations, i.e.,
those close to right and left Fermi momenta. Those processes are shown in Fig.  6.4,
and are associated  to the coupling
 constants g4  V (0), g2  V (0) − V kRF +
k L F and g3  V k R F + k L F in the g-ology jargon. We observe that the umklapp
scattering g4 is the only one that does not conserve separately N R and N L . However,
it is allowed by momentum conservation only if 2k R F ≡ −2k L F , where the equiv-
alence holds apart from the primitive reciprocal lattice vector 2π, which requires
k R F + k L F = π or, equivalently, N R + N L = L/2, thus half-filled density.
We start assuming that the density is away from half-filling, so that only the
scattering processes g4 and g2 are active at low energy. In this case N R and N L
are still conserved quantity. Even though that is not a true symmetry property of
Hint , it is asymptotically so at low enough energy and temperature. In other words,
exactly like in a Fermi liquid, at low energy the system recovers the large sym-
metry of the non-interacting Hamiltonian, which includes independent U R (1) and
U L (1) symmetries for R and L moving fermions, respectively. Here, we are actu-
ally assuming such dynamical symmetry recovery, while in Landau’s Fermi liquids
6.2 Interacting Spinless Fermions 285

V(0) V(0)
g4 : +
L R L R

V(0)
g2 : −
L R L V(kRF + kLF) R

g3 : +
L V(kRF + kLF) R L V(kRF + kLF) R

only if kRF + kLF =

Fig. 6.4 Leading scattering processes yielded by interaction. By assumption, each excitation drawn
as an arrow carries small momentum |q| k R(L)F , thus both initial and final states are close to a
Fermi point

we proved that explicitly. In reality, also in Luttinger liquids one can convincingly
demonstrate that away from half-filling the low-energy effective theory possesses
the large U R (1) × U L (1) of the non-interacting Hamiltonian. In that case, we can
assume that the effective interaction at low energy has the expression

1 
Hint  V (q) ρ(q) ρ(−q)
2L q
 
1  g4 
 g2 ρ R (q) ρ L (−q) + ρ R (q) ρ R (−q) + ρ L (q) ρ L (−q) ,
L q 2
(6.55)
where g4 and g2 might  be different from their bare values, g4  V (0), g2 
V (0) − V k R F + k L F , and just correspond to effective parameters like the f -ones
in Landau’s Fermi liquid theory. Comparing with (6.19), we note that g4 simply yields
a correction to the Fermi velocity v F → v F + g4 /2π. If we redefine v F accordingly,
the full Hamiltonian reads
   
1 
H= π v F ρ R (q) ρ R (−q) + ρ L (q) ρ L (−q) + g2 ρ R (q) ρ L (−q) − μ R N R − μ L N L
L q
   
1 
= π v F ρ R (q) ρ R (−q) + ρ L (q) ρ L (−q) + g2 ρ R (q) ρ L (−q)
L
q =0
πv F  2  g
2
+ N R + N L2 + N R N L − μ R N R − μL N L ,
L L
(6.56)
which implies that now

g2 g2
μR = vF k R F + k L F , μL = vF k L F + kRF , (6.57)
2πv F 2πv F
286 6 Brief Introduction to Luttinger Liquids

in order to have the desired

L
N R(L) = k R(L) F .
2π
We can equivalently write
   
vF ∂φ R (x) 2 ∂φ L (x) 2 g2 ∂φ R (x) ∂φ L (x)
H= dx + + 2 dx
4π ∂x ∂x 4π ∂x ∂x
  
1 ∂φ R (x) ∂φ L (x)
− d x μR + μL .
2π ∂x ∂x
(6.58)
We introduce the linear combinations
1   1  
(x) ≡ √ φ R (x) + φ L (x) , (x) ≡ √ φ L (x) − φ R (x) ,
4π 4π
(6.59)
which, through (6.32), satisfy
  1  
(x) , (y) = − 2i π sign(x − y) − 2i π = −i θ(x − y) ,
4π
so that, introducing the field,

∂(x)
(x) ≡ , (6.60)
∂x
then
 
(x) , (y) = i δ(x − y) .

namely, (x) and (x) are conjugate fields. Correspondingly,

√   √  
φ R (x) = π (x) − (x) , φ L (x) = π (x) + (x) , (6.61)

which, substituted in the Hamiltonian (6.58), lead to

  2   2 
vF ∂(x) g2 ∂(x)
H= dx (x)2 + + dx − (x)2
2 ∂x 4π ∂x
  
1 ∂(x)
− d x μ +μ (x) ,
2π ∂x
(6.62)
where
√   √  
μ = π μ R + μL , μ = − π μ R − μL .
6.2 Interacting Spinless Fermions 287

We apply the canonical transformation

√ 1 2πv F − g2
(x) = K (x) , (x) = √ (x) , K2 = , (6.63)
K 2πv F + g2
after which
   2    
v ∂(x) 1 √ ∂(x) μ
H= dx (x) +
2
− d x μ K +√ (x) ,
2 ∂x 2π ∂x K
(6.64)
where
g22
v = vF 1− . (6.65)
4π 2 v 2F
We now recall that the Fermi velocity is actually v F + g4 /2π, so that

g4 2 g2 g4 g2 g4 g2
v2 = v F + − 22 = v F + + vF + −
2π 4π 2π 2π 2π 2π
   
2V (0) − V k R F + k L F V k R F + kL F
 vF + vF + ,
2π 2π

which is always finite if the interaction in real space V (x) is repulsive, thus K < 1.
If instead V (x) < 0 is attractive, thus K > 1, the effective velocity vanishes when
 
g4 + g2  2V (0) − V k R F + k L F = −2πv F ,

which signals a thermodynamic instability towards phase separation.

Following (6.61), we define
√   √  
φ R (x) = π (x) − (x) , φ L (x) = π (x) + (x) , (6.66)

which satisfy the same commutation relations as φ R (x) and φ L (x) and allow rewrit-
ing (6.64) as
  2  2 
v ∂φ R (x) ∂φ L (x)
H= dx +
4π ∂x ∂x
   (6.67)
1 ∂φ R (x) ∂φ L (x)
− d x μR + μL ,
2π ∂x ∂x

where
1 √ 1 √
1
μR = K + √ μR + K − √ μL ≡ v k RF ,
2 K K
(6.68)
1 √ 1 √ 1
μL = K + √ μL + K − √ μR ≡ v kLF .
2 K K
288 6 Brief Introduction to Luttinger Liquids

The Hamiltonian (6.67) has the form of a non-interacting one for the new Fermi
operators

1 1
 R (x) = √ eiφ R (x) ,  L (x) = √ eiφ L (x) ,
2πα 2πα

apart from a different Fermi velocity v.

6.2.6 Interacting Green’s Functions and Correlation Functions

We note that
√   √ √ 1 
φ R (x) = π (x) − (x) = π K (x) − √ (x)
K
1 √ 1 √ 1
= K + √ φ R (x) + K − √ φ L (x) , (6.69)
2 K K
1 √ 1 √ 1
φ L (x) = K + √ φ L (x) + K − √ φ R (x) ,
2 K K

which also implies, through (6.68) and (6.57) that

x √ 1 √ 1
0 | φ R (x) | 0 = K + √ k RF + K − √ kLF = kRF x ,
2 K K
x √ 1 √ 1
0 | φ L (x) | 0 = K + √ kLF + K − √ k RF = kL F x ,
2 K K

as expected since both are conserved quantities.

6.2.6.1 Green’s Functions

Through (6.48) and (6.49) we readily find that the interacting Green’s functions are
 β 2
1 eik R F x α2
G R (x, t) =    ,
2π x − v t + iα sign(t) x − v t + iα sign(t) x + v t − iα sign(t)
 β 2
1 e−ik L F x α2
G L (x, t) =    ,
2π −x − v t + iα sign(t) x − v t + iα sign(t) x + v t − iα sign(t)
(6.70)
where
1 √ 1 2
β2 = K − √ .
4 K
6.2 Interacting Spinless Fermions 289

Since the Green’s functions at t = 0− are non-analytic on both sides of the complex
plane x → z = x + i y, the momentum distributions

n R(L) (k) = −i d x e−ikx G R(L) (x, 0− ) ,

does not have anymore a jump at the Fermi momenta. However,


∂n R(L) (k)
= − d x e−ik R(L)F x x G R(L) (x, 0− ) ,
∂k k=k R(L)F

is singular due to the large |x| convergence of the integrand provided 2β 2 ≤ 1. In

that case, n R(L) (k) has no jump at k R(L)F but it has a diverging slope. If n ≤ 2β 2 ≤
n + 1 with integer n ≥ 1, all derivatives of order less than n + 1 are finite, but those
with order ≥ n + 1 are singular. Therefore, even though n R(L) (k) is smooth, the
singularity of its derivatives still allows identifying a Fermi surface.

6.2.6.2 Correlation Functions

The long-wavelength component of the local density is

√ ∂(x) √ √ ∂(x) √
ρ(x) = ρ R (x) + ρ L (x) = 4π = K 4π = K ρ(x)
∂x ∂x
so that the interacting density-density response function at small |q| k F can be
readily found through (6.52) and is

|q| 1 1
χ(q, ω) = K − , (6.71)
π ω − v |q| + iη ω + v |q| + iη

which shows that the particle-hole excitations at small |q| are still exhausted by an
acoustic mode with renormalised velocity v. The interacting compressibility

1 vF
κ = −χ(q → 0, 0) = K =K κ0 , (6.72)
2π v v
and is lower than the non-interacting one (6.53) for repulsive interaction, whereas
it is bigger for attractive one diverging for v → 0. That, as mentioned, signals the
onset of phase separation.
Let us consider now the component of the local density at q ∼ −k R F − k L F ,
namely,
1 −iφ R (x) −iφ L (x)
ρ−k R F −k L F (x) =  R† (x)  L (x) = e e
2πα
 
1 i
= e−i(φ R (x)+φ L (x)) e+ −iφ R (x),−iφ L (x) /2 = e−i(φ R (x)+φ L (x))
2πα 2πα
√ √ √  
i i i
= e−i 4π (x)
= e−i 4π K (x)
= e−i K φ R (x)+φ L (x) ,
2πα 2πα 2πα
(6.73)
290 6 Brief Introduction to Luttinger Liquids

and thus
√  
i φ R (x)+φ L (x)
ρk R F +k L F (x) =  L† (x)  R (x) = − ei K . (6.74)
2πα
It follows that
 
k R F +k L F x K
ei α2
ρ−k R F −k L F (x, t) ρk R F k L F (0) =    ,
4π α
2 2 x − vt + iα x + vt − iα

which vanishes for x → ∞ with an interaction dependent exponent K . In particular,

it vanishes more slowly than in absence of interaction if the latter is repulsive, which,
as mentioned, corresponds to a quasi long-range CDW order; the true long-range one
prevented by Mermin-Wagner theorem.
Let us consider now the Cooper pair operator at zero total momentum, assuming
zero current and thus k R F = k L F = k F , which is
† 1 −iφ R (x) iφ L (x)
† (x) =  R (x)  L† (x) = e e
2πα
 
1 i(φ L (x)−φ R (x)) + −iφ R (x),iφ L (x) /2 i
= e e =− ei(φ L (x)−φ R (x))
2πα 2πα
√ √  √
i i i
=− ei 4π (x) = − ei 4π/K (x) = − e−i φ R (x)−φ L (x) / K ,
2πα 2πα 2πα
(6.75)
hence its correlation function
1/K
1 α2
† (x, t) (0) =    ,
4π 2 α2 x − vt + iα x + vt − iα
decays more slowly than in absence of interaction when the latter is attractive, a quasi
long-range superconducting order, whereas it decays faster when the interaction is
repulsive.
It follows that the corresponding CWD and superconducting (SC) susceptibilities,
both of which diverge as ln 1/T in absence of interaction, behave as

κC DW ∼ T 2K −2 , κ SC ∼ T 2/K −2 . (6.76)

6.2.7 Umklapp Scattering

Let us now consider the case in which the umklapp scattering g3 in Fig. 6.4 becomes
allowed at low-energy. That occurs at half-filling when k R F + k L F = π ≡ −k R F −
k L F , and corresponds to an interaction
 
Humklapp  g3 d x  R† (x + )  R† (x)  L (x)  L (x + ) + H .c.
  √
g3   g3
=− 2 2 d x cos 2 φ R (x) + φ L (x) = − 2 2 d x cos 16 π (x)
2π α 2π α
 √
g3
=− 2 2 d x cos 16 π K (x) ,
2π α
(6.77)
6.2 Interacting Spinless Fermions 291

where the bare value of g3 is V (q = π). For simplicity, we assume that k R F = k L F ≡

k F = π/2. The fully interacting Hamiltonian at half-filling thus reads, dropping for
simplicity the chemical potential terms and assuming zero current,
     
vF ∂(x) 2 g2 ∂(x) 2
H= dx (x) +2
+ dx − (x) 2
2 ∂x 4π ∂x
 √
g3
− 2 2 d x cos 16 π (x)
2π α
   2  
v ∂(x) g3 √
= dx (x) +
2
− 2 2 d x cos 16 π K (x) .
2 ∂x 2π α
(6.78)
This Hamiltonian is known as sine-Gordon model, and its physics is equivalent to
the model shown in Fig. 6.5. If we denote the position of the atom n in that figure as
rn = n a + u(na), where u(na) is the displacement with conjugate variable p(na),
the Hamiltonian of the model in Fig. 6.5 is

 C  2 
L L
p(na)2 
H∗ = + 2 u(na) − u (n + 1)a −U cos u(na) .
2M 2a
n=1 n=1

If a → 0 such that n a → x with x a continuous variable ∈ [0, La], where La goes

to infinity in the thermodynamic limit, then
  2 
dx p(x)2 C ∂u(x) U
H∗ = + − d x cos u(x)
a 2M 2 ∂x a
  2  (6.79)
v∗ ∂ Q(x) U
= d x P(x)2 + − d x cos β Q(x) ,
2a ∂x a

where, taking as usual  = 1,

1 1 C
p(x) = P(x) , u(x) = β Q(x) , β 4 = , v∗2 = .
β MC M

The Hamiltonian (6.79) coincides with (6.78) upon renaming the parameters and
defining a dimensionless variable x → x/a. However, in the model of Fig. 6.5 the
physics is more transparent. The acoustic mode with velocity v∗ is just the acoustic
phonon of the lattice of atoms, which is gapless since a uniform shift of all atomic
positions does not cost elastic energy. However, that shift costs potential energy. If
the mass M or the elastic constant C are very small, thus β is large, the gapless
acoustic mode must survive. In the opposite case, the potential pins the atoms, which
can only oscillate in each potential well without drifting. We thus expect a transition
as function of β from a phase at large β where the model is gapless into a phase at
small β where the atoms localise and the spectrum has a gap.
292 6 Brief Introduction to Luttinger Liquids

a
n-1 n n+1

U(x)

Fig. 6.5 A simple model whose Hamiltonian is equivalent to that in (6.78): L atoms of mass M on
a one-dimensional rail are coupled to each other by springs whose equilibrium length and stiffness
are a and C/a 2 , respectively. Specifically, atom n is coupled to atoms n − 1 and n + 1. In addition,
the atoms feel a periodic cosine potential with period a: U (x) = −U cos(x)

The simplest way to identify such transition in the Hamiltonian (6.78) is recognis-
ing that in the gapless phase the precise value of the short-distance cutoff α, namely
of the lattice spacing a in the model of Fig. 6.5, must not matter and therefore can
be sent to zero provided the umklapp operator is properly normal-ordered, namely,

1 √ 1 √ 1 √  
cos 16 π (x) = 16 π K (x) = 2 cos 2 K φ R (x) + φ L (x)
cos
α2 α2 α
4K −2
√
=α : cos 16 π K (x) : .
(6.80)
In this way, we readily find that α can be safely sent to zero only if K > 1/2.
If, instead, K < 1/2, which may occur only if the interaction is sufficiently repul-
sive, the umklapp term grows in strength as α → 0, which implies that it is more
appropriate first to minimise the potential, and next add quantum effects. Assuming
a nearest neighbour repulsion, thus V (q) = V cos q and g3  V (π) = −V < 0, the
potential is minimum when

g3 √ V √
− cos 16 π (x)  cos 16 π (x) = −1 ,
2π α2 2
2π α
2 2

thus
√    
16 π (x) = 2 φ R (x) + φ L (x) = 2n + 1 π , n ∈ Z ,
 
noticing that the uniform term 2 k R F + k L F x = 2π x is also multiple of 2π because
x is actually an integer in units of the lattice spacing.
Since √φ R(L) (x) is defined modulo 2π, there are actually only two inequivalent
minima: 16 π (x) = ±π. That actually corresponds to the fact that, for K < 1/2,
the fermions localise either on the even sites or on the odd ones, thus a double
6.2 Interacting Spinless Fermions 293

degenerate state. More specifically, the local density that includes both uniform
component 1/2 and staggered component at 2k F = π, is

1
ρ(x = ) = + ρ2k F (x) + ρ−2k F (x)
2
1  i √ i √  1 1 √ 
= + e−i 4π (x) − ei 4π (x) = + sin 4π (x)
2 2πα 2πα 2 πα
1 eiπ x  π 1
= + sin ± = ± (−1) δρ ,
2 πα 2 2
(6.81)
thus a staggered modulation of the local density.

6.2.8 Behaviour Close to the K = 1/2 Marginal Case

We intentionally have not discussed the case of K = 1/2, where, according to (6.80)
the umklapp term remains finite for α → 0, which implies it is a marginal operator.
To assess whether it is marginally irrelevant or relevant, we write that

1 1 − g/2πv
K = , |g| 2πv ,
2 1 + g/2πv

where g quantifies the deviation of K from 1/2, and assume that also g3 2πv. In
(6.78) we perform the following canonical transformation

1 √
(x) =  (x) , (x) = 2K 
(x) ,
2K

where
g 1 g
2K  1 − , 1+ ,
2πv 2K 2πv
so that the Hamiltonian becomes
   
v  1 ∂ (x)
2
g3 √
H= d x 2K (x)2 + − d x cos 8 π  (x)
2 2K ∂x 2π α
2 2
  2
v  ∂ (x)
 dx (x)2 +
2 ∂x
   
g ∂ (x)
2
g3 √
+ dx −  (x)2 − 2 2 d x cos 8 π  (x)
4π ∂x 2π α
     
g ∂ (x)
2
g3 √
≡ H0 + v dx −  (x)2 − 2 2 d x cos 8 π  (x)
4πv ∂x 2π vα
≡ H0 + v δ H .
(6.82)
294 6 Brief Introduction to Luttinger Liquids

We shall treat δ H in perturbation theory on the imaginary time-axis ivt → τ , with

τ having dimension of a length. The S-operator is therefore
 ! vβ 
Sα = Tτ e− 0 dτ δ H (τ ) , (6.83)

where δ H (τ ) is the imaginary time evolution with the unperturbed Hamiltonian

H0 , and we explicitly indicate the short-distance cutoff α. The reason is that, in
the ‘semiclassical’ approach we have adopted, α represents the indeterminacy of
the position (x, τ ), as we earlier mentioned. Therefore, when expanding the expo-
nential in Sα , the multidimensional integral must be performed excluding regions
where pair of coordinates (x, τ ) and (x  , τ  ) get closer than α. The procedure we
shall follow will be to integrate out regions where the distance between any pair
of coordinates (x, τ ) and (x  , τ  ) is greater than α but lower than λα, with λ > 1,
making use of the operator product expansion of Sect. 6.2.3. The outcome of such
partial integration will be interpreted in terms of a new effective Sλα defined with a
larger short-distance cutoff λα. Since τ ∈ [0, vβ] and x ∈ [0, L], should we iterate
that procedure till λα ∼ vβ  L without encountering any singularity, we would
end up in a theory where perturbation theory is well behaved. Put differently, we
mentioned that perturbation theory in one dimension is ill defined because of the
proliferation of logarithmic singularities like ln vk F β  ln vβ/α at finite tempera-
ture, or ln L/α on a finite size. Evidently, should α ∼ vβ  L, the logarithms would
be vanishingly small and perturbation theory could be safely performed. This pro-
cedure essentially implements a Renormalisation Group (RG) scheme, here akin a
real space decimation.
In imaginary time, φR (x, τ ) + φL (x, τ ) = φR (x + i τ ) + φL (x − i τ ). We can
therefore define new variables z ≡ x + iτ and z̄ ≡ z ∗ = x − iτ , so that φR (x +
i τ ) ≡ φR (z) is holomorphic, while φL (x − i τ ) ≡ φL (z̄) antiholomorphic. It fol-
lows that
1   
 (z, z̄) = √ φ R (z) + φL (z̄) ,
4π
so that

∂ (z, z̄) 1 ∂φ R (z) ∂ (z, z̄) 1 ∂φ L (z̄)

=√ , =√ .
∂z 4π ∂z ∂ z̄ 4π ∂ z̄

Therefore

∂ (x, τ )
2
 1 ∂φ R (x, τ ) ∂φ L (x, τ ) 1 ∂φ R (z) ∂φ L (z̄)
− (x, τ )2 = =
∂x π ∂x ∂x π ∂z ∂ z̄
∂ (z, z̄) ∂ (z, z̄)
=4 ,
∂z ∂ z̄
6.2 Interacting Spinless Fermions 295

and thus the first order correction of the S-operator can be simply written as
 vβ 
i
Sα(1) = − dτ δ H (τ ) ≡ − dz d z̄ δ H (z, z̄)
0 2
  
i g ∂ (z, z̄) ∂ (z, z̄) g3 √ 
=− dz d z̄ − 2 2 cos 8π  (z, z̄) .
2 πv ∂z ∂ z̄ 2π vα
√ (6.84)
We √observe that Sα(1) is in reality independent of α, since cos 8π  (z, z̄)/α2 =:
cos 8π  (z, z̄) :, and the normal ordered product is what really enters the pertur-
bative calculations.
To simplify notations, we shall drop the apex and redefine  →  and  → ,
as well as
g g3
→ g, → g3 ,
2πv 2πv
and finally write δ H (z, z̄) = δ H2 (z, z̄) + δ H3 (z, z̄) where

∂(z, z̄) ∂(z, z̄) g3 √

δ H2 (z, z̄) = 2g , δ H3 (z, z̄) = − 2 cos 8π (z, z̄) .
∂z ∂ z̄ πα
The second order term, fixing a specific time order, is instead

1  
Sα(2) = − dz d z̄ dz  d z̄  θ τ − τ  δ H (z, z̄) δ H (z  , z̄  ) . (6.85)
4

As discussed above, we divide Sα(2) into a component Sλα (2)

where |z − z  | = |z̄ −
 (2)  
z̄ | > λ α > α, and another δS where α < |z − z | = |z̄ − z̄ | ≤ λ α, in which case
we make use of the operator product expansion. Specifically, in δS (2) we set z  =
z +  eiθ and z̄  = z̄ +  e−iθ where θ ∈ [−π, 0] since τ > τ  in the time-ordered
equation (6.85), and  ∈ [α, λα]. That implies a phase space vanishing as 2 , so that
only singular terms diverging as least as 1/2 can contribute to δS (2) .
To accomplish our goal, we first derive some preliminary results. Since
         
(+) (−) (+) (−)
φ R (z) , φ R (z  ) = ln α + i z − z  , φ L (z̄) , φ L (z̄  ) = ln α − i z̄ − z̄  ,

the following expressions readily follow for z ∼ z 

√ √ √  
 α2 
e±i 2 φ R (z) e∓i 2 φ R (z ) =   2 : e±i 2 φ R (z)−φ R (z ) :
α − i z − z

α2 √ ∂φ R (z)  
   2 1±i 2 z − z
∂z
α − i z − z

α2 √ ∂(z, z̄)  
=   2 1 ± i 8π z − z ,
∂z
α − i z − z
296 6 Brief Introduction to Luttinger Liquids
√
∂φ R (z) ±i √2 φ R (z  ) ∂φ R (z) ±i √2 φ R (z  ) 2 √ 
e = : e : ∓    e±i 2 φ R (z ) ,
∂z ∂z α−i z−z
√ √ 2 √  
 α ±i 2 φ L (z̄)−φ L (z̄  )
e±i 2 φ L (z̄) e∓i 2 φ L (z̄ ) =   2 : e :
α + i z̄ − z̄ 

α2 √ ∂φ L (z̄)  
   2 1±i 2 z̄ − z̄ 
∂ z̄
α + i z̄ − z̄ 

α2 √ ∂(z, z̄)  
=   2 1 ± i 8π z̄ − z̄  ,
∂ z̄
α + i z̄ − z̄ 
√
∂φ L (z̄) ±i √2 φ L (z̄  ) ∂φ L (z̄) ±i √2 φ L (z̄  ) 2 √ 
e = : e : ±    e±i 2 φ L (z̄ ) .
∂ z̄ ∂ z̄ α + i z̄ − z̄
(6.86)

By means of the above results, we readily find that for z = z  +  eiθ , with the OPE
prescription and dropping constant terms as well as terms non-singular as  → 0,

g32 √ g2 1
δ H3 (z, z̄) δ H3 (z  , z̄  )  cos 32π (z, z̄) + 32  2  2
2π α
2 4
4π z−z 
z̄ − z¯

√ ∂(z, z̄)   √ ∂(z, z̄)  
1 + i 8π z − z 1 + i 8π z̄ − z̄ 
∂z ∂ z̄

√ ∂(z, z̄)   √ ∂(z, z̄)  
+ 1 − i 8π z − z 1 − i 8π z̄ − z̄ 
∂z ∂ z̄
g32 √ 4g 2 1 ∂(z, z̄) ∂(z, z̄) 4g 2 1 ∂(z, z̄) ∂(z, z̄)
 cos 32π (z, z̄) − 3 2 − 3 2 .
2π α
2 4 π  ∂z ∂ z̄ π  ∂z ∂ z̄
(6.87)
Similarly,
 
δ H2 (z, z̄) δ H2 (z  , z̄  ) ∼ O 0  0 ,
2 gg3 1 √
δ H2 (z, z̄) δ H3 (z  , z̄  ) + δ H3 (z, z̄) δ H2 (z  , z̄  ) = 2 2 cos 8π (z, z̄)
π α 2
We still have to integrate over  and θ:
  λα  0
1 1
dz  d z̄  = −2i  d dθ = −2π i ln λ ,
 2
α  2
−π

so that, finally,
 √
i ∂(z, z̄) ∂(z, z̄) 2 gg3
δS (2) = − ln λ dz d z̄ 4g32 − cos 8π (z, z̄) .
2 ∂z ∂ z̄ πα 2
(6.88)
6.2 Interacting Spinless Fermions 297

In conclusion, we have found that

(2)
Sα = Sα(1) + Sα(2) + · · · = Sα(1) + δS (2) + Sλα + ··· ,

(1)
which allows defining Sλα ≡ Sα(1) + δS (2) . Comparing (6.88) with (6.84) we can
formally define up to second order the new coupling constants that refer to the larger
short-distance cutoff λα,

g + δg = g + 2g32 ln λ , g3 + δg3 = g3 + 2g g3 ln λ .

Similarly, writing λ = es and denoting as g(s) and g3 (s) the coupling constants with
short-distance cutoff es α, we could repeat the same procedure of integrating over
the regions es α < |z − z  | < es+δs α, and thus find

g(s + δs) = g(s) + 2g3 (s)2 δs , g3 (s + δs) = g3 (s) + 2g(s) g3 (s) δs ,

which can be recast as the following differential equations

∂g(s) ∂g3 (s)

= 2g3 (s)2 , = 2g(s) g3 (s) , (6.89)
∂s ∂s

with boundary conditions the bare coupling constants g(0) = g and g3 (0) = g3 .1
We observe that (6.89) is invariant under g3 → −g3 . That is not surprising since

1 We emphasise that the validity of the equations (6.89) is not guaranteed by the second order
calculation we performed. Indeed, suppose that s 1, and we write
∞
 ∞
 (n)
g(s)  g (n) s n , g3 (s)  g3 s n ,
n=0 n=0

(n) (n+1)
where g (n) and g3 derive from Sα , then, if (6.89) were exact, for n ≥ 0


n 
n
(n + 1) g (n+1) = 2 g3(n−m) g3(m) , (n + 1) g3(n+1) = 2 g (n−m) g3(m) ,
m=0 m=0

implying that higher order terms are all determined from the first
order correction we have calculated. For instance, the third order Sα(3)
yields corrections ∝ ln2 λ which must correspond to

δg = 4 g(0) g3 (0)2 ln2 λ , δg3 = 2 g3 (0)3 ln2 λ + 2 g(0)2 g3 (0) ln2 λ .

If that indeed happens at all orders, we say that the theory is renormalisable. In the present case,
renormalisability can be indeed proven at few orders in perturbation theory. We shall assume it does
hold at any order. We also note that the right hand sides of the equations (6.89) are just the leading
order terms. In general,
∂g(s)   ∂g3 (s)  
= β g(s), g3 (s) , = β3 g(s), g3 (s) ,
∂s ∂s
298 6 Brief Introduction to Luttinger Liquids

g3

Fig. 6.6 Renormalisation group flow corresponding to the equations (6.89). Starting from the initial
values of g(0) = g and g3 (0) = g3 , the solutions g(s) and g3 (s) move along the paths that are shown
as s grows. Any initial point within the blue coloured region flows at g(∞) = g3 (∞) = ∞, while
within the red coloured one at g(∞) = g3 (∞) = −∞. All initial points within the green coloured
region flow to g3 (∞) = 0 and g(∞) finite. We remark that the horizontal axis g3 = 0 corresponds
to a gapless system, and is stable for g < 0 and unstable otherwise

g3 → −g3 under the unitary transformation  L (x) → i  L (x). Therefore the RG

flow is independent of the sign of g3 , though the physical properties do depend on
it, as we discuss later. The equations in (6.89) can be readily solved noticing that

∂ 2 g(s) ∂g(s) ∂g(s)2

= 4 g3 (s)2 g(s) = 2 g(s) = .
∂s 2 ∂s ∂s
We shall not explicitly solve the above simple equation, but just show the renormal-
isation group flow diagram in Fig. 6.6. For positive g, thus K < 1/2, both coupling
constants flows to strong coupling, as expected from (6.80). However, the flow is
still towards strong coupling even if g < 0 provided g3 > −g.
We already discussed the fixed point at g3 → −∞, which corresponds to a pinned
CDW where the fermions occupy either the even or the odd sites, see top panels in
Fig. 6.7. The fixed point at g3 → +∞ implies instead that

g3 √
− cos 16 π (x) = −1 ,
2π α
2 2

where β and β3 can be expanded in powers of the coupling constants. For instance, Sα(3) contains
terms ∼ g 3 ln λ that contribute to the higher order corrections of β. Equation (6.89) is therefore
valid provided g, g3 1.
6.2 Interacting Spinless Fermions 299

CDW

2n-1 2n 2n+1 2n-1 2n 2n+1

BDW

2n-1 2n 2n+1 2n-1 2n 2n+1

Fig. 6.7 Charge density wave (CDW), top panels, versus bond density wave (BDW), bottom panels.
The density profile is the blue curve, and is peaked when there is excess charge, red dots. The BDW
is associate to spontaneous dimerisation where the bond strengths alternate along the chain, strong
bonds in green and weak one in black

thus
√  
16 π (x) = 2 φ R (x) + φ L (x) = 2n π , n ∈ Z ,
and still admits only two inequivalent minima at n = 0, 1. This state actually cor-
responds to a spontaneously dimerised chain, or bond density wave (BDW), where
the density is peaked in the middle of a bond, and minimum in the nearest neigh-
bour ones, see bottom panels in Fig. 6.7. The two minima at n = 0, 1 correspond
to the two dimerisation patterns shown in that figure. To better understand the dif-
ference between CDW and BDW from the bosonization viewpoint, let us redefine
φ R(L) (x) → k F x + φ R(L) (x) so that φ R(L) (x) are now slowly varying fields with
vanishing space average. It follows that the Fermi fields become
π π
 R (x) ∼ eik F x eiφ R (x) = ei 2 x
eiφ R (x) ,  L (x) ∼ e−ik F x e−iφ L (x) = e−i 2 x
e−iφ L (x) .

The CDW, C DW (x), and   B DW (x),

 BDW,  order
 parameters are the slowly varying
(s-v) components of cos πx ρ(x) and sin πx ρ(x), as one can easily realise from
Fig. 6.7. Therefore
   
   
C DW (x) = cos πx ρ(x)  cos πx  R† (x)  L (x) +  L† (x)  R (x)
s-v s-v
1 1 √
= e−iφ R (x) e−iφ L (x) + eiφ L (x) eiφ R (x) = sin 4π (x) ,
2πα πα
    (6.90)
   
 B DW (x) = sin πx ρ(x)  sin πx  R† (x)  L (x) +  L† (x)  R (x)
s-v s-v
1 −iφ R (x) −iφ L (x) iφ L (x) iφ R (x) 1 √
= −ie e +ie e = cos 4π (x) .
2πα πα
√ √
We recall that the CDW state is characterised
√ by 16π (x) = 2 4π (x) =
±π,√and indeed C DW (x) ∼ sin 4π (x) = ±1 is finite, while  √B DW (x) ∼
cos 4π (x) = 0. On the contrary, the BDW state is characterised by 16π (x)
300 6 Brief Introduction to Luttinger Liquids
√ √
= 2 4π (x) =√0, 2π, and indeed C DW (x) ∼ sin 4π (x) = 0 is zero, while
 B DW (x) ∼ cos 4π (x) = ±1 is finite.
Let us end discussing the transition from the green to either the red or blue
regions in the RG flow diagram, Fig. 6.6. The transition point corresponds to any
initial g and g3 along the black lines in Fig. 6.6, i.e., g3 = g or g3 = −g with g < 0.
Along these lines, the flow tends towards g(∞) = g3 (∞) = 0, which, in the original
model, corresponds to K (∞) = 1/2. On the contrary, beyond the transition, the
system develops a single particle gap, so that the compressibility vanishes and thus
K (∞) = 0, see (6.72). The jump of K (∞) is one of the features of such transition,
which actually belongs to the Berezinskii-Kosterlitz-Thouless universality class.
For simplicity, let us consider an initial point g3 = −g, in which case such rela-
tionship persists along the flow so that the RG equation (6.89) simplifies into

∂g(s) s dg g g
= 2g(s)2 ⇒ = 2s ⇒ g(s) = = .
∂s 0 g 2 1 − 2gs 1 − 2g ln λ

Indeed, if g < 0, g(s) → 0 for s → ∞. On the contrary, if 0  g 1, g(s) diverges

at λ corresponding to a length scale ξ = α e1/2g , which is actually the correlation
length in the CDW phase.

6.2.8.1 SU(2) Invariant Line

The generic spinless fermion Hamiltonian (6.6) plus (6.7) is invariant under the global
charge U (1) symmetry. Here we show that the line g3 = g in the flow diagram of
Fig. 6.6 actually corresponds to models with enlarged SU (2) symmetry.
Let us consider again (6.86) but now at equal times τ = τ  , and x ∼ y

 √ √   2  2 
1 i 2 φ (x) −i 2 φ (y) 1 α + i(x − y) α − i(x − y)
e R ,e R = −
4π 2 α2 4π 2 (x − y)2 + α2 (x − y)2 + α2
√ ∂φ R (x)  
1+i 2 x−y
∂x
1 x−y √ ∂φ R (x)  
=i δα (x − y) 1 + i 2 x−y
2π (x − y)2 + α2 ∂x
√
2 ∂φ R (x)
 −δ(x − y) 2 ,
4π ∂x
 √ √    √
2 ∂φ R (x) ei 2 φ R (x) 1 −1 1
, =   +   ei 2 φ R (y)
4π ∂x 2πα 4π 2 α α−i x − y α+i x − y
√
ei 2 φ R (x)
= −δ(x − y) .
2πα

We note that the operators

√ √ √
2 ∂φ R (x) + e−i 2 φ R (x) − ei 2 φ R (x)
Jz R (x) ≡ , J R (x) ≡ i , J R (x) ≡ −i ,
4π ∂x 2πα 2πα
(6.91)
6.2 Interacting Spinless Fermions 301

realise an SU (2) algebra of right-moving fields. Similarly,

√ √ √
2 ∂φ L (x) + ei 2 φ L (x) − e−i 2 φ L (x)
Jz L (x) ≡ , JL (x) ≡ i , JL (x) ≡ −i ,
4π ∂x 2πα 2πα
(6.92)
do the same but for left-moving ones. We also note that, using the OPE,

1 + − − +
 1 ∂φ R(L) (x) 2
J R(L) (x) J R(L) (x) + J R(L) (x) J R(L) (x) = = 2 Jz R(L) (x)2 ,
2 4π 2 ∂x

so that
2
v v ∂φ R(L) (x)
= 2π
J R(L) (x) · J R(L) (x) ,
4π 3 ∂x
 
±
where J R(L) (x) = Jx R(L) (x), Jy R(L) (x), Jz R(L) (x) , and, as usual, J R(L) (x) =
Jx R(L) (x) ± i Jy R(L) (x). Therefore, H0 in (6.82) can be simply written as
  
v
H0 = 2π d x J R (x) · J R (x) + J L (x) · J L (x) , (6.93)
3

which makes the SU (2) symmetry explicit. In addition,2

∂φ R (x) ∂φ L (x)
g
= 2g J Rz (x) JLz (x) ,
4π ∂x ∂x 2

g3 √  
− 2 2 cos 2 φ R (x) + φ L (x) = g3 J R+ (x) JL− (x) + J R− (x) J + L(x)
2π α
 
= 2g3 J Rx (x) JL x (x) + J Ry (x) JL y (x) ,
(6.94)
so that the Hamiltonian (6.82) can be written as
  
v
H = 2π d x J R (x) · J R (x) + J L (x) · J L (x)
3
   
+ 2 d x g J Rz (x) JLz (x) + g3 J Rx (x) JL x (x) + J Ry (x) JL y (x) ,
(6.95)
and is SU (2) symmetric if g3 = g. Moreover, if g = g3 < 0, the black line in the
lower half-plane of Fig. 6.6, the model asymptotically flows to g(∞) = g3 (∞) = 0,
which thus describes a gapless SU (2) invariant phase right at K (∞) = 1/2. If, on the

2 Note that
√ √ √     √  
e−i 2 φ R (x)
e−i 2 φ L (x)
= e−i 2 φ R (x)+φ L (x)
e−2 φ R (x),φ L (x) /2
= −e−i 2 φ R (x)+φ L (x)
.
302 6 Brief Introduction to Luttinger Liquids

contrary, g = g3 > 0, the model flows to g(∞) = g3 (∞) = ∞. As we discussed,

this point corresponds to a BDW which, in spin language, is commonly known as a
spin Peierls state where the strong bonds in Fig. 6.7 represent pairs of spins coupled
into singlets.
The line g3 = −g, though behaves under RG as that with g = g3 , corresponds
instead to an anisotropic system that still flows to the gapless SU (2) point if g < 0,
but to g(∞) = −g3 (∞) = −∞ for g > 0.

6.2.8.2 Luther-Emery Point

For initial g and g3 within the red and blue regions in Fig. 6.6, the RG equations
(6.89) predict that g flows to infinity for any g3 = 0. Therefore, g will unavoidably
reach along its evolution a value g∗ /2πv = 3/5 such that

1 1 − g∗ /2πv 1
K∗ = = .
2 1 + g∗ /2πv 4

At this point, known as the Luther-Emery point, the Hamiltonian becomes, see (6.78),
   2  
v∗ ∂(x) g3∗ √
H∗ = dx (x) + 2
− 2 2 d x cos 4 π (x)
2 ∂x 2π α
  2  2    
v∗ ∂φ R (x) ∂φ L (x) g3∗
= dx + − 2 2 d x cos φ R (x) + φ L (x) ,
4π ∂x ∂x 2π α
(6.96)
where v∗ and g3∗ are the renormalised
 velocity and umklapp strength at g∗ . We
observe that φ R (x) + φ L (x) = 2 φ R (x) + φ L (x) , which implies that φ R(L) (x) has
a Fermi momentum k R(L)F = 2k R(L)F = π. If we define new Fermi fields

1 1
 R (x) = √ eiφ R (x) ,  L (x) = √ e−iφ L (x) , (6.97)
2πα 2πα
 
assuming that φ R (x), φ L (y) = −iπ, then

1  
† †
cos φ R (x) + φ L (x) = −i  R (x)  L (x) + i  L (x)  R (x) .
2π 2 α2
Therefore, through (6.41), the Hamiltonian can be written as

† ∂ R (x) † ∂ L (x)
H∗ = −i v∗ d x  R (x) −  L (x)
∂x ∂x
  
† †
+ i g3 d x  R (x)  L (x) −  L (x)  R (x) ,
6.3 Spin-1/2 Heisenberg Chain 303

which reads in momentum space

  
H∗ = v∗ k d † d Rk+k †
− d Lk−k d Lk−k
Rk+k RF RF LF LF
k

(6.98)

† †
+ i g3 d Rk+k d Lk−k − d Lk−k d Rk+k .
RF LF LF RF

The Hamiltonian (6.98) describes an insulator with conduction and valence band
dispersions, respectively,
" "
E c (k) = v∗2 k 2 + g32 , E v (k) = − v∗2 k 2 + g32 .

In the spirit of the renormalisation group, the Hamiltonian (6.98) should be repre-
sentatives of all Hamiltonians with parameters inside the red and blue regions of
Fig. 6.6.

6.3 Spin-1/2 Heisenberg Chain

Let us now consider an anisotropic spin-1/2 Heisenberg model in one dimension,

L−1 
 
H=J Si x Si+1x + Si y Si+1y + Si z Si+1z , (6.99)
i=1

where  quantifies the anisotropy. The Hamiltonian for  = ±1 has only U (1) sym-
metry of rotations around the spin z-axis. On the contrary, at  = ±1 the symmetry
enlarges to SU (2), and the model becomes the isotropic Heisenberg antiferromagnet
for  = 1 and ferromagnet for  = −1.3
We can map the Hamiltonian (6.99) onto a spinless fermion model via the so
called Jordan-Wigner transformation. Specifically, we write
1 1
Si z = − ci† ci = − n i , (6.100)
2 2
so that Si z = 1/2 and Si z = −1/2 correspond, respectively, to an empty and occupied
site. We note that

Si+ = (−1)i ξi ci , Si− = (−1)i ci† ξi , (6.101)

3 We observe that it is always possible to change the sign of Si x Si+1x + Si y Si+1y by a staggered
rotation around z such that

Si x → (−1)i Si x , Si y → (−1)i Si y .
304 6 Brief Introduction to Luttinger Liquids

are good representation for spin operators, where the ‘string‘ operator is defined as

#i−1 $
i−1
 
ξi ≡ eiπ j=1 nj
= 1 − 2n j (6.102)
j=1

   
Indeed, since ξi2 = 1 and ξi , ci† ci = ξi , ci ci† = 0, then
     
Si+ , Si− = ci ci† − ci† ci = 1 − 2n i = 2Si z , Si+ , Si z = −(−1)i ξi ci , n i = −Si† .

Moreover, noticing that, for i > j, ξi c j = −c j ξi while, for i ≤ j, ξi c j = c j ξi , one

can readily verify that the operators (6.101) commute at different sites. Let us rewrite
the Hamiltonian (6.99) in terms of the above representation of the spin operators.
We note that
1 + −  1 
Si x Si+1x + Si y Si+1y = S S + Si− Si+1
+
= − ξi ci ci+1†
ξi+1 + ci† ξi ξi+1 ci+1
2 i i+1 2
1   †    1 
= − ci 1 − 2n i ci+1 + ci† 1 − 2n i ci+1 = − ci† ci+1 + ci+1 †
ci ,
2 2
1 1 1 1 1
Si z Si+1z = − ni − n i+1 = + n i n i+1 − n i − n i+1 .
2 2 4 2 2

Therefore, the Hamiltonian (6.99) can be also written as

J  † †
  1 1 1
H =− ci ci+1 + ci+1 ci + J  + n i n i+1 − n i − n i+1
2 4 2 2
i i
J  † †
 
≡− ci ci+1 + ci+1 ci + J  n i n i+1
2
i i
 
= (k) ck† ck + V (q) ρ(q) ρ(−q) ,
k q
(6.103)
where we have dropped a constant as well as a term proportional to the total number N
of fermions, which is conserved, and (k) = J cos k while V (q) = J  cos(q). The
Hamiltonian (6.103) is therefore a particular case of the spinless fermion Hamiltonian
we have earlier discussed, so that we can simply borrow all previous results.
Specifically, in the subspace of vanishing total magnetisation, i.e.,

1   L
0 = Sz = 1 − 2n i = − N ,
2 2
i
6.4 The One-Dimensional Hubbard Model 305

the model has half-filled density of spinless fermions, and therefore the bosonized
Hamiltonian reads
   
vF ∂φ R (x) 2 ∂φ L (x) 2 g2 ∂φ R (x) ∂φ L (x)
H= dx + + 2 dx
4π ∂x ∂x 4π ∂x ∂x

g3  
− 2 2 d x cos 2 φ R (x) + φ L (x)
2π α
  
g2 ∂φ R (x) ∂φ L (x) g3  
≡ H0 + d x − cos 2 φ R (x) + φ L (x) ,
4π 2 ∂x ∂x 2π 2 α2
(6.104)
where now the bare values of the coupling constants are g2 = 2J  and g3 = −J .
If we absorb g2 through the canonical transformation (6.63), we arrive at the same
conclusion as before that the umklapp scattering is irrelevant as long as K > 1/2. On
the ferromagnetic side,  < 0, K > 1 and the umklapp is always irrelevant. There,
the worst can happen is phase separation when the effective velocity (6.65) vanishes.
The only region where we physically expect phase separation is for  ≤ −1, where
the easy axis anisotropy would favour a ferromagnetic state with all spin polarised
parallel or anti parallel to z. Since we force to have equal number of up and down
spins, all equal spins have to cluster together, thus the phase separation. Therefore,
we expect a gapless phase for any −1 <  ≤ 0.
Considering instead the antiferromagnetic side,  > 0 and K < 1, the only region
where we expect a gap opening is for  > 1, which corresponds to an easy axis anti-
ferromagnet. That is essentially an Ising antiferromagnet that can order at zero tem-
perature even in one dimension, since the ordered phase breaks a discrete symmetry.
Therefore   1 corresponds to K  1/2, and its gapless nature implies that the
model is within the green region in Fig. 6.6, which also implies that, as  → 1 from
below, the model flows to the origin g(∞) = g3 (∞) = 0 in Fig. 6.6. As discussed
in Sect. 6.2.8.1, g(∞) = g3 (∞) = 0 with K (∞) = 1/2 is an SU (2) invariant point,
consistently with the fact that at  = 1 the Hamiltonian (6.99) describes an isotropic
antiferromagnetic Heisenberg model.

6.4 The One-Dimensional Hubbard Model

We now consider the spinful Hubbard model

  
H = −t ci†σ ci+1σ + ci+1
†
σ iσ + U
c ni ↑ ni ↓ , (6.105)
iσ i

with U > 0, as representative of generic models of electrons with short-range inter-

action and two Fermi points, ±k F , and apply the same bosonization scheme as above.
For that, we need to introduce two copies of right and moving electrons, one for each
spin, thus φ R↑ (x), φ R↓ (x), φ L↑ (x) and φ L↓ (x), as well as the fields ↑ (x), ↑ (x),
↓ (x) and ↓ (x). Each of them have the same properties of φ R (x), φ L (x), (x)
306 6 Brief Introduction to Luttinger Liquids

and (x) previously discussed. In addition, in order to make different Fermi fields
anticommute, we assume that
     
φ R↑ (x) , φ L↑ (y) = φ R↑ (x) , φ R↓ (y) = φ R↑ (x) , φ L↓ (y) = −iπ ,
   
φ L↑ (x) , φ R↓ (y) = φ L↑ (x) , φ L↓ (y) = −iπ ,
 
φ R↓ (x) , φ L↓ (y) = −iπ .
(6.106)
We further introduce the combinations
1  
φ R(L)ρ (x) = √ φ R(L)↑ (x) + φ R(L)↓ (x) ,
2
1   (6.107)
φ R(L)σ (x) = √ φ R(L)↑ (x) − φ R(L)↓ (x) ,
2
referring to charge, ρ, and spin, σ, degrees of freedom, as well as the fields
1  
ρ(σ) (x) = √ φ Rρ(σ) (x) + φ Lρ(σ) (x) ,
4π
1   (6.108)
ρ(σ) (x) = √ φ Lρ(σ) (x) − φ Rρ(σ) (x) ,
4π
or, equivalently,
1   1  
ρ (x) = √ ↑ (x−)55 + 5↓ (x) , ρ (x) = √ ↑ (x) + ↓ (x) ,
2 2
1   1  
σ (x) = √ ↑ (x) − ↓ (x) , σ (x) = √ ↑ (x) − ↓ (x) .
2 2
We note that the total electron number
  
1 ∂φa (x) 2 ∂ρ (x)
N= dx = dx , (6.109)
2π ∂x π ∂x
a=R↑,L↑,R↓,L↓

so that the non-interacting Hamiltonian, assuming

N kF
N R↑ = N L↑ = N R↓ = N L↓ = = L,
4 2π
is simply
  2 2
vF ∂ρ (x) ∂σ (x)
H0 = dx ρ (x)
2
+ + σ (x)
2
+
2 ∂x ∂x

2 ∂ρ (x)
−μ dx ,
π ∂x

where μ must be fixed by imposing the desired electron number.

6.4 The One-Dimensional Hubbard Model 307

In order to bosonize the interaction, we follow a simpler approach than we did

before for spinless fermions. In the long wavelength limit the interaction can be
written as
 
U n i↑ n i↓ → U d x ρ↑ (x) ρ↓ (x) ,
i
where
† †
ρ↑ (x) = ρ R↑ (x) + ρ L↑ (x) +  R↑ (x)  L↑ (x) +  L↑ (x)  R↑ (x)
   
1 ∂φ R↑ (x) 1 ∂φ L↑ (x) i i
= + + e−i φ R↑ (x)+φ L↑ (x) − ei φ R↑ (x)+φ L↑ (x)
2π ∂x 2π ∂x 2πα 2πα
1 ∂↑ (x) i √ i √
−i 4π ↑ (x) i 4π ↑ (x)
= √ + e − e
π ∂x 2πα 2πα
1 ∂↑ (x) 1 √
≡ √ + sin 4π ↑ (x) ,
π ∂x πα
(6.110)
and similarly for ρ↓ (x). It follows that
  
1 ∂↑ (x) ∂↓ (x)
U d x ρ↑ (x) ρ↓ (x)  U dx
π ∂x ∂x

1 √ √
+ 2 2 sin 4π ↑ (x) sin 4π ↓ (x)
π α
  2 2
1 ∂ρ (x) 1 ∂σ (x)
=U dx −
2π ∂x 2π ∂x

1 √ 1 √
+ cos 8π σ (x) − cos 8π ρ (x)
2π 2 α2 2π 2 α2
  2
gρ ∂ρ (x) gσ ∂σ (x) 2
≡ dx +
2π ∂x 2π ∂x

g3σ √ g3ρ √
− cos 8π σ (x) − cos 8π ρ (x) ,
2π 2 α2 2π 2 α2

where gρ = U , gσ = −U , g3ρ = U and g3σ = −U . Therefore, the fully interacting

Hamiltonian can be written as

H = Hρ + Hσ ,

where the charge component is

  2 
vF ∂ρ (x) 2 ∂ρ (x)
Hρ = dx ρ (x)
2
+ −μ dx
2 ∂x π ∂x
   (6.111)
gρ ∂ρ (x) 2
g3ρ √
+ dx − cos 8π ρ (x) ,
2π ∂x 2π 2 α2
308 6 Brief Introduction to Luttinger Liquids

with
gρ
μ = vF 1 +
kF ,
π vF
so as to have the desired electron number, while the spin component reads
  2
vF ∂σ (x)
Hσ = dx σ (x) +
2
2 ∂x
   (6.112)
gσ ∂σ (x) 2
g3σ √
+ dx − cos 8π σ (x) .
2π ∂x 2π 2 α2

We note that both Hρ and Hσ look like the Hamiltonian of interacting spinless
fermions around K = 1/2 that we studied in Sect. 6.2.8. This observation allows us
to draw several conclusions.
First, Hσ , with gσ = g3σ , refers to a model endowed with an SU (2) symmetry,
as discussed in Sects. 6.2.8.1 and 6.3.
In addition, since gσ < 0, Hσ lies on the black line in the upper half-plane of
Fig. 6.6. Therefore, under RG it will flow to gσ (∞) = g3σ (∞) = 0, thus to an
effective non-interacting spinless fermion model with K σ (∞) = 1.
Let us now discuss more closely the charge Hamiltonian (6.111). The chemical
potential term can be eliminated by the canonical transformation

2
ρ (x) → ρ (x) + kF x , ρ (x) → ρ (x) ,
π
after which
  2
vF ∂ρ (x)
Hρ = dx ρ (x) +
2
2 ∂x
   
gρ ∂ρ (x) 2
g3ρ √
+ dx − cos 4k F x + 8π ρ (x) ,
2π ∂x 2π 2 α2
(6.113)
If 4k F = 2π, i.e., away from half-filling, the umklapp oscillates fast and thus can
be neglected in the asymptotic long-wavelength limit. In that case, Hρ describes a
gapless system with

1 vρ
Kρ = , vρ = v F 1+ . (6.114)
1 + gρ /πv F πv F

On the contrary, at half-filling 4k F = 2π, the umklapp cannot be neglected. Since

gρ > 0 and g3ρ > 0, the system will flow to the strong-coupling fixed point gρ (∞) =
g3ρ (∞) = ∞. This implies that we must fix ρ (x) to the value that minimises the
umklapp, thus
√ √
8π ρ (x) = 4π ↑ (x) + ↓ (x) = 2πn .
6.4 The One-Dimensional Hubbard Model 309

In this case there is only one independent value n = 0, which implies that ↑ (x) =
− ↓ (x) and therefore

1 √ √
ρ(x) = 1 + sin 4π ↑ (x) + sin 4π ↓ (x) = 1 . (6.115)
πα
This phase thus corresponds to a charge insulator, the electrons localised one at
each site, with gapless spin excitations, which behaves asymptotically just alike
the isotropic antiferromagnetic Heisenberg model of Sect. 6.3. As we know, the
two models exactly maps onto each other at large U t, hence in that limit the
correspondence holds not only asymptotically at low energy and long distances.
If N /L = 1 + δ, thus 4k F = 2π + 2π δ ≡ 2π + 2k F , with δ 1, the model
asymptotically approaches a doped Luther-Emery model, see Sect. 6.2.8.2, at
K ρ = 1/2 with Hamiltonian

   2  
v∗ ∂ρ (x) g∗  √ 
Hρ  dx ρ (x)
2
+ − d x cos 2k F x + 4π ρ (x) ,
2 ∂x 2π 2 α2

which describes an insulator with a doping of δ > 0 electrons in the conduction band
or −δ > 0 holes in the valence one.

6.4.1 Luttinger Versus Fermi Liquids

Away from half-filling, the one-dimensional Hubbard model has gapless charge and
spin modes that behave as acoustic waves with different velocities vρ > v F > vσ >
0, and with K ρ < 1 and K σ = 1. The single-particle Green’s function can be readily
calculated as we previously did for spinless fermions,

1 eik R F x
G R (x, t) = % %
2π x − vρ t + iα sign(t) x − vσ t + iα sign(t)
 β 2
α2
   ,
x − vρ t + iα sign(t) x + vρ t − iα sign(t)
1 e−ik L F x
G L (x, t) = % %
2π −x − vρ t + iα sign(t) −x − vσ t + iα sign(t)
 β 2
α2
   ,
x − vρ t + iα sign(t) x + vρ t − iα sign(t)
(6.116)
where now
1 % 1 2
β2 = Kρ − % .
8 Kρ
310 6 Brief Introduction to Luttinger Liquids

These expressions clearly highlight the spin-charge separation; the electron decou-
ples into independent charge and spin components that move with different velocities.
In addition, the Green’s functions at t = 0− are non analytic on both sides of the
complex z = x + i y plane, and thus the momentum distribution does not jump at the
Fermi momenta.
Although the Fourier transform G R(L) ( + iη, k) of those heavily non-analytic
Green’s functions is rather cumbersome and does not allow easily extracting the
self-energies  R(L) ( + iη, k), spin-charge separation makes hardly conceivable that
filtering G R(L) ( + iη, k) by

∂ R(L) ( + iη, k)
Z R(L) (, k)−1 = 1 − , (6.117)
∂
may yield well defined quasiparticles. Therefore, the Landau-Fermi liquid theory dis-
cussed in Chap. 5 cannot be applied in one dimensional interacting electron systems,
all the more since all correlation functions at momentum transferred ∼ 2k F decay
with anomalous interaction-dependent exponents, which, as discussed in Sect. 6.1,
signals the unavoidable quasi-long range order that characterises interacting elec-
trons in one-dimension.
Nonetheless, through ρ(σ) (x) and ρ(σ) (x) we can define new Fermi fields
 Rρ(σ) (x) and  Lρ(σ) (x) whose low-energy long-wavelength dynamics is deter-
mined by free Hamiltonians. Those fields thus play the role of the ‘quasiparticles’ in
Luttinger liquids. As a consequence, the specific heat receives linear in temperature
contributions from both charge and spin fermions, each having its own density of
states 1/2πvρ(σ) . Moreover, the charge, χρ , and spin, χσ , density-density response
function at long wavelengths also look like those of non-interacting electrons, specif-
ically,

|q| 1 1
χρ(σ) (q, ω) = 2K ρ(σ) − ,
π ω − vρ(σ) |q| + iη ω + vρ(σ) |q| + iη

yielding charge compressibility

2K ρ vF
κ= = Kρ κ0 < κ0 ,
2πvρ vρ

and spin susceptibility, recalling that K σ = 1,

2 vF
χ= = κ0 > κ0 .
2πvσ vσ

All those long-wavelength properties, as well as their temperature dependence,

behave as in ordinary Fermi liquids, with K ρ , vρ and vσ playing the role of Lan-
dau’s parameters.
Therefore, Luttinger and Fermi Liquids, in spite of their differences, share a
very unique feature: both of them recovers at low energy the larger symmetry of
Problems 311

the non-interacting system. However, there is an important caveat. Landau’s Fermi

liquids are asymptotically described by quasiparticles whose Hamiltonian conserves
their occupation numbers in momentum space, n k↑ and n k↓ , separately for each
momentum k on the quasiparticle Fermi surface and for each spin. On the contrary,
‘quasiparticles’ in Luttinger liquids are spin-charge separated. Therefore, only the
charge n k↑ + n k↓ and the spin n k↑ − n k↓ are separately conserved at each Fermi
point, k R F and −k L F , while each spin component is not.

Problems

6.1 Spin-charge separation—Consider the following one-dimensional Hamilto-

nian:
     
H0 = v F k − k F c†Rkσ c Rkσ − k + k F c†Lkσ c Lkσ
kσ
g4      (6.118)
+ ρ p↑ (q) + ρ p↓ (q) ρ p↑ (−q) + ρ p↓ (−q) ,
2L q
p=R,L

which contains only a g4 charge scattering process, see Fig. 6.4, and where the
dispersion (k) of Fig. 6.2 has been linearised around the two Fermi points ±k F .

• Calculate the single-particle Green’s function.

Now assume to have two chains, each described by the Hamiltonian (6.118), so that

       
H0 = vF k − k F cn† Rkσ cn Rkσ − k + k F cn† Lkσ cn Lkσ
n=1,2 kσ
g4     
+ ρnp↑ (q) + ρnp↓ (q) ρnp↑ (−q) + ρnp↓ (−q) ,
2L q p=R,L
(6.119)
where n = 1, 2 is the chain label. The two chains are mutually coupled by an inter-
chain hopping
    
H⊥ = −t d x 1†pσ (x) 2 pσ (x) + 2†pσ (x) 1 pσ (x) ,
σ p=R,L
(6.120)
†
where 1R(L)σ (x) creates a right(left) moving fermion with spin σ on chain 1, while
†
2R(L)σ (x) does the same but on chain 2.

• Rewrite the interchain hopping (6.120) using the bosonized expressions of right
and left moving fields.
312 6 Brief Introduction to Luttinger Liquids

• Introduce the charge (c) and spin (s) symmetric (S) and antisymmetric (A) com-
binations of the Bose fields φnpσ (x), n = 1, 2, p = R, L and σ = ↑, ↓:

1 
φc S p (x) = φ1 p↑ (x) + φ2 p↑ (x) + φ1 p↓ (x) + φ2 p↓ (x) ,
2
1 
φs S p (x) = φ1 p↑ (x) + φ2 p↑ (x) − φ1 p↓ (x) − φ2 p↓ (x) ,
2
1  (6.121)
φc A p (x) = φ1 p↑ (x) − φ2 p↑ (x) + φ1 p↓ (x) − φ2 p↓ (x) ,
2
1 
φs A p (x) = φ1 p↑ (x) − φ2 p↑ (x) − φ1 p↓ (x) + φ2 p↓ (x) ,
2
and through them the new Fermi fields

1 1
c S R (x) = √ eiφc S R (x) , s S R (x) = √ eiφs S R (x) ,
2π α 2π α
1 iφc A R (x) 1
c A R (x) = √ e , s A R (x) = √ eiφc A R (x) ,
2π α 2π α
1 −iφc S L (x) 1
c S L (x) = √ e , s S L (x) = √ e−iφs S L (x) ,
2π α 2π α
1 1
c A L (x) = √ e−iφc A L (x) , s A L (x) = √ e−iφc A L (x) .
2π α 2π α

Rewrite the unperturbed Hamiltonian H0 (6.119) plus the interchain hopping

(6.120) in terms of those Fermi fields, similarly to Sect. 6.2.8.2 but here for generic
k F , and show that H = H0 + H⊥ can be easily diagonalised in this representation.
• Calculate the intrachain and interchain Green’s functions.

References
1. A. Gogolin, A. Nersesyan, A. Tsvelik, Bosonization and Strongly Correlated Systems (Cambridge
University Press, 1998)
2. T. Giamarchi, Quantum Physics in One Dimension (Oxford University Press, 2004)
Kondo Effect and the Physics of the
Anderson Impurity Model 7

The behaviour of magnetic impurities (Fe, Mn, Cr) diluted into non-magnetic metals,
e.g., Cu, Ag, Au, Al, is the simplest manifestation of strong electron-electron corre-
lations that escapes a description in terms of independent particles. Moreover, that
same phenomenology has progressively become the paradigm of strongly correlated
metals close to a Mott transition as unveiled by Dynamical Mean Field Theory.
We know that the behaviour of a good metal with a large Fermi temperature, TF ∼
104 K, is dominated at low temperature T  TF by Pauli principle. For instance, the
magnetic susceptibility is roughly constant, χ ∼ 1/TF , and one should in principle
heat the sample to very high temperatures T  TF , where the metal likely melts, to
release the spin entropy and recover a Curie-Weiss behaviour χ ∼ 1/T . Moreover,
the resistivity is an increasing function of temperature, since the channels that may
dissipate current, the coupling to phonons and the electron decay into particle-hole
excitations brought by interaction, become available only upon heating.
At odds with that expectation, if one introduces a very diluted (few parts per mil-
lion) concentration of magnetic impurities the above behaviour changes drastically.
We just mention three distinct features.

(1) The magnetic susceptibility shows a Curie-Weiss behaviour well below TF , pro-
portional to the impurity concentration n i and roughly with the same g-factor
of the isolated magnetic impurity, apart from corrections due to the crystal field.
Around a very low temperature, called Kondo-temperature TK , the Curie-Weiss
behaviour turns into a logarithmic one, and finally the susceptibility saturates at
low temperature to a value χ (0) ∼ n i /TK , with χ (T ) − χ (0) ∼ −T 2 .
(2) The resistivity R(T ) displays a minimum around TK , followed at T < TK by
a logarithmic increase. At very low temperatures, R(T ) approaches a constant
value R(0) ∝ n i with R(T ) − R(0) ∼ −T 2 . The value of the residual resistivity
R(0) suggests very strong scattering potential, near the so-called unitary limit.

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 313
M. Fabrizio, A Course in Quantum Many-Body Theory, Graduate Texts in Physics,
https://doi.org/10.1007/978-3-031-16305-0_7
314 7 Kondo Effect and the Physics of the Anderson Impurity Model

(3) The entropy which is released above the Kondo temperature, which can be
extracted from the specific heat, includes the spin degrees of freedom of the
isolated impurity but not the charge ones, which indicates that the magnetic
impurities behave as local moments above TK .

Explaining this behaviour amounts actually to understand three different prob-

lems.

• The first concerns the region TK  T  TF and can be formulated as follows:

How is it possible to sustain a local moment inside a metal? We shall see that to
answer this question one is obliged to abandon conventional independent particle
schemes.
• The second issue regards what happens around the Kondo temperature, namely
to understand the logarithmic crossover.
• Finally, the last question concerns the low temperature behaviour, the way local
moments get screened and why resistivity is a decreasing function of temperature.

The last two questions turn out to represent a complicated many-body problem which
we shall here discuss using the tools we have presented so far.

7.1 Brief Introduction to Scattering Theory

We start by showing why the existence of local moments in a metal for T  TF is

so puzzling.
Let us consider an impurity imbedded in a wide-band metal. We only consider
the valence band, which can be safely described by a non-interacting Hamiltonian
 †
H0 = k ckσ ckσ . (7.1)
k,σ

The scattering potential provided by the impurity has the general form
 †
V = Vkp ckσ cpσ , (7.2)
σ k,p

where Vkp are the matrix elements of the impurity potential V̂ between the valence
band Block waves. The single-particle Green’s function in complex frequency for
the full Hamiltonian H = H0 + V can be formally written as the matrix

1
Ĝ(z) = , (7.3)
z − Ĥ
7.1 Brief Introduction to Scattering Theory 315

where Ĥ = Ĥ0 + V̂ , with Ĥ0 diagonal with elements k . Similarly, the unperturbed
Green’s function reads
1
Ĝ 0 (z) = . (7.4)
z − Ĥ0
The above operators have the following interpretation. If we consider for instance
the inverse of the Green’s function Ĝ, namely

Ĝ(z)−1 = z − Ĥ ,

in the basis set of the valence band Block waves, this is the matrix
   
Ĝ(z)−1 
= δσ σ  δkp z − δkp k − Vkp , (7.5)
kσ,pσ

hence the Green’s function is the inverse of the above matrix. Suppose we have
diagonalised the full Hamiltonian and obtained the eigenvalues a . In the diagonal
basis the Green’s function has matrix elements
1
G(z)ab = δab .
z − a

Therefore, we readily find that for η > 0 infinitesimal

1   
− Im Tr Ĝ(z = ω + iη) = δ(ω − a ) ≡ ρ(ω),
π aσ

where ω is a real frequency and ρ(ω) is the density of states (DOS). Since the trace
is invariant under unitary transformations, hence also under the transformation that
diagonalises the Hamiltonian, it is generally true that

1  
− Im Tr Ĝ(z = ω + iη) = ρ(ω). (7.6)
π
Similarly,

1   
− Im Tr Ĝ 0 (z = ω + iδ) = ρ0 (ω) = δ(k − ω), (7.7)
π
kσ

gives the DOS of the host metal.

Let us formally write the full Green’s function as

1 1 1
Ĝ = = = Ĝ 0
z − Ĥ0 − V̂ Ĝ −1
− V̂ ˆI − Ĝ 0 V̂
0 (7.8)
 −1
= Ĝ 0 + Ĝ 0 V̂ Iˆ − Ĝ 0 V̂ Ĝ 0 ≡ Ĝ 0 + Ĝ 0 T̂ Ĝ 0 ,
316 7 Kondo Effect and the Physics of the Anderson Impurity Model

which provides the definition of the so-called T -matrix, namely

 −1
T̂ (z) = V̂ Iˆ − Ĝ 0 V̂ . (7.9)

Hereafter, we shall assume z = ω + iη with η > 0 infinitesimal. We note that

∂
ln Ĝ(z) = −Ĝ(z),
∂z

so that
1 ∂   
ρ(ω) = Im Tr ln Ĝ(z) . (7.10)
π ∂z
We are interested in the variation of the DOS induced by the impurity, which, by
making use of (7.8) and (7.9), is given by

1 ∂   
ρ(ω) = ρ(ω) − ρ0 (ω) = Im Tr ln Ĝ(z) Ĝ 0 (z)−1
π ∂z
1 ∂ 1 1 ∂  
= Im Tr ln = Im Tr ln V̂ −1 T̂ (z) .
π ∂z 1̂ − Ĝ 0 (z) V̂ π ∂z
(7.11)
We define the matrix of the scattering phase shifts
 
δ̂(z) = Im ln V̂ −1 T̂ (z) = Arg V̂ −1 T̂ (z) , (7.12)

through which
1 ∂  
ρ(ω) = Tr δ̂(ω) . (7.13)
π ∂ω
We note that for |z| → ∞, Ĝ 0 (z) → 1/z → 0, hence T̂ (z) → V̂ and δ̂(z) → 0.
The variation of the total number of electrons, Nels , induced by the impurity at
fixed chemical potential μ is therefore
 μ 1  
Nels = dω ρ(ω) = Tr δ̂(μ) , (7.14)
−∞ π

the so-called Friedel sum rule.

Let us go back to the T -matrix defined in (7.9). One readily finds its inverse
 
T̂ (z)−1 = V̂ −1 − Ĝ 0 (z) .

Since the Hamiltonian is hermitean, then

 †  
T̂ (z)−1 = V̂ −1 − Ĝ 0 (z ∗ ) ,
7.1 Brief Introduction to Scattering Theory 317

so that
 †  
T̂ (z)−1 − T̂ (z)−1 = Ĝ 0 (z) − Ĝ 0 (z ∗ ) .

Therefore
 †  
T̂ (z)† T̂ (z)−1 − T̂ (z)−1 T̂ (z) = T̂ (z) − T̂ (z)† = T̂ (z)† Ĝ 0 (z) − Ĝ 0 (z ∗ ) T̂ (z).
(7.15)
Since
 
Ĝ 0 (z) − Ĝ 0 (z ∗ ) = −2πi δ ω − Ĥ0 ,

Equation (7.15) implies the following identity

†
 †
Tkp (ω) − Tkp (ω) = −2πi Tkq (ω) δ(ω − q ) Tqp (ω), (7.16)
q

which is the so-called optical theorem. It also shows that the imaginary part of the
T -matrix is finite only within the conduction band.
Let us now analyse the on-shell T -matrix Tkp (), where k = p = . We can
rewrite the on-shell optical theorem as follows
  †
   †
−2πi δ k − p Tkp () − Tkp () = (−2πi)2 δ k − p Tkq () δ(k − q ) Tqp ().
q

The above equation implies that, if we introduce the so-called on-shell S-matrix
through
 
Skp () = δkp − 2πi δ k − p Tkp (), (7.17)

where, as before, k = p = , then it follows that the S-matrix is unitary, i.e., Ŝ Ŝ † =

Iˆ. Skp () is the transition probability that an electron in state k scatters elastically
into state p. Since it is unitary, it follows that only elastic scattering survives.
Let us consider the simpler case of a spherical Fermi surface, i.e. k = k depend-
ing only on the modulous of the wavevector. In addition we assume that the matrix
elements Vkp only depend on the angle between the two wavevectors, θkp , hence can
be expanded in Legendre polynomials
  
Vkp = Vl (2l + 1) Pl cos θkp , (7.18)
l

as well as the T -matrix

  
Tkp (z) = tl (z) (2l + 1) Pl cos θkp . (7.19)
l
318 7 Kondo Effect and the Physics of the Anderson Impurity Model

Since

d q     1  
Pl cos θkq Pl  cos θqp = δll  Pl cos θkp ,
4π 2l + 1
the optical theorem transforms into
 2
tl (ω) − tl∗ (ω) = 2i Im tl (ω) = −2πi ρ0 (ω) tl (ω) , (7.20)

where ρ0 (ω) is the bare conduction electron density of states per spin. Since
 
tl (ω) = tl (ω) eiδl (ω) ,

it follows that
1
tl (ω) = − sin δl (ω) eiδl (ω) . (7.21)
πρ0 (ω)
If the concentration of impurities is very low, then the scattering rate suffered by the
electrons is given by the Fermi-golden rule with the T -matrix playing the role of the
operator driving electron scattering, namely

1  †  
= 2π n i Tkq (ω) δ(ω − q ) Tqk (ω) 1 − cos θkq ,
τk q

where n i is the impurity concentration and the last term is a geometric factor which
guarantees that forward scattering, θ = 0, does not contribute to the current dissipa-
tion. Since
(2l + 1) z Pl (z) = (l + 1) Pl+1 (z) + l Pl−1 (z),
we obtain
 
1 d q
= 2π ρ(ω) tl∗ (ω) tl  (ω) (2l + 1) Pl (cos θkq )
n i τk (ω) 4π
ll 
  
(2l + 1) Pl  (cos θkq ) + (l  + 1) Pl  +1 (cos θkq ) + l  Pl  −1 (cos θkq )
  2 
= 2π ρ(ω) (2l + 1) tl (ω) + l tl∗ (ω) tl−1 (ω) + (l + 1) tl∗ (ω) tl+1 (ω) ,
l
(7.22)
which as expected gives a scattering time τk (ω) ≡ τ (ω) independent of momentum,
whose value at zero energy determines the residual resistivity

1
R(T = 0) ∝ . (7.23)
τ (0)
7.1 Brief Introduction to Scattering Theory 319

7.1.1 General Analysis of the Phase-Shifts

Let us keep assuming a spherically symmetric case. In addition we assume that the
scattering components Vl are non-zero only for a well defined l = L. For instance,
in the case of magnetic impurities with partially filled d-shell, L = 2. Then the only
non-zero phase shift is
Im t L (ω)
δ L (ω) = tan−1 .
Re t L (ω)
An important role is played by the frequencies at which the real part vanishes. The
first possibility is that occurs outside the conduction band, either below or above.
In this case we already know that the imaginary part is zero. This implies that the
phase-shift jumps by π at these energies so that the variation of the DOS is δ-like. In
this case, one speaks of bound states that appear outside the conduction band. Clearly
this possibility can not explain a Curie-Weiss behaviour for T  TF . Indeed, at very
low temperature the bound state is either doubly occupied, if below the conduction
band, or empty, if above, and one needs a temperature larger than the conduction
bandwidth, hence larger that TF , to release its spin entropy.
The other possibility is that the real part vanishes at a frequency ω∗ within the
conduction band, where the imaginary part is therefore finite. Around ω∗ , assuming
a real part linearly vanishing and an imaginary part roughly constant, we get


δ L (ω) tan−1 ,
ω − ω∗

leading to a Lorentzian DOS variation yielded by the impurity

1 
ρ(ω) , (7.24)
π (ω − ω∗ )2 +  2

which is called a resonance. Clearly, in other for this resonance to contribute at T 

TF , ω∗ should be very close to the chemical potential, in particular |ω∗ − μ| ≤ TK ,
and, in addition,  TK .
Let us therefore assume  = TK and, for simplicity, ω∗ = 0. The contribution of
the resonant state to the magnetic susceptibility is then given by

∂ f () 1 TK
χ (T ) = −μ B g (2L + 1) d , (7.25)
∂ π  2 + TK2

where f () is the Fermi distribution function. We readily find that for T  TK ,

2L + 1
χ μB g ,
T
while for T  TK
2L + 1
χ μB g ,
π TK
320 7 Kondo Effect and the Physics of the Anderson Impurity Model

consistent with the observed behaviour. In addition, the resistivity would be given
through (7.22) and (7.23) by

1 2(2L + 1) 2(2L + 1)
R(0) ∝ = ni sin2 δ L (0) = n i , (7.26)
τ (0) πρ0 πρ0

where ρ0 = ρ0 (0), again compatible with the almost unitary limit, δ(0) = π/2,
observed experimentally. Therefore, the existence of a narrow resonance near the
chemical potential with width exactly given by TK seems the natural explanation to
what experiments find.
However, this conclusion is rather strange, since the Kondo behaviour is observed
in many different host metals and for different magnetic impurities, hence it would
be really surprising that in all those cases a resonance appears and it is always pinned
near the chemical potential.
Moreover, this simple single-particle scenario fails to explain the entropy released
above TK . Indeed, for T  TK , the resonance is effectively like an isolated level
which can be empty, singly occupied with a spin up or down, or doubly occupied.
Therefore its entropy per impurity should be Simp = 2(2L + 1) ln 2. On the other
hand the experiments tells us that only the spin degrees of freedom are released
above TK , which amounts in the above simplified model to an entropy Simp = ln S,
where S is the spin of the isolated atom. In other words, a resonance at the chemical
potential and at temperatures higher than its width is not at all the same as a local
moment, since the former does have valence fluctuations which are absent in the
latter. Therefore, although the resonance scenario is suggestive, it is not at all the
solution to the puzzle.

7.2 The Anderson Impurity Model

Something which is obviously missing in the above analysis are the electron-electron
correlations at the magnetic ions. A narrow resonance induced by an impurity is quite
localised around it, and is essentially akin an atomic level slightly broadened by the
hybridizsation with the conduction electrons of the host metal. Therefore, like an
atomic level, such resonance can accomodate only a fixed number of electrons, say
N , paying a finite amount of energy by adding or removing electrons with respect
to the reference valency N . As usual, it is convenient to define a so-called Hubbard
repulsion U through

U = E(N + 1) + E(N − 1) − 2E(N ), (7.27)

where E(M) is the total energy when the resonance is forced to have M electrons, N
being the equilibrium value. In principle one might include this additional ingredient
by adding to the Hamiltonian a term

U  2
n̂ − N , (7.28)
2
7.2 The Anderson Impurity Model 321

with n̂ the occupation number of the resonance. If it were possible to define such an
operator, then (7.28) would actually solve one of the puzzles. Indeed, if U  TK and
for temperatures TK  T  U , valence fluctuations with respect to the equilibrium
value N would be suppressed and the only degrees of freedom contributing to the
entropy would be those related to the degeneracy of the N -electron configurations at
the resonance. Assuming the angular momentum quenched by the crystal field, those
residual degrees of freedom are just the spin ones, now consistent with experiments.
Unfortunately the resonance occupation number operator is an ill defined object.
To overcome such difficulty, Anderson had the idea to represent the resonance as
an additional level inside the conduction band, even if that would be, rigorously
speaking, an overcomplete basis. This leads to the so-called Anderson Impurity
Model (AIM) [1] defined, in the simplest case of a single-orbital impurity, as
    U 2
† †
H= k ckσ ckσ + Vk ckσ dσ + dσ† ckσ + n − 1 + d n.
2
kσ kσ
(7.29)
This model describes a band of conduction electrons with energy dispersion k ,
hybridised with a single level, dσ , with energy d , both k and d being measured
with respect to the chemical potential, and n = n ↑ + n ↓ with n σ = dσ† dσ . Because
of U > 0, this level tends to accomodate no more no less than a single electron.
Before switching the hybridisation Vk and the interaction U , the conduction elec-
tron Green’s function in complex frequency z is

1
G (0) (z, k) = , (7.30)
z − k
and is represented as a dashed tiny line in Fig. 7.1, while the impurity one, solid tiny
line in Fig. 7.1, is
1
G (0) (z) = . (7.31)
z − d
The impurity Green’s function at finite hybridisation, the solid black line in Fig. 7.1,
can be readily calculated through Dyson’s equation,

1
G (z) = , (7.32)
z − d − (z)
where the so-called hybridisation function is defined through

 Vk2
(z) = . (7.33)
z − k
k

A finite U yields an impurity self-energy (z), so that the fully interacting impurity
Green’s function G (z), blue bold line in Fig. 7.1, reads read

1 1
G (z) = = . (7.34)
G0 (z)−1 − (z) z − d − (z) − (z)
322 7 Kondo Effect and the Physics of the Anderson Impurity Model

Correspondingly, the interacting conduction electron Green’s function, G(z, k, k ),

not anymore diagonal in momentum due to the impurity breaking translational sym-
metry, and the mixed conduction electron-impurity Green’s function, G(z, k), are,
see Fig. 7.1

G(z, k, k ) = δk,k G (0) (z, k) + G (0) (z, k) Vk G (z) Vk G (0) (z, k )
≡ δk,k G (0) (z, k) + G (0) (z, k) Tkk (z) G (0) (z, k ) , (7.35)
(0)
G(z, k) = G (z, k) Vk G (z) .

The hybridisation function has a branch cut on the real axis,

  
(z →  ± i0+ ) = ∓ iπ Vk2 δ  − k ≡ ∓ i(),
k

with () > 0, so that


d ()
(z) = .
π z−
Since the physics of interest occurs on energy scales  TF , we can safely assume that
() =  for || < W and () = 0 otherwise, with W the conduction bandwidth.
In other words, we assume that both conduction electron DOS, ρ0 (), and () are
particle-hole symmetric, i.e., are even function of , which is measured with respect
to the chemical potential. Therefore, if |z|  W ,
 
(z) −i sgn Im z , (7.36)

and thus
1
G (z) =   . (7.37)
z − d − (z) + i sgn Im z

7.2.1 Non Interacting Impurity

If U = 0, thus (z) = 0, the impurity Green’s function (7.37) is simply

1
G0 (z) =   , (7.38)
z − d + i sgn Im z

from which the impurity DOS readily follows

1 1 
A0 () = − Im G0 ( + i0+ ) =  
π π  − d 2 +  2
(7.39)
1 ∂ − 1 ∂ Im ln G0 ( + i0+ )
= tan−1 = ,
π ∂  − d π ∂
7.2 The Anderson Impurity Model 323

k
G (0)(z, k) = (0)
(z) = Vk = =

0(z) = = +

(z) = = +

k k k′
G(z, k, k′) = δk,k′ +

G(z, k) = =

Fig. 7.1 Green’s functions of the Hamiltonian (7.29). G (0) (z, k) (dashed line) and G (0) (z) (solid
line) are, respectively, the Green’s functions of conduction electrons and impurity in the absence of
hybridisation, the black dot vertex with dashed and solid external legs, and for U = 0. G (z) is the
impurity Green’s function that includes the hybridisation but still at U = 0, which is obtained by
the shown Dyson equation. A finite U yields an impurity self-energy (z), in red, that defines the
fully interacting impurity Green’s function G (z), blue bold line, through Dyson’s equation. In turn,
G (z) determines the interacting electron Green’s function G(z, k, k ), not anymore diagonal in k,
and the mixed impurity-conduction electron Green’s function G(z, k)
 
which is just a resonance within the conduction band for d  < W . Therefore, at U =
0, the Anderson impurity model (7.29) does yield the desired physical behaviour.
Correspondingly, the non interacting conduction electron and mixed Green’s func-
tions are

G 0 (z, k, k ) = δk,k G (0) (z, k) + G (0) (z, k) Vk G0 (z) Vk G (0) (z, k ) ,
G 0 (z, k) = G (0) (z, k) Vk G0 (z) .

The variation of electron number at fixed chemical is therefore

  + 
N= T ei0 G0 (i) + G 0 (i, k, k)
σ  k
 +  Vk2  + ∂ (i)
= 2T ei0 G0 (i) 1 +  2 = 2T ei0 G0 (i) 1 − (7.40)
 i −  k 
∂i
k
  + 
+ ∂ ln G0 (i) d ∂Im ln G0 ( + i0 )
= −2T ei0 =2 f () = 2 d f () A0 () ,

∂i π ∂

and is just the expectation value  n  of the number of electrons n occupying the
impurity. That is a consequence of our assumption (7.36). More generally, N is
equal to  n  plus corrections of order /W , which we assume vanishingly small.
At zero temperature, (7.40) becomes

d ∂ 0 − 2 −
N =  n  =  n↑ + n↓  = 2 tan−1 = tan−1 +π
π ∂ −∞  − d π −d
(7.41)
2  2 d
= tan−1 =1− tan−1 .
π d π 
324 7 Kondo Effect and the Physics of the Anderson Impurity Model

If a magnetic field is present that splits d↑ from d↓ , then

1 1 dσ
 nσ  = − tan−1 . (7.42)
2 π 

7.2.2 Hartree-Fock Approximation

When U is finite, the Hamiltonian (7.29) includes the interaction term

U U
(n − 1)2 = U n ↑ n ↓ − n .
2 2
The second term renormalises d → d − U /2. Let us treat the interaction U n ↑ n ↓
within the Hartree-Fock approximation [1], allowing for spontaneous spin-
polarisation of the impurity, thus G↑ (z) = G↓ (z). The Hartree-Fock self-energy in
Matsubara frequencies and for spin-σ impurity electrons contains in our case only
the Hartree term
 +
σ (i) = U T ei0 G−σ (i) = U  n −σ  , (7.43)


where  n σ  must be determined self-consistently. Therefore the impurity Green’s

function is
1 1
Gσ ( + i0+ ) = ≡ , (7.44)
 − d + U /2 − U  n −σ  + i   − dσ + i 

from which it readily follows through (7.42) that

1 1 dσ 1 1 d − U /2 + U  n −σ 
 nσ  = − tan−1 = − tan−1 .
2 π  2 π 
(7.45)
We define  n ↑  = (n + m)/2 and  n ↓  = (n − m)/2, with n ∈ [0, 2] and m =
[−1, 1], so that the self-consistency conditions are
1 2d − U + U n − U m 1 2d − U + U n + U m
n =1− tan−1 − tan−1 ,
π 2 π 2
1 2d − U + U n + U m 1 2d − U + U n − U m
m= tan−1 − tan−1 ≡ F(m) .
π 2 π 2
(7.46)
In order to assess under which conditions a spontaneous magnetisation emerges, i.e.,
a solution with m = 0 exists, we note that the function F(m) on the right hand side of
the second equation increases monotonically in m, and, in addition, F(0) = 0 while
F(±∞) = ±1. Therefore a solution m = ±m ∗ , with m ∗ > 0, exists if and only if
F  (0) ≥ 1, namely

1 
  U = Am=0 (0) U ≥ 1 , (7.47)
π d − U /2 + U n/2 2 +  2
7.3 From the Anderson Impurity Model to the Kondo Model 325

Fig. 7.2 Impurity density of

states of the Anderson
impurity model (7.29) within
the Hartree-Fock
approximation
  for

AHF (ω)
U  , d 

lower Hubbard upper Hubbard

band band

−U/2 0 U/2
ω

where Am=0 (0) is the Hartree-Fock impurity DOS at the chemical   potential in
absence of spontaneous magnetisation. Specifically, if U  , d , (7.46) has solu-
tion n = 1 and m = ±1, i.e., the impurity is half-filled and spin polarised. Since the
Hamiltonian (7.29) has spin SU (2) symmetry, we can freely rotate the axis of spin
polarisation and get always the same result. Therefore, if U is large the impurity
effectively behaves as S = 1/2 local moment, in accordance with the experimental
observation at TK  T  TF . In other words, the Curie-like
  behaviour is consistent
with the Anderson impurity model (7.29) if U  , d , which we assume hereafter.
Since, as we mentioned, the direction of the impurity spin is arbitrary, averaging
over that direction yields a spin-independent impurity Green’s function

1 1 1
G ( + i0+ ) + , (7.48)
2  + U /2 + i  + U /2 + i

and thus the impurity DOS is composed by two Lorentzian functions, one centred at
−U /2 below the chemical potential, and the other at U /2 above, which are called the
Mott-Hubbard bands, see Fig. 7.2. In conclusion, the large U Hartree-Fock solution
of the Anderson impurity model (7.29) seems to account for the behaviour above the
Kondo temperature, and satisfactorily explains how local moments appear. The key
ingredient is a large Coulomb repulsion at the magnetic impurity as compared to the
resonance broadening , which suppresses valence fluctuations. However, we still
have to understand what happens as temperature decreases.

7.3 From the Anderson Impurity Model to the Kondo Model

The Hartree-Fock solution of the Anderson impurity model (7.29) does explain in the
large U / -limit the existence of free local moments in a metal well below its Fermi
temperature. Nonetheless, the SU (2) symmetry breaking of the mean-field solution
is an artefact, since SU (2) symmetry cannot be locally broken by the Mermin-
Wagner theorem. The issue in that the Hamiltonian still allows for a matrix element
coupling two solutions corresponding to different orientations of the impurity spin.
326 7 Kondo Effect and the Physics of the Anderson Impurity Model

In order to understand the role of such coupling, it is more convenient to introduce

the conduction wave-function at the impurity site through

1 
c0σ = Vk ckσ , (7.49)
t
k

where

t2 = Vk2 ,
k
 
so that the impurity Hamiltonian, neglecting d   U , plus the hybridisation
becomes
 †  U
t c0σ dσ + dσ† c0σ + (n − 1)2 . (7.50)
σ
2
In the large U limit, the impurity traps a single electron, either with spin up or down,
and we can treat the hybridisation as a small perturbation lifting the degeneracy
between the two spin direction. Second order perturbation theory through interme-
diate states in which the impurity is either empty or doubly occupied, with energy
difference U /2 for large U , yields the following effective Hamiltonian
 †
HK = k ckσ ckσ + JK S0 · S , (7.51)
kσ

where
8V 2
JK = ,
U
and

1  † 1  † 1  †
S0 = c0α σ αβ c0β = 2 Vk Vp ckα σ αβ cpβ , S= dα σ αβ dβ ,
2 2t 2
αβ kpαβ αβ
(7.52)
are the conduction electron spin-density operator at the impurity site, S0 , and the
impurity spin-1/2 operator S. The Hamiltonian (7.51) is known as the Kondo
model [2], and describes conduction electrons antiferromagnetically coupled to a
local moment.
We note that HK has built-in a local moment, hence by construction correctly
describes the regime TK  T  TF . That local moment provides a scattering poten-
tial for the conduction electrons. The major difference with respect to a conventional
scalar potential is that the Kondo exchange has a non-trivial structure yielded by
the commutation relations between spin operators, whose consequences we now
investigate.
7.3 From the Anderson Impurity Model to the Kondo Model 327

Fig. 7.3 First order

corrections to the Kondo α γ, −iϵ β
exchange. Solid and dashed
lines indicate impurity and
(a) Ji Jj
conduction electron Green’s
a c, iϵ b
functions. Also indicated are
the spin labels. All external
lines are assumed to be at
zero frequency  = 0. The
β γ, −iϵ α
internal frequency, which is
going to be summed over, is (b) Ji Jj
also indicated a c, iϵ b

7.3.1 The Emergence of Logarithmic Singularities and the Kondo

Temperature

Let us analyse the role of the Kondo exchange

JK S0 · S , (7.53)

in perturbation theory [2]. Since (7.53) conserves independently the number of con-
duction and d-electrons, we treat the impurity spin in terms of d-electrons with an
unperturbed Green’s function
1
G (0) (z) = ,
z
namely as a localised state right at the chemical potential, which thus contains just
a single electron, number that is not going to be changed by (7.53). Since a single
electron has a well defined spin, it indeed acts like a local spin-1/2 moment. Moreover,
the Green’s function for the conduction electron c0 is

(0) 1  Vk2 1
G 0 (z) = = 2 (z) . (7.54)
t2 z − k t
k

For convenience let us rewrite (7.53) in a spin-asymmetric form as


Ji Si0 S i , (7.55)
i=x,y,z

and calculate the first order corrections to the exchange as given by the diagrams
in Fig. 7.3. For simplicity we assume that all external lines are at zero frequencies.
Each diagram is multiplied by (−1) because of first order perturbation theory. The
328 7 Kondo Effect and the Physics of the Anderson Impurity Model

explicit expression of diagram (a) is

  
G (0) (0)
j j
(a) = − Ji J j i
Sαγ i
Sac Sγβ Scb T 0 (i) G (−i)
i, j=x,y,z γc 
 Ji J j  i j j
 (i)
=− i
Sαγ Sac Sγβ Scb T
t 2
γc 
−i
i, j=x,y,z
 Ji J j  i j j
= −I (T ) S S Si S ,
t 2 γ c αγ γβ ac cb
i, j=x,y,z

where the function of temperature I (T ) is defined through

 
(i) d ( + i0+ )
I (T ) = −T = f () Im

i π 
 W  W  
d 1 d 1
= − f () = − f () − f (−) (7.56)
−W π  0 π 
  W
W d 1  d 1  W
= tanh  = ln .
0 π  2T T π  π T

Similarly, diagram (b) is

  
G (0) (0)
j j
(b) = − Ji J j i
Sαγ i
Scb Sγβ Sac T 0 (i) G (i)
i, j=x,y,z γc 
 Ji J j  i j j
= I (T ) i
S S Sac Scb ,
t 2 γ c αγ γβ
i, j=x,y,z

hence the sum is

 Ji J j  i j
 j j

(a) + (b) = I (T ) 2
Sαγ Sγβ i
Sac Scb − Sac
i
Scb
t γ c
i, j=x,y,z
 Ji J j  i j
= I (T ) S S i  jik Sab
k
,
t 2 γ αγ γβ
i, j,k=x,y,z

where we used the commutation relations between the spin operators

 
S i , S j = i i jk S k ,
7.3 From the Anderson Impurity Model to the Kondo Model 329

where i jk is the antisymmetric tensor. Since we sum over i and j, we can also write
 
1 Ji J j  j
(a) + (b) = I (T ) Sαγ Sγβ i  jik Sab + (i ↔ j)
i k
2 t2 γ
i, j,k=x,y,z
1  Ji J j  j

= I (T ) i  jik S k
ab S i
αγ Sγβ − S j
αγ S i
γβ
2 t2 γ
i, j,k=x,y,z
⎡ ⎤
 1  Ji J j  2
= k
Sab k ⎣
Sαβ I (T ) i jk  ⎦ ,
2 t2
k=x,y,z i, j

which provides the first order correction to the exchange constants according to
1  Ji J j  2
Jk + δ Jk = Jk + I (T ) i jk 
2 t2
i, j
(7.57)
1  W   2
= Jk + 2
ln Ji J j i jk  ,
2π t T
i, j

namely,
 W  W  W
δ Jx = lnJy Jz , δ Jy = 2 ln Jz Jx , δ Jz = 2 ln Jx J y .
πt 2 T πt T πt T
(7.58)
Therefore, perturbation theory generates logarithmic singularities [2] that become
visible roughly around a temperature, which has to be identified with the Kondo
temperature TK , at which the correction exceeds the bare exchange. In the isotropic
case, Jx = Jy = Jz = J , that implies
 2 W
J= J ln ,
πt 2 TK
thus
πt2 πU
TK = W exp − = W exp −  W ∼ TF , (7.59)
J 8
since U  . For instance, the resistivity calculated through the Fermi golden rule,
see (7.22) and (7.23), at first order in perturbation theory would looks like
2
 2 W
R(T ) − R0 (T ) ∝ n i ρ0 J + J ln ,
πt 2 TK
where R0 (T ) is the value in absence of magnetic impurities and decreases monoton-
ically lowering T . It follows a minimum metallic resistivity around TK , followed by
a logarithmic raise, just like in experiments. In agreement with the latter, we do find
that the Kondo temperature is much smaller than the host-metal Fermi temperature.
Since perturbation theory becomes meaningless below TK , the next obvious question
is how to proceed further.
330 7 Kondo Effect and the Physics of the Anderson Impurity Model

7.3.2 Anderson’s Poor Man’s Scaling

If Jx = Jy ≡ J⊥ = Jz , through (7.58) we can write

 W
Jz [W ] + δ Jz [W ] = Jz [W ] + J⊥ [W ]2 ln ,
πt 2 T (7.60)
 W
J⊥ [W ] + δ J⊥ [W ] = J⊥ [W ] + J⊥ [W ] Jz [W ] ln .
π t2 T
In (7.60) we explicitly indicate the dependence upon the high-energy cutoff W . Now
suppose we have another model with a smaller conduction electron bandwidth cut-
off W (λ) = W /λ with λ > 1, different J ’s but equal /t 2 . In this case, we would
get the first order corrections

 W
Jz [W (λ)] + δ Jz [W (λ)] = Jz [W (λ)] + J⊥ [W (λ)]2 ln ,
π t2 λT
 W
J⊥ [W (λ)] + δ J⊥ [W (λ)] = J⊥ [W (λ)] + J⊥ [W (λ)] Jz [W (λ)] ln .
π t2 λT
On the other hand, it makes not really a big difference for the low energy behaviour
whether those electrons close to the chemical potential derive from a bandwidth W
or W (λ) provided they suffer the same scattering off the impurity. Therefore we can
ask the following question [3]: What should J [W (λ)] be in order for the effective
exchange up to first order to be the same as that with bandwidth W ?
The answer is quite simple, since we can, for instance, write

  W
Jz [W ] + δ Jz [W ] = Jz [W ] + J⊥ [W ]2 ln λ + J⊥ [W ]2 ln
π t2 π t2 λT

Jz [W ] + J⊥ [W ]2 ln λ
π t2
  2
W  
+ Jz [W ] + J⊥ [W ]2 ln λ ln + O J3 ,
π t2 πt 2 λT

the equality being valid up to the order at which we stop the expansion. Therefore
the two models with W and W /λ have the same spin exchange up to first order in
perturbation theory provided the bare exchange constants satisfy


Jz [W (λ)] = Jz [W ] + J⊥ [W ]2 ln λ,
π t2

J⊥ [W (λ)] = J⊥ [W ] + J⊥ [W ] Jz [W ] ln λ ,
π t2
which can be cast in a differential form
d jz (s) d j⊥ (s)
= 2 j⊥ (s)2 , = 2 j⊥ (s) jz (s) , (7.61)
ds ds
7.3 From the Anderson Impurity Model to the Kondo Model 331

having defined λ = es with s ≥ 0, and


ji (s) ≡ Ji [W (λ)] .
2π t 2
The equations (7.61) are exactly the renormalisation group equations (6.89) in
Sect. 6.2.8, with g replaced by jz and g3 by j⊥ .1 These equations describe how the
bare exchange constants have to be modified in order for the model with a reduced
bandwidth W e−s to have the low-energy scattering amplitudes equal to those of the
original model [3]. The idea behind is the same as in Sect. 6.2.8: if we are able to
follow the evolution of ji (s) until W e−s ∼ T , at this point we can rely on pertur-
bation theory since ln(W /λT ) 0. This is more or less how Anderson formulated
his poor man’s scaling theory for the Kondo problem in 1970, in essence the first
implementation of a Renormalisation Group transformation.
The results can be simply borrowed from Fig. 6.6 recalling that g = jz and
g3 = j⊥ , which we can take positive. In particular, if j⊥ < − jz > 0, the case of
an easy-axis ferromagnetic Kondo exchange, j⊥ (∞) = 0 and jz (∞) = jz∗ < 0. In
this case, the impurity polarises along z and provides an effective local magnetic
field for the conduction electrons. This model can be readily solved. In the particular
case j⊥ = − jz > 0, both exchange constants flows to zero; the magnetic impurity
asymptotically decouples from the conduction electrons.
Everywhere else in the flow diagram 6.6, the exchange constants flows to +∞. As
discussed in Sect. 6.2.8, in this case we must first diagonalise the Kondo exchange
JK S0 · S, and next treat what is left in perturbation theory. The Kondo exchange
is minimised by locking the conduction electron c0σ in (7.49) with the impurity
spin into a spin-singlet state. After that, the impurity site becomes inaccessible to
other conduction electrons, which would break the Kondo singlet, in that equivalent
to a scattering potential yielding a π/2 phase shift. More precisely, the conduction
electron local Green’s function at the impurity site is, see (7.54),

(0) 
G 0 ( + i0+ ) = −i .
t2
If the magnetic impurity asymptotically behaves as a conventional scalar potential
V∗ , the local Green’s function changes into, see (7.8),2

G 0 ( + i0+ ) = G (0) + (0) + + (0) +

0 ( + i0 ) + G 0 ( + i0 ) T ( + i0 ) G 0 ( + i0 ) ,

where now
V∗
T ( + i0+ ) = (0)
.
1− V∗ G 0 ( + i0+ )

1 We note that the sign of j⊥ is unimportant as it can be changed by an 180◦ rotation around the
z-axis.
2 The conduction electron combination c
0σ in (7.49) is the only coupled to the impurity, and thus
the T -matrix is diagonal with the only finite element referring to c0σ .
332 7 Kondo Effect and the Physics of the Anderson Impurity Model

Since the impurity site in inaccessible after the Kondo singlet is formed, then G 0 ( +
i0+ ) must vanish, which implies V∗ = ∞ and thus

1 t2
T ( + i0+ ) = − = −i ,
G (0) +
0 ( + i0 )


hence a π/2 phase shift, consistent with experiments, as well as the low-temperature
screening of the magnetic impurity yielding a Pauli-like magnetic susceptibility.

7.4 Noziéres’s Local Fermi Liquid Theory

A better description of the low temperature behaviour can be obtained through the
diagrammatic many-body perturbation theory presented in Chap. 4, especially in con-
nection with the microscopic derivation of Landau’s Fermi liquid theory in Chap. 5.
Indeed, a similar Fermi liquid picture can be also derived for Anderson impurity
models, in this case known as Noziéres local Fermi liquid theory [4].
Let us consider again the Anderson impurity model (7.29). The non-interacting
and fully-interacting impurity Green’s functions are

1 A0 (ω)
G0 (i) = = dω ,
i − d + i  sign() i − ω
 (7.62)
1 A(ω)
G (i) = = dω ,
i − d − (i) + i  sign() i − ω

where A0 (ω) and A(ω) are the corresponding densities of states, and we correctly
assume unbroken SU (2) symmetry.
We start calculating the contribution to the retarded self-energy given by the
second order diagram in Fig. 7.4, exactly as we did in Sect. 4.3.4. We can simply
borrow the result of (4.58) and adapt it to the present case, thus finding

      1
(i) = U 2 d0 d1 dω1 A0 0 A0 1 A0 ω1
i + ω1 − 0 − 1

             
f ω1 1 − f 0 1 − f 1 + 1 − f ω1 f 0 f 1 .
(7.63)

Fig. 7.4 Second order i ω1

self-energy diagram of the
Anderson impurity model
(7.29)

iϵ 1

iϵ iϵ + i ω 1 − iϵ 1 iϵ
7.4 Noziéres’s Local Fermi Liquid Theory 333
 
The Im   + i0+ for very small || and temperature can be readily obtained as in
Sect. 4.3.4,

         
Im   + i0+ = −π U 2 d0 d1 dω1 A0 0 A0 1 A0 ω1 δ  + ω1 − 0 − 1

             
f ω1 1 − f 0 1 − f 1 + 1 − f ω1 f 0 f 1

 
−π U 2 A0 (0)3 d0 d1 dω1 δ  + ω1 − 0 − 1

             
f ω1 1 − f 0 1 − f 1 + 1 − f ω1 f 0 f 1
π  
− U 2 A0 (0)3  2 + π 2 T 2 .
2
(7.64)
Therefore, exactly like in the case of bulk interacting electrons,
  we arrive at the
conclusion that, at any order in perturbation theory, Im   + i0+ vanishes at least
quadratically in  at T = 0. Since at U = 0 as well as at large U the Hamiltonian
(7.29) supposedly describes a resonant level model with interaction dependent width,
we can safely
 assume  that perturbation theory never breaks down, and thus that the
result Im   + i0+ −γ  2 holds true beyond the perturbative regime. It follows
that, for very small || at T = 0, and upon defining the conventional quasiparticle
residue
 
∂Re   + i0+ −1
Z () ≡ 1 − , (7.65)
∂
the inverse of the retarded impurity Green’s function
 
G()−1 =  − d −   + i0+ + i Z −1 (0)  − d − (0) + i
    
= Z −1 (0)  − Z (0) d + (0) + i Z (0)  ≡ Z −1 (0)  − d∗ + i∗ ,

where we have neglected γ  2 which is much smaller than , and we recall that
(0) ∈ R. In conclusion,

Z (0)
G () + Ginc () ≡ Z (0) Gqp () + Ginc () , (7.66)
 − d∗ + i∗

where Ginc () is the high-energy incoherent component that we know must describe
lower and upper Hubbard bands, on top of which a narrow resonance of weight Z (0)
appears near zero energy, see Fig. 7.5. Filtering out Z () yields the quasiparticle
Green’s function, Gqp (), that describes a resonant level with width Z (0)  that must
be identified with the Kondo temperature TK , and centred at
  TK  
d∗ = Z (0) d + (0) = d + (0) . (7.67)

334 7 Kondo Effect and the Physics of the Anderson Impurity Model

Kondo resonance

A(ω)

lower Hubbard upper Hubbard

band band

−U/2 0 U/2
ω
Fig. 7.5 Sketch of the interacting impurity density of states at low temperature. Besides the high
energy Hubbard bands, a narrow resonance emerges at low temperature that is pinned at zero energy
and whose width is the Kondo temperature TK . This behaviour is believed to be the paradigm of the
local density of states of correlated metals close to a Mott transition, where the resonance correspond
to a very narrow band of quasiparticles

To get an order of magnitude of d + (0), let us recall the Hartree-Fock result (7.46)
at m = 0, i.e., when (0) = HF = −U (1 − n)/2 ≡ −U δn/2, and assuming δn
0,

2 2d − U δn U π δn π δn
δn = tan−1 ⇒ d − δn =  tan  .
π 2 2 2 2
In this case
U d π
d + HF = d − δn d .
2 1 + U /π  U

Therefore, if we approximate (0) HF , then

π TK  
d∗ d ⇒ d∗   TK , (7.68)
U
namely, the resonance is essentially pinned at the chemical potential, consistently
with the experimental observations.
After the above preamble, let us apply the whole diagrammatic technique to better
analyse the Anderson impurity model, which we here rewrite for convenience,
    U  2
† †
H= k ckσ ckσ + Vk ckσ dσ + dσ† ckσ + d n + n−1
2
kσ kσ
 †
 †  U  2
(7.69)
= k ckσ ckσ +t c0σ dσ + dσ† c0σ + d n + n−1 ,
σ
2
kσ
7.4 Noziéres’s Local Fermi Liquid Theory 335


with c0σ defined in (7.49), and n = σ dσ† dσ . Even in the present case, we can intro-
duce a Luttinger-Ward
  functional, here functional of the impurity Green’s function
only, i.e.,  G , so that
  
δ G = T eiη (i) δ G (i) ,
i

and, as usual, η > 0 is infinitesimal. Similarly, in the generic case in which, because
of external fields, Gσ1 σ2 (i1 , i2 ) ≡ G (1, 2) becomes non diagonal in frequency and
spin, as well as (1, 2),

δ(1, 2)  
= 0 (1, 2; 3, 4) ≡ 0 σ1 ,σ2 ;σ3 ,σ4 i1 , i2 ; i3 , i4 , (7.70)
δ G (3, 2)

with 1 + 2 = 3 + 4 and σ1 + σ2 = σ3 + σ4 , is by definition the irreducible inter-

action vertex in the particle-hole channel, which, through the Bethe-Salpeter equation
in Fig. 4.32, where now the internal lines are the fully interacting
 impurity Green’s

functions, yields the reducible interaction vertex σ1 ,σ2 ;σ3 ,σ4 i1 , i2 ; i3 , i4 .
For later convenience, we recall that
  
1
−Tτ dσ (τ ) c0σ†
(τ  )  = dτ1 G (τ − τ1 ) (τ1 − τ  ) ,
t
   (7.71)
†  1 
−Tτ c0σ (τ ) dσ (τ )  = dτ1 (τ − τ1 ) G (τ1 − τ ) ,
t

where

(τ ) = T e−iτ (i) ,

is the hybridisation function in imaginary time.

7.4.1 Ward-Takahashi Identity

The Hamiltonian (7.69) admits as conserved quantities the total charge and any
component of the total spin. For simplicity we shall consider the z-component, so
that we can use the following two conserved quantities
 †
 †
Nσ = ckσ ckσ + dσ† dσ = ckσ ckσ + n σ , σ = ↑, ↓ . (7.72)
k k

We observe that
∂n σ    
†
− = n σ , H = −t c0σ dσ − dσ† c0σ ≡ Jσ . (7.73)
∂τ
336 7 Kondo Effect and the Physics of the Anderson Impurity Model

Fig. 7.6 Diagrammatic

representation of the vertex
function  in (7.74), and of
the corresponding
Ward-Takahashi identity
(7.75), using (7.71). The
dashed line between the two
black circles represents the
hybridisation function (τ )

Following what we did in Sect. 4.9, we define, see Fig. 7.6,

     
(τ σ ; τ1 , τ2 , σ  ) ≡ − Tτ n σ (τ ) dσ†  (τ1 ) dσ  (τ2 )  = δσ,σ  G τ2 − τ G τ − τ1

   
+ dτ1 dτ2 dτ3 dτ4 G τ4 − τ G τ − τ1
 
σ,σ  ;σ  ,σ τ1 , τ2 ; τ3 , τ4 G (τ2 − τ2 ) G (τ1 − τ1 ) ,
(7.74)
and find that [5,6]

∂  
− (τ σ ; τ1 , τ2 , σ  ) = − Tτ Jσ (τ ) dσ†  (τ1 ) dσ  (τ2 ) 
∂τ
        
+ δσ,σ  δ τ − τ1 G τ2 − τ − δ τ − τ2 G τ − τ1 ,
(7.75)
shown diagrammatically in Fig. 7.6. If we now take the Fourier transform, we find
 
δσ,σ  iω − (i + iω) + (i) G (i + iω) G (i)
 
+T iω − (i  + iω) + (i  ) G (i  + iω) G (i  )

σ,σ  ;σ  ,σ (i  + iω, i; i + iω, i  ) G (i + iω) G (i)
 
= δσ,σ  G (i) − G (i + iω) ,
7.4 Noziéres’s Local Fermi Liquid Theory 337

so that, dividing by G (i + iω) G (i),

 
δσ,σ  iω − (i + iω) + (i)
 
+T iω − (i  + iω) + (i  ) G (i  + iω) G (i  )

σ,σ  ;σ  ,σ (i  + iω, i; i + iω, i  )
 
= δσ,σ  G (i + iω)−1 − G (i)−1
 
= δσ,σ  iω − (i + iω) + (i) − (i + iω) + (i) ,

thus the final expression of the Ward-Takahashi identity [5,6]

   
δσ,σ  (i + iω) − (i) = −T iω − (i  + iω) + (i  ) G (i  + iω) G (i  )

σ,σ  ;σ  ,σ (i  + iω, i; i + iω, i  ) ,
(7.76)
which can also be written as

∂(i)  iω − (i  + iω) + (i  )

δσ,σ  = − lim T G (i  + iω) G (i  )
∂i ω→0+ iω


σ,σ  ;σ  ,σ (i  + iω, i; i + iω, i  ) .

(7.77)
The limit ω → 0+ of the right hand side requires some care, similar to that we take
in Sect. 5.2.1. The main issue is the term

iω − (i  + iω) + (i  )

,
iω
 
which is simply one if     + ω > 0, the semi infinite lines of the imaginary axis
within the red regions in Fig. 7.7, while it is

iω + 2i 2i
,
iω iω
for −ω <   < 0, the segment of the imaginary axis within the light green region in
Fig. 7.7, thus also leading to a finite result since the sum over   is proportional to ω.
The sum over the over the Matsubara frequency   in (7.77) can be transformed as
usual in the contour integrals shown in Fig. 7.7. The two contours C1 and C2 involve
singularities of the Green’s functions but also of the reducible interaction vertex.
However, their sum is simply the right hand side of (7.77) calculated right at ω = 0,
thus

C1 + C2 = −T G (i  )2 σ,σ  ;σ  ,σ (i  , i; i, i  ) .

338 7 Kondo Effect and the Physics of the Anderson Impurity Model

On the contrary,

d  2i 
C3 = − lim f (  ) G (  + iω) G (  − i0+ )
ω→0+ 2πi iω
σ,σ  ;σ  ,σ (  + iω, i; i + iω,   − i0+ )

− G (  + i0+ ) G (  − iω) σ,σ  ;σ  ,σ (  + i0+ , i; i + iω,   − iω) .

We take the analytic continuation iω → ω + i0+ , after which


d  2i 
C3 = − lim f (  ) G (  + ω + i0+ ) G (  − i0+ )
ω→0 2πi ω
σ,σ  ;σ  ,σ (  + ω + i0+ , i; i + ω,   − i0+ )

− G (  + i0+ ) G (  − ω − i0+ ) σ,σ  ;σ  ,σ (  + i0+ , i; i + ω,   − ω − i0+ )


d  f (  ) − f (  + ω)
= − lim G (  + ω + i0+ ) G (  − i0+ )
ω→0 π ω
σ,σ  ;σ  ,σ (  + ω + i0+ , i; i + ω + i0+ ,   − i0+ )
  
d ∂ f ( )
 G (  + i0+ ) G (  − i0+ ) σ,σ  ;σ  ,σ (  , i; i,   ) ,
π ∂ 

where we assumed that the interaction vertex is independent of the infinitesimal

imaginary part of the external frequencies. In conclusion, if we analytically continue
i →  + i0+ with small  and recalling that Im ( + i0+ ) ∼ − 2 , we find

∂( + i0+ )   
δσ,σ  δσ,σ  1 − Z ()−1 = −T G (i  )2 σ,σ  ;σ  ,σ (i  , ; , i  )
∂


d  ∂ f (  )
+ G (  + i0+ ) G (  − i0+ ) σ,σ  ;σ  ,σ (  , ; ,   ) .
π ∂ 

Because of the Fermi distribution function derivative,   0, so that

1 1
G (  + i0+ ) G (  − i0+ )  
 − d − Re () + i  − d − Re () − i
1   π
= G (  − i0+ ) − G (  + i0+ ) = A(  ) ,
2i 
and thus [5,6]

∂( + i0+ )   
δσ,σ  δσ,σ  1 − Z ()−1 = −T G(i  )2 σ,σ  ;σ  ,σ (i  , ; , i  )
∂

 
∂ f ( )
+ d  A(  ) σ,σ  ;σ  ,σ (  , ; ,   ) .
∂ 
(7.78)
7.4 Noziéres’s Local Fermi Liquid Theory 339

Im(z)

C1
Re(z)

C3
−iω

C2

Fig. 7.7 Contour integral used to evaluate the sum over the Matsubara frequency   in (7.77). The
contour actually comprises three different ones, C1 , C2 and C3 , running anticlockwise and avoiding
the imaginary axis, not explicitly shown

7.4.2 Luttinger’s Theorem and Thermodynamic Susceptibilities

Since there is no breakdown of perturbation theory whatever the value of U is,

Luttinger’s theorem must be valid, and thus
 +  
Nσ = T ei0 G (0) (i, k) + Vk2 G (0) (i, k)2 G (i) + G (i)
 k
 + ∂ (i)  + ∂ ln G (i)
= N0σ + T ei0 1− G (i) ≡ N0σ − T ei0 ,

∂i 
∂i

where N0σ is the value in absence of the impurity. Therefore

 
+ ∂ ln G (i) d ∂Im ln G ()
Nσ ≡ Nσ − N0σ = −T ei0 = f ()

∂i π ∂

d ∂ f ()
=− Im ln G () ,
π ∂
(7.79)
where, by definition, G () ≡ G ( + i0 + ). Suppose we add to the Hamiltonian the

perturbation δ H = − σ h σ Nσ , in which case all Green’s functions as well as the
impurity self-energy become dependent on h ≡ (h ↑ , h ↓ ) and different for both spin
directions. Specifically,
1 1
Gσ (i, h) = = ,
i − d + h σ − σ (i, h) − σ (i, h) G 0σ (i, h) − σ (i, h)
(7.80)
where
 Vk2  Vk2
σ (i) = (i) → σ (i, h) = ,
i − k i − k + h σ
k k
340 7 Kondo Effect and the Physics of the Anderson Impurity Model

which, in the large bandwidth limit we have so far assumed, is actually independent
of h σ , thus
∂ σ (i, h)
= 0.
∂h σ
By definition, the impurity contribution to the susceptibility is

imp ∂ Nσ (h) d ∂ f () ∂Im ln Gσ (, h)
χσ σ  ≡  =− 
∂h σ  h=0 π ∂ ∂h σ  h=0
  (7.81)
d ∂ f () ∂σ (, h)
= Im G () δσ,σ  −  .
π ∂ ∂h σ  h=0

On the other hand3


−1
δσ (i, h) = −T σ,σ  ;σ  ,σ (i, i  ; i  , i) Gσ  (i  , h)2 δ G0σ 
 (i , h) ,
 σ

and thus
 −1 
∂σ (i, h) ∂ G0σ  (i , h)
 = −T σ,σ  ;σ  ,σ (i, i  ; i  , i) Gσ  (i  )2 
∂h σ  h=0 ∂h σ  h=0
  σ 

= −T σ,σ  ;σ  ,σ (i, i  ; i  , i) Gσ  (i  )2 ,


which, using (7.78), and after analytic continuation i →  + i0+ , can be also writ-
ten as
∂σ (, h)    ∂ f (  )
 = δσ,σ  1 − Z () − d 
−1
A(  ) σ,σ  ;σ  ,σ (  , ; ,   ) .
∂h σ  h=0 ∂ 
(7.82)

3 In compact notations,
 
δ = 0  δ G = −0  G 2  δ G0−1 − δ ,

thus
   −1
1 − 0  G 2  δ = −0  G 2  δG0−1 ⇒ δ = − 1 − 0  G 2  0  G 2  δG0−1 .

Since the Bethe-Salpeter equation reads

 −1
 = 0 + 0  G 2   ⇒ 1 − 0  G 2   = 0 ,

substituting in the previous equation we get the desired result

δ = −  G 2  δ G0−1 .
7.4 Noziéres’s Local Fermi Liquid Theory 341

Substituting (7.82) in (7.81) we find


d ∂ f ()
Im G ( + i0+ ) δσ,σ  Z ()−1
imp
χσ σ  =
π ∂
  
 ∂ f ( )   
+ d A( ) σ,σ  ;σ  ,σ ( , ; ,  ) .
∂ 
(7.83)
Since both  and   are small due to the derivatives of the Fermi distribution func-
tions, we expect the interaction vertex to be real and thus the imaginary part is only
contributed by Im G ( + i0+ ) = −π A(), so that
  
∂ f () ∂ f (  )
A() δσ,σ  Z ()−1 + d    
imp
χσ σ  = − d A( ) σ,σ  ;σ  ,σ ( , ; ,  ) .
∂ ∂ 
(7.84)
For small , by definition
A() ≡ Z () Aqp () ,
where the quasiparticle DOS Aqp () is normalised to one and represents a narrow
resonance pinned close to zero energy. Therefore, upon defining the quasiparticle
scattering amplitudes

Aσ,σ  ;σ  ,σ (  , ; ,   ) ≡ Z () Z (  ) σ,σ  ;σ  ,σ (  , ; ,   ) , (7.85)

Equation (7.84) can be written as

  
∂ f () ∂ f (  )
d  Aqp (  ) Aσ,σ  ;σ  ,σ (  , ; ,   ) ,
imp
χσ σ  = − d Aqp () δσ,σ  + 
∂ ∂
(7.86)
where the analogy with Landau’s Fermi liquid theory is self-evident, see (5.57).
If we assume that h ↑ = h c + h s and h ↓ = h c − h s , where h c refers to the charge
compressibility and h s to the magnetic susceptibility, we readily find [5,6]
  
 ∂ f () ∂ f (  )
d  Aqp (  ) A S (  , ; ,   )
imp
κ imp = χσ σ = −2 d Aqp () 1 + 
∂ ∂
σσ
 
2 Aqp (0) 1 − A S ,
  
 ∂ f () ∂ f (  )
σ σ  χσ σ  = −2 d   ) A A (  , ; ,   )
imp
χ imp = d Aqp () 1 + A (
∂ 
qp
∂
σσ
 
2 Aqp (0) 1 − A A ,
(7.87)
where
1 
A S (  , ; ,   ) ≡
A↑,↑;↑,↑ (  , ; ,   ) + A↑,↓;↓,↑ (  , ; ,   ) ,
2
A   1 
A ( , ; ,  ) ≡ A↑,↑;↑,↑ (  , ; ,   ) − A↑,↓;↓,↑ (  , ; ,   ) ,
2
342 7 Kondo Effect and the Physics of the Anderson Impurity Model

and

∂ f () ∂ f (  )
Aqp (0) A S(A) ≡ d d  Aqp () Aqp (  ) A S(A) (  , ; ,   ) .
∂ ∂ 

We could readily follow the derivation of the specific heat in a Fermi liquid, see
Sect. 5.4.3, through the Ward-Takahashi identity for the heat density, and find in the
present case that

imp 2 ∂ f () 2π 2
cV − d  2 Aqp () = T Aqp (0) , (7.88)
T ∂ 3

namely, the impurity contributes to the specific heat with a linear in temperature
term. Similarly, in the Kondo regime U  , valence fluctuations of the impurity
are suppressed thus

1  A↑,↓;↓,↑
AS = A↑,↑;↑,↑ + A↑,↓;↓,↑ 1,
2 2
1  (7.89)
AA = A↑,↑;↑,↑ − A↑,↓;↓,↑ −1 ,
2
yielding a Wilson ratio (5.68) RW 2. We mention that the above procedure can
be readily extended to multi-orbital Anderson impurity model that better describe
realistic magnetic ions with partially filled d or f shells.
We end remarking that a necessary condition for the local Fermi liquid description
to apply is
 
∂ f () ∂ f ()
0 = Aqp (0) = d − Aqp () = d − Z ()−1 A()
∂ ∂
 (7.90)
d ∂ f () ∂Re()  − Im()
= − 1−  2  2 ,
π ∂ ∂
 − d − Re() +  − Im()

and is clearly verified in the single-orbital Anderson impurity model, where Z (0)
and A() are both finite, but remains valid also when Re() ∼ 1/, in which case
Z ()−1 ∼ 1/ 2 compensates the vanishing A() ∼  2 , which represents the local
counterpart of the quasiparticles at Luttinger surfaces discussed in Sect. (5.1). Since
the conduction electron T -matrix is t 2 G (), a Re() ∼ 1/ corresponds to van-
ishing phase shift.
References 343

Problems

7.1 Mean-field approximation of two coupled Anderson impurities—Consider

two Anderson impurities with Hamiltonian


2    
† † †
H= k ckσ ckσ + Vk ckσ dσ + dσ ckσ
=1 kσ kσ (7.91)
U 2  † †

+ n − 1 − t⊥ d1σ d2σ + d2σ d1σ ,
2 σ

where  = 1, 2 labels the two impurities.

• Find under which condition the Hartree-Fock approximation yields spin-

polarised impurities.
• Calculate the effective Kondo model in the limit of large U .

References
1. P.W. Anderson, Phys. Rev. 124, 41 (1961). https://doi.org/10.1103/PhysRev.124.41
2. J. Kondo, Prog. Theor. Phys. 32, 37 (1964)
3. P.W. Anderson, J. Phys. C: Solid State Phys. 3(12), 2436 (1970)
4. P. Nozières, J. Low Temp. Phys. 17(1), 31 (1974). https://doi.org/10.1007/BF00654541
5. A. Yoshimori, A. Zawadowski, J. Phys. C: Solid State Phys. 15(25), 5241 (1982)
6. L. Mihaly, A. Zawadowski, J. Phys. Lett. (France) 39, L483 (1978)